Methodological proposal for constructing a classifier algorithm in clinical diagnostics of diseases using Bayesian methods

Abstract

Abstract Objectives To develop a methodological proposal to build clinical classifiers using information about signs and symptoms reported by the patient in initial the consultation and laboratory test results. Methods The proposed methodology considers procedures typical of the Bayesian paradigm of statistics as predictive probabilities and the sequential use of the Bayes formula. Additionally, some procedures belonging to classical statistics, such as Youden’s index and ROC curves, are applied. The method assumes two possible scenarios; when the patient only reports the signs and symptoms and the physician does not have access to information from laboratory tests. The other one is when the physician, besides the patient’s information, knows the blood test results. The method is illustrated using data from patients diagnosed with dengue. Results The performance of the proposed method depends of the set of signs and symptoms and the laboratory tests considered by the doctor as good indicators of presence of the sick in the individual. Conclusions The classifier can be used as a screening tool in scenarios where there is no extensive experience treating sick individuals, or economic and social conditions do not allow laboratory methods or gold standard procedures to complete the diagnosis.