Does the onset of tuberculosis in AIDS predict shorter survival? Results of a cohort study in 17 European countries over 13 years

Introduction

The HIV epidemic has had a major impact on the incidence of tuberculosis and on mortality and the case fatality rate of that disease.1 2 3 It is unclear, however, whether tuberculosis affects the course of HIV disease. In particular, we do not know whether tuberculosis shortens survival in patients with HIV infection. A recent review concluded that there was no noticeable decrease in survival attributable to tuberculosis in patients with HIV infection.1 That conclusion rested on two kinds of evidence. Firstly, in patients who had died with both AIDS and tuberculosis the cause of death was almost invariably attributed to AIDS.4 However, the validity of a cause of death diagnosis may be limited when there are several possible causes of death. Secondly, studies from Brazil,5 Spain,6 and Europe7 indicate that patients with tuberculosis as the AIDS defining disease survive longer than other AIDS patients. But because clinical tuberculosis may be manifested earlier than other AIDS defining diseases in HIV infection, apparently longer survival in these patients may be due entirely to earlier diagnosis of AIDS.

We analysed data from a large European follow up study of AIDS patients (AIDS in Europe study) to assess the impact of tuberculosis on mortality in this population. To avoid using tuberculosis both to define AIDS and to predict survival after the diagnosis of AIDS we restricted analysis to patients who were free of tuberculosis at the time of their AIDS diagnosis.

Patients and methods

This non-concurrent prospective study included AIDS patients followed up at 52 centres in 17 European countries.7 The cohort comprised patients diagnosed as having AIDS between 1979 and 1989 and followed up until 1992. The collaborating centres, except those in Italy, enrolled all of their AIDS patients who were aged 16 or over; the Italian centres enrolled only a predefined proportion of patients. Information was collected from patients' charts on a standardised form in 1991-2 under the supervision of the coordinating centre (University of Copenhagen). Of 6655 registered patients, 6546 had complete data. Of these patients, 5249 who were alive and free of tuberculosis 31 days after their diagnosis of AIDS were included in the analysis.

Study variables—AIDS was defined according to Centers for Disease Control criteria, 1987 revision.8 Patient follow up started on day 32 after the diagnosis of AIDS and was terminated when the patient was last seen alive or at death. Dates of death, time the patient was last seen alive, and diagnosis of tuberculosis (pulmonary or extrapulmonary) were based on medical charts. Eighty five per cent of tuberculosis cases (n=171) were confirmed by culture, 5% (11) were diagnosed at necropsy, and 9% (19) were diagnosed by direct smear examination, clinical assessment, and response to treatment. Potential confounders of the relation between the incidence of tuberculosis and mortality included age, sex, time of diagnosis of AIDS, region of Europe (northern, central, southern), and whether the patient took zidovudine at baseline.

Analysis—We compared patients who went on to develop active tuberculosis during follow up with those who remained free of tuberculosis. Firstly, we compared survival in these two groups (Kaplan-Meier method) ignoring the time of diagnosis of tuberculosis and adjusting for potential confounders in a proportional hazards model.8 To take into account that the longer a patient survives the more likely that patient is to develop a disease (including tuberculosis) we also modelled tuberculosis as a time dependent variable.9 In such models a patient with AIDS contributes to the estimation of the hazard of dying in the unexposed group as long as he or she is free of tuberculosis and to the estimation of the hazard of dying in the exposed group from the moment tuberculosis is diagnosed. The variable “tuberculosis” took a value of zero before the diagnosis of tuberculosis and a value of 1 thereafter. Analyses were by the SAS procedure PHREG.

Results

A total of 5249 patients were included in the study (table 1). Patients who developed tuberculosis were similar to the others in sex and main risk factors for HIV. Patients who developed tuberculosis were slightly younger than the other patients (34.2 versus 35.9 years; P=0.04), less likely to live in northern European countries (relative to central and southern countries), less likely to be taking zidovudine at the time of their AIDS diagnosis, and less likely to have developed AIDS in 1989 (relative to previous years).

During a mean follow up period of 15 months 3889 (74%) patients died and 201 (4%) developed a new episode of active tuberculosis. The tuberculosis was pulmonary in 80 cases and extrapulmonary in 107; 14 patients developed both forms. Tuberculosis was diagnosed on average 13 months after the start of follow up. Survival after the diagnosis of AIDS was significantly longer in patients who developed tuberculosis (median 22 months) than in the other patients (median 16 months; log rank test: P<0.001). The apparently lower mortality after the diagnosis of AIDS in patients who developed tuberculosis remained significant after adjustment for confounders in proportional hazards models (unadjusted relative hazard of death 0.72 (95% confidence interval 0.61 to 0.85); relative hazard of death adjusted for age, calendar year, region of Europe, and zidovudine treatment 0.69 (0.59 to 0.81)).

In analyses that allowed the risk of death to change in a given patient at the time tuberculosis was diagnosed (proportional hazards models with tuberculosis as a time dependent variable) the onset of tuberculosis appeared to increase the mortality of AIDS patients (table 2). This increase was 40% in a univariate analysis and 34% after adjustment for potential confounders. The excess risk of death was somewhat greater for pulmonary tuberculosis than for extrapulmonary tuberculosis.

The increased mortality after the onset of tuberculosis can also be expressed in terms of reduced patient survival. Under reasonable simplifying assumptions, expected survival would be divided by 1.34 (the relative hazard estimate) by the onset of tuberculosis. This corresponds to a reduction of 25%. For example, an AIDS patient who might live 12 months without tuberculosis would be expected to survive only nine months after the onset of tuberculosis.

These estimations of months of life lost assume that tuberculosis increases mortality in the same way, regardless of how much time has elapsed between the diagnosis of AIDS and the onset of tuberculosis. We tested this assumption by using two separate time dependent variables for tuberculosis in proportional hazards models—one for tuberculosis occurring 365 days or less after the start of follow up and another for tuberculosis occurring after more than one year of follow up. The adjusted relative hazards of death were 1.30 (95% confidence interval 1.05 to 1.60) for tuberculosis diagnosed in the first year of follow up and 1.54 (1.12 to 2.11) for tuberculosis diagnosed thereafter.

Discussion

This study of tuberculosis and mortality based on a large population based cohort of European AIDS patients yielded a seemingly paradoxical result. Patients who developed active tuberculosis at some point during their follow up survived longer after their diagnosis of AIDS than patients who remained free of tuberculosis. On the other hand, the development of tuberculosis predicted a substantial and statistically significant increase in the risk of death. This apparent contradiction may be explained as follows. Patients who survive for a long period have a greater chance of developing tuberculosis than do patients who die early after the diagnosis of AIDS. But among patients who survive free of tuberculosis for any length of time, those who contract tuberculosis die sooner than the others.

Observing long survival in AIDS patients who develop tuberculosis may have led clinicians to underestimate the detrimental impact that tuberculosis may have on vital prognosis. Such underestimation can result from faulty heuristic interpretation of the association between long survival and tuberculosis. Causation may be taken as flowing in one direction (tuberculosis results in long survival) instead of the other (long survival increases the chance of developing tuberculosis). Our analysis avoided this particular pitfall.

Survival was reduced by about 25% in AIDS patients who developed tuberculosis. Preliminary evidence suggests that the relative reduction in survival may be greater for pulmonary tuberculosis than for extrapulmonary tuberculosis and greater when tuberculosis occurs late rather than early after the diagnosis of AIDS. The relative hazard of death estimated in this study was weaker than the twofold increase in mortality reported in a study of 106 AIDS patients with tuberculosis compared with controls matched for CD4 cell count10; however, the confidence intervals of the two estimates overlapped.

Our study does not prove that the onset of tuberculosis increases mortality. It is possible that an unmeasured variable (such as severe immunosuppression or malnutrition) may induce both active tuberculosis and death in AIDS patients. Several arguments, however, favour a causal association. Though active tuberculosis is in principle curable, several studies point to dismal rates of compliance with antituberculous treatment in HIV positive patients.11 12 This problem is compounded by the increasing frequency of multidrug resistant strains of Mycobacterium tuberculosis in patients with HIV.13 Preliminary evidence also suggests that tuberculosis may accelerate the progression of HIV disease.14

There are few predictors of death among AIDS patients that may lend themselves to preventive intervention. Hence the possibility that active tuberculosis may increase by one third the mortality in AIDS patients deserves consideration. Preventive chemotherapy may reduce not only the incidence of tuberculosis15 but also mortality in HIV infected patients at high risk of mycobacterial infections. Early detection and appropriate treatment of tuberculosis may be particularly effective in HIV infected populations that now have limited access to health services. The effectiveness of such interventions cannot be inferred from this observational study but requires verification in experimental trials.