Increasing multidrug-resistant tuberculosis in Ho Chi Minh City: a retrospective study of 2,266 cases from 2011 to 2015

Background: Multidrug resistant tuberculosis (MDR-TB) remains a serious public health problem with poor treatment outcome. Predictors of poor outcomes vary in different regions. Vietnam is among the 30 countries with high burden of MDR-TB. We aim to describe demographic characteristics and identify risk factors for poor outcome of MDR-TB in Ho Chi Minh City (HCMC), the most populous city in Vietnam. Methods: This retrospective study included 2,266 patients who initiated MDR-TB treatment from 2011 to 2015 in HCMC. Treatment outcomes were available in 2,240 patients. Data was collected from standardized paper-based treatment cards and electronic records. Kruskal Wallis test was used to diagnose the change of median of age and body mass index (BMI) over 5 years, and Wilcoxon test to compare median BMI of patients with and without diabetes mellitus. Chi squared test was used to compare categorical variables. Multivariate logistic regression on multiple imputation was used to identify risk factors for poor outcomes. Statistical analysis was performed using R program. Results: Among 2,266 eligible cases, 60.2% were failure of category I or II regimen, 57.7% were underweight, 30.2% had diabetes mellitus and 9.6% were HIV positive. Notification rate increased 24.7% from 2011 to 2015.Treatment success rate was 73.3%. Risk factors for poor treatment outcome included HIV co-infection (adjusted odds ratio (aOR): 2.94), advanced age (aOR: 1.45 for every increase of 5 years for patients 60 years or older), having history of MDR-TB treatment (aOR: 5.53), sputum smear grade scanty and 1+ (aOR: 1.47), smear grade 2+ or 3+ (aOR: 2.06), low BMI (aOR: 0.83 for every increase of 1kg/m2 of BMI for patients with BMI<21). Conclusion: Our study describes the increasing cases of MDR-TB in HCMC during

previous MDR-TB treatment should receive additional care.
Keywords: multidrug resistant tuberculosis; retrospective; treatment outcome; risk factors; Vietnam Background Multidrug resistant tuberculosis (MDR-TB), defined as tuberculosis (TB) with resistance to at least rifampicin and isoniazid, is a serious public health problem. In 2017, there were an estimated 558,000 incident cases and 230,000 deaths due to MDR/Rifampicin resistant (RR)-TB worldwide. Treatment of MDR-TB is lengthy, toxic and expensive, with success rates reported at 55% in 2017 (1). HIV co-infection, low body mass index (BMI) and positive sputum smear are predictors of poor MDR-TB outcomes, but the effects of these predictors may vary in different regions (2) (3) (4) (5) (6).
Vietnam is among high TB and MDR-TB burden countries. Its estimated incidence of TB in 2017 was 129 per 100,000 people. A national survey in 2011 showed resistance to any drug was 32.7% in new TB patients and 54.2% in previously treated patients (7). The prevalence of MDR-TB was 4.1% in new patients and 17% in previously treated patients with 4900 estimated new cases annually countrywide (1). Ho Chi Minh City (HCMC) is the most populous city in Vietnam (around 8 million people) and also the center for TB and drug-resistant TB management in Southern

Study setting and population
Patients were hospitalized in PNTH for 7 to 14 days to initiate treatment, and then referred to DTUs for outpatient follow-up. Treatment modalities did not change during the study time, with the standardized combination of 6 drugs for a total of 18 to 24 months of treatment (2).
Sputum samples from suspected MDR-TB patients or from patients with MDR/RR-TB detected by XpertMTB/RIF or line probe assay (LPA) were sent to the laboratory in PNTH for confirming MDR-TB by drug susceptibility testing (DST). Culture and DST in solid and liquid media were the main initial diagnostic method prior 2012.
Pyrazinamide and ethambutol were components of standardized MDR-TB regimen, but detecting these drugs' resistance were challenging because of unreliable phenotypic DST results (9). GenoType MTBDRplus (Hain Lifescience GmbH, Germany -LPA) since 2010and XpertMTB/RIF (Cepheid, USA) since 2012 were used to detect We included all patients who initiated treatment with second-line drugs in HCMC under PMDT from January 2011 to December 2015. Patients who started their MDR-TB treatment after 2015 might not have treatment outcomes available at the time of data collection. We then excluded the patients who 1) had evidence of irrelevant diagnosed MDR either by DST, Xpert or LPA test, or 2) were enrolled in the STREAM trial (10) to receive a 9-month regimen, or 3) did not start treatment ( Figure 1).

Data collection
Demographic and clinical information, radiographs, acid-fast bacilli (AFB) staining, DST results, treatment regimens and treatment outcomes were recorded into structured paper forms. To improve reliability, we collected data from both standardized paper-based treatment cards and electronic records and verified data during the data collection, entry and analysis processes.

Statistical analysis
Data analysis was performed using R program version 3.5.2 (11). The baseline characteristics were summazied as number of cases and percentage for categorical variables and median with interquartile range (IQR) for continuous variables. We used Kruskal Wallis test to identify the change of median of age and BMI over 5 years, and Wilcoxon test to compare median BMI of diabetes mellitus (DM) and non-DM patients. Chi squared test was used to compare categorical variables.
To identify the risk factors which contribute to the poor treatment outcome, we performed a multivariate logistic regression model. The outcome is a binary treatment outcome variable with "success" (cured, completed) and "non-success" (death, failure, lost to follow-up) as defined in (12). All the risk factors (HIV co-infection, history of previous MDR-TB treatment, AFB smear grade and BMI) were included as covariates in the model. The model also adjusted for potential confounders (gender, age and DM status). In the model, we model covariate age with and BMI with a piecewise linear form. In particular, for covariate age, we used index variable (age ≤ 60 years old) and linear pattern for age greater than 60 years old. Similarly, for BMI covariate, we used index variable (BMI ≥ 21kg/m 2 ) and a linear pattern for BMI less than 21kg/m 2 . To minimize bias caused by missing data, we imputed them by using multiple imputation by chained equation ("mice" package in R (13)) and we performed multivariate logistic regression models using both imputed data analysis and complete case analysis. There were only small differences in the results between the two analyses. Therefore, we presented the Forrest plot of results from imputed data analysis and provided results of both imputed and non-imputed analysis.

Results
Characteristics of MDR/RR-TB patients From 2,395 electronic records and 1,913 paper-based records available, 2,266 MDR/RR-TB cases were included ( Figure 1). Twenty-three patients had two episodes of treatment during the study time, including eight relapse cases, 13 retreatment after lost to follow-up cases and two retreatment after failure cases. Baseline characteristics are presented in Table 1. Of 2,266 cases, the median age was 43 years (IQR: 33-53 years) and did not change during 2011 to 2015 (p = 0.481). A total of 204 patients (9.6% of tested patients) were HIV co-infected; of these, 33 (16.1%) were registered as new patients, 21 (10.3%) had extra-pulmonary MDR-TB including 10 MDR-TB meningitis cases (4.9%), which was higher than in non-HIV patients (p<0.001). Eighty four of 204 HIV co-infected patients (41.2%) already received antiretroviral therapy before the start of MDR-TB treatment. The remaining HIV co-infected patients (58.8%) started ART after at least two weeks from the start of MDR-TB treatment. Among 1,815 cases with BMI information available, 57.8% was classified as underweight with 25.1% severe underweight. Median of BMI did not differ during 5 years (p = 0.966). DM status was available for 1,189 patients (52.5%), 359 of whom (30.2%) had DM. Median BMI of DM patients (20.0 kg/m 2 ) was higher than non-DM patients (17.8 kg/m 2 ) (p<0.001) and HIV co-infection in DM patients (0.9%) was lower than in non-DM patients (9.8%) (p<0.001).   Table 3 summarizes the treatment outcomes of 2,240 MDR-TB patients whose treatment outcomes were retrievable. Successful outcomes were achieved in 1,641 (73.3 %) patients, including 55.6% cured and 17.7% completed. Among those with unsuccessful outcomes, 10.1% died, 5% failed treatment and 11.6% lost to follow- In this RR/MDR-TB cohort, almost all strains were resistant to streptomycin (96.3%), comparable to previous studies in Vietnam (14) (15) (16). This could be explained by the fact that the streptomycin was widely used in regimen 1 and 2 during the intensive phase for both new and retreated patients.

Treatment outcomes
We found high rates of resistance to pyrazinamide (55.0%) and ethambutol (63.1%) in our MDR-TB cohort, as reported by other studies (17) (18). This may reflect the fact that the majority of MDR-TB patients (94.3%) already had exposure to first line anti-TB drugs and might have developed resistance to pyrazinamide and ethambutol during previous treatment. This questions the effectiveness of empirical use of these two drugs in the standardized MDR-TB regimen (19) and emphasizes the need for an approved genotypic DST to rapidly detect pyrazinamide resistance.
Resistance rates to fluoroquinolones (12.7%) and injectable agents (8.1%) were comparable to those of the survey in Vietnam in 2011 (20)

but lower than in South
Korea (17) and average global rates (1). Resistance to second-line drugs were high although they are not used in the regimen 1 and 2. It might partly due to easy access to antibiotics without prescription in Vietnam (21).
HIV co-infection, positive baseline AFB smear, older age and previous treatment with second-line drugs are main risk factors for poor treatment outcomes in our cohort, which were also observed in cohorts in Estonia, Latvia, Philippines, Russia, Peru (4), Ukraine (3). Malnutrition was common (57.8%) and also a risk factor for poor outcome (OR: 0.81 for every 1kg/m 2 increase of BMI). Low BMI might be a consequence of severe disease and low social-economic status, which are wellknown risk factors for poor outcome of TB. PMDT should focus on nutrition support to improve treatment outcomes.
The prevalence of DM in our cohort (30.2%) was double that of other TB patients (13.7%) in Hanoi, Vietnam (22) and was almost 6 times higher that of general Vietnamese population in 2013 (5.4%) (23). Although DM is a known risk factor for poor treatment outcome of TB, developing MDR-TB and reducing sputum conversion rate during MDR-TB treatment (24) (25) (26), whether DM also leads to poor treatment outcome of MDR-TB is still controversial (24) (27) (28). After adjusted for other factors, DM was not an independent risk for poor outcomes in our cohort, which agrees with pooled data analysis from cohorts in Latvia, Korea and Italy (29).
Due to the unavailability of DM treatment information, we do not know whether the effect of DM on MDR-TB treatment was influenced by the use of metformin, a hypoglycemic agent that might improve TB treatment outcomes (30) (31).
Furthermore, levels of glycemic control of MDR-TB patients with DM were unable to retrieve to access its effect on treatment outcome. Despite these limitations, DM is a common but neglected comorbidity in MDR-TB patients and should be screened for prior MDR-TB treatment.
This study has several limitations. This is a retrospective study and some records were irretrievable at the study time. Demographic information and records of smear, culture and DST was not completely recorded on the electronic database.
The majority of patients (78.4%) did not have DST results, and we could not include drug resistance information into multivariate logistic regression models. Finally, the information on smoking and alcohol use were not available in our cohort, although they are known risk factors for poor outcome (32) (33). Therefore, a prospective study is necessary to provide a comprehensive assessment of risk factors for poor treatment outcome of MDR-TB. Individual consent to participate was waived by the IRB because this is a retrospective collection and data were recorded and analyzed anonymously.

Consent to publish
Not applicable.

Availability of data and materials
The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Competing interests
All authors declare no competing interests.