Beyond Clinical Acuity: Cardiac Comorbidity and Complication Profiles and the Prehospital Hospitalization/Transport Decision in Ischemic Heart Disease — A Five-Year Retrospective Emergency Medical Service Study in Astana, Kazakhstan

1. Introduction

Ischemic heart disease (IHD) is the single largest contributor to global cardiovascular mortality. The most recent Global Burden of Disease estimates attribute roughly 19.2 million deaths to cardiovascular disease in 2023, with IHD remaining the leading cardiovascular cause of death and of disability-adjusted life-years worldwide [1,2]. The absolute burden continues to grow as populations age and cardiometabolic risk factors accumulate, so that health systems must now manage an ever-expanding pool of patients living with both chronic and acute coronary disease [3,4]. The burden is unevenly distributed: age-standardized rates are highest in low- and middle-income and transitioning settings, and registries from these regions remain comparatively sparse [4,5]. In Kazakhstan and the wider Central Asian region, cardiovascular disease dominates the mortality profile and risk-factor control remains incomplete [6]. In this context the emergency medical service (EMS) is frequently the first - and sometimes the only - point of medical contact for patients with cardiac symptoms, and the ambulance crew must decide at the scene whether a patient with IHD requires emergency hospitalization or can safely be left at home. This conveyance decision is consequential for the patient and for a stretched emergency system, yet the clinical factors that drive it in routine IHD care have rarely been quantified.

For time-sensitive coronary events, outcomes are tightly coupled to the speed and appropriateness of the prehospital response, and a large body of work has used routinely collected data to identify the factors that delay or shape that response. A recurring methodological theme is the use of multivariable logistic regression on registry or chart data to derive adjusted odds ratios for delay or for adverse outcomes. Khaled and colleagues studied patients across 27 hospitals in Bangladesh and used multivariable logistic regression to show that older age, female sex, rural residence and ignorance of symptoms were significant determinants of prolonged prehospital delay in acute coronary syndrome (ACS) [7]. In a single-center ST-elevation myocardial infarction (STEMI) cohort, Ohtaroglu Tokdil et al. applied univariate and multivariable logistic regression and found that body-mass index, diabetes, hypertension, smoking and pain intensity independently predicted longer delay times, which in turn were linked to adverse clinical end points [8]. At the system level, Chang et al. analyzed 142,474 acute myocardial infarction (AMI) events (ICD-10 I21-I22) from the Beijing surveillance system, combining geographic driving-time estimates with logistic regression to demonstrate that each additional band of travel time to a percutaneous-coronary-intervention (PCI)-capable hospital increased case-fatality odds in graded fashion [9]. Collectively, these studies establish multivariable logistic regression on routinely collected registry data, reporting adjusted odds ratios with confidence intervals, as the analytical standard for identifying prehospital risk factors in coronary disease - and directly motivate the modeling approach adopted here.

A second, fast-moving literature focuses on the prehospital decision itself - whether and with what priority a patient should be conveyed - and increasingly compares classical regression with machine-learning techniques. Al-Zaiti et al. applied machine-learning classifiers to the prehospital 12-lead electrocardiogram and improved field detection of ACS, evaluating performance with receiver-operating-characteristic (ROC) analysis and the area under the curve (AUC) [10]. Wibring et al. and Spangler et al. developed logistic-regression-based triage and dispatch models for chest pain, reporting AUCs of roughly 0.74-0.79 and showing that variables routinely available to EMS can separate high-risk from low-risk presentations [11,12]. Beyond regression, random-forest and gradient-boosting models have been used to predict early mortality and to flag adverse events after EMS non-conveyance decisions, generally matching or modestly exceeding logistic regression while being harder to interpret [13,14]. A parallel patient-safety literature has characterized non-conveyance itself: a systematic review reported non-conveyance rates ranging widely across systems and documented measurable rates of re-contact and short-term adverse events, underscoring that both over-triage and under-triage carry risk [15,16]. These methods - logistic regression, discrimination metrics, registry-based modeling and risk-stratification scores - frame the present analysis.

A third strand links the comorbidities recorded by EMS to downstream hospitalization and outcome. Comorbidity-based risk scores such as the DICER-score have been built specifically to stratify emergency-department demand using drugs, income and comorbidities [17], and population studies show that multimorbidity is strongly associated with emergency admission and 30-day mortality, with effects that are proportionally largest in younger patients [18]. For the specific conditions captured in our data, the evidence is consistent and recent. Heart failure is a dominant driver of emergency presentation and admission, and both regression- and machine-learning-based models have been developed to predict heart-failure ED visits and hospitalization [19,20,21,22]. Atrial fibrillation and other arrhythmias carry high admission rates that rise further in the presence of concomitant heart failure and multimorbidity [23,24]. Cardiogenic shock, though rare, is the archetypal high-acuity complication: contemporary staging systems and registries report in-hospital mortality of the order of 20-40%, and dedicated scores have been proposed to recognize and stage it early, including in the prehospital phase [25,26,27,28]. Finally, demographic effects are well documented but context-dependent: sex differences in symptom presentation and in delayed hospitalization are established for ACS and NSTEMI, and rural-urban differences shape invasive management and outcomes [29,30,31].

Despite this progress, an important gap remains. Most prehospital studies concentrate on the acute, time-critical presentation (STEMI/NSTEMI) and on the timing of care, whereas the majority of EMS calls for IHD in routine practice involve chronic, stable coronary disease against a background of cardiac and non-cardiac comorbidity. It is not well established to what extent the comorbidity and complication profile recorded by the ambulance crew - heart failure, atrial fibrillation, other arrhythmias, cardiogenic shock and related conditions - independently drives the field hospitalization decision, over and above the acuity of the index event. Moreover, prior work rarely models the binary at-scene outcome that crews actually face (hospitalize versus leave at scene), tending instead to model delay, in-hospital mortality or downstream readmission [7,8,9,19,20]. Evidence from Central Asian urban EMS systems is particularly scarce [5,6], limiting the external validity of risk-stratification tools developed elsewhere and leaving local services without quantified, data-driven support for conveyance decisions.

To address this gap, we analyzed five years of EMS calls for IHD in Astana, Kazakhstan. Building on the methodological standard established by the studies above, we adopted multivariable logistic regression on routinely collected dispatch data. The specific objectives of this study were: (i) to describe the demographic, clinical and comorbidity characteristics of EMS calls for IHD according to call outcome; (ii) to identify, using multivariable logistic regression, the independent predictors of emergency hospitalization, expressed as adjusted odds ratios with 95% confidence intervals; and (iii) to quantify the relative contribution of comorbidity and complication profiles compared with clinical acuity, sex, age and operational EMS factors. By isolating routinely available predictors, we aim to inform field risk-stratification and EMS resource planning in urban Central Asian settings.

2. Materials and Methods

2.1. Study Design and Setting

We performed a retrospective, observational cohort study of emergency medical service (EMS) calls attended for ischemic heart disease in Astana, the capital of Kazakhstan, over an approximately five-year period from 19 February 2020 to 30 June 2024. The final year (2024) therefore covers the first half of the calendar year only, which should be borne in mind when interpreting annual call volumes. Astana is served by a centralized municipal ambulance dispatch system that records each call electronically, including dispatch priority, crew type, response and on-scene time intervals, working diagnosis (coded to ICD-10) and final disposition. The study followed the principles of the Declaration of Helsinki, and only de-identified routinely collected operational data were analyzed. Reporting adheres to the STROBE recommendations for observational studies.

2.2. Participants and Case Definition

Eligible records were all EMS calls with a working diagnosis of ischemic heart disease, defined by ICD-10 codes I20-I25. In keeping with registry-based AMI studies that identify cases through principal discharge or working diagnoses [9], calls were classified into two clinical forms: acute/unstable IHD (unstable angina and acute coronary syndromes, including I20-I24) and chronic IHD (I25). Comorbidities and complications documented by the crew were captured as binary indicators using their respective ICD-10 codes: diabetes mellitus (E10-E14), chronic obstructive pulmonary disease and related obstructive disease (J40-J47), arterial hypertension (I10-I15), heart failure (I50), atrial fibrillation (I48), other arrhythmias (I47, I49), cardiogenic shock (R57.0) and pneumonia (J12-J18). After removal of 2080 duplicate call records from the 12,304 raw IHD-coded dispatch records and exclusion of 239 records with a final disposition of death at scene or handover that did not permit a hospitalize-versus-leave classification, 9985 calls were available for descriptive comparison and 9860 for the fully adjusted regression model. The complete data-processing and cleaning cascade is summarized in Figure 1. The unit of analysis was the individual EMS call rather than the unique patient: each dispatch record represents one ambulance activation and the conveyance decision taken at that encounter. Because the dispatch dataset did not contain a stable patient identifier that could be reliably linked across calls, repeat calls by the same patient could not be collapsed to the patient level, and the analysis therefore treats calls as the sampling unit (see Limitations).

2.3. Outcome and Variables

The primary outcome was emergency hospitalization (coded 1) versus being left at the scene (coded 0). This binary at-scene disposition reflects the decision the ambulance crew makes in routine practice and mirrors the conveyance/non-conveyance outcomes modeled in prior prehospital studies [12,14,15]. Candidate predictors, selected a priori on clinical and operational grounds, comprised: age (modeled both continuously and in categories <45, 45-59, 60-74 and ≥75 years), sex, clinical form of IHD (acute/unstable vs chronic I25), the eight comorbidity/complication indicators listed above, the presence of any documented complication (DS3), dispatch urgency category (1-4, dichotomized into high urgency [categories 1-2] for the model), crew type (specialized vs general/line) and EMS response time (modeled per additional 10 minutes). Two operational time intervals - call-to-arrival and arrival-to-hospital-delivery - were summarized descriptively.

2.4. Statistical Analysis

Continuous variables were summarized as median (interquartile range, IQR) and compared between hospitalized and left-at-scene groups with the Mann-Whitney U test, given their skewed distributions. Categorical variables were summarized as counts and percentages and compared with the Pearson chi-square test. Following the analytical approach established in comparable registry studies of coronary disease [7,8,9], independent associations with hospitalization were estimated with an explanatory multivariable logistic regression model entering all pre-specified covariates simultaneously. The model was specified to quantify independent associations (explanatory aim) rather than to derive a deployable prediction rule. Results are reported as adjusted odds ratios (aOR) with 95% confidence intervals (CI). Overall model significance was assessed with the log-likelihood-ratio test, and explained variation with McFadden's pseudo-R². Model performance was characterized in terms of discrimination (area under the receiver-operating-characteristic curve, AUC), overall accuracy (Brier score, together with a scaled Brier score relative to a prevalence-only model) and calibration (calibration intercept and slope, with a calibration plot of observed versus predicted probability across deciles of predicted risk). A 95% confidence interval for the AUC was obtained by percentile bootstrap (2000 resamples), and internal validity was assessed by bootstrap internal validation (500 resamples, Harrell's optimism-correction procedure) to estimate the optimism-corrected AUC. A two-sided p-value < 0.05 was considered statistically significant. Adjusted odds ratios were visualized as a forest plot on a logarithmic scale, and the relationship between response time and the cumulative probability of hospitalization was displayed separately by clinical form of IHD. Analyses were performed in Python using the statsmodels, scikit-learn and pandas libraries. Reporting followed the STROBE statement [37]. The flow of records from the full set of EMS dispatch entries to the final analytical samples, including each exclusion step and the corresponding numbers, is summarized in Figure 1.

Figure 1. Data-processing and cleaning flow from the raw EMS dispatch dataset to the final analytical samples. Starting from 12,304 raw EMS dispatch records coded as ischemic heart disease (ICD-10 I20–I25) over the five-year study period, 2080 duplicate call records were removed, leaving 10,224 unique call records. A further 239 calls with an ambiguous final disposition (death at scene, handover, or any disposition that did not permit a hospitalize-versus-leave classification) were excluded, yielding the descriptive analytical sample of 9985 calls (2676 hospitalized/transported [26.8%] and 7,309 left at scene/ambulatory [73.2%]). Finally, 125 calls with missing values in one or more pre-specified model covariates were excluded to give the complete-case sample of 9860 calls used in the multivariable logistic regression model.

Figure 1. Data-processing and cleaning flow from the raw EMS dispatch dataset to the final analytical samples. Starting from 12,304 raw EMS dispatch records coded as ischemic heart disease (ICD-10 I20–I25) over the five-year study period, 2080 duplicate call records were removed, leaving 10,224 unique call records. A further 239 calls with an ambiguous final disposition (death at scene, handover, or any disposition that did not permit a hospitalize-versus-leave classification) were excluded, yielding the descriptive analytical sample of 9985 calls (2676 hospitalized/transported [26.8%] and 7,309 left at scene/ambulatory [73.2%]). Finally, 125 calls with missing values in one or more pre-specified model covariates were excluded to give the complete-case sample of 9860 calls used in the multivariable logistic regression model.

3. Results

3.1. Analytical Sample and Overall Disposition

After duplicate removal and exclusion of calls without an unambiguous final disposition, the analytical sample comprised 9985 EMS calls for ischemic heart disease (IHD). Of 12,304 raw EMS dispatch records coded to ICD-10 I20–I25, 2080 duplicate call records were removed to leave 10,224 unique calls, and a further 239 calls with an ambiguous disposition were excluded; the full cleaning cascade is shown in Figure 1. Overall, 2676 calls (26.8%) resulted in hospitalization/transport to hospital, whereas 7,309 calls (73.2%) were managed at the scene or resulted in ambulatory care. This distribution indicates that most IHD-coded EMS activations did not lead to hospital transfer, supporting the use of disposition as a clinically and operationally relevant outcome.

3.2. Demographic, Clinical, Comorbidity and Operational Characteristics

Baseline characteristics by call outcome are presented in Table 1. The median age of the study population was 68 years (IQR 59-77), and women accounted for 55.1% of calls. Compared with patients left at the scene, hospitalized/transported patients were younger (median 65 years, IQR 56-73, vs. 70 years, IQR 60-79; p < 0.001) and more frequently male (56.1% vs. 40.8%; p < 0.001). The age gradient was most pronounced among patients aged ≥75 years, who represented 22.3% of hospitalized/transported calls but 34.4% of calls left at the scene.

Chronic IHD (ICD-10 I25) accounted for most EMS calls (95.8%), while acute/unstable forms of IHD accounted for 4.2% of the sample. Nevertheless, the clinical form differed markedly by disposition: acute/unstable IHD represented 11.3% of hospitalized/transported calls but only 1.7% of calls left at the scene (p < 0.001). Urgency category was also associated with disposition (p < 0.001), whereas crew type did not differ significantly in the univariable comparison (p = 0.593).

The comorbidity and complication profile was dominated by cardiac conditions. Heart failure was the most common coded comorbidity/complication (30.8% overall) and was more frequent among hospitalized/transported patients than among those left at the scene (37.8% vs. 28.2%; p < 0.001). Other arrhythmias were also more frequent in the hospitalized/transported group (25.0% vs. 21.0%; p < 0.001), and cardiogenic shock, although uncommon overall (1.0%), was strongly concentrated among hospitalized/transported calls (3.3% vs. 0.2%; p < 0.001). In contrast, diabetes mellitus, COPD/obstructive disease, arterial hypertension, atrial fibrillation and pneumonia did not differ significantly between the outcome groups in unadjusted comparisons. Median response and delivery intervals were similar between groups.

Table 1. Demographic, clinical, comorbidity and operational characteristics by EMS call outcome.

Characteristic	Total	Hospitalized/ transported	Left at scene/ ambulatory	p-value	Test
n (% of sample)	9985	2676 (26.8%)	7309 (73.2%)
Age, median (IQR), years	68 (59-77)	65 (56-73)	70 (60-79)	<0.001	Mann-Whitney U
Sex, n (%)				<0.001	Pearson chi-square
Male	4484 (44.9%)	1502 (56.1%)	2982 (40.8%)
Female	5501 (55.1%)	1174 (43.9%)	4327 (59.2%)
Age group, n (%)				<0.001	Pearson chi-square
<45	436 (4.4%)	193 (7.2%)	243 (3.3%)
45–59	2205 (22.1%)	719 (26.9%)	1486 (20.3%)
60–74	4231 (42.4%)	1166 (43.6%)	3065 (41.9%)
≥75	3113 (31.2%)	598 (22.3%)	2515 (34.4%)
Urgency category, n (%)				<0.001	Pearson chi-square
Category 1	186 (1.9%)	65 (2.4%)	121 (1.7%)
Category 2	8091 (81.0%)	2106 (78.7%)	5985 (81.9%)
Category 3	1444 (14.5%)	452 (16.9%)	992 (13.6%)
Category 4	264 (2.6%)	53 (2.0%)	211 (2.9%)
Main diagnosis, n (%)				<0.001	Pearson chi-square
Acute/unstable IHD	423 (4.2%)	302 (11.3%)	121 (1.7%)
Chronic IHD (I25)	9562 (95.8%)	2374 (88.7%)	7188 (98.3%)
Crew type, n (%)				0.593	Pearson chi-square
Specialized	1074 (10.8%)	280 (10.5%)	794 (10.9%)
General/line	8911 (89.2%)	2396 (89.5%)	6515 (89.1%)
Comorbidities/complications, n (%)
Diabetes mellitus (E10-E14)	83 (0.8%)	23 (0.9%)	60 (0.8%)	0.949	Pearson chi-square
COPD/obstructive disease (J40-J47)	109 (1.1%)	29 (1.1%)	80 (1.1%)	1.000	Pearson chi-square
Arterial hypertension (I10-I15)	1118 (11.2%)	279 (10.4%)	839 (11.5%)	0.149	Pearson chi-square
Heart failure (I50)	3071 (30.8%)	1011 (37.8%)	2060 (28.2%)	<0.001	Pearson chi-square
Atrial fibrillation (I48)	2043 (20.5%)	555 (20.7%)	1488 (20.4%)	0.696	Pearson chi-square
Other arrhythmias (I47, I49)	2204 (22.1%)	669 (25.0%)	1535 (21.0%)	<0.001	Pearson chi-square
Cardiogenic shock (R57.0)	99 (1.0%)	88 (3.3%)	11 (0.2%)	<0.001	Pearson chi-square
Pneumonia (J12-J18)	106 (1.1%)	32 (1.2%)	74 (1.0%)	0.495	Pearson chi-square
Time intervals, median (IQR), min
Call to crew arrival	14 (11-16)	14 (11-16)	14 (11-16)	0.589	Mann-Whitney U
Arrival to hospital delivery	62 (52-74)	62 (52-74)	63 (47-78)	0.694	Mann-Whitney U

Values are n (%) unless otherwise indicated. Continuous variables are summarized as median (interquartile range, IQR) and compared using the Mann-Whitney U test. Categorical variables are compared using the Pearson chi-square test. Exact p-values are shown when p ≥0.001; values below 0.001 are reported as p < 0.001. Calls with death before arrival, handover or another ambiguous disposition were excluded from outcome comparisons. COPD: chronic obstructive pulmonary disease; EMS: emergency medical service; IHD: ischemic heart disease.

Table 1. Demographic, clinical, comorbidity and operational characteristics by EMS call outcome.

Characteristic	Total	Hospitalized/ transported	Left at scene/ ambulatory	p-value	Test
n (% of sample)	9985	2676 (26.8%)	7309 (73.2%)
Age, median (IQR), years	68 (59-77)	65 (56-73)	70 (60-79)	<0.001	Mann-Whitney U
Sex, n (%)				<0.001	Pearson chi-square
Male	4484 (44.9%)	1502 (56.1%)	2982 (40.8%)
Female	5501 (55.1%)	1174 (43.9%)	4327 (59.2%)
Age group, n (%)				<0.001	Pearson chi-square
<45	436 (4.4%)	193 (7.2%)	243 (3.3%)
45–59	2205 (22.1%)	719 (26.9%)	1486 (20.3%)
60–74	4231 (42.4%)	1166 (43.6%)	3065 (41.9%)
≥75	3113 (31.2%)	598 (22.3%)	2515 (34.4%)
Urgency category, n (%)				<0.001	Pearson chi-square
Category 1	186 (1.9%)	65 (2.4%)	121 (1.7%)
Category 2	8091 (81.0%)	2106 (78.7%)	5985 (81.9%)
Category 3	1444 (14.5%)	452 (16.9%)	992 (13.6%)
Category 4	264 (2.6%)	53 (2.0%)	211 (2.9%)
Main diagnosis, n (%)				<0.001	Pearson chi-square
Acute/unstable IHD	423 (4.2%)	302 (11.3%)	121 (1.7%)
Chronic IHD (I25)	9562 (95.8%)	2374 (88.7%)	7188 (98.3%)
Crew type, n (%)				0.593	Pearson chi-square
Specialized	1074 (10.8%)	280 (10.5%)	794 (10.9%)
General/line	8911 (89.2%)	2396 (89.5%)	6515 (89.1%)
Comorbidities/complications, n (%)
Diabetes mellitus (E10-E14)	83 (0.8%)	23 (0.9%)	60 (0.8%)	0.949	Pearson chi-square
COPD/obstructive disease (J40-J47)	109 (1.1%)	29 (1.1%)	80 (1.1%)	1.000	Pearson chi-square
Arterial hypertension (I10-I15)	1118 (11.2%)	279 (10.4%)	839 (11.5%)	0.149	Pearson chi-square
Heart failure (I50)	3071 (30.8%)	1011 (37.8%)	2060 (28.2%)	<0.001	Pearson chi-square
Atrial fibrillation (I48)	2043 (20.5%)	555 (20.7%)	1488 (20.4%)	0.696	Pearson chi-square
Other arrhythmias (I47, I49)	2204 (22.1%)	669 (25.0%)	1535 (21.0%)	<0.001	Pearson chi-square
Cardiogenic shock (R57.0)	99 (1.0%)	88 (3.3%)	11 (0.2%)	<0.001	Pearson chi-square
Pneumonia (J12-J18)	106 (1.1%)	32 (1.2%)	74 (1.0%)	0.495	Pearson chi-square
Time intervals, median (IQR), min
Call to crew arrival	14 (11-16)	14 (11-16)	14 (11-16)	0.589	Mann-Whitney U
Arrival to hospital delivery	62 (52-74)	62 (52-74)	63 (47-78)	0.694	Mann-Whitney U

Values are n (%) unless otherwise indicated. Continuous variables are summarized as median (interquartile range, IQR) and compared using the Mann-Whitney U test. Categorical variables are compared using the Pearson chi-square test. Exact p-values are shown when p ≥0.001; values below 0.001 are reported as p < 0.001. Calls with death before arrival, handover or another ambiguous disposition were excluded from outcome comparisons. COPD: chronic obstructive pulmonary disease; EMS: emergency medical service; IHD: ischemic heart disease.

3.3. Multivariable Predictors of Emergency Hospitalization/Transport

The multivariable logistic regression model was statistically significant (likelihood-ratio test p < 0.001; McFadden pseudo-R2 = 0.093). Adjusted odds ratios (aORs) are shown graphically in Figure 2 and reported numerically in Table 2. The model identified both acuity-related variables and comorbidity-related variables as independent predictors of EMS disposition.

Cardiogenic shock was the strongest independent predictor of hospitalization/transport (aOR 15.06; 95% CI 7.79-29.13; p < 0.001), followed by acute/unstable IHD compared with chronic I25-coded disease (aOR 8.52; 95% CI 6.74-10.76; p < 0.001). Among cardiac comorbidity indicators, heart failure (aOR 2.46; 95% CI 2.17-2.79), other arrhythmias (aOR 1.84; 95% CI 1.62-2.10) and atrial fibrillation (aOR 1.60; 95% CI 1.40-1.82) remained independently associated with higher odds of hospitalization/transport after adjustment for demographic, clinical and operational variables.

Demographic and operational covariates also contributed to the final model. Male sex was associated with increased odds of hospitalization/transport (aOR 1.65; 95% CI 1.49-1.82), as was age <45 years compared with the 60-74-year reference group (aOR 1.88; 95% CI 1.51-2.33). By contrast, age ≥75 years was associated with lower adjusted odds (aOR 0.61; 95% CI 0.54-0.69). Each additional 10 minutes of EMS response time was associated with a modest increase in the adjusted odds of hospitalization/transport (aOR 1.05; 95% CI 1.01-1.09).

Figure 2. Adjusted odds ratios for hospitalization/transport among EMS calls for ischemic heart disease. Points represent aORs and horizontal bars represent 95% confidence intervals. The vertical dashed line denotes OR = 1; estimates to the right indicate higher adjusted odds and estimates to the left indicate lower adjusted odds. *p < 0.05.

Figure 2. Adjusted odds ratios for hospitalization/transport among EMS calls for ischemic heart disease. Points represent aORs and horizontal bars represent 95% confidence intervals. The vertical dashed line denotes OR = 1; estimates to the right indicate higher adjusted odds and estimates to the left indicate lower adjusted odds. *p < 0.05.

Table 2. Multivariable logistic regression model for predictors of hospitalization/transport.

Predictor	aOR	95% CI	p-value
Cardiogenic shock	15.06	7.79-29.13	<0.001
Acute/unstable IHD (vs. chronic I25)	8.52	6.74-10.76	<0.001
Heart failure	2.46	2.17-2.79	<0.001
Age <45 years (vs. 60-74 years)	1.88	1.51-2.33	<0.001
Other arrhythmias	1.84	1.62-2.10	<0.001
Male sex (vs. female)	1.65	1.49-1.82	<0.001
Atrial fibrillation	1.60	1.40-1.82	<0.001
Pneumonia	1.32	0.84-2.08	0.228
COPD	1.24	0.79-1.94	0.352
Age 45-59 years (vs. 60-74 years)	1.18	1.05-1.33	0.006
Arterial hypertension	1.18	0.98-1.41	0.080
Diabetes mellitus	1.12	0.65-1.92	0.681
Complication present (DS3)	1.05	0.93-1.19	0.407
Response time (per +10 min)	1.05	1.01-1.09	0.012
High urgency (category 1-2)	0.84	0.73-0.96	0.012
Specialized crew	0.84	0.71-0.98	0.029
Age ≥75 years (vs. 60-74 years)	0.61	0.54-0.69	<0.001

Dependent variable: hospitalization/transport to hospital = 1. Reference categories: female sex, age 60–74 years, chronic IHD (I25), non-high urgency, general/line crew and absence of the respective comorbidity/complication. Model N = 9860. aOR: adjusted odds ratio; CI: confidence interval; COPD: chronic obstructive pulmonary disease; IHD: ischemic heart disease.

Table 2. Multivariable logistic regression model for predictors of hospitalization/transport.

Predictor	aOR	95% CI	p-value
Cardiogenic shock	15.06	7.79-29.13	<0.001
Acute/unstable IHD (vs. chronic I25)	8.52	6.74-10.76	<0.001
Heart failure	2.46	2.17-2.79	<0.001
Age <45 years (vs. 60-74 years)	1.88	1.51-2.33	<0.001
Other arrhythmias	1.84	1.62-2.10	<0.001
Male sex (vs. female)	1.65	1.49-1.82	<0.001
Atrial fibrillation	1.60	1.40-1.82	<0.001
Pneumonia	1.32	0.84-2.08	0.228
COPD	1.24	0.79-1.94	0.352
Age 45-59 years (vs. 60-74 years)	1.18	1.05-1.33	0.006
Arterial hypertension	1.18	0.98-1.41	0.080
Diabetes mellitus	1.12	0.65-1.92	0.681
Complication present (DS3)	1.05	0.93-1.19	0.407
Response time (per +10 min)	1.05	1.01-1.09	0.012
High urgency (category 1-2)	0.84	0.73-0.96	0.012
Specialized crew	0.84	0.71-0.98	0.029
Age ≥75 years (vs. 60-74 years)	0.61	0.54-0.69	<0.001

Dependent variable: hospitalization/transport to hospital = 1. Reference categories: female sex, age 60–74 years, chronic IHD (I25), non-high urgency, general/line crew and absence of the respective comorbidity/complication. Model N = 9860. aOR: adjusted odds ratio; CI: confidence interval; COPD: chronic obstructive pulmonary disease; IHD: ischemic heart disease.

3.4. Model Performance and Internal Validation

Because the model was specified to explain rather than to predict the at-scene disposition, its performance is reported to describe model fit rather than to support a deployable prediction tool. Discrimination was moderate: the area under the receiver-operating-characteristic curve (AUC) was 0.69 (95% CI 0.68–0.70; Figure 3A). Overall accuracy of the predicted probabilities was modest, with a Brier score of 0.18 and a scaled Brier score of 0.10 relative to a prevalence-only model. Calibration across deciles of predicted risk was close to the identity line over the observed probability range (Figure 3B), with an apparent calibration slope of 1.00 and intercept of 0.00 on the development data. Bootstrap internal validation (500 resamples, Harrell’s optimism-correction procedure) indicated minimal overfitting, with a mean optimism of 0.004 and an optimism-corrected AUC of 0.69. These values are consistent with the modest McFadden pseudo-R² of 0.093 and sit somewhat below the AUCs of roughly 0.74–0.79 reported for dedicated prehospital chest-pain triage models [11,12], reinforcing that the recorded dispatch variables explain only part of the variation in the hospitalization/transport decision and that the predictors identified here should be regarded as candidate variables for future risk-stratification work rather than as a validated prediction instrument.

3.5. Graphical Summary of Clinical Disposition and Comorbidity Burden

The graphical summaries complement the tabular findings by showing how disposition differs by clinical form and comorbidity profile. Acute/unstable IHD calls had a substantially higher hospitalization/transport proportion than chronic IHD calls (71.4% vs. 24.8%; Figure 4A). The hospitalized/transported group also had a higher prevalence of heart failure, other arrhythmias and cardiogenic shock, whereas diabetes mellitus, COPD/obstructive disease and pneumonia were rare in both outcome groups (Figure 4B).

3.6. Age Distribution, Annual Dynamics and Response-Time Pattern

Age and annual patterns are summarized in Figure 5. The median-IQR plot confirms the younger profile of hospitalized/transported patients compared with those left at the scene (Figure 5A), consistent with the values reported in Table 1. Annual call volume varied across the observation period, while the hospitalization/transport rate declined in 2021 and then partially recovered in subsequent years (Figure 5B).

The response-time visualization in Figure 6 shows a steeper cumulative curve for acute/unstable IHD than for chronic IHD, indicating earlier accumulation of hospitalized/transported acute/unstable cases across the EMS response-time scale. This pattern is consistent with the strong adjusted association between acute/unstable IHD and hospitalization/transport observed in the multivariable model.

Overall, the probability of hospitalization/transport among EMS calls for IHD was driven by both the acute clinical form of IHD and specific cardiac comorbidity/complication profiles. Cardiogenic shock and acute/unstable IHD were the dominant acuity-related predictors, while heart failure, other arrhythmias and atrial fibrillation remained independent comorbidity-related predictors after adjustment. Diabetes mellitus and COPD were rare in the coded EMS fields and did not show statistically significant adjusted associations in this dataset, suggesting limited capture of non-cardiac comorbidity in the available DS2/DS3 coding structure.

4. Discussion

In this five-year analysis of nearly 10,000 EMS calls for IHD in Astana, the field decision to hospitalize/transport was governed by a clear hierarchy: clinical acuity first, but stable cardiac comorbidity independently and substantially second. The central contribution of this study is the demonstration that, even in a population overwhelmingly composed of chronic coronary disease, the comorbidity and complication profile documented by the crew shapes the hospitalization/transport decision after adjustment for the acuity of the index event. Below we interpret these findings in the light of the wider literature, rather than restating them, and then reflect critically on the strengths and weaknesses of the data and methods used.

4.1. Acuity, Comorbidity and the Wider Evidence

That hemodynamic instability and acute presentation dominated the model is concordant with the broader coronary literature and with mechanism: cardiogenic shock is the archetypal high-acuity complication, and contemporary staging systems and registries consistently report in-hospital mortality of the order of 20-40%, justifying near-universal escalation of care [25,26,27]. Position statements on AMI complicated by cardiogenic shock likewise frame immediate transfer to definitive care as the default [28]. Our more novel observation concerns stable comorbidity. The independent association of heart failure with hospitalization aligns with a large and recent body of work in which heart failure is a principal driver of emergency presentation, admission and readmission, and in which both regression- and machine-learning-based tools have been built specifically to anticipate heart-failure ED visits and hospitalization [19,20,22]. The independent contributions of atrial fibrillation and other arrhythmias are similarly supported: nationwide and statewide analyses report high admission rates for atrial fibrillation that rise further with concomitant heart failure and multimorbidity [23,24]. More generally, multimorbidity is strongly associated with emergency admission and short-term mortality, and comorbidity-based scores such as the DICER-score have been designed to translate exactly this signal into operational risk stratification [17,18]. Our findings extend this evidence to the prehospital hospitalization/transport decision in an under-studied Central Asian setting, suggesting that the comorbidity information crews already record carries decision-relevant prognostic weight.

4.2. Response Time, Demography and the Crew-Side Decision

The association between longer response time and higher hospitalization/transport odds parallels the Beijing driving-time study, in which longer travel to PCI-capable hospitals raised case-fatality odds in graded fashion [9]; in both settings, time behaves partly as a proxy for case severity and geographic remoteness rather than as a simple causal exposure. This interpretation is also consistent with local GIS-based evidence from Astana showing that urban environmental and spatiotemporal factors shape EMS response patterns [32], while telemedicine- and system-delay studies have similarly emphasized the importance of operational delay [33]. The demographic pattern we observed merits particular comment. Studies of patient-driven prehospital delay generally find that older age and female sex are associated with later presentation and poorer access [7,29,30], whereas our crew-side outcome showed younger and male patients more likely to be conveyed and the oldest patients conveyed less often. The divergence is informative rather than contradictory: delay studies capture help-seeking behavior, while our outcome captures the hospitalization/transport decision made once the crew is on scene. Lower hospitalization/transport odds among patients aged ≥75 years may reflect a higher prevalence of chronic stable disease managed at home, ceilings-of-care considerations, frailty-related decisions or patient preference, and should be interpreted cautiously rather than read as simple under-triage - particularly because multimorbidity confers proportionally greater risk in younger patients [18,24]. Sex differences in symptom presentation and in delayed hospitalization reported for ACS and NSTEMI reinforce that demographic effects on prehospital outcomes are context-dependent and should be modeled explicitly [29,30].

Two operational associations ran counter to naive expectation and warrant explicit interpretation. First, high dispatch urgency (categories 1–2) was associated with slightly lower adjusted odds of hospitalization/transport (aOR 0.84). This is most plausibly an artifact of the urgency distribution rather than a true protective effect: category 2 alone accounted for the large majority of calls (81.0%), so the dichotomized “high urgency” group is dominated by category-2 activations and contrasts with a small, heterogeneous lower-urgency group, compressing the contrast and pulling the estimate below 1. Dispatch urgency is also assigned before the crew assessment from limited call-taker information, so it is an upstream triage signal rather than a measure of on-scene severity. A sensitivity analysis modeling urgency as a four-level categorical variable, rather than the operational 1–2 versus 3–4 dichotomy, would clarify whether a monotonic gradient is present and is a worthwhile robustness check in future work. Second, attendance by a specialized crew was associated with marginally lower adjusted odds of hospitalization/transport (aOR 0.84). Rather than implying that specialized crews convey less appropriately, this likely reflects triage-after-stabilization and selection effects: specialized crews can deliver definitive on-scene treatment that resolves the acute problem and averts transport, and their deployment is itself governed by dispatch protocols that allocate them non-randomly across call types. Both estimates are therefore at least partly endogenous and should be read as associations conditional on the dispatch and deployment process rather than as causal effects of urgency coding or crew type on the conveyance decision.

4.3. Methodological Context and Implications

Methodologically, our study sits within an active stream of work building prehospital risk-stratification and dispatch tools. Logistic-regression and machine-learning models for chest-pain triage and dispatch have achieved areas under the curve of roughly 0.74-0.79 using variables routinely available to EMS [11,12], and machine-learning approaches applied to the prehospital electrocardiogram or to non-conveyance follow-up have matched or modestly exceeded regression while sacrificing interpretability [10,13,14]. Our results suggest that comorbidity indicators - already captured in dispatch records - are candidate features for such tools, and that models omitting them may under-represent the real drivers of hospitalization/transport decisions for chronic IHD. From a practical standpoint two implications follow for urban EMS systems such as Astana's. First, because heart failure, arrhythmias and cardiogenic shock are independently associated with hospitalization, structured field documentation of these conditions could feed into comorbidity-aware risk scores, helping to standardize hospitalization/transport decisions and reduce unwarranted variation, while remaining mindful that both over- and under-triage carry measurable risk [15,16]. Second, the modest but consistent effects of response time, crew type and dispatch urgency point to system-level levers - crew allocation and dispatch prioritization - that interact with patient case mix; aligning specialized crews and high-priority dispatch with the highest-risk comorbidity profiles, in line with current ACS guidance [34,35], may improve the efficiency of an increasingly stretched service [36].

4.4. Strengths and Limitations

The principal strengths of this study are its large, consecutive, population-based sample drawn from a centralized municipal dispatch system, its five-year span, its STROBE-compliant reporting [37], and the use of a pre-specified multivariable model that mirrors the analytical standard of comparable coronary registries [7,8,9]. Capturing the real at-scene hospitalization/transport decision, rather than a downstream hospital outcome, is a further strength with direct operational relevance, and the setting addresses a genuine evidence gap for Central Asian EMS systems [5,6].

Several limitations temper these strengths and concern the data and methods directly. First, the analysis relies on working diagnoses and comorbidity indicators recorded by ambulance crews under time pressure; misclassification and under-documentation are likely, and the very low recorded prevalence of diabetes (0.8%) almost certainly reflects incomplete field coding rather than true prevalence, which would bias its effect toward the null. Second, as a retrospective observational study it can establish association but not causation, and residual confounding by unmeasured factors - symptom severity, vital signs, point-of-care troponin, frailty, social circumstances and patient preference - is probable; the modest McFadden pseudo-R² (0.093) makes explicit that most of the variation in the hospitalization/transport decision is not captured by the recorded variables, an inherent ceiling for models built on routine dispatch fields rather than richer clinical data [17,19]. Third, response time and dispatch urgency are partly endogenous markers of severity, so their odds ratios should be read as associations rather than direct effects. Fourth, we lacked linkage to in-hospital and post-discharge outcomes, so we cannot judge whether individual decisions were clinically optimal; non-conveyance safety in particular can only be assessed with follow-up data on re-contact, admission and mortality [14,15,16]. Fifth, the unit of analysis was the EMS call rather than the unique patient. Because the dispatch records lacked a stable cross-call patient identifier, repeat calls from the same individual could not be identified and aggregated; if such repeat activations are present, observations may not be fully independent, which can understate standard errors and narrow confidence intervals. The estimates should therefore be read as call-level associations, and future work with patient-level linkage and clustered or mixed-effects models will be needed to confirm them. Finally, the data derive from a single capital-city EMS system, and external validity to rural Kazakhstan or to other health systems should be tested before extrapolation [31].

4.5. Future Directions

Future work should link prehospital records to hospital and mortality registries to evaluate the appropriateness and safety of at-scene decisions, incorporate vital signs and point-of-care biomarkers to lift the explanatory ceiling observed here, and prospectively derive and externally validate a comorbidity-aware field risk-stratification tool - benchmarking interpretable logistic regression against machine-learning alternatives [13,14,20]. Comparative studies across urban and rural EMS systems in Central Asia would help establish the generalizability of the predictors identified here.

5. Conclusions

Among nearly 10,000 EMS calls for ischemic heart disease in Astana, roughly one in four resulted in emergency hospitalization. Clinical acuity - cardiogenic shock and acute/unstable IHD - was the dominant driver of the decision, but stable cardiac comorbidities, especially heart failure, atrial fibrillation and other arrhythmias, independently increased the odds of hospitalization after full adjustment. Male sex and younger age raised these odds, while the oldest patients, specialized crews and higher dispatch urgency were associated with lower odds. Because these predictors are routinely captured in dispatch records, they offer a practical basis for comorbidity-aware field risk-stratification and for more efficient allocation of EMS resources in urban Central Asian settings.