Censored mixed-effects models for irregularly observed repeated measures with applications to HIV viral loads