SSRN Electronic Journal, Journal Year: 2021, Volume and Issue: unknown
Published: Jan. 1, 2021
Background: The new coronavirus pandemic emphasized the relevance of initial screening patients, due to emerging need for its results. Predictive diagnosis combining artificial and geospatial intelligence can accelerate prediction COVID-19 tests Several studies used only clinical variables in predictive modeling, disregarding socioeconomic aspects, human development social vulnerabilities, than be as markers probability a patient presenting positive results SARS-CoV-2. Those modeling done by machine learning models subsidize management public health policies.Objective: objective present work is apply an automated mechanism, order test first those with high result among samples, complaints socioeconomics scores, without taking into consideration exams.Methods: Data from patients who underwent polymerase chain reaction (PCR) testing SARS-CoV-2 were collected between July September 2020, symptomatology onset under 40 days prior testing. Patient registration data, data collected, such age, sex, address, symptoms onset, index (HDI) vulnerability (SVI). final sample consisted 3743 (lines) 231 (columns). using autoML H2O framework, different techniques balancing inputs. performance was measured AUCROC curves. generated all and, then, ten most relevant ones.Results: best version itself variables, resulting model Stacked Ensemble (Training: AUC=0.872, Validation: AUC=0.617 Testing: AUC=0,627). individual model, ranked relevance, are, respectively, ageusia, fever, myalgia, % people aged 15 24 years old don't study or have per capita household income equal less half minimum wage (from 2010), index, ageing rate, inadequate householding water supply sanitation os children 0 5 that attend school. Among 10 are socioeconomic, which demonstrates diagnosis.Conclusions: achieved indicate there may benefits bringing together inpatient basic assistance. Exploring optimized detection, we achieve better, faster decision making policies distribution, especially facing variants considering unknown events, like pandemics endemics.
Language: Английский
