Statistical Analyses

Our analysis was conducted in several phases. First, we performed quality checks on the data extracted from DHIS2, aggregated the data on new users, and created summary statistics on new acceptors overall and bymethod. Next, we descriptively analyzed trends in the number of new acceptors, overall and by method, conducting chi‐square to test for significant differences in method mix. We then conducted interrupted time series (ITS) analysis to estimate both the immediate and longer‐term impact of COVID‐19 on the number of new acceptors, overall, and by method. ITS segmented ordinary least square regression models were estimated to assess the impact of COVID‐19 and related lockdowns on the number of new FP acceptors using the onset month of March 2020 as the “intervention”. The period of March 2019–February 2020 is the preintervention period as provides the “counterfactual” estimate. Due to model fit and evidence of autocorrelation (data not shown), we chose to use Prais–Winsten models to account for the autocorrelation observed in the data (Prais and Winsten 1954). To adjust for the potential effect of seasonality in contraceptive uptake, we used moving averages for smoothing. Postintervention linear trends were estimated for all ITS models. We also tested the robustness of our results by region (data not shown).

The ITS regression model (single group) takes the form:

where Y is the count of new users each month, (T). β0 represents the intercept or starting level of the outcome variable (estimated number of new FP acceptors in March 2019). β1 estimates the average monthly change in the number of new FP acceptors until the onset of the intervention or the counterfactual slope (March 2019 to February 2020). Tt is the time (in months) since the start of the study, β2 represents the change in the number of new FP users that occurred in March 2020 with the beginning of the COVID‐19 pandemic, and β3 represents the difference between the trend in new FP acceptors before the first announced case of COVID‐19 compared to the trend in the number of new FP acceptors following the initial case. Xt is a dummy (indicator) variable representing the event (pre‐event periods 0, otherwise 1), XtTt is an interaction term between the event and time, and εt the random error term. Data were unweighted and analyzed using Stata version 16 (StataCorp 2019).