To the best of our knowledge, this is the first study to identify groups of weight change and to determine factors associated with these groups. Furthermore, these data also suggest that the management of HIV infection and depression status, as well as more therapeutic education to improve treatment adherence may reduce the risk of community transmission from patients with MDR-TB. In addition, the results provide more information to help with patient selection and stratification for the design of future interventional clinical trials.
The mechanism underlying weight loss in patients with MDR-TB is well known . Poverty-induced malnutrition is one of the main causes of weight loss in countries with a high prevalence of TB, such as Guinea. By decreasing the concentration of immunoglobulins, interleukin-2 receptor, and T-cell subset (helper, suppressor-cytotoxic, and natural killer cells), malnutrition further alters the immunity of patients with TB, making them vulnerable to infections such as HIV, and prone to severe clinical presentation and a higher proportion of positive sputum cultures . In addition, socioeconomic status, including the number of household contacts, may increase the risk of the MDR-TB infection. The report of a study conducted in Guinea between 1 January 2017 and 30 September 2018 showed that of 4,255 people who underwent the GeneXpert MDR/RIF test, 339 (8%) were identified as household contacts, and 105 (31%) of them were positive for TB (17 MDR-TB and 88 TB sensitive) (data not shown). This prevalence is probably underestimated because only the symptomatic household contacts are depicted. A similar result was reported in China where the positive rate of household contacts was 28% . Furthermore, others risk factors for MDR-TB were reported; they were social determinants of health (monthly low income of the family [< 100 €], stigma, unemployment, prison homelessness, alcoholism and substance abuse), health system weakness (poor organization of TB program, absence or inappropriate clinical guidelines), mental health factors (subjective feeling of sadness, use of sedatives), and clinical factors (history of prior TB treatment, HIV infection, chronic obstructive pulmonary, lung cavitation, and larger burden of bacilli on sputum microscopy) [16–18].
Two profiles of BMI increase were identified: rapid and slow, with the average probability of belonging to the two LCM models being higher, ranging from 0.82 to 0.99, suggesting unambiguous classification (appendix). From a LM model, we found that the BMI increase over time differed according to the treatment outcome. After controlling for lung cavities on X-ray, patients who were cured had gained on average 2.62 kg/m2 in BMI at the end of treatment. A previous study reported that patients who were cured had gained on average 3.9 kg at the end of the sixth month . Unlike the LM model, which shows an average gain BMI over time, our analysis showed that the speed of this weight gain was not identical for all patients. The most interesting finding was that the patients in the slow BMI increase group had a poor response to the MDR-TB treatment, suggesting that weight may serve as a potential biomarker to monitor treatment outcome. These patients were characterized by a positive HIV infection, depression symptoms, poor adherence to the MDR-TB treatment, and delay to the culture conversion. This is a relevant finding in public health, particularly in resource-limited settings because it allows better targeting of patients with a high risk of treatment failure and hence better channeling of the resources needed to improve treatment success rates. Strategies such as close monitoring of these patients, therapeutic education to improve treatment adherence, and the setting up of psychiatric consultations to manage depression will help improve the prognosis of these patients and increase their chance of success.
Furthermore, in patients with slow BMI increase, the likelihood of culture conversion was reduced by 65% (HR = 0.35, 95% CI [0.13 – 0.96]; p = 0.0087). This finding was higher than those reported from studies evaluating the impact of baseline weight loss (BMI < 18.5 kg/m2) and delay in culture conversion [4,5]. The reduced chances of culture conversion were 43% and 45% for Indonesian and South Korean patients, respectively. In addition, 89% of our patients showed culture conversion in a median of 2 months, which was higher than the rates of culture conversion reported in Indonesia (80%)  and South Korea (70%) , suggesting that our MDR-TB treatment program performed reasonably well.
Recently, the superiority of culture conversion over smear conversion in predicting MDR-TB treatment outcomes was demonstrated, with an optimum time point between four and six months after treatment commencement. This conclusion supports the WHO recommendation to add culture examination to the sputum smear for the monitoring of MDR-TB patients for better prediction of successful treatment outcomes . Nevertheless, in resource-constrained settings, the sputum culture is resource-intensive, takes time to obtain, is costly, and requires specialized laboratories, equipment and trained staff. We found that in patients with MDR-TB, a stable or decreased weight between two visits is probably a sign of a poor response to treatment, especially in an HIV-infected, depressed woman with lung cavities on X-ray whose treatment adherence was poor. Since the measurement of body weight is easy, rapid, inexpensive, and accessible everywhere, the association between a faster increase in BMI and shorter time to initial culture conversion suggests that weight measurement is a useful surrogate of culture conversion in predicting an early MDR-TB treatment response.
Our study has a number of strengths. First, patients from three referral centers for MDR-TB management in Guinea were evaluated, which reduces selection bias and increases the validity of the extrapolation of our findings to the entire population of Guinean patients with MDR-TB. Second, unlike the conventional mixed linear model used to describe the change in weight over time, our analysis identified a group of patients with poor prognosis (slow BMI increase) as well as the characteristics of these patients. Third, we used a compromise criterion to select the best BMI change groups instead of using only the Bayesian information criterion. As mentioned above, the model with two classes has a higher average posterior probability of up 0.80, suggesting unambiguous classification. Fourth, to account for informative dropout, we applied a sensitivity analysis using a joint model for longitudinal and time to dropout . The results obtained from this joint model were similar to estimations using the standard LM model, suggesting an absence of bias in parameter estimations (data not shown). However, the limits of our study were its retrospective design and small sample size, which had an impact on the internal validity of the study, some missing factors such as diabetes status and other metabolic factors, smoking and alcohol use, information on second-line drug susceptibility, and other anthropometric measurements, such as skin-fold thickness and waist circumference, which could possibly serve as a proxy for weight assessment. Further prospective cohort studies with patient numbers are needed to confirm our findings.