Infection with M. tuberculosis is the most common bacterial infection in humans, and HIV infection is the strongest risk factor for tuberculosis (TB). According to World Health Organization estimates, over 4 million people, the majority of whom reside in Africa, are coinfected with both organisms . The risk of death in HIV-infected individuals with clinically symptomatic TB is about three-sevenfold higher than in individuals without HIV infection [2-4]. In AIDS patients previously free of TB, the clinical manifestation of TB increases overall mortality by one-third .
Randomized controlled trials have demonstrated that isoniazid (INH) prophylaxis may reduce the incidence of TB in HIV-negative populations at high risk of developing active disease . However, the altered immune state of HIV-infected patients, differences in the underlying risk for TB, and differences in susceptibility to drug toxicity make generalization to HIV populations questionable. Observational studies in HIV-positive injecting drug users (IDU) first suggested a potential benefit of INH prophylaxis [7-11]. More recently, investigators have published randomized controlled trials examining INH prophylaxis in HIV-infected persons [12-14].
We have conducted a meta-analysis of randomized controlled trials to assess the overall efficacy of INH prophylaxis in HIV-infected persons. We specifically investigated whether prophylaxis with INH is equally effective in tuberculin skin test (TST)-positive and negative persons and its effects on both TB and mortality, because individual trials may lack the power to address these issues.
We identified randomized controlled trials through a search of Medline, Embase, CAB Health, Biosis, Health Star, IDIS Drug File, DHSS-Data, Medical Toxicology and Health, Drug Information Full Text, AIDSLINE, AIDSTRIAL, AIDSDRUG, Cochrane database, by screening the proceedings of the International and European Conferences on AIDS, and by reviewing the references of identified trials. We used the following search terms: isoniazid, tuberculosis, human immunodeficiency syndrome, acquired immunodeficiency syndrome, HIV infection, and the textword randomized, as listed in titles and abstracts, and randomized controlled trial as a publication type.
Pairs of two reviewers (H.C.B., M.N. and H.C.B, P.S.) independently and blindly assessed studies for relevance, and for studies that proved eligible, for methodological quality. There was complete agreement between reviewers for relevance of eligible studies. Agreement for methodological quality of included studies was good (κ = 0.80). Reviewers resolved disagreement by consensus. We included studies that met the following criteria: randomized controlled trials comparing INH with placebo or no prophylaxis, intention-to-treat analysis, reporting of the number of patients who developed TB and the number of patients who died. Reviewers abstracted data from eligible studies in duplicate.
We rated the methodological quality of the included trials by a validated score that includes the following items : (i) randomization of participants, (ii) double-blind evaluation, (iii) full description of withdrawals and dropouts. The scoring gives one point to each item if present. If randomization is concealed, and if the method of double-blinding is appropriate (e.g., identical placebo, active placebo, dummy), the study receives one additional point, thus yielding a score with a range from 0 to 5 points.
Endpoints considered for this analysis were TB, death and drug-related adverse effects. One trial considered only culture-proven cases of TB . Three trials [13,14,16] used the following classification scheme: ‚Definite‚ TB was defined as a culture-confirmed disease. ‚Probable‚ TB was defined as a clinical illness with at least two of the following characteristics: chest radiography consistent with TB, acid-fast bacilli- positive smear of tissue or secretion, or response to anti-TB therapy. One trial  defined TB according to the American Thoracic Society standards, which require a clinical response to anti-TB therapy in addition to at least two out of three clinical, microbial or radiological criteria . Two trials did not provide information on the case definition of TB [19,20]. All trials where information was available defined a positive TST as an induration ≥5 mm.
Four trials provided information about drug adherence of study participants. The methods used were urinary testing for INH in a random sample of study participants [13,14], pill count [14,16], or interviews with patients [12,16]. According to each trial‚s definition of drug adherence, between 70 and 75% of the study participants were drug-adherent [13,14,16].
We tried to contact authors of trials with missing information. We were able to identify fax numbers or email addresses of all but one author. Three out of five authors complied with our request to provide missing information from their studies.
Risk ratios (RR) and computed summary estimates with 95% confidence intervals (CI) were calculated using a random effect method . All reported P values are two-tailed. We calculated summary incidence ratios for the progression to any endpoint using events by patient-years of follow-up . Because summary incidence ratios were very similar to RR, we do not report them.
Two subgroup analyses were conducted, a comparison of the efficacy of INH in people with or without a positive TST at study onset, and an average effect size was calculated for each subgroup. We tested the difference in combined estimates of TST-positive and negative patient groups and used the z-score for each subgroup by dividing the difference in the subgroup summary log-relative risk by the SE of the difference . In the second subgroup analysis, we excluded two unpublished trials.
We identified 10 randomized controlled trials. Two trials were excluded because they compared two different drug regimens [24,25], and one trial from Thailand is ongoing (J. Perriens, personal communication, 1998). Thus, seven randomized controlled trials are included in our analysis. Table 1 provides information about important baseline characteristics and event rates of the included trials. Investigators used double-blinding with placebos in four trials [12-14,16], whereas three were unblinded [17,19,20]. Two trials received the lowest possible quality score of 1, four trials received intermediate scores of 3 and 4, and one trial received the top score of 5. The two studies with the quality score of 1 were never fully published. Only one study used a 12-month regimen of INH ; the remaining trials used 6-month regimens. Africa provided the site for four studies [13,14,16,20], and Haiti , Mexico , and the United States  the locale for the other three studies. The US study restricted enrolment to TST-negative persons; the study from Mexico  was restricted to TST-positive persons, all other trials included TST-positive and TST-negative persons (information was unavailable in one trial ). Only the US investigators administered antiretroviral therapy and measured CD4 cell count in all participants at baseline.
This analysis includes 2367 persons treated with INH and 2162 persons in control and placebo groups. The proportion of those who developed TB in whom the diagnosis was confirmed by culture varied between 51.1 and 100%. The risk of TB in control group participants varied between 3.4 and 10.0 per 100 patient-years. Follow-up varied from 0.4 to 3.2 years.
INH prophylaxis and incidence of TB
The RR of INH versus placebo for TB (seven trials) was 0.58 (95% CI, 0.43-0.80; Fig. 1). In trials with separate information on TST-positive participants (five trials), the RR for TB was 0.40 (95% CI, 0.24-0.65), and in TST-negative participants (five trials), 0.84 (95% CI, 0.54-1.30). For both summary estimates, the test of heterogeneity was not statistically significant (P > 0.20, for each estimate). We found a statistically significant difference for the effectiveness of INH versus placebo prevention between groups of TST-positive and negative individuals (P = 0.03, for the difference of summary estimates).
INH prophylaxis and mortality
The RR of INH versus placebo for death (seven trials) was 0.94 (95% CI, 0.83-1.07; Fig. 2). In TST-positive participants (five trials) the RR of death was 0.79 (95% CI, 0.37-1.70), and in TST-negative participants (five trials), 1.02 (95% CI, 0.90-1.17). For both summary estimates, the test of heterogeneity was not statistically significant (P > 0.10, for each estimate). There was no statistically significant difference between the summary estimates in TST-positives and negatives (P = 0.52, for the difference of summary estimates). When we excluded the unpublished studies with the quality score of 1, the results were virtually identical to the overall analyses, and we therefore do not present the details here.
Side-effects and drug-limiting toxicity of INH prophylaxis
Four trials provided more detailed data concerning side-effects and drug toxicity. The summary estimate pooling all side-effects indicated a trend towards an increased risk in INH compared with placebo (RR, 1.36; 95% CI, 1.00-1.86; P = 0.27, test of heterogeneity; Fig. 3). The RR for drug-limiting toxicity of INH compared with placebo was 1.66 (95% CI, 0.83-3.32; P = 0.18, test of heterogeneity). The RR for hepatotoxicity (defined as at least a twofold increase of the serum aspartate aminotransferase level) of INH compared with placebo was 1.80 (95% CI, 1.05-3.10; P = 0.11, test of heterogeneity). Only one trial reported on drug-related mortality and the study physicians attributed three and five deaths in the INH and placebo groups, respectively, to the study medication .
This meta-analysis shows that preventive therapy with INH for 6 months effectively reduces the incidence of TB in HIV-infected people. The effect is statistically significant, and the magnitude is clinically relevant. In a subgroup analysis, we further showed that the risk reduction is larger in TST-positive than in TST-negative persons. Indeed, the data suggest that INH prophylaxis in HIV infection may be limited to individuals with positive TST. The results of the overall and subgroup summary estimates were similar in the trials, and tests for heterogeneity thus yielded high P values.
A recently published meta-analysis including only four trials . Our meta-analysis, which includes an additional large trial , confirms a positive effect of INH prophylaxis in TST-positive individuals, provides a more precise estimate of the magnitude of the effect and shows that the effect differs significantly in TST-positive and negative individuals with HIV infection.
The previous meta-analysis reported a reduction in mortality in TST-positive individuals. The present systematic review, however, showed no impact of INH prophylaxis on mortality. We found a trend suggesting that patients receiving active drug experienced more side-effects, particularly related to hepatotoxicity. It is possible that any mortality reduction due to TB prevention was counterbalanced by mortality associated with drug toxicity.
Strengths of our study included explicit inclusion and exclusion criteria, limiting eligible studies to randomized trials, a comprehensive literature search, and contacting researchers for unpublished trials and study results. Our success in obtaining data from unpublished studies reduced the likelihood of publication bias.
The subgroup analysis demonstrating differences between TST-positive and negative people gains considerable strength from the within-trial comparisons and the statistical significance of the differences in effect in the two subgroups of patients . Concern about this subgroup analysis comes from the observation that HIV-infected individuals with negative TST represent a heterogenous group. Some may not have been exposed to TB, whereas others may have been TST-negative because they were anergic. Since authors do not characterize response according to demonstrated anergy, or a surrogate for anergy such as CD4 cell counts, we could not explore this issue directly. Thus, the decrease in response to INH may represent lack of efficacy in the unexposed, the anergic, or both. From a mechanistic view, this is very unsatisfying. From a clinical point of view, however, the implications are clear. Whatever the mechanism, the evidence does not support a benefit of INH in TST-negative HIV-infected patients.
There are several limitations of the results and their interpretation. The quality of the trials varied considerably, which could lead to additional bias . Two trials that had never been fully published had very low quality scores. However, when we excluded these two studies and restricted our analysis to trials with higher quality and specific case definitions for TB, summary estimates for TB in the overall study population and in the subgroups of TST-positive individuals remained virtually unchanged and statistically significant.
A specific aspect of study quality relates to case definitions of TB and blinding of those assessing outcome. The studies used different case definitions of TB and the proportion of cases with culture verification differed considerably. In unblinded trials, biased ascertainment of outcome, particularly in patients classified as having TB without culture positivity, could have inflated the treatment effect. However, the similar estimates of treatment effect in blinded and unblinded studies make this less likely. The presentation of the data did not allow us to conduct an analysis restricted to culture-proven TB.
Trials also varied in the duration of follow-up. One trial with long follow-up indicated a decreased effectiveness of INH prophylaxis over extended observation periods , raising questions about long-term effectiveness. Loss of follow-up was considerable and where reported varied from 6.6 to 30.7%. This high rate of loss to follow-up may introduce further bias.
What are the practical implications from this systematic review? The implications depend on one‚s interpretation of the strength of inferences that the data warrants, the setting, and the target population for preventive therapy. In general, the case for prophylaxis is stronger when the incidence of TB is high than when it is low. In placebo groups of TST-positive persons from trials included in this analysis, the incidence of TB was in the range of 3.4- 10.0 per 100 person-years. The number of patients we must treat to prevent one patient from developing TB (the number needed to treat; NNT) provides a useful index of the impact of therapy . Assuming that the pooled estimate of relative risk reduction is unbiased, in the lowest risk group (3.4% incidence) we would need to treat 70 persons to prevent one case of TB. In the highest risk group (10.0% incidence) the NNT would be 24.
Several studies of the incidence of TB amongst HIV-infected individuals were performed in IDU in New York, Madrid and Barcelona, and found incidence rates that varied from 9.7 to 16.2 cases per 100 person-years [9,10,30]. However, multicentre studies from the United States and France that included a lower proportion of IDU found much lower incidence rates of TB in the range of 0.45-2.6 cases per 100 person-years [31,32]. Other groups at higher risk include HIV-infected persons who have had contact with people with known TB and patients with TST conversion [9,33].
Cohort studies in more-industrialized countries have shown that preventive therapy with INH is feasible and effective in HIV-infected individuals who are living in areas with a relatively high prevalence of TB [11,30]. In less-industrialized countries, the data from feasibility studies on preventive therapy against TB in HIV-infected populations are limited. In a study from Uganda, 27% of approximately 5600 individuals testing positive at an HIV counselling and testing centre in Kampala were referred to a TB prevention programme and 1344 (88%) had a TST . Of these, 19% did not return for test reading and 520 patients with positive TST received INH prophylaxis. Of these, 62% collected at least 80% of their drugs. Investigators studying 412 HIV-infected persons recruited from a voluntary HIV counselling and testing centre in Thailand found a similar adherence rate (69.4%) .
In conclusion, this meta-analysis suggests that preventive therapy with INH may be effective in TST- positive individuals with HIV infection, but it does not reduce mortality from HIV infection. The case for preventive therapy will be strongest in TST-positive patients at high risk of TB. Preventive therapy programmes are likely to do most good when resources for the exclusion of active TB and proper monitoring of patients with dual infection are available.
The authors would like to thank Dr A. Mwinga and Dr J.D.P. Porter for their generosity in providing us with the full detail of their study prior to publication. The authors thank Dr M.P. Hawken and Dr C. Whalen for additional data from their trials, and Dr P. Godfrey-Faussett, World Health Organization, for his help in identifying relevant unpublished clinical trials. The authors are indebted to Dr P. Wolf for the systematic literature search.
1. Narain JP, Raviglione MC, Kochi A: HIV-associated tuberculosis in developing countries: epidemiology and strategies for prevention.Tuber Lung Dis
2. Chaisson RE, Schecter GF, Theuer CP, Rutherford GW, Echenberg DF, Hopewell PC: Tuberculosis in patients with the acquired immunodeficiency syndrome.Am Rev Respir Dis
3. Perriens JH, Colebunders RL, Karahunga C, et al.
: Increased mortality and tuberculosis treatment failure rate among human immunodeficiency virus (HIV) seropositive compared with HIV seronegative patients with pulmonary tuberculosis treated with ‚standard‚ chemotherapy in Kinshasa, Zaire.Am Rev Respir Dis
4. Nunn P, Brindle R, Carpentier L, et al.
: Cohort study of human immunodeficiency virus infection in patients with tuberculosis in Nairobi, Kenya.Am Rev Respir Dis
5. Perneger TV, Sudre P, Lundgren JD, Hirschel B: Does the onset of tuberculosis in AIDS predict shorter survival? Results of a cohort study in 17 European countries over 13 years. BMJ
6. Ferebee SH: Controlled chemoprophylaxis trials in tuberculosis: a general review.Adv Tuberc Res
7. Selwyn PA, Hartel D, Lewis VA: A prospective study of the risk of tuberculosis among intravenous drug users with human immunodeficiency virus infection.N Engl J Med
8. Gulear A, Gatell JM, Verdejo J: A prospective study of the risk of tuberculosis among HIV-infected patients.AIDS
9. Guelar A, Gatell JM, Verdejo J, et al.
: A prospective study of the risk of tuberculosis among HIV-infected patients.AIDS
10. Moreno S, Baraia-Etxaburu J, Bouza E, et al.
: Risk for developing tuberculosis among anergic patients infected with HIV.Ann Intern Med
11. Graham NMH, Galai N, Nelson KE, et al.
: Effect of isoniazid chemoprophylaxis on HIV-related mycobacterial disease. Arch Intern Med
12. Gordin FM, Matts JP, Miller C, et al.
: A controlled trial of isoniazid in persons with anergy and human immunodeficiency virus infection who are at high risk for tuberculosis.N Engl J Med
13. Whalen CC, Johnson JL, Okwera A, et al.
: A trial of three regimens to prevent tuberculosis in Ugandan adults infected with the human immunodeficiency virus. N Engl J Med
14. Hawken MP, Meme HK, Elliott LC, et al.
: Isoniazid preventive therapy for tuberculosis in HIV-1-infected adults: results of a randomized controlled trial. AIDS
15. Jadad AR, Moore RA, Jenkinson C, Reynolds DJ, Gavagahn DJ: Assessing the quality of reports of randomized clinical trials: is blinding necessary? Control Clin Trials
16. Mwinga A, Godfrey-Faussett P, Quigley M, et al.
: Twice weekly tuberculosis preventive therapy in HIV infection in Zambia.AIDS
1998 (in press).
17. Pape JW, Jean SS, Ho JL, Hafner A, Johnson WD Jr: Effect of isoniazid prophylaxis on incidence of active tuberculosis and progression of HIV infection. Lancet
18. American Thoracic Society/Medical Section of the American Lung Association: Diagnostic standards and classification of tuberculosis. Am Rev Respir Dis
19. Valdespino JL, Garcia ML, Daniels E, et al.
: Outcomes of the pilot study of TB chemoprophylaxis trials.IX International Conference on AIDS
. Berlin, June 1993 [abstract PO-BO7-114].
20. Wadhawan D, Hira S, Mwansa N, Sunkutu R, Adera T, Perine P: Preventive tuberculosis chemotherapy with isoniazid among patients infected with HIV-1. XI International Conference on AIDS.
Berlin, June 1993 [abstract PO-BO7-1133].
21. Der Simonian R, Laird N: Meta-analysis in clinical trials.Contr Clin Trials
22. Rothman KJ: Modern Epidemiology.
Boston: Little, Brown and Company; 1986.
23. Greenland S: Quantitative methods in the review of epidemiologic literature. Epidemiol Rev
24. Halsey NA, Coberly JS, Desormeaux J, et al.: Randomised trial of isoniazid versus rifampicin and pyrazinamide for prevention of tuberculosis in HIV-1 infection. Lancet
25. Matteelli A, Carosi G, Signorini L, Tebaldi A, Caligaris S, Gulletta M: Multinational study for the preventive therapy of latent tuberculosis in HIV positive individuals: isoniazid versus rifabutin. 12th World AIDS Conference
. Geneva, June-July 1998 [abstract 60858].
26. Wilkinson D, Squire SB, Garner P: Effect of tuberculosis preventive therapy in adults infected with HIV: systematic review of randomised placebo controlled trials.BMJ
27. Oxman AD, Guyatt GH: A consumer‚s guide to subgroup analyses.Ann Intern Med
28. Schulz KF, Chalmers I, Grimes DA, Altman DG: Assessing the quality of randomization from reports of controlled trials published in obstetrics and gynecology. JAMA
29. Jaeschke R, Guyatt GH, Shannon H, Walter SD, Cook DJ, Heddle N: Basic statistics for clinicians. III. Assessing the effects of treatment: measures of association. Can Med Assoc J
30. Selwyn PA, Sckell BM, Alcabes P, Friedland GH, Klein RS, Schoenbaum EE: High risk of active tuberculosis in HIV-infected drug users with cutaneous anergy. JAMA
31. Melnick SL, Sherer R, Louis TA, et al.
: Survival and disease progression according to gender of patients with HIV infection.JAMA
32. Dupon M, Texier-Maugein J, Leroy V, et al.
: Tuberculosis and HIV infection: a cohort study of incidence and susceptibility to antituberculous drugs, Bordeaux, 1985-1993.AIDS
33. Markowitz N, Hansen NI, Hopewell PC, et al.
: Incidence of tuberculosis in the United States among HIV-infected persons. Ann Intern Med
34. Aisu T, Raviglione MC, van Praag E, et al.
: Preventive chemotherapy for HIV-associated tuberculosis in Uganda: an operational assessment at a voluntary counselling and testing center.AIDS
35. Ngamvithayapong J, Uthaivoravit W, Yanai H, Akarasewi P, Sawanpanyalert P: Adherence to tuberculosis preventive therapy among HIV-infected persons in Chiang Rai, Thailand.AIDS