Table 3. Cox regression modelling of baseline characteristics and attrition comparing baseline CD4 categories > 500 cells/μL and 0–500 cells/μL.

		Crude hazard ratio (95% CI)	P	Adjusted hazard ratio (95% CI)	P
Baseline CD4	> 500	1,24(1,03–1,48)	0,022	1,26(1,05–1,52)	0,014
(cells/μL)	0–500	1		1
gender	Male	1,06(0,9–1,25)	0,483	1,23(1,03–1,47)	0,025
	Femaile	1		1
Agecategory	18–25	1,29(1,07–1,56)	<0.001	1,3(1,07–1,58)	0,002
	26–35	1		1
	36–45	0,91(0,73–1,12)		0,9(0,72–1,11)
	46–55	0,7(0,5–0,97)		0,69(0,49–0,96)
	>55	0,82(0,46–1,46)		0,8(0,45–1,44)
Pregnant at baseline	Yes	1,36(1,01–1,82)	0,045	1,27(0,93–1,72)	0,134
clinic	Metro 1	1	0,279	1	0,137
	Metro 2	1,05(0,88–1,26)		1,16(0,95–1,43)
	Rural 1	0,89(0,71–1,11)		0,96(0,75–1,23)
Baseline TB	Yes	0,91(0,51–1,61)	0,744	0,94(0,53–1,68)	0,841
Previous ART of > 3 mths	Yes	0,88(0,5–1,56)	0,667	0,91(0,51–1,61)	0,739
Year ART start	2014	0,86(0,72–1,02)	0,088	0,81(0,66–1)	0,051
	2015	1		1

Proportional hazards assumptions were checked with scaled Schoenfeld residuals. Likelihood ratios were used to estimate P values in regression models where categorical variables had more than two strata. Model fits were assessed as good based on the likelihood ratio test statistic. Selection of baseline variable category for comparison (HR = 1) was based on sample size and clinical significance.