• Adriana C Diamantino, Bruno R Nascimento, Maria Carmo P Nunes, Craig A Sable, Kaciane K B Oliveira, Lara C Rabelo, Juliane Franco, Luciana C Diamantino, Marcia M Barbosa, Alison T Reese, Laura Olivieri, de Lima Emilly Malveira EM Department of Statistics, Instituto de Ciência Exatas, Universidade Federal de Minas Gerais, Belo Horizonte, Brazil., Larissa Natany A Martins, Enrico Antonio Colosimo, Andrea Z Beaton, Antonio L P Ribeiro, and PROVAR+ (Programa de RastreamentO da VAlvopatia Reumática e outras Doenças Cardiovasculares) investigators.
• Serviço de Cardiologia e Cirurgia Cardiovascular e Centro de Telessaúde do Hospital das Clínicas da UFMG, Belo Horizonte, Brazil.
• Int. J. Clin. Pract. 2021 Mar 1; 75 (3): e13686.

IntroductionAccess to public subspecialty healthcare is limited in underserved areas of Brazil, including echocardiography (echo). Long waiting lines and lack of a prioritisation system lead to diagnostic lag and may contribute to poor outcomes. We developed a prioritisation tool for use in primary care, aimed at improving resource utilisation, by predicting those at highest risk of having an abnormal echo, and thus in highest need of referral.MethodsAll patients in the existing primary care waiting list for echo were invited for participation and underwent a clinical questionnaire, simplified 7-view echo screening by non-physicians with handheld devices, and standard echo by experts. Two derivation models were developed, one including only clinical variables and a second including clinical variables and findings of major heart disease (HD) on echo screening (cut point for high/low-risk). For validation, patients were risk-classified according to the clinical score. High-risk patients and a sample of low-risk underwent standard echo. Intermediate-risk patients first had screening echo, with a standard echo if HD was suspected. Discrimination and calibration of the two models were assessed to predict HD in standard echo.ResultsIn derivation (N = 603), clinical variables associated with HD were female gender, body mass index, Chagas disease, prior cardiac surgery, coronary disease, valve disease, hypertension and heart failure, and this model was well calibrated with C-statistic = 0.781. Performance was improved with the addition of echo screening, with C-statistic = 0.871 after cross-validation. For validation (N = 1526), 227 (14.9%) patients were classified as low risk, 1082 (70.9%) as intermediate risk and 217 (14.2%) as high risk by the clinical model. The final model with two categories had high sensitivity (99%) and negative predictive value (97%) for HD in standard echo. Model performance was good with C-statistic = 0.720.ConclusionThe addition of screening echo to clinical variables significantly improves the performance of a score to predict major HD.© 2020 John Wiley & Sons Ltd.

9 keywords...

hide…