Development of a Prognosis Nomogram of Treatment Outcomes for MDR-Tuberculosis in Guinea (Conakry): A retrospective cohort analysis

Abstract Background: Despite the availability of the drug treatment for tuberculosis (TB) more than 75 years, mortality and drug resistance are increasing. Therefore, little data is available in Guinea. We aimed to develop and validate a prognosis nomogram of MDR-TB treatment outcomes. Methods:A retrospective cohort study was conducted among men and women, aged 18 years or older, with MDR-TB, from three major drug-resistance TB centres in Guinea. We used the logistic regression to analyse treatment outcomes. Prognostic factors with a p value less than 0.05 from a multivariate model were used to build nomogram and assessed their performance based on discriminative c-index, and calibration using the Hosmer-Lemeshow (H-L) test. To derive the optimal cut-off point score, the Youden’s index method was used. Results:Among 232 patients with MDR-TBenrolled and followed between June 07, 2016 and June 22, 2018, 218 were analyzed. All patients were resistant to rifampicin, which diagnosed by the Xpert MTB/RIF. The overall rate of success was 73%.Factors associated with successful treatment in drug-resistant TB patients were higher BMI more than 18.5 kg/m2(p = 0.0253; aOR = 2.94), good adherence to treatment (p = < 0.0001; aOR = 33.92), normal platelets count (p = 0.0053; OR = 1.004), and the absence of clinical symptoms such as chest pain (p = 0.0083; aOR = 3.19) and depression (p = 0.0308; aOR = 8.62). The discrimination (c-index= 0.848 95% bootstrap CI, 0.780 – 0.916 in the derivation sample and 0.803 after correction for optimism) and calibration (H-LX2= 2.91 p = 0.94) were good. The optimal absolute risk threshold was 20%, corresponding to a sensibility of 95% and specificity of 58%. Health Organization (75%). We recommend to improve the MDR-TB patient monitoring during treatment, nutritional status, and considering the psychological state. Our prognosis nomogram needs to be validated in an external population before it can be used in clinical practice.

treated with a standardized shorted treatment regimen of 9-months in Guinea. We aimed to develop and validate a prognosis nomogram of MDR-TB treatment outcomes during follow-up.

Study design and population
We retrospectively analyzed 218 patients with MDR-TB enrolled between June 07, 2016 and June 22, 2018 in the multicenter longitudinal cohort study conducted in three major drug-resistance TB centres in Guinea (Ignace-Deen, Carrière and Tombolia). For MDR-TB diagnosis, both sputum smear and culture were performed in these centres. Patients were seen at baseline and followed by monthly visit for 9 months according to the WHO standardized 9 -12 months shorter treatment regimen guidelines recommendations [13].
Patients who were younger than 18 years were excluded from the analysis. Ethical approval was obtained from the Guinean Ethics Committee for Heath Research.

Bacteriology, and drug susceptibility testing
The diagnosis of drug-resistance was done by the Xpert MTB/RIF test. For others anti-TB drugs, the drug susceptibility testing (DST), based on solid culture (Lowenstein-Jensen) were done late, and in a very partial way. Second-line anti-TB drugs were not routinely tested. Xpert MTB/RIF test and DST were only available at the main TB centre located in Ignace Deen.

Treatment regimens for MDR-TB in Guinea,
According to the guideline for the MDR management in Guinea [3], naive patients for second-line anti-TB drugs were treated with a short 9-month regimen, consisting of an intensive phase lasting a minimum of 4 months including moxifloxacin, kanamycin, clofazimine, prothionamide, pyrazinamide, ethambutol, and INH at high dose. The intensive phase was then followed by the continuation phase during 5 months and consisting of administration of four drugs: moxifloxacin, clofazimine, pyrazinamide, and ethambutol. Sputum smears and cultures were obtained monthly during the MDR-TB treatment duration.

Data collection
Data was collected using a case report form (CRF) from the MDR-TB registry that containing sociodemographic, clinical data, and the laboratory test results (sputum smear or culture conversion) for all patients who were admitted in the three major TB centres outpatient care. Additional information was completed with data from the patients' clinical files. The following clinical and demographic data record were extracted: age, gender, residence, comorbidity, HIV status, history of the previously treated TB, the presence of cavities on chest X-ray determined by the senior radiologist, baseline data on weight, sputum smear and culture, clinical symptoms (as chest pain or cough), and biological data. Additionally, we extracted depression status where the patient was asked if he/she was depressed or anxious, and adherence to MDR-TB treatment status during follow-up based on the proportion of days covered (PDC). For each patient, we calculated a PDC by dividing the number of days covered with MDR-TB treatment delivered over one month by 31. Then, we considered a minimal value of PDC during follow-up for each patient as a marker of adherence to MDR-TB treatment. Conventionally, the PDC was dichotomized between good adherence if the PDC was 0.8 or more and poor adherence otherwise [14].

Treatment outcomes
Using the revised WHO recommendations in 2013 [15], we classified the treatment outcomes into two categories (successful or unsuccessful). Successful treatment outcomes corresponded to patient who declared as either "cured" or "treatment completed".
Unsuccessful treatment outcomes included "treatment failure", or "death form any reason" or "lost to follow-up" or "not evaluated".

Statistical analysis
Frequencies (percentage) or means (standard deviation; SD) were used to describe categorical and continuous variables at baseline. Univariate logistic regression was used to identify prognostic factors associated with successful MDR-TB treatment, and then candidates with a p-value less than 0.10 were entered into the multivariate logistic regression. The independent predictors of successful MDR-TB treatment from multivariate regression were selected through a backward procedure based on the lowest Akaike information criterion. Odds ratio (OR) together with their 95% confidence interval (CI) were used as association measures.
We constructed a nomogram that included the selected prognostic factors from final logistic model to estimate the probability of successful MDR-TB treatment at 9-months. A raw prognostic score was computed by summing the contribution of each individual factor, based on the estimate for each factor in the nomogram. Then, we divided patients into four prognostic chance groups: "low chance", "moderate chance", "intermediate chance", and "high chance". We did calibration plots with the Hosmer and Lemeshow statistic test [16], and computed the discriminative c-index to assess the performance of the nomogram. 1,000 random samples of the population were used to drive the 95% CI bootstrap percentile for the c-index, and to correct the optimism. We used the receiveroperating characteristic (ROC), applying the Youden's index method [17], to derive the optimal cut-off point score. Then, at this optimal threshold, the performance measures including the sensitivity, specificity, and positive and negative predictive values (PV) were estimated. The final model was internally validated using the 1,000 samples bootstrap procedure. All data analyses were done in R (version 3.5.1). Significance was defined as a p-value less than 0.05, and all tests were two-sided.

Socio-demographic and clinical characteristics at baseline
A total of 218 patients with MDR-TB who meet the inclusion criterion were analyzed.  (Table 1).

Predictors of successful treatment outcomes
In univariate logistic regression (Table 2), predictors associated with successful treatment outcomes were younger patients (p = 0.0441; OR = 1.03), higher BMI more than 18.5 kg/m 2 (p = 0.0386; OR = 1.18), good adherence to treatment (p < 0.0001; OR = 19.52), and the absence of clinical symptoms such as dyspnea (p = 0.0006; OR = 2.97), chest pain (p = 0.0372; OR = 1.90), vomiting (p = 0.0139; OR = 3.08), and depression (p = 0.0060; OR = 7.0). In addition, for patients with negative HIV (p = 0.0552; OR = 1.93), and patients who resided in an urban area (p = 0.0591; OR = 2.22), we noted a trend association with the successful treatment outcomes. For the remaining predictors, we failed to show any association with the treatment success (p ³ 0.0600). The strongest contributing risk for higher successful treatment rates (Table 3) (Table 3), suggesting robustness of the final model.
The nomogram to predict the probability of 9-months successful treatment ( Figure 1) showed that the normal plaquettes count and good adherence to treatment contributed the most strongly to the prognosis. Whereas, the increased BMI more than 18.5 kg/m 2 , and the absence of chest pain and depression at baseline had little effect on the probability of successful treatment. A row score was computed from the nomogram, and   Table S1 in appendix.

Discussion
In this study, we identified predictors and derived prognosis nomogram of 9-months MDR-TB treatment success.
During the treatment period, the success rate was 73%, which was slightly below than the success rate recommended by the WHO (75%  [4,7,9], first episode of MDR-TB [4,9], cavitary lesions on X-ray [6,7], resistance to streptomycin and Ethambutol [6] had been reported as independent predictors to successful treatment. While, we not found any association between cavitary lesions and treatment outcomes. These discordant finding suggest the conduct of an international meta-analysis on individual data to definitively identify factors associated with successful treatment.
Based on the number of easily accessible predictors identified in the final logistic model, we constructed a nomogram in a rigorous methodology framework that allowed prediction of individual treatment success with high precision. The nomogram had good discriminatory capacity, and there was good agreement between the nomogram prediction and actual rates of success. Moreover, using the ROC curve methodology, the threshold of 135 points was identified to be the optimal cut-off to select patients with high chance of being successful treatment. However, the nomogram needs to be externally validated on independent samples including non-Guinean patients to establish the generalisability of the model. While, some missing parameters such as diabetes status, smoking and alcohol use, and biomarkers were limits of this study, they might be further used to improve nomogram.

Conclusion
The current study allowed to develop and internally validate a prognosis nomogram to estimate the chance of successful treatment in MDR-TB patients. This nomogram was developed from the easily accessible predictors including body weight, adherence to treatment, platelets count, chest pain, and depression status.

Ethics approval and consent to participate
The study was approved by the National Ethics Committee for Heath Research (NECHR) reattached to the Ministry of Health (Conakry, Guinea). The study was based in accordance to the Declaration of Helsinki, and the confidentiality of the data was guaranteed.      Discriminative curve to predict the probability of successful MDR-TB treatment outcomes.

Supplementary Files
This is a list of supplementary files associated with the primary manuscript. Click to download. Appendix.pdf