The study site, described previously (5,6,9), is located in a predominantly urban region in the Orizaba Health Jurisdiction in the state of Veracruz, which encompasses 134 square km and has a population of 284,728 (10). The incidence rate of TB during the year 2000 for the state was higher than that for the nation (28.0 vs. 15.9 per 100,000 inhabitants) (11).

Community-based screening of chronic coughers (>2 weeks) was performed from March 1995 to October 2000. Patients with positive AFB sputum smears underwent epidemiologic, clinical (standardized questionnaire, physical exam, chest x-ray, and HIV test), and mycobacteriologic evaluation. Treatment was provided in accordance with official norms (12,13). Treatment outcomes were classified as previously described (6). Annual follow-up was performed for treatment outcome and vital status. Deaths were confirmed with death certificates. A close caregiver was interviewed to elicit signs and symptoms of the terminal illness and “probable cause of death” was assigned by two of the authors (JF, LF). Informed consent was obtained from participants. The study was approved by the appropriate institutional review boards.

Deaths were attributed to TB based on the death certificate (TB listed as the cause of death), interview of close caregiver (TB identified as probable cause of death), and active TB at the time of death (positive AFB smear after last treatment or during treatment if the patient did not complete treatment performed <6 months before death). A patient’s death was attributed to TB if two or more of these conditions were met. Bivariate and multivariate analyses were performed to describe association between sociodemographic (age, sex, household characteristics, occupation, ethnicity, years of formal education, place of residence, and social security), behavioral (usage of drugs and alcohol, and previous incarceration), clinical (previous diagnosis of TB; other associated diseases, previous hospitalizations; HIV infection, duration of symptoms previous to diagnosis; time elapsed between diagnosis and initiation of treatment and between initiation of treatment and smear conversion; and symptoms such as cough, hemoptysis, fever, night sweats, weight loss, and general malaise), bacteriologic (number of bacilli per microscopic field, drug resistance, RFLP, or spoligotype pattern), therapeutic (compliance, treatment outcome, and retreatment) variables, and death due to TB and other causes. Survival analyses included Kaplan-Meier curves and Cox proportional hazards model for TB and non-TB–related deaths using as reference time the period elapsed from diagnosis. Variables were entered into the models according to their statistical significance in univariate analysis and their biological relevance. The percentage of population attributable risk was calculated for variables that were included in the final model (21). DBASE IV and STATA 5.0 programs were used for data analysis (22).

During the study period, 454 patients were diagnosed with pulmonary TB. Most patients were men (270 [59.5%] of 454) and the median age was 42 years (range 12–97 years). Most came from lower socioeconomic status as indicated by household characteristics, formal years of education, and occupation. Of 438 patients, 74 (17%) lived in households with earthen floor, 214 (49.3%) of 434 had no access to potable water within the household, 293 (65.4%) of 448 had <6 years of formal education, and 104 (23.4%) of 445 were manual workers. The prevalence of HIV was 2.1% (9 of 429), combined resistance to isoniazid and rifampin was seen in 26 (6.7%) of 388 patients, and other resistance patterns in an additional 61 (15.7%) of 388. One hundred fifty-one (37%) of 413 patients had chest radiographs with evidence of pulmonary cavitation, and 72 (17.4%) of 413 patients had interstitial pulmonary infiltrates. Median time from initiation of symptoms to treatment was 101 days (range 3–2,307 days), median time from diagnosis to initiation of treatment was 5 days (range 0–594 days), and median time for sputum conversion was 42 days (range 15–1,228 days).

This study describes high mortality rates from TB in a cohort of pulmonary TB patients who resided in an area with a low rate of HIV infection and were treated in the context of a well-functioning DOTS program. Patients were followed for an average of 2.3 years after diagnosis. Tuberculosis was associated with high rates of deaths both during treatment and after treatment completion, default, or failure. The main independent risk factors for death due to TB were treatment default and being infected with multidrug-resistant M. tuberculosis. Additionally, data indicate that cases due to ongoing transmission of TB may have higher mortality rates than cases due to reactivation of latent disease. These results suggest that current techniques underestimate death associated with TB and provide further impetus not only to treat but also to prevent TB.

Case completion rates are the standard by which the effectiveness of DOTS-based treatment programs are judged. A large body of evidence collected under diverse settings demonstrates completion rates of 81% and cure rates of 73% (23). The epidemiology of death in these programs is less well studied. Studies performed in HIV-endemic settings constitute an exception as the high rate of deaths occurring while patients are still on antituberculous therapy are cited as an impediment to achieving desired cure rates (24).

Our study reports TB mortality rates both during and after treatment completion, default, or failure. Of the 443 patients who were started on therapy, 16 died before completing therapy. This frequency (3.6%) compares favorably to that reported in other studies (4% in the Gambia [25], 15% in Bolivia [26], 25% in Malawi [27], and 28% in India [28]). However, in contrast to previous reports, we had a median of 833 days of follow-up after the completion of therapy, during which 21 additional patients (4.6%) died of TB, comparable with reports from the 1960s and 1970s (29). This mortality rate exceeds that expected according to age-adjusted state mortality statistics (30).

Although not novel, the risk factors for death due to TB identified in this cohort are noteworthy; these findings suggest that, even in a high-quality TB control program, additional efforts could yield important benefits. Treatment default has been previously described as associated with higher mortality rates in Mexico (31) and elsewhere (32,33). Concerted efforts to further reduce default may disproportionately decrease deaths. We also found that drug resistance, and particularly multidrug resistance, are associated with death (6,14,34). Although the most appropriate strategy for managing drug-resistant TB is debated, these data provide additional impetus for evaluating novel approaches, such as those recently introduced in Mexico, for managing drug-resistant TB (13,35). We estimate that 60% of deaths due to TB in this setting were attributable to treatment default and multidrug-resistance. Additional studies are needed to quantitate the presumably greater mortality rates that result from default and drug resistance in other settings.

Although relatively uncommon in this setting, HIV was associated with a high overall mortality rate. Most of these deaths were due to non-TB–related death. HIV/AIDS was the main predictor for non-TB deaths in this cohort. In our study, eight of the nine HIV-seropositive persons died, three of them a year or more after completing antituberculous therapy. Several studies have demonstrated excess mortality rates after successful TB treatment in HIV-infected patients (24,36,37); the high rates have been attributed to HIV-related disease (38). These data emphasize the need to improve integration of quality treatment for both TB- and HIV-infected persons in populations that suffer from both diseases.

The most striking finding of the molecular epidemiologic component of this study is the association between clustering, which we interpret as indicative of recently transmitted disease, and TB-related death. Death due to TB was significantly more common among those who had recently transmitted disease than those with reactivated disease (28.6% vs. 7%, p=0.01). This association was independent of treatment default, multidrug resistance, time of occurrence of death, weight loss, and years of formal education. Furthermore, the effect was not modified when other variables indicative of sociodemographic level (such as occupation, characteristics of the household, or ethnicity) or clinical variables indicating other diseases, including HIV infection were introduced in the model. Inferring causality from such associations is difficult, it seems biologically implausible that being more likely to die made people more likely to acquire recently transmitted TB. A more plausible explanation is that rapidly progressing to disease contributes to deterioration of the health of these patients and thus increases their likelihood of death. The phenomenon is well described for other diseases such as measles. However, why this would be more pronounced for recently transmitted disease is unclear.

The association between clustering and death could be spurious because of limitations in molecular or conventional epidemiology. The validity of clustering as a proxy for recent transmission might be challenged by the fact that we did not perform fingerprint analysis on all isolates (39). Comparison between patients whose isolates were genotyped and those whose isolates were not genotyped showed that results of this study may not be generalizable to indigenous or lower socioeconomic groups. Our selection of the time interval allowed us to consider TB that was transmitted very recently and progressed to disease. The validity of using a 1-year interval was confirmed with the sensitivity analysis using different time periods (6, 18, 24, 30, and 36 months) as association between clustered cases and death persisted, despite modification of the time interval and by the fact that almost 50% of isolates with matching DNA fingerprint patterns occurred within 1 year of identification of the previous case. Conventional epidemiologic approaches to studying the cause of death are difficult. Death certificate data are notoriously unreliable, and whether patients died of TB or other causes is not certain (40). Therefore, we added other criteria that included the interview of a close caregiver and activity of TB at the time of death to validate our definition of death due to TB. The cause-of-death profile derived from interview of a close caregiver has been demonstrated to be useful for planning purposes (41). We consider that this definition adequately identified TB-related death as characteristics of patients dying from TB differed from patients dying from other causes in several important aspects (drug resistance, coexisting chronic conditions, and clinical severity) and allowed the identification of different risk factors for TB-related death and for death due to other causes.

If confirmed in other settings, the conclusions of this study have important implications for control programs. Most importantly, given that current surveillance data are collected at the conclusion of therapy, this method probably underestimates the true impact of TB on a population’s death. Given the increasing role of cost-efficacy modeling in setting health-care priorities, this oversight has important consequences. In addition, if recently acquired TB exhorts a greater mortality rate than that due to reactivated infection, the importance of interrupting TB transmission is further elevated.