In total, 3752 ART-naïve individuals started ART between 1 January 2004 and 31 December 2007. 213 individuals were excluded due to starting IPT later than 3 months after ART start, and 14 who started IPT greater than 12 months prior to ART. 255 individuals were on TB treatment at the start of ART and were excluded. Of the 3270 included individuals, 93% were male, mean age was 44.9 years, median baseline CD4 155 cells/μl (interquartile range [IQR] 87 – 221), and median viral load 48 817 copies/ml (IQR 17 860 – 128 629). Total follow-up time was 2910 person-years (339 (9.03%) censored before 12 months). WHO stage at ART initiation (available for 3 270 individuals), was WHO stage 2 (1803, 55.1%), stage 3 (1123, 34.3%) or stage 4 (344, 10.5%).
922 (28%) individuals started IPT either before or during the first three months of starting ART; the vast majority (n=882, 96%) within 0–7 days of starting ART. 234 individuals (7.2%) had previously been treated for tuberculosis prior to starting ART. 1338 (40.9%) individuals started cotrimoxazole either before starting ART (94% within a month of starting ART) or during the follow up time on ART. Baseline and 6 week viral load results were available for 1986 individuals (60.6%), of whom 1648 (83%) achieved a greater than 1 log10 reduction in HIV RNA by 6 weeks.
Baseline characteristics among those who started versus did not start IPT are compared in . Individuals who were started on IPT tended to have less advanced HIV disease as they had higher CD4 counts (CD4 <100cells/μl, 26.2% vs. 32.9%), lower viral load (viral load >100 000copies/ml, 24.3% vs. 33.1%), and higher haemoglobin levels (haemoglobin >10g/dl, 94.1% vs. 87%) at ART initiation and fewer were in WHO stage 3 or 4 (29.8% vs. 50.3%). The groups also differed with regards to whether they had had an episode of TB (8.9% in those never started on IPT and 2.5% in those who had started on IPT) and markedly by whether they were also on cotrimoxazole (84.8% in those who had started on IPT and 29.9% in those never started on IPT).
Baseline characteristics of individuals starting antiretroviral therapy, comparing those who did vs. did not receive isoniazid preventive therapy (n=3270).
In the first 12 months following initiation of ART, 259 (7.9%) individuals died giving an overall mortality rate of 8.90/100 person years (py) (95% confidence interval (CI) 7.88 – 10.05). The mortality rate of those never started on IPT was 11.08/100py (227/2047, 95% CI 9.74 – 12.63) and those who were started IPT was 3.71/100py (32/863, 95% CI 2.62 – 5.24). The unadjusted hazard ratio (HR) for mortality from the Cox model for those started on IPT compared to those not started on IPT was 0.34 (95% CI 0.24 – 0.49).
shows associations of various risk factors with mortality. The final adjusted model included age group, baseline CD4 group, WHO stage at ART initiation, year of starting ART, individual company and receipt of IPT ). In the adjusted model, IPT use remained strongly associated with decreased mortality (adjusted HR: 0.50, 95% CI 0.32 – 0.80). Adding cotrimoxazole use, baseline haemoglobin and baseline viral load to the model did not change the hazard ratio substantially. As an indicator of early response to ART, change in viral load at 6 weeks was included in a model restricted to individuals with available data, who therefore survived at least this long (); however there was no evidence that this caused confounding in the association between IPT and mortality. All factors met the proportional hazards assumption (P<0.01). Other risk factors for mortality that remained in the adjusted model were older age group (Ptrend<0.001), lower baseline CD4 count (Ptrend<0.001), earlier year of ART start (Ptrend=0.001) and company (P=0.30). We graphically illustrate the association between IPT and mortality using a Kaplan-Meier survival curve ().
Factors associated with mortality, unadjusted and adjusted analyses
Figure Kaplan Meier curve comparing survival among those who started or did not start isoniazid preventive therapy (IPT)
To assess whether the association between IPT and mortality differed by time period after starting ART we assessed for an interaction between IPT use and time period. The data were consistent with no interaction (unadjusted analysis p=0.89, adjusted analysis p=0.94). In the first three months following ART initiation, overall mortality rate was 16.28/100py, higher in those not on IPT (20.15/100py : 113/561py) compared to those started on IPT (6.65/100pyrs :15/226py), HR 0.33 (95% CI 0.19 – 0.57). In the period after three months on ART, overall mortality rate was 6.17/100py, higher in those not on IPT (7.67/100py : 114/1487py) compared to those started on IPT (2.67/100pyrs :17/638py), HR 0.35 (95% CI 0.21 – 0.58). In the multivariable model, the adjusted hazard ratio IPT in the time period less than 3 months was 0.50 (95% CI 0.27 – 0.93) and in the time period greater than 3 months was 0.51 (95% CI 0.29 – 0.91) (). Similarly we tested for effect modification with cotrimoxazole. When stratifying by whether patients were also on cotrimoxazole, those on IPT continued to have lower mortality rates than those not on IPT, both among individuals taking cotrimoxazole (HR 0.27 (0.11 – 0.56)) and not taking cotrimoxazole (HR 0.33 (0.21 – 0.51); P-value for interaction P=0.68)
Interaction between CD4 count and isoniazid preventive therapy use on mortality (n=2540)
shows the association between IPT and mortality in sensitivity analyses. When analysis was restricted to only those who had not had previous tuberculosis (n= 3036) the adjusted HR was 0.52 (95%CI 0.32 – 0.82). When restricted to those who had data available regarding symptoms at baseline, and had no symptoms suggesting tuberculosis (n=1923), the adjusted HR for the association of IPT and mortality was 0.48 (95%CI 0.24 – 0.96). Using a combined mortality end-point which classified individuals who left employment on medical grounds (n=99) as deaths, the adjusted HR for the association of IPT and mortality was 0.72 (95%CI 0.47 – 1.10). We also looked at the association of IPT with mortality, restricted to those who had no previous TB and also no symptoms suggesting tuberculosis with the period of ART initiation (N=1795); the adjusted HR for the association of IPT and mortality was 0.44 (95% CI 0.22 – 0.89).
Sensitivity analyses for associations between isoniazid preventive therapy and mortality using various scenarios
To assess whether the association between IPT and mortality differed by CD4 count strata we assessed for an interaction between IPT use and CD4 count. The data were consistent with no interaction (unadjusted analysis p=0.84, adjusted analysis p=0.81). The effect of IPT use on mortality, stratified by CD4 group at ART initiation was as follows: CD4 <50 cells/mm3, HR 0.80 (95% CI 0.36–1.78); CD4 50–100 cells/mm3, HR 0.48 (95% CI 0.19 – 1.19); CD4 101 – 200cells/mm3, HR 0.48, (95% CI: 0.22 – 1.07), and >200 cells/mm3, HR 0.45, (95% CI: 0.15–1.33) ().