Advancing geospatial preconception health research in primary care through medical informatics and artificial intelligence.

Established life course approaches suggest that health status in adulthood can be influenced by events that occurred during the prenatal developmental period. Yet, interventions such as diet and lifestyle changes performed during pregnancy have had a small impact on both maternal and offspring health outcomes. Currently, there is a growing body of literature that highlights the importance of maternal health before conception (months or years before pregnancy occurs) for the future health of offspring. While some studies have explored factors such as maternal body composition, nutrition, and lifestyle in this area, location-based environmental and socioeconomic exposures before conception may also contribute to future offspring health. In this line, the study of a patient's geographic history presents a promising avenue. To foster research in this direction, the integration of geospatial health, medical informatics and artificial intelligence techniques offers great potential. Importantly, novel sources of big health data sets such as electronic health records registered at the primary care level provide a unique framework due to its inherent longitudinal nature. Nonetheless, a number of privacy, ethical, and technical challenges need to be overcome for this kind of longitudinal analysis to mature and succeed. In the long-term, we support the vision of incorporating a patient's geographic history into her clinical history to equip clinicians with useful contextual information to explore.

Applying two approaches to detect unmeasured confounding due to time-varying variables in a self-controlled risk interval design evaluating COVID-19 vaccine safety signals, using myocarditis as a case example.

We test the robustness of the self-controlled risk interval (SCRI) design in a setting where time between doses may introduce time-varying confounding, using both negative control outcomes (NCOs) and quantitative bias analysis (QBA). All vaccinated cases identified from 5 European databases between 1 September 2020 and end of data availability were included. Exposures were doses 1-3 of the Pfizer, Moderna, AstraZeneca, and Janssen COVID-19 vaccines; outcomes were myocarditis and otitis externa (NCO). The SCRI used a 60-day control window and dose-specific 28-day risk windows, stratified by vaccine brand and adjusted for calendar time. The QBA included two scenarios: (i) baseline probability of the confounder was higher in the control window and (ii) vice versa. The NCO was not associated with any of the COVID-19 vaccine types or doses except Moderna dose 1 (IRR = 1.09, 95%CI 1.01-1.09). The QBA suggested even the strongest literature-reported confounder (COVID-19; RRmyocarditis = 18.3) could only explain away part of the observed effect from IRR = 3 to IRR = 1.40. The SCRI seems robust to unmeasured confounding in the COVID-19 setting, although a strong unmeasured confounder could bias the observed effect upward. Replication of our findings for other safety signals would strengthen this conclusion.

A comparison of four self-controlled study designs in an analysis of COVID-19 vaccines and myocarditis using five European databases.

INTRODUCTION: The aim of this study was to assess the possible extent of bias due to violation of a core assumption (event-dependent exposures) when using self-controlled designs to analyse the association between COVID-19 vaccines and myocarditis. METHODS: We used data from five European databases (Spain: BIFAP, FISABIO VID, and SIDIAP; Italy: ARS-Tuscany; England: CPRD Aurum) converted to the ConcePTION Common Data Model. Individuals who experienced both myocarditis and were vaccinated against COVID-19 between 1 September 2020 and the end of data availability in each country were included. We compared a self-controlled risk interval study (SCRI) using a pre-vaccination control window, an SCRI using a post-vaccination control window, a standard SCCS and an extension of the SCCS designed to handle violations of the assumption of event-dependent exposures. RESULTS: We included 1,757 cases of myocarditis. For analyses of the first dose of the Pfizer vaccine, to which all databases contributed information, we found results consistent with a null effect in both of the SCRI and extended SCCS, but some indication of a harmful effect in a standard SCCS. For the second dose, we found evidence of a harmful association for all study designs, with relatively similar effect sizes (SCRI pre = 1.99, 1.40 - 2.82; SCRI post 2.13, 95 %CI - 1.43, 3.18; standard SCCS 1.79, 95 %CI 1.31 - 2.44, extended SCCS 1.52, 95 %CI = 1.08 - 2.15). Adjustment for calendar time did not change these conclusions. Findings using all designs were also consistent with a harmful effect following a second dose of the Moderna vaccine. CONCLUSIONS: In the context of the known association between COVID-19 vaccines and myocarditis, we have demonstrated that two forms of SCRI and two forms of SCCS led to largely comparable results, possibly because of limited violation of the assumption of event-dependent exposures.