Background: Electronic health records (EHR) are increasingly used to investigate epidemiologic research questions, as they provide high power and easily available data. Recent studies found a dramatic reduction of healthcare consumptions during the C19 pandemic. Recommendations were published on how to conduct clinical studies during the C19 pandemic, but no equivalents exist for observational studies using EHR.
Objectives: To give insights and tentative recommendations on statistical analyses to perform for coming observational studies not assessing C19 using EHR during the C19 pandemic period.
Methods: Sources of bias will be highlighted in the context of C19 pandemic with two different focuses: the measure of incidence of pathologies and the association between an exposure and an outcome. The statistical analyses to evaluate the magnitude of these biases and to account for them will be described. An illustration on surgical data will be provided.
Results: The incidence estimation of diseases is expected to decrease in 2020-2021 compared to previous years, due to a bias of detection. Instead of providing the observed number of incident cases, the expected number of new cases could be produced using projections with temporal series or age-cohort-period models. The temporal series of weekly/monthly number of newly exposed patients and patients with outcome are compared with the same series identified on several previous years. Hypotheses behind significant changes should be evaluated (pandemic burden on medical resources, lock-down, social distancing). Confounding bias may occur if C19 affects outcome and exposure. Before/after designs should be avoided and case-control preferred. The estimation should be performed overall and by civil year using mixed models. A dramatic change in association in 2020 and 2021 compared to other years will prevent to use the pandemic period in estimations.
Conclusions: EHR are a powerful tool that should continue to be harnessed to investigate varied medical research questions and not be limited to their already proven usefulness in evaluating the C19 pandemic. While extra care is warranted when examining endpoints unrelated to C19 during the pandemic, statistical tools already exist to carefully address methodological challenges that may arise in observational study settings.
