This is a prospective cohort study, which recruited adult patients who started treatment for TB at three referral centers in the city of Juiz de Fora, Minas Gerais, Brazil, which comprised more than 50% of all treatments performed in this municipality according to the database of the Brazilian Notifiable Diseases Information System (NDIS).
Two follow-ups were performed: the first, with patients who started treatment from November 2007 to March 2010 (RH-FDC + Z), and the second from July 2018 to November 2019 (4-FDC). The cases were monitored from the first day of medication use until the end of treatment.
Study setting and population
Juiz de Fora has the highest incidence of TB in the state of Minas Gerais, reaching 41.8 cases per 100 000 inhabitants in 2018, being considered a priority for TB prevention and care in Brazil. Primary health care diagnosed less than a third of TB cases in the same year in the municipality, while the three centers involved in this study, secondary and tertiary care providers, remained the main treatment units. The Directly Observed Treatment Short-course (DOTS) strategy, recommended worldwide to strengthen adherence to TB treatment, covered only 9.5% of cases .
Centers 1 and 2 are of secondary complexity and attend general TB patients and HIV/aids patients, respectively; center 3 is a hospital unit.
This research included only new TB cases to avoid including patients with longer treatment regimens, and to better compare the two cohorts focusing on only changing in the first-line drugs for TB treatment. Other inclusion criteria were patients aged 18 years or older, with a diagnosis confirmed by bacteriological tests (smear or positive culture), molecular biology (rapid molecular test for positive TB) or histopathology suggestive of TB (presence of granuloma with caseous necrosis).
Patients were provided with adherence counselling and TB health education by health care workers at the start of TB treatment and on recurrent visits.
The data were collected through a semi-structured questionnaire for interviewing patients and a form to verify additional information in medical records and in the NDIS. The data collection instrument used was previously validated in a Brazilian study .
The patients were approached at the treatment centers and invited to participate in the research, by signing the informed consent form. In the interview, sociodemographic, economic, behavioral, and clinical information was collected.
The outcome of interest was defined as the time of occurrence of LTFU from TB treatment, as classified by the Ministry of Health of Brazil: “a patient who used medication for 30 days or more and interrupted treatment for 30 consecutive days or more”, or “a patient who used medication for less than 30 days and interrupted 30 consecutive days or more, or when the diagnosed patient does not start treatment” .
The time variable was defined in days between the treatment start date and the occurrence of LTFU, or the cure (censorship).
The explanatory variables were grouped into three hierarchical levels:
- distal (socio-demographic and economic variables).
- middle (health care variables).
- proximal (individual and behavioral variables).
While in the treatment period 1, RH-FDC + Z was the only basic regimen used for new TB cases; in the period 2, they used exclusively the 4-FDC one.
The instruments to assess illicit drug consumption were already tested for validity and reliability in previous Brazilian studies . The Cut down, Annoyed, Guilty, Eye-opener (CAGE) questionnaire was used to screen persons at high risk of alcoholism, considering two affirmative responses or more as a high risk . Tobacco addiction was considered to be the consumption of 10 or more cigarettes per day.
Descriptive epidemiology was done, and the chi-square test was used to compare characteristics of patients included and excluded who started TB treatment in both periods.
Survival analysis techniques were used to identify the variables associated with the time of occurrence of LTFU from TB treatment, especially the two treatment periods, which bring intrinsic each regime used. The magnitude of association was determined by the Hazard Ratio (HR).
Survival analysis considered LTFU and cure. The other outcomes, such as patients who failed or died during the study period were excluded from the analysis, because these explanatory variables could be associated with both LTFU and death or other negative outcomes. Thus, the use of the participation times of individuals censored with negative outcomes could bias resulting towards the null, as previously stated [12, 13].
The survival curves were estimated using the Kaplan-Meier method. Then, a comparison between the patients treated in the two treatment periods was done using the Log-rank statistical test. These survival curves were analyzed using the total of patients treated per period and stratifying them by treatment center.
Cox's proportional hazards model was used for univariate and multivariate analyses. All variables associated with the time of LTFU with a p-value ≤ 0.20 in the univariate analysis were included in the multiple regression model, performed hierarchically from the distal to the proximal level. The variables with significance p ≤ 0.05 remained in the final model.
The assumption of proportional hazards for the final model was verified by graphic inspection and adjustment test by Schoenfeld residuals. This adjustment was considered adequate if p > 0.05.
Also, we compared by chi square whether each treatment period differed from the other with respect to the variables sex, skin color, age, education, dwelling type, institutionalization, place of residence, family income, alcoholism, illicit drugs, hospitalization during TB treatment and treatment center. Complementary, the variables considered different (p ≤ 0.05) were added to the final multivariate model to assess whether even in an over-saturated model, adjusted for all possible confounding variables, the treatment period variable would still prevail as a significant predictor of LTFU, our main hypothesis tested in this study.
The IBM Statistical Package for the Social Sciences 21 software (IBM Corp., Armonk, NY, USA) was used for survival analysis and R 3.0.2 (R Foundation for Statistical Computing Platform, Vienna, Austria) for the evaluation of the goodness of fit of the multivariate model.