Trajectory of Type 2 Diabetes in Sepsis Outcome : Impacts of Diabetic Complication Burdens , Initial Glucose Level , and HbA 1 c : Population-Based Cohort Study Combining with Nationwide and Hospital-Based Database

1Institute of Occupational Medicine and Industrial Hygiene, National Taiwan University College of Public Health, Taipei, Taiwan; 2Department of Emergency Medicine, Taipei Veterans General Hospital, Taoyuan Branch, Taoyuan, Taiwan; 3Department of Emergency Medicine, Taipei Veterans General Hospital, Taipei, Taiwan; 4School of Medicine, National Yang-Ming University, Taipei, Taiwan; 5Department of Emergency Medicine, Taichung Veterans General Hospital, Taichung, Taiwan; 6Department of Emergency Medicine, Lin-Kou Medical Center, Chang Gung Memorial Hospital, Taoyuan, Taiwan; 7Department of Health Services Administration, China Medical University, Taichung, Taiwan; 8Management Office for Health Data, China Medical University, Taichung, Taiwan; 9Department of Medical Research, Taichung Veterans General Hospital, Taichung, Taiwan; 10Department of Environmental and Occupational Medicine, National Taiwan University College of Medicine and Hospital, Taipei, Taiwan;


Introduction
Sepsis is a leading cause of mortality in critical care worldwide.[1][2][3] In addition to mortality, sepsis may also induce acute organ dysfunction, and even cause long-term post-sepsis cardiovascular diseases.[4] The reported incidence of sepsis varies; however, it presents an undoubtedly increasing trend that is reflected in the aging population and greater recognition of this condition.
Treating sepsis patients causes significant national financial burdens.
Diabetes is an important comorbid condition in sepsis because of its high prevalence.[5] It has generally been believed that diabetic patients are more prone to infections than the general population.[6] However, the influence of diabetes on the outcome of sepsis remains inconclusive.
Higher mortality rates in the patients with diabetes were reported [7][8][9][10][11][12]; however, some others found no influence [13][14][15][16], and even protective effects of diabetes in sepsis.[17][18][19][20] The most frequently proposed study limitations of this debate were the study designs: the epidemiological studies using large cohorts can avoid the selection bias that is frequently observed in hospital-based studies, however, detailed clinical information is usually not available.Most importantly, many studies failed to consider the influence of diabetic complication severity.
HbA1c is commonly used to measure blood glucose control in diabetic patients and has also been proposed as an independent predictor for hospital mortality in sepsis patients.[21] However, its importance in diabetic sepsis patients requires further study due to limited information.
Hyperglycemia was shown to impair polymorphonuclear neutrophil function and cytokine production.However, it was reported that high initial glucose levels were not associated with increased mortality in the diabetic sepsis patients.[22] Furthermore, tight glucose control did not seem to be significantly associated with reduced hospital mortality in critical patients.[23,24] The influence of HbA1c and initial glucose levels on the outcome of sepsis deserves further investigation.
In the current study, using the representative nationwide database and the hospital-based database from multi-centers with laboratory data, we examined the association between type 2 diabetes and sepsis outcomes, and specifically focused on (1) whether diabetes itself increases the risk of mortality in hospitalized sepsis patients or depends on the diabetes complication burdens and (2) whether initial blood glucose levels and HbA1c affect the hospital course and outcomes.

Nationwide database
In the first part of this study, we conducted a nationwide cohort study using data from the National Health Insurance Research Database (NHIRD).The diagnosis codes of the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) are used in the NHIRD to identify specific diagnoses.Sepsis patients were retrieved using the ICD-9-CM codes 038.The specificity of a sepsis diagnosis in the NHIRD has been validated in previous studies.[25] The infection site classification was performed similar to Angus et al. (Supplement Table 1).[26] The patients were classified as using certain drugs if they took them for more than one month within a one-year period prior to the index hospitalization.The index date was defined as the first date of index hospitalization.The drug codes are shown in Supplement Table 2.The procedures during hospitalization were defined by using the claims data (Supplement Table 3).
Initially, we used the Longitudinal Cohort of Diabetes Patients (LHDB) of the NHIRD, which contains randomized selected data (120,000 patients / year) from patients with newly diagnosed diabetes to retrieve the study cohort of type 2 diabetic first episode sepsis patients.[27] The patients in the study cohort should have been diagnosed to have type 2 diabetes prior to the index hospitalization to allow for the evaluation of diabetic complication status by using the adapted Diabetes Complications Severity Index score (aDCSI score) (Supplement Table 4).[28,29] (The original form of DCSI score had some missing and we had corrected in the supplement table ) The Diabetes Complications Severity Index (DCSI) was first developed by Young et al. [28] The DCSI is a useful tool to adjust for the baseline severity of diabetic complications and to predict hospitalization and mortality.The aDCSI score was modified from the DCSI score and had been validated in the NHIRD.[30] The aDCSI score included seven categories of complications: cardiovascular disease, nephropathy, neuropathy, retinopathy, peripheral vascular disease, stroke, and metabolic emergency events.The comparison cohort, which was composed of non-diabetic first episode sepsis patients, was retrieved from the Longitudinal Health Insurance Database 2000 (LHID2000).The LHID2000 used in this study contains the medical information of 1 million beneficiaries, randomly sampled from the registry of all beneficiaries in 2000.This study was approved by the Institutional Review Board of China Medical University (CMUH104-REC2-115).

Hospital-based database
In the second part of this study, we retrieved data on the type 2 diabetic first episode sepsis patients between 2006 and 2012 from the electronic databases of three medical centers.Most of these diabetic sepsis patients were included in the LHDB.Laboratory data, including initial blood glucose level, HbA1c, and initial lactate level; hospital outcomes, including ICU (intensive care unit) admission, hospital and 28-day mortalities; the received procedures (including mechanical ventilation and emergent hemodialysis); blood culture results were collected for further analysis.
This study was approved by the Institutional Review Board of Taichung Veterans General Hospital (No. CE13233).
The participant selection process of the nationwide and hospital-based databases was showed in Supplement Figure 1 and Supplement Figure 2.Only type 2 diabetic patients were included for further study.In the hospital-based database, the initial blood glucose level was obtained on the admission day ether in the emergency department or ward before receiving any acute glucose-lowering injection therapy, that is, insulin; the HbA1c level should be obtained in a period of three days around the admission day.

Statistical analyses
In the hospital-based database, the type 2 diabetic and non-diabetic sepsis patients were matched by age and gender.In the nationwide database, the study cohort from the LHDB and the comparison cohort from the LHID2000 were matched by the propensity score matching method.
For each patient, we calculated the propensity score using the multivariate logistic regression by entering age, gender, income, urbanization level, hospital level, baseline comorbidities, and infection site.
Differences in demographic characteristics, comorbidities, medications, and laboratory data were examined using Chi-square, Mann-Whitney and two-sample t-tests.Odds ratios (ORs) with 95% confidence intervals (95% CIs) were calculated in the logistic regression model.The Kaplan-Meier analysis with log-rank test was conducted to compare the hospital outcome in patients with different initial blood glucose levels and HbA1c.The statistical analyses were performed using SAS 9.4 statistical package (SAS Institute Inc., Cary, NC, USA).A P value of 0.05 was considered significant.

Sensitivity analysis
Because of limited cases with the data of HbA1c using the strict criteria of three days around the admission date, we further conducted a sensitivity analysis.The HbA1c data was re-collected with a wider time period of one month prior to the admission date to re-examine the effect of HbA1c.

First part: nationwide database
From the LHDB and LHID2000 between 1999 and 2012, after propensity score matching, 19,719 type 2 diabetic and an equal number of non-diabetic first episode sepsis patients were retrieved as the study and comparison cohorts.The demographic characteristics, comorbidities, medications, infection sites, and received procedures of the study and comparison cohorts were showed in Supplement Table 5.
After matching, in the further multivariate analysis, type 2 diabetic sepsis patients had an increased OR of 1.14 (95% confidence level [CI] 1.1-1.19,P<0.0001) for mortality after adjusting for age, gender, insurance premium (as a proxy for household income), urbanization level, and hospital level (Supplement Table 6).

Second part: hospital-based database
From the hospital-based database, we initially included 4,984 sepsis patients between 2006 and 2012.After matching for age and gender, there were 1,054 type 2 diabetic and 2,108 non-diabetic sepsis patients included for further analysis.The type 2 diabetic sepsis patients had a higher prevalence of receiving hemodialysis (23.2% versus 16.9%, P<0.001) during hospitalization (Table 1).The type 2 diabetic sepsis patients had higher hospital mortality rate (45.2% versus 42.3%, P=0.138) and 28-day mortality rate (35.5% versus 32.8%, P=0.147) compared to the non-diabetic sepsis patients.The type 2 diabetic sepsis patients had higher prevalence of gram-positive coccus bacteremia (16.8% versus 14.4%, P=0.089) but lower prevalence of gram-negative bacillus bacteremia (19.1% versus 20.7%, P=0.294) compared to the non-diabetic sepsis patients, although the P value did not reach the significance.Acute physiologic and chronic health (APACH), Charlson comorbidity index (CCI),Chronic kidney disease (CKD), Chronic liver disease (CLD), Chronic obstructive pulmonary disease (COPD), Gram-negative bacillus (GNB), Gram-positive coccus (GPC), Hypertension (HTN), Intensive care unit (ICU), Ischemic heart disease (IHD), Peripheral arterial occlusion disease (PAOD) In the univariate and further multivariate logistic regression analyses, type 2 diabetes was associated with an increased risk of hospital mortality during the sepsis course (adjusted OR=1.31, 95% CI, 1.11-1.54,P=0.002) (Supplement Table 7).But the Kaplan-Meier analysis with log-rank test did not show a significant difference of hospital course of mortality between the type 2 diabetic and non-diabetic sepsis patients (P=0.122)(Figure 2A).The 1,054 type 2 diabetic sepsis patients were divided into surviving and dead groups for further comparison (Supplement Table 8).Initial blood glucose levels in the surviving and dead diabetic sepsis patients did not differ significantly: 273.9 ± 180.3 versus 266.1 ± 200.2 (mg/dL) (P=0.095)(Figure 2B).Furthermore, the surviving diabetic sepsis patients did not have lower HbA1c (%) than the dead diabetic sepsis patients: 8.4 ± 2.6 versus 8.0 ± 2.5 (P=0.078).In the further logistic regression analysis, the univariate analysis which included age, gender, CCI score and important laboratory data, showed an OR of 1.00 (95% CI, 1.00-1.00,P=0.532) for initial glucose levels and 0.94 (95% CI, 0.86-1.02,P=0.143) for HbA1c (Supplement Table 9).The Kaplan-Meier analysis with log-rank test also showed that the hospital mortality during the sepsis course did not differ between type 2 diabetic sepsis patients with different initial blood glucose levels (≤200, 201-400, and >400 mg/dL) and HbA1c (≤7 and >7%) (Figure 3) (Supplement Table 10 and Supplement Table 11).In the sensitivity analysis, using a wider time period of one month prior to the admission date for HbA1c collection, the hospital mortality during the sepsis course between the type 2 diabetic sepsis patients with HbA1c ≤7 and >7% did not differ significantly (47.2% versus 44.3%, P=0.471) (Supplement Table 12) (Supplement Figure 4).

Discussion
In the current study, we demonstrated that the outcome of type 2 diabetic sepsis patients was mainly determined by the diabetic complication burdens (represented as the aDCSI score).
Somewhat surprisingly, neither the recent glucose control (HbA1c) nor the initial glucose level was associated with hospital mortality in the sepsis course.Physicians should not infer the outcome of a diabetic sepsis patient merely via the recent poor glucose control or initial high glucose level; rather, they should consider the diabetic complication burdens.Donnelly et al. demonstrated that diabetes was associated with an increased risk of hospitalization due to infection diseases.However, diabetes itself and insulin use were not associated with increased 28-day hospital mortality.[31] Dianna et al. demonstrated that diabetic patients had an excess risk of dying from a range of infection diseases.[32] Both studies used a large cohort; however, their conclusions were conflicting.We infer that the difference was induced by the lack of severity classification of diabetic complications.In our current study, we introduced the utilization of aDCSI score and the results showed that the sepsis outcomes of diabetic patients were mainly determined by the complication burdens of diabetes.This inference was supported by the dose-responsive effect in the trend test in our study.
In this study, we found that the OR for hospital mortality elevated as the aDCSI score increased, and presented in a dose-responsive manner.In the type 2 diabetic sepsis patients with an aDCSI score of ≤ 1 even had a reverse OR for hospital mortality.However, we did not found an obviously shorter disease course of diabetes (from the first diagnosis date of type 2 diabetes to the index hospitalization date) in the patients with an aDCSI score of ≤1 (median duration = 294 days) compared to those with an aDCSI score of >2 (median duration = 306 days).From this result, we can infer that in type 2 diabetic sepsis patients, diabetes-related complication burdens are the major determinant of hospital mortality, but not merely the diabetes itself.
HbA1c is a widely used marker that reflects the average glucose level within 120 days.It was reported that HbA1c was a major outcome predictor in diabetic sepsis patients.[21] However, our study results did not support this argument.There are many studies supporting the influence of long-term glycemic control on diabetic complication development.[33,34] Long-term poor glycemic control makes diabetic patients prone to infection diseases because of their impaired immune functions.[31] However, the hospital outcome of diabetic sepsis patients presenting with higher HbA1c may be not as poor as we initially thought because these patients may receive more aggressive blood sugar control in the initial stage of sepsis with insulin.
Hyperglycemia frequently occurs in sepsis patients as a stress response by stimulating However, our study demonstrated that a high blood glucose level at admission was not associated with the hospital outcome.
Our study has the strengths.We examined the association between type 2 diabetes and sepsis outcomes by concomitantly using the representative nationwide database (LHDB and LHID2000) and the hospital-based database from medical centers (details in the appendix) with laboratory data.Using this method, we longitudinally linked the long-term accumulated diabetic complication burdens, HbA1c (as a representation of recent glucose control), and initial blood glucose levels at admission to describe the complex trajectory of the disease course in these type 2 diabetic sepsis patients.Our study which evaluated the diabetic complication burdens by using aDCSI score rather than simply adjusting for baseline comorbidities was of important innovation in this topic.Also, in this study of the nationwide database, we used the claims data for procedures such as mechanical ventilation, hemodialysis and so on; therefore the accuracy is much better than the use of ICD codes for acute organ dysfunction.Finally, the detailed information such as blood culture results in the hospital-based database provided a richer understanding in the complex interplay between type 2 diabetes and sepsis, rather than simple taxonomy.
This study has limitations.First, although we were able to link the individual patient's medical information between the hospital-based database and the nationwide database to create a convincing longitudinal cohort study.However, due to the increasing conflict of health database utilization in Taiwan, we abandoned this idea to avoid further severe conflicts.Second, HbA1c was usually measured at the physician's discretion and we initially included only one third of the whole type 2 diabetic sepsis patients with data of HbA1c (366 in 1054), that might induce a significant bias.We therefore use a wider time period of one month to substitute three days as the inclusion criteria to examine the effect of HbA1c on sepsis outcome.However, the OR of HbA1c and hospitalization course remained unchanged in using the larger sample size (953 in 1054).

Conclusion
In type 2 diabetic sepsis patients, diabetes-related complication burdens are the major determinant of hospital mortality, rather than the diabetes itself.Initial blood glucose levels and HbA1c may not be associated with the hospital outcome of sepsis.

Figure 1
Figure 1 Logistic regression analysis analyzing the odds ratios of hospital mortality in sepsis patients with type 2 diabetes and different diabetic complication burdens (aDCSI scores).

Figure
Figure 2AThe Kaplan-Meier analysis with log-rank test for hospital course of mortality between the type 2 diabetic and non-diabetic sepsis patients.Figure2BComparison of Initial blood glucose levels in the surviving and dead diabetic sepsis patients.

Figure 3
Figure 3 Kaplan-Meier analysis with log-rank test for type 2 diabetic sepsis patients with different initial blood glucose levels and HbA1c.

Figure 1
Figure 1 Logistic regression analysis analyzing the odds ratios of hospital mortality in sepsis patients with type 2 diabetes and different diabetic complication burdens (aDCSI scores).

Figure 2A
Figure 2AThe Kaplan-Meier analysis with log-rank test for hospital course of mortality between the type 2 diabetic and non-diabetic sepsis patients.Figure2BComparison of Initial blood glucose levels in the surviving and dead diabetic sepsis patients.

Figure 3
Figure3 Kaplan-Meier analysis with log-rank test for type 2 diabetic sepsis patients with different initial blood glucose levels and HbA1c.

Table 1 Demographic characteristics, comorbidities, laboratory data, hospital course, and outcomes of matched type 2 diabetic and non-diabetic sepsis patients
¶ Mann-Whitney test.Continuous data were expressed as mean ± SD.Categorical data were expressed as number (percentage).

Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 July 2018 doi:10.20944/preprints201807.0398.v1 gluconeogenesis
[38][36][37] recycled pyruvate and lactate.[35][36][37]Hyperglycemiamayhave protective effects in patients because high blood glucose levels increase the diffusion gradient in tissues with abnormal microvasculature caused by sepsis.Our study may indirectly support the above argument.The study by van Vught et al. demonstrated that admission hyperglycemia was associated with adverse outcomes in sepsis course, irrespective of the presence of diabetes.[38]

Table Table 1
Demographic characteristics, comorbidities, laboratory data, hospital course, and outcomes of matched type 2 diabetic and non-diabetic sepsis patients.