Prioritizing patients for stomach cancer screening programs: a machine learning approach

Abstract

Stomach cancer ranks fifth in incidence and is the fourth cause of death by cancer in the world. Since usually this disease is asymptomatic or the symptoms are shared with other diseases, it is diagnosed when the probabilities of recovery are low or null. In this context, performing endoscopy screenings and biopsy follow-ups during early stages could allow the detection of stomach cancer when the patient has a higher probability of recovery. Hence, a proper prioritizing of patients can make feasible the implementation of endoscopy screening programs. This work presents a Decision Support System (DSS) to support the prioritization of patients for endoscopy screening programs. For this purpose, we use the information available in the national healthcare system of Colombia (Sistema General de Seguridad Social en Salud, SGSSS). Our contribution to literature is twofold. First, we identify variables that explain the probability of being diagnosed with stomach cancer, including clinical pathways modeled from a Process Mining approach. Second, we assess the effectiveness of two machine learning approaches for classifying patients and their performance in terms of coverage. Our results show a feasible way to design prevention programs for patient prioritization in a cost-effective approach.