Table 3

Sensitivity analyses, demonstrating results of the main regression model and alternate models

All visitsMean daily visits per clinic at time of lockdown periodStepwise change in clinic visits/day at start of level 5 lockdownP value
Primary model90.3 (67.1 to 113.5)−6.7 (−16.4 to 3.0)0.18
Poisson* mixed effects model92.0 (59.3 to 124.7)−6.9 (−11.0 to −2.8)0.001
Linear GEE (exchangeable correlation matrix)89.2 (67.0 to 111.4)−6.4 (−16.8 to 4.08)0.23
Poisson GEE* (exchangeable correlation matrix)89.2 (84.7 to 93.6)−6.6 (−8.7 to −4.5)<0.001
Linear GEE (autoregressive correlation matrix)90.2 (73.2 to 107.2)−5.4 (−27.4 to 16.6)0.63
Poisson GEE* (autoregressive correlation matrix)88.4 (85.0 to 91.9)−5.0 (−9.4 to −0.6)0.03
Difference-in-differences96.3 (63.6 to 129.0)3.4 (−5.5 to 12.4)0.45
Childcare visits
 Primary model11.9 (8.6 to 15.1)−7.1 (−8.9 to −5.3)<0.001
 Poisson* mixed effects model12.3 (7.1 to 17.5)−7.7 (−11.1 to −4.4)<0.001
 Linear GEE (exchangeable correlation matrix)11.8 (8.6 to 15.0)−7.1 (−9.0 to −5.2)<0.001
 Poisson GEE* (exchangeable correlation matrix)12.0 (10.5 to 13.5)−7.5 (−8.4 to −6.6)<0.001
 Linear GEE (autoregressive correlation matrix)11.9 (9.5 to 14.4)−6.4 (−9.9 to −2.9)<0.001
 Poisson GEE* (autoregressive correlation matrix)11.9 (10.7 to 13.1)−6.7 (−8.1 to −5.4)<0.001
 Difference-in-differences†11.8 (8.0 to 15.7)−4.0 (−5.5 to −2.5)<0.001
HIV visits
 Primary model37.5 (24.4 to 50.7)8.0 (2.3 to 13.7)0.01
 Poisson* mixed effects model39.2 (22.7 to 55.8)9.0 (4.5 to 13.5)<0.001
 Linear GEE (exchangeable correlation matrix)37.7 (25.3 to 50.1)8.1 (2.2 to 14.0)0.007
 Poisson GEE* (exchangeable correlation matrix)37.7 (34.8 to 40.6)8.7 (7.2 to 10.3)<0.001
 Linear GEE (autoregressive correlation matrix)38.9 (29.3 to 48.5)6.1 (−6.2 to 18.5)0.33
 Poisson GEE* (autoregressive correlation matrix)37.9 (35.7 to 40.1)5.7 (2.6 to 8.8)0.002
 Difference-in-differences†43.6 (25.1 to 62.1)4.8 (−0.5 to 10.1)0.08
  • *Poisson GEE results are presented as predictive margins and marginal effects so they represent changes on the same additive scale as the linear models.

  • †Difference-in-differences estimates are estimated as the mean of the level 5 lockdown period minus the mean of the prelockdown period, comparing 2020 with 2019. Estimates are based on a period-by-year interaction term fit via linear mixed models.

  • .GEE, generalised estimating equations.