Identifying CA patients from a large cross-database (EHR and institutional registry of amyloidosis) in Buenos Aires, Argentina

POSADAS MARTINEZ, MARIA LOURDES; ET AL

Congreso; XVII INTERNATIONAL SYMPOSIUM ON AMYLOIDOSIS (ISA); 2020

N. QUIROZ1, E.ROSSI2, E. NUCIFORA2, M.A. AGUIRRE2, D. GIUNTA2 , M. RISK1 & M. L POSADAS MARTINEZ2Introduction A rare, underdiagnosed disease with an estimated global incidence between 5-9 cases per million. The Italian Hospital of Buenos Aires has an institutional registry of Amyloidosis and an Electronic Medical Record system that functions as a repository of the patient's entire medical history from consultations, complementary studies, hospitalizations, drug purchases, etc. , the HCE is used to determine certain types of alerts about diseases other than the rare ones. There is limited information on the detection of cases of amyloidosis through HCE in Latin America.Objective is to design, develop and evaluate a predictive model to identify patients with cardiac amyloidosis.Design Dynamic prospective cohort of all adult patients older than 17 years of the Italian Hospital in the period 1/1/2018 to 12/31/2018. Splitting was used with 70/30% for training and testing of the model. Supervised predictive models were developed and evaluating machine learning tools. A treebag predictive model was used to identify incident cases with cardiac amyloidosis from hospital bases based on the HIBA population. Taking into account relevant characteristics extracted from the HCE such as age, sex, blood pressure, biomarkers, echocardiographic data, medical consultations. Model performance was evaluated with AUC and its c95% confidence intervalResultsOf 131 patients with complete data, with an incidence of amyloidosis was 14.5% (25/131). Two final models were evaluated. Model 1 included demographics, laboratory and imaging. The most important variables for the model were FEE, SISM, age, pl sm, carpal tunnel syndrome, tas, body mass index, sex with a performance with an AUC 0.75 (xx). Model 2 included demographics, laboratory, imaging and specialist appointments. The most important variables for the model were hematologist, SISM, bnp, traumatology, sexo withth a performance with an AUC 0.67