Feature selection and prediction of treatment failure in tuberculosis

Abstract

BackgroundTuberculosis is a major cause of morbidity and mortality in the developing world. Drug resistance, which is predicted to rise in many countries worldwide, threatens tuberculosis treatment and control.ObjectiveTo identify features associated with treatment failure and to predict which patients are at highest risk of treatment failure.MethodsOn a multi-country dataset managed by the National Institute of Allergy and Infectious Diseases we applied various machine learning techniques to identify factors statistically associated with treatment failure and to predict treatment failure based on baseline demographic and clinical characteristics alone.ResultsThe complete-case analysis database consisted of 587 patients (68% males) with a median (p25-p75) age of 40 (30–51) years. Treatment failure occurred in approximately one fourth of the patients. The features most associated with treatment failure were patterns of drug sensitivity, imaging findings, findings in the microscopy Ziehl-Nielsen stain, education status, and employment status. The most predictive model was forward stepwise selection (AUC: 0.74), although most models performed at or above AUC 0.7. A sensitivity analysis using the 643 original patients filling the missing values with multiple imputation showed similar predictive features and generally increased predictive performance.ConclusionMachine learning can help to identify patients at higher risk of treatment failure. Closer monitoring of these patients may decrease treatment failure rates and prevent emergence of antibiotic resistance. The use of inexpensive basic demographic and clinical features makes this approach attractive in low and middle-income countries.

Prediction of treatment failure and compliance in patients with tuberculosis

“…The first model for predicting treatment failure by Kalhori et al [14] used clinical data including old age, male sex, body weight, nationality, prisoner status, and previous history of TB, and achieved an AUC of 0.70. Recently, Sauer et al [15] tried to predict treatment failure by machine learning using demographic and laboratory data and reported a best AUC of 0.74. However, this model lacked information about comorbidities; our model included such variables and yielded considerably high prediction power, an AUC of 0.79.…”

FLOating-Window Projective Separator (FloWPS): A Data Trimming Tool for Support Vector Machines (SVM) to Improve Robustness of the Classifier

“…It was seen previously that SVM sometimes fails when it is intended for distinguishing fine biomedical properties such as disease progression prognosis or assessment of clinical efficiency of drugs for an individual patient, using high throughput molecular data, e.g., complete DNA mutation or gene expression profiles (Ray and Zhang, 2009; Babaoglu et al, 2010). Particularly, for many biologically relevant applications, SVM occurred either fully incapable to predict drug sensitivity (Turki and Wei, 2016), or demonstrated poorer performance than competing method for machine learning (Davoudi et al, 2017; Cho et al, 2018; Jeong et al, 2018; Leite et al, 2018; Sauer et al, 2018; Yosipof et al, 2018). Thus, the tool for improvement of SVM performance is certainly needed.…”

Prediction of Treatment Failure and Compliance in Patients with Tuberculosis

“…The rst model for predicting treatment failure by Kalhori et al (15) used clinical data including old age, male sex, body weight, nationality, prisoner status, and previous history of TB, and achieved an AUC of 0.70. Recently, Sauer et al (16) tried to predict treatment failure by machine learning using demographic and laboratory data and reported a best AUC of 0.74. However, this model lacked information about comorbidities; our model included such variables and yielded considerably high prediction power, an AUC of 0.79.…”