Effects of Dipeptidyl Peptidase-4 Inhibitors and Sulfonylureas on Cognitive and Physical Function in Nursing Home Residents

Study Design and Data Source

This was a retrospective new-user cohort study that used the following linked national datasets for the years 2007 – 2010: 100% Medicare fee-for-service enrollment information, Part A inpatient claims, and Part B outpatient claims; 20% Part D prescription drug claims; 100% Minimum Data Set (MDS) version 2.0 assessment records; and 100% Online Survey Certification and Reporting System (OSCAR) data. The MDS is a comprehensive, clinical assessment instrument used to document health status of NH residents, including functional status, psychological, and cognitive status information.³¹ MDS assessments occur at a minimum of every 3 months, and are conducted more frequently for residents with a major recent change in clinical status. OSCAR data provided NH-level information. The study was designed to mimic the hypothetical target trial detailed in Supplementary Table S1.³² This study was approved by the Brown University Institutional Review Board.

Study Population

The study population was adults aged ≥65 years who were long-stay NH residents (>100 days in the NH) on January 1, 2008, or who became a long-stay resident between January 1, 2008 and September 30, 2010. The index date (time zero) was the date of the first eligible dispensing of a DPP4I or SU between January 1, 2008 and September 30, 2010 after becoming a long-stay resident. Data from July 1, 2007 to December 31, 2007 was available to ascertain prior glucose-lowering treatment use. All individuals were required to have one year of continuous enrollment in Medicare fee-for-service Parts A, B, and D immediately preceding the index date. Cohort exclusions are shown in Supplementary Figure S1 and Table S1.

Exposures and Causal Contrast of Interest

Exposures of interest were new use of DPP4Is (saxagliptin, sitagliptin) or SUs (glimepiride, glipizide, glyburide) in NHs. New use was defined as the first Part D claim for a DPP4I or SU after 6 months without a dispensing for any glucose-lowering treatment other than metformin. As metformin is generally first-line therapy but is commonly contraindicated in the NH setting³³, its use was permitted and adjusted for but not required for inclusion. The causal contrast of interest was defined as the effect of initiating DPP4Is versus SUs regardless of subsequent treatment discontinuation or switching (i.e., the observational study analog of the intention-to-treat [ITT] estimand).

Outcomes

The outcomes were decline in physical functioning, decline in cognitive functioning, and hospitalizations or emergency department (ED) visits for altered mental status. We defined physical functional decline as an increase of 3 points on a validated 28-point scale of independence in activities of daily living (ADLs) between the pre-initiation baseline and any available MDS assessment following initiation, up to 180 days after initiation.³⁴ A 3-point increase corresponds to a major loss of independence in 1 ADL or incremental loss in 2 or more ADLs.³⁵ Cognitive decline was defined as an increase of 1 point on the validated 6-point MDS Cognitive Performance Scale (CPS). A 1-point cognitive decline is considered clinically significant.³⁶ Lastly, we defined a hospitalization or ED visit for altered mental status using International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes 780.0, 780.02, 780.09, or 780.97 in any position on a Part A inpatient hospital claim or a Part B ED claim, though inclusion of ED visits contributed only a small number of additional cases. While no validated algorithm exists for altered mental status events in claims, our definition is consistent with prior literature.³⁷ We analyzed each outcome individually. We also explored a pre-specified composite outcome: time to functional or cognitive decline (3-point ADL or 1-point CPS).

Follow-Up

The start of follow-up (baseline or time zero) for each individual was the day of the first DPP4I or SU dispensing and continued until the earliest event of the following: Medicare disenrollment (from Parts A, B, or D), enrollment in a health maintenance organization (Medicare Advantage), death, an outcome (each evaluated separately), or administrative end of follow-up (September 30, 2010 or 180 days of follow-up). We chose a 180-day outcome period because it is long enough to be clinically meaningful, but short enough that many of these highly vulnerable residents have not yet died. Death is a prevalent competing event that complicates interpretation of longer-term outcomes in the NH setting.

Baseline characteristics

We identified 190 baseline characteristics that could potentially confound the relationship between receiving DPP4Is versus SUs and the outcomes (Supplementary Table S2). All variables were pre-specified and measured using the most recent available record on or before the index date (time zero).^{21, 22, 33} These variables were obtained from Medicare claims and MDS v2.0 data, which has well-established reliability and validity for assessing the clinical status of NH residents. The MDS v2.0 also provided data on other patient characteristics including pre-treatment physical function³⁴ and cognitive status, geriatric syndromes, nutrition, social characteristics, care preferences, and a mortality risk index: Changes in Health, End-stage Disease, and Symptoms Scale (CHESS) score.³⁸ We used the OSCAR data to evaluate a variety of NH facility characteristics such as staffing, resident mix, and quality indicators (Supplementary Table S2).

Statistical Analyses

We adjusted for potential confounding by baseline covariates by estimating the propensity scores using a logistic regression that included 190 baseline characteristics to predict the initiation of DPP4Is versus SUs. We matched DPP4I to SU users on the propensity scores using a 1:1 greedy (nearest neighbor) 5-to-1 digit matching algorithm without replacement.³⁹ Cox regression models with robust standard errors to account for clustering within the matched sets were used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) comparing DPP4I versus SU users.⁴⁰ No covariates were included in the model. Testing for violations of the proportional hazards assumption was unnecessary because we interpret the HRs as a weighted average of the true HRs over the entire follow-up period.⁴¹ We used SAS, version 9.4 (SAS Institute Inc) for data processing, and Stata version 14.0 (Stata Corp., College Station, TX) and R version 3.4.4 (The R Foundation) for data analyses.

Stability Analyses

We conducted several stability analyses to test the robustness of our treatment effect estimates to study design and analysis decisions. First, we evaluated more substantial declines by re-defining functional decline as a 4-point decrease and cognitive decline as a 2-point decrease along with a composite outcome of time to larger functional declines (4-point ADL or 2-point CPS). Second, to assess the impact of missing data, we performed multiple imputation with chained equations to impute missing covariate data for 582 residents excluded from the primary analysis. Third, we estimated the propensity scores using generalized boosted regression models to evaluate whether our parametric propensity score estimation model might have been mis-specified. Fourth, we used Fine and Gray competing risks regression models to account for the potential competing risk of death. At the end of each follow-up period, subjects were classified as alive without one of the outcomes of interest, having had the outcome documented in the MDS or claims in that period, or having died without evidence of an outcome event. Lastly, because metformin is often the preferred first-line medication used to treat T2DM, we restricted our study cohort to individuals using metformin at baseline and performed the analyses in this subpopulation.

Cohort

Before matching, the initial cohort included 903 new DPP4I users and 6,075 new SU users, and these groups had the same average baseline values of ADL (mean=16) and CPS scores (mean=3). DPP4I initiators had a higher medication burden, used other glucose-lowering treatments more frequently, and had more angiotensin receptor blocker, clopidogrel, fibrate, and omega-3 fatty acid medication use in the prior 12 months than new SU users (Table 1 and Supplementary Table S2). Additionally, DPP4I users were more likely to have abnormal laboratory results, ischemic heart disease, and a prior ED visit for hyperglycemia, but less likely to have a do not resuscitate order.

Propensity score matching yielded a cohort of 892 new DPP4I users and an equal number of SU users (N=1,784; Table 1). The mean age (SD) was 80 (SD=8) years. Approximately 73% of the cohort was female and 72% was White race, and 46% had no or mild cognitive impairment (CPS score 0-2) and 35% had no or mild functional impairment (ADL score 0-14) before DPP4I or SU initiation. The matching procedure balanced covariates closely⁴² with all but 2 variables having absolute standardized mean differences of 0.06 or less (Supplementary Table S2). The propensity score distribution between new initiators of DPP4Is and of SUs overlapped substantially before and after matching (Supplementary Figure S2).

Effects on Outcomes

The rates of cognitive decline and altered mental status were not statistically different between DPP4I users and SU users (Table 2, Figure 1, Supplementary Table S3). The rate of cognitive decline among DPP4I users was 0.61 (95%CI 0.31-1.19) times that of SU users, and the rate of an altered mental status event was 0.71 (95%CI 0.39-1.27) times that of SU users (Table 2, Figure 2). DPP4I users were as likely as SU users to have a physical functional decline at 180 days after treatment initiation (HR=0.89, 95%CI 0.51-1.56). The rates of the composite outcome of cognitive or functional decline were not statistically different between DPP4I user and SU users (180 days, HR=0.83, 95%CI 0.53-1.31).

• Download figure
• Open in new tab

Figure 1. Survival Curves for Cognitive Decline, Altered Mental Status Event, Functional Decline, and Cognitive or Functional Decline Composite Outcomes over 180 Days of Follow-up Stratified by Dipeptidyl Peptidase-4 Inhibitor versus Sulfonylurea Use after Propensity Score Matching among Nursing Home Residents.

Panel A shows time to cognitive decline among DPP4I users and SU users. Panel B shows time to hospitalization or emergency department visit for altered mental status. Panel C shows time to functional decline. Panel D shows time to the first of either a cognitive or functional decline. Red lines are DPP4I users; blue lines are SU users. Please refer to Supplementary Table S3 for the corresponding risk table for each survival curve.

• Download figure
• Open in new tab

Figure 2. Effects of Dipeptidyl Peptidase-4 Inhibitor versus Sulfonylurea Use on 180-day Outcomes After Propensity Score Matching among Nursing Home Residents.

Note: Composite denotes a composite outcome of time to either cognitive or functional decline. Abbreviations: HR, hazard ratio; CI, confidence interval; DPP4I, dipeptidyl peptidase-4 inhibitor; SU, sulfonylurea.

Stability Analyses

Analyses re-defining cognitive and functional decline using thresholds representing larger changes (2-point increase in CPS score or 4-point increase in the Morris ADL scale) resulted in similar point estimates with decreased precision for the individual outcomes as well as the composite that reflected the most severe measures of cognitive or functional decline (Supplementary Table S4). The main results were generally similar when implementing multiple imputation of missing baseline covariate information (Supplementary Table S5), estimating the propensity score using generalized boosted regression models (Supplementary Table S6), employing Fine and Gray regression models to account for the competing risk of death (Supplementary Table S7), and restricting to metformin users (Supplementary Table S8).