Submitted:
21 October 2026
Posted:
21 October 2026
You are already at the latest version
Abstract
Hospital closure is concentrated in some communities, but it is not known through which hospital characteristics county poverty, racial composition, and rurality are linked to closure. We analyzed Medicare cost reports for short-term acute care and critical access hospitals in the 50 states and the District of Columbia (reports of 150–365 days; 4,785 hospitals, 45,107 hospital-years, 325 closures, 2011–2020), linked to county poverty and percentage Black from the American Community Survey. A parallel multiple-mediator model with 16 mediators estimated associations with closure in the following year as risk differences per 1000 hospital-years, with a hospital-clustered bootstrap. Each standard deviation of county poverty was associated with 1.64 (95% CI 0.72, 2.60) more closures, 59% of it indirect, through critical access status, total margin, occupancy, and for-profit ownership, partly offset by larger hospital size. County percentage Black had a direct association (1.58; 0.52, 2.74) masked by opposing indirect associations (−0.83). Rural location showed no net association because pathways through smaller size and lower margins were offset by critical access status. Days cash, payer mix, charity care, and bad debt did not mediate any association. Community poverty is linked to closure mainly through margins, utilization, structure, and ownership, not uncompensated care; monitoring should address margins and occupancy.
Keywords:
hospital closure
; county poverty
; racial composition
; rural hospitals
; mediation analysis
; hospital finances
; uncompensated care
1. Introduction
Hospital closures have become more frequent in the United States, particularly in rural areas. The rate of rural hospital closures rose in the early 2010s [1], and a federal review documented rural hospital closures between 2013 and 2017 and the factors associated with them [2]. A substantial share of rural hospitals operate with thin or negative margins, and their financial viability has varied widely across states and years [3,4]. Closures are not confined to rural areas: urban hospitals, including trauma centers and emergency departments, have also closed [5,6,7].
Closure harms the communities that lose a hospital. It increases the distance to care and has been linked to worse outcomes for nearby residents [8], longer emergency medical service response and transport times [9], loss of the local physician workforce [10], higher inpatient mortality [5], and changes in the cost and quality of the care that remains [11]. Closure also weakens local economies that depend on the hospital as an employer [12], and case studies describe the access problems that follow [13].
County poverty is a plausible contributor to closure. Hospitals in poorer communities serve more patients who are uninsured or covered by Medicaid and provide more uncompensated care [14,15]. Closures of trauma centers and emergency departments have disproportionately burdened vulnerable populations [6,7], and urban public hospitals that serve poor communities have faced threats to their survival [16].
The racial composition of the community is also linked to hospital survival. Racial residential segregation concentrates poverty and disadvantage in neighborhoods and shapes where services are located [17,18]; it has been associated with the availability of health care resources such as primary care physicians [19] and with the survival of urban public hospitals [16]. Black patients are concentrated in a small number of lower-quality hospitals [20], which makes the finances of hospitals in predominantly Black communities relevant to health inequalities [18].
Rural location is the best established correlate of closure. Rural hospitals are smaller, serve lower-volume populations, and depend heavily on Medicare, which leaves them exposed to payment changes [4,21], and rural closures have been studied extensively [1,2,22]. Critical access hospital designation, which pays small rural hospitals on a cost basis, was created to protect access in rural communities [23,24], and state Medicaid expansion has been linked to a lower probability of rural hospital closure [25].
These community characteristics may be linked to closure through the finances of the hospital. Hospitals that close have lower operating and total margins in the years before closure [22,26], and indicators of liquidity and leverage help to predict financial distress and closure in rural hospitals [22,23]. Margins reflect payer mix, uncompensated care, and the cost structure of the hospital [3,27], and weaker financial condition has implications for the quality of care [28]. Assets, meaning the balance-sheet position summarized by days cash on hand, the current ratio, and leverage, indicate how long a hospital can absorb losses and are distinct from performance, meaning the margins earned in a year [22,23].
Hospital size and utilization are also candidate mediators. Small hospitals and hospitals with low occupancy are more likely to close [1,21,22]. Staffing per bed is a cost driver that differs by type of hospital [29]. Structural features may also matter: critical access hospitals are paid on a cost basis and have a distinct financial profile [4,23], and teaching hospitals tend to be large urban hospitals with additional payments and costs [30].
Payer mix and uncompensated care are the most direct channels from community poverty to the hospital. A high Medicare share of days exposes rural hospitals to federal payment policy [4], whereas Medicaid coverage changes the share of patients who cannot pay and has been linked to closure [25]. Uncompensated care, the sum of charity care and bad debt, is a large financial burden for hospitals in poor communities [14,15]; charity care is a policy commitment whereas bad debt reflects failed collection [31,32], and the relation of charity care to tax treatment differs between nonprofit and government hospitals [33]. Finally, ownership shapes finances, charity care, and closure: for-profit, nonprofit, and government hospitals differ in how resources are obtained and used [29,34], and for-profit ownership has been associated with closure of hospital services [6].
What is not known is which of these hospital characteristics transmit the community-level associations of poverty, racial composition, and rurality to closure. The three community factors are correlated, so they must be estimated together, and the pathways may work in opposite directions, so that an indirect association hides a direct one [35]. Mediation analysis decomposes the association of an exposure with an outcome into direct and indirect components through one or several mediators [36,37,38,39,40], and it can be extended to many mediators measured together [38,40].
Aim 1: to estimate the joint total, direct, and indirect associations of county poverty, county percentage Black, and rurality with hospital closure. Closure monitoring and policy responses are usually framed around hospital finances and location [1,2,22]. How much of the community-level association is explained by measurable hospital characteristics shows whether community risk is carried by features of hospitals that can be observed and acted on, or whether it remains after those characteristics are accounted for and so points to other community mechanisms [40]. Estimating the three exposures together is necessary because each is correlated with the others [16,17].
Aim 2: to identify which groups of mediators, and which individual mediators, carry each association. The policy levers differ by pathway: margins and utilization respond to payment and market conditions [4,27], uncompensated care to financial assistance and coverage policy [14,25,32], critical access and teaching status to statute and designation [23,30], and ownership to organizational decisions [29,34]. Which pathways carry the associations therefore determines which responses are plausible, and an analysis that examines uncompensated care, assets, and performance side by side can show which of them matter for closure and which, despite differing by community, do not.
2. Materials and Methods
2.1. Data Source and Eligibility
Data came from the Centers for Medicare & Medicaid Services Healthcare Provider Cost Reporting Information System public use files for federal fiscal years 2011 through 2023 [41]. Every Medicare-certified institutional provider files an annual cost report, so the file is a census of participating hospitals. Hospitals were eligible if they were short-term acute care hospitals (certification numbers 0001–0879) or critical access hospitals (1300–1399) according to the facility-type classification in the file; psychiatric, rehabilitation, long-term care, children's, and religious nonmedical institutions and a small category labeled ORD were excluded. Where a hospital filed more than one report for a fiscal year, the longer report was kept. Reports were restricted to periods of 150 to 365 days, measured as the fiscal-year end date minus the begin date, so that full years, including leap years, were retained and brief reports arising when hospitals open, close, change ownership, or change fiscal year were excluded. Hospitals with a classifiable ownership code (1–13) in the 50 states and the District of Columbia were retained; US territories were excluded. The closure analysis used hospitals with a report in 2011–2013 and observation years 2011–2020, and hospital-years were retained when the hospital could be linked to a county (Table 1).
2.2. Outcome
Closure was defined as cessation of Medicare cost-report filing. A hospital was classified as closed in year Y if its final cost report, of any length, was for fiscal year Y (2021 or earlier), it filed no later full-year report under the same certification number, and no successor certification number existed. A successor was identified when the street address with ZIP code, or the name with city, appeared under a different certification number in the same or a later year, which removes mergers, acquisitions, relicensure, and renumbering. An observation in year t carries the outcome closure in year t+1. Cessation of filing does not establish that inpatient services permanently ended, and a hospital may close for reasons unrelated to financial distress.
2.3. Exposures
The exposures were county poverty (percentage of the population below the poverty level), county percentage Black (Black alone), and rural location. County characteristics are the 2015–2019 American Community Survey 5-year estimates [42], matched to hospitals by state and county name; 48 hospitals (1.0%) could not be matched and were excluded. County poverty and percentage Black were standardized (per standard deviation) and rural location is the cost-report rural indicator (versus urban). All three exposures are time-invariant and were entered together.
2.4. Mediators
Sixteen mediators were measured in the observation year, before the outcome, in eight groups. Size: beds (log). Utilization and staffing: occupancy (inpatient days divided by bed days available) and payroll full-time equivalents per bed. Performance: total margin (net income divided by total revenue) and operating margin (net patient revenue less operating expense, divided by net patient revenue). Assets: days cash on hand (cash and short-term investments divided by average daily operating expense), current ratio, and leverage (liabilities divided by assets). Payer mix: Medicare share of inpatient days and Medicaid share of charges. Uncompensated care: charity care and bad debt, each as a percentage of total costs (total uncompensated care is their sum and was not entered separately). Structure: critical access designation and teaching status (any intern or resident full-time equivalent). Ownership: for-profit and government ownership (versus nonprofit). Ratios were winsorized at the 1st and 99th percentiles, missing values were set to the sample median, continuous mediators were standardized, and binary mediators were coded 0/1.
2.5. Covariates
All models included year fixed effects, census region, state Medicaid expansion status in the observation year, and standardized log county population, which adjusts for the overlap of rurality with urbanicity of the county.
2.6. Statistical Analysis
Associations were estimated in a parallel multiple-mediator model [37,38,39,40]. For each exposure and mediator, the a-path is the coefficient of the exposure in a linear regression of the mediator on the three exposures, the covariates, and year fixed effects. The outcome model is a linear probability model of closure in the following year on the exposures, all 16 mediators, the covariates, and year fixed effects, and the coefficient of a mediator is its b-path. The indirect association of an exposure through a mediator is the product of its a-path and the b-path of the mediator, the direct association is the exposure coefficient with the mediators in the model, and the total association is the exposure coefficient without them. In linear models the total association equals the direct association plus the sum of the indirect associations exactly, which makes the decomposition additive and allows the mediators to be summed by group; this was verified in the data (largest discrepancy below 10⁻¹³ per 1,000). Estimates are risk differences in closures per 1,000 hospital-years for a one standard deviation increase in poverty or percentage Black, or for rural versus urban location, and for one standard deviation of a standardized mediator. The proportion mediated is the total indirect association divided by the total association and is unstable when the total association is small or has the opposite sign to the indirect association; it is reported but interpreted cautiously. Confidence intervals are 2.5th and 97.5th percentiles of 400 bootstrap draws that resample hospitals with all of their hospital-years.
Mediation estimates have a causal interpretation only if there is no unmeasured confounding of the exposure–outcome, mediator–outcome, and exposure–mediator relationships and no mediator–outcome confounder affected by the exposure [40]. These conditions are not guaranteed by the design, so the estimates are interpreted as a descriptive decomposition of associations. Because individual mediators are correlated, the indirect association of a single mediator is conditional on the others, and grouped estimates are emphasized.
Path diagrams (Figure 2 to 4) display, for each exposure, the direct association and the indirect associations of the mediators whose indirect association is individually significant, with the remaining mediators combined in one box. Paths whose 95% confidence interval includes zero are drawn as dotted lines.
2.7. Sensitivity Analyses
We repeated the model (a) with odds ratios from logistic outcome models for the total and direct associations and the mediator paths; (b) without the ownership mediators, because ownership could be upstream of community context; (c) without the county population covariate; and (d) for each exposure alone, to show the effect of estimating the three together. Sensitivity models used 100 bootstrap draws.
3. Results
3.1. Sample and Descriptive Characteristics
The analytic sample contained 45,107 hospital-years from 4,785 hospitals, of which 21,319 (47.3%) were rural, with 325 closures (Table 2). Closure occurred in 7.2 per 1,000 hospital-years, 5.5 in rural and 8.7 in urban hospitals (118 and 207 closures). The community exposures were correlated: poverty with percentage Black ρ = 0.30, poverty with rural location ρ = 0.16, and percentage Black with rural location ρ = -0.43.
Rural hospitals were located in counties with higher poverty (15.6% versus 13.5%) and a lower percentage Black (7.4% versus 13.5%). They were smaller (median 25.0 versus 166.0 beds), had lower occupancy (33.3% versus 57.0%) and lower margins (total 3.1% versus 5.0%; operating −4.9% versus −0.1%), a higher Medicare share, more bad debt but less charity care, and more days cash. They were more often critical access (56.0% versus 4.5%) and government-owned and less often for-profit or teaching hospitals.
3.2. Total, Direct, and Indirect Associations
Regarding Aim 1, county poverty had a total association with closure of 1.64 (0.72, 2.60) closures per 1,000 hospital-years per standard deviation (5.5 percentage points), equivalent to 22.8% of the overall rate (Table 3). The direct association was 0.68 (−0.34, 1.67) and the indirect association through all mediators was 0.96 (0.50, 1.48), so 59% (28%, 132%) of the total association was mediated; the upper bound above 100% reflects the width of the interval around a modest total association. County percentage Black had a total association of 0.74 (−0.27, 1.90), a direct association of 1.58 (0.52, 2.74), and an indirect association of −0.83 (−1.31, −0.37) in the opposite direction, a pattern in which the mediators suppress rather than carry the association. Rural hospitals had a lower closure rate than urban hospitals, but the total association was imprecise (−2.10 (−4.51, 0.20)); the direct association was −2.10 (−4.53, 0.04) and the net indirect association was −0.00 (−1.07, 1.00), which hides large offsetting indirect associations described below.
The logistic models gave the same pattern: the odds ratio for closure per standard deviation of county poverty was lower after the mediators were added, whereas that for percentage Black was higher, and rural location was not significantly associated with closure in either model. The estimates are therefore not specific to the linear probability model.
3.3. Paths from the Exposures to the Mediators and from the Mediators to Closure
Closure was less common in critical access hospitals (−10.51 (−14.18, −7.26) per 1,000), larger hospitals (−6.84 (−8.83, −4.89) per standard deviation of log beds), and hospitals with higher occupancy (−3.83 (−5.15, −2.61)), higher total margin (−4.43 (−6.18, −2.70)), and higher operating margin (−1.99 (−3.52, −0.41)), and more common in for-profit hospitals (6.71 (3.87, 9.29)) and those with higher leverage (1.74 (0.50, 3.11)) (Table 4). Days cash, current ratio, Medicare and Medicaid shares, charity care, and bad debt were not significantly associated with closure once the other mediators were included.
County poverty was associated with a lower probability of critical access status (−0.05 (−0.06, −0.03)), lower total margin (−0.05 (−0.08, −0.03) standard deviations per standard deviation), lower occupancy (−0.06 (−0.09, −0.03)), more bad debt (0.07 (0.04, 0.09)), more for-profit hospitals, and somewhat larger hospitals. County percentage Black was associated with higher occupancy (0.11 (0.08, 0.14)), larger hospitals, fewer for-profit and more government and teaching hospitals, higher leverage, and lower total margin. Rural location was associated with a higher probability of critical access status (0.24 (0.21, 0.27)), much smaller size (−0.34 (−0.39, −0.28)), lower occupancy and margins, and fewer for-profit and teaching hospitals.
3.4. Specific Mediators of each Association
Regarding Aim 2, the poverty association operated through four mediators that raised closure risk: lower probability of critical access status (0.48 (0.30, 0.70) per 1,000), lower total margin (0.24 (0.12, 0.40)), lower occupancy (0.23 (0.12, 0.37)), and more for-profit hospitals (0.16 (0.06, 0.27)); larger hospital size in poorer counties offset part of this (−0.22 (−0.42, −0.06)) (Table 5). Bad debt, although higher in poorer counties, was not a mediator (−0.05 (−0.13, 0.03)), and neither were days cash, payer mix, or charity care.
For county percentage Black the indirect associations were mostly negative: higher occupancy (−0.42 (−0.62, −0.26)), larger hospitals (−0.41 (−0.67, −0.23)), and fewer for-profit hospitals (−0.28 (−0.44, −0.14)) lowered closure risk and offset positive indirect associations through lower total margin (0.17 (0.05, 0.33)) and higher leverage (0.10 (0.02, 0.22)). For rural location, critical access status (−2.57 (−3.49, −1.75)) and fewer for-profit hospitals (−0.56 (−0.86, −0.30)) lowered the risk and were offset by smaller size (2.31 (1.59, 3.17)), lower occupancy (0.71 (0.42, 1.05)), and lower total (0.63 (0.32, 1.06)) and operating margins (0.32 (0.05, 0.67)).
3.5. Mediation by Group of Mediators
Grouped, the poverty association was carried by structure (0.51 (0.32, 0.74) per 1,000), performance (0.30 (0.15, 0.49)), utilization and staffing (0.24 (0.12, 0.39)), and ownership (0.15 (0.04, 0.26)), partly offset by size (−0.22 (−0.42, −0.06)); assets, payer mix, and uncompensated care did not mediate it (Table 6). The percentage Black association was suppressed by size (−0.41 (−0.67, −0.23)), utilization and staffing (−0.40 (−0.61, −0.21)), and ownership (−0.37 (−0.57, −0.18)), while performance (0.25 (0.09, 0.43)) and assets (0.10 (0.02, 0.23)) pointed the other way. For rural location, size (2.31 (1.59, 3.17)) and performance (0.95 (0.56, 1.45)) raised and structure (−2.77 (−3.72, −1.89)) and ownership (−0.68 (−1.02, −0.37)) lowered the risk.
Figure 1 shows the grouped indirect associations. For county poverty the confidence intervals for structure, performance, utilization and staffing, and ownership exclude zero and lie above it, whereas those for assets, payer mix, and uncompensated care include zero; for county percentage Black and rural location the indirect associations lie on both sides of zero.
3.6. Path Models for Each Exposure
The path models in this section show, for each exposure, how the total association with closure divides into a direct association, drawn as the arc over the top, and indirect associations through the mediators, drawn as the boxes in the middle. Each diagram draws only the mediators whose indirect association is individually significant and combines the others in a dotted box. A solid line means that the 95% bootstrap confidence interval of the path excludes zero, and a dotted gray line means that it includes zero. A red box is an indirect association that raises closure risk and a blue box is one that lowers it. The number on an arrow from the exposure is the change in the mediator per unit of exposure, the number on an arrow to closure is the change in closures per 1,000 hospital-years per standard deviation of the mediator, and the indirect association in the box is their product.
Figure 2 shows the path model for county poverty. The direct path is dotted (direct association 0.68 (−0.34, 1.67)), so the total association of 1.64 (0.72, 2.60) is carried mainly by the indirect paths. The pathways that raise closure risk are: critical access (a −0.05, b −10.51, indirect 0.48); total margin (a −0.05, b −4.43, indirect 0.24); occupancy (a −0.06, b −3.83, indirect 0.23); for-profit (a 0.02, b 6.71, indirect 0.16); operating margin (a −0.03, b −1.99, indirect 0.06). The pathway that lowers it is: beds (log) (a 0.03, b −6.84, indirect −0.22). In words, poorer counties are less likely to have critical access hospitals, which close much less often, so poverty is linked to more closure through fewer critical access hospitals; hospitals in poorer counties also have lower total and operating margins and lower occupancy, each of which is associated with closure, and are more often for-profit. Larger hospitals in poorer counties partly offset these pathways. The other 10 mediators, including days cash, payer mix, charity care, and bad debt, are combined in the dotted box and none is significant.
Figure 2.
Path model for county poverty. Solid lines: the 95% bootstrap confidence interval excludes zero; dotted gray lines: it includes zero. Red boxes: the indirect association raises closure risk; blue boxes: it lowers it. Numbers on the arrows from the exposure are changes in the mediator (standard deviations; probability for binary mediators) per unit of exposure, those on the arrows to closure are changes in closures per 1,000 hospital-years per standard deviation of the mediator, and each box gives their product (the indirect association). The arc over the top is the direct association.
Figure 2.
Path model for county poverty. Solid lines: the 95% bootstrap confidence interval excludes zero; dotted gray lines: it includes zero. Red boxes: the indirect association raises closure risk; blue boxes: it lowers it. Numbers on the arrows from the exposure are changes in the mediator (standard deviations; probability for binary mediators) per unit of exposure, those on the arrows to closure are changes in closures per 1,000 hospital-years per standard deviation of the mediator, and each box gives their product (the indirect association). The arc over the top is the direct association.

Figure 3 shows the path model for county percentage Black. The direct path is solid (direct association 1.58 (0.52, 2.74)), whereas the total association is not significant (0.74 (−0.27, 1.90)) because the indirect paths act in the opposite direction (total indirect −0.83 (−1.31, −0.37)). The pathways that lower closure risk are: for-profit (a −0.04, b 6.71, indirect −0.28); beds (log) (a 0.06, b −6.84, indirect −0.41); occupancy (a 0.11, b −3.83, indirect −0.42). Counties with a higher percentage Black have larger and busier hospitals, which close less often, and fewer for-profit hospitals, which close more often. The pathways that raise closure risk are: total margin (a −0.04, b −4.43, indirect 0.17); leverage (a 0.06, b 1.74, indirect 0.10); operating margin (a −0.04, b −1.99, indirect 0.08), because hospitals in these counties have lower margins and higher leverage. The direct path therefore represents the part of the association that is not explained by any of the measured mediators, and the lowering pathways hide it in the total association.
Figure 4 shows the path model for rural location. The direct path is dotted (direct association −2.10 (−4.53, 0.04)) and the total association is imprecise (−2.10 (−4.51, 0.20)), because large indirect paths in opposite directions cancel (total indirect −0.00 (−1.07, 1.00)). The pathways that raise closure risk are: beds (log) (a −0.34, b −6.84, indirect 2.31); occupancy (a −0.19, b −3.83, indirect 0.71); total margin (a −0.14, b −4.43, indirect 0.63); operating margin (a −0.16, b −1.99, indirect 0.32); rural hospitals are much smaller, less busy, and have lower total and operating margins. The pathways that lower it are: staff per bed (a 0.21, b −1.33, indirect −0.28); for-profit (a −0.08, b 6.71, indirect −0.56); critical access (a 0.24, b −10.51, indirect −2.57); rural hospitals are far more often critical access hospitals, which close much less often, are less often for-profit, and have more staff per bed. Because the raising and lowering pathways are of similar size, the net indirect association is close to zero even though each individual pathway is large.
3.7. Sensitivity Analyses
The pattern was robust (Table 7). Without the ownership mediators the indirect association for county poverty was 0.88 (54% of the total), and the direct associations for percentage Black (1.34) and rural location (−2.50) were similar to the main estimates. Without the county population covariate the total association of rural location was larger (−2.96) and the poverty association was more strongly mediated (76%). Estimating each exposure alone gave a total association for county poverty of 1.77 and a larger total association for percentage Black (1.48) than when the three were estimated together (0.74), showing that part of the unadjusted association of percentage Black with closure reflects poverty and rurality.
4. Discussion
4.1. Principal Findings
This study asked, regarding Aim 1, how county poverty, county percentage Black, and rurality are jointly associated with hospital closure and how much of each association runs through hospital characteristics, and, regarding Aim 2, which characteristics carry it. County poverty was associated with more closure, and about 59% of that association was indirect, running through critical access status, margins, occupancy, and for-profit ownership, with larger size in poorer counties offsetting part of it. County percentage Black had a direct association with closure that was masked by indirect associations in the opposite direction, chiefly through larger size, higher occupancy, and fewer for-profit hospitals. Rural location showed no net association because pathways of similar size in opposite directions cancelled: smaller size and lower margins raised closure risk, while critical access designation and fewer for-profit hospitals lowered it.
4.2. Interpretation
The pattern for poverty is consistent with hospitals in poorer communities operating on thinner margins and lower occupancy, which are associated with closure [3,22,26], and with the protective role of critical access designation [23,24] that poorer counties are less likely to have. The finding that bad debt is higher in poorer counties but does not carry closure risk, and that charity care and days cash do not either, is informative: uncompensated care differs by community [14,32,33], but within this sample it did not account for closure once margins and utilization were considered, which suggests that the financial consequences of uncompensated care matter through margins and occupancy and not as a separate pathway. The contribution of for-profit ownership to the poverty association is modest but consistent with ownership composition differing by community.
The negative indirect associations for county percentage Black show that counties with higher percentage Black have hospitals that are, on measured characteristics, less exposed to closure: larger, busier, and less often for-profit. This is an example of suppression, in which an indirect association masks a direct one [35]. The direct association that remains is not explained by any of the 16 mediators and may reflect mechanisms not measured here, such as the market and payer environment of the local area, segregation of services, or the financial position of the owning system. The estimate should not be read as an effect of race, but as a description of how the racial composition of the county is linked to hospital closure through and beyond hospital characteristics, in the tradition of research on residential segregation and hospital survival [16,17,18].
The null net association for rural location is a case of inconsistent mediation: smaller size and lower margins raise closure risk, whereas critical access designation, which was created for small rural hospitals, and the lower presence of for-profit hospitals lower it. That the net association is near zero should not be read as rural hospitals being unaffected; it shows that policy design, in this case cost-based reimbursement, offsets the vulnerabilities that rurality brings [23,24].
4.3. Implications
Monitoring of hospitals at risk of closure in poor communities should focus on margins and occupancy, which carried most of the modifiable indirect association, and not on uncompensated care or cash reserves. Support for hospitals in poorer and rural communities that targets utilization and margins, and extension of the protections that critical access designation provides to similar hospitals that do not qualify, are consistent with these pathways. Because the associations of racial composition and rurality involve opposing pathways, any single indicator that summarizes community risk would miss part of the pattern, and monitoring should examine the combination of community context and hospital characteristics.
4.4. Limitations
The study is observational and the mediation estimates depend on assumptions that cannot be verified, including no unmeasured confounding of the mediator–closure relationship; the mediators were measured in the year before the outcome, so reverse causation from impending closure to margins and occupancy is possible. The mediators are correlated and measured with error, so individual indirect associations are conditional on the other mediators and should be read together with the grouped estimates. The exposures are measured at the county level from 2015–2019 estimates and do not describe the patients of a hospital. The proportion mediated is unstable when the total association is small, and the confidence interval for poverty exceeds 100%. Closure is inferred from cessation of cost-report filing and may include conversions and exclude services that closed within continuing hospitals. Cost-report measures such as days cash are facility-level and may reflect system accounting. The linear probability model is a close approximation for a rare outcome and gave the same pattern as logistic models, but absolute risk differences should not be extrapolated beyond the observed range. Hospitals that could not be linked to a county were excluded (1.0%).
5. Conclusions
Regarding Aim 1, county poverty was associated with hospital closure, mostly indirectly; county percentage Black had a direct association that was masked by opposing indirect associations; and rural location showed offsetting pathways with no net association. Regarding Aim 2, the poverty association ran through critical access status, margins, occupancy, and for-profit ownership, and not through uncompensated care, payer mix, or cash reserves. Monitoring and support for hospitals in communities at risk should therefore address margins and utilization and should examine community context and hospital characteristics together.
Author Contributions
Conceptualization, S.A.; methodology, S.A.; software, S.A.; formal analysis, S.A.; investigation, S.A.; data curation, S.A.; writing—original draft preparation, S.A.; writing—review and editing, S.A.; visualization, S.A. The author has read and agreed to the published version of the manuscript.
Funding
Shervin Assari is partially supported by NIH grants to Charles R. Drew University of Medicine and Science: U54MD007598, S21MD018960, and R25MD007610. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.
Institutional Review Board Statement
Ethical review and approval were waived for this study because it analyses de-identified, publicly available administrative and vital statistics data and does not constitute human subjects research as defined at 45 CFR 46.102.
Informed Consent Statement
Not applicable.
Data Availability Statement
Hospital cost report data are publicly available from the Centers for Medicare & Medicaid Services (https://www.cms.gov/data-research) and county data from the American Community Survey. The list of included hospitals (Appendix A) and the complete analysis code (Appendix B), which regenerates every number, table, and figure from the raw files, accompany this submission.
Conflicts of Interest
The author declares no conflicts of interest.
Appendix A. Hospitals Included in the Analyses
Hospitals are identified by their six-digit CMS certification number (provider number). 4,785 hospitals in the 50 states and the District of Columbia (3,455 short-term acute care and 1,330 critical access; 2,519 rural) entered the closure analysis after the eligibility steps in Table 1. The full list, with hospital name, type, ownership, county, and the exposure values, is provided as Supplementary File S1 (hospitals_included_mediation.csv). A * marks a hospital with a closure event (325).
- AK (n = 21): 020001, 020006, 020008, 020012, 020017, 020018, 020024, 020026, 020027, 021301, 021302, 021303*, 021304, 021305, 021306, 021307, 021308, 021309, 021310, 021312, 021313
- AL (n = 97): 010001, 010005, 010006, 010007, 010008, 010009, 010010, 010011, 010012, 010016, 010018, 010019, 010021, 010022, 010023, 010024, 010025, 010027*, 010029, 010032, 010033, 010034, 010035, 010036, 010038*, 010039, 010040, 010043, 010044, 010045, 010046, 010047*, 010049, 010050, 010051, 010052, 010054*, 010055, 010056, 010058, 010059, 010061, 010062, 010065, 010066*, 010069, 010073, 010078, 010079, 010083, 010085, 010086, 010087, 010089, 010090, 010091, 010092, 010095, 010097, 010099, 010100, 010101, 010102, 010103, 010104, 010108, 010109*, 010110, 010112, 010113, 010114, 010118, 010120, 010125, 010126, 010128, 010129, 010130, 010131, 010137*, 010138, 010139, 010144, 010145, 010146*, 010148, 010149, 010150, 010152*, 010157, 010158, 010164, 010168, 010169, 011300, 011302, 011304
- AR (n = 75): 040001, 040002, 040004, 040007, 040010, 040011, 040014, 040015, 040016, 040017, 040018, 040019, 040020, 040022, 040026, 040027, 040029, 040036, 040039, 040041, 040042*, 040047, 040050, 040051, 040055, 040062, 040067, 040069, 040071, 040072, 040074, 040076, 040078, 040080*, 040084, 040085, 040088, 040091, 040114, 040118, 040119, 040132, 040134, 040137, 040147, 040152*, 041300, 041302, 041303, 041304, 041305, 041306, 041307, 041308, 041309, 041310, 041311, 041312, 041313, 041314, 041316, 041317, 041318, 041319*, 041320, 041321, 041322, 041323, 041324, 041326, 041327, 041328, 041329, 041330, 041331
- AZ (n = 80): 030001*, 030002, 030006, 030007, 030010, 030011, 030012, 030013, 030014, 030016, 030022, 030023, 030024, 030030, 030033, 030036, 030037, 030038, 030043, 030055, 030061, 030062, 030064, 030065, 030067, 030068, 030069, 030071, 030073, 030074, 030077, 030078, 030083, 030084, 030085, 030087, 030088, 030089, 030092, 030093, 030094, 030100*, 030101, 030103, 030105, 030107, 030108, 030110, 030111, 030112, 030113, 030114, 030115, 030117, 030118*, 030119, 030120, 030121, 030122, 030123, 030126*, 030128, 030129, 030130, 030131, 030132*, 031300, 031301, 031302, 031303*, 031304, 031307, 031308, 031309, 031311, 031312, 031313, 031314, 031315, 031316
- CA (n = 342): 050002, 050006, 050007, 050008, 050009, 050013, 050014, 050016*, 050017, 050018*, 050022, 050024, 050025, 050026, 050028, 050030, 050036, 050038, 050039, 050040, 050042, 050043, 050045, 050046, 050047, 050054, 050055, 050056, 050057, 050058, 050060, 050063, 050067, 050070, 050071, 050072, 050073, 050076, 050077, 050078, 050079*, 050082, 050084, 050089, 050090, 050091, 050093, 050096, 050099, 050100, 050101, 050102, 050103, 050104, 050107, 050108, 050110, 050111*, 050112, 050113, 050115, 050116, 050118, 050121, 050122, 050124, 050125, 050126, 050127, 050128, 050129, 050131, 050132, 050133, 050135, 050136, 050137, 050138, 050139, 050140, 050145, 050146, 050149, 050150, 050152, 050153*, 050158, 050159, 050167, 050168, 050169, 050174, 050179, 050180, 050189, 050191, 050192, 050194, 050195, 050196*, 050197, 050204, 050205, 050211, 050219, 050222, 050224, 050225*, 050226, 050228, 050230, 050231, 050232, 050234, 050235, 050236, 050238, 050239, 050242, 050243, 050245, 050248, 050254, 050257, 050261, 050262, 050272, 050276, 050277, 050278, 050279, 050280, 050281, 050283, 050289, 050290, 050291, 050292, 050295, 050296, 050298, 050300, 050301, 050305, 050308, 050309, 050313, 050315, 050320, 050324, 050327, 050329, 050334, 050335, 050336, 050342, 050348, 050349*, 050350, 050351, 050352, 050353, 050357, 050359, 050360, 050367, 050373, 050376, 050378, 050380, 050382, 050385*, 050390, 050393, 050394, 050396, 050397, 050407, 050411, 050414, 050417, 050423, 050424, 050425, 050426, 050434*, 050435, 050438, 050441, 050444, 050448, 050454, 050455, 050457, 050464, 050468, 050471, 050481, 050485, 050488, 050492, 050496, 050498, 050502*, 050503, 050506, 050510, 050512, 050515, 050516, 050517, 050523, 050526, 050528, 050531*, 050534, 050537, 050541, 050543, 050545*, 050546, 050547*, 050548*, 050549, 050551, 050552, 050557, 050561, 050567, 050568*, 050570, 050573, 050575, 050580, 050581, 050586, 050588, 050589, 050590, 050597, 050599, 050603, 050604, 050608, 050609, 050616*, 050618, 050624, 050625, 050633, 050636, 050641, 050644*, 050660, 050663, 050668, 050674, 050677, 050678, 050684, 050686, 050688*, 050689, 050690, 050696, 050697, 050701, 050704, 050708, 050709, 050710, 050714, 050717, 050723, 050724, 050726, 050727, 050732*, 050735, 050736, 050737, 050738, 050739, 050740, 050742*, 050744, 050745, 050746, 050747, 050748, 050749*, 050751*, 050752*, 050754*, 050755, 050757, 050758, 050760, 050761, 050763, 050764, 050765, 050766, 050767, 050768*, 050769, 050770, 050771, 050772, 050773*, 051300, 051301, 051302, 051303, 051304, 051305, 051306, 051307, 051308, 051309, 051310, 051311, 051312, 051314, 051315, 051316, 051317, 051318, 051319, 051320, 051321, 051323, 051324, 051325, 051326, 051327, 051328, 051329, 051330, 051331, 051332, 051333
- CO (n = 75): 060001, 060003, 060004, 060006, 060008, 060009, 060010, 060011, 060012, 060014, 060015, 060016, 060020, 060022, 060023, 060024, 060027, 060028, 060030, 060031, 060032, 060034, 060036, 060043, 060044, 060049, 060054, 060064, 060065, 060071, 060075, 060076, 060096, 060100, 060103, 060104, 060107, 060112, 060113, 060114, 060116, 060117, 060118, 060119, 060124, 060125, 061300, 061301, 061302, 061303, 061304, 061305, 061306, 061307, 061308, 061309, 061310, 061311, 061312, 061313, 061314, 061315, 061316, 061317, 061318, 061319, 061320, 061321, 061322, 061323, 061324, 061325, 061326, 061327, 061328
- CT (n = 31): 070001*, 070002, 070003, 070004, 070005, 070006, 070007, 070008, 070010, 070011, 070012, 070016, 070017, 070018, 070019*, 070020, 070021, 070022, 070024, 070025, 070027, 070028, 070029, 070031, 070033, 070034, 070035, 070036, 070038, 070039, 070040
- DC (n = 7): 090001, 090003, 090004, 090005, 090006*, 090008, 090011
- DE (n = 6): 080001, 080003, 080004, 080006, 080007, 080009
- FL (n = 186): 100001, 100002, 100006, 100007, 100008, 100009, 100012, 100014, 100017, 100018, 100019, 100022, 100023, 100025, 100026, 100028, 100029, 100030, 100032, 100034, 100035, 100038, 100039, 100040, 100043, 100044, 100045, 100046, 100047, 100048, 100049, 100050, 100051, 100052, 100053, 100054, 100055, 100057, 100062, 100063, 100067, 100068, 100069, 100070, 100071, 100072, 100073, 100075, 100076*, 100077, 100079, 100080, 100081*, 100084, 100086, 100087, 100088, 100090, 100092, 100093, 100099, 100102*, 100105, 100106, 100107, 100109, 100110, 100113, 100117, 100118, 100121, 100122, 100124, 100125, 100126, 100127, 100128, 100130, 100131, 100132, 100134, 100135, 100137, 100139*, 100140, 100142, 100150, 100151, 100154, 100156, 100157, 100161, 100166, 100167, 100168, 100173, 100175, 100176, 100177, 100179, 100180, 100181, 100183, 100187, 100189, 100191, 100200, 100204, 100206, 100209, 100211, 100212, 100213, 100217, 100220, 100223, 100224, 100226, 100228, 100230, 100231, 100232, 100234*, 100236, 100238, 100239*, 100240*, 100242, 100243, 100244, 100246, 100248, 100249, 100252, 100253, 100254, 100255*, 100256, 100258, 100259, 100260, 100264, 100265, 100266, 100267, 100268, 100269, 100271, 100275, 100276, 100277, 100281, 100284, 100285, 100286, 100287, 100288, 100289, 100290, 100291, 100292, 100296, 100298, 100299, 100302, 100307, 100313, 100314, 100315, 100316, 100317, 100320, 100321, 101300, 101301*, 101302*, 101303, 101304, 101305, 101307, 101308, 101309, 101310*, 101311, 101312, 101313
- GA (n = 143): 110001, 110002, 110003, 110004, 110005, 110006, 110007, 110008, 110010, 110011, 110015, 110016, 110018, 110023, 110024, 110025, 110026, 110027, 110028, 110029, 110030, 110031, 110032, 110034, 110035, 110036, 110038, 110039*, 110040*, 110041, 110042, 110043, 110044, 110045, 110046, 110050, 110051, 110054, 110059*, 110064, 110069, 110071, 110073, 110074, 110075, 110076, 110078, 110079, 110082, 110083, 110086, 110087, 110089, 110091, 110092, 110095, 110100, 110101, 110104, 110105, 110107, 110109, 110111, 110112, 110113, 110115, 110121, 110122, 110124, 110125, 110128, 110129, 110130, 110132, 110135, 110142, 110143, 110146, 110150, 110153, 110161, 110163*, 110164, 110165, 110168, 110177, 110183*, 110184, 110186*, 110187, 110189, 110190, 110191, 110192, 110194, 110198, 110200, 110201, 110203, 110205*, 110209, 110212, 110215, 110219*, 110225, 110226, 110229, 110230, 110233, 111300*, 111302, 111303, 111304, 111305, 111306, 111308, 111309*, 111310, 111311, 111312, 111313, 111314, 111315*, 111316, 111318, 111319, 111320, 111321*, 111322*, 111323, 111324, 111325, 111326, 111327, 111328, 111329, 111330, 111331, 111332, 111333, 111334, 111335, 111336
- HI (n = 17): 120002, 120004, 120006, 120007, 120011, 120014, 120022, 120026, 120028, 121300, 121301, 121303, 121304, 121305, 121306, 121307, 121308
- IA (n = 116): 160001, 160005, 160008, 160013, 160016, 160024*, 160028, 160029, 160030, 160032, 160033, 160040, 160045, 160047, 160057, 160058, 160064, 160067, 160069, 160079, 160080, 160082, 160083, 160089, 160101, 160104, 160110, 160112, 160117, 160122*, 160124, 160146, 160147, 160153, 161300, 161301, 161302, 161303, 161304, 161305, 161306, 161307, 161308, 161309, 161310, 161311, 161312, 161313, 161314, 161315, 161316, 161317, 161318, 161319, 161320, 161321, 161322, 161323, 161324, 161325, 161326, 161327, 161328, 161329, 161330, 161331, 161332, 161333, 161334, 161335, 161336, 161337, 161338, 161339, 161340, 161341, 161342, 161343, 161344, 161345, 161346, 161347, 161348, 161349, 161350, 161351, 161352, 161353, 161354, 161355, 161356, 161357, 161358, 161359, 161360, 161361, 161362, 161363, 161364, 161365, 161366, 161367, 161368, 161369, 161370, 161371, 161372, 161373, 161374, 161375, 161376, 161377, 161378, 161379, 161380, 161381
- ID (n = 41): 130002, 130003, 130006, 130007, 130013, 130014, 130018, 130025, 130028, 130049, 130063, 130065, 130066, 130067, 131301, 131302, 131303, 131304, 131305, 131307, 131308, 131309, 131310, 131311, 131312, 131313, 131314, 131315, 131316, 131317, 131318, 131319, 131320, 131321, 131322, 131323, 131324, 131325, 131326, 131327, 131328
- IL (n = 180): 140001, 140002, 140007, 140008, 140010, 140011, 140012, 140013, 140015, 140018, 140019, 140026*, 140029, 140030, 140032, 140033, 140034, 140040*, 140043, 140046, 140048, 140049, 140051*, 140052, 140053, 140054, 140058, 140059, 140062, 140063, 140064, 140065, 140067, 140068, 140077, 140080, 140082, 140083, 140084, 140088, 140089, 140091, 140093, 140095, 140100, 140101, 140103, 140110, 140113, 140114, 140115, 140116, 140117, 140118*, 140119, 140120, 140122, 140124, 140125, 140127, 140130, 140133, 140135, 140137, 140143, 140145, 140147, 140148, 140150, 140151, 140155, 140158, 140160, 140161, 140162, 140164, 140166, 140167, 140172, 140174, 140176*, 140177, 140179, 140180, 140181, 140182, 140184, 140185, 140187, 140189, 140191, 140197, 140200, 140202, 140206, 140208, 140209, 140210, 140211, 140213, 140217, 140223, 140224, 140228, 140231, 140233, 140234, 140239, 140240*, 140242, 140250*, 140251, 140252, 140258, 140275, 140276, 140280, 140281, 140286, 140288, 140289, 140290, 140291, 140292, 140294, 140300, 140301, 140304, 140306, 141300, 141301, 141302, 141303, 141304, 141305, 141306, 141307, 141308, 141309, 141310, 141311, 141312, 141313, 141315, 141316, 141317, 141318, 141319, 141320, 141321, 141322, 141323, 141324, 141325, 141326, 141327, 141328, 141329, 141330, 141331, 141332, 141333, 141334, 141335, 141336, 141337, 141338, 141339, 141340, 141341, 141342, 141343, 141344, 141345, 141346, 141347, 141348, 141349, 141350, 141351
- IN (n = 124): 150001, 150002, 150003, 150004, 150005, 150006, 150007, 150008, 150009, 150010, 150011, 150012, 150015, 150017, 150018, 150021, 150022, 150023, 150024, 150026, 150030, 150033, 150034, 150035, 150037, 150038*, 150042, 150044, 150045, 150046, 150047, 150048, 150051, 150056, 150057, 150058, 150059, 150061, 150064*, 150065, 150069, 150072, 150074, 150075, 150076, 150082, 150084, 150086, 150088, 150089, 150090, 150091, 150097, 150100, 150101, 150102, 150104, 150109, 150112, 150113, 150115, 150125, 150128, 150129*, 150133, 150146, 150149, 150150, 150153, 150154*, 150157, 150158, 150160, 150161, 150162, 150163*, 150164, 150165, 150166, 150167, 150168, 150169, 150172, 150173, 150175*, 150176*, 150177, 150179, 150182*, 151300, 151301, 151302, 151303, 151304, 151305, 151306, 151307, 151308, 151309, 151310, 151311, 151312, 151313, 151314, 151315, 151316, 151317, 151318, 151319, 151320, 151322, 151323, 151324, 151325, 151326, 151327, 151328, 151329, 151330, 151331, 151332, 151333, 151334, 151335
- KS (n = 139): 170001, 170006, 170009, 170010*, 170012, 170013, 170014, 170016, 170017, 170020, 170023, 170027, 170039, 170040, 170049, 170058*, 170068, 170074, 170075, 170086, 170094, 170103, 170104, 170105, 170109, 170110, 170120, 170122, 170123, 170133*, 170137, 170142, 170145, 170146, 170150, 170166, 170175, 170176, 170180, 170182, 170183, 170185, 170186, 170187, 170188, 170190, 170191, 170194, 170195*, 170196, 170197, 170198, 170199, 170200, 170201*, 170203, 171300, 171301, 171302*, 171303, 171304, 171306, 171307, 171308, 171309, 171310, 171311, 171312, 171313, 171314, 171315, 171316, 171317, 171318, 171319, 171320*, 171321, 171322, 171323, 171324, 171325, 171326, 171327, 171328, 171329, 171330, 171331, 171332, 171333, 171334, 171335, 171336, 171337, 171338, 171339, 171340*, 171341, 171342, 171343, 171344, 171345, 171346, 171347, 171348, 171349, 171350, 171351, 171352, 171353, 171354, 171355, 171356, 171357, 171358, 171359, 171360, 171361, 171362, 171363, 171364, 171365, 171366*, 171367, 171368, 171369, 171370, 171371, 171372, 171373, 171374, 171375, 171376, 171377, 171378, 171379, 171380, 171381, 171382, 171383
- KY (n = 94): 180001, 180002, 180004, 180005, 180009, 180010, 180011, 180012, 180013, 180016, 180017, 180018, 180019, 180020, 180021, 180024, 180025, 180027, 180029, 180035, 180036*, 180038, 180040, 180043, 180044, 180045, 180046, 180048, 180049, 180050, 180051, 180053, 180056, 180064, 180066, 180067, 180069, 180070, 180078, 180079, 180080, 180087, 180088, 180092, 180093, 180095, 180101, 180102, 180103, 180104, 180105, 180106, 180115, 180116, 180117*, 180124, 180127, 180128, 180130, 180132, 180138, 180139, 180141, 180143, 180149, 181301, 181302, 181303*, 181304, 181305, 181306, 181307, 181308, 181309, 181310, 181311, 181312, 181314, 181315, 181316, 181317, 181318, 181319, 181320, 181321, 181322, 181323, 181324, 181325, 181327, 181328, 181329, 181330, 181331
- LA (n = 126): 190001, 190002, 190003*, 190004, 190005, 190006, 190007, 190008, 190009*, 190011, 190013, 190014, 190015, 190017, 190019, 190020, 190025, 190026, 190027, 190034, 190036, 190039, 190040, 190041, 190044, 190045, 190046, 190050, 190053, 190054, 190060, 190065, 190079, 190081, 190086, 190088, 190090, 190098, 190099, 190102, 190106, 190111, 190114, 190116, 190118, 190122*, 190125, 190128, 190133, 190135*, 190140, 190144, 190145, 190146, 190151, 190160, 190161*, 190164, 190167, 190175, 190176, 190177*, 190183, 190184, 190190, 190199*, 190201, 190202, 190204, 190205*, 190208, 190218, 190236*, 190241, 190242*, 190245, 190246*, 190250, 190251, 190255, 190256, 190257*, 190259, 190261*, 190263*, 190266, 190267, 190268, 190270, 190273*, 190274, 190278, 190297*, 190298, 190300, 190303, 190304, 190307, 190308, 191300, 191301, 191302, 191303, 191304, 191305, 191306, 191307, 191308, 191309, 191310, 191311, 191312, 191313, 191314, 191315, 191316, 191317, 191318, 191319, 191320, 191321, 191322, 191323, 191324, 191325, 191326
- MA (n = 65): 220001, 220002, 220008, 220010, 220011, 220012, 220015, 220016, 220017, 220019, 220020, 220024, 220031, 220033, 220035, 220036, 220046, 220049, 220050*, 220051*, 220052, 220058*, 220060, 220062, 220063, 220065, 220066, 220067*, 220070, 220071, 220073, 220074, 220075, 220077, 220080, 220082*, 220083, 220084, 220086, 220088, 220090, 220095, 220098, 220100, 220101, 220105, 220108, 220110, 220111, 220116, 220119, 220126, 220135, 220153, 220154*, 220162, 220163, 220171, 220174*, 220175, 220176, 220177, 221300, 221302, 221303
- MD (n = 48): 210001, 210002, 210003, 210004, 210005, 210006, 210007, 210008, 210009, 210011, 210012, 210013*, 210015, 210016, 210017, 210018, 210019, 210022, 210023, 210024, 210027, 210028, 210029, 210030, 210032, 210033, 210034, 210035, 210037, 210038, 210039, 210040, 210043, 210044, 210045*, 210048, 210049, 210051, 210054, 210055, 210056, 210057, 210058, 210060, 210061, 210062, 210063, 210064
- ME (n = 36): 200001, 200002*, 200008, 200009, 200018, 200019, 200020, 200021, 200024, 200025*, 200031, 200033, 200034, 200037, 200039, 200040, 200041, 200050, 200052, 200063, 201300, 201301, 201302, 201303, 201304, 201305, 201306, 201307, 201308, 201309, 201310, 201311, 201312, 201313, 201314, 201315
- MI (n = 133): 230002, 230003, 230004*, 230005, 230013, 230015, 230017, 230019, 230020, 230021, 230022, 230024, 230029, 230030, 230031, 230035, 230036, 230037, 230038, 230040, 230041, 230046, 230047, 230053, 230054, 230055, 230058, 230059, 230066, 230069, 230070, 230071, 230072, 230075, 230077, 230078, 230080, 230081, 230085, 230089, 230092, 230093, 230095, 230096, 230097, 230099, 230100, 230101, 230104, 230105, 230106, 230108, 230110, 230117, 230118, 230121, 230130, 230132, 230133, 230141, 230142, 230144, 230146, 230151, 230156, 230165, 230167, 230174, 230176, 230180, 230193, 230195, 230197, 230207, 230208, 230212, 230216, 230217, 230222, 230227, 230230, 230236, 230239, 230241, 230244, 230254, 230259, 230264, 230269, 230270, 230273, 230275, 230277, 230279, 230297, 230301, 230302, 231300, 231301, 231303, 231304, 231305, 231306, 231307, 231308, 231309, 231310, 231311, 231312, 231313, 231314, 231315, 231316, 231317, 231318, 231319, 231320, 231321, 231322, 231323, 231324, 231325, 231326, 231327, 231328, 231329, 231330, 231331, 231332, 231333, 231334*, 231335, 231337
- MN (n = 131): 240001, 240002, 240004, 240006, 240010, 240014, 240018, 240019, 240020, 240022, 240030, 240036, 240038, 240040, 240043, 240044, 240047, 240050, 240052, 240053, 240056, 240057, 240059, 240061*, 240063, 240064, 240066, 240069, 240071, 240075, 240076, 240078, 240080, 240084, 240088, 240093, 240100, 240101, 240104, 240106, 240115, 240117, 240132*, 240141, 240166, 240187, 240196*, 240206, 240207, 240210, 240213, 240214, 241300, 241301, 241302, 241303, 241304, 241305, 241306, 241308, 241309, 241311, 241312, 241313, 241314, 241315, 241316, 241317, 241318, 241319, 241320, 241321, 241322, 241323, 241325, 241326, 241327, 241328, 241329, 241330, 241331*, 241332, 241333, 241334, 241335, 241336, 241337, 241338, 241339, 241340, 241341, 241342, 241343, 241344, 241345, 241346, 241347, 241348, 241349, 241350, 241351, 241352*, 241353, 241354, 241355, 241356, 241357, 241358, 241359, 241360, 241361, 241362, 241363, 241364, 241365, 241366, 241367, 241368, 241369, 241370, 241371, 241372, 241373, 241374, 241375, 241376, 241377, 241378, 241379, 241380, 241381
- MO (n = 117): 260001, 260004, 260005, 260006, 260009, 260011, 260015*, 260017, 260020, 260022, 260023, 260024, 260025, 260027, 260032, 260034, 260040, 260047, 260048, 260050, 260052, 260057, 260059, 260061, 260062, 260064*, 260065, 260068, 260070, 260074, 260077, 260078, 260080*, 260081, 260085, 260091, 260094, 260095, 260096, 260097, 260102, 260104, 260105, 260108, 260110, 260113, 260115, 260116, 260119, 260137, 260138, 260141, 260142, 260147*, 260160, 260162, 260163, 260175, 260176, 260177, 260179, 260180, 260183, 260186, 260190, 260191, 260193, 260195, 260200, 260207*, 260209, 260210, 260211*, 260214, 260216, 260219, 260227*, 260228, 260229, 260230, 261301, 261302, 261303, 261304*, 261305, 261306, 261307, 261308, 261309, 261310, 261311, 261312, 261313, 261314, 261315, 261316, 261317, 261318, 261319, 261320, 261321, 261322, 261323, 261324, 261325, 261326, 261327, 261328, 261329, 261330, 261331, 261332, 261333, 261334*, 261335, 261336, 261337
- MS (n = 99): 250001, 250002, 250004, 250006, 250007, 250009, 250010, 250012, 250015, 250017, 250018, 250019, 250020, 250025, 250027, 250031, 250034, 250036, 250038, 250040, 250042, 250043, 250044, 250048, 250049, 250050, 250051*, 250057, 250058, 250059, 250060, 250061, 250067, 250069, 250072, 250077, 250078, 250079, 250081, 250082, 250084, 250085, 250093, 250094, 250095, 250096, 250097, 250099, 250100, 250102, 250104, 250117, 250122, 250123, 250124, 250126, 250127, 250128, 250134, 250136, 250138, 250141, 250151, 250152, 250162, 250163, 250166, 251300, 251302, 251303*, 251304, 251305, 251306, 251307, 251308, 251309, 251311*, 251312, 251313, 251314, 251315, 251316, 251317, 251318, 251319, 251320, 251322, 251323, 251324, 251325, 251326, 251327, 251329, 251330, 251331, 251332*, 251333, 251334, 251335
- MT (n = 61): 270003, 270004, 270012, 270014, 270017, 270023, 270032, 270049, 270051, 270057, 270074, 270086, 270087*, 271300, 271301, 271302, 271303, 271304, 271305, 271306, 271307, 271308, 271309, 271310, 271311, 271312, 271313, 271314, 271315, 271316, 271317, 271318, 271319, 271320, 271321, 271322, 271323, 271324, 271325, 271326, 271327, 271328, 271329, 271330, 271331, 271332, 271333, 271334, 271335, 271336, 271337, 271338, 271339, 271340, 271341, 271342, 271343, 271344, 271345, 271346, 271347
- NC (n = 112): 340001, 340002, 340003, 340004, 340008, 340010, 340011, 340013, 340014, 340015, 340016, 340017, 340020, 340021, 340023, 340024, 340027, 340028, 340030, 340032, 340036*, 340037*, 340038, 340039, 340040, 340041, 340042, 340047, 340049, 340050, 340051, 340053, 340055*, 340060, 340061, 340064, 340068, 340069, 340070, 340071, 340073, 340075, 340084, 340085, 340087, 340090, 340091, 340096, 340097, 340098, 340099, 340106*, 340107, 340109, 340113, 340114, 340115, 340116, 340119, 340120, 340123, 340126, 340127, 340129, 340130, 340131, 340132, 340133, 340141, 340142, 340143, 340144, 340145, 340147, 340148, 340151, 340153*, 340155, 340156, 340158, 340159, 340160, 340166, 340168, 340171, 340173, 340183, 340184, 340186*, 341302*, 341303, 341304, 341305, 341307, 341308*, 341310*, 341311, 341313*, 341314, 341315, 341316, 341317, 341318, 341319, 341320, 341321*, 341322, 341323, 341324, 341325, 341326, 341327
- ND (n = 44): 350002, 350006, 350011, 350015, 350019, 350063, 350064, 350070, 351300, 351301, 351302, 351303, 351304, 351305, 351306, 351307, 351308, 351309, 351310, 351311, 351312, 351313, 351314, 351315, 351316, 351318, 351319, 351320, 351321, 351322, 351323, 351324, 351325, 351326, 351327, 351328, 351329, 351330, 351331, 351332, 351333, 351334, 351335, 351336
- NE (n = 88): 280003, 280009, 280013, 280020, 280023, 280030*, 280032, 280040, 280060, 280061, 280065, 280077, 280081, 280105, 280111, 280119*, 280125, 280127, 280128, 280129, 280130, 280131, 280132, 280133, 281300, 281301, 281302, 281303, 281304, 281305, 281306, 281307, 281308, 281309, 281310, 281311, 281312, 281313, 281314, 281315, 281316, 281317*, 281318, 281319, 281320, 281321*, 281322, 281323, 281324, 281325, 281326, 281327, 281328, 281329, 281331, 281332, 281333, 281334, 281335, 281336, 281338, 281339, 281340, 281341, 281342, 281343, 281344, 281345, 281346, 281347, 281348, 281349, 281350, 281351, 281352, 281353, 281354, 281355, 281356, 281357, 281358, 281359, 281360, 281361, 281362, 281363, 281364, 281365
- NH (n = 25): 300001, 300003, 300005, 300011, 300012, 300014, 300017, 300018, 300019, 300020, 300023, 300029, 300034, 301300, 301301, 301302, 301303, 301304, 301306, 301307, 301308, 301309, 301310, 301311, 301312
- NJ (n = 63): 310001, 310002, 310003, 310005, 310006, 310008, 310009, 310010, 310011, 310012, 310015, 310016, 310017, 310019, 310021, 310022, 310024, 310025, 310027, 310028, 310029, 310031, 310032, 310034, 310038, 310039, 310040, 310041, 310044, 310045, 310047, 310048, 310050, 310051, 310052, 310054, 310057, 310058, 310060, 310061, 310064, 310069, 310070, 310073, 310074, 310075, 310076, 310081*, 310083, 310084, 310086, 310091, 310092, 310096, 310108, 310110, 310111, 310112, 310113, 310115, 310118, 310119, 310120*
- NM (n = 43): 320001, 320002, 320003, 320004, 320005, 320006, 320009, 320011, 320013, 320014, 320016, 320017, 320018, 320021, 320022, 320030, 320033, 320038, 320057, 320058, 320059, 320060, 320061, 320062, 320063, 320065, 320067, 320070, 320074, 320083, 320084, 320085, 320086, 320089, 321300, 321301, 321302, 321304, 321305, 321306, 321307, 321308, 321309
- NV (n = 30): 290001, 290002, 290003, 290005, 290006, 290007, 290008, 290009, 290012, 290019, 290020*, 290032, 290039, 290042, 290045, 290047, 290049, 290051*, 290053, 291300, 291301, 291302, 291303, 291304, 291306, 291307, 291308, 291309, 291311, 291312
- NY (n = 177): 330002*, 330003*, 330004*, 330005, 330006, 330008, 330009, 330011, 330013, 330014, 330019, 330023, 330027, 330028, 330030, 330033, 330037*, 330043, 330044, 330045, 330046, 330047, 330049, 330053, 330055, 330056, 330058, 330059, 330061*, 330064*, 330065, 330067*, 330073, 330074, 330078, 330079, 330080, 330084, 330085, 330086, 330088*, 330090, 330094, 330096, 330100, 330101, 330102, 330103, 330106, 330107, 330108*, 330111, 330115, 330125, 330126, 330127, 330128, 330132*, 330135, 330136, 330140, 330141, 330144, 330151, 330153, 330154, 330157, 330158, 330159, 330160, 330162, 330163, 330164, 330166, 330167*, 330169, 330175, 330177, 330180, 330181, 330182, 330184, 330185, 330188, 330191, 330193, 330194, 330195, 330196, 330197, 330198, 330199, 330201*, 330202, 330203, 330204, 330205, 330208, 330211, 330213, 330214, 330215, 330218, 330219, 330221, 330222, 330223, 330224, 330225, 330226, 330229, 330231, 330232*, 330234, 330235, 330236, 330238, 330239, 330240, 330241, 330245, 330246, 330249, 330250, 330259, 330261, 330263, 330264, 330265, 330267, 330268, 330273, 330276, 330277, 330279, 330285, 330286, 330304, 330306*, 330307, 330314, 330316*, 330331, 330332, 330340*, 330350, 330353, 330354, 330372, 330385*, 330386, 330393, 330394, 330395, 330396, 330397*, 330399, 330401, 330403*, 330404, 330405, 330406, 330408*, 331301, 331302, 331303, 331304, 331305, 331306*, 331307, 331309, 331310, 331311, 331312, 331313, 331314, 331315
- OH (n = 175): 360001, 360002, 360003, 360006, 360008, 360009, 360010, 360011, 360012, 360013, 360014, 360016, 360017, 360019*, 360020, 360025, 360026, 360027, 360029, 360032, 360035, 360036, 360037, 360039, 360040, 360041, 360044, 360046, 360048, 360051, 360052*, 360054, 360055, 360056, 360058, 360059, 360064, 360065, 360066, 360068, 360070, 360071, 360072, 360074*, 360075, 360076, 360077, 360078, 360079, 360080, 360081, 360082, 360084, 360085, 360086, 360087, 360089, 360090, 360091, 360092, 360095, 360096, 360098, 360101, 360107, 360109, 360112, 360113*, 360115, 360116, 360118, 360121, 360123, 360125, 360130, 360131, 360132, 360133, 360134, 360137, 360141*, 360143, 360144, 360145, 360147, 360148, 360150, 360151*, 360152, 360153, 360155, 360156, 360159, 360161, 360163, 360170, 360172, 360174, 360175, 360179, 360180, 360185, 360189, 360192, 360195*, 360197, 360203, 360210, 360211, 360212*, 360218, 360230, 360234, 360236, 360239, 360241, 360242, 360245, 360247, 360259, 360261*, 360262, 360263, 360266, 360269*, 360270, 360271*, 360274, 360276, 360347*, 360348, 360350, 360351, 360352, 360354, 360355, 360356*, 360358, 360359, 360360, 360361, 360362*, 361300, 361301, 361302, 361303, 361304, 361305*, 361306, 361307, 361308, 361309, 361310, 361311, 361312, 361313, 361314, 361315, 361316, 361318, 361319, 361320, 361321, 361322, 361323, 361324, 361325, 361326, 361327, 361328, 361329, 361330, 361331, 361332, 361334
- OK (n = 126): 370001, 370002, 370004, 370006, 370007, 370008, 370011*, 370013, 370014, 370015, 370016, 370018, 370019, 370020, 370022, 370023, 370025, 370026, 370028, 370029, 370030, 370032*, 370034, 370036, 370037, 370039, 370040, 370041, 370047, 370048, 370049, 370051*, 370054, 370056, 370057, 370065, 370072*, 370078, 370080, 370083, 370089, 370091, 370093, 370094, 370097, 370099, 370100, 370103*, 370106, 370112, 370113, 370114, 370138, 370139, 370149, 370153, 370156*, 370158, 370166, 370169*, 370170, 370171, 370172, 370173, 370178, 370180, 370183, 370190*, 370192, 370199, 370201, 370202, 370203, 370206, 370210, 370211, 370212, 370214, 370215, 370216, 370218, 370220, 370222, 370225, 370227, 370228, 370229, 370232*, 370233*, 370234, 370235, 370236, 371300, 371301, 371302, 371303, 371304, 371305, 371306, 371307, 371309, 371310, 371311, 371312, 371313, 371314, 371316, 371317, 371318, 371319, 371320, 371321, 371322, 371323, 371324, 371325, 371326, 371327, 371328, 371329, 371330, 371331, 371332, 371333, 371334, 371335
- OR (n = 58): 380001, 380002, 380004, 380005, 380007, 380009, 380014, 380017, 380018, 380020, 380021, 380022, 380025, 380027, 380029, 380033, 380037, 380038, 380040, 380047, 380050, 380051, 380052, 380056, 380060, 380061, 380071, 380075, 380082, 380089, 380090, 380091, 380102, 381301, 381302, 381303, 381304, 381305, 381306, 381307, 381308, 381309, 381310, 381311, 381312, 381313, 381314, 381315, 381316, 381317, 381318, 381319, 381320, 381321, 381322, 381323, 381324, 381325
- PA (n = 170): 390001, 390002, 390003, 390004, 390006, 390008, 390009, 390012, 390013, 390016, 390019*, 390025, 390026, 390027, 390028, 390030, 390031, 390032, 390035, 390036, 390037, 390039, 390041, 390042, 390043, 390044, 390045, 390046, 390048, 390049, 390050, 390052*, 390056, 390057, 390058, 390061*, 390062, 390063, 390065, 390066, 390067, 390068, 390070, 390071, 390072, 390073, 390076*, 390079, 390080*, 390081, 390084*, 390086, 390090, 390091, 390093, 390095, 390096, 390097, 390100, 390101, 390102, 390104, 390107, 390108*, 390110, 390111, 390112, 390113, 390114, 390115, 390116, 390117, 390118, 390119, 390122, 390123, 390125, 390127, 390130, 390132*, 390133, 390137, 390138, 390139, 390142, 390145, 390146, 390150, 390151, 390153, 390154, 390157, 390160, 390162, 390163, 390164, 390168, 390173, 390174, 390178, 390179, 390180, 390183, 390184, 390185, 390189, 390192*, 390194*, 390195, 390196, 390197*, 390198, 390199, 390201, 390203, 390204, 390211, 390217, 390219, 390220*, 390222, 390223, 390225, 390226, 390228, 390231, 390233, 390236*, 390237, 390256, 390258, 390263*, 390265, 390266, 390267, 390268, 390270, 390272, 390278, 390290*, 390302*, 390304, 390307, 390312*, 390313*, 390314*, 390316, 390318*, 390321, 390322, 390323, 390324, 390325, 390326, 390327, 390328, 390329, 391300, 391301, 391302, 391303, 391304, 391305, 391306, 391307, 391308, 391309, 391311*, 391312, 391313
- RI (n = 10): 410001*, 410004, 410005, 410006, 410007, 410008, 410009, 410010, 410011, 410012
- SC (n = 62): 420002, 420004, 420005, 420007, 420009, 420010, 420011, 420015, 420016*, 420018, 420019, 420020, 420023, 420026, 420027, 420030, 420033, 420036, 420037, 420038, 420039, 420043, 420048, 420049, 420051, 420053, 420054*, 420055, 420056*, 420057*, 420062, 420064, 420065, 420066, 420067, 420068, 420069, 420070, 420071, 420072, 420073, 420078, 420079, 420080, 420082, 420083*, 420085, 420086, 420087, 420089, 420091, 420098, 420101, 420102, 420103, 420104, 420105, 421300, 421301, 421302*, 421303, 421304
- SD (n = 61): 430005, 430008, 430012, 430013, 430014, 430015, 430016, 430027, 430048, 430077, 430081, 430082*, 430083, 430084, 430089, 430090, 430091, 430092*, 430093, 430094, 430095, 430096*, 430097, 431300, 431301, 431302, 431305, 431306, 431307, 431308, 431309, 431310, 431311, 431312, 431313, 431314, 431315, 431316, 431317, 431318, 431319, 431320, 431321, 431322, 431323, 431324, 431325, 431326, 431327, 431328, 431329, 431330, 431331, 431332, 431333, 431334, 431335, 431336, 431337, 431338, 431339
- TN (n = 117): 440001, 440002, 440003, 440006, 440007, 440008, 440009, 440010, 440011, 440012, 440015, 440016, 440017, 440018, 440020, 440025*, 440029, 440030, 440031, 440032, 440033, 440034, 440035, 440039, 440040*, 440046, 440047*, 440048, 440049, 440050, 440051*, 440053, 440054*, 440056, 440057, 440058, 440059, 440060, 440061, 440063, 440064*, 440065, 440067*, 440068, 440070*, 440072, 440073, 440081, 440082, 440083, 440084, 440091, 440102, 440104, 440109, 440110, 440111, 440115*, 440120, 440125, 440130, 440131, 440132, 440133*, 440137, 440141, 440144, 440147*, 440148, 440150, 440151, 440152, 440153, 440156, 440159, 440161, 440162, 440168*, 440173, 440174*, 440175, 440176, 440180, 440181, 440182*, 440183, 440184, 440185, 440187, 440189*, 440192, 440193, 440194, 440197, 440200, 440218, 440227, 440228, 440231, 440232, 441300, 441301, 441303, 441304, 441305, 441306, 441307, 441308, 441309, 441310, 441311, 441312, 441313, 441314, 441315*, 441316, 441317
- TX (n = 402): 450002, 450005*, 450007, 450008, 450010, 450011, 450015, 450018, 450021, 450023, 450024, 450028, 450029, 450032, 450033, 450034, 450035, 450037, 450039, 450040, 450042, 450044, 450046, 450051, 450052, 450054, 450055, 450056, 450058, 450064, 450068, 450072, 450073, 450076, 450078, 450079, 450080, 450082, 450083, 450085, 450087, 450090, 450092, 450097, 450099, 450101, 450102, 450104, 450107, 450108, 450119, 450123, 450124, 450128, 450130*, 450132, 450133, 450135, 450137, 450143, 450144, 450147, 450148, 450152, 450154, 450155, 450162, 450163, 450165, 450176, 450177, 450178, 450184, 450187, 450188*, 450192, 450193, 450194, 450196, 450200, 450203, 450209, 450210, 450211, 450213, 450214*, 450219, 450221, 450222, 450229, 450231, 450234, 450235, 450236, 450237, 450241, 450243*, 450253, 450270*, 450271, 450272, 450280, 450283*, 450289, 450292, 450293, 450296*, 450299*, 450306*, 450324, 450330, 450340, 450346, 450347, 450348, 450351, 450352, 450358, 450369, 450370, 450372, 450373, 450379, 450388, 450389, 450395, 450399, 450400, 450403, 450411, 450419, 450422, 450424, 450431, 450438, 450446*, 450447, 450451, 450460, 450462, 450465, 450469, 450475, 450484, 450488*, 450489, 450497*, 450498, 450508, 450518, 450537, 450539, 450558, 450563, 450565, 450571, 450573, 450578, 450580*, 450584, 450586, 450587, 450591*, 450596, 450597, 450604, 450605*, 450610, 450615*, 450617, 450620, 450630*, 450634, 450638, 450639, 450641, 450643, 450644, 450647, 450651, 450653, 450654, 450656, 450658, 450659, 450661, 450662, 450668, 450669, 450670, 450672, 450674, 450675, 450677, 450678, 450683, 450684, 450686, 450688, 450690, 450694, 450697, 450698, 450702, 450709, 450711, 450713, 450716*, 450718, 450723, 450730, 450742, 450743, 450746, 450747, 450749*, 450754, 450755, 450766*, 450770*, 450771, 450774, 450775, 450779, 450780*, 450788, 450795*, 450796*, 450797*, 450801, 450803*, 450804, 450808, 450809, 450813, 450820, 450822, 450825, 450827, 450828, 450831, 450832, 450833, 450834, 450839, 450840, 450841, 450844, 450845*, 450847, 450848, 450851, 450853, 450855, 450856, 450860, 450862, 450864, 450865, 450867, 450869, 450871, 450872, 450874, 450875, 450876, 450877, 451300, 451301, 451302*, 451303, 451304, 451305, 451306, 451307, 451308, 451309, 451310, 451311, 451312, 451313, 451314, 451315, 451316, 451317, 451318, 451319, 451320, 451321*, 451322, 451323, 451324, 451325, 451326, 451328, 451329, 451330, 451331, 451332, 451333, 451334, 451335, 451337, 451338, 451339, 451340, 451341, 451342, 451343, 451344, 451345, 451346, 451347, 451348, 451349, 451350, 451351, 451352, 451353, 451354, 451355, 451356, 451357*, 451358, 451359, 451360, 451361, 451362, 451363, 451364, 451365, 451366, 451367, 451369, 451370, 451371, 451372, 451373, 451374, 451375, 451376, 451377, 451378, 451379, 451380, 451381, 451382, 670002*, 670004, 670005, 670006, 670007*, 670008, 670010, 670012*, 670018, 670019*, 670021*, 670023, 670024*, 670025, 670029*, 670031, 670034, 670041, 670043, 670044, 670045*, 670046, 670047, 670049, 670050, 670052*, 670053, 670054, 670055, 670056, 670058, 670059, 670060, 670062, 670063, 670064, 670067, 670068, 670069, 670071, 670072, 670073, 670074, 670075, 670076, 670077, 670078, 670079*, 670080, 670081, 670082, 670083*, 670085, 670087*, 670088
- UT (n = 43): 460001, 460003, 460004, 460005, 460006, 460007, 460009, 460010, 460011, 460014, 460015, 460017, 460019, 460021, 460023, 460026, 460030, 460033, 460035, 460039, 460041, 460042, 460043, 460044, 460047, 460049, 460051, 460052, 460054, 460057, 460058, 460059, 461300, 461301, 461302, 461303, 461304, 461305, 461306, 461307, 461308, 461309, 461310
- VA (n = 85): 490001*, 490002, 490004, 490005, 490007, 490009, 490011*, 490012*, 490013, 490017, 490018, 490019, 490020, 490021, 490022, 490023, 490024, 490027, 490032, 490033, 490037, 490038, 490040, 490041, 490042, 490043, 490044, 490045, 490046, 490048, 490050, 490052, 490053, 490059, 490060, 490063, 490066, 490069, 490075, 490077, 490079*, 490084, 490088, 490089, 490090, 490092, 490093, 490094, 490097, 490098, 490101, 490104, 490105, 490106, 490107, 490108, 490109, 490110, 490111, 490112, 490113, 490114, 490115, 490116, 490117, 490118, 490120, 490122, 490123, 490126, 490127, 490129*, 490130, 490134*, 490135, 490136, 490140, 490141, 491300, 491302, 491303, 491304, 491305, 491306*, 491307
- VT (n = 14): 470001, 470003, 470005, 470011, 470012, 470024, 471300, 471301, 471302, 471303, 471304, 471305, 471306, 471307
- WA (n = 88): 500001*, 500002, 500003, 500005, 500007, 500008, 500011, 500012, 500014, 500015, 500016, 500019, 500021, 500024, 500025, 500026, 500027, 500030, 500031, 500033, 500036, 500037, 500039, 500041, 500044, 500049*, 500050, 500051, 500052, 500053, 500054, 500058, 500060, 500064, 500072, 500077, 500079, 500084, 500088, 500108, 500119, 500124, 500129, 500138, 500139, 500141, 500148, 500150, 500151, 500152, 501301, 501302, 501304, 501305, 501307, 501308, 501309, 501310, 501311, 501312, 501313, 501314, 501315, 501316, 501317, 501318, 501319, 501320, 501321, 501322, 501323, 501324, 501325, 501326, 501327, 501328, 501329, 501330, 501331, 501332, 501333, 501334, 501335, 501336, 501337, 501338, 501339, 501340
- WI (n = 124): 520002, 520004, 520008, 520009, 520011, 520013, 520017, 520019, 520021, 520027*, 520028, 520030, 520033, 520034, 520035, 520037, 520038, 520041, 520044, 520045, 520048*, 520049, 520051, 520057, 520059, 520062, 520063, 520066, 520070, 520071, 520075, 520076, 520078, 520083, 520087, 520088, 520089, 520091, 520095, 520096, 520097, 520098, 520100, 520102, 520103, 520107, 520109, 520113, 520116, 520136, 520138, 520139, 520160, 520177, 520189, 520193, 520194, 520195*, 520196, 520198, 520202, 520204*, 520205, 520206, 520207, 520208, 521300, 521302, 521303, 521304, 521305, 521307, 521308, 521309, 521310, 521311, 521312, 521313, 521314, 521315, 521316, 521317, 521318, 521319, 521320, 521321, 521322, 521323, 521324, 521325, 521326, 521327, 521328, 521329, 521330, 521331, 521332, 521333, 521334, 521335, 521336, 521337, 521338, 521339, 521340, 521341, 521342, 521343, 521344, 521345, 521346, 521347, 521348, 521349, 521350, 521351, 521352, 521353, 521354, 521355, 521356, 521357, 521358, 521359
- WV (n = 51): 510001, 510002, 510006, 510007, 510008, 510012, 510013, 510018, 510022, 510023, 510024, 510029, 510030, 510031, 510038, 510039*, 510046, 510047*, 510048, 510050, 510053, 510055, 510058, 510059, 510062, 510070, 510071, 510072, 510077*, 510082, 510085*, 510086, 511300, 511301, 511303, 511304, 511306, 511307, 511308, 511309, 511310, 511311, 511312, 511313, 511314, 511315, 511316, 511317, 511318, 511319, 511320
- WY (n = 27): 530002, 530006, 530008, 530010*, 530011, 530012, 530014, 530015, 530025, 530032, 530033*, 531301, 531302, 531303, 531304, 531305, 531306, 531307, 531308, 531309, 531310, 531311, 531312, 531313, 531314, 531315, 531316
Appendix B. Analytic Code
The complete analytic code follows (Supplementary File S2, mediation_closure_analysis.py). It starts from the raw cost-report files and the county file with the eligibility of hospitals and reports and then constructs the closure outcome, the exposures and mediators, the mediation models, the bootstrap, and all tables, figures, and the hospital list. Every number in this paper is generated by this code. The path diagrams in Figure 2 to 4 are drawn by path_model.py, also in Supplementary File S2.
#!/usr/bin/env python3
"""
Mediators of the associations of county poverty, county percentage Black, and rurality with hospital closure.
Raw CMS HCRIS cost-report CSVs + ACS county file -> eligibility -> closure outcome -> multiple-mediator models
(product-of-coefficients, logistic outcome, hospital-clustered bootstrap) -> tables, figures, hospital list.
Usage: python mediation_closure_analysis.py [DATA_DIR] [OUT_DIR] [N_BOOT_MAIN] [N_BOOT_SENS]
DATA_DIR must hold CostReport_2011_Final.csv ... CostReport_2023_Final.csv and acs5_2019_county_derived.csv
"""
import sys, glob, json, os, re, difflib
import numpy as np, pandas as pd
from scipy import stats
import matplotlib; matplotlib.use('Agg'); import matplotlib.pyplot as plt
DATA_DIR = sys.argv[1] if len(sys.argv) > 1 else '/mnt/user-data/uploads'
OUT_DIR = sys.argv[2] if len(sys.argv) > 2 else './output_med'
B_MAIN = int(sys.argv[3]) if len(sys.argv) > 3 else 400
B_SENS = int(sys.argv[4]) if len(sys.argv) > 4 else 100
os.makedirs(OUT_DIR, exist_ok=True)
# ============ 1. LOAD RAW COST REPORTS ============
frames = []
for f in sorted(glob.glob(os.path.join(DATA_DIR, 'CostReport_*_Final.csv'))):
d = pd.read_csv(f, low_memory=False); d['file_year'] = int(f.split('_')[-2]); frames.append(d)
raw = pd.concat(frames, ignore_index=True)
raw = raw.rename(columns={'Provider CCN': 'ccn', 'State Code': 'state', 'Fiscal Year Begin Date': 'fbeg', 'Fiscal Year End Date': 'fend',
'Type of Control': 'ctrl', 'Rural Versus Urban': 'rural_cls', 'Hospital Name': 'name', 'Street Address': 'addr', 'Zip Code': 'zip'})
raw['fbeg'] = pd.to_datetime(raw.fbeg, errors='coerce'); raw['fend'] = pd.to_datetime(raw.fend, errors='coerce')
raw['days'] = (raw.fend - raw.fbeg).dt.days; raw['yr'] = raw.file_year; raw['suf'] = raw.ccn % 10000
# ============ 2. ELIGIBILITY ============
# (a) short-term acute care (STH) and critical access (CAH) hospitals; (b) one report per provider-year (longer report kept);
# (c) reporting period 150-365 days (end minus begin date); (d) ownership classifiable (control codes 1-13); (e) 50 states + DC
FLOW = {}
e0 = (raw[raw['CCN Facility Type'].isin(['STH', 'CAH'])].sort_values(['ccn', 'yr', 'days'], ascending=[True, True, False]).drop_duplicates(['ccn', 'yr']))
FLOW['source_reports'] = len(e0); FLOW['source_hosp'] = int(e0.ccn.nunique())
e0 = e0[e0.ctrl.between(1, 13)].copy(); e_p = e0[(e0.days >= 150) & (e0.days <= 365)].copy()
FLOW['after_period'] = len(e_p); FLOW['after_period_hosp'] = int(e_p.ccn.nunique())
# ============ 3. CLOSURE OUTCOME ============
def nrm(s):
return s.fillna('').astype(str).str.upper().str.replace(r'[^A-Z0-9 ]', '', regex=True).str.replace(r'\s+', ' ', regex=True).str.strip()
raw['a'] = nrm(raw.addr).str.slice(0, 22) + '|' + raw.zip.astype(str).str.slice(0, 5); raw['n'] = nrm(raw.name).str.slice(0, 18) + '|' + nrm(raw['City'])
hos = e_p.groupby('ccn').yr.agg(['min', 'max']).reset_index(); rawlast = raw.groupby('ccn').yr.max()
by_addr = raw.groupby('a').apply(lambda x: list(zip(x.ccn, x.yr))).to_dict(); by_name = raw.groupby('n').apply(lambda x: list(zip(x.ccn, x.yr))).to_dict()
lastrow = raw.sort_values('yr').groupby('ccn').tail(1).set_index('ccn')
def has_successor(c, L):
for key, idx in (('a', by_addr), ('n', by_name)):
k = lastrow.loc[c, key]
if isinstance(k, str) and not k.startswith('|'):
for c2, y2 in idx.get(k, []):
if c2 != c and y2 >= L: return True
return False
hos['cand'] = hos['max'] <= 2022
hos['succ'] = [bool(c and has_successor(cc, L)) for cc, L, c in zip(hos.ccn, hos['max'], hos.cand)]
lr = raw.merge(hos[['ccn', 'max']], on='ccn'); lr = lr[lr.file_year > lr['max']]; later_full = lr[lr.days >= 150].groupby('ccn').size()
cl = hos.copy(); cl['Y'] = cl.ccn.map(raw.groupby('ccn').yr.max()); cl['later_full'] = cl.ccn.map(later_full).fillna(0)
cl['closed'] = cl.cand & ~cl.succ & (cl.later_full == 0) # closure: final report (any length) in Y<=2021, no later full-year report, no successor CCN
# ============ 4. COUNTY EXPOSURES (ACS 2015-2019 5-year) matched to hospitals by state + county name ============
FIPS = {'01':'AL','02':'AK','04':'AZ','05':'AR','06':'CA','08':'CO','09':'CT','10':'DE','11':'DC','12':'FL','13':'GA','15':'HI','16':'ID','17':'IL','18':'IN','19':'IA','20':'KS','21':'KY','22':'LA','23':'ME','24':'MD','25':'MA','26':'MI','27':'MN','28':'MS','29':'MO','30':'MT','31':'NE','32':'NV','33':'NH','34':'NJ','35':'NM','36':'NY','37':'NC','38':'ND','39':'OH','40':'OK','41':'OR','42':'PA','44':'RI','45':'SC','46':'SD','47':'TN','48':'TX','49':'UT','50':'VT','51':'VA','53':'WA','54':'WV','55':'WI','56':'WY'}
def cnorm(s):
s = str(s).lower().replace('ñ', 'n').replace('.', '').replace("'", '').replace('-', ' '); s = re.sub(r'\bsaint\b', 'st', s)
s = re.sub(r'\b(county|parish|borough|census area|municipality|city and borough|and borough)\b', '', s); return re.sub(r'\s+', ' ', s).strip()
cty = pd.read_csv(os.path.join(DATA_DIR, 'acs5_2019_county_derived.csv'), dtype={'state_fips': str, 'county_fips': str}); cty['st'] = cty.state_fips.map(FIPS); cty = cty.dropna(subset=['st'])
cty['base'] = cty.county_name.str.split(',').str[0]; cty['key'] = cty.st + '|' + cty.base.map(cnorm); cty['is_city'] = cty.base.str.contains(r' city$', case=False)
cty = cty.sort_values('is_city').drop_duplicates('key'); cty['fips'] = cty.state_fips + cty.county_fips
first = e0.sort_values('yr').groupby('ccn').first(); HC = pd.DataFrame({'state': first.state, 'county_raw': first['County']}); HC['key'] = HC.state + '|' + HC.county_raw.map(cnorm)
CC = ['fips', 'pct_poverty', 'pct_black_alone', 'log_population']; HC = HC.join(cty.set_index('key')[CC], on='key')
byst = {s_: cty[cty.st == s_].set_index('key') for s_ in cty.st.unique()}
ALIAS = {'DC|dc': 'DC|district of columbia', 'AK|north star': 'AK|fairbanks north star', 'AK|municipality of anchorage': 'AK|anchorage', 'AK|wrangellpetersburg': 'AK|petersburg', 'AK|kobuk': 'AK|northwest arctic'}
for i, r in HC[HC.fips.isna()].iterrows():
if not isinstance(r.county_raw, str) or r.state not in byst: continue
k = ALIAS.get(r.key, r.key); keys = list(byst[r.state].index)
cand = [x for x in keys if x == k] or [x for x in keys if (x.split('|')[1] in k.split('|')[1] or k.split('|')[1] in x.split('|')[1]) and len(x.split('|')[1]) > 3]
if not cand: cand = difflib.get_close_matches(k, keys, n=1, cutoff=0.8)
if cand: HC.loc[i, CC] = byst[r.state].loc[cand[0], CC].values
# ============ 5. ANALYTIC PANEL: measures ============
e = e_p.copy(); g = lambda c: e[c]; npr = g('Net Patient Revenue'); oi = g('Total Other Income').fillna(0); tr = npr + oi
e['tm'] = g('Net Income') / tr * 100; e['om'] = (npr - g('Less Total Operating Expense')) / npr * 100
e['dcoh'] = (g('Cash on Hand and in Banks') + g('Temporary Investments').fillna(0)) / ((g('Less Total Operating Expense') - g('Depreciation Cost')) / 365)
e['cur'] = g('Total Current Assets') / g('Total Current Liabilities'); e['lev'] = g('Total Liabilities') / g('Total Assets') * 100
e['occ'] = g('Total Days (V + XVIII + XIX + Unknown)') / g('Total Bed Days Available') * 100; e['lbeds'] = np.log(g('Number of Beds').clip(lower=1))
e['cah'] = ((e.suf >= 1300) & (e.suf <= 1399)).astype(int); e['rural'] = (e.rural_cls == 'R').astype(int); e['teach'] = (g('Number of Interns and Residents (FTE)').fillna(0) > 0).astype(int)
e['mcare'] = g('Total Days Title XVIII') / g('Total Days (V + XVIII + XIX + Unknown)') * 100; e['mcaid'] = g('Medicaid Charges').fillna(0) / g('Combined Outpatient + Inpatient Total Charges') * 100
e['uc'] = g('Cost of Uncompensated Care') / g('Total Costs') * 100; e['charity'] = g('Cost of Charity Care') / g('Total Costs') * 100
e['baddebt'] = (g('Cost of Uncompensated Care') - g('Cost of Charity Care')) / g('Total Costs') * 100; e['fte'] = g('FTE - Employees on Payroll') / g('Number of Beds').clip(lower=1)
WIN = ['tm', 'om', 'dcoh', 'cur', 'lev', 'occ', 'mcare', 'mcaid', 'uc', 'charity', 'baddebt', 'fte']
for v in WIN:
e[v] = e[v].replace([np.inf, -np.inf], np.nan); lo, hi = e[v].quantile([.01, .99]); e[v + '_w'] = e[v].clip(lo, hi)
e['own'] = np.where(e.ctrl.isin([1, 2]), 'Nonprofit', np.where(e.ctrl.isin([3, 4, 5, 6]), 'For-profit', 'Government'))
NE = ['CT','ME','MA','NH','RI','VT','NJ','NY','PA']; MW = ['IL','IN','MI','OH','WI','IA','KS','MN','MO','NE','ND','SD']
SO = ['DE','DC','FL','GA','MD','NC','SC','VA','WV','AL','KY','MS','TN','AR','LA','OK','TX']; WE = ['AZ','CO','ID','MT','NV','NM','UT','WY','AK','CA','HI','OR','WA']
e['region'] = e.state.map(lambda s: 'Northeast' if s in NE else 'Midwest' if s in MW else 'South' if s in SO else 'West' if s in WE else 'Territory')
EXP = {s: 2014 for s in ['AZ','AR','CA','CO','CT','DE','DC','HI','IL','IA','KY','MD','MA','MI','MN','NV','NH','NJ','NM','NY','ND','OH','OR','RI','VT','WA','WV']}
EXP.update({'PA': 2015, 'IN': 2015, 'AK': 2015, 'MT': 2016, 'LA': 2016, 'VA': 2019, 'ME': 2019, 'ID': 2020, 'UT': 2020, 'NE': 2020, 'MO': 2021, 'OK': 2021})
e['expansion'] = [1 if (s in EXP and y >= EXP[s]) else 0 for s, y in zip(e.state, e.yr)]
e = e[e.region != 'Territory']; FLOW['analytic_reports'] = len(e); FLOW['analytic_hosp'] = int(e.ccn.nunique())
base = set(e[e.yr <= 2013].ccn); Ymap = dict(zip(cl[cl.closed].ccn, cl[cl.closed].Y)); e['y'] = (e.ccn.map(Ymap) == e.yr + 1).astype(int)
D = e[(e.yr <= 2020) & e.ccn.isin(base)].copy().join(HC[CC], on='ccn')
FLOW['closure_hy_before_county'] = len(D); FLOW['closures_before_county'] = int(D.y.sum()); FLOW['hosp_before_county'] = int(D.ccn.nunique())
D = D.dropna(subset=['pct_poverty', 'pct_black_alone', 'log_population']).copy() # county exposures required
FLOW['closure_hy'] = len(D); FLOW['closures'] = int(D.y.sum()); FLOW['closure_hosp'] = int(D.ccn.nunique()); FLOW['hosp_unmatched'] = FLOW['hosp_before_county'] - FLOW['closure_hosp']
for v in WIN: D[v + '_i'] = D[v + '_w'].fillna(D[v + '_w'].median())
D['lbeds'] = D.lbeds.fillna(D.lbeds.median()); D['fp'] = (D.own == 'For-profit').astype(int); D['gov'] = (D.own == 'Government').astype(int)
# ============ 6. MEDIATION MODEL ============
z = lambda s: (s - s.mean()) / s.std()
EXPOS = {'Poverty': z(D.pct_poverty), 'PctBlack': z(D.pct_black_alone), 'Rural': D.rural.astype(float)}
EXPLAB = {'Poverty': 'County poverty (per SD)', 'PctBlack': 'County % Black (per SD)', 'Rural': 'Rural (vs urban)'}
MED = {'Beds (log)': ('Size', z(D.lbeds)), 'Occupancy': ('Utilization and staffing', z(D.occ_i)), 'Staff per bed': ('Utilization and staffing', z(D.fte_i)),
'Total margin': ('Performance', z(D.tm_i)), 'Operating margin': ('Performance', z(D.om_i)),
'Days cash': ('Assets', z(D.dcoh_i)), 'Current ratio': ('Assets', z(D.cur_i)), 'Leverage': ('Assets', z(D.lev_i)),
'Medicare share of days': ('Payer mix', z(D.mcare_i)), 'Medicaid share of charges': ('Payer mix', z(D.mcaid_i)),
'Charity care': ('Uncompensated care', z(D.charity_i)), 'Bad debt': ('Uncompensated care', z(D.baddebt_i)),
'Critical access': ('Structure', D.cah.astype(float)), 'Teaching': ('Structure', D.teach.astype(float)),
'For-profit': ('Ownership', D.fp.astype(float)), 'Government': ('Ownership', D.gov.astype(float))}
GROUPS = ['Size', 'Utilization and staffing', 'Performance', 'Assets', 'Payer mix', 'Uncompensated care', 'Structure', 'Ownership']
yy = pd.get_dummies(D.yr, prefix='y', drop_first=True).astype(float); rg = pd.get_dummies(D.region).drop(columns=['Northeast']).astype(float)
COVA = pd.concat([rg, D[['expansion']].astype(float), z(D.log_population).rename('logpop'), yy], axis=1)
def logit(X, y, iters=30):
X = np.c_[np.ones(len(X)), X]; b = np.zeros(X.shape[1]); b[0] = np.log(y.mean() / (1 - y.mean()))
for _ in range(iters):
p = 1 / (1 + np.exp(-np.clip(X @ b, -30, 30))); W = p * (1 - p); st = np.linalg.solve(X.T @ (X * W[:, None]) + 1e-8 * np.eye(X.shape[1]), X.T @ (y - p)); b += st
if np.abs(st).max() < 1e-8: break
return b
def ols(A, Y):
return np.linalg.lstsq(A, Y, rcond=None)[0]
def paths(idx, mnames, expn, cova_cols, drop_mediators=(), logit_flag=False):
"""Parallel multiple-mediator model. Outcome = closure in the following year (linear probability model, so that
total = direct + sum of indirect effects exactly). Mediators are standardized; exposures are per SD (poverty, % Black) or 0/1 (rural)."""
mn = [m for m in mnames if m not in drop_mediators]; k = len(expn); nm = len(mn)
Xe = np.column_stack([EXPOS[x].values[idx] for x in expn]); Cv = COVA[cova_cols].values[idx]; M = np.column_stack([MED[m][1].values[idx] for m in mn]); y = D.y.values[idx].astype(float)
one = np.ones((len(idx), 1)); a = ols(np.c_[one, Xe, Cv], M)[1:1 + k, :] # a-paths: exposure -> mediator
tot = ols(np.c_[one, Xe, Cv], y)[1:1 + k]; full = ols(np.c_[one, Xe, M, Cv], y); dire = full[1:1 + k]; b = full[1 + k:1 + k + nm]
r = dict(total=tot, direct=dire, a=a, b=b, ind=a * b[None, :], mn=mn)
if logit_flag: # logistic sensitivity: odds-ratio scale
r['total_logit'] = logit(np.c_[Xe, Cv], y)[1:1 + k]; bm = logit(np.c_[Xe, M, Cv], y); r['direct_logit'] = bm[1:1 + k]; r['b_logit'] = bm[1 + k:1 + k + nm]
return r
def run(expn, cova_cols, nboot, drop_mediators=(), seed=11, label='main', logit_flag=False):
mn = [m for m in MED if m not in drop_mediators]; idx_all = np.arange(len(D)); est = paths(idx_all, list(MED), expn, cova_cols, drop_mediators, logit_flag)
groups = {g_: [i for i, m in enumerate(mn) if MED[m][0] == g_] for g_ in GROUPS}; groups = {g_: v for g_, v in groups.items() if v}
rng = np.random.default_rng(seed); ids = D.ccn.values; ccn_idx = {c: np.where(ids == c)[0] for c in pd.unique(ids)}; keys = list(ccn_idx.keys()); store = []
for bnum in range(nboot):
samp = np.concatenate([ccn_idx[k_] for k_ in rng.choice(keys, len(keys))]); store.append(paths(samp, list(MED), expn, cova_cols, drop_mediators, logit_flag))
pct = lambda arr: np.percentile(arr, [2.5, 97.5], axis=0)
res = dict(expn=expn, mediators=mn, label=label, nboot=nboot, scale='risk difference per 1,000 hospital-years' )
f = 1000.0
for key in ['total', 'direct', 'a', 'b', 'ind']:
sc = 1.0 if key == 'a' else f
res[key] = (est[key] * sc).tolist(); res[key + '_ci'] = (pct(np.array([s_[key] for s_ in store])) * sc).tolist()
res['groups'] = {g_: (est['ind'][:, v].sum(axis=1) * f).tolist() for g_, v in groups.items()}
gb = {g_: np.array([s_['ind'][:, v].sum(axis=1) * f for s_ in store]) for g_, v in groups.items()}; res['groups_ci'] = {g_: pct(gb[g_]).tolist() for g_ in groups}
res['total_indirect'] = (est['ind'].sum(axis=1) * f).tolist(); res['total_indirect_ci'] = pct(np.array([s_['ind'].sum(axis=1) * f for s_ in store])).tolist()
res['prop_mediated'] = (est['ind'].sum(axis=1) / est['total']).tolist(); res['prop_mediated_ci'] = pct(np.array([s_['ind'].sum(axis=1) / s_['total'] for s_ in store])).tolist()
res['identity_check_max_abs'] = float(np.abs(est['total'] - est['direct'] - est['ind'].sum(axis=1)).max() * f)
if logit_flag:
for key in ['total_logit', 'direct_logit', 'b_logit']:
res[key] = est[key].tolist(); res[key + '_ci'] = pct(np.array([s_[key] for s_ in store])).tolist()
return res
COV_MAIN = list(COVA.columns); COV_NOPOP = [c for c in COVA.columns if c != 'logpop']
R = {'flow': FLOW, 'exposures': EXPLAB}
R['main'] = run(list(EXPOS), COV_MAIN, B_MAIN, label='main', logit_flag=True)
R['sens_noownership'] = run(list(EXPOS), COV_MAIN, B_SENS, drop_mediators=('For-profit', 'Government'), seed=12, label='without ownership mediators')
R['sens_nopop'] = run(list(EXPOS), COV_NOPOP, B_SENS, seed=13, label='without county population covariate')
R['sens_single'] = {k: run([k], COV_MAIN, B_SENS, seed=14 + i, label='exposure ' + k + ' alone') for i, k in enumerate(EXPOS)}
R['groups_map'] = {m: MED[m][0] for m in MED}
# ============ 7. DESCRIPTIVES, OUTPUTS ============
desc = {}
def ms(s): return [float(s.mean()), float(s.std()), float(s.median())]
for lab, col in [('County poverty, %', 'pct_poverty'), ('County % Black', 'pct_black_alone'), ('County population (log)', 'log_population'), ('Beds (median)', None)]:
if col: desc[lab] = {'all': ms(D[col]), 'rural': ms(D[D.rural == 1][col]), 'urban': ms(D[D.rural == 0][col])}
for lab, col in [('Beds', 'Number of Beds'), ('Occupancy, %', 'occ_i'), ('Staff per bed', 'fte_i'), ('Total margin, %', 'tm_i'), ('Operating margin, %', 'om_i'), ('Days cash', 'dcoh_i'), ('Current ratio', 'cur_i'), ('Leverage, %', 'lev_i'),
('Medicare share of days, %', 'mcare_i'), ('Medicaid share of charges, %', 'mcaid_i'), ('Charity care, % of costs', 'charity_i'), ('Bad debt, % of costs', 'baddebt_i')]:
desc[lab] = {'all': ms(D[col]), 'rural': ms(D[D.rural == 1][col]), 'urban': ms(D[D.rural == 0][col])}
for lab, col in [('Critical access, %', 'cah'), ('Teaching, %', 'teach'), ('For-profit, %', 'fp'), ('Government, %', 'gov'), ('Closure in following year, per 1,000', 'y')]:
k = 1000 if col == 'y' else 100; desc[lab] = {'all': [float(D[col].mean() * k)], 'rural': [float(D[D.rural == 1][col].mean() * k)], 'urban': [float(D[D.rural == 0][col].mean() * k)]}
R['desc'] = desc; R['n'] = dict(hy=len(D), hosp=int(D.ccn.nunique()), closures=int(D.y.sum()), rural_hy=int((D.rural == 1).sum()), rural_closures=int(D[D.rural == 1].y.sum()), urban_closures=int(D[D.rural == 0].y.sum()))
R['corr'] = {a + '|' + b_: float(stats.spearmanr(D[ca], D[cb])[0]) for (a, ca), (b_, cb) in [(('Poverty', 'pct_poverty'), ('PctBlack', 'pct_black_alone')), (('Poverty', 'pct_poverty'), ('Rural', 'rural')), (('PctBlack', 'pct_black_alone'), ('Rural', 'rural'))]}
def clean(o):
if isinstance(o, dict): return {str(k): clean(v) for k, v in o.items()}
if isinstance(o, (list, tuple)): return [clean(x) for x in o]
if isinstance(o, (np.floating, float)): return None if np.isnan(o) else float(o)
if isinstance(o, np.integer): return int(o)
return o
json.dump(clean(R), open(os.path.join(OUT_DIR, 'results_mediation.json'), 'w'), indent=1)
hl = D.groupby('ccn').agg(state=('state', 'first'), facility_type=('CCN Facility Type', 'first'), name=('name', 'last'), n_reports=('yr', 'size'), first_year=('yr', 'min'), last_year=('yr', 'max'), ownership=('own', 'last'), rural=('rural', 'last'),
county_fips=('fips', 'first'), county_poverty=('pct_poverty', 'first'), county_pct_black=('pct_black_alone', 'first'), closure_event=('y', 'max'))
hl.reset_index().rename(columns={'ccn': 'provider_number'}).to_csv(os.path.join(OUT_DIR, 'hospitals_included_mediation.csv'), index=False)
# figure: grouped indirect effects (risk difference per 1,000 hospital-years) by exposure
m = R['main']; fig, axs = plt.subplots(1, 3, figsize=(11, 3.8), dpi=250, sharey=True); cols = ['#1f3b63', '#b5192f', '#2a7f62']; gl = list(m['groups'].keys())
for ki, (k, ax) in enumerate(zip(['Poverty', 'PctBlack', 'Rural'], axs)):
for i, gname in enumerate(gl):
est = m['groups'][gname][ki]; lo, hi = m['groups_ci'][gname][0][ki], m['groups_ci'][gname][1][ki]; yv = len(gl) - 1 - i
ax.errorbar(est, yv, xerr=[[est - lo], [hi - est]], fmt='o', color=cols[ki], ms=4, capsize=2.5, lw=1.3)
ax.axvline(0, color='k', ls=':', lw=1); ax.set_title(EXPLAB[k], loc='left', fontsize=9, fontweight='bold'); ax.tick_params(labelsize=8)
ax.set_xlabel('Indirect effect on closure, per 1,000 hospital-years (95% CI)', fontsize=7.5)
for s_ in ['top', 'right']: ax.spines[s_].set_visible(False)
axs[0].set_yticks(range(len(gl))); axs[0].set_yticklabels(gl[::-1], fontsize=8.5)
plt.tight_layout(); plt.savefig(os.path.join(OUT_DIR, 'fig_mediation_groups.png')); plt.close()
print('FLOW', FLOW); print('n', R['n']); print('Done. Outputs in', OUT_DIR)
References
- Kaufman, B.G.; Thomas, S.R.; Randolph, R.K.; Perry, J.R.; Thompson, K.W.; Holmes, G.M.; et al. The rising rate of rural hospital closures. J. Rural Health 2016, 32, 35–43. [Google Scholar] [CrossRef]
- Government Accountability Office. Rural Hospital Closures: Number and Characteristics of Affected Hospitals and Contributing Factors; GAO-18-634; U.S. Government Accountability Office: Washington, DC, USA, 2018. [Google Scholar]
- Bai, G.; Yehia, F.; Chen, W.; Anderson, G.F. Varying trends in the financial viability of US rural hospitals, 2011–2017. Health Aff. 2020, 39, 942–948. [Google Scholar] [CrossRef]
- Holmes, G.M.; Pink, G.H.; Friedman, S.A. The financial performance of rural hospitals and implications for elimination of the Critical Access Hospital program. J. Rural Health 2013, 29, 140–149. [Google Scholar] [CrossRef]
- Gujral, K.; Basu, A. Impact of rural and urban hospital closures on inpatient mortality. In NBER Working Paper 26182; National Bureau of Economic Research: Cambridge, MA, USA, 2019. [Google Scholar]
- Hsia, R.Y.; Kellermann, A.L.; Shen, Y.C. Factors associated with closures of emergency departments in the United States. JAMA 2011, 305, 1978–1985. [Google Scholar] [CrossRef]
- Hsia, R.Y.; Shen, Y.C. Rising closures of hospital trauma centers disproportionately burden vulnerable populations. Health Aff. 2011, 30, 1912–1920. [Google Scholar] [CrossRef]
- Buchmueller, T.C.; Jacobson, M.; Wold, C. How far to the hospital? The effect of hospital closures on access to care. J. Health Econ. 2006, 25, 740–761. [Google Scholar]
- Miller, K.E.M.; James, H.J.; Holmes, G.M.; Van Houtven, C.H. The effect of rural hospital closures on emergency medical service response and transport times. Health Serv. Res. 2020, 55, 288–300. [Google Scholar] [CrossRef]
- Germack, H.D.; Kandrack, R.; Martsolf, G.R. When rural hospitals close, the physician workforce goes. Health Aff. 2019, 38, 2086–2094. [Google Scholar] [CrossRef]
- Carroll, C. Impeding Access or Promoting Efficiency? Effects of Rural Hospital Closure on the Cost and Quality of Care; NBER Working Paper 25437; National Bureau of Economic Research: Cambridge, MA, USA, 2019. [Google Scholar]
- Holmes, G.M.; Slifkin, R.T.; Randolph, R.K.; Poley, S. The effect of rural hospital closures on community economic health. Health Serv. Res. 2006, 41, 467–485. [Google Scholar] [CrossRef]
- Wishner, J.; Solleveld, P.; Rudowitz, R.; Paradise, J.; Antonisse, L. A Look at Rural Hospital Closures and Implications for Access to Care: Three Case Studies; Kaiser Family Foundation: Washington, DC, USA, 2016. [Google Scholar]
- Young, G.J.; Chou, C.H.; Alexander, J.; Lee, S.Y.; Raver, E. Provision of community benefits by tax-exempt U.S. hospitals. N. Engl. J. Med. 2013, 368, 1519–1527. [Google Scholar] [CrossRef]
- Garthwaite, C.; Gross, T.; Notowidigdo, M.J. Hospitals as insurers of last resort. Am. Econ. J. Appl. Econ. 2018, 10, 1–39. [Google Scholar] [CrossRef]
- Ko, M.; Needleman, J.; Derose, K.P.; Laugesen, M.J.; Ponce, N.A. Residential segregation and the survival of U.S. urban public hospitals. Med. Care Res. Rev. 2014, 71, 243–260. [Google Scholar] [CrossRef]
- Williams, D.R.; Collins, C. Racial residential segregation: A fundamental cause of racial disparities in health. Public Health Rep. 2001, 116, 404–416. [Google Scholar] [CrossRef]
- Phelan, J.C.; Link, B.G. Is racism a fundamental cause of inequalities in health? Annu. Rev. Sociol. 2015, 41, 311–330. [Google Scholar] [CrossRef]
- Gaskin, D.J.; Dinwiddie, G.Y.; Chan, K.S.; McCleary, R.R. Residential segregation and the availability of primary care physicians. Health Serv. Res. 2012, 47, 2353–2376. [Google Scholar] [CrossRef]
- Jha, A.K.; Orav, E.J.; Li, Z.; Epstein, A.M. Concentration and quality of hospitals that care for elderly black patients. Arch. Intern. Med. 2007, 167, 1177–1182. [Google Scholar] [CrossRef]
- Mullner, R.M.; Rydman, R.J.; Whiteis, D.G.; Rich, R.F. Rural community hospitals and factors correlated with their risk of closing. Public Health Rep. 1989, 104, 315–325. [Google Scholar]
- Holmes, G.M.; Kaufman, B.G.; Pink, G.H. Predicting financial distress and closure in rural hospitals. J. Rural Health 2017, 33, 239–249. [Google Scholar] [CrossRef]
- Pink, G.H.; Holmes, G.M.; D’Alpe, C.; Strunk, L.A.; McGee, P.; Slifkin, R.T. Financial indicators for critical access hospitals. J. Rural Health 2006, 22, 229–236. [Google Scholar] [CrossRef]
- Joynt, K.E.; Harris, Y.; Orav, E.J.; Jha, A.K. Quality of care and patient outcomes in critical access rural hospitals. JAMA 2011, 306, 45–52. [Google Scholar] [CrossRef]
- Lindrooth, R.C.; Perraillon, M.C.; Hardy, R.Y.; Tung, G.J. Understanding the relationship between Medicaid expansions and hospital closures. Health Aff. 2018, 37, 111–120. [Google Scholar] [CrossRef]
- Ly, D.P.; Jha, A.K.; Epstein, A.M. The association between hospital margins, quality of care, and closure or other change in operating status. J. Gen. Intern. Med. 2011, 26, 1291–1296. [Google Scholar] [CrossRef]
- Bai, G.; Anderson, G.F. A more detailed understanding of factors associated with hospital profitability. Health Aff. 2016, 35, 889–897. [Google Scholar] [CrossRef]
- Bazzoli, G.J.; Chen, H.F.; Zhao, M.; Lindrooth, R.C. Hospital financial condition and the quality of patient care. Health Econ. 2008, 17, 977–995. [Google Scholar] [CrossRef]
- Horwitz, J.R. Making profits and providing care: Comparing nonprofit, for-profit, and government hospitals. Health Aff. 2005, 24, 790–801. [Google Scholar] [CrossRef]
- Mechanic, R.; Coleman, K.; Dobson, A. Teaching hospital costs: Implications for academic missions in a competitive market. JAMA 1998, 280, 1015–1019. [Google Scholar]
- Zare, H.; Eisenberg, M.; Anderson, G. Charity care and community benefit in non-profit hospitals: Definition and requirements. Inquiry 2021, 58, 469580211028180. [Google Scholar] [CrossRef]
- Zare, H.; Eisenberg, M.D.; Anderson, G. Comparing the value of community benefit and tax-exemption in non-profit hospitals. Health Serv. Res. 2022, 57, 270–284. [Google Scholar] [CrossRef]
- Bai, G.; Zare, H.; Eisenberg, M.D.; Polsky, D.; Anderson, G.F. Analysis suggests government and nonprofit hospitals’ charity care is not aligned with their favorable tax treatment. Health Aff. 2021, 40, 629–636. [Google Scholar] [CrossRef]
- Sloan, F.A. Not-for-profit ownership and hospital behavior. In Handbook of Health Economics; Culyer, A.J., Newhouse, J.P., Eds.; Elsevier: Amsterdam, The Netherlands, 2000; Volume 1B, pp. 1141–1174. [Google Scholar]
- MacKinnon, D.P.; Krull, J.L.; Lockwood, C.M. Equivalence of the mediation, confounding and suppression effect. Prev. Sci. 2000, 1, 173–181. [Google Scholar] [CrossRef]
- Baron, R.M.; Kenny, D.A. The moderator–mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. J. Pers. Soc. Psychol. 1986, 51, 1173–1182. [Google Scholar] [CrossRef]
- MacKinnon, D.P. Introduction to Statistical Mediation Analysis; Routledge: New York, NY, USA, 2008. [Google Scholar]
- Preacher, K.J.; Hayes, A.F. Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models. Behav. Res. Methods 2008, 40, 879–891. [Google Scholar] [CrossRef]
- Imai, K.; Keele, L.; Tingley, D. A general approach to causal mediation analysis. Psychol. Methods 2010, 15, 309–334. [Google Scholar] [CrossRef]
- VanderWeele, T.J. Explanation in Causal Inference: Methods for Mediation and Interaction; Oxford University Press: New York, NY, USA, 2015. [Google Scholar]
- Centers for Medicare; Medicaid Services. Hospital Provider Cost Report Methodology. Available online: https://data.cms.gov/resources/hospital-provider-cost-report-methodology (accessed on 16 September 2026).
- United States Census Bureau. American Community Survey 5-Year Estimates, 2015–2019: Tables B01003, B02001, and B17001; United States Census Bureau: Washington, DC, USA, 2020. [Google Scholar]
Figure 1.
Indirect associations of county poverty, county percentage Black, and rurality with closure through groups of mediators. Risk differences per 1,000 hospital-years with 95% bootstrap confidence intervals.
Figure 1.
Indirect associations of county poverty, county percentage Black, and rurality with closure through groups of mediators. Risk differences per 1,000 hospital-years with 95% bootstrap confidence intervals.

Figure 3.
Path model for county percentage Black. Solid lines: the 95% bootstrap confidence interval excludes zero; dotted gray lines: it includes zero. Red boxes: the indirect association raises closure risk; blue boxes: it lowers it. Numbers on the arrows from the exposure are changes in the mediator (standard deviations; probability for binary mediators) per unit of exposure, those on the arrows to closure are changes in closures per 1,000 hospital-years per standard deviation of the mediator, and each box gives their product (the indirect association). The arc over the top is the direct association.
Figure 3.
Path model for county percentage Black. Solid lines: the 95% bootstrap confidence interval excludes zero; dotted gray lines: it includes zero. Red boxes: the indirect association raises closure risk; blue boxes: it lowers it. Numbers on the arrows from the exposure are changes in the mediator (standard deviations; probability for binary mediators) per unit of exposure, those on the arrows to closure are changes in closures per 1,000 hospital-years per standard deviation of the mediator, and each box gives their product (the indirect association). The arc over the top is the direct association.

Figure 4.
Path model for rural location. Solid lines: the 95% bootstrap confidence interval excludes zero; dotted gray lines: it includes zero. Red boxes: the indirect association raises closure risk; blue boxes: it lowers it. Numbers on the arrows from the exposure are changes in the mediator (standard deviations; probability for binary mediators) per unit of exposure, those on the arrows to closure are changes in closures per 1,000 hospital-years per standard deviation of the mediator, and each box gives their product (the indirect association). The arc over the top is the direct association.
Figure 4.
Path model for rural location. Solid lines: the 95% bootstrap confidence interval excludes zero; dotted gray lines: it includes zero. Red boxes: the indirect association raises closure risk; blue boxes: it lowers it. Numbers on the arrows from the exposure are changes in the mediator (standard deviations; probability for binary mediators) per unit of exposure, those on the arrows to closure are changes in closures per 1,000 hospital-years per standard deviation of the mediator, and each box gives their product (the indirect association). The arc over the top is the direct association.

Table 1.
Derivation of the analytic sample.
| Step | Reports / hospital-years | Hospitals | Closures |
| Short-term acute care and critical access hospital cost reports, FY2011–2023 (one per provider-year) | 60,992 | 5,250 | — |
| Reporting period 150–365 days | 60,501 | 5,212 | — |
| 50 states and District of Columbia (territories excluded) | 59,773 | 5,147 | — |
| Closure analysis before county linkage: baseline cohort, observation years 2011–2020 | 45,535 | 4,833 | 330 |
| Final analytic sample (county poverty and percentage Black available) | 45,107 | 4,785 | 325 |
Note. Closure is closure in the following year; observation years 2011–2020, so closures are ascertained through 2021.
Table 2.
Characteristics of hospital-years, overall and by rural location.
| Characteristic | All hospital-years | Rural | Urban |
| County poverty, % (mean, SD) | 14.5 (5.5) | 15.6 (6.1) | 13.5 (4.7) |
| County % Black (mean, SD) | 10.6 (13.6) | 7.4 (13.5) | 13.5 (13.1) |
| Beds (median) | 74.0 | 25.0 | 166.0 |
| Occupancy, % (median) | 45.4 | 33.3 | 57.0 |
| Staff per bed (median) | 5.7 | 5.7 | 5.7 |
| Total margin, % (median) | 4.1 | 3.1 | 5.0 |
| Operating margin, % (median) | −2.3 | −4.9 | −0.1 |
| Days cash (median) | 20.9 | 31.3 | 13.8 |
| Current ratio (median) | 2.0 | 2.1 | 1.9 |
| Leverage, % (median) | 44.0 | 42.7 | 44.8 |
| Medicare share of days, % (median) | 42.4 | 53.3 | 35.4 |
| Medicaid share of charges, % (median) | 13.6 | 13.7 | 13.6 |
| Charity care, % of costs (median) | 1.7 | 1.5 | 1.9 |
| Bad debt, % of costs (median) | 2.2 | 2.7 | 1.9 |
| Critical access, % | 28.9 | 56.0 | 4.5 |
| Teaching, % | 24.9 | 7.4 | 40.5 |
| For-profit, % | 19.2 | 11.3 | 26.2 |
| Government, % | 23.0 | 34.2 | 12.9 |
| Closure in following year, per 1,000 hospital-years | 7.2 | 5.5 | 8.7 |
Note. Hospital-year level; medians for hospital characteristics (the median is shown for each mediator), mean and standard deviation for county variables.
Table 3.
Total, direct, and indirect associations of county poverty, county percentage Black, and rurality with closure.
Table 3.
Total, direct, and indirect associations of county poverty, county percentage Black, and rurality with closure.
| Exposure | Total, per 1,000 (95% CI) | Direct, per 1,000 | Indirect, per 1,000 | Proportion mediated | Total OR | Direct OR |
| County poverty (per SD) | 1.64 (0.72, 2.60) | 0.68 (−0.34, 1.67) | 0.96 (0.50, 1.48) | 59% (28%, 132%) | 1.26 (1.12, 1.41) | 1.17 (1.01, 1.33) |
| County % Black (per SD) | 0.74 (−0.27, 1.90) | 1.58 (0.52, 2.74) | −0.83 (−1.31, −0.37) | −113% (−1562%, 961%) | 1.05 (0.95, 1.16) | 1.12 (1.01, 1.27) |
| Rural (vs urban) | −2.10 (−4.51, 0.20) | −2.10 (−4.53, 0.04) | −0.00 (−1.07, 1.00) | 0% (−103%, 107%) | 0.74 (0.51, 1.03) | 0.80 (0.56, 1.11) |
Note. Risk differences in closures per 1,000 hospital-years in the following year, from a linear probability model with 16 mediators, year fixed effects, region, Medicaid expansion, and county population; the three exposures were entered together. Odds ratios (OR) are from the corresponding logistic models. Confidence intervals are from 400 hospital-clustered bootstrap draws.
Table 4.
Paths from the exposures to the mediators (a) and from the mediators to closure (b).
| Mediator | Poverty → mediator | % Black → mediator | Rural → mediator | Mediator → closure, per 1,000 | Mediator → closure, OR |
| Beds (log) | 0.03 (0.01, 0.05) | 0.06 (0.04, 0.09) | −0.34 (−0.39, −0.28) | −6.84 (−8.83, −4.89) | 0.57 (0.46, 0.67) |
| Occupancy | −0.06 (−0.09, −0.03) | 0.11 (0.08, 0.14) | −0.19 (−0.25, −0.13) | −3.83 (−5.15, −2.61) | 0.57 (0.47, 0.70) |
| Staff per bed | −0.01 (−0.05, 0.02) | −0.01 (−0.04, 0.01) | 0.21 (0.15, 0.28) | −1.33 (−2.20, −0.48) | 0.82 (0.69, 0.93) |
| Total margin | −0.05 (−0.08, −0.03) | −0.04 (−0.06, −0.01) | −0.14 (−0.21, −0.09) | −4.43 (−6.18, −2.70) | 0.75 (0.65, 0.87) |
| Operating margin | −0.03 (−0.06, −0.00) | −0.04 (−0.07, −0.01) | −0.16 (−0.22, −0.10) | −1.99 (−3.52, −0.41) | 0.93 (0.82, 1.08) |
| Days cash | −0.07 (−0.09, −0.04) | 0.01 (−0.02, 0.03) | 0.11 (0.06, 0.17) | −0.12 (−0.84, 0.56) | 0.82 (0.61, 1.01) |
| Current ratio | −0.06 (−0.09, −0.04) | 0.02 (−0.01, 0.04) | 0.03 (−0.02, 0.08) | 0.12 (−0.59, 0.89) | 0.95 (0.79, 1.08) |
| Leverage | 0.02 (−0.01, 0.05) | 0.06 (0.02, 0.09) | 0.03 (−0.03, 0.09) | 1.74 (0.50, 3.11) | 1.15 (1.03, 1.32) |
| Medicare share of days | −0.07 (−0.10, −0.04) | −0.00 (−0.03, 0.02) | 0.35 (0.30, 0.40) | −0.40 (−1.60, 0.80) | 1.09 (0.94, 1.27) |
| Medicaid share of charges | 0.26 (0.23, 0.30) | −0.01 (−0.04, 0.02) | 0.10 (0.05, 0.15) | −0.13 (−1.24, 0.97) | 1.05 (0.93, 1.18) |
| Charity care | 0.02 (−0.00, 0.05) | −0.02 (−0.05, 0.01) | −0.07 (−0.12, −0.02) | −0.31 (−1.18, 0.68) | 1.00 (0.87, 1.11) |
| Bad debt | 0.07 (0.04, 0.09) | 0.07 (0.03, 0.10) | 0.23 (0.17, 0.28) | −0.73 (−1.90, 0.47) | 1.02 (0.89, 1.15) |
| Critical access | −0.05 (−0.06, −0.03) | 0.01 (−0.01, 0.02) | 0.24 (0.21, 0.27) | −10.51 (−14.18, −7.26) | 0.29 (0.18, 0.42) |
| Teaching | 0.02 (0.01, 0.03) | 0.06 (0.04, 0.07) | −0.10 (−0.13, −0.08) | 1.90 (−0.37, 4.21) | 1.12 (0.78, 1.57) |
| For-profit | 0.02 (0.01, 0.04) | −0.04 (−0.06, −0.03) | −0.08 (−0.11, −0.06) | 6.71 (3.87, 9.29) | 1.37 (1.02, 1.80) |
| Government | 0.01 (−0.01, 0.02) | 0.05 (0.03, 0.06) | 0.06 (0.04, 0.09) | −1.89 (−3.86, 0.19) | 0.88 (0.59, 1.30) |
Note. The a-paths are the change in the mediator (standard deviations; change in probability for binary mediators) per standard deviation of poverty or percentage Black, or for rural versus urban location. The b-paths are the change in closures per 1,000 hospital-years per standard deviation of the mediator (or for the binary mediator), and the corresponding odds ratio from the logistic model. 95% bootstrap confidence intervals.
Table 5.
Indirect associations of the exposures with closure through each mediator, per 1,000 hospital-years.
Table 5.
Indirect associations of the exposures with closure through each mediator, per 1,000 hospital-years.
| Mediator | County poverty | County % Black | Rural (vs urban) |
| Beds (log) | −0.22 (−0.42, −0.06) | −0.41 (−0.67, −0.23) | 2.31 (1.59, 3.17) |
| Occupancy | 0.23 (0.12, 0.37) | −0.42 (−0.62, −0.26) | 0.71 (0.42, 1.05) |
| Staff per bed | 0.02 (−0.03, 0.07) | 0.02 (−0.02, 0.07) | −0.28 (−0.49, −0.10) |
| Total margin | 0.24 (0.12, 0.40) | 0.17 (0.05, 0.33) | 0.63 (0.32, 1.06) |
| Operating margin | 0.06 (0.00, 0.15) | 0.08 (0.01, 0.16) | 0.32 (0.05, 0.67) |
| Days cash | 0.01 (−0.04, 0.05) | −0.00 (−0.01, 0.01) | −0.01 (−0.11, 0.07) |
| Current ratio | −0.01 (−0.05, 0.04) | 0.00 (−0.01, 0.02) | 0.00 (−0.03, 0.04) |
| Leverage | 0.04 (−0.02, 0.11) | 0.10 (0.02, 0.22) | 0.05 (−0.04, 0.17) |
| Medicare share of days | 0.03 (−0.06, 0.12) | 0.00 (−0.01, 0.03) | −0.14 (−0.58, 0.29) |
| Medicaid share of charges | −0.03 (−0.33, 0.27) | 0.00 (−0.02, 0.03) | −0.01 (−0.13, 0.12) |
| Charity care | −0.01 (−0.04, 0.02) | 0.01 (−0.02, 0.03) | 0.02 (−0.05, 0.09) |
| Bad debt | −0.05 (−0.13, 0.03) | −0.05 (−0.14, 0.03) | −0.17 (−0.46, 0.11) |
| Critical access | 0.48 (0.30, 0.70) | −0.07 (−0.24, 0.05) | −2.57 (−3.49, −1.75) |
| Teaching | 0.04 (−0.00, 0.09) | 0.11 (−0.02, 0.24) | −0.20 (−0.45, 0.04) |
| For-profit | 0.16 (0.06, 0.27) | −0.28 (−0.44, −0.14) | −0.56 (−0.86, −0.30) |
| Government | −0.01 (−0.04, 0.01) | −0.09 (−0.18, 0.01) | −0.11 (−0.25, 0.01) |
| All mediators (total indirect) | 0.96 (0.50, 1.48) | −0.83 (−1.31, −0.37) | −0.00 (−1.07, 1.00) |
Note. Product of the a-path and b-path (Table 4), risk difference per 1,000 hospital-years in the following year; 95% bootstrap confidence intervals. A positive value means that the pathway raises closure risk. Individual indirect associations are conditional on the other mediators.
Table 6.
Indirect associations through groups of mediators, per 1,000 hospital-years.
| Group of mediators | County poverty | County % Black | Rural (vs urban) |
| Size | −0.22 (−0.42, −0.06) | −0.41 (−0.67, −0.23) | 2.31 (1.59, 3.17) |
| Utilization and staffing | 0.24 (0.12, 0.39) | −0.40 (−0.61, −0.21) | 0.44 (0.03, 0.87) |
| Performance | 0.30 (0.15, 0.49) | 0.25 (0.09, 0.43) | 0.95 (0.56, 1.45) |
| Assets | 0.04 (−0.04, 0.12) | 0.10 (0.02, 0.23) | 0.04 (−0.09, 0.19) |
| Payer mix | −0.01 (−0.30, 0.29) | 0.00 (−0.03, 0.03) | −0.16 (−0.61, 0.28) |
| Uncompensated care | −0.06 (−0.15, 0.03) | −0.04 (−0.13, 0.04) | −0.15 (−0.45, 0.13) |
| Structure | 0.51 (0.32, 0.74) | 0.03 (−0.18, 0.20) | −2.77 (−3.72, −1.89) |
| Ownership | 0.15 (0.04, 0.26) | −0.37 (−0.57, −0.18) | −0.68 (−1.02, −0.37) |
Note. Sum of the indirect associations of the mediators in each group (95% bootstrap confidence intervals); groups are defined in Section 2.4.
Table 7.
Sensitivity analyses.
| Analysis | Exposure | Total (95% CI) | Direct | Indirect | Proportion mediated |
| Main model | County poverty | 1.64 (0.72, 2.60) | 0.68 | 0.96 | 59% |
| County % Black | 0.74 (−0.27, 1.90) | 1.58 | −0.83 | −113% | |
| Rural | −2.10 (−4.51, 0.20) | −2.10 | −0.00 | 0% | |
| Without ownership mediators | County poverty | 1.64 (0.86, 2.60) | 0.76 | 0.88 | 54% |
| County % Black | 0.74 (−0.36, 2.03) | 1.34 | −0.60 | −81% | |
| Rural | −2.10 (−3.75, 0.02) | −2.50 | 0.40 | −19% | |
| Without county population covariate | County poverty | 1.60 (0.86, 2.48) | 0.39 | 1.21 | 76% |
| County % Black | 0.87 (−0.07, 2.09) | 2.01 | −1.14 | −131% | |
| Rural | −2.96 (−4.53, −1.33) | −4.39 | 1.42 | −48% | |
| Each exposure alone | County poverty | 1.77 (1.07, 2.78) | 1.10 | 0.67 | 38% |
| County % Black | 1.48 (0.47, 2.41) | 1.87 | −0.39 | −26% | |
| Rural | −1.43 (−3.44, 0.64) | −1.89 | 0.46 | −32% |
Note. Risk differences in closures per 1,000 hospital-years; sensitivity models used 100 bootstrap draws.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2026 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license.
Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.