Abstract
BACKGROUND: Observational studies play an important role in assessing the comparative effectiveness of competing treatments. In clinical trials the randomization of participants to treatment and control groups generally results in balanced groups with respect to possible confounders, which makes the analysis straightforward. However, when analysing observational data, the potential for unmeasured confounding makes comparing treatment effects more challenging. METHODS: Causal inference methods such as Instrumental Variable and Prior Event Rate Ratio approaches enable the estimation of causal effects even in the presence of unmeasured or imperfectly measured confounding factors. Direct confounder adjustment via multivariable regression and propensity score matching also have considerable utility. Each method relies on a different set of assumptions and leverages different aspects of the data.The assumptions of each method are described, and the impact of their violation is assessed in a simulation study. We propose the prior outcome augmented Instrumental Variable method that leverages data from before and after treatment initiation and is robust to key assumption violations. Finally, we propose a heterogeneity statistic to decide if two or more estimates are statistically dissimilar, considering their correlation. We illustrate our framework in an application study assessing the risk of genital infection in type 2 diabetes patients prescribed SGLT2-inhibitors versus DPP4-inhibitors using UK primary care data. RESULTS: Our proposed approach can estimate treatment effects without bias in scenarios where assumptions of other methods are violated. Furthermore, the application study exemplified the usefulness of discussing the consistency of estimation results from different estimation methods using triangulation. CONCLUSION: Triangulating results of different estimation methods is important in observational data to derive high quality evidence. The proposed triangulation framework and heterogeneity statistic are valuable tools to discuss the consistency of estimation results from different methods to shed light on possible sources of bias.