Prescribing trends, utilisation and comparative effectiveness of antidiabetic drug therapies prescribed in UK primary care

Background: Over the last decade, several new treatments for type 2 diabetes have launched for routine care in the United Kingdom (UK). Little was known however about how general practitioners (GPs) prescribed these drugs, why GPs chose each therapy, or how they compared in terms of effectiveness. Methods: This thesis describes prescribing trends and comparative effectiveness of type 2 diabetes drugs, using routine UK clinical care data and observational study designs. Results: All analyses used a cohort of 280,241 individuals treated in UK primary care between 2000 and 2017. First, I show rapid changes in prescribing practice for type 2 diabetes drugs. After metformin monotherapy, GPs have increasingly added new drug classes, in particular the Dipeptidyl peptidase 4 inhibitors (DPP4is) and sodium glucose cotransporter-2 inhibitors (SGLT2is). From their launch in 2007 and 2012, by 2017 prescriptions for DPP4is and SGLT2is grew to 42% (95% CI: 38, 47) and 22% (95% CI: 17, 27) of therapy intensifications. To describe patient characteristics associated with treatment intensification with a sulfonylurea (SU), a DPP4i or a SGLT2i, I applied multinomial regression analysis. I found inequalities by socioeconomic status and ethnicity. People of South Asian ethnicity had lower odds of receiving SGLT2is compared to SUs, odds ratio: 0.6 (95% CI: 0.4, 0.9). Finally, I used a new-user propensity score matched design to contrast intensification with SUs, DPP4is and SGLT2is and reveal differences in clinical variables. People prescribed SGLT2is had the greatest falls in glycated haemoglobin (HbA1c) at 60 weeks, of 16.1 mmol/mol (95% CI: 18.7, 13.5), compared to SUs and DPP4is (13.8 mmol/mol (95% CI: 15.4, 12.2) and 9.8 mmol/mol (95% CI: 11.6, 7.9) respectively). Conclusions: Electronic health records offer important clinical information for clinicians, and answer questions not investigated by randomised controlled studies. This work addresses methodological challenges in drug research using electronic health records