Table 5.

Sensitivity analyses: LTFU in the second year

	Model D (original)		Model E (age and year of start as continuous)
Characteristic	HR (95% CI)	p	HR (95% CI)p	p
Age at 12 months
1 to <3 years	1
3 to <5 years	1.19 (0.74–1.91)	0.471
5 to 12 years	0.88 (0.58–1.35)	0.566
Age (as continuous variable)			0.94 (0.89–0.99)	0.047
Year of Starting ART
2004–2005	1
2006–2008	1.85 (1.12–3.07)	0.017
2009–2011	2.68 (1.58–4.55)	<0.001
Year of starting ART (as a continuous variable)			1.21 (1.10–1.34)	<0.001
Primary caregiver
Mother	1		1
Grandmother	0.30 (0.15–0.59)	0.001	0.32(0.16–0.64)	0.001
Other family	0.49 (0.30–0.82)	0.006	0.54(0.32–0.89)	0.017
Foster/Institution/Neighbour/Guardian	0.59 (0.28–1.21)	0.150	0.63(0.30–1.29)	0.206
Weight-for-age Z score
> −2 (Not underweight)	1		1
−2 to −3 (Moderately underweight)	1.97 (1.28–3.04)	0.004	1.97 (1.29–3.04)	0.002
< −3 (Severely underweight)	2.93 (1.91–4.47)	<0.001	2.65(1.74–4.04)	<0.001
CD4 cell percentage	0.97 (0.96–0.99)	0.004	0.97 (0.95–0.99)	0.001

Likelihood ratio test between Model D and Model E showed that there was no difference between the model (p=1.000). No imputation of missing CD4% was done for the LTFU in the second year model since only 4.1% had missing CD4% at 12 months.

Tests for linear trend for age group at 12 months were significant (p=0.003) and those for departure from linear trend were not significant (p=0.525), hence there was a linear relationship between age group at 12 months and LTFU in the second year.

Tests for linear trend for year of ART initiation were significant (p<0.001) and those for departure from linear trend were not significant (p=0.559), hence there was a linear relationship between year of ART start and LTFU in the second year.