Machine learning e análise preditiva em saúde: um estudo de caso sobre detecção de anomalias em contas médicas do Exército

• Ano de defesa: 2022
• Autor(a) principal: Vitoria, Sergio Ricardo Pacheco da
• Orientador(a): Paiva, Matheus S. de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Católica de Brasília
• Programa de Pós-Graduação: Programa Stricto Sensu em Governança, Tecnologia e Inovação
• Departamento: Escola de Educação, Tecnologia e Comunicação
• País: Brasil
• Palavras-chave em Português: Contas médicas; SAMMED; Auditoria
• Palavras-chave em Inglês: Machine Learnng; Medical bills; Isolation forest; DBSCAN; Auditing
• Área do conhecimento CNPq: CNPQ::CIENCIAS SOCIAIS APLICADAS
• Link de acesso: https://bdtd.ucb.br:8443/jspui/handle/tede/3198

Medical costs tend to increase more than inflation measured for the period, which has a direct impact on services provided and amounts transferred to beneficiaries of health plans. In addition to the growth in medical inflation, measured by the Medical-Hospital Cost Variation Index (VCMH), other concerns have plagued health plans, like fraud or inconsistencies in charges made by private or public service providers. The Brazilian Army has a health care plan (SAMMED), which is very similar to a private health plan, with: monthly contribution; outpatient and hospital coverage; nationwide coverage; 24-hour urgent and emergency service; exam coverage, etc. However, it is not regulated by the National Supplementary Health Agency (ANS). The responsibility for managing this assistance plan is the responsibility of the Health Directorate (D Sau) of the Army. The purpose of SAMMED is to serve the military, pensioners and their dependents, totaling about 730 thousand beneficiaries throughout the national territory. For this purpose, it uses as support a network formed by 62 military hospitals, approximately 3,000 contracted civil hospitals and 2,000 Autonomous Health Professionals. All external services are evaluated by the medical bills auditing section, with the objective of verifying the existence of some type of anomaly, whether intentional or not, if found to be in order, pay the invoice sent by the providers. Today in the Brazilian Army the audit process is done manually, there is no automated system to support the auditor for this purpose. Therefore, this research aimed to build a model capable of identifying anomalous points in medical bills, the result of external assistance to SAMMED beneficiaries in units accredited with the EB, taking into account values determined in contracts. To this end, this research proposed to apply and compare two machine learning algorithms, DBSCAN and ISALATION FOREST, in order to verify which best fits the correction of the study problem, in addition to refuting the hypothesis that new technologies may or may not assist auditors in this process of auditing medical accounts, for this purpose, the hypothetical/deductive method will be used, supported by bibliographic research on the subject. It was concluded in this study that the unsupervised model using the ISOLATION FOREST algorithm, in a database consisting of 4 (four) OCS/PSA and only one procedure obtained almost 100% of correct answers when finding points with values outside the normal curve. For future work, it is recommended to build a WEB application for auditors to make the necessary filters to improve the search capability of the model in question.