Decision Tree Based Diagnostic System for Moderate to Severe Obstructive Sleep Apnea

Abstract

One of the major modern medical issues, obstructive sleep apnea (OSA), particularly at moderate to severe levels, may potentially cause cardiovascular morbidity and mortality. However, polysomnography (PSG), a gold standard tool in diagnosing OSA, is cumbersome, has limited availability, and is costly and time-consuming. Clinical prediction models thus are absolutely necessary in screening patients with OSA. Furthermore, the performance of the published prediction formulas is not satisfactory for Chinese populations. The aim of this study was to develop and validate a simple and accurate prediction system for the diagnosis of moderate to severe OSA by integrating an expert-based feature extraction technique with decision tree algorithms which have automatic feature selection capability in screening the moderate to severe OSA cases in Taiwan. Moreover, the backward stepwise multivariable logistic regression model and four other decision tree algorithms were also employed for comparison. The results showed that the proposed best prediction formula, with an overall accuracy reaching to 96.9 % in sensitivity = 98.2 % and specificity = 93.2 %, could present a good tool for screening moderate and severe Taiwanese OSA patients who require further PSG evaluation and medical intervention. Results also indicate that the proposed best prediction formula is simple, accurate, and reliable, and outperforms all the other prediction formulae considered in the present study. The proposed clinical prediction formula derived from three non-invasive features (Sex, Age, and AveSBP) may help prioritize patients for PSG studies as well as avoid a diagnosis of PSG in subjects who have a low probability of having the disease.

Predicting Long-Term Outcome After Traumatic Brain Injury Using Repeated Measurements of Glasgow Coma Scale and Data Mining Methods

“…This knowledge discovery methodology has been increasingly used in medical studies to explore the patterns or relationships among medical variables and make predictions about unknown values at the individual level by mining historical medical data [28][29][30][31]. Some predictive data mining methods have been introduced to predict prognostic outcomes of patients with TBI and are potentially more accurate than conventional statistical methods [7,9,32,33].…”

Platelets, from sample to big data

“…Also, 5-FU is known to produce a significant reduction in platelet aggregation and platelet factor-3 (PF(3)) availability without being associated with thrombocytopenia [40]. 5-FU does contribute to a potentially pro-thrombotic environment through the depletion of protein C and increased thrombin activity [41].…”

“…(Table 1) The Spearman correlation is nonparametric with the exact sampling distribution obtained without requiring knowledge of the joint probability distribution of the parameters. Correlation analysis using Spearman's correlation coefficients (Table 1) [35][36][37][38][39][40][41][42][43][44][45]. PLTEM, was calculated by subtracting FIBTEM from EXTEM to distinguish thrombocytopenia from hypofibrinogenemia [8].…”

“…The procedure is recursively iterated, until a stop criterion is met. Greedy strategy for DT building is often criticized, but the ability to build strait forward and the highly interpretable rules on massive data, led to many successful applications in medical applications [39,40]. In this research we used J48 algorithms which is the Java implementation of the C4.5 algorithm [41].…”

“…). The chilling-induced (0°C, ice water), reversible shape-change in platelets correlates with phosphorylation of myosin, subsequent interaction on actin filaments and free cytosolic calcium increase [40]. Human platelets can be maintained in a discoid shape in the cold, in vitro, using a cell-permeable calcium chelator to attenuate calcium mobilization and cytochalasin B to prevent barbed-end actin assembly [39].…”