Targeted inference to identify drug repositioning candidates in the Danish health registries

Abstract

Electronic health records can be used to track diagnoses and drug prescriptions in large heterogeneous populations over time. Coupled with recent advances in causal inference from observational data, these records offer new opportunities to emulate clinical trials and identify potential targets for drug repositioning. Here, we run a hypothesis generating cohort study of Danes aged 50 to 80 years from 2001 to 2015 (n = 2,512,380), covering a total of 23,371,354 years of observations. We examine prescription drugs at ATC level-4 and their effect on 9 major disease outcomes. Using Bayesian time-varying Cox regression and longitudinal minimum loss estimation, our analysis successfully reproduces known drug-disease associations from clinical trials, such as the reduction in the 3-year absolute risk of death associated with Statins (ATC:C10AA) -0.8% (95% CI =[-1.2%, -0.5%]) and -0.8% (95% CI =[-1.3%, -0.2%]) for females and males, respectively. Additionally, we discovered novel associations that suggest potential repositioning opportunities. For instance, Statins were associated with a reduction in the 3-year absolute risk of dementia by -0.3% (95% CI =[-0.5%, -0.1%]) for females and -0.2% (95% CI =[-0.4%, 0.1%]) for males. Furthermore, Biguanides (ATC:P01BB) stands out as a particularly interesting candidate with absolute risk reductions across various outcomes. In total, we identified 76 potential drug-disease pairs for further investigation. However, it should be stressed that the emulation of clinical trials here is solely of hypothesis generating nature and identified effects need to be corroborated with additional evidence, preferably from RTCs, as the risk of confounding by indication in this study is substantial. In summary, this study provides a large-scale screen of prescribed drugs and their effect on major debilitating disease in the Danish health registries. This provides an additional source of information that can be used in the search for possible repositioning candidates.