Análise multivariada de características clínicas de PET/MAH e níveis de expressão gênica e derivação de modelos de predição diagnóstica em pacientes infectados com o HTLV-1

Abstract

Human T-cell lymphotropic virus type 1 (HTLV-1)- associated myelopathy/tropical spastic paraparesis (HAM/TSP) is a debilitating condition resulting from inflammation of the nerve tissue of the spinal cord caused by the action of HTLV-1. The aim of the present study was to evaluate the classification of individuals infected with HTLV-1 and propose a clinical prediction model for the occurrence of HAM/TSP. A database composed of 63 infected individuals was used, 23 of whom were diagnosed with HAM/TSP using the criteria recommended by the World Health Organization. Functional predictors (ordinal variables), gene expression levels (continuous variables) and sex (demographic variable) were also used. A mixed principal component analysis was employed, followed by hierarchical cluster analysis to determine the allocation of individuals into groups in an unsupervised fashion and compare the results to the classifications defined by clinicians. Diagnostic prediction models were then derived based on penalized binary logistic regression, which is suitable when the sample size is small. The unsupervised analysis showed that the patients were arranged into three groups: patients with HAM/TSP, patients without HAM/TSP and an intermediate group composed of individuals with and without the disease. Two models were derived from the statistical modeling – one with a penalization criterion of 0.032 and another with a criterion of 0.1 (more extreme). Both models were evaluated by internal validation using 10-fold crossvalidation. The variables that composed the final models were degree of gait alteration, derived Tinetti score, left and right adductor muscle tone and left triceps surae muscle tone. Statistical prediction methods may constitute a useful tool to support the diagnoses of HAM/TSP, especially in settings with limited resources.