Multivariable Cox proportional hazards regression was used to assess the effect of baseline predictors on the survival of children on ART.
We also examined baseline predictors of the CBT outcomes.
Results of the analyses of baseline predictors on percent changes from baseline are displayed in Fig. 1 and supplementary Fig. S1 (categorical) and supplementary Table D (continuous).
For both progression and incidence, a multiple logistic regression model was used to determine baseline predictors of on-study retinopathy.
For both measures of renal disease progression, a multiple logistic regression model was used to determine baseline predictors of on-study renal disease progression.
Classes of individuals with different social recovery trajectories were compared on baseline predictor variables using one-way ANOVAs and chi-squared tests.
This means that the estimate of the regression of follow-up health on baseline predictor controlled for baseline health on average over the 500 samples was underestimated and that only 67% of the 500 samples had a 95% C.I. containing the true population value of bpred = .10.10
The effect of baseline predictors and the treatment on survival of the patients was assessed using a uni- and multivariate analysis.
The estimated bpred was then the same as when attrition rate was 50%, except for the second situation where liability of dropping out was dependent on baseline predictor and baseline health to the same degree (both bdrops = .30), but not on follow-up health.
Actually, when attrition was equally dependent on follow-up health, baseline predictor, and baseline health, the result again was unbiased.
