Paradigm Shift in Digital Health Assessment: Reconstructing the Causal Logic of Telemedicine Effectiveness Research

Abstract

The COVID-19 pandemic resulted in a dramatic change to healthcare around the world due to the rapid scaling of the use of telemedicine into a central position within the healthcare systems. However, with this rapid scaling, the available scientific evidence as it relates to the clinical effectiveness, cost-effectiveness, and equity of telemedicine presents disturbing fragmentation and heterogeneity. The crisis in the evidence is largely due to the statistical bias generated by the bidirectional selection processes inherently present in much of the observational data that exists on telemedicine, as well as the structural bias resulting from the digital divide. The objective of this article is to systematically reorganize the methodological landscape concerning causal inference within the field of telemedicine research. To accomplish this goal, this article follows four steps. First, it describes how "healthy user bias" and "severity bias" interfere with traditional association analyses. Second, it assesses the evolution of empirical paradigms from classic causal adjustment techniques to quasi-experimental methods. Third, it examines the potential of Causal Machine Learning (Causal ML) in capturing heterogeneous treatment effects. Finally, it proposes the Target Trial Emulation (TTE) framework as the "gold standard" for enhancing evidence quality. As shown in this review of current research practices, the more rigorous the causal inference design, the better the elimination of complex confounding factors and thereby the more evidence available for formulating digital health policy.