Nutritional Indices Predict All Cause Mortality in Patients with Multi-/Rifampicin-Drug Resistant Tuberculosis

Introduction

Tuberculosis (TB) is a global health concern, with high morbidity and mortality rates. Compared with drug-sensitive TB, multidrug- and rifampicin-resistant TB (MDR/RR-TB) has longer treatment time, higher treatment cost, poorer treatment outcomes and a higher risk of mortality, which is a serious obstacle to TB control.¹ The Global Tuberculosis Report 2023 by the World Health Organization (WHO) reported that estimated 410,000 new cases of MDR/RR-TB in 2022.² A previous study reported that the mortality of MDR/RR-TB patients reached 14.68%.³ Therefore, decreasing the mortality of MDR/RR-TB patients is gradually becoming a crucial problem in the treatment of anti-TB medicines for TB.

Malnutrition is strongly associated with the incidence, treatment, and prognosis of TB. Many studies have shown that malnutrition increases the risk of incidence of TB,^4,5 treatment failure⁶ and mortality.^7–11 On the contrary, a randomized controlled trial of Reducing the Activation of TB by Improving Nutritional Status (RATIONS) from India indicated that nutritional support can reduce the incidence of TB and reduce mortality.^12,13 In addition, some previous studies and meta-analyses have provided evidence that undernutrition increases the risk of unsuccessful treatment outcomes and mortality in patients with MDR-TB.^14–16 Hence, the assessment and management of nutritional status is greatly significant for the long-term prognosis of TB patients.

At present, many methods, including BMI,¹⁷ nutritional risk screening (NRS) 2002,¹⁸ geriatric nutritional risk index (GNIRI),⁴ hemoglobin-albumin-lymphocyte-platelet (HALP) score,¹⁹ prognostic nutritional index (PNI),⁴ and controlling nutritional status (CONUT)⁴ can assess the nutritional status. However, few studies have explored the correlation between nutritional status and mortality in patients with TB, let alone in MDR/RR-TB patients. Studies of 1075 sample cohorts revealed that the NRS 2002, COUNT score, GNIRI, and PNI were associated with all-cause mortality (ACM) in TB patients.⁴ A retrospective study of 93 TB patients suggested that the NRS 2002 and PNI could predict mortality in TB patients.²⁰ Additionally, previous studies have explored the risk factors related to nutritional status of poor treatment outcomes or mortality in MDR-TB patients, such as anemia,²¹ low BMI,²² and decreased albumin (ALB).²³

In conclusion, the predictive value of nutritional indices for mortality in MDR/RR-TB patients is unclear. Therefore, the purpose of the present study was to investigate the predictive value of nutritional indices, including the HALP score, PNI and CONUT score, for ACM in patients with MDR/RR-TB.

Materials and Methods

Patients and Study Design

This retrospective cohort study enrolled 728 participants diagnosed with MDR/RR-TB according to drug susceptibility testing (DST) or GeneXpert MTB/RIF at Wuhan Jinyintan Hospital, Tongji Medical College, Huazhong University of Science and Technology (Infectious Disease Hospital) between January 2015 and December 2021. Patients were treated according to the Treatment Guidelines for Drug Resistant TB.²⁴ The inclusion criteria were: (1) laboratory-confirmed diagnosis of TB with at least rifampicin resistance; (2) more than 18 years old; (3) had documented treatment outcome in the dataset. The exclusion criteria were:(1) death before treatment; (2) non-tuberculous mycobacteria infection; (3) refusal to undergo treatment.

Data Collection

Baseline data, including demographic characteristics (age and sex), history of treatment, smoking, drinking, co-infection (bacteria, fungi, HIV, hepatitis B virus, and Hepatitis C virus), underlying disease (diabetes mellitus, hypertension, and malignancy), Underlying Pulmonary disease (pulmonary heart disease and COPD), and laboratory parameters [hemoglobin (HGB) (g/L), ALB (g/L), lymphocyte count (10^9/L), platelet count (10^9/L), and total cholesterol (mmol/L), were obtained from the electronic medical record system. In addition, the computing methods for the HALP score, PNI, and CONUT score are shown in Table 1. The primary outcome of interest was mortality. The definition of censoring and overall survival time (OS) reference the previous literature.³ Data regarding follow-up outcomes were extracted from the national electronic case registry.

Table 1 The Computing Method of HALP Score, PNI, and COUNT

Grouping

All patients with MDR/RR-TB were randomly classified into training and validation cohorts to ensure study reproducibility. In addition, patients with MDR/RR-TB were classified into high nutritional risk groups (HNRGs) and low nutritional risk groups (LNRGs) according to the optimal cut-off values of the HALP score, PNI, and CONUT score, calculated using the receiver operating characteristic (ROC) curve.

Statistical Analysis

Statistical analyses were performed using Statistical Package for the SPSS (IL SPSS) version 22 and R software version 4.2.1. The numerical variables of normal distribution or approximate normal distribution are presented as mean ± standard deviation, and differences were compared using the Student’s t-test. Categorical variables are presented as n (%), and the chi-square test was used for comparisons between the groups. The differences in OS between the groups were compared using Kaplan-Meier survival curves and Log rank tests. Cox risk proportion regression analyses were used to explore the risk factors of ACM in patients with MDR/RR-TB. Statistical significance was set at P<0.05. Receiver operating characteristic (ROC) curves were used to calculate the optimal cutoff values of nutritional indices (HALP score, PNI and CONUT scores) and to evaluate the predictive performance of ACM in patients with MDR/RR-TB. In addition, we also operated the ROC curves for single parameters such as ALB HGB and lymphocyte count to further evaluate the predictive value of the three nutritional indices. DeLong’s test was used to compare the area under the curve (AUCs).

Result

Clinical Characteristics and Nutrition Indices

A total of 524 patients with MDR/RR-TB diagnosed and treated at the Wuhan Jinyintan Hospital were randomly classified into training (n=255) and validation (n=269) cohorts. The median follow-up time of the cohort was 2 years (24.30 months; IQR, 13.42–24.47). The age was 42.66±15.06 (rang, 18–82 years). In addition, 79 (15.08%) patients with MDR/RR-TB died during the follow-up period. As shown in Supplementary Table 1, although there were significant differences in bronchiectasis and pulmonary cavitation (P<0.05), the remaining characteristics showed no significant differences between the training and validation cohorts (P>0.05), indicating that the two cohorts were independent of each other and that the data from the two cohorts could be used for mutual validation.

In the training cohort, Tables 2–4 showed that the clinical characteristics of patients with MDR/RR-TB in the HNRGs and LNRGs according to the optimal cut-off values of the HALP, PNI, and CONUT scores. The accuracy of the cutoff values, sensitivity, specificity, and Youden index of nutritional indices in the training cohort are shown in Supplementary Table 2. MDR/RR-TB Patients in the LNRGs, including those with a HALP score>16.85, PNI>38.7, and COUNT score<6.5, had higher OS, lower mortality, and lower prevalence of fungal co-infection, pulmonary heart disease, bronchiectasis, destroyed lung, pulmonary cavitation, and pleural effusion than those in the HNRGs (P<0.05). Additionally, MDR/RR-TB patients with a PNI>38.7 had a lower prevalence of smoking, HIV coinfection, and COPD (P<0.05). MDR/RR-TB patients with a CONUT score≥6.5 had a higher prevalence of HIV co-infection (P<0.05).

Table 2 Baseline Characteristics of Participants with Low and High Nutritional Risk According to HALP Score in the Training Set

Table 3 Baseline Characteristics of Participants with Low and High Nutritional Risk According to PNI in the Training Set

Table 4 Baseline Characteristics of Participants with Low and High Nutritional Risk According to CONUT Score in the Training Set

Kaplan-Meier Curves for OS in Training Set and Validation Set

In training cohort, survival analyses suggested that patients with MDR/RR-TB in the HNRGs had lower OS than those in the LNRGs (Figure 1A–C), with the HALP score (median OS:20.02 vs 24.37 months, P<0.05), PNI (median OS:19.23 vs 24.37 months, P<0.05), and CONUT score (median OS:18.33 vs 24.35 months, P<0.05). Similarly, in the validation cohort, the OS of MDR/RR-TB patients in HNRGs were lower than that in LNRGs (Figure 1D–F), with HALP score (median OS:22.83 vs 24.37 months, P<0.05); PNI (median OS:19.20 vs 24.37 months, P<0.05), and CONUT score (median OS:22.60 vs 24.33 months, P<0.05).

Figure 1 Overall survival of patients with MDR/RR-TB according to nutritional indices in the training cohort and the validation cohort. Abbreviations: HALP score, hemoglobin albumin-lymphocyte-platelet score; PNI, prognostic nutritional index; CONUT score, controlling nutritional status score. Note: (A) HALP score, (B) PNI, and (C) CONUT score in the training cohort; (D) HALP score, (E) PNI, and (F) CONUT score in the validation cohort.

Cox Regression Analyses for All-Cause Mortality in the Training Set and Validation Set

The HALP score, PNI, CONUT, age, sex, drinking, bacterial co-infection, fungal co-infection, malignancy, pulmonary heart disease, bronchiectasis, COPD, destroyed lung, and pleural effusion were significantly associated with ACM in unadjusted and multivariable-adjusted Cox regression analyses (Table 5 and Supplementary Table 3). The adjusted multivariate Cox risk proportion regression revealed that a low HALP score (adjusted HR: 3.405, 95% CI: 1.411–8.215, P=0.006), low PNI (adjusted HR: 3.970, 95% CI: 1.662–9.485, P=0.002), and high CONUT score (adjusted HR: 4.734, 95% CI: 1.946–11.516, P=0.001) were independent risk factors for ACM in patients with MDR/RR-TB, which was also consistent with the results in the validation cohort (Table 5).

Table 5 Unadjusted and Multivariate Cox Risk Proportion Regression Analyses for All-Cause Mortality

Diagnostic Efficiency of Three Nutritional Indices for Mortality in the Training Set and Validation Set

ROC analyses revealed that the area under the curve of the HALP score was 0.765(95% CI:0.681–0.850; cut-off value: 16.85; sensitivity: 0.643; specificity: 0.797), PNI was 0.783(95% CI:0.693–0.873; cut-off value: 38.7; sensitivity: 0.755; specificity: 0.719) and CONUT score was 0.807(95% CI:0.733–0.880; cut-off value: 6.5; sensitivity: 0.684; specificity: 0.748) in the training cohort (Figure 2A). Similarly, the validation cohort showed equally good predictive performance, with AUCs of 0.764(95% CI:0.689–0.839; cut-off value: 16.85; sensitivity: 0.624; specificity: 0.814), 0.782(95% CI:0.712–0.852; cut-off value: 38.7; sensitivity: 0.757; specificity: 0.698), and 0.742(95% CI:0.662–0.823; cut-off value: 5.5; sensitivity: 0.610; specificity: 0.767), respectively (Figure 2B). In addition, AUC of HGB was 0.688(95% CI:0.592–0.784; cut-off value: 118.5; sensitivity: 0.548; specificity: 0.452), ALB was 0.748 (95% CI:0.644–0.852; cut-off value: 33.2; sensitivity: 0.776; specificity: 0.75) and lymphocyte count was 0.701(95% CI:0.609–0.794; cut-off value: 1.105; sensitivity: 0.594; specificity: 0.806) in the training cohort (Figure 2A) and in the validation cohort (Figure 2B) was 0.705(95% CI:0.623–0.787; cut-off value: 115.5; sensitivity: 0.624; specificity: 0.721), 0.723 (95% CI: 0.645–0.801; cut-off value: 32.6; sensitivity: 0.761; specificity: 0.581) and 0.731 (95% CI:0.652–0.809; cut-off value: 1.295; sensitivity: 0.553; specificity: 0.930), respectively. HALP score, PNI and CONUT score have better predictive performance than simple indicators including HGB, ALB and lymphocyte count for ACM in patients with MDR/RR-TB. Furthermore, the combination of the three nutrition indices showed a better AUC of 0.811(95% CI:0.743–0.879; cut-off value: 0.143; sensitivity: 0.758; specificity: 0.722) and 0.796 (95% CI:0.725–0.867; cut-off value: 0.211; sensitivity: 0.796; specificity: 0.698) than any single index in the two cohorts (Figure 2). DeLong ‘s test showed that there were statistically significant differences in AUC between combination index and HGB, combination index and lymphocyte count in the training set, while there were statistically significant differences in AUC between combination index and ALB, combination index and lymphocyte count in the validation set (Supplementary Table 4).

Figure 2 ROC curves for nutritional indices (HALP score, PNI, CONUT score and their combination) and simple parameters (HGB, ALB and lymphocyte count) in the training cohort and the validation cohort. Abbreviations: HALP score, hemoglobin-albumin-lymphocyte-platelet score; PNI, prognostic nutritional index; CONUT score, controlling nutritional status score; HGB, hemoglobin; ALB, albumin. Note: (A) Training cohort: The AUC (95% CI) of the HALP score, PNI, CONUT score and their combination, HGB, ALB and lymphocyte count were 0.765 (95% CI: 0.681–0.850; cut-off value: 16.85; sensitivity: 0.643; specificity: 0.797), 0.783 (95% CI: 0.693–0.873; cut-off value: 38.7; sensitivity: 0.755; specificity: 0.719), 0.807 (95% CI:0.733–0.880; cut-off value: 6.5; sensitivity: 0.684; specificity: 0.748), 0.811 (95% CI:0.743–0.879; cut-off value: 0.143; sensitivity: 0.758; specificity: 0.722), 0.688 (95% CI:0.592–0.784; cut-off value: 118.5; sensitivity: 0.548; specificity: 0.452), 0.748 (95% CI:0.644–0.852; cut-off value: 33.2; sensitivity: 0.776; specificity: 0.75) and 0.701 (95% CI:0.609–0.794; cut-off value: 1.105; sensitivity: 0.594; specificity: 0.806) respectively. (B) Validation cohort: AUC (95% CI) of the HALP score, PNI, CONUT score and their combination, HGB, ALB and lymphocyte count were 0.764 (95% CI:0.689–0.839; cut-off value: 16.85; sensitivity: 0.624; specificity: 0.814), 0.782 (95% CI:0.712–0.852; cut-off value: 38.7; sensitivity: 0.757; specificity: 0.698), 0.742 (95% CI:0.662–0.823; cut-off value: 5.5; sensitivity: 0.610; specificity: 0.767), 0.796 (95% CI:0.725–0.867; cut-off value: 0.211; sensitivity: 0.796; specificity: 0.698), 0.705 (95% CI:0.623–0.787; cut-off value: 115.5; sensitivity: 0.624; specificity: 0.721), 0.723 (95% CI: 0.645–0.801; cut-off value: 32.6; sensitivity: 0.761; specificity: 0.581) and 0.731 (95% CI:0.652–0.809; cut-off value: 1.295; sensitivity: 0.553; specificity: 0.930), respectively.

Discussion

MDR/RR-TB is still a serious issue in global TB control. Studies have shown that malnutrition is associated with MDR mortality,^14,15 and nutritional support can improve the prognosis of TB patients with malnutrition.^5,12,13 Hence, it is important to identify and intervene in malnourished MDR/RR-TB patients with a high risk of death.

Some studies, including our previous studies, have found that some laboratory indicators can predict complications and prognosis in patients with TB,^25,26 however, there were limited studies on nutritional indicators in TB patients. Recently, several nutritional indices, including HALP score, PNI, and CONUT score, have been proven to be used for assessing malnutrition and predicting the prognosis of some diseases,^4,27–31 which are relatively simple, convenient, effective, and practical. However, the effectiveness of these nutritional indices on prognosis prediction in MDR/RR-TB patients is unclear. To the best of our knowledge, this study is the first to explore the relationship between HALP score, PNI, and CONUT score and ACM in MDR/RR-TB patients and to further evaluate their predictive value for ACM in MDR/RR-TB patients.

The present study revealed that patients in the LNRGs had higher OS than those in the HNRGs. In addition, low HALP score, low PNI, and high CONUT score were independent risk factors for ACM in MDR/RR-TB patients in the adjusted multivariate Cox risk proportion regression analyses. Furthermore, compared with simple parameters such as HGB, ALB and lymphocyte count, the HALP score, PNI, and CONUT have more powerful predictive capabilities for predicting mortality in MDR/RR-TB patients, especially when combined. The possible reason for the inconsistency of AUCs comparison between the combination index and ALB, the combination index and lymphocyte count in the training and validation set is the insufficient sample size.

The HALP, PNI, and CONUT scores were calculated using HGB, ALB, lymphocyte count, platelet count, and total cholesterol (TC). HGB plays an important role in gas exchange in human organs and tissues, and is a diagnostic indicator of anemia. Previous evidence revealed a high prevalence of anemia among TB patients, and TB patients with anemia had an increased risk of death.³² ALB is a common component of the HALP score, PNI, and CONUT score, and has been proven to reflect nutritional status and systemic inflammation.³³ Franch AG suggested that lower ALB levels had a higher risk of mortality in populations with diseases or healthy populations.³⁴ As for lymphocyte count and platelet count, lymphocytes reflected the immune regulatory response,³⁵ and its count was proven to be the predictor of mortality in various diseases.³⁶ In addition, numerous observational studies report thrombocytosis in patients with TB, and platelet count correlated with disease severity.³⁷ TC was also an index, reflected nutritional status and systemic inflammation. Previous literature revealed that a cholesterol-rich diet accelerated the sterilization rate of sputum cultures in pulmonary TB patients,³⁸ and higher TC levels could reduce the risk of TB incidence and mortality.³⁹ Therefore, HALP score, PNI and CONUT scores could represent nutritional status and reflect the mortality of MDR/RR-TB patients and had good predictive value for mortality, especially when combined.

Furthermore, previous studies have shown that nutritional support had a commendable impact on improvement of lymphocyte count, hemoglobin, and albumin levels in patients with TB,^40,41 and robust nutritional intervention would be highly cost-effective in reducing TB mortality.⁵ Therefore, nutrition monitoring should be strengthened and nutritional interventions should be used early to minimize the mortality of MDR/RR-TB patients.

Nevertheless, this study had some limitations. Firstly, the present study was a retrospective single center study. Secondly, this study did not include a control group of non-drug resistant TB and could not compare the difference in nutritional indices between MDR/RR-TB and drug-sensitive TB patients. Therefore, Future prospective, multicenter, large-sample studies are needed to confirm and improve our conclusions.

Conclusion

In conclusion, the HALP, PNI, and CONUT scores were objective and simple nutritional indices and powerful predictors of ACM in MDR/RR-TB patients. Routine screening for the three nutritional indices, especially their combination, should be strengthened in clinical practice to identify high-risk MDR/RR-TB patients with mortality, and could help decrease the mortality of MDR/RR-TB patients using early nutritional interventions.