Systematic review of prediction models for pulmonary tuberculosis treatment outcomes in adults

A pharmacometric multistate model for predicting long-term treatment outcomes of patients with pulmonary TB

BACKGROUND

Studying long-term treatment outcomes of TB is time-consuming and impractical. Early and reliable biomarkers reflecting treatment response and capable of predicting long-term outcomes are urgently needed.

OBJECTIVES

To develop a pharmacometric multistate model to evaluate the link between potential predictors and long-term outcomes.

METHODS

Data were obtained from two Phase II clinical trials (TMC207-C208 and TMC207-C209) with bedaquiline on top of a multidrug background regimen. Patients were typically followed throughout a 24 week investigational treatment period plus a 96 week follow-up period. A five-state multistate model (active TB, converted, recurrent TB, dropout, and death) was developed to describe observed transitions. Evaluated predictors included patient characteristics, baseline TB disease severity and on-treatment biomarkers.

RESULTS

A fast bacterial clearance in the first 2 weeks and low TB bacterial burden at baseline increased probability to achieve conversion, whereas patients with XDR-TB were less likely to reach conversion. Higher estimated mycobacterial load at the end of 24 week treatment increased the probability of recurrence. At 120 weeks, the model predicted 55% (95% prediction interval, 50%-60%), 6.5% (4.2%-9.0%) and 7.5% (5.2%-10%) of patients in converted, recurrent TB and death states, respectively. Simulations predicted a substantial increase of recurrence after 24 weeks in patients with slow bacterial clearance regardless of baseline bacterial burden.

CONCLUSIONS

The developed multistate model successfully described TB treatment outcomes. The multistate modelling framework enables prediction of several outcomes simultaneously, and allows mechanistically sound investigation of novel promising predictors. This may help support future biomarker evaluation, clinical trial design and analysis.

Machine learning algorithms using national registry data to predict loss to follow-up during tuberculosis treatment

BACKGROUND

Identifying patients at increased risk of loss to follow-up (LTFU) is key to developing strategies to optimize the clinical management of tuberculosis (TB). The use of national registry data in prediction models may be a useful tool to inform healthcare workers about risk of LTFU. Here we developed a score to predict the risk of LTFU during anti-TB treatment (ATT) in a nationwide cohort of cases using clinical data reported to the Brazilian Notifiable Disease Information System (SINAN).

METHODS

We performed a retrospective study of all TB cases reported to SINAN between 2015 and 2022; excluding children (< 18 years-old), vulnerable groups or drug-resistant TB. For the score, data before treatment initiation were used. We trained and internally validated three different prediction scoring systems, based on Logistic Regression, Random Forest, and Light Gradient Boosting. Before applying our models we splitted our data into training (~ 80% data) and test (~ 20%) sets, and then compared the model metrics using the test data set.

RESULTS

Of the 243,726 cases included, 41,373 experienced LTFU whereas 202,353 were successfully treated. The groups were different with regards to several clinical and sociodemographic characteristics. The directly observed treatment (DOT) was unbalanced between the groups with lower prevalence in those who were LTFU. Three models were developed to predict LTFU using 8 features (prior TB, drug use, age, sex, HIV infection and schooling level) with different score composition approaches. Those prediction scoring systems exhibited an area under the curve (AUC) ranging between 0.71 and 0.72. The Light Gradient Boosting technique resulted in the best prediction performance, weighting specificity and sensitivity. A user-friendly web calculator app was developed ( https://tbprediction.herokuapp.com/ ) to facilitate implementation.

CONCLUSIONS

Our nationwide risk score predicts the risk of LTFU during ATT in Brazilian adults prior to treatment commencement utilizing schooling level, sex, age, prior TB status, and substance use (drug, alcohol, and/or tobacco). This is a potential tool to assist in decision-making strategies to guide resource allocation, DOT indications, and improve TB treatment adherence.

Prediction Models for Adverse Drug Reactions During Tuberculosis Treatment in Brazil

BACKGROUND

Tuberculosis (TB) treatment-related adverse drug reactions (TB-ADRs) can negatively affect adherence and treatment success rates.

METHODS

We developed prediction models for TB-ADRs, considering participants with drug-susceptible pulmonary TB who initiated standard TB therapy. TB-ADRs were determined by the physician attending the participant, assessing causality to TB drugs, the affected organ system, and grade. Potential baseline predictors of TB-ADR included concomitant medication (CM) use, human immunodeficiency virus (HIV) status, glycated hemoglobin (HbA1c), age, body mass index (BMI), sex, substance use, and TB drug metabolism variables (NAT2 acetylator profiles). The models were developed through bootstrapped backward selection. Cox regression was used to evaluate TB-ADR risk.

RESULTS

There were 156 TB-ADRs among 102 of the 945 (11%) participants included. Most TB-ADRs were hepatic (n = 82 [53%]), of moderate severity (grade 2; n = 121 [78%]), and occurred in NAT2 slow acetylators (n = 62 [61%]). The main prediction model included CM use, HbA1c, alcohol use, HIV seropositivity, BMI, and age, with robust performance (c-statistic = 0.79 [95% confidence interval {CI}, .74-.83) and fit (optimism-corrected slope and intercept of -0.09 and 0.94, respectively). An alternative model replacing BMI with NAT2 had similar performance. HIV seropositivity (hazard ratio [HR], 2.68 [95% CI, 1.75-4.09]) and CM use (HR, 5.26 [95% CI, 2.63-10.52]) increased TB-ADR risk.

CONCLUSIONS

The models, with clinical variables and with NAT2, were highly predictive of TB-ADRs.

Integrative analysis of multimodal patient data identifies personalized predictors of tuberculosis treatment prognosis

Tuberculosis (TB) afflicted 10.6 million people in 2021, and its global burden is increasing due to multidrug-resistant TB (MDR-TB) and extensively resistant TB (XDR-TB). Here, we analyze multi-domain information from 5,060 TB patients spanning 10 countries with high burden of MDR-TB from the NIAID TB Portals database to determine predictors of TB treatment outcome. Our analysis revealed significant associations between radiological, microbiological, therapeutic, and demographic data modalities. Our machine learning model, built with 203 features across modalities outperforms models built using each modality alone in predicting treatment outcomes, with an accuracy of 83% and area under the curve of 0.84. Notably, our analysis revealed that the drug regimens Bedaquiline-Clofazimine-Cycloserine-Levofloxacin-Linezolid and Bedaquiline-Clofazimine-Linezolid-Moxifloxacin were associated with treatment success and failure, respectively, for MDR non-XDR-TB. Drug combinations predicted to be synergistic by the INDIGO algorithm performed better than antagonistic combinations. Our prioritized set of features predictive of treatment outcomes can ultimately guide the personalized clinical management of TB.

Activities of Daily Living, Hypoxemia, and Lymphocytes Score for Predicting Mortality Risk in Patients With Pulmonary TB

BACKGROUND

A clinically applicable mortality risk prediction system for pulmonary TB may improve treatment outcomes, but no easy-to-calculate and accurate score has yet been reported. The aim of this study was to construct a simple and objective disease severity score for patients with pulmonary TB.

RESEARCH QUESTION

Does a clinical score consisting of simple objective factors predict the mortality risk of patients with pulmonary TB?

STUDY DESIGN AND METHODS

The data set from our previous prospective study that recruited patients newly diagnosed with pulmonary TB was used for the development cohort. Patients for the validation cohort were prospectively recruited between March 2021 and September 2022. The primary end point was all-cause in-hospital mortality. Using Cox proportional hazards regression, a mortality risk prediction model was optimized in the development cohort. The disease severity score was developed by assigning integral points to each variate.

RESULTS

The data from 252 patients in the development cohort and 165 patients in the validation cohort were analyzed, of whom 39 (15.5%) and 17 (10.3%), respectively, died in the hospital. The disease severity score (named the AHL score) included three clinical parameters: activities of daily living (semi-dependent, 1 point; totally dependent, 2 points); hypoxemia (1 point), and lymphocytes (< 720/μL, 1 point). This score showed good discrimination with a C statistic of 0.902 in the development cohort and 0.842 in the validation cohort. We stratified the score into three groups (scores of 0, 1-2, and 3-4), which clearly corresponded to low (0% and 1.3%), intermediate (13.5% and 8.9%), and high (55.8% and 39.3%) mortality risk in the development and validation cohorts.

INTERPRETATION

The easy-to-calculate AHL disease severity score for patients with pulmonary TB was able to categorize patients into three mortality risk groups with great accuracy.

CLINICAL TRIAL REGISTRATION

University Hospital Medical Information Network Center; No. UMIN000012727 and No. UMIN000043849; URL: www.umin.ac.jp.

The effect of machine learning algorithms in the prediction, and diagnosis of meningitis: A systematic review

Background and Aims

This systematic review aimed to evaluating the effectiveness of machine learning (ML) algorithms for the prediction and diagnosis of meningitis.

Methods

On November 12, 2022, a systematic review was carried out using a keyword search in the reliable scientific databases PubMed, EMBASE, Scopus, and Web of Science. The recommendations of Preferred Reporting for Systematic Reviews and Meta-Analyses (PRISMA) were adhered to. Studies conducted in English that employed ML to predict and identify meningitis were deemed to match the inclusion criteria. The eligibility requirements were used to independently review the titles and abstracts. The whole text was then obtained and independently reviewed in accordance with the eligibility requirements.

Results

After all the research matched the inclusion criteria, a total of 16 studies were added to the systematic review. Studies on the application of ML algorithms in the three categories of disease diagnosis ability (8.16) and disease prediction ability (8.16) (including cases related to identifying patients (50%), risk of death in patients (25%), the consequences of the disease in childhood (12.5%), and its etiology [12.5%]) were placed. Among the ML algorithms used in this study, logistic regression (LR) (4.16, 25%) and multiple logistic regression (MLR) (4.16, 25%) were the most used. All the included studies indicated improvements in the processes of diagnosis, prediction, and disease outbreak with the help of ML algorithms.

Conclusion

The results of the study showed that in all included studies, ML algorithms were an effective approach to facilitate diagnosis, predict consequences for risk classification, and improve resource utilization by predicting the volume of patients or services as well as discovering risk factors. The role of ML algorithms in improving disease diagnosis was more significant than disease prediction and prevalence. Meanwhile, the use of combined methods can optimize differential diagnoses and facilitate the decision-making process for healthcare providers.

Machine learning-based radiomics to distinguish pulmonary nodules between lung adenocarcinoma and tuberculosis

BACKGROUND

Radiomics is increasingly utilized to distinguish pulmonary nodules between lung adenocarcinoma (LUAD) and tuberculosis (TB). However, it remains unclear whether different segmentation criteria, such as the inclusion or exclusion of the cavity region within nodules, affect the results.

METHODS

A total of 525 patients from two medical centers were retrospectively enrolled. The radiomics features were extracted according to two regions of interest (ROI) segmentation criteria. Multiple logistic regression models were trained to predict the pathology: (1) The clinical model relied on clinical-radiological semantic features; (2) The radiomics models (radiomics+ and radiomics-) utilized radiomics features from different ROIs (including or excluding cavities); (3) the composite models (composite+ and composite-) incorporated both above.

RESULTS

In the testing set, the radiomics+/- models and the composite+/- models still possessed efficient prediction performance (AUC ≥ 0.94), while the AUC of the clinical model was 0.881. In the validation set, the AUC of the clinical model was only 0.717, while that of the radiomics+/- models and the composite+/- models ranged from 0.801 to 0.825. The prediction performance of all the radiomics+/- and composite+/- models were significantly superior to that of the clinical model (p < 0.05). Whether the ROI segmentation included or excluded the cavity had no significant effect on these models (radiomics+ vs. radiomics-, composite+ model vs. composite-) (p > 0.05).

CONCLUSIONS

The present study established a machine learning-based radiomics strategy for differentiating LUAD from TB lesions. The ROI segmentation including or excluding the cavity region may exert no significant effect on the predictive ability.

Tuberculosis in Sudan: systematic review and meta analysis

Every year, 10 million people fall ill with tuberculosis (TB). Despite being a preventable and curable disease, 1.5 million people die from TB each year -making it the world's top infectious disease. TB is the leading cause of death of people with HIV and also a major contributor to antimicrobial resistance. Its presumed that TB was the cause of 1% of the total deaths among inpatients in Sudan in 2017. The current study is aimed to provide pooled prevalence of Mycobacterium tuberculosis among Sudanese as well as to determine any socio-cultural risk factors associated. A systematic review of the literature was conducted and regulated in accordance with the PRISMA Statement. After abstract and full text screening only twenty-six articles met our inclusion criteria and passed the quality assessment procedure. Pulmonary tuberculosis prevalence was assessed in sixteen included studies among participants from Khartoum, Gezira, Kassala, Blue Nile, River Nile, White Nile, Gadarif, Red sea, North Kordofan, Northern State, Sennar and West Darfur States, representing a total sample size of 11,253 participants of suspected individuals such as febrile outpatients, TB patients' contacts and other groups such as HIV/AIDS patients, hemodialysis patients, School adolescents as well as pregnant women. The pooled prevalence was 30.72% [CI: 30.64, 30.81]. Moreover, Khartoum State recorded the highest pooled prevalence as 41.86% [CI: 14.69, 69.02] based on a total sample size of 2,737 participants. Furthermore, male gender and rural residence were found to be significantly associated with TB infection. Further research with larger sample sizes targeting prevalence and risk factors of TB among Sudanese population is needed to be conducted.

Association of indicators of extensive disease and rifampin-resistant tuberculosis treatment outcomes: an individual participant data meta-analysis

BACKGROUND

Indicators of extensive disease-acid fast bacilli (AFB) smear positivity and lung cavitation-have been inconsistently associated with clinical rifampin-resistant/multidrug-resistant tuberculosis (RR/MDR-TB) outcomes. We evaluated the association of these indicators with end-of-treatment outcomes.

METHODS

We did an individual participant data meta-analysis of people treated for RR/MDR-TB with longer regimens with documented AFB smear and chest radiography findings. We compared people AFB smear-negative without cavities to people: (1) smear-negative with lung cavities; (2) smear-positive without lung cavities and (3) AFB smear-positive with lung cavities. Using multivariable logistic regression accounting for demographic, treatment and clinical factors, we calculated adjusted ORs (aOR) for any unfavourable outcome (death, lost to follow-up, failure/recurrence), and mortality and treatment failure/recurrence alone.

RESULTS

We included 5596 participants; included participants significantly differed from excluded participants. Overall, 774 (13.8%) were AFB smear-negative without cavities, 647 (11.6%) only had cavities, 1424 (25.4%) were AFB smear-positive alone and 2751 (49.2%) were AFB smear-positive with cavities. The median age was 37 years (IQR: 28-47), 3580 (64%) were male and 686 (12.5%) had HIV. Compared with participants AFB smear-negative without cavities, aOR (95% CI) for any unfavourable outcome was 1.0 (0.8 to 1.4) for participants smear-negative with lung cavities, 1.2 (0.9 to 1.5) if smear-positive without cavities and 1.6 (1.3 to 2.0) if AFB smear-positive with lung cavities. Odds were only significantly increased for mortality (1.5, 95% CI 1.1 to 2.1) and failure/recurrence (2.2, 95% CI 1.5 to 3.3) among participants AFB smear-positive with lung cavities.

CONCLUSION

Only the combination of AFB smear-positivity and lung cavitation was associated with unfavourable outcomes, suggesting they may benefit from stronger regimens.

Machine learning algorithms using national registry data to predict loss to follow- up during tuberculosis treatment

Background

Identifying patients at increased risk of loss to follow-up (LTFU) is key to developing strategies to optimize the clinical management of tuberculosis (TB). The use of national registry data in prediction models may be a useful tool to inform healthcare workers about risk of LTFU. Here we developed a score to predict the risk of LTFU during anti-TB treatment (ATT) in a nationwide cohort of cases using clinical data reported to the Brazilian Notifiable Disease Information System (SINAN).

Methods

We performed a retrospective study of all TB cases reported to SINAN between 2015-2022; excluding children (<18 years-old), vulnerable groups or drug-resistant TB. For the score, data before treatment initiation were used. We trained and internally validated three different prediction scoring systems, based on Logistic Regression, Random Forest, and Light Gradient Boosting. Before applying our models we split our data into train (~80% data) and test (~20%), and then we compare model metrics using a test data set.

Results

Of the 243,726 cases included, 41,373 experienced LTFU whereas 202,353 were successfully treated and cured. The groups were different with regards to several clinical and sociodemographic characteristics. The directly observed treatment (DOT) was unbalanced between the groups with lower prevalence in those who were LTFU. Three models were developed to predict LTFU using 8 features (prior TB, drug use, age, sex, HIV infection and schooling level) with different score composition approaches. Those prediction scoring system exhibited an area under the curve (AUC) ranging between 0.71 and 0.72. The Light Gradient Boosting technique resulted in the best prediction performance, weighting specificity, and sensibility. A user-friendly web calculator app was created (https://tbprediction.herokuapp.com/) to facilitate implementation.

Conclusions

Our nationwide risk score predicts the risk of LTFU during ATT in Brazilian adults prior to treatment commencement. This is a potential tool to assist in decision-making strategies to guide resource allocation, DOT indications, and improve TB treatment adherence.

Deep learning on longitudinal CT scans: automated prediction of treatment outcomes in hospitalized tuberculosis patients

Three deep learning (DL)-based prediction models (PMs) using longitudinal CT images were developed to predict tuberculosis (TB) treatment outcomes. The internal dataset consists of 493 bacteriologically confirmed TB patients who completed the anti-tuberculosis treatment with three-time CT scans, including a pretreatment CT scan and two follow-up CT scans. PM1 was trained using only pretreatment CT scans, and PM2 and PM3 were developed by adding follow-up scans. An independent testing was performed on external dataset comprising 86 TB patients. The area under the curve for classifying success and drug-resistant (DR)-TB was improved on both internal (0.609 vs. 0.625 vs. 0.815) and external (0.627 vs. 0.705 vs. 0.735) dataset by adding follow-up scans. The accuracy and F1-score also showed an increasing tendency in the external test. Regular follow-up CT scans can aid in the treatment prediction, and special attention should be given to early intensive phase of treatment to identify high-risk DR-TB patients.

Target product profiles: tests for tuberculosis treatment monitoring and optimization

The World Health Organization has developed target product profiles containing minimum and optimum targets for key characteristics for tests for tuberculosis treatment monitoring and optimization. Tuberculosis treatment optimization refers to initiating or switching to an effective tuberculosis treatment regimen that results in a high likelihood of a good treatment outcome. The target product profiles also cover tests of cure conducted at the end of treatment. The development of the target product profiles was informed by a stakeholder survey, a cost-effectiveness analysis and a patient-care pathway analysis. Additional feedback from stakeholders was obtained by means of a Delphi-like process, a technical consultation and a call for public comment on a draft document. A scientific development group agreed on the final targets in a consensus meeting. For characteristics rated of highest importance, the document lists: (i) high diagnostic accuracy (sensitivity and specificity); (ii) time to result of optimally ≤ 2 hours and no more than 1 day; (iii) required sample type to be minimally invasive, easily obtainable, such as urine, breath, or capillary blood, or a respiratory sample that goes beyond sputum; (iv) ideally the test could be placed at a peripheral-level health facility without a laboratory; and (v) the test should be affordable to low- and middle-income countries, and allow wide and equitable access and scale-up. Use of these target product profiles should facilitate the development of new tuberculosis treatment monitoring and optimization tests that are accurate and accessible for all people being treated for tuberculosis.

Prognostic prediction models for clinical outcomes in patients diagnosed with visceral leishmaniasis: protocol for a systematic review

INTRODUCTION

Visceral leishmaniasis (VL) is a neglected tropical disease responsible for many thousands of preventable deaths each year. Symptomatic patients often struggle to access effective treatment, without which death is the norm. Risk prediction tools support clinical teams and policymakers in identifying high-risk patients who could benefit from more intensive management pathways. Investigators interested in using their clinical data for prognostic research should first identify currently available models that are candidates for validation and possible updating. Addressing these needs, we aim to identify, summarise and appraise the available models predicting clinical outcomes in VL patients.

METHODS AND ANALYSIS

We will include studies that have developed, validated or updated prognostic models predicting future clinical outcomes in patients diagnosed with VL. Systematic reviews and meta-analyses that include eligible studies are also considered for review. Conference abstracts and educational theses are excluded. Data extraction, appraisal and reporting will follow current methodological guidelines. Ovid Embase; Ovid MEDLINE; the Web of Science Core Collection, SciELO and LILACS are searched from database inception to 1 March 2023 using terms developed for the identification of prediction models, and with no language restriction. Screening, data extraction and risk of bias assessment will be performed in duplicate with discordance resolved by a third independent reviewer. Risk of bias will be assessed using the Prediction model Risk Of Bias ASsessment Tool (PROBAST). Tables and figures will compare and contrast key model information, including source data, participants, model development and performance measures, and risk of bias. We will consider the strengths, limitations and clinical applicability of the identified models.

ETHICS AND DISSEMINATION

Ethics approval is not required for this review. The systematic review and all accompanying data will be submitted to an open-access journal. Findings will also be disseminated through the research group's website (www.iddo.org/research-themes/visceral-leishmaniasis) and social media channels.

PROSPERO REGISTRATION NUMBER

CRD42023417226.

T-BACCO SCORE: A predictive scoring tool for tuberculosis (TB) loss to follow-up among TB smokers

INTRODUCTION

Loss to follow-up (LTFU) and smoking during TB treatment are major challenges for TB control programs. Smoking increases the severity and prolongs TB treatment duration, which lead to a higher rate of LTFU. We aim to develop a prognostic scoring tool to predict LTFU among TB patients who smoke to improve successful TB treatment outcomes.

MATERIALS AND METHODS

The development of the prognostic model utilized prospectively collected longitudinal data of adult TB patients who smoked in the state of Selangor between 2013 until 2017, which were obtained from the Malaysian Tuberculosis Information System (MyTB) database. Data were randomly split into development and internal validation cohorts. A simple prognostic score (T-BACCO SCORE) was constructed based on the regression coefficients of predictors in the final logistic model of the development cohort. Estimated missing data was 2.8% from the development cohort and was completely at random. Model discrimination was determined using c-statistics (AUCs), and calibration was based on the Hosmer and Lemeshow goodness of fit test and calibration plot.

RESULTS

The model highlights several variables with different T-BACCO SCORE values as predictors for LTFU among TB patients who smoke (e.g., age group, ethnicity, locality, nationality, educational level, monthly income level, employment status, TB case category, TB detection methods, X-ray categories, HIV status, and sputum status). The prognostic scores were categorized into three groups that predict the risk for LTFU: low-risk (<15 points), medium-risk (15 to 25 points) and high-risk (> 25 points). The model exhibited fair discrimination with a c-statistic of 0.681 (95% CI 0.627-0.710) and good calibration with a nonsignificant chi-square Hosmer‒Lemeshow's goodness of fit test χ2 = 4.893 and accompanying p value of 0.769.

CONCLUSION

Predicting LTFU among TB patients who smoke in the early phase of TB treatment is achievable using this simple T-BACCO SCORE. The applicability of the tool in clinical settings helps health care professionals manage TB smokers based on their risk scores. Further external validation should be carried out prior to use.

A nomogram for predicting mortality of patients initially diagnosed with primary pulmonary tuberculosis in Hunan province, China: a retrospective study

Objective

According to the Global Tuberculosis Report for three consecutive years, tuberculosis (TB) is the second leading infectious killer. Primary pulmonary tuberculosis (PTB) leads to the highest mortality among TB diseases. Regretfully, no previous studies targeted the PTB of a specific type or in a specific course, so models established in previous studies cannot be accurately feasible for clinical treatments. This study aimed to construct a nomogram prognostic model to quickly recognize death-related risk factors in patients initially diagnosed with PTB to intervene and treat high-risk patients as early as possible in the clinic to reduce mortality.

Methods

We retrospectively analyzed the clinical data of 1,809 in-hospital patients initially diagnosed with primary PTB at Hunan Chest Hospital from January 1, 2019, to December 31, 2019. Binary logistic regression analysis was used to identify the risk factors. A nomogram prognostic model for mortality prediction was constructed using R software and was validated using a validation set.

Results

Univariate and multivariate logistic regression analyses revealed that drinking, hepatitis B virus (HBV), body mass index (BMI), age, albumin (ALB), and hemoglobin (Hb) were six independent predictors of death in in-hospital patients initially diagnosed with primary PTB. Based on these predictors, a nomogram prognostic model was established with high prediction accuracy, of which the area under the curve (AUC) was 0.881 (95% confidence interval [Cl]: 0.777-0.847), the sensitivity was 84.7%, and the specificity was 77.7%.Internal and external validations confirmed that the constructed model fit the real situation well.

Conclusion

The constructed nomogram prognostic model can recognize risk factors and accurately predict the mortality of patients initially diagnosed with primary PTB. This is expected to guide early clinical intervention and treatment for high-risk patients.

Development and validation of a prediction model for unsuccessful treatment outcomes in patients with multi-drug resistance tuberculosis

BACKGROUND

The World Health Organization has reported that the treatment success rate of multi-drug resistance tuberculosis is approximately 57% globally. Although new drugs such as bedaquiline and linezolid is likely improve the treatment outcome, there are other factors associated with unsuccessful treatment outcome. The factors associated with unsuccessful treatment outcomes have been widely examined, but only a few studies have developed prediction models. We aimed to develop and validate a simple clinical prediction model for unsuccessful treatment outcomes in patients with multi-drug resistance pulmonary tuberculosis (MDR-PTB).

METHODS

This retrospective cohort study was performed between January 2017 and December 2019 at a special hospital in Xi'an, China. A total of 446 patients with MDR-PTB were included. Least absolute shrinkage and selection operator (LASSO) regression and multivariate logistic regression were used to select prognostic factors for unsuccessful treatment outcomes. A nomogram was built based on four prognostic factors. Internal validation and leave-one-out cross-validation was used to assess the model.

RESULTS

Of the 446 patients with MDR-PTB, 32.9% (147/446) cases had unsuccessful treatment outcomes, and 67.1% had successful outcomes. After LASSO regression and multivariate logistic analyses, no health education, advanced age, being male, and larger extent lung involvement were identified as prognostic factors. These four prognostic factors were used to build the prediction nomograms. The area under the curve of the model was 0.757 (95%CI 0.711 to 0.804), and the concordance index (C-index) was 0.75. For the bootstrap sampling validation, the corrected C-index was 0.747. In the leave-one-out cross-validation, the C-index was 0.765. The slope of the calibration curve was 0.968, which was approximately 1.0. This indicated that the model was accurate in predicting unsuccessful treatment outcomes.

CONCLUSIONS

We built a predictive model and established a nomogram for unsuccessful treatment outcomes of multi-drug resistance pulmonary tuberculosis based on baseline characteristics. This predictive model showed good performance and could be used as a tool by clinicians to predict who among their patients will have an unsuccessful treatment outcome.

Diagnostic models predicting paediatric viral acute respiratory infections: a systematic review

OBJECTIVES

To systematically review and evaluate diagnostic models used to predict viral acute respiratory infections (ARIs) in children.

DESIGN

Systematic review.

DATA SOURCES

PubMed and Embase were searched from 1 January 1975 to 3 February 2022.

ELIGIBILITY CRITERIA

We included diagnostic models predicting viral ARIs in children (<18 years) who sought medical attention from a healthcare setting and were written in English. Prediction model studies specific to SARS-CoV-2, COVID-19 or multisystem inflammatory syndrome in children were excluded.

DATA EXTRACTION AND SYNTHESIS

Study screening, data extraction and quality assessment were performed by two independent reviewers. Study characteristics, including population, methods and results, were extracted and evaluated for bias and applicability using the Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies and PROBAST (Prediction model Risk Of Bias Assessment Tool).

RESULTS

Of 7049 unique studies screened, 196 underwent full text review and 18 were included. The most common outcome was viral-specific influenza (n=7; 58%). Internal validation was performed in 8 studies (44%), 10 studies (56%) reported discrimination measures, 4 studies (22%) reported calibration measures and none performed external validation. According to PROBAST, a high risk of bias was identified in the analytic aspects in all studies. However, the existing studies had minimal bias concerns related to the study populations, inclusion and modelling of predictors, and outcome ascertainment.

CONCLUSIONS

Diagnostic prediction can aid clinicians in aetiological diagnoses of viral ARIs. External validation should be performed on rigorously internally validated models with populations intended for model application.

PROSPERO REGISTRATION NUMBER

CRD42022308917.

Tuberculosis Treatment Outcomes in Brazil: Different Predictors for Each Type of Unsuccessful Outcome

BACKGROUND

Successful tuberculosis (TB) treatment is necessary for disease control. The World Health Organization (WHO) has a target TB treatment success rate of ≥90%. We assessed whether the different types of unfavorable TB treatment outcome had different predictors.

METHODS

Using data from Regional Prospective Observational Research for Tuberculosis-Brazil, we evaluated biological and behavioral factors associated with each component of unsuccessful TB outcomes, recently updated by WHO (death, loss to follow-up [LTFU], and treatment failure). We included culture-confirmed, drug-susceptible, pulmonary TB participants receiving standard treatment in 2015-2019. Multinomial logistic regression models with inverse probability weighting were used to evaluate the distinct determinants of each unsuccessful outcome.

RESULTS

Of 915 participants included, 727 (79%) were successfully treated, 118 (13%) were LTFU, 44 (5%) had treatment failure, and 26 (3%) died. LTFU was associated with current drug-use (adjusted odds ratio [aOR] = 5.3; 95% confidence interval [CI], 3.0-9.4), current tobacco use (aOR = 2.9; 95% CI, 1.7-4.9), and being a person with HIV (PWH) (aOR = 2.0; 95% CI, 1.1-3.5). Treatment failure was associated with PWH (aOR = 2.7; 95% CI, 1.2-6.2) and having diabetes (aOR = 2.2; 95% CI, 1.1-4.4). Death was associated with anemia (aOR = 5.3; 95% CI, 1.4-19.7), diabetes (aOR = 3.1; 95% CI, 1.4-6.7), and PWH (aOR = 3.9; 95% CI, 1.3-11.4). Direct observed therapy was protective for treatment failure (aOR = 0.5; 95% CI, .3-.9) and death (aOR = 0.5; 95% CI, .2-1.0).

CONCLUSIONS

The treatment success rate was below the WHO target. Behavioral factors were most associated with LTFU, whereas clinical comorbidities were correlated with treatment failure and death. Because determinants of unsuccessful outcomes are distinct, different intervention strategies may be needed to improve TB outcomes.

High variability in tuberculosis treatment outcomes across 15 health facilities in a semi-urban area in central Ethiopia

Background

Despite reported tuberculosis (TB) treatment success rate of 86%, TB remains a leading cause of death in Ethiopia. We investigated patient and provider-specific factors associated with unfavorable treatment outcomes in Ethiopian health facilities providing TB care.

Methods

Data on characteristics and treatment outcomes of patients registered for TB treatment at 15 public health facilities (4 hospitals and 11 health centres) were collected from clinic registers. Proportions of unfavorable outcomes (defined as deaths, loss-to-follow-up [LTFU] and treatment failure), were compared across facilities using multivariable logistic regression, with separate analyses for death and LTFU.

Results

Among 3359 patients (53.5 % male, median age 28 years, 19.6 % HIV-positive), 296 (8.8 %) had unfavorable treatment outcome. Proportions of unfavorable outcomes across facilities ranged from 2.0 % to 21.1 % (median 8.3 %). Median proportions of death and LTFU among facilities were 3.3 % (range 0-10.9 %) and 2.6 % (range 0.6 %-19.2 %), respectively. Three facilities had significantly higher rates of LTFU, whereas two facilities had higher rates of death. The two facilities with full-time TB-nurses had higher proportions of successful outcomes (95.2 % vs 90.1 %, adjusted odds ratio 2.27, p < 0.0001).

Conclusion

Substantial variability of TB treatment outcomes was observed across the assessed health facilities providing TB care, independently of age and HIV co-infection, reflecting possible differences in service structure and related quality of care.

Bedaquiline resistance probability to guide treatment decision making for rifampicin-resistant tuberculosis: insights from a qualitative study

BACKGROUND

Bedaquiline (BDQ) is a core drug for rifampicin-resistant tuberculosis (RR-TB) treatment. Accurate prediction of a BDQ-resistant phenotype from genomic data is not yet possible. A Bayesian method to predict BDQ resistance probability from next-generation sequencing data has been proposed as an alternative.

METHODS

We performed a qualitative study to investigate the decision-making of physicians when facing different levels of BDQ resistance probability. Fourteen semi-structured interviews were conducted with physicians experienced in treating RR-TB, sampled purposefully from eight countries with varying income levels and burden of RR-TB. Five simulated patient scenarios were used as a trigger for discussion. Factors influencing the decision of physicians to prescribe BDQ at macro-, meso- and micro levels were explored using thematic analysis.

RESULTS

The perception and interpretation of BDQ resistance probability values varied widely between physicians. The limited availability of other RR-TB drugs and the high cost of BDQ hindered physicians from altering the BDQ-containing regimen and incorporating BDQ resistance probability in their decision-making. The little experience with BDQ susceptibility testing and whole-genome sequencing results, and the discordance between phenotypic susceptibility and resistance probability were other barriers for physicians to interpret the resistance probability estimates. Especially for BDQ resistance probabilities between 25% and 70%, physicians interpreted the resistance probability value dynamically, and other factors such as clinical and bacteriological treatment response, history of exposure to BDQ, and resistance profile were often considered more important than the BDQ probability value for the decision to continue or stop BDQ. In this grey zone, some physicians opted to continue BDQ but added other drugs to strengthen the regimen.

CONCLUSIONS

This study highlights the complexity of physicians' decision-making regarding the use of BDQ in RR-TB regimens for different levels of BDQ resistance probability.. Ensuring sufficient access to BDQ and companion drugs, improving knowledge of the genotype-phenotype association for BDQ resistance, availability of a rapid molecular test, building next-generation sequencing capacity, and developing a clinical decision support system incorporating BDQ resistance probability will all be essential to facilitate the implementation of BDQ resistance probability in personalizing treatment for patients with RR-TB.

The World of Organoids: Gastrointestinal Disease Modelling in the Age of 3R and One Health with Specific Relevance to Dogs and Cats

One Health describes the importance of considering humans, animals, and the environment in health research. One Health and the 3R concept, i.e., the replacement, reduction, and refinement of animal experimentation, shape today’s research more and more. The development of organoids from many different organs and animals led to the development of highly sophisticated model systems trying to replace animal experiments. Organoids may be used for disease modelling in various ways elucidating the manifold host–pathogen interactions. This review provides an overview of disease modelling approaches using organoids of different kinds with a special focus on animal organoids and gastrointestinal diseases. We also provide an outlook on how the research field of organoids might develop in the coming years and what opportunities organoids hold for in-depth disease modelling and therapeutic interventions.

Couples-based interventions and postpartum contraceptive uptake: A systematic review

OBJECTIVE

Systematically review the existing evidence about couples-based interventions and postpartum contraceptive uptake and generate recommendations for future research.

DATA SOURCES

PubMed, Web of Science, PsycINFO, Embase, and CINAHL through June 7, 2021.

STUDY SELECTION AND DATA EXTRACTION

Studies with a couples-based intervention assessing postpartum contraceptive uptake. Two independent reviewers screened studies, extracted data, and assessed risk of bias with RoB-2 (Cochrane Risk of Bias 2) for randomized and ROBINS-I (Risk of Bias in Non-Randomized Studies - Interventions) for observational studies. Data were synthesized in tables, figures, and a narrative review.

RESULTS

A total of 925 papers were identified, 66 underwent full text review, and 17 articles, which included 18 studies - 16 randomized, 2 observational - were included. The lack of intervention and outcome homogeneity precluded meta-analysis and isolating the effect of partner involvement. Four studies were partner-required, where partner involvement was a required component of the intervention, and 14 were partner-optional. Unadjusted risk differences ranged from 0.01 to 0.51 in favor of couples-based interventions increasing postpartum contraceptive uptake versus standard of care. Bias assessment of the 16 randomized studies classified 8, 3, and 5 studies as at a high, some concern, and low risk of bias. Common sources of bias included intervention non-adherence and missing outcome data. One observational study was at a high and the other at a low risk of bias.

CONCLUSIONS

Future studies that assess couples-based interventions must clearly define and measure how partners are involved in the intervention and assess how intervention adherence impacts postpartum contraceptive uptake.

How to conduct a systematic review and meta-analysis of prognostic model studies

BACKGROUND

Prognostic models are typically developed to estimate the risk that an individual in a particular health state will develop a particular health outcome, to support (shared) decision making. Systematic reviews of prognostic model studies can help identify prognostic models that need to further be validated or are ready to be implemented in healthcare.

OBJECTIVES

To provide a step-by-step guidance on how to conduct and read a systematic review of prognostic model studies and to provide an overview of methodology and guidance available for every step of the review progress.

SOURCES

Published, peer-reviewed guidance articles.

CONTENT

We describe the following steps for conducting a systematic review of prognosis studies: 1) Developing the review question using the Population, Index model, Comparator model, Outcome(s), Timing, Setting format, 2) Searching and selection of articles, 3) Data extraction using the Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies (CHARMS) checklist, 4) Quality and risk of bias assessment using the Prediction model Risk Of Bias ASsessment (PROBAST) tool, 5) Analysing data and undertaking quantitative meta-analysis, and 6) Presenting summary of findings, interpreting results, and drawing conclusions. Guidance for each step is described and illustrated using a case study on prognostic models for patients with COVID-19.

IMPLICATIONS

Guidance for conducting a systematic review of prognosis studies is available, but the implications of these reviews for clinical practice and further research highly depend on complete reporting of primary studies.

Integrative analysis of clinical health records, imaging and pathogen genomics identifies personalized predictors of disease prognosis in tuberculosis

Tuberculosis (TB) afflicts over 10 million people every year and its global burden is projected to increase dramatically due to multidrug-resistant TB (MDR-TB). The Covid-19 pandemic has resulted in reduced access to TB diagnosis and treatment, reversing decades of progress in disease management globally. It is thus crucial to analyze real-world multi-domain information from patient health records to determine personalized predictors of TB treatment outcome and drug resistance. We conduct a retrospective analysis on electronic health records of 5060 TB patients spanning 10 countries with high burden of MDR-TB including Ukraine, Moldova, Belarus and India available on the NIAID-TB portals database. We analyze over 200 features across multiple host and pathogen modalities representing patient social demographics, disease presentations as seen in cChest X rays and CT scans, and genomic records with drug susceptibility features of the pathogen strain from each patient. Our machine learning model, built with diverse data modalities outperforms models built using each modality alone in predicting treatment outcomes, with an accuracy of 81% and AUC of 0.768. We determine robust predictors across countries that are associated with unsuccessful treatmentclinical outcomes, and validate our predictions on new patient data from TB Portals. Our analysis of drug regimens and drug interactions suggests that synergistic drug combinations and those containing the drugs Bedaquiline, Levofloxacin, Clofazimine and Amoxicillin see more success in treating MDR and XDR TB. Features identified via chest imaging such as percentage of abnormal volume, size of lung cavitation and bronchial obstruction are associated significantly with pathogen genomic attributes of drug resistance. Increased disease severity was also observed in patients with lower BMI and with comorbidities. Our integrated multi-modal analysis thus revealed significant associations between radiological, microbiological, therapeutic, and demographic data modalities, providing a deeper understanding of personalized responses to aid in the clinical management of TB.

Tuberculosis Disease Diagnosis Based on an Optimized Machine Learning Model

Computer science plays an important role in modern dynamic health systems. Given the collaborative nature of the diagnostic process, computer technology provides important services to healthcare professionals and organizations, as well as to patients and their families, researchers, and decision-makers. Thus, any innovations that improve the diagnostic process while maintaining quality and safety are crucial to the development of the healthcare field. Many diseases can be tentatively diagnosed during their initial stages. In this study, all developed techniques were applied to tuberculosis (TB). Thus, we propose an optimized machine learning-based model that extracts optimal texture features from TB-related images and selects the hyper-parameters of the classifiers. Increasing the accuracy rate and minimizing the number of characteristics extracted are our goals. In other words, this is a multitask optimization issue. A genetic algorithm (GA) is used to choose the best features, which are then fed into a support vector machine (SVM) classifier. Using the ImageCLEF 2020 data set, we conducted experiments using the proposed approach and achieved significantly higher accuracy and better outcomes in comparison with the state-of-the-art works. The obtained experimental results highlight the efficiency of modified SVM classifier compared with other standard ones.

Integrating landmark modeling framework and machine learning algorithms for dynamic prediction of tuberculosis treatment outcomes

OBJECTIVE

This study aims to establish an informative dynamic prediction model of treatment outcomes using follow-up records of tuberculosis (TB) patients, which can timely detect cases when the current treatment plan may not be effective.

MATERIALS AND METHODS

We used 122 267 follow-up records from 17 958 new cases of pulmonary TB in the Republic of Moldova. A dynamic prediction framework integrating landmark modeling and machine learning algorithms was designed to predict patient outcomes during the course of treatment. Sensitivity and positive predictive value (PPV) were calculated to evaluate performance of the model at critical time points. New measures were defined to determine when follow-up laboratory tests should be conducted to obtain most informative results.

RESULTS

The random-forest algorithm performed better than support vector machine and penalized multinomial logistic regression models for predicting TB treatment outcomes. For all 3 outcome classes (ie, cured, not cured, and died after 24 months following treatment initiation), sensitivity and PPV of prediction models improved as more follow-up information was collected. Specifically, sensitivity and PPV increased from 0.55 to 0.84 and from 0.32 to 0.88, respectively, for the not cured class.

CONCLUSION

The dynamic prediction framework utilizes longitudinal laboratory test results to predict patient outcomes at various landmarks. Sputum culture and smear results are among the important variables for prediction; however, the most recent sputum result is not always the most informative one. This framework can potentially facilitate a more effective treatment monitoring program and provide insights for policymakers toward improved guidelines on follow-up tests.

Empiric Addition of Quinolones to First-Line Tuberculosis Treatment Is Associated With Increased Odds of XDR-TB

Background

Multidrug-resistant tuberculosis (MDR-TB) represents a significant clinical and public health challenge worldwide. Out of concern for possible resistance, some providers prescribe first- and second-line tuberculosis treatment together before completing drug susceptibility testing (DST), which may increase emergent resistance.

Methods

MDR-TB patients at an Indian referral center were enrolled in an observational cohort. Participants with drug susceptibility test (DST) results were categorized as prescribed fluoroquinolones, streptomycin, both, or neither with first-line treatment before DST. Odds of additional resistance to fluoroquinolones and aminoglycosides (XDR-TB) were calculated in association with empiric combined first- and second-line treatment before DST.

Results

Of 494 participants, 130 (26.3%) received a fluoroquinolone or streptomycin with first-line drugs before DST. Odds of XDR-TB were associated with fluoroquinolone prescription before DST [odds ratio (OR): 2.19, 95% confidence interval (CI): 1.26–3.76). The association with XDR-TB persisted in multivariable analysis (adjusted OR: 2.43, 95% CI: 1.19-4.91). Combined empiric first- and second-line treatment before DST was not associated with eventual outcomes.

Conclusion

Many participants received empiric combined first- and second-line drugs before DST, which was associated with XDR-TB. To minimize emerging resistance, treatment-associated side effects, and provide the best possible care, this approach should be discouraged in favor of early DST and DST-guided TB treatment.

Social and health factors associated with unfavourable treatment outcome in adolescents and young adults with tuberculosis in Brazil: a national retrospective cohort study

BACKGROUND

Tuberculosis elimination strategies in Brazil might neglect adolescents and young adults aged 10-24 years, hampering tuberculosis control. However, little is known about factors associated with tuberculosis treatment outcomes in this underserved group. In this study, we aimed to investigate social and health factors associated with unfavourable treatment outcomes in young people with tuberculosis in Brazil.

METHODS

A national retrospective cohort study was done using data from Sistema de Informação de Agravos de Notificação (SINAN), the national tuberculosis registry in Brazil. People aged 10-24 years (young people) with tuberculosis registered in SINAN between Jan 1, 2015, and Dec 31, 2018, were included. Unfavourable outcomes were defined as loss to follow-up, treatment failure, and death. Favourable outcome was defined as treatment success. Multiple logistic regression models estimated the association between social and health factors and tuberculosis treatment outcomes.

FINDINGS

67 360 young people with tuberculosis were notified to SINAN, and we included 41 870 young people in our study. 7024 (17%) of the 41 870 included individuals had unfavourable treatment outcomes. Young people who received government cash transfers were less likely to have an unfavourable outcome (adjusted odds ratio 0·83, 95% CI 0·70-0·99). Homelessness (3·03, 2·07-4·42), HIV (2·89, 2·45-3·40), and illicit drug use (2·22, 1·93-2·55) were the main factors associated with unfavourable treatment outcome.

INTERPRETATION

In this national cohort of young people with tuberculosis in Brazil, tuberculosis treatment success rates were lower than WHO End TB Strategy targets, with almost a fifth of participants experiencing unfavourable treatment outcomes. Homelessness, HIV, and illicit drug use were the main factors associated with unfavourable outcome. In Brazil, strategies are required to support this underserved group to ensure favourable tuberculosis treatment outcomes.

FUNDING

Wellcome Trust, UK Medical Research Council, and UK Foreign Commonwealth and Development Office.

Impact of COVID-19 on Tuberculosis Case Detection and Treatment Outcomes in Sierra Leone

The COVID-19 pandemic has adversely affected tuberculosis (TB) care delivery in high burden countries. We therefore conducted a retrospective study to assess the impact of COVID-19 on TB case detection and treatment outcomes at the Chest Clinic at Connaught Hospital in Freetown, Sierra Leone. Overall, 2300 presumptive cases were tested during the first three quarters of 2020 (intra-COVID-19) versus 2636 in 2019 (baseline), representing a 12.7% decline. Testing declined by 25% in women, 20% in children and 81% in community-initiated referrals. Notwithstanding, laboratory-confirmed TB cases increased by 37.0% and treatment success rate was higher in 2020 (55.6% vs. 46.7%, p = 0.002). Multivariate logistic regression analysis found that age < 55 years (aOR 1.74, 95% CI (1.80, 2.56); p = 0.005), new diagnosis (aOR 1.69, 95% CI (1.16, 2.47); p = 0.007), pulmonary TB (aOR 3.17, 95% CI (1.67, 6.04); p < 0.001), HIV negative status (aOR 1.60, 95%CI (1.24, 2.06); p < 0.001) and self-administration of anti-TB drugs through monthly dispensing versus directly observed therapy (DOT) (aOR 1.56, 95% CI (1.21, 2.03); p = 0.001) independently predicted treatment success. These findings may have policy implications for DOTS in this setting and suggest that more resources are needed to reverse the negative impact of the COVID-19 pandemic on TB program activities in Sierra Leone.

Development of a nomogram for predicting treatment default under facility-based directly observed therapy short-course in a region with a high tuberculosis burden

Background:

Poor adherence to tuberculosis (TB) treatment is a substantial barrier to global TB control. The aim of this study was to construct a nomogram for predicting the probability of TB treatment default.

Methods:

A total of 1185 TB patients who had received treatment between 2010 and 2011 in Peru were analyzed in this study. Patient demographics, social, and medical information were recorded. Predictors were selected by least absolute shrinkage and selection operator (LASSO) regression analysis, and a nomogram for predicting TB treatment default was constructed by using multivariable logistic regression analysis. Bootstrapping method was applied for internal validation. Calibration and clinical utility of the nomogram was also evaluated.

Results:

The incidence of TB treatment default among the study patients was 11.6% (138/1185). Six predictors (secondary education status, alcohol use, illegal drug use, body mass index, multidrug-resistant tuberculosis, and human immunodeficiency virus serostatus) were selected through the LASSO regression analysis. A nomogram was developed based on the six predictors and it yielded an area under the curve (AUC) value of 0.797 [95% confidence interval (CI), 0.755–0.839]. In the internal validation, the AUC achieved 0.805 (95% CI, 0.759–0.844). Additionally, the nomogram was well-calibrated, and it showed clinical utility in decision curve analysis.

Conclusion:

A nomogram was constructed that incorporates six characteristics of the TB patients, which provides a good reference for predicting TB treatment default.

A clinical prediction model for unsuccessful pulmonary tuberculosis treatment outcomes

BACKGROUND

Despite widespread availability of curative therapy, tuberculosis treatment outcomes remain suboptimal. Clinical prediction models can inform treatment strategies to improve outcomes. Using baseline clinical data, we developed a prediction model for unsuccessful TB treatment outcome and evaluated the incremental value of HIV-related severity and isoniazid acetylator status.

METHODS

Data originated from the Regional Prospective Observational Research for Tuberculosis Brazil cohort, which enrolled newly-diagnosed tuberculosis patients in Brazil from 2015-2019. This analysis included participants with culture-confirmed, drug-susceptible pulmonary tuberculosis who started first-line anti-tuberculosis therapy and had ≥12 months of follow-up. The endpoint was unsuccessful tuberculosis treatment: composite of death, treatment failure, regimen switch, incomplete treatment, or not evaluated. Missing predictors were imputed. Predictors were chosen via bootstrapped backward selection. Discrimination and calibration were evaluated with c-statistics and calibration plots, respectively. Bootstrap internal validation estimated overfitting, and a shrinkage factor was applied to improve out-of-sample prediction. Incremental value was evaluated with likelihood ratio-based measures.

RESULTS

Of 944 participants, 191 (20%) had unsuccessful treatment outcomes. The final model included seven baseline predictors: hemoglobin, HIV-infection, drug use, diabetes, age, education, and tobacco use. The model demonstrated good discrimination (c-statistic=0.77; 95% confidence interval: 0.73-0.80) and was well-calibrated (optimism-corrected intercept and slope: -0.12 and 0.89, respectively). HIV-related factors and isoniazid acetylation status did not improve prediction of the final model.

CONCLUSIONS

The prediction model, using information readily available at treatment initiation, performed well in this population. The findings may guide future work to allocate resources or inform targeted interventions for high-risk patients.

Benchmarking Machine Learning Models to Assist in the Prognosis of Tuberculosis

Tuberculosis (TB) is an airborne infectious disease caused by organisms in the Mycobacterium tuberculosis (Mtb) complex. In many low and middle-income countries, TB remains a major cause of morbidity and mortality. Once a patient has been diagnosed with TB, it is critical that healthcare workers make the most appropriate treatment decision given the individual conditions of the patient and the likely course of the disease based on medical experience. Depending on the prognosis, delayed or inappropriate treatment can result in unsatisfactory results including the exacerbation of clinical symptoms, poor quality of life, and increased risk of death. This work benchmarks machine learning models to aid TB prognosis using a Brazilian health database of confirmed cases and deaths related to TB in the State of Amazonas. The goal is to predict the probability of death by TB thus aiding the prognosis of TB and associated treatment decision making process. In its original form, the data set comprised 36,228 records and 130 fields but suffered from missing, incomplete, or incorrect data. Following data cleaning and preprocessing, a revised data set was generated comprising 24,015 records and 38 fields, including 22,876 reported cured TB patients and 1139 deaths by TB. To explore how the data imbalance impacts model performance, two controlled experiments were designed using (1) imbalanced and (2) balanced data sets. The best result is achieved by the Gradient Boosting (GB) model using the balanced data set to predict TB-mortality, and the ensemble model composed by the Random Forest (RF), GB and Multi-Layer Perceptron (MLP) models is the best model to predict the cure class.