Study design

A retrospective cohort study was conducted at hospitals and TB control health centers chosen by convenience of the National Strategy for Prevention and Control of Tuberculosis (ESNPCT) in Lima, Peru. The data collected were all the records of patients enrolled in the ESNPCT and their follow-up information during their anti-tuberculosis treatment until therapy completion or death (censorship).

Study settings and patients

We retrieved information from all patients aged 18 years and above that began and completed their anti-tuberculosis treatment for a non-XDR pulmonary TB disease in the San Juan de Miraflores—Villa María del Triunfo Health Network (SJM-VMT) in Lima, Peru between January 2000 and December 2012. If the anti-tuberculosis treatment initiation date was not available in the clinical record, the patient was excluded from analysis. In addition, treatment failures were also excluded due to the risk of misclassification in the exposure of interest.

Variable definition

The main outcome of interest for both objectives was time to death. Death during follow-up was defined as the deceased condition of a patient during treatment confirmed by the presence of a death notification form or by explicit indication in the clinical record. For the survival analysis, the follow-up time was censored at 180 days. Therefore, the time until event was defined as the time elapsed, in days, from the beginning of the anti-tuberculosis treatment to the date of completion or death. If the patient was transferred or abandoned therapy, then time until event was censored at the moment of transfer of therapy abandonment.

For the first objective, type of tuberculosis was dichotomized as MDR-TB or drug-susceptible TB according to information of clinical records, based especially upon clinical assessment and the use of first- or second-line drugs, at the beginning of treatment. Thus, in the case of MDR-TB forms, diagnosis and classification was made following international standards and national guideline protocols (culture and drug susceptibility tests). This information was confirmed by checking treatment schemes as well as laboratory results. The remaining patients, those without a culture or drug-susceptibility test, were categorized as presumed drug-susceptible TB cases.

For both objectives, several other variables were included: sex (male vs. female), age (in tertiles), education level (≤6 years vs. 7–10 years vs. ≥11 years), previous TB episodes (0 vs. 1 vs. 2 or more), diabetes history (yes vs. no), anemia (positive vs. negative), HIV infection (positive vs. negative), self-reported smoking (yes vs. no), self-reported alcohol use (yes vs. no), self-reported illicit drug use (yes vs. no) and body mass index (BMI) (underweight vs. normal vs. overweight/obesity). All variables were assessed at the beginning of the anti-tuberculosis treatment.

Procedures and data collection

Sputum smear is indicated to those in whom tuberculosis is suspected, according to Peruvian National Guidelines [23]. The ESNPCT enrolls any patient with a positive result and the patient starts first-line anti-tuberculosis treatment. Culture and drug-susceptibility tests are performed in order to investigate a possible MDR-TB case if any of the following risk factors are present in a patient: confirmed MDR-TB contact, immunodeficiency (HIV or diabetes history), treatment relapse within six months after an active TB disease, TB cases treated multiple times, health workers, convicts, illegal drug users, contact with a person who died with TB or failed to complete treatment, hospitalization, irregular treatment schemes and adverse drug reaction to the treatment [23]. The patient’s treatment scheme is then adapted accordingly, if necessary. Patients without these risk factors are given first-line drugs. They are considered drug-susceptible TB cases for this analysis. Patients are followed and monitored during the duration of their whole treatment.

Data was collected directly from the patient’s clinical onto a single database. Only the investigators performed data collection during this study.

Sample size and statistical analysis

Power and Sample Size (PASS) software was used for sample size calculation. Assuming a confidence level of 95% and a power of 80%, a minimum total of 1090 patients were required to find a Hazard Ratio (HR) of 2.0 or over for any variable of interest. Calculations assumed a MDR-TB prevalence of 5.4% and an expected mortality of 6%.

After a double data entry process, statistical analysis was performed using STATA for Windows (STATA Corp, College Station, TX, USA). Patients who failed treatment during follow-up were excluded from analysis. Mortality curves were obtained through the Kaplan-Meier method. Each variable was evaluated with a Log-Rank test to verify potential association with mortality. Cox Regression models (crude and adjusted) were used to compare MDR-TB vs. drug-susceptible TB mortality, reporting Hazard Ratio (HR) with a 95% confidence interval (95%CI). Factors associated with mortality among MDR-TB cases were also assessed using Cox regression methods, assessing proportional hazard assumptions.

Ethics

This investigation was approved by the Ethics Committee of the Universidad Peruana de Ciencias Aplicadas (UPC), Lima, Peru and by the Institutional Ethics and Investigation Committee of Hospital María Auxiliadora, head of the SJM-VMT Health Network. The identity of participants was not recorded in order to guarantee anonymity.

MDR-TB and mortality

Our results demonstrate that multidrug-resistant strains of tuberculosis substantially increase the risk of death when compared to presumed drug-susceptible TB cases during anti-tuberculosis treatment. Recent studies have found similar findings where MDR-TB is associated with a higher mortality rate [24,25], with HR estimates ranging from 7.8 to 8.5. Our results have been controlled for several confounders, which makes a stronger adjusted association.

The cause behind this increased risk of death attributable to multidrug-resistant forms is multifactorial. MDR-TB is commonly seen as a co-morbidity to other conditions such as HIV infection, diabetes and chronic kidney disease, as well as particular lifestyles including alcohol abuse, smoking, drugs use, etc [21]. Furthermore, MDR-TB has been associated with elevated rates of treatment failure and relapse, which directly increases the proportions of death amongst these patients [10]. The longer exposure to anti-tuberculosis drugs might also increase the risk of death as antibiotics used for MDR-TB have proven higher toxicity profiles and greater incidence of adverse effects [10]. There is no doubt, then, that a multidisciplinary management of drug-resistant TB is needed and of great importance.

Scientific literature indicates that drug-resistant tuberculosis has increased over time. In 1990, Peru was believed to have a MDR-TB prevalence of 1%; however, recent publications have reported a prevalence of up to 5.3% [5], and our study suggests approximately 20% might have MDR-TB. This rise in resistant tuberculosis proportion over time might be related to many factors, namely treatment abandonment and other epidemiologic circumstances such as the existence of clusters or overcrowded households—especially in resource-limited settings [26]; nonetheless, the use of new diagnostic tools for detecting drug-resistant strains may be exposing the actual context of MDR-TB, reassuring the existence of this public health problem.

Factors independently associated with mortality among MDR-TB cases

Education level: According to our results, MDR-TB patients with lower education level have a greater risk of death during specific treatment. Of importance, education has been associated with a better adherence to treatment as it increases awareness regarding the disease [26]. Education has also been recognized as an economic status marker; in this case, lower education may be associated with lack of resources, overcrowding and unsanitary conditions.

Previous TB episodes: A greater number of previous TB episodes has been similarly associated with a higher risk of death. Patients under multiple regimens of anti-tuberculosis treatment might create greater antibiotic resistance with the subsequent development of MDR- and XDR-TB cases [2]. This causal link between number of past episodes and the appearance of drug-resistant strains has been thoroughly mentioned in scientific literature [6, 12]. However, the reasons for the association reported in this study require further investigation.

Diabetes history: Diabetes history increases the risk of mortality because it reduces cellular immunity, which favors the progression of the disease [28]. This association has been demonstrated in previous studies [19, 20]. Given the high frequency of diabetes as a co-morbidity among MDR-TB patients [27], screening for this disease should be encouraged in order to guarantee an appropriate treatment and control of both pathologies, as its awareness can reach up to 60% [28].

HIV Infection: HIV infection was also associated with a greater death proportion in MDR-TB patients. Investigations regarding this topic state that around 12% to 20% of all tuberculosis-related deaths may be attributed to an HIV co-infection [22,29], even if highly active antiretroviral therapy (HAART) has shown to reduce the overall mortality [18,30]. The results in our analysis might be the reflection of the HIV screening protocol done to all MDR-TB cases that was recently implemented by the ESNPCT.

Strengths and limitations

This study includes over 200 cases of MDR-TB under a directly-observed treatment scheme (DOTS), which is a notable strength. In addition, sample size allowed us to control for several variables in our proposed models. Nevertheless, this study has several limitations. First, non-TB related deaths such as accidents or other chronic diseases could have been included as the death certificate stating specific cause of death was not available. Although this might propose a bias in our findings, the results are greatly compatible and comparable with literature. Second, it was not possible to obtain drug-susceptibility test data for non-MDR TB participants. Drug-susceptibility and cultures are performed only in those with high risk or suspected MDR-TB, yet are not done in patients without such risk factors. Third, the starting point of evaluation, especially in the situation of MDR-TB cases, can be an issue as potentially resistant forms of tuberculosis may need longer periods of diagnosis and have previously administered treatment. Therefore, our results might be overestimated. Nonetheless, our findings are comparable to the investigations previously mentioned—many of which consider the beginning of treatment as their starting point of analysis. In addition, data from the ESNPCT maintains a certain level of objectivity and uniformity as it bases its clinical records on clinical forms and notifications. Fourth, there is a possibility of misclassification of MDR-TB into the drug-susceptible group as MDR-TB is widely under-diagnosed. Nevertheless, our findings show a high proportion of MDR-TB cases (about 20%), greater than previous studies using cultures to detect resistant cases [11] and even greater than the official ESNPCT surveillance report [5]. Finally, the data concerning smoking, alcohol and drug use was collected as a self-report notification. No clinical parameters, questionnaires or scales were applied to properly define the use of these substances. Due to its small amount, we believe that the risk of misclassification is non-differential, thus allowing the report of a true association.