For patients with diabetes, hemoglobin A(HbAtreatment goals are based primarily on evaluating the outcomes and responses of the average patient. However, the American Diabetes Association and the American Association of Clinical Endocrinologists now recommend individualizing treatment regimens and HbA goals by including patient-centered and sociodemographic characteristics that might impact heterogeneity of treatment effect (HTE) for diabetes treatment.
We used pooled data from randomized controlled trials (RCTs) in diabetes and from Partners HealthCare electronic health records (EHRs) to estimate HTE and to provide performance benchmarks across clinical, sociodemographic, quality-of-life (QOL), and treatment satisfaction (SAT) patient subgroups to identify potential management strategies for improving the quality of diabetes care.
Data from 19 previously conducted diabetes RCTs (6954 patients, 1002 clinics, 38 diabetes treatments) provided HbA best-practice, high benchmark probabilities (HBPs) for reaching HbA goals of <7.0% and <8.0%. EHRs, including those from 8107 diabetes medication-naive patients followed between 2000 and 2014, provided usual-care clinical performance probabilities (CPPs) for reaching the same HbA end points before any treatment modification. Linear, logistic, and survival regression models were used to estimate the effects of clinical and sociodemographic variables, such as age, sex, body mass index (BMI), baseline HbA, race/ethnicity, and diabetes treatments, on change in HbA, odds of achieving HbAtarget, and time to HbAtarget. Patient-reported QOL and SAT outcomes, HbA levels, and sociodemographic variables were analyzed simultaneously using latent variable structural equation modeling (SEM) to assess the impact of mediators and treatment effect modifiers on HTE. Using the estimates obtained from these statistical models, we deployed interactive web-based, online calculators on a social media learning platform and used them to facilitate the dissemination, implementation, and evaluation of the study findings.
The results for 594 patients from 62 clinics showed that after 12 weeks on either diet and exercise (DE) plus a placebo (DE+P) or DE plus a sulfonylurea (DE+S), the overall unadjusted mean (SD) HBP estimate for reaching an HbA <8.0% for DE+P was 0.28 (0.29) (n = 184), and for DE+S, it was 0.67 (0.32) (n = 359). The adjusted HBP estimates for a subgroup of White men (age, 50 years; fasting plasma glucose [FPG], 150 mg/dL; diabetes duration, 1 year; BMI, 30; and baseline HbA, 9%) were 0.06 and 0.48 for HbA goals of <7.0% and <8.0%, respectively, with DE+P, compared with 0.51 and 0.94 with DE+S, respectively. For Black women of the same age and with the same FPG and duration of diabetes, but with a BMI of 36 and baseline HbA of 9.5%, the HBPs for HbA <7.0% and <8.0% were 0.04 and 0.18 with DE+P and 0.41 and 0.78 with DE+S, respectively. In the same study, mean (SD) clinic-specific HBPs for the DE+S group were 0.35 (0.18) and 0.67 (0.17) for HbA goals of <7.0% and <8.0%, respectively. RCT models (2927 patients, 413 clinics, 18 arms) showed that improvements in QOL, SAT, and subscales of mental health and perceived health status were associated with increased odds of achieving HbA target goals of <7.0% and <8.0% (odds ratios [ORs] between 1.7 and 1.9 for QOL and SAT scales; all < .05). For each SD unit improvement in overall QOL, there was a 10% increase in the odds of reaching an HbA goal of <8.0% (OR, 1.10; 95% CI, 1.03-1.18). Improved perceived effectiveness, symptom interference, mental health, health perceptions, cognitive performance, and functional health (all < .02) facilitated achieving a more favorable HbA response. Latent variable SEM was undertaken for the full RCT sample of 5209 patients with type 1 and type 2 diabetes, 712 centers, and 32 regimens. Estimated path coefficients indicated that a 1-SD higher (better) score in baseline psychological distress corresponded to an 11% SD greater improvement in the diabetes Side Effects and Symptoms (SES) Distress Scale. In turn, a 1-SD improvement in this scale corresponded to a 15% SD decrease (better glycemic control) in HbA. (2477 patients). The CPPs of reaching HbAtargets of <7.0% and <8.0% before any medication change were 0.41 and 0.54 with sulfonylurea and 0.56 and 0.72 with metformin, respectively, for the Black female group described above. These CPPs could serve as benchmarks for a clinical practice to strive to meet. EHR models for the 8107 medication-naive patients indicated that the largest variability in TEH was among the different classes of diabetes medications.
Multiple RCT and EHR models confirmed the substantial variability in treatment response that is influenced by demographic, socioeconomic, and patient-reported QOL and SAT characteristics. Even in a well-controlled, highly resourced, and strict-protocol RCT setting, treatment response varies substantially. The probability of reaching HbA goals in benchmark practices is a measure of quality against which clinical practices can measure themselves. Moreover, potentially modifiable patient factors that drive success in achieving HbA goals, such as anxiety, depression, and symptom distress, can become targets for intervention to improve diabetes care.
Best-practice and usual-care benchmarks were based on clinical trials and the Partners HealthCare EHR data; neither is a nationally representative sample. Due to the fixed-visit-interval study design of the RCT in contrast to the EHR observational design, a direct head-to-head comparison of the 2 benchmark metrics cannot be made.
