The Cardiovascular Health Study (CHS) is a National Heart, Lung, and Blood Institute (NHLBI)-funded prospective study designed to assess the traditional and non-traditional cardiovascular risk factors among community-dwelling older adults.8
The CHS recruited 5888 Medicare-eligible community-dwelling adults ≥65 years of age from four US communities in two phases. A mostly white initial cohort of 5201 participants (1989–1990) was later supplemented by 687 African-Americans from three of those four communities (1992–1993). We used a de-identified public-use copy of the CHS dataset obtained from the NHLBI which contained information on 5795 participants who consented to be included in that dataset. After excluding 63 participants without data on DM status and 268 participants with prevalent HF at baseline, the final sample size for the current analysis was 5464 participants.
Baseline DM was defined by fasting plasma glucose (FPG) level >126 mg/dl or treatment with insulin or hypoglycemic drugs, and 16% (862/5464) of the CHS participants had DM. Data on socio-demographic, clinical, sub-clinical, and laboratory variables including serum insulin, triglyceride, interleukin-6 (IL-6), and C-reactive protein (CRP) levels were measured at baseline.8
If the value of a continuous variable was found to be missing, then predicted values based on age, sex and race were imputed. The primary outcome for this study was incident HF, which was centrally adjudicated by the CHS Events Committee. Data on self-reports of physician diagnosis of HF were obtained during semi-annual visits, which was then verified via review of medical records.2, 9, 10
Secondary outcomes included all-cause and cause-specific mortalities, acute myocardial infarction (AMI), stroke, and peripheral arterial disease (PAD).
Propensity scores, or the conditional probability of having DM, were estimated for each of the 5,464 participants using a non-parsimonious multivariable logistic regression model in which DM was the dependent variable and the 65 baseline characteristics were covariates.11–14
We then used the propensity scores to match 717 (83% of the 862) individuals with DM with 717 of those without DM who had similar propensity scores.15–18
Pre- and post-match absolute standardized differences for all 65 covariates were estimated and presented as a Love plot ().19–23
An absolute standardized difference of less than 10% indicates inconsequential imbalances, while 0% indicates no between-group imbalances on that covariate.24, 25
Figure 1 Absolute standardized differences comparing 65 baseline characteristics between CHS participants with and without diabetes mellitus, before and after propensity score matching.(ACE = angiotensin-converting enzyme; COPD = chronic obstructive pulmonary (more ...)
For between-group comparisons for pre- and post-match data, we used Pearson chi-square tests, Wilcoxon rank-sum tests, McNemar’s tests and paired sample t-tests, as appropriate. Kaplan-Meier and matched Cox proportional hazard analyses were used to estimate the associations between DM and outcomes. Formal sensitivity analyses were conducted to determine the impact of a potential hidden confounder on the association between DM and incident HF in the matched cohort.26
Subgroup analyses were performed to determine the homogeneity of this association. Two-tailed statistical tests with 95% confidence intervals were employed with a p-value <0.05 considered to be significant. All data analysis was completed using SPSS for Windows (Version 15).