Prior BCG vaccination improves survival of Gambian patients treated for pulmonary tuberculosis

Authors


correspondence Dr Tumani Corrah, MRC Laboratories Fajara, PO Box 273, Banjul, The Gambia. E-mail tcorrah@mrc.gm

Abstract

Summary The protection provided by BCG against pulmonary tuberculosis ranges from nil to over 90%. While BCG protects against the more serious forms of tuberculosis, it is not known whether or not it protects patients with pulmonary tuberculosis from death. In a study designed to look at the effects of immunotherapy with M. vaccae as an adjunct to chemotherapy in 285 adult Gambian patients treated for proven pulmonary tuberculosis, we examined the association between the presence or absence of a BCG scar and mortality. The data showed that subjects who had a BCG scar were significantly younger than those who did not, and were less likely to have nutritional oedema. During the course of treatment, none of the 85 patients who had a BCG scar died compared to 35 of 200 patients (17.5%) who did not (P < 0.001). In these Gambian patients with pulmonary tuberculosis, prior vaccination with BCG may have provided substantial protection against death. However, there is the possibility that this finding is the result of confounding by other factors or has arisen from bias. Researchers with similar data need to investigate this question as this association, if true, could have major implications for BCG vaccination.

Introduction

For reasons that are poorly understood, the protection provided by BCG vaccination against pulmonary tuberculosis (TB) varies widely between populations, ranging from being possibly detrimental to over 90% ( Bloom & Fine 1994; Colditz et al. 1995 ; Fine 1995). Recently, repeated vaccination with BCG gave disappointing results ( Karanga Prevention Trial Group 1996). Against leprosy, BCG provides between 20% and 90% protection ( Fine 1988). It also protects against the more serious types of TB (tuberculous meningitis and disseminated tuberculosis), but these account only for the minority of cases and are less important in terms of transmission ( Rodrigues et al. 1993 ). Whether or not BCG protects adult TB patients against mortality is not known. As part of a study on immunotherapy, we examined the survival of 285 patients treated for pulmonary tuberculosis, and our results show that patients who had a BCG scar were significantly less likely to die than those who did not.

Subjects and methods

This study was part of a larger investigation into the effect of immunotherapy with M. vaccae as an adjunct to chemotherapy in adult Gambian patients treated for proven sputum-positive pulmonary TB. Patients diagnosed with pulmonary tuberculosis at the Medical Research Council Laboratories Hospital at Fajara, The Gambia, between 1985 and 1989 were invited to take part in the study. All patients were clinically examined on enrolment, inspected for a BCG vaccination scar, chest X-rayed (CXR), had their sputum examined for acid fast bacilli (AFB) and cultured for M. tuberculosis. The deltoid region of both arms was examined for the presence of a BCG scar by the study physician (TC); typically, a BCG is a round, slightly depressed area of 4–7 mm diameter with irregular edges ( Fine et al. 1989 ). Sputum and CXR examinations were repeated after 8 weeks and at the end of a course of chemotherapy. Radiographs were coded to exclude bias and assessed blindly by an independent radiologist who took no further part in the study. Each film was divided into upper, middle and lower zones and graded with respect to area and type of disease distribution. Effusion was given the same score as infiltration. A low score indicated severe disease; a high score, mild disease. Criteria of progression were derived from comparisons between initial and subsequent films and classified as greatly improved, improved, no change and regressed. All patients were tested for HIV-1 and HIV-2 using two type-specific competitive enzyme linked immunosorbent assays (ELISA; Wellcome Laboratories, Dartford, Kent, UK). Positive sera were confirmed by Western Blotting (New LAV Blot I and New LAV Blot II, Diagnostics Pasteur, Marnes-la-Coquette, France).

During the course of the study, standard chemotherapy regimens in The Gambia changed from long-term chemotherapy (daily streptomycin injections, oral isoniazid and thiacetazone for 2 months, followed by daily isoniazid and thiacetazone for 16 months) to short course oral chemotherapy (supervised thrice weekly rifampicin, isoniazid, ethambutol and pyrazinamide for 2 months, followed by thrice weekly rifampicin and isoniazid for 4 months). Patients who agreed to join the study were allocated to the chemotherapy regimen that was in force in their own region at the time. Six weeks after initial presentation patients were randomly allocated to receive either a single immunotherapeutic injection of killed Mycobacterium vaccae or a saline placebo.

Patients were followed throughout the course of their treatment; those who failed to return were traced by fieldworkers and encouraged to attend clinic. Patients who could not be traced or whose survival status could not be confirmed were considered lost to follow-up.

After the study, to assess the confounding effect of socio-economic status, we determined retrospectively the socio-economic parameters for all of the patients who died and for a group of 70 study patients who survived, half of whom had a BCG scar. The 70 survivors were selected using a list of random numbers. Fieldworkers traced and visited their households and those of the 35 who had died and identified factors relating to socio-economic status.

Statistical methods

Discrete data were compared between groups using the χ2 test with Yate's correction or by Fisher's exact test, with stratified analysis by the Mantel–Haenszel χ2 test. Multivariate analysis was by logistic regression modelling. This procedure breaks down if the sum of events in any category is zero as was the case for deaths among patients who had a BCG scar. Therefore, to make this procedure work, we re-classified the one patient who had a BCG scar and had been lost to follow-up as having died. This assumption will have under-estimated the adjusted protective effect of a BCG scar. Survival analysis was conducted using Kaplan–Meier plots and the log-rank test. Continuous normally distributed data were compared between groups using t-tests for which homogeneity of variance was checked by the F-test. All other continuous data were compared using the Wilcoxon test. The SAS System for Windows and Epi-Info were used for all analyses.

Ethical approval

The study was approved by the Gambia Government/Medical Research Council Ethical Committee.

Results

We recruited 285 patients, ranging in age from 14 to 68 years, 85 of whom had a BCG scar. The presenting features of the two groups of patients are shown in Table 1. Subjects who had a BCG scar were significantly younger and less likely to have nutritional oedema than subjects who did not. Other presenting features were similar between the two groups.

Table 1.  Baseline characteristics according to presence or absence of a BCG scar Thumbnail image of

The outcome of illness is shown in Table 2. None of 85 subjects who had a BCG scar died, whereas 35 (17.5%) of the 200 who did not have a BCG scar (≤ 0.0001, χ2 test) did. 86% of those who had a BCG scar were cured compared to 61% of those without (< 0.0001, χ2 test). A significantly higher proportion of people who did not have a BCG scar were lost to follow-up (= 0.02, χ2 test). There was no significant difference between the groups in the proportion of subjects who were treatment failures but who survived.

Table 2.  The effect of chemotherapy, immunotherapy, and age on outcome according to the presence or absence of a BCG scar Thumbnail image of

After stratifying by age (as grouped in Table 2), analysis showed that patients who had a BCG scar were less likely to die than patients who did not have a BCG scar (= 0.0004, χ2M–H). Type of chemotherapy and immunotherapy were not related to BCG scar or to outcome. Ten subjects had nutritional oedema. Of 275 patients who did not have nutritional oedema, those who had a BCG scar were less likely to die than those who did not (= 0.0002, χ2 test). This effect remained after stratifying by age (= 0.001, χ2M–H–test).

In the group of patients without a BCG scar death occurred throughout the period of treatment but most frequently during the first 3 months: the probability of survival was 89% after 3 months and 79% after 1 year of follow-up ( Fig. 1). In this group, the mean age of patients who died was higher than the mean age of patients who survived although this was not statistically significant (difference in mean age = 3.2 years, 95% confidence interval =− 1.8, 8.2; = 0.21, t-test). Those who were cured were on average slightly younger than those who were not, although this was not statistically significant (difference in mean age = 2.0 years, 95% confidence interval =− 3.0, 7.1; = 0.4, t-test). This result was virtually unchanged after adjusting for scar status.

Figure 1.

Kaplan–Meier survival plot of 200 patients who had a BCG-scar and 85 patients who did not have a scar.

A multivariate logistic regression was conducted assuming that the one patient with a BCG scar who was lost to follow-up had died (see methods). This analysis showed that, after adjusting for age (considered as a continuous variable recorded in years), type of immunotherapy, type of treatment, and presence or absence of nutritional oedema, patients who had a BCG scar were 10.2 times (95% CI 1.3, 81.3) less likely to die than patients who did not have a BCG scar (= 0.0033, likelihood ratio test). Similar results were found when this analysis was repeated using only the data from those who did not have nutritional oedema.

We compared several socio-economic factors among those who died and among a random sample of subjects who survived (see methods): type of housing, number of inhabitants, education level, and ownership of domestic animals including cows or goats, land ownership, possession of household items such as television, floor mats, or a refrigerator, and distance to the nearest health centre. We first assessed whether or not the prevalence of these factors was the same among those who had a scar and those who did not have scar. The 70 subjects who had survived, half of whom had a scar and half did not, were used in this analysis. Subjects who did not have a scar were less well educated (χ2 = 10.3 on 1 d.f.; = 0.001), had poorer housing (χ2 = 13.5 on 2 d.f., = 0.001) and lived further away from a health centre (= 0.04, Wilcoxon test). We next assessed whether these three factors might have confounded the association between scar status and death, using the 70 subjects without a scar, half of whom had died. There was no association between death and either level of education (χ2 = 0.04 on 1 d.f.; = 0.83), type of housing (χ2 = 3.1 on 2 d.f., = 0.22) or distance to the health centre (= 0.9, Wilcoxon test).

Discussion

This hospital-based study has shown that no patient with pulmonary tuberculosis who had a BCG scar died, in contrast to 17.5% of those who did not have a BCG scar. While this is a powerful effect, there are several important reasons why this should be interpreted with caution. The first problem is the possibility that scar status was misclassified. The BCG scar can disappear over time, or other scars, such as from smallpox, can be mistaken as being BCG scars ( Fine et al. 1989 ). In our population, the proportion of vaccinated people who develop and retain a recognizable scar following vaccination is not known, but this has been reported to vary from 98.9% in a vaccine trial in South India 4 years after vaccination ( Tuberculosis Prevention Trial 1980) to 60% in Swedish children assessed 14 years after vaccination at birth ( Beskow et al. 1980 ). Conversely, a proportion of our nonvaccinated subjects may have had a scar, as traumatic and decorative scars in some populations may be mistaken for BCG scars ( Pyakalia & Pust 1977). However, this figure is even harder to measure as it is difficult to verify a negative vaccination history. To minimize the possibility of this error, scar status was assessed by one person (TC). On the whole, because of the size of the survival effect, it may not be possible to explain the difference in mortality between the scar and nonscar groups on the basis of misclassification alone unless this was substantial.

A second reason for caution is that this result might be due to confounding which could have occurred in a variety of ways. A significantly higher number of people without scars were lost to follow-up than subjects who had a scar. However, even if all those lost to follow-up in the non-BCG scar group are considered to have lived and the ones in the BCG scar group are considered to have died, the difference in mortality between the groups remains statistically highly significant. Subjects who did not have a BCG scar were on average 11 years older than subjects who did. This is probably because younger adults were vaccinated through mass campaigns before the Gambian EPI was implemented in 1978. However, as age was not related significantly to fatal outcome among patients who did not have a scar, the difference in mortality between subjects with and without a BCG scar may not be explained just by the difference in mean age between the two groups. Indeed among all patients, the difference in mortality between those with a scar and those without was consistent after stratifying for age. However, the difference in age was large and may be hiding differences between the two groups on which we have no information and which may be related to a poor prognosis. We looked further for confounding, by ascertaining and comparing the socio-economic status and education of the 35 nonvaccinated people who died with 70 randomly selected vaccinated subjects who survived. This showed that none of these factors could have confounded the protective effect of a BCG scar against death, although the number of patients in these analyses was small. A further consideration in interpreting these results is that mass BCG campaigns were usually targeted through schools which are attended by children from affluent families more frequently than from children from poorer, at-risk households. However, had this been the case, one might have expected more vaccinated boys than girls, but in fact the converse was true.

We have found only one study in which the clinical course of tuberculosis has been compared among adult patients with and without evidence of BCG vaccination ( Nunn et al. 1992 ). In this, 174 HIV-negative patients were studied over a period of 6 months. One hundred and three had a scar, 63 did not and the scar status of 10 patients was not known. Two patients who had a scar, as compared to eight who did not, died, but because follow-up data were limited, these findings were inconclusive; the risk of death was 3.3 times less for patients with a scar than for patients without (95% confidence interval 0.5, 10.0; not significant).

We do not know the mechanism by which BCG vaccination would have protected these patients from death. One possible explanation is that it might have limited haematogenous spread, thus facilitating cure. This is supported by the finding that BCG has generally been observed to provide good protection against meningitis and miliary TB even in trials where the vaccine offered little protection against other forms of TB ( Fine 1988). At the molecular biology level, BCG has been shown to be necessary for the control of mycobacterial infections by triggering a TH1 response ( Ravn et al. 1997 ; Ottenhof et al. 1998 ). Thus the combination of the controlling effect of BCG and the effect of chemotherapy could explain the protection that BCG may offer against death.

In summary, this result is difficult to interpret because of possible confounding or bias, but the magnitude of the effect, 35 deaths among 200 people without a scar compared with none among 85 people who had a scar, is such that the survival benefit may not be explained away easily. Other groups of researchers with similar data need to address this question and new studies need to be designed, for if this association is true, it could have major implications for BCG vaccination.

Acknowledgements

We thank Mamadi Njie and Haddy Fye for their careful fieldwork and bacteriology and Mike Pearson for reading the X-rays. The constant support of Professor Brian Greenwood was invaluable; we are also grateful to Professor Peter Smith. JS was partly funded by the British Council. SJ was funded by the British Medical Research Council.

Ancillary