Comparative Effectiveness of Metformin for Type 2 Diabetes With Chronic Kidney Disease

Specific aims are as follows: Aim 1. For patients with Type 2 Diabetes Mellitus(T2DM) and Chronic Kidney Disease (CKD), compare metformin to alternative non-insulin diabetes drugs (sulfonylureas, DPP-4 inhibitors, GLP-1 receptor agonists, and SGLT-2 inhibitors)) with respect to the key safety outcome of severe hypoglycemia. The primary hypothesis is that metformin will be superior to sulfonylurea in terms of severe hypoglycemia rates, and non-inferior to DPP-4 inhibitors. Secondary outcomes will include hospitalization for acidosis, hospitalization for hyperglycemia, acute myocardial infarction, stroke, heart failure hospitalization, and heart failure emergency room visit. Aim 2. For patients with T2DM and CKD, compare metformin to alternative non-insulin diabetes drugs with respect to HbA1c reduction. The primary hypothesis is that metformin will be superior to DPP-4 inhibitors and non-inferior to sulfonylureas for HbA1c reduction (i.e., improvement in blood sugar). Secondary outcomes will include change in body-mass index (BMI) and kidney function, non-persistence to treatment, and progression to insulin use. Aim 3. Examine the heterogeneity of treatment effects on hypoglycemia risk and HbA1c response across patient subgroups. The primary hypothesis is that metformin's advantages will be more pronounced in more severe CKD. Aim 4 (data completeness): Using the linked Medicare-CDRN dataset, assess completeness of CDRN data for drugs and hospitalization among Medicare recipients. Specifically, we will use Medicare data from 2013-2016 as a gold standard and assess the sensitivity, specificity, negative predictive value (NPV), and positive predictive value (PPV) of INSIGHT CDRN data from those years in identifying hospitalizations and prevalent diabetes drug use among Medicare patients with T2DM (type 2 diabetes mellitus), and develop and validate an algorithm to identify patients for whom NPV and PPV for hospitalizations and all major diabetes drug classes exceed 80%. We hypothesize that such an algorithm will identify a population of a quarter of eligible Medicare patients in the CDRN who meet or exceed these standards for completeness of data. Aim 5. Conduct a cohort study to test the hypothesis that poorly controlled baseline HbA1c is not independently associated with the primary outcome (hospitalization with COVID). Primary analysis will be restricted to Medicare patients identified by aim 1 as likely to surpass 80% NPV and PPV for the primary outcome. A finding that HbA1c is not associated with worse outcomes would support relaxing glycemic targets during the pandemic when this allows patients to avoid unnecessary risks associated with aggressive treatment, monitoring, and exposure to the health care system. Aim 6: Conduct a comparative cohort study to test the null hypothesis that metformin, SGLT-2 inhibitors, DPP-4 inhibitors, GLP-1 receptor agonists, and sulfonylureas do not have class-specific effects on the primary outcome. Primary analysis will be restricted to Medicare patients identified by aim 1 as likely to surpass 80% NPV and PPV both for the primary outcome and exposure to the drug of interest. Two sub-hypotheses would be of special interest: if SGLT-2 inhibitors are associated with increased risk, this would support suggestions that they be temporarily stopped in high risk patients; if DPP-4 inhibitors are associated with decreased risk, this would argue for prospective research into this class as protective agents.