• Ambarish Chandra, Tracy Beth Høeg, Shamez Ladhani, Vinay Prasad, and Ram Duriseti.
• Department of Management, University of Toronto Scarborough, and Rotman School of Management, University of Toronto, Toronto, Ontario, Canada (A.C.).
• Ann. Intern. Med. 2024 Nov 1; 177 (11): 156615721566-1572.

BackgroundThere are considerable challenges when using difference-in-differences (DiD) analysis of ecological data to estimate the effectiveness of public health interventions in rapidly changing situations.ObjectiveTo discuss the shortcomings of DiD methodology for the estimation of the effects of public health interventions using ecological data.DesignAs an example, the authors consider an analysis that used DiD methodology and reported a causal reduction in COVID-19 cases due to the maintenance of school mask mandates. They did alternate analyses using various control groups to assess the robustness of the prior analysis.SettingSchool districts in the greater Boston area and Massachusetts during the 2021-to-2022 academic year.ParticipantsStudents and school staff.MeasurementsChanges in COVID-19 case rates in districts that did and did not lift mask mandates.ResultsImportant potential confounders rendered DiD methodology inappropriate for causal inference, including prior immunity, temporal variation in rates of infection, and changes in testing practices. The racial composition and income of intervention and control groups also differed substantially. Compared with maintaining the mask requirement, dropping the requirement was associated with anywhere from an increase of 5.64 cases (95% CI, 3.00 to 8.29 cases) per 1000 persons to a decrease of 2.74 cases (CI, 0.63 to 4.85 cases) per 1000 persons, depending on choice of control group and whether students or staff were examined.LimitationEcological data were used; detailed data on all potential confounders were unavailable.ConclusionAlternate analyses yielded estimates consistent with a wide range of both negative and positive associations in COVID-19 case rates after removal of mask mandates. The findings highlight the challenges of using DiD analysis of ecological data to estimate the effectiveness of interventions in divergent intervention and control groups during rapidly changing circumstances.Primary Funding SourceNone.

20 keywords...

hide…