Study design
This was a retrospective secondary analysis of routine standard National Leprosy and Tuberculosis and Lung Disease (NTLD) register data. The outcome was TB treatment outcomes categorized into successful outcome (cured or completed six months of treatment) or poor outcome (lost-to-follow-up, death, transferred out, treatment failure or development of drug resistance). The exposures examined were demographic (age, sex), sub-county of resident, year of starting TB treatment, nutritional status (body mass index), nutritional support provided and clinical features (HIV status, underlying comorbidities, type of TB (pulmonary or extra-pulmonary), TB diagnosis (bacteriological confirmed TB or empirically treated), treatment regimen and direct observed treatment.
Setting
TB Electronic surveillance data was collected from health facilities in seven sub-counties including Kilifi North, Kilifi South, Malindi, Magarini, Kaloleni, Rabai, and Ganze in Kilifi County within the coast region of Kenya. The county had a population of 1.4 million people in 2019 census[11]. More than 70% of Kilifi county residents live in rural areas, are majorly poor, lacks formal education, and makes a living from subsistence farming and fishing [12]. The estimated prevalence of TB is 122/100,000 cases according to the national survey while that of HIV is 5% [13, 14]. In Kilifi County, there were three health facilities with GeneXpert machines with the capacity to diagnose TB during the period of this study. Not all rural health facilities have laboratory services to run sputum smear test for TB as the golden standard of TB diagnosis, nonetheless, facilities leverage on the existing sputum sample referral to the high-volume health facilities for sample examination.
Participants
The study population was all adult TB patients (≥18 years) who were on anti-TB treatment from January 2012 to December 2019 within the Kilifi County.
Variables
Pulmonary TB was defined as any bacteriologically confirmed or clinically diagnosed case of TB involving the lung parenchyma or the tracheobronchial tree. Extra-pulmonary TB were any bacteriologically confirmed or clinically diagnosed case of TB involving organs other than the lungs. A patient was classified as transferred out if the treatment outcome was not known as a result of moving away from Kilifi County. Patients were classified as having poor treatment outcomes if they i) failed treatment (i.e., remaining smear-positive after 5 months of treatment), ii) had defaulted, iii) died during treatment or iv) transferred out. Cured patients were those with pulmonary TB and bacteriologically confirmed at the beginning of treatment but had smear- or culture-negative test in the last one previous occasion. Deaths included all-cause mortality within the six months of follow-ups. TB patient who completed treatment without evidence of failure BUT with no record to show that sputum smear or culture results in the last month of treatment and on at least one previous occasion were negative, either because tests were not done or because results are unavailable were classified as having completed treatment. A TB patient whose sputum smear or culture was positive at month 5 or later during treatment was defined as treatment failure. A TB patient who did not start treatment or whose treatment was interrupted for 2 consecutive months or more was defined as lost-to-follow-up while those who had initiated treatment but defaulted from treatment before completing the regimen were the defaulters. New TB cases were patients newly registered who had never been treated for TB before or had been on anti-TB treatment less than 4 weeks. Retreated patients were patients who had been treated for any form of TB before but had initiated treatment again following relapse or default or failure to cure of the 1st regimen.
Data sources/measurements
Data were extracted from the TB Electronic surveillance system known as Treatment Information from Basic Unit (TIBU). This system stores individual patient episodes of TB including demographic characteristics, location, clinical details, laboratory results, and treatment outcomes [15]. De-identified data were extracted directly into a Microsoft Excel spreadsheet that was designed to capture the relevant variables. Data extractions were done by the researchers in the presence of the County TB Coordinator.
Study size
The study used all available eligible patient data from 2012 to 2019. A total of 14,706 patients were eligible. Assuming 14% probability of a poor outcome[16], a two-sided alpha level of 0.05, the study has power >90% to estimate a crude hazard ratio of at least 1.5 of HIV positive being associated with poor treatment outcome in the first three months of follow-up [16].
Quantitative variables
Bacteriological confirmed TB were patients with positive smear microscopy, culture or GeneXpert MTB/RIF. Empirically treated patients did not have any positive TB bacteriological test but had clinical signs suggestive of TB including abnormal chest radiograph, chronic cough, fever, night sweats, weight loss, suggestive histology or extrapulmonary cases.
We created four age groups:18 to 30, 31 to 40, 41 to 50 and 51+ years. Body Mass Index (BMI) was calculated as weight (Kg) divided by square of height (meters) and further recorded into three groups according to WHO guidelines: undernourished (BMI<18.5), normal (BMI 18.5 to 25) and overweight (BMI ≥25) [17].
We assumed the missing BMI and HIV status were not missing at random. To include all patients in the regression analysis, we added extra category (missing for BMI and unknown for HIV) and used the categorical variables in the analysis.
Statistical methods
All study patients’ characteristics were summarised using frequencies and percentages. We calculated the annual proportion of poor outcome and tested for trend across the years (from 2012 to 2019)[18].
To examine factors associated with poor treatment outcomes, we run single event survival analysis with time under observation starting from date of starting TB treatment up to 180 days later or date of any of the outcomes. All patients who completed treatment or were under treatment after 180 days were right censored at day 180. All other patients who did not complete treatment and experienced any of the poor outcomes were right censored at their last date seen alive or last follow-up. We tested the presence of heterogeneity across the seven sub-counties using likelihood ratio test in the final regression model. We found evidence for presence of sub-county heterogeneity (P<0.001) and included the sub-county as random intercept in all the survival regression models using the shared frailty models [19]. We tested the Proportional-hazards assumption using the scaled Schoenfeld residuals in each independent variable and in the multivariable cox proportional hazard model with all the independent variables. Because of the violation of the PH assumption (P<0.05), we performed time-stratified survival regression analyses. We chose to stratify the analysis at month three follow-up because this was the halfway of the follow-up time and from operational perspective, it would inform interventions targeting the early poor outcomes. However, we provided survival analyses results for the first three months and last three months of follow-up separately. We tested proportional-hazards assumption for the two time points and found no evidence of violation (the scaled Schoenfeld residuals global test for the first 3 months was P= 0.0936 and the last 3 months was P= 0.0656). We therefore used the Cox Proportional hazard regression model, running univariate model for each independent variable. To build the multivariable regression models, we used a backward stepwise approach retaining independent variables with a P<0.1 and reported their hazard ratios and 95% confidence intervals. We assessed predictive values of the multivariable models using area under receiver operating curves (AUCs).
In sub-analysis, we repeated the multivariable regression models amongst bacteriological confirmed TB cases only and explored interaction between the year of starting TB treatment and various independent variables considered (age, gender, HIV status, TB diagnosis, underlying comorbidities, patient type) by comparing models with and without interaction terms using likelihood ratio test. The statistical analyses were performed using STATA version 15.1 (StataCorp, College Station, TX, USA).