The role of BCG in prevention of leprosy: a meta-analysis

Transcription

1 The role of BCG in prevention of leprosy: a meta-analysis Maninder Singh Setia, Craig Steinmaus, Christine S Ho, George W Rutherford Lancet Infect Dis 2006; 6: MSS and CSH are at the Division of Epidemiology, and CS is at the Division of Occupational and Environmental Medicine, School of Public Health, University of California, Berkeley, CA, USA; GWR is Salvatore Pablo Lucia Professor and Director of the Institute for Global Health, Department of Epidemiology and Biostatistics, University of California, San Francisco, San Francisco, CA, USA. Correspondence to: Dr George W Rutherford, Institute for Global Health, Department of Epidemiology and Biostatistics, University of California, San Francisco, 50 Beale Street, Suite 1200, San Francisco, CA 94105, USA. Tel ; fax ; GRutherford@psg.ucsf.edu The present meta-analysis investigates the role of BCG a widely used yet controversial vaccine in the prevention of leprosy. The electronic databases Medline, Embase, the Cochrane Library, and LILACS were searched to identify studies assessing the protective effect of BCG against leprosy. We included seven experimental studies and 19 observational studies. The experimental studies demonstrated an overall protective effect of 26% (95% CI 14 37%). At 61% (95% CI 51 70%), the observational studies overestimated the protective effect. The age at vaccination did not predict the protective effect of BCG. An additional dose of BCG was more protective in the prevention of leprosy compared with a single dose. An additional dose of BCG may be warranted for contacts of leprosy patients in areas where leprosy continues to be a public-health problem. Introduction Although eliminated from most of the countries where it was considered a public-health problem, leprosy a chronic infection caused by Mycobacterium leprae is still present in countries in Africa, Asia, and Latin America. With the introduction of multidrug therapy in 1982 as a response to resistance to dapsone, 1,2 the duration of treatment for leprosy reduced, and lower relapse rates were observed. 3 5 However, the WHO estimated that there were registered cases worldwide at the beginning of 2005 and that about 52% of these cases were from India. 6 Efforts to reduce the prevalence of leprosy below the elimination target of less than one case per need to be stepped up in target countries like Brazil (figure 1) and India. 7,8 The primary public-health strategy for the control of leprosy has been the diagnosis and treatment of symptomatic patients to make them less infectious to susceptible individuals. The different methods considered for primary prevention of leprosy are chemoprophylaxis and immunoprophylaxis. 9 Apart from dapsone and multidrug therapy as methods of treatment and secondary prevention, vaccination of susceptible individuals has often been an area of discussion for primary prevention efforts. Some of the potential vaccines for prevention of leprosy are BCG alone, BCG and killed Mycobacterium leprae, Indian Cancer Research Centre (ICRC) bacillus, Mycobacterium w, and Mycobacterium vaccae Among these potential vaccines, studies with BCG are most common. BCG is recommended by WHO s Expanded Programme on Immunization for protection against tuberculosis caused by a mycobacterium closely related to M leprae. Thus it is routinely used in the vaccination of children in countries where leprosy still continues to be a public-health problem and is currently also recommended for the household contacts of leprosy patients in Brazil. 14 In 1939, Fernandez 15 was the first to demonstrate the induction of a positive Mitsuda reaction a marker for improved cell-mediated immunity against leprosy following vaccination with BCG, a finding later discussed by others. 16,17 Subsequently studies were done to assess the protective effect of BCG in leprosy and controlled clinical trials to examine the role of BCG in leprosy were initiated in the 1960s. The studies done in endemic areas in different countries have demonstrated varying protective effects against leprosy. BCG is the world s most widely used vaccine, yet its role in the prevention of tuberculosis and leprosy is controversial. Given that BCG is easily available and widely used compared with the other vaccines useful in leprosy, we did a meta-analysis of existing published work to determine the protective effect of BCG in leprosy and the factors that impact its efficacy by doing a subgroup analysis. Pietro Cenini/Panos Pictures Rights were not granted to include this image in electronic media. Please refer to the printed journal Figure 1: Leprosy care in Brazil Methods Search strategy and identification of studies We assessed the role of BCG in protection of clinical leprosy, our study hypothesis being that rates of leprosy cases were different in BCG vaccinated and unvaccinated populations. We did a search of Medline, limited to human studies, using the index terms BCG and leprosy, leprosy vaccines, BCG and Hansen s disease, and BCG and hanseniasis. All papers until February 2005 were included in the meta-analysis. Other sources searched for relevant articles using the same index terms included Embase, the Cochrane Library, and LILACS. Review articles on the use of BCG in leprosy and leprosy vaccines were identified and searched for any additional Vol 6 March 2006

2 Reference Place of study Follow-up Cases of leprosy Point estimate of effect of % weight in fixed (years) in BCG group BCG on leprosy (95% CI) effects model Experimental studies Stanley et al 25 Uganda 8 41/8085* 0 20 ( ) 2 Lwin et al 26 Burma / ( ) 16 Bagshawe et al 27 Papua New Guinea / ( ) 3 Fine et al 28 Malawi /23 456* 0 51 ( ) 0 Gupte et al 29 India / ( ) 76 Gupte et al 30 India 5 5../ ( ) 2 Truoc et al 31 Vietnam 8 8/100* 0 42 ( ) 0 Observational studies Cohort studies Convit 32 Venezuela 5 3/584* 0 11 ( ) 1 Chatterjee et al 33 India 5 9/678* 0 04 ( ) 2 Ponnighaus et al 34 Malawi 4 80/ ( ) 10 Duppre 35 Brazil 5 39/468* 0 38 ( ) 7 Cunha et al 36 Brazil 4 31/ ( ) 4 Case-control studies Abel et al 37 Vietnam ( ) 2 Muliyil et al 38 India ( ) 12 Rodrigues et al 39 Brazil ( ) 3 Alvim 40 Brazil ( ) 4 Convit et al 41 Venezuela ( ) 5 Baker et al 42 Malawi ( ) 6 Orege et al 43 Kenya ( ) 3 Thuc et al 44 Vietnam ( ) 6 Boelens et al 45 Indonesia ( ) 1 Lombardi et al 46 Brazil ( ) 2 Bertolli et al 47 Burma ( ) 5 Zodpey et al 48 India ( ) 11 Zodpey at al 49 India ( ) 4 Zodpey et al 50 India ( ) 14..=not reported. *Cases of leprosy/patient population. Cases of leprosy/person-years. Table 1: Experimental and observational studies used for meta-analysis along with point estimates and 95% CIs of effect of BCG on leprosy studies not identified by the electronic search. The search was not limited to English language articles, and we translated studies in other languages if required. A list of all the articles reviewed was maintained. We scanned conference abstracts to identify any studies that were not published. Inclusion and exclusion criteria Inclusion criteria were established before identifying the articles to avoid any selection bias. All the studies (experimental and observational) that described the efficacy of BCG with prevention of clinical leprosy (defined according to clinical and/or microscopic criteria) were eligible for inclusion for meta-analysis. We excluded studies that assessed only the change in immunological status of individuals after BCG vaccination without assessing clinical infection. We excluded studies that measured the protective effect of BCG combined with other therapies (eg, chemotherapy, killed M leprae) and restricted the meta-analysis to studies that measured the protective effect of BCG alone. We also excluded review articles and case reports from the meta-analysis, although these papers were identified and searched for additional references. If the trials had multiple published reports, the latest reports with longest duration of follow-up were included for meta-analysis. If a study reported the effects of different doses of BCG in an adult population, we included the effect due to 0 1 mg (the recommended adult dose for BCG). We excluded case-control studies that did not describe selection criteria for cases and controls or assessment of the exposure among cases and controls (BCG vaccination). A list of all the included and excluded studies was maintained. Data abstraction For each clinical trial we collected information about the year of publication, the year when the trial began, the place of trial, the design of the trial (ie, randomisation, use of placebo, and blinding procedures), the number of study participants, the duration of follow-up, withdrawal of study participants, type of BCG vaccine used in the trial, the dose of BCG vaccine used in the trial, and the protective effect of BCG for the longest reported period. For the cohort studies identified we collected information on the year of publication of the study, the place of study, the number of study participants in the vaccinated and unvaccinated groups, the duration of follow-up, and the protective effect of the vaccine. For case-control studies we identified the year of publication, the place of study, number of cases and controls, the assessment of vaccination status, and the protective effect of BCG. Vol 6 March

3 Reference Age (years) Sex Type Type of contact Male Female Pauci Multi Indet Household Other Stanley et al 25 ( ) ( ) ( ) ( ) Lwin et al ( ) Bagshawe et al 27 ( ) ( ) ( ) ( ) Fine et al ( ) ( ) Gupte et al 30 ( ) ( ) Chatterjee et al 33 ( ) ( ) Ponnighaus et al 34 ( ) ( ) Cunha et al ( ) ( ) ( ) Muliyil et al ( ) ( ) ( ) Convit et al ( )* ( )* ( )* ( )* ( )* ( )* Orege et al ( ) ( ) ( ) ( ) Thuc et al ( ) Boelens et al 45 ( ) ( ) Lombardi et al ( ) ( ) ( ) ( ) ( ) ( ) Bertolli et al 47 ( ) ( ) Zodpey et al 48 ( ) ( ) ( ) ( ) ( ) ( ) ( ) Zodpey et al 49 ( ) ( ) ( ) ( ) ( ) ( ) ( ) Zodpey et al 50 ( ) ( ) ( ) ( ) ( )..=unreported; pauci=paucibacillary; multi=multibacillary; indet=indeterminate. *Unadjusted confidence intervals calculated. Table 2: Point estimates and 95% CIs of effect of BCG on leprosy for age, sex, type of leprosy, and type of contact subgroup analyses The estimates and the CIs were extracted by two reviewers (MSS and CSH). Data analysis and statistical methods We extracted the effect measure and its 95% CI for each study. The general principles for fixed effects models as described by Greenland 18 were used to calculate the pooled summary estimate. The standard error (SE) for each study was calculated by the dividing the difference in natural logarithms of the upper and lower CI by 3 92 (ie, SE=[ln CI upper ln CI lower ]/3 92). The weight (W i ) for each relative risk was calculated based on the inverse square of its standard error (ie, 1/SE 2 ). Missing CIs were calculated using unadjusted data for case-control studies, cohort studies, and trials. The summary measure (W i b i ) was calculated by multiplying the weight by the natural logarithm of the risk ratio (b i ). The pooled summary (b) was calculated by dividing the sum of summary measures by the sum of weights (ie, b= W i b i / W i ). The summary relative risk was calculated by exponentiating the pooled summary (ie, RR sum =exp[b]). The studies were also assessed for heterogeneity by the 2 test statistic ( 2 = W i [b b i ] 2 ) with a degree of freedom (df) equal to one less than the number of studies that were combined in the meta-analysis. If the studies were found to be heterogeneous (ie, 2 df), the 95% CI was recalculated using the method described by Shore and colleagues. 19 The adjusted variance was calculated by multiplying the ratio of the 2 statistic to its degrees of freedom by the variance of the logarithm of the pooled relative risk (ie, Var adj =[ 2 /df] Var). The 95% CIs were recalculated using this adjusted variance. The relative risks obtained in subgroup analysis were compared using the method described by Altman. 20 The natural logarithms for the two relative risks (E1 and E2) and their 95% CIs were calculated. The standard error for each log relative risk was calculated by dividing the width of the CI by We calculated the difference in log relative risks (d) and its standard error (SE d ). The test of interaction was calculated using the formula d/se d. The vaccine efficacy was calculated using the formula 100(1 RR sum ) for the trials and 100(1 OR sum ) for the case-control studies Vol 6 March 2006

4 We calculated separate summary estimates for experimental and observational studies. We did a quality scoring of the trials according to the scale described by Jadad and colleagues. 22 The trials were scored on a scale of 0 5 based on their reporting of randomisation, blinding procedures, and withdrawals in the study. We did subgroup analysis to assess the factors that might impact the efficacy of BCG vaccination. We calculated summary relative risks for men and women and for different age groups. We calculated the summary relative risks for effect with a single dose of BCG and more than one dose of BCG in individuals. We also calculated the summary relative risks for different types of leprosy reported by the various studies (ie, paucibacillary, multibacillary, or indeterminate). Some authors reported relative risks for paucibacillary and multibacillary types of leprosy, whereas others reported them for lepromatous and tuberculoid forms of leprosy. For calculating the pooled estimate we used the relative risks for paucibacillary and multibacillary cases (if specifically mentioned by the authors). However, we also calculated the summary estimates by including the lepromatous cases in the multibacillary group and tuberculoid cases in the paucibacillary group. We also calculated the summary relative risks among different types of contacts of leprosy patients that were included in the studies (ie, household contacts or other contacts). Forest plots and publication bias The point estimates and CIs were entered in Intercooled Stata Version 8 2 (College Station, TX, USA) and forest plots were obtained for experimental and observational studies. Publication bias was assessed by a funnel plot of the log of relative risk for the individual studies by its standard error. It was also assessed by the Begg s 23 and Egger s tests. 24 Results Search results We scanned 986 articles obtained from all the electronic searches and reviewed 115 complete articles (including studies, reviews, editorials, case reports, and WHO reports). Of the articles reviewed in detail, we found 26 studies eligible for inclusion in our meta-analysis, which included seven experimental studies (clinical trials) and 19 observational studies (cohort and case-control studies). Table 1 describes all the studies that were included in the meta-analysis, along with the relative risks, odds ratio, and 95% CIs for each study and the weight of the studies in the fixed effects model. Table 2 describes the point estimates and the 95% CIs for age, sex, type of leprosy, and type of contacts from the included studies. We excluded earlier reports of the studies in Uganda, Papua New Guinea, 55 Burma, Malawi, India, 64 and Venezuela, 65 and included the latest reports for each study. A B Stanley et al 25 Lwin et al 26 Bagshawe et al 27 Fine et al 28 Gupte et al 29 Gupte et al 30 Truoc et al 31 Combined Convit 32 Chatterjee et al 33 Ponnighaus et al 34 Duppre 35 Cunha et al 36 Abel et al 37 Muliyil et al 38 Rodrigues et al 39 Alvim 40 Convit et al 41 Baker et al 42 Orege et al 43 Thuc et al 44 Boelens et al 45 Lombardi et al 46 Bertolli et al 47 Zodpey et al 48 Zodpey et al 49 Zodpey et al 50 Combined Relative risk Point estimate Figure 2: Forest plots of the studies included in meta-analysis (A) Forest plot of the experimental studies. (B) Forest plot of the observational studies. Horizontal lines=95% CI. The rectangles represent the point estimates of the study and the size of the rectangle represents the weight given to each study in the meta-analysis. The diamond and vertical broken line represent the summary estimate; the size of the diamond represents the CIs of the summary estimate. The solid vertical line is the null value. Summary estimates The summary estimate for all the experimental studies was 0 74 and the heterogeneity 2 statistic was (p ). Therefore the Shore-adjusted 95% CI was calculated to be The summary estimate for observational studies was 0 39 with a heterogeneity 2 statistic of (p ); hence the Shoreadjusted 95% CI was calculated to be The summary estimate for experimental studies was significantly different from the summary estimate calculated from the observational studies (p 0 001). Figure 2 shows the forest plots for the fixed effects model for all the experimental and observational studies included for meta-analysis. Among the observational studies, the summary estimate of 14 case-control studies was 0 41 and of the five cohort studies was 0 31, not significantly different from each other. Vol 6 March

5 Summary Unadjusted Hetero- p Adjusted estimate 95% CI geneity ( 2 ) 95% CI* Experimental studies All studies (7) Age groups (age at inclusion in the trial) 15 years (5) years (4) Sex Male (2) Female (2) Scoring of studies on scale of (3) (4) Observational studies All studies (19) < Age groups 10 years (2) years (2) years (3) Sex Male (7) Female (7) Type of leprosy Paucibacillary (10) Multibacillary (10) Indeterminate (3) Type of contact Household contact (3) Other contact (3) *Shore-adjusted 95% CI. According to score by Jadad and colleagues. 22 Table 3: Summary estimates of effect of BCG on leprosy in subgroup analyses in experimental and observational studies Subgroup analyses Three trials received a quality score of 3 or more. The summary estimate of these trials was The summary estimate of other trials (quality score of less than 3) 25 27,31 was BCG was given to scar-positive individuals in the Karonga trial, 28 and study participants were revaccinated in the Karimuri study. 27 The summary estimate of these two trials with more than one dose of BCG was The summary estimate of the other four trials 25,26,29,31 was 0 75, which was significantly different from the trials with more than one dose of BCG (p=0 05). We excluded a study from India 30 to calculate the summary estimate for single dose because the population under study had a baseline prevalence of BCG scars of 17 4%. The results of the subgroup analyses are shown in table 3. The summary estimate for individuals under the age of 15 years enrolled in trials was 0 69, which was not significantly different (p=0 8) from the summary estimate for individuals over the age of 15 years (summary estimate of 0 66). The observational studies, however, showed maximum protective effect among individuals less than 10 years of age. There was no significant difference in the summary estimates of men and women either in the observational or experimental studies. In the experimental studies, the estimate for paucibacillary leprosy (including tuberculoid) was 0 80 and was not significantly different from the estimate for multibacillary leprosy (lepromatous included). Both these estimates were not significantly different from the summary estimate for indeterminate leprosy, which was However, the observational studies demonstrated an increased protective effect for multibacillary leprosy compared with paucibacillary leprosy with (p=0 04) or without (p=0 02) the inclusion of tuberculoid and lepromatous cases in the analysis. These studies also demonstrated an increased risk against indeterminate forms of leprosy. Although the protective effect was higher in the household contacts compared with other contacts in the observational studies (24% vs 12%), the summary estimates were not significantly different (p=0 5; table 3). However, these two subgroups had significant heterogeneity owing to one of three studies in the household contact subgroup and the same one of three studies in the other contact subgroup. After excluding this one study 41 from the analyses for both subgroups, the summary estimates changed to 0 05 for household contacts and to 0 07 for other, non-household contacts. Publication bias There appeared to be asymmetry in the funnel plot for experimental (figure 3A) and observational studies (figure 3B). For the experimental studies, the co-efficient for bias in Egger s test was 3 18 with p= The Begg s test had a p value of for correlation between the point estimate and its standard error. For the observational studies the co-efficient bias in Egger s test was 4 11 (p=0 003). The Begg s test had a p value of for correlation between the point estimate and its standard error. Discussion The meta-analysis of these studies provides useful information on the overall protective effect of BCG in the prevention of leprosy. The results from the experimental studies indicate that BCG vaccination offered an average protection of 26% (95% CI 14 37%) against leprosy, with significant heterogeneity between the trials (p ). The overall average protective effect of BCG estimated by the observational studies was 61% (95% CI 51 70%), also with significant heterogeneity between these studies (p ). Thus, the observational studies overestimated the protective effect of BCG vaccine in leprosy. In our meta-analysis, the protection was better for multibacillary forms of leprosy compared with paucibacillary forms; this difference was significantly higher in the observational studies. The experimental studies demonstrated a protective effect of 31% for indeterminate leprosy; however, the observational studies showed an increased risk for it. The observational studies demonstrated a reduced protection with increasing age; however, the protection was not significantly different among individuals under 15 years of age or 15 years or Vol 6 March 2006

6 over in the experimental studies. The protection was greater among women compared with men in both the experimental and observational studies; the difference, however, was not significant. BCG offered more protection among household contacts compared with other contacts, although the difference was also not significant. More than one dose of BCG was significantly more protective in the experimental studies. BCG was thought of as a vaccine for preventing leprosy after it was demonstrated that it led to lepromin conversion, a type of delayed hypersensitivity to lepra bacilli in human beings. Although we did not include the studies that measured changes in immune responses to skin testing for this meta-analysis, we reviewed these articles. Enhancement of immune response following BCG vaccination was demonstrated by Bottaso and colleagues 66 in Argentina among contacts of leprosy patients; however, Ganapati and co-workers 67 have argued that BCG and killed M leprae is a better combination in enhancing the immune response. Animal studies have also demonstrated that BCG is protective against M leprae. 68 Shephard and co-workers 69 have shown that BCG was most protective under two circumstances: just before the challenge with M leprae and just before the logarithmic multiplicative phase of M leprae in mice. Although some studies found BCG to be more protective in leprosy compared with tuberculosis, 34,64 the metaanalysis of the experimental studies demonstrated an overall protection of 26%, which was lower than the average protection of 50% in tuberculosis shown by Colditz and colleagues. 70 The protective effect of BCG in tuberculosis wanes with time after vaccination, 71 which was also seen in our meta-analysis of observational studies. The observational studies demonstrated a reduced protection with increasing age; however, we could not include all the studies as they had different age groups. Mulyil and colleagues 38 have suggested that change in host immune status after BCG vaccination may be responsible for a differential protection against multibacillary and paucibacillary leprosy. BCG could lead to an increased occurrence of milder forms such as paucibacillary (tuberculoid) and indeterminate leprosy due to improved host immunity, a finding that was observed in our meta-analysis. Multiple doses of BCG offered greater protection against leprosy. Although we just used the experimental studies to assess the role of multiple vaccinations, increased protection was also demonstrated among individuals with more than one BCG scar in observational studies. 41,47 As with all meta-analyses, our findings may be influenced by three important factors heterogeneity among studies, bias, and publication bias. Heterogeneity between studies is an important concern in meta-analyses. Studies can be heterogeneous by virtue of their designs, the population under study, the methods of exposure, and outcome assessment, etc. The fixed effects model takes into account the within-study variance and hence weighs A B log (RR) log (RR) SE of log (RR) Figure 3: Funnel plots to assess publication bias in the meta-analysis (A) Funnel plot for experimental studies. (B) Funnel plot for observational studies. RR=relative risk. the study according to the study size and precision. DerSimonian and Laird 72 have described a random effects model that accounts for the inter-study variation, the underlying assumption in this model being that the studies are from a hypothetical population and that the inter-study variance can be represented by a single estimate. Compared with the fixed effects model, the random effects model gives greater weight to less precise studies, which can have undesired effects on the summary risk estimates. 73 We felt that heterogeneity between the studies could be explained by differing study populations and methods. Hence, we chose to account for inter-study variance by a method initially described by Armitage 74 and later by Shore and colleagues. 19 In this method all the studies are weighed primarily on precision and a more conservative CI (wider CI) is obtained to account for the heterogeneity between the studies. Many factors have been considered to explain the varying protective effect of BCG in these studies. The study population in the experimental studies varied Vol 6 March

7 widely. In Uganda, 25 the population was contacts of leprosy patients. In Burma, 26 only children under the age of 14 years were included. In other sites, it was the general population in an endemic area. The trials that scored highly in terms of quality demonstrated a higher protective effect of BCG compared with the trials that scored poorly (ie, those with inadequate randomisation or blinding). Another factor discussed for varying protection of BCG is the presence of environmental bacteria in the study sites. Stanford and colleagues 75 and Rook and coworkers 76 have discussed how environmental mycobacteria could modify the type of cell-mediated response developed to BCG vaccination and hence the protective effect against tuberculosis and leprosy. By contrast with the trials in India 29 and Burma, 26 which were done in areas with high prevalence of leprosy and tuberculosis, the study in Karimuri, Papua New Guinea, 27 was done in an area with high leprosy prevalence but relatively free from other environmental mycobacteria, including tubercle bacilli. Colditz and co-workers 70 showed that the protective effect of BCG in tuberculosis increased with higher latitudes ie, it was effective further away from the equator. In leprosy, however, Fine 77 argued that studies that were done in sites closer to the equator had a greater protective effect and hence environmental and geographic factors could have a role in varying protection. BCG currently represents a group of vaccines with phenotypically and genotypically different strains. 78 Different BCG strains were used in vaccination of the various populations under study, which might account for some of the variation observed in different studies. Another potential reason could be the genetic differences in the populations under study, often resulting in varying susceptibility to infection. 79 Fine and colleagues 80 have discussed that varying degrees of tuberculin sensitivity was associated with reduced rates of leprosy infection in individuals. In addition, Foster and co-workers 81 have reviewed the literature on the role of nutrition in leprosy. They have suggested the role of adequate diet, vitamins (A, D, E, B complex, and C), calcium, and zinc in the aetiopathogenesis and prognosis of leprosy. Thus, differing baseline nutritional status of the population in these studies would be a potential factor for the observed heterogeneity. Of note is that our subanalyses of the protective effects of BCG among household and other contacts of patients with leprosy both exhibited significant heterogeneity. The heterogeneity in both analyses was caused by one study. 41 We had calculated unadjusted CIs for this study (adjusted odds ratio was 1 and unadjusted was 1). When this study was removed from the analysis, heterogeneity disappeared. However, the study that contributed to the heterogeneity, like the other two studies, demonstrated a protective effect both among household contacts (0 62) and other contacts (0 40), and this protective effect was statistically significant among other contacts (95% CI ). There are other biases that one needs to consider in meta-analyses. Knowledge of immunisation status of the trial participants could lead to over diagnosis of leprosy, especially in the early stages of clinical infection in unvaccinated individuals, leading to an ascertainment bias, in this case ascertainment of the outcome. Thus, the protective effect of BCG would be overestimated in these trials. In our meta-analysis, observational studies (cohort and case-control) overestimated the protective effect of BCG. Although case-control studies are less expensive and suited for a rare disease such as leprosy, there are limitations of these studies, and they are prone to biases. Selecting controls from households that have a leprosy case or close contacts of leprosy patients often leads to selection bias. 41 These controls are at a greater risk for infection than controls selected from the general population 38 and there could be a potential confounding, since people from the same household tend to have similar vaccination histories. It was often difficult to establish a confirmed history of vaccination in most of the populations where the case-control studies were done and hence the exposure was determined by the presence of a BCG scar. The sensitivity of BCG scar reading itself could vary as discussed by various authors. 43,82 There is potential for observer bias in these situations. Investigators reading and interpreting the presence of BCG scars, if aware of the study hypothesis, may be more likely to report the presence of scars in controls compared with the cases. This type of bias could lead to an odds ratio away from the null and hence an increased vaccine efficacy of BCG in these studies. The other potential source of bias in these studies is misclassification. However, a non-differential misclassification would tend to bias the effect towards null and hence underestimate the protective effect in leprosy. The case-control studies are usually designed to take into account the potential confounding factors, some of them being age, sex, and geographic location. If the cases and controls are matched on these confounding factors it is essential to take in to account the matching in analysis of the data. In addition to the above biases, an association between the presence of a BCG scar and protection against infections has also been postulated. Vaccinations that are more likely to leave scars probably are more likely to offer substantial protection. 82 Thus diseases would be reported more in unscarred individuals causing the results to be biased away from the null effect. We only used published works and hence an important bias we have to consider in the meta-analysis is publication bias. There is a tendency not to publish smaller studies that have a negative result or results closer to null effect, which often leads to publication bias in the existing literature. We assessed the publication bias by a funnel plot. The point estimates for each study were plotted against the standard error of the point estimates. An asymmetry in the funnel plot is indicative of a publication bias. We had few experimental studies and Vol 6 March 2006

8 there was asymmetry in the plot (figure 3A). We formally tested the funnel plot by a simple linear regression of the effect size, as described by Egger and colleagues. 24 In the test described by Begg and co-workers, 23 the Kendall s rank order test is used to assess the correlation between the effect size of the study and its standard error. There was no evidence of publication bias either with Egger s test (p=0 114) or Begg s test (p=0 133) for the experimental studies. Even though the larger studies were closer to the null, they still demonstrate protective effects of BCG in leprosy and reported relative risk estimates close to the summary estimate of 0 7 that we had calculated from the overall meta-analysis. The funnel plot was asymmetrical for the observational studies (figure 3B) and there was evidence of publication bias by Egger s test (p=0 003) and Begg s test (p=0 006). Thus, our meta-analysis of seven experimental studies and 19 observational studies demonstrated that BCG reduces clinical leprosy among vaccinees. The average protective effect among the experimental studies was 26% and among the observational studies was 61%. BCG vaccination offered greater protection against multibacillary leprosy. Although the effect waned with age, age at vaccination was not a predictor of BCG efficacy and an additional dose was more protective. However, BCG vaccine is not without adverse events. Fatal disseminated BCG has been reported among people with HIV infection 83 and this risk will need to be weighed against concerted efforts to provide additional BCG vaccine to prevent leprosy among certain subgroups eg, contacts. However, given the protective effects and its easy availability, additional BCG vaccination may be warranted, at least among the household contacts of multibacillary patients who are at greater risk of acquiring the infection, in regions where leprosy still continues to be public-health problem and elimination efforts need to be boosted. Conflicts of interest We declare that we have no conflicts of interest. Acknowledgments MSS was supported by Fogarty Grant number 1-D43-TW00003 for his MPH at the School of Public Health, University of California, Berkeley. We thank Professor Lee Riley, University of California, Berkeley, for his comments and feedback. We acknowledge Gail E Kennedy and Karishma Busgeeth, for assistance with the Embase search. References 1 Anon. Chemotherapy of leprosy for control programmes. World Health Organ Tech Rep Ser 1982; 675: Ji BH. Drug resistance in leprosy a review. Lepr Rev 1985; 56: Meima A, Smith WC, van Oortmarssen GJ, Richardus JH, Habbema JD. The future incidence of leprosy: a scenario analysis. Bull World Health Organ 2004; 82: Ji B. Why multidrug therapy for multibacillary leprosy can be shortened to 12 months. Lepr Rev 1998; 69: Visschedijk J, van de Broek J, Eggens H, Lever P, van Beers S, Klatser P. Mycobacterium leprae millennium resistant! Leprosy control on the threshold of a new era. Trop Med Int Health 2000; 5: WHO. Global leprosy situation, Wkly Epidemiol Rec 2005; 34: World Health Assembly. Elimination of leprosy: resolution of the 44th World Health Assembly. Geneva: WHO, WHO. Leprosy. factsheets/fs101/en/ (accessed Feb 3, 2006). 9 Fine PE. Primary prevention of leprosy. Int J Lepr Other Mycobact Dis 1996; 64 (suppl 4): S Fine P, Dockrell H. Leprosy vaccines. Vaccine 1991; 9: Gupte MD. Vaccines against leprosy. Indian J Lepr 1991; 63: Stanford JL. The history and future of vaccination and immunotherapy for leprosy. Trop Geogr Med 1994; 46: Talwar GP. Vaccines against leprosy. Asian Pac J Allergy Immunol 1985; 3: Cunha SS, Dourado I, Barreto ML, et al. Design of the leprosy component of the Brazilian BCG revaccination trial for assessing BCG effectiveness against leprosy in school children. Int J Lepr Other Mycobact Dis 2004; 72: Fernandez J. Estudio comparativo de la reaccion de Mitsuda con las reacciones tuberculinicas. Revista Argentina Dermatosifilis 1939; 23: Doull JA, Guinto RS, Mabalay MC. Effect of BCG vaccination, lepromin testing and natural causes in inducing reactivity to lepromin and to tuberculin. Int J Lepr 1957; 25: Yanagisawa K. On the immunological relationship between tuberculosis and leprosy with special reference to the effect of BCG administration upon the prophylaxis of leprosy. La Lepro 1960; 26 28: Greenland S. Quantitative methods in the review of epidemiologic literature. Epidemiol Rev 1987; 9: Shore RE, Gardner MJ, Pannett B. Ethylene oxide: an assessment of the epidemiological evidence on carcinogenicity. Br J Ind Med 1993; 50: Altman DG, Bland JM. Interaction revisited: the difference between two estimates. BMJ 2003; 326: Smith PG. Epidemiological methods to evaluate vaccine efficacy. Br Med Bull 1988; 44: Jadad AR, Moore RA, Carroll D, et al. Assessing the quality of reports of randomized clinical trials: is blinding necessary? Control Clin Trials 1996; 17: Begg CB, Mazumdar M. Operating characteristics of a rank correlation test for publication bias. Biometrics 1994; 50: Egger M, Davey Smith G, Schneider M, Minder C. Bias in metaanalysis detected by a simple, graphical test. BMJ 1997; 315: Stanley SJ, Howland C, Stone MM, Sutherland I. BCG vaccination of children against leprosy in Uganda: final results. J Hyg (Lond) 1981; 87: Lwin K, Sundaresan T, Gyi MM, et al. BCG vaccination of children against leprosy: fourteen-year findings of the trial in Burma. Bull World Health Organ 1985; 63: Bagshawe A, Scott GC, Russell DA, Wigley SC, Merianos A, Berry G. BCG vaccination in leprosy: final results of the trial in Karimui, Papua New Guinea, Bull World Health Organ 1989; 67: Karonga Prevention Trial Group. Randomised controlled trial of single BCG, repeated BCG, or combined BCG and killed Mycobacterium leprae vaccine for prevention of leprosy and tuberculosis in Malawi. Lancet 1996; 348: Gupte MD. Field trials of antileprosy vaccines. Indian J Lepr 1998; 70: Gupte MD, Vallishayee RS, Anantharaman DS, et al. Comparative leprosy vaccine trial in south India. Indian J Lepr 1998; 70: Truoc LV, Ly HM, Thuy NK, Trach DD, Stanford CA, Stanford JL. Vaccination against leprosy at Ben San Leprosy Centre, Ho Chi Minh City, Vietnam. Vaccine 2001; 19: Convit J. Studies of leprosy in the German ethnic group of Colonia Tovar, Venezuela. V. The morbidity rates in BCG-vaccinated and unvaccinated groups during five years. Int J Lepr 1956; 24: Chatterjee KR, Soucou P, Saint-Rose M. Prophylactic value of BCG vaccination against leprosy: a preliminary report. Bulletin of the Calcutta School of Tropical Medicine 1958; 6: Ponnighaus JM, Fine PE, Sterne JA, et al. Efficacy of BCG vaccine against leprosy and tuberculosis in northern Malawi. Lancet 1992; 339: Duppre NC. Efetividade do BCG-ID em comunicantes de pacientes om as formas multibacilares da hanseniase (MS thesis). Escola Nacional de Saude Publica, Rio de Janeiro, Brazil: Vol 6 March

9 36 Cunha SS, Rodrigues LC, Pedrosa V, Dourado IM, Barreto ML, Pereira SM. Neonatal BCG protection against leprosy: a study in Manaus, Brazilian Amazon. Lepr Rev 2004; 75: Abel L, Cua VV, Oberti J, et al. Leprosy and BCG in southern Vietnam. Lancet 1990; 335: Muliyil J, Nelson KE, Diamond EL. Effect of BCG on the risk of leprosy in an endemic area: a case control study. Int J Lepr Other Mycobact Dis 1991; 59: Rodrigues ML, Silva SA, Neto JC, de Andrade AL, Martelli CM, Zicker F. Protective effect of intradermal BCG against leprosy; a casecontrol study in central Brazil. Int J Lepr Other Mycobact Dis 1992; 60: Alvim MFS. Efeito protetor do BCG intradermico em contatos de pacientes de hanseniase: estudo de caso-controle (MS thesis). Universidade do Estado do Rio de Janeiro, Rio de Janeiro, Brazil: Convit J, Smith PG, Zuniga M, et al. BCG vaccination protects against leprosy in Venezuela: a case-control study. Int J Lepr Other Mycobact Dis 1993; 61: Baker DM, Nguyen-Van-Tam JS, Smith SJ. Protective efficacy of BCG vaccine against leprosy in southern Malawi. Epidemiol Infect 1993; 111: Orege PA, Fine PE, Lucas SB, Obura M, Okelo C, Okuku P. Casecontrol study of BCG vaccination as a risk factor for leprosy and tuberculosis in western Kenya. Int J Lepr Other Mycobact Dis 1993; 61: Thuc NV, Abel L, Lap VD, Oberti J, Lagrange PH. Protective effect of BCG against leprosy and its subtypes: a case-control study in southern Vietnam. Int J Lepr Other Mycobact Dis 1994; 62: Boelens JJ, Kroes R, van Beers S, Lever P. Protective effect of BCG against leprosy in South Sulawesi, Indonesia. Int J Lepr Other Mycobact Dis 1995; 63: Lombardi C, Pedrazzani ES, Pedrazzani JC, Filho PF, Zicker F. Protective efficacy of BCG against leprosy in Sao Paulo. Bull Pan Am Health Organ 1996; 30: Bertolli J, Pangi C, Frerichs R, Halloran ME. A case-control study of the effectiveness of BCG vaccine for preventing leprosy in Yangon, Myanmar. Int J Epidemiol 1997; 26: Zodpey SP, Shrikhande SN, Salodkar AD, Maldhure BR, Kulkarni SW. Effectiveness of bacillus Calmette-Guerin (BCG) vaccination in the prevention of leprosy: a case-finding control study in Nagpur, India. Int J Lepr Other Mycobact Dis 1998; 66: Zodpey SP, Bansod BS, Shrikhande SN, Maldhure BR, Kulkarni SW. Protective effect of bacillus Calmette Guerin (BCG) against leprosy: a population-based case-control study in Nagpur, India. Lepr Rev 1999; 70: Zodpey SP, Ambadekar NN, Thakur A. Effectiveness of bacillus Calmette Guerin (BCG) vaccination in the prevention of leprosy: a population-based case-control study in Yavatmal District, India. Public Health 2005; 119: Brown JA, Stone MM. A trial Of BCG vaccination in the prophylaxis of leprosy. Lepr Rev 1963; 34: Brown JA, Stone MM. B.C.G. vaccination of children against leprosy: first results of a trial in Uganda. Br Med J 1966; 5478: Brown JA, Stone MM, Sutherland I. B.C.G. vaccination of children against leprosy in Uganda: results at end of second follow-up. Br Med J 1968; 1: Brown JA, Stone MM, Sutherland I. Trial of BCG vaccination against leprosy in Uganda. Lepr Rev 1969; 40: Russell DA, Scott GC, Wigley SC. BCG vaccination in leprosy. A preliminary report of a blind controlled trial. Int J Lepr 1964; 32: Bechelli LM, Garbajosa G, Uemura K, et al. BCG vaccination of children against leprosy. Preliminary findings of the WHO-controlled trial in Burma. Bull World Health Organ 1970; 42: Bechelli LM, Garbajosa PG, Gyi MM, et al. BCG vaccination of children against leprosy. Preliminary findings of the WHO-controlled trial in Burma up to January Int J Lepr Other Mycobact Dis 1971; 39: Bechelli LM, Garbajosa PG, Gyi MM, et al. BCG vaccination of children against leprosy: seven-year findings of the controlled WHO trial in Burma. Bull World Health Organ 1973; 48: Bechelli LM, Lwin K, Gallego Garbajosa P, et al. BCG vaccination of children against leprosy: nine-year findings of the controlled WHO trial in Burma. Bull World Health Organ 1974; 51: Ponnighaus JM, Fine PE. The Karonga prevention trial which BCG? Lepr Rev 1986; 57 (suppl 2): Ponnighaus JM, Fine PE, Bliss L, et al. The Karonga prevention trial: a leprosy and tuberculosis vaccine trial in northern Malawi. I. Methods of the vaccination phase. Lepr Rev 1993; 64: Fine PE, Ponnighaus JM, Maine N, Clarkson JA, Bliss L. Protective efficacy of BCG against leprosy in northern Malawi. Lancet 1986; 2: Fine PE, Ponnighaus JM. Leprosy in Malawi. 2. Background, design and prospects of the Karonga prevention trial, a leprosy vaccine trial in northern Malawi. Trans R Soc Trop Med Hyg 1988; 82: Tripathy SP. The case for B.C.G. Ann Natl Acad Med Sci 1983; 19: Convit J, Sampson C, Zuniga M, et al. Immunoprophylactic trial with combined Mycobacterium leprae/bcg vaccine against leprosy: preliminary results. Lancet 1992; 339: Bottasso O, Merlin V, Cannon L, et al. Studies of vaccination of persons in close contact with leprosy patients in Argentina. Vaccine 1998; 16: Ganapati R, Revankar CR, Lockwood DN, et al. A pilot study of three potential vaccines for leprosy in Bombay. Int J Lepr Other Mycobact Dis 1989; 57: Gormus BJ, Baskin GB, Xu K, et al. Antileprosy protective vaccination of rhesus monkeys with BCG or BCG plus heat-killed Mycobacterium leprae: immunologic observations. Int J Lepr Other Mycobact Dis 2000; 68: Shepard CC. Vaccination against human leprosy bacillus infections of mice: protection by BCG given during the incubation period. J Immunol 1966; 96: Colditz GA, Brewer TF, Berkey CS, et al. Efficacy of BCG vaccine in the prevention of tuberculosis. Meta-analysis of the published literature. JAMA 1994; 271: Sterne JA, Rodrigues LC, Guedes IN. Does the efficacy of BCG decline with time since vaccination? Int J Tuberc Lung Dis 1998; 2: DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials 1986; 7: Poole C, Greenland S. Random-effects meta-analyses are not always conservative. Am J Epidemiol 1999; 150: Armitage P. Statistical considerations: conclusion. In: Wald NJ, Doll R, eds. Interpretation of negative epidemiological evidence for carcinogenicity. Lyon: International Agency for Research on Cancer, 1985: Stanford JL, Shield MJ, Rook GA. How environmental mycobacteria may predetermine the protective efficacy of BCG. Tubercle 1981; 62: Rook GA, Bahr GM, Stanford JL. The effect of two distinct forms of cell-mediated response to mycobacteria on the protective efficacy of BCG. Tubercle 1981; 62: Fine PE. The Kellersberger memorial lecture, The role of BCG in the control of leprosy. Ethiop Med J 1985; 23: Behr MA, Small PM. A historical and molecular phylogeny of BCG strains. Vaccine 1999; 17: Shields ED, Russell DA, Pericak-Vance MA. Genetic epidemiology of the susceptibility to leprosy. J Clin Invest 1987; 79: Fine PE, Sterne JA, Ponnighaus JM, Rees RJ. Delayed-type hypersensitivity, mycobacterial vaccines and protective immunity. Lancet 1994; 344: Foster RL, Sanchez AL, Stuyvesant W, Foster FN, Small C, Lau BH. Nutrition in leprosy: a review. Int J Lepr Other Mycobact Dis 1988; 56: Fine PE, Ponnighaus JM, Maine N. The distribution and implications of BCG scars in northern Malawi. Bull World Health Organ 1989; 67: Grange JM. Complications of bacille Calmette-Guerin (BCG) vaccination and immunotherapy and their management. Commun Dis Public Health 1998; 1: Vol 6 March 2006