Discussion
This meta-analysis brings together trials done in varied settings. The children in the various trials were of diverse ages and varying HIV status. Upon pooling the studies, INH was found to provide substantial protection against TB among children aged ≤15 years (RR = 0.65, 95% CI 0.47, 0.89 p = 0.004), however, there was marked heterogeneity in the efficacy of INH (Chi 18.56, df = 7, I = 62.3% p = 0.01). We conducted a sensitivity analysis to explore the observed heterogeneity. Most of the heterogeneity was noted to be due to the two studies by Madhi et al.(also referred to as the infant studies in this article), primarily due to two reasons. First, the results of the two studies by Madhi et al. were notably different from the results of the other studies, in that these two studies yielded point estimates much closer to the null compared to the results of the other studies, with confidence intervals that included the null, while the rest of the studies had point estimates and confidence intervals that showed that INH conferred a protective effect. Second, both the random effects and the fixed effects models gave these two studies more weight because of their relatively large size. One plausible explanation for the null results in the Madhi studies may be due to overdiagnosis of TB given that only few of the cases were confirmed microbiologically and many cases only met minimal clinical criteria. After excluding the two studies by Madhi et al., the heterogeneity disappeared (I = 0.0%, p = 0.799) and our analysis yielded an even greater effect estimate (RR = 0.41 (95% CI 0.31, 0.55) p <0.001).
One possible explanation for the discrepancy between findings from Madhi et al. and from the remaining studies has to do with the age of the participants at initiation of INH. Whereas subjects in the Madhi et al. trials were infants (median age 96 days, range 91–120 days) at the time of INH initiation, the remaining studies involved mostly older children, suggesting that age may modify the effect of INH prophylaxis on the development of TB. This hypothesis is supported by subgroup analyses based on five studies that investigated differential INH efficacy by age. INH was found to be more effective among children aged 5–15 years (RR = 0.53, 95% CI 0.30, 0.94 p = 0.014) than among children less than five years of age (RR = 0.85, 95% CI 0.54, 1.35 p = 0.25) ( Table 2 ). As mentioned previously, INH did not appear to confer any protective effect against TB in the infant studies (RR = 0.93, 95% CI 0.71, 1.21 p = 0.29).The lack of effect observed may be due to the relative immaturity of the innate immune system among younger as opposed to older children considering existing evidence suggesting that the dosage delivered in these studies was sufficient to achieved therapeutic levels. We also acknowledge that age might be a confounding factor for prior TB exposure, given that in the Madhi et al. study, INH was used for primary prophylaxis compared to the other studies in which it was used for either secondary prophylaxis or both primary and secondary prophylaxis. INH is known to be effective for the preventive treatment of latent TB which would be more prevalent among older children but it might not be as effective in preventing TB infection. We therefore cannot fully rule out the possibility that age may be a surrogate marker for TB infection thus explaining the higher efficacy seen in older children consistent with the results of the primary versus secondary prophylaxis subgroup analysis. While age may, in fact, be an effect modifier of the efficacy of INH, additional alternative explanations proposed by Madhi et al. for the null results in their studies include INH resistance, poor compliance, and the specificity of the study end points. Other differences in the studies by Madhi et al. compared to the rest were that the population had no known history of TB exposure and the participant ages were more uniform compared to the other studies.
Minimal differences in the efficacy of INH prophylaxis were seen in the subgroup analyses based on duration of administration of INH and study quality. Six trials administered INH for a duration of six months or longer and showed a 34% reduction in the risk of developing TB among those randomized to receive INH, RR = 0.66 (95% CI 0.45, 0.98) p = 0.02; the rest of the trials administered INH for less than six months and still demonstrated a protective effect against TB, RR = 0.53 (95% CI 0.30, 0.94) p = 0.014. These results suggest that administration of INH for less than six months may be as effective as longer courses of prophylaxis among non-HIV infected children. This finding, which is consistent with that of other studies demonstrating only a small advantage of a 12-month over a 6-month course of INH preventive therapy, may have important public health implications. This would be useful if applied in high TB endemic areas where the duration and intensity of exposure might be a critical factor affecting effectiveness of INH prophylaxis. While current guidelines for isoniazid preventive therapy recommend six months of prophylaxis, an equally effective but shorter course could result in improved adherence to therapy in children without HIV. While the studies in this meta-analysis that administered a shorter course of prophylaxis were of generally low quality ( Table 1 ) and should therefore be interpreted with caution, these findings justify further research into the optimal duration of INH preventive therapy in children, which remains uncertain.
Subgroup analyses based on HIV infection status found a strong protective effect of INH against TB among HIV-negative children (RR = 0.55, 95% CI 0.40, 0.75 p = 0.001), but no evidence of an effect among HIV-positive children (RR = 0.86, 95% CI 0.41, 1.81 p = 0.187). However, it is worth noting that the analysis of HIV-infected children was limited by the inclusion of only two studies; one in which INH was used for primary prophylaxis in HIV-exposed infants (Madhi et al.) and the other for secondary prophylaxis in older HIV-infected children (Zar et al.). We note, however that the Zar et al. study found INH to be highly effective and is further discussed below. As mentioned above, age may, itself, modify INH efficacy, complicating interpretation of these results. Therefore, we are unable to make a conclusive statement concerning INH efficacy among HIV-infected children and urge further research into this question. To our knowledge, the two studies we included in this subgroup analysis are currently the only studies on the efficacy of INH in prevention of TB that have been conducted in HIV-infected children.
The results also suggest that the efficacy of INH was greater in TB non-endemic regions (RR = 0.40, 95% CI 0.27, 0.57 p < 0.001) compared to TB endemic regions (RR = 0.78, 95% CI 0.55, 1.11 p = 0.08). One plausible explanation as to why isoniazid may have been less efficacious in endemic regions is reinfection of the children with a different strain of Mycobacterium tuberculosis. However, it is important to note that the results of this subgroup analysis may be misleading because the studies in TB endemic regions contained a large proportion of very young children compared to studies in the TB non-endemic regions, and the studies that contained this category of participants (Madhi et al.) yielded null results; this finding, combined with the fact that the studies by Madhi et al. received more weight in both the fixed effects and the random effects models, may explain the null result in this subgroup analysis. Upon excluding these two studies, we obtained a summary estimate of RR 0.44(0.27, 0.72) p < 0.001, suggesting that INH confers a protective effect in TB endemic areas as well.
Four of the eight studies included in this meta-analysis reported results on all-cause mortality. Combining the results of these studies (comprising 2,391 children) showed little evidence of an effect of INH on all-cause mortality RR = 0.86 (95% CI 0.63, 1.17) p = 0.168, consistent with the findings of studies conducted on adults. This estimate however may be misleading, because two of the four studies included were the infant studies described above, where no protective effect against TB or mortality benefit had been demonstrated; these two studies received 77.8% of the weight in this subgroup analysis. We thus excluded these two studies to get a more reliable estimate (RR = 0.58 (95% CI 0.31, 1.09) p = 0.092); this estimate, though imprecise, may suggest a mortality benefit of INH prophylaxis. We are limited in our ability to draw firm conclusions on mortality benefits of INH prophylaxis due to the limited data available, and thus recommend further studies.
A legitimate concern with expanded use of INH prophylaxis among children is the potential for serious adverse events resulting from therapy. This meta-analysis, however, suggests that INH is relatively safe in children. The most serious adverse events reported in the studies included were the deaths of two children following overdoses of INH in the study by Comstock et al.. In the same study, however, only 0.8% of the participants experienced adverse events that necessitated discontinuation of the medication. The adverse events reported in the other studies were generally mild and did not warrant discontinuation of INH. The study by Zar et al. reported that the safety and tolerability of INH prophylaxis in children were excellent, even among the HIV-infected subgroup receiving highly active antiretroviral therapy. Widespread use of INH over several years has shown that a dosage of 5 mg/kg of body weight is unlikely to cause serious toxic effects, a conclusion substantiated by this meta-analysis and the results of other INH prophylaxis trials. Another concern is INH resistance; previous reviews have not found a statistically significant elevated risk of isoniazid-resistant TB among individuals previously treated with isoniazid preventive therapy.
The study by Zar et al. demonstrated a marked protective effect in terms of prevention of TB (RR 0.28 (95% CI 0.1, 0.78) p = 0.005) and TB mortality (RR 0.46 (95% CI 0.22, 0.95) p = 0.015) among HIV-infected children and is therefore worth commenting on. The study by Zar et al. was a landmark study that was conducted on HIV- infected children in a TB endemic region and demonstrated marked benefits of INH in terms of reducing morbidity as well as mortality and was instrumental in informing policy on INH prophylaxis among HIV-infected children. We note that most of the differences in mortality and incidence of TB were observed within two months after randomization, presenting the possibility that some of the children selected to participate in the study already had subclinical TB and the INH was actually treating (as monotherapy) rather than preventing TB. This scenario, if true, may have led to an overestimation of the efficacy of INH as preventive therapy. However, it is worth noting that if the aforementioned concern is true, the findings by Zar et. al present valuable evidence demonstrating that INH is useful in reducing TB and mortality in HIV-infected children with advanced disease (WHO stage 3 or 4), regardless of whether a diagnosis of TB is made or not. Diagnosis of TB in this category of children presents a challenge, especially in resource limited settings.
The risk of bias in this meta-analysis arises mostly from inclusion of two studies that were of generally low quality. The study by Debre et al. was not conducted in a blinded fashion, while the study by Gupta et al. did not give details of how the randomized controlled trial was conducted; the authors acknowledged irregularities and defaults in drug intake in the latter study which may have influenced their results. Concerning all of the studies, generally, diagnosis of TB is very difficult in children, particularly in infants and this may have led to failure to make a diagnosis, misdiagnosis or over diagnosis of the outcome, leading to a biased estimate of the efficacy of INH preventive therapy. This scenario is highly likely due to the diverse diagnostic criteria used in the different studies which is a major weakness in this analysis. Misdiagnosis is equally likely in the treatment and control groups so the effect was likely to be non-differential and therefore bias estimates towards the null.
Publication bias has the potential to bias the results of a meta-analysis. We used a funnel plot to look for evidence of publication bias. While there was some suggestion of asymmetry in the funnel plot, we are not convinced that publication bias is a serious threat to our findings. However, it is important to note that other factors, apart from publication bias, can cause asymmetry in a funnel plot and affect the outcome of the Egger and Begg tests; in addition, the validity and interpretation of the Egger and Begg tests have been debated. It should also to be noted that the sensitivity of both the Egger and Begg tests is low when the analysis contains fewer than ten studies, as was the case with our meta-analysis. To establish whether publication bias was a concern, we estimated the pooled RR from the logarithm of the RR corresponding to the largest studies, which was approximately −0.4; we exponentiated this value, yielding an estimate of RR = 0.67, close to the estimate of our pooled summary measure (RR 0.65) showing that in this case, publication bias may not be influential in altering the overall summary estimate. In addition, because the studies by Madhi et al. were different from the other studies because of the ages of the participants at INH initiation and were responsible for the asymmetry on the funnel plot, we excluded them and re-did a funnel plot; after exclusion of these two studies there was no asymmetry (Egger test p = 0.598, Begg test p = 0.26), providing no evidence of publication bias.
Limitations associated with this meta-analysis include the fact that only eight studies were deemed suitable for inclusion, compromising the statistical power of the subgroup analyses. Diverse TB diagnostic criteria were used in the different studies which is a major weakness in this analysis due to lack of standardization across studies. Attempts have been made to standardize TB case definition in children for research purposes to prevent this scenario in the future. In addition, the studies were of a mixed nature with some evaluating efficacy of INH in primary TB prevention while others evaluating secondary TB prophylaxis yet others evaluating efficacy of preventing both primary and secondary TB, this limits comparability across the studies. Another limitation was that we restricted our search to English language articles and thus may have missed studies published in other languages. Furthermore, we included only RCTs in this meta-analysis, which poses the challenge of generalizability because trial participants are different from the general population, owing to the fact that trials often have strict eligibility criteria and those who agree to participate may be very different from the general population.