Risk of type 2 diabetes mellitus in patients
with acute critical illness: a populationbased
cohort study
Chin-Wang Hsu, Chin-Sheng Lin,
Sy-Jou Chen, Shih-Hua Lin, Cheng-Li Lin &
Chia-Hung Kao
Introduction
Stress-induced hyperglycemia (SIH) is common in patients with critical conditions such as sepsis, acute stroke, and acute myocardial infarction (AMI). Previous studies have suggested that elevated blood glucose levels with acute stress is part of the adaptive response to survive critical illness, which is mediated by
inflammation, immune, and neuroendocrine mechanisms [1]. However, increasing degrees of SIH may indicate the severity of the physiological stress response, implying a possible link between SIH and adverse outcomes of critical illness. Tamita et al. proposed that abnormal glucose tolerance is a major risk factor for future cardiovascular events in patients with AMI [2]. Moreover, previous studies have suggested the association of hyperglycemia with worsened outcomes in
patients with sepsis [3–5]. Eslami et al. analyzed 12 studies in a systemic review and suggested that blood glucose variability might be associated with mortality in critical ill patients [6, 7]. Such evidence indicated the possible interaction between critical illness and hyperglycemia.
Hyperglycemia during critical illness may represent a known diagnosis of diabetes mellitus (DM), previously unrecognized DM, or SIH. Although SIH usually
resolves spontaneously, it may indicate impaired glucose metabolism or tolerance, which is a high risk factor for DM [3, 8]. Gornik et al. suggested that patients with
hyperglycemia during sepsis are at risk of DM [3].
MacIntyre et al. reported that 14 % of patients hospitalized with pneumonia and SIH developed newly
diagnosed DM within 5 years [9]. Okosieme et al. reported a high prevalence of DM (27 %) and impaired glucose tolerance (39 %) in patients with acute coronary syndrome and no history of DM [10]. These results have indicated that patients with critical illnesses are at a high risk of DM.
To date, no large population-based study has evaluated the association between critical illness and DM
risk. Therefore, this study investigated the risk of developing DM in patients with certain critical illnesses by analyzing the Taiwan National Health Insurance Research Database (NHIRD). The following statistical models were adopted for the analysis: Kaplan–Meier
analysis, univariate and multivariate Cox proportionalhazard regression models, and competing-risk regression
models. We hypothesized that critical illnesses are associated with DM risk; accordingly, both patients and clinicians should be aware of the potential risk of DM following diagnosis of septicemia, septic shock, AMI, and stroke.
Methods
Data source
The Taiwan National Health Insurance (NHI) program is a single-payer universal insurance system with more than 99 % of Taiwan’s population covered (http://nhird. nhri.org.tw/en/background.html). The National Health Research Institutes established and manages the NHIRD. Data such as identification numbers are scrambled before being released to researchers. In this study, we used the Longitudinal Health Insurance Database 2000 (LHID2000), which contains original claims data of 1,000,000 individuals randomly sampled from the 2000 Registry for Beneficiaries (23.75 million residents of Taiwan) of the NHIRD.
According to the NHRI, no statistically significant differences exist regarding the distribution of sex, age or health
care costs between cohorts in LHID2000 and all insurants. The Bureau of NHI has established an ad hoc committee consisting ofmedical experts in various disciplines to review the claims on a random sample basis. The LHID2000 contains medical information including inpatient and outpatient
care facilities, drug prescriptions, sex, date of birth, visitation and hospitalization dates, and diagnosis codes in the format of the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM). The diagnostic
ICD-9-CM codes were determined by qualified clinical physicians based on laboratory, imaging, and pathological
data. All insurance claims are scrutinized by medical reimbursement specialists and undergo peer review. Severe
penaltieswould be imposed on false claims and unnecessary treatment. Therefore, the definition of diagnostic codes are accurate and reliable even if they were diagnosed by different doctors. This study was approved by the Ethics
Review Board of China Medical University (CMU-REC-101-012).
Research participants
From the 2000–2011 claims data, we identified patients aged 20 years and older with the following newly diagnosed critical illnesses: septicemia (ICD-9 code 038),
septic shock (ICD-9 code 785.52), AMI (ICD-9 code 410), hemorrhagic stroke (ICD-9 codes 430-432), and ischemic stroke (ICD-9 codes 433-438). The date of initial critical illness diagnosis was defined as the index date. We excluded patients with a medical history of type 2 diabetes mellitus (T2DM) (ICD-9 codes 250.x0 and 250.x2) or antidiabetic agents use. Patients diagnosed with T2DM and those who died within 90 days after the
index date were also excluded. For each patient with critical illness, 2 comparison controls with no critical
illness were randomly selected from the LHID2000 and frequency-matched according to age (every 5 years), sex, and year of index date on the basis of the same exclusion criteria. A second method of comparison included the critical illness and control cohorts matched with propensity
score (PS) at a 1:1 ratio was performed to minimize
selection bias [11]. We used logistic regression to estimate the probability of the disease status by calculating
the PS. Baseline variables for calculating the PS were the year of critical illness diagnosis, age, sex, urbanization level, Charlson comorbidity index score (CCI score) and the following comorbidities and medications: hypertension (ICD-9 codes 401–405), hyperlipidemia (ICD-9 code
272), congestive heart failure (ICD-9 codes 428, 398.91, and 402.x1), chronic kidney diseases (ICD-9 codes 585, 586, 588.8, and 588.9), chronic obstructive pulmonary disease (ICD-9 codes 491, 492, and 496), cancer (ICD-9 codes 140–208), intensive care unit (ICU) hospitalization, endotracheal tube (E-T) intubation, obesity (ICD-9 code 278), and mechanical ventilation, and medications of norephinephrine, dopamin, and heparin. Each patient was followed until diagnosis of T2DM, death, withdrawal
from the NHI program, or until December 31, 2011. Statistical analysis
Two matching methods were adopted to compare the risk of T2DM and mortality between the critical illness and control cohorts: an age- and sex-matched design and PS matching. Demographic characteristics and the prevalence of comorbidities were compared using the chisquare
test for the categorical variables and a nonparametric Mann–Whitney test for the continuous variables.
We calculated the incidence density of T2DM and mortality according to person-years in each cohort. The
incidence rate ratio (IRR) of the critical illness cohort to the control cohort and 95 % confidence interval (CI) were estimated using Poisson regression. Multivariate Cox proportional-hazard regression models were employed to determine the effect of critical illness on the risk of T2DM and mortality by using hazard ratios (HRs) with
95 % CIs. Furthermore, we considered death as a competing risk for estimating the risk of T2DM. By
controlling the competing risk of death, the Fine and Gray model, which extends the standard univariate and multivariate
Cox proportional-hazard regression model, was
used to estimate the T2DM risk [12]. The multivariate models were simultaneously adjusted for age, sex,
urbanization level, CCI score, comorbidities of hypertension, hyperlipidemia, CKD, CHF, COPD, cancer, ICU
hospitalization, E-T intubation, obesity, mechanical ventilation, and medications of norephinephrine, dopamin,
and heparin. To assess the difference in the T2DM-free probability curve and survival curve between the 2 cohorts, Kaplan–Meier analysis and a log-rank test were used. All analyses were performed using SAS v.9.3 (SAS Institute, Cary, NC, USA), and the Kaplan–Meier survival curve was plotted using R software (R Foundation for Statistical Computing, Vienna, Austria). A 2-tailed P\0.05 was considered statistically significant.
Results
We identified 26,382 patients with certain newly diagnosed critical illness and 52,087 control patients as the age- and sex-matched cohorts. In addition, 9528 patients in the critical illness cohort were matched with 9528 control
patients according to the PS. Both of the age- and sexmatched cohorts predominantly comprised patients older
than 65 years (approximately 53 %), men (59.2 %), and CCI score C2 (30.1 %); comorbidities, ICU hospitalization, E-T intubation, and medications were significantly
more common in the critical illness cohort (P\0.001). By contrast, after PS matching, the two cohorts were more similar in the baseline characteristics (Table 1).
Figure 1 shows the T2DM- and mortality-free probabilities, which were obtained using the Kaplan–Meier
analysis, over 12 years of follow-up of the age- and sexmatched cohorts. The T2DM- and mortality-free probabilities
were significantly lower in the critical illness
cohort than in the control cohort (log-rank test: P\0.001 for both).
The median follow-up periods (interquartile range,
IQR) for the patients with T2DM in the age- and sexmatched certain critical illnesses and control cohorts were
3.86 (1.64–6.93) and 5.12 (2.51–8.13) years, respectively (Table 2). The overall incidence density of T2DM was higher in the critical illness patients than in the controls (13.9 vs. 9.46 per 1000 person-years), with an IRR of 1.47 (95 % CI 1.41–1.52). After adjusting for age, sex, urbanization level, CCI score, comorbidities of hypertension,
hyperlipidemia, CKD, CHF, COPD, cancer, ICU hospitalization, E-T intubation, obesity, mechanical ventilation,
and medications of norephinephrine, dopamin, and heparin, critical illness patients had an adjusted HR (aHR) of
1.31 (95 % CI 1.20–1.42) for T2DM, compared with the controls. The risk of T2DM was significantly higher in the critical illness cohort (32 %) than in the control cohort according to the PS (aHR = 1.32, 95 % CI 1.16–1.50). Moreover, the risk of mortality was 2.85-fold greater in the PS-matched critical illness cohort compared with that in the control cohort (aHR = 2.85, 95 % CI 2.61–3.11). Compared with the control patients, the aHRs of T2DM for the aged B49 group and the aged 65? group age- and sex-matched critical illness patients were 1.40 and 1.39, respectively (95 % CI 1.13–1.74; 95 % CI 1.23,
1.56) (Table 3). Compared with the adjusted HRs for the control patients, those for T2DM in the PS-matched
critical illness patients were significantly higher for all age groups, except for those aged B49 group. The risk of mortality among critical illness patients were statistically significant higher in all age groups than the control patients. Similar trends were observed in PS-matched cohorts.
The incidence and HRs of T2DM and mortality were calculated according to the various subtypes of critical illness in the age-and-sex-matched cohort (Table 4). Compared with the patients in control cohort, those with septicemia or septic shock were 1.51-fold more likely to develop T2DM (95 % CI 1.48–1.64), followed by patients with AMI. Compared with the controls, the risk of mortality was highest among the critical illness patients with
4.69–5.32), followed by hemorrhagic stroke, and
ischemic stroke. After adjustment for all relevant confounding factors in the competing risk regression model,
the risk of T2DM remained significantly increased in the critical illness cohort (aHR = 1.58, 95 % CI 1.45–1.72) (Table 5).
Discussion
In this study, the association of T2DM risk with septicemia, septic shock, stroke, and AMI was
comprehensively evaluated using 2 analysis models. The results derived from all models consistently revealed a
significantly increased T2DM risk, irrespective of underlying comorbidities in the patients with certain
critical illnesses. Among the patients with critical illness, those with septicemia or septic shock exhibited the highest risk of T2DM, followed by those with AMI (Table 4). These results indicate that both patients and clinicians should consider testing for T2DM following diagnosis of septicemia, septic shock, and AMI. A strength of this study was the use of a nationwide population-based cohort analysis of the risk of developing T2DM in Taiwanese patients with critical illnesses.
Because participation in the NHI program is mandatory and all Taiwanese residents can access medical care with low copayments, the loss to follow-up is low. Moreover, as mentioned above, the diagnostic ICD-9-CM codes are determined by qualified clinical physicians and all
insurance claims are scrutinized by medical reimbursement specialists and undergo peer review. Therefore, the
definition of diagnostic codes are accurate and reliable in this study.
DM is often asymptomatic for up to 7 years before
diagnosis [13]. Ginde et al. reported a 14 % prevalence of newly diagnosed DM in emergency department (ED)
patients who were unaware that they had the disease [14].
Furthermore, 29 % of cases were confirmed by abnormal HbA1c during their ED visit, and 74 % of cases were
1–6 weeks after visiting the ED. This evidence emphasizes that DM may be overlooked or unrecognized in
asymptomatic patients when they visit a general practitioner. Moreover, previous studies have suggested that
newly diagnosed DM or newly detected abnormal glucose tolerance is a strong risk factor for future cardiovascular events in patients with AMI [2, 15]. DM can be recognized early by measuring HbA1c, serial fasting glucose
testing, and oral glucose tolerance testing. Lifestyle modification and appropriate glycemic control can reduce possible complications [15]. Although our study does not firmly establish a temporal relationship between critical illness and T2DM, and it is entirely possible that the critically ill cohort became critically ill because they had undiagnosed and unmanaged T2DM, we did suggest the association between T2DM and certain critical illnesses. Therefore, target screening and frequent follow-ups for DM increase the likelihood of an early diagnosis and can prevent chronic complications from developing, which is crucial for managing patients with critical illness.
A large population-based retrospective cohort study
from McAllister et al. suggested patients with initial plasma glucose level more than 15 mmol/l have the
highest risk of developing T2DM, especially those not admitted to ICU. Accordingly, they concluded that initially elevated plasma glucose levels are associated with
subsequent risk of developing T2DM and mortality in hospitalized patients [16]. However, among patients with initial plasma glucose levels less than 5 mmol/l, who
account for the most-studied patients, patients admitted to ICU have higher risk of developing T2DM compared with those not admitted to ICU. Such evidence is in agreement with our study and further supports the effects of critical illness on the risk of developing T2DM.
The underlying mechanisms of critical illness related to T2DM include adaptive or maladaptive response to stress through neuroendocrine, inflammatory, and immune mechanisms [1, 3, 17]. When the human body
senses a stressor, several systems and organs are activated (e.g., the central nervous system, the immune system, adipose tissue, and the gastrointestinal tract) [1, 17]. During acute stress, these responses ensure that sufficient glucose is available for the brain and blood cells, which rely on glucose as a metabolic substrate [17]. Stress activates the hypothalamic–pituitary–adrenal axis,
releasing cortisol and other anti-insulin hormones such as catecholamines, glucagon, and growth hormones [18, 19]. Moreover, the inflammatory and immunological processes activated during acute stress are partially regulated by the central nervous system as well as cytokine and inflammatory mediators, such as the tumor necrosis factor and interleukin-1 [1, 3, 20]. These inflammatory reactions not only promote the anti-insulin hormones but also alter insulin signaling, leading to insulin resistance, which decreases glucose uptake in muscle and fat cells but causes gluconeogenesis and increased insulin release in hepatocytes [3, 21, 22]. An uncontrolled catabolism and resistance to anabolism during acute stress can contribute to insulin resistance [1]. Finally, the combination of
oxidative and endoplasmic reticulum stress and impaired autophagy during acute stress may contribute to pancreatic b-cell death or dysfunction, resulting in T2DM [23].
In this study, the patients with septicemia or septic shock were at the highest risk of T2DM among the patients with critical illness (Table 4). The possible underlying reasons must be explored. In patients with prolonged critical illness, SIH can be severe and persist for a long time, resulting in T2DM [3, 8, 17]. Moreover,
stress may uncover latent impairment of glucose metabolism leading to overt hyperglycemia [3]. Gornik et al.
proposed that metabolic impairment related to sepsis and the inflammatory response is persistent and can progress to an overt disturbance of glucose intolerance or T2DM following a septic episode, even after infection and stress are controlled [3]. Although this phenomenon may not be specific to infection or sepsis, which probably occur in
most patients with critical illnesses [8], our results imply that the contribution of critical illness to T2DM risk is different and that sepsis has a marked effect on T2DM risk among the critical illnesses investigated in this study. Our study had several limitations. First, we had no
data regarding the plasma glucose concentrations, degrees of SIH, proportions of patients with SIH, severity of the critical illness, and the causes of mortality. Moreover, a subgroup analysis on the risk factors for critical illness related to T2DM was not conducted. Second, although we observed an increased risk of T2DM and mortality in the patients with critical illness (Fig. 1), the association of T2DM with survival outcomes following diagnosis of critical illness was not evaluated. Third, patients with other critical conditions, such as major trauma, surgery, and burn injuries, who require ICU hospitalization and intubation, may be included in the control group and are not evaluated for their risk of developing T2DM. Fourth, although we excluded patients with newly diagnosed T2DM within 90 days of the index date, some patients with pre-existing unrecognized DM could not be excluded. Fifth, the NHIRD contains no detailed information
on lifestyle or health-related factors of patients, such as smoking status, alcohol consumption, body mass index, socioeconomic status, prescription history, and family history of T2DM, all of which can increase the risk of T2DM and were potential confounding factors. Finally, the methodological quality of evidence derived from cohort studies is generally lower than that of evidence derived from randomized trials because retrospective cohort studies are subject to various biases because of the lack of necessary adjustments or possibly unmeasured or unknown confounding factors. Nevertheless, despite these limitations, our study provides critical evidence supporting the association of T2DM with critical illness.
In conclusion, based on the consistent results obtained through various approaches, the findings of this large
patients with certain critical illnesses.
Additional observational studies to establish
whether critically ill patients have undiagnosed T2DM prior to the onset of critical illness and to better control for established risk factors are necessary. Accordingly, we recommend intensive screening for T2DM following
