diabetes research and clinical practice 106 (2014) 373–382

Contents available at ScienceDirect

Diabetes Research and Clinical Practice journ al h ome pa ge : www .elsevier.co m/lo cate/diabres

Hemoglobin A1c, comorbid conditions and all-cause mortality in older patients with diabetes: A retrospective 9-year cohort study§ David Grembowski b,*, James D. Ralston a,1, Melissa L. Anderson a,2 a

Group Health Research Institute, 1730 Minor Ave., Suite 1600, Seattle, WA 98101, United States Department of Health Services, School of Public Health, University of Washington, 1959 NE Pacific Street, Box 357660, Seattle, WA 98195-7660, United States b

article info

abstract

Article history:

Aims: To examine whether hemoglobin A1c levels and comorbid conditions are related to

Received 21 August 2013

all-cause mortality in a cohort of patients with type 1 or 2 diabetes receiving continuous care

Received in revised form

for 9 years. In patients with comorbid congestive heart failure (CHF), we test for ‘reverse

25 April 2014

epidemiology,’ or whether greater HbA1c values are associated with lower risk of mortality.

Accepted 20 July 2014

Methods: The population for this longitudinal cohort study was 8820 Group Health enrollees

Available online 29 July 2014

in the Seattle area with type 1 or 2 diabetes in 1997 and enrolled continuously from 1997 to 2006. Comorbid conditions were hypertension, coronary artery disease, congestive heart

Keywords:

failure, depression, and chronic pulmonary disease. Mistimed HbA1c scores were addressed

Diabetes

by multiple imputation, and Cox proportional hazards models estimated associations

Comorbid conditions

controlling for other risk factors.

Glycated hemoglobin

Results: About 30% of the enrollees died in 1998–2006. CHF had the strongest association

Reverse epidemiology

with all-cause mortality. Compared to enrollees with HbA1c  7.1% (54 mmol/mol) and

Mortality

7.5% had HR < 1.0 but were not significant. For enrollees with diabetes and CHF at baseline, HbA1c scores  8.7% (72 mmol/ mol) had a significantly lower risk of death (HR range: 0.64–0.69). Conclusions: In our patient population, HbA1c scores < 6.4% have significantly higher allcause mortality. CHF is a major determinant of all-cause mortality. Adults with comorbid CHF and high HbA1c scores have lower all-cause mortality. # 2014 Elsevier Ireland Ltd. All rights reserved.

§

Funded by Grant No. R21 HS017657 from the Agency for Healthcare Research and Quality. * Corresponding author. Tel.: +1 206 616 2921; fax: +1 206 543 3964. E-mail addresses: [email protected] (D. Grembowski), [email protected] (J.D. Ralston), [email protected] (M.L. Anderson). 1 Tel.: +1 206 287 2076; fax: +1 206 288 2871. 2 Tel.: +1 206 287 2647; fax: +1 206 288 2871. http://dx.doi.org/10.1016/j.diabres.2014.07.017 0168-8227/# 2014 Elsevier Ireland Ltd. All rights reserved.

374

1.

diabetes research and clinical practice 106 (2014) 373–382

Introduction

There are few long-term studies of the benefits of intensive glycemic control in older adults with comorbid conditions [1]. In 2008 the ACCORD trial reported increased all-cause mortality and cardiovascular mortality in the intensive treatment group (target HbA1c < 6.0%, or 42 mmol/mol) compared to the standard treatment group (target HbA1c 7.0–7.9%, or 53–63 mmol/mol) [2,3]. The increased mortality may not be caused by tighter glycemic control per se, but rather may likely reflect the treatment strategies to reach these A1c levels [4,5]. Since 2008 six meta-analyses [6–11], one systematic review [12], one systematic review/meta-analysis [13], and five reviews [4,5,14–16] have examined the effects of intensive glycemic control on health outcomes reported in randomized trials (range in number of trials studied: 4–14). Most conclude that intensive glycemic control had reduced rates of microvascular complications and non-fatal myocardial infarction, had no effect on stroke and all-cause or cardiovascular mortality, and increased hypoglycemia. However, the type 2 diabetes UKPDS trial and the type 1 diabetes EDIC trials with 10+ year follow-ups had lower all-cause and cardiovascular mortality, respectively, as a result of previous intensive control (‘legacy effect’) [4,5,17,18]. Observational studies report mixed evidence about intensive glycemic control and health outcomes in older patients with diabetes and comorbid conditions. In a 5-year longitudinal observational study, intensive glycemic control was not associated with cardiovascular benefits in adults with mean age over 61 and high levels of comorbidity [19]. In a 1986–2008 cohort study with mean age > 59 in the United Kingdom, HbA1c deciles had a nuanced, U-shaped relationship: low and high HbA1c values were associated with increased all-cause mortality and cardiac events, with HbA1c values of about 7.5% (58 mmol/mol) having the lowest hazard ratios [20]. A similar U-shaped relationship was found in a cohort study of older patients with diabetes [21], and in a cohort study of heart failure patients with diabetes [22]. Other studies of adults with diabetes and congestive heart failure (CHF) find support for the ‘reverse epidemiology’ hypothesis, or the paradox that higher HbA1c values are associated with better survival in populations with CHF or end stage renal disease [23–25], suggesting that patients with advanced comorbidity may have higher morbidity and mortality risks with tight glycemic control. For patients with limited life expectancy, advanced microor macrovascular complications or extensive comorbidity, an HbA1c goal < 8% (64 mmol/mol) may be more appropriate [1]. Other emerging clinical policies also suggest that target HbA1c < 8% may be appropriate for older adults with long duration of diabetes and multiple or severe comorbid conditions [26,27]. Few long-term studies have examined the contribution of HbA1c to mortality in a population of older patients with advanced comorbidity. Our purpose is to examine whether HbA1c is associated with all-cause mortality in a cohort of patients with type 1 or 2 diabetes and receiving continuous care for 9 years in one integrated delivery system in the United States. We also examine comorbid hypertension, CAD, CHF, and depression because they are common and correlated with higher

mortality rates in patients with diabetes [28]. In a sub-analysis of patients with diabetes and CHF, we also test whether greater HbA1c values are associated with lower risk of mortality.

2.

Methods

2.1.

Population and study design

Patients were from Group Health (GH), a mixed-model health care system with over 580,000 enrollees and a multispecialty group practice of 800 physicians who work in its owned and operated facilities located mainly in Washington State (USA). The patient population for this retrospective, longitudinal cohort study consisted of Group Health patients, or ‘‘enrollees,’’ with type 1 or 2 diabetes in 1997 and listed on the GH Diabetes Registry, and alive on December 31, 2007. To enter the registry, enrollees met at least one of the following criteria in the previous 12 months: (1) filled prescriptions for insulin or an oral hypoglycemic agent; (2) had one HbA1c > 7.0 (53 mmol/ mol); (3) had 2 or more fasting plasma glucose levels  126 mg/ dl; (4) had 2 or more random plasma glucose levels 200 mg/dl; (5) had any combination of 2 fasting plasma glucose and random plasma glucose over the previous limits; or (6) had two outpatient diagnoses of diabetes, or any inpatient diagnoses of diabetes. Enrollees with gestational diabetes were ineligible. The patient population for our retrospective, longitudinal cohort study consisted of 9871 enrollees who were on the GH Diabetes Registry in 1997 and obtained care in western Washington GH practices in all years from 1997 until 2006, or until their deaths. We excluded enrollees with end stage renal disease (n = 158). If competing risks of mortality existed, estimates of the relative contributions of diabetes complications and comorbid hypertension, CAD, CHF and depression to mortality may be biased. Therefore, we identified and excluded 893 enrollees (9%) with residual comorbid conditions (other than hypertension, CAD, CHF and depression) at baseline (1997) that may be predictive of death. Enrollees with the following non-mutually exclusive, residual comorbid conditions were excluded using the Deyo version of the Charlson comorbidity index [28], a validated index predicting the likelihood of mortality based on 17 comorbid disorders: dementia (N = 118), peptic ulcer disease (N = 117), mild-to-severe liver disease (N = 86), malignancy (including leukemia and lymphoma, N = 606), metastatic solid tumor (N = 100), and AIDS (N = 5). The final analytic sample size was 8820 enrollees and 270,463 enrollee-quarters. Study protocols were approved by Group Health Human Subjects Review Committee.

2.2.

Measures and data sources

Enrollee characteristics and mortality. We measured the enrollee’s age and gender from Group Health records. Mortality was measured using date of death in the Group Health Death Register. The original source of death data was the Washington State Death Certificate System in the Washington State Department of Health.

diabetes research and clinical practice 106 (2014) 373–382

Diabetes complications and comorbid conditions. Diabetes with ophthalmic or neurologic complications at baseline was measured using codes in the Charlson index [29]. Common comorbid conditions were measured from medical records as documented below:  Depression: antidepressant prescription associated with a visit diagnosis of depressive disorder (including physician diagnoses of major depression, dysthymia), and depressive disorder not otherwise specified [30–32];  Hypertension: at least 2 outpatient visits with primary or secondary diagnosis of hypertension (ICD-9 codes 401.0– 401.9), 1 hypertension prescription, not pregnant, and without end-stage renal disease [33,34];  Coronary artery disease (CAD): patient enrolled on the GH Secondary Heart Care Registry. Registry eligibility criteria include a hospital discharge diagnosis of either myocardial infarction, coronary bypass grafting, percutaneous transluminal coronary angioplasty, or unstable angina with age 30; and  Congestive heart failure (CHF): enrollees with an outpatient or hospital discharge ICD-9 diagnosis of CHF (401.91, 402.01, 402.11, 402.91, 404.01, 404.03, 404.11, 404.13, 404.93, 428.0) [35,36]. Because of the chronicity of hypertension, CAD and CHF, we assumed that after condition onset, the condition persisted for the remainder of the time series. For each calendar quarter, we assigned enrollees to one of six mutually exclusive comorbidity groups: (1) no comorbid conditions (diabetes only); (2) hypertension and/or CAD; (3) CHF with or without hypertension or CAD; (4) depression; (5) depression and hypertension and/or CAD; and (6) depression and CHF with or without hypertension or CAD. Comorbid groups for diabetes and depression with hypertension, CAD or CHF were created because coexisting depression contributes to mortality beyond cardiovascular deaths [37]. We also measured chronic pulmonary disease (CPD) using the Charlson index because CPD was common in enrollees, and CPD may be related to mortality [28,29]. Diabetes management. Diabetes management was measured quarterly by HbA1c. If a patient had 2+ HbA1c scores in a quarter, the last score was chosen. We computed HbA1c deciles and constructed 10 binary variables indicating an enrollee’s decile in each quarter [20]. Utilization. Enrollee face-to-face utilization was collected from the Center for Health Studies’ Data Warehouse. Enrollees had financial incentives to seek care within the integrated system, and in those instances where care was received from contract providers, utilization was measured through claims. Quarterly in-person utilization measures for 1998–2006 included primary care visits, specialty visits, emergency room visits, inpatient admissions, and prescriptions filled. Enrollee utilization of secure messaging was measured in 2003–2006 after installation of the EpicCare1 electronic medical record system. The EpicCare1 system has an enrollee Web site, ‘‘MyGroupHealth,’’ that allows Group Health enrollees to exchange electronic messages with their primary providers, as well as their entire healthcare team (see URL: https:// The patient-provider member.ghc.org/open/index.jhtml). emails are called ‘‘secure messages’’ because of the privacy

375

protection safeguards in the Web site. In each quarter we measured the number of secure message threads for each enrollee (a thread is a set of messages including an original message plus any messages descending from it as replies); an ‘‘original’’ message is one that is not created in reply to another message [38].

2.3.

Data analysis

Descriptive statistics were computed for enrollee characteristics at baseline (quarter 1 of 1998). We calculated 9-year survival curves by the Kaplan–Meier method by time-varying comorbid group, and differences between the curves were tested with the log rank statistic. Frequency distributions revealed that HbA1c was observed in 46% of quarters. Enrollee quarters with ‘mistimed’ (that is, missing) observations occurred because all enrollees did not have clinic visits and clinical assessments of HbA1c in all quarters [39]. We assigned HbA1c values to quarters with mistimed observations through a clinical guideline approach, followed by multiple imputation. Clinical guidelines recommended HbA1c tests at least twice a year (6 months on average) for patients with stable glycemic control [1]. Applying the guideline, if an enrollee had HbA1c < 7% (53 mmol/mol) in the current quarter and HbA1c was missing in the next quarter, we carried forward the current quarter HbA1c score to the next quarter. This protocol carried forward HbA1c scores for 22,229 quarters, or 18% of the missing quarters (n = 22,229/122,438). We used Stata version 12 statistical software to perform multiple imputation for the remaining mistimed clinical assessments for HbA1c, following guidelines in Sterne et al. [40]. In brief, multiple imputation consisted of three steps. In the first step, multiple copies of the dataset were created, with the missing values for HbA1c replaced by imputed values. In the second step, the regression analysis was performed for each one of the imputed datasets. In the third step, the regression estimates from each imputed dataset were averaged together to produce overall estimated associations, with standard errors computed using Rubin’s rules, which take account of the variation in results across the imputed datasets. In the first step the imputation models included HbA1c and non-missing variables (age, gender, indicators for the six comorbidity groups, diabetes complications, death, time to death (log scale), and utilization measures) to increase the plausibility of the missing-at-random assumption [40]. Twenty imputed datasets were created, each containing 270,463 enrollee quarters, or a total of 5,409,260 enrollee quarters. HbA1c decile measures were calculated using data from the 20 imputed datasets combined. We used Cox proportional hazards models to estimate the association of HbA1c with all-cause mortality in the 20 multiple imputed datasets. We estimated three multivariate models, with each sequential model including additional covariates. In the first model we estimated the relative contribution of HbA1c deciles (reference group: HbA1c scores 7.1% (54 mmol/mol) and

Hemoglobin A1c, comorbid conditions and all-cause mortality in older patients with diabetes: a retrospective 9-year cohort study.

To examine whether hemoglobin A1c levels and comorbid conditions are related to all-cause mortality in a cohort of patients with type 1 or 2 diabetes ...
502KB Sizes 2 Downloads 6 Views