How well can electronic health records from primary care identify Alzheimer's disease cases?

• Published in: Clinical epidemiology Vol. 11; pp. 509 - 518
• Main Authors: Ponjoan, Anna, Garre-Olmo, Josep, Blanch, Jordi, Fages, Ester, Alves-Cabratosa, Lia, Martí-Lluch, Ruth, Comas-Cufí, Marc, Parramon, Dídac, García-Gil, María, Ramos, Rafel
• Format: Journal Article
• Language: English
• Published: New Zealand Dove Medical Press Limited 01.07.2019; Taylor & Francis Ltd; Dove; Dove Medical Press
• Subjects: Accuracy; Advertising executives; Aging; algorithm; Algorithms; Alzheimer's disease; Clinical medicine; data accuracy; Dementia; Electronic health records; electronic medical records; Electronic records; Epidemiology; Family; Family medicine; family physician; Family physicians; General practitioners; Health aspects; Health services; Health surveillance; Information systems; Medical care quality; Medical records; Medical research; Original Research; Physicians; Population; Primary care; Public health; real world data; survey; validation; United Kingdom; Spain; electronic medical records; family physician; dementia; survey; real-world data; data accuracy; validation; algorithm
• ISSN: 1179-1349; 1179-1349
• DOI: 10.2147/CLEP.S206770

Electronic health records (EHR) from primary care are emerging in Alzheimer's disease (AD) research, but their accuracy is a concern. We aimed to validate AD diagnoses from primary care using additional information provided by general practitioners (GPs), and a register of dementias. This retrospective observational study obtained data from the System for the Development of Research in Primary Care (SIDIAP). Three algorithms combined International Statistical Classification of Diseases (ICD-10) and Anatomical Therapeutic Chemical codes to identify AD cases in SIDIAP. GPs evaluated dementia diagnoses by means of an online survey. We linked data from the Register of Dementias of Girona and from SIDIAP. We estimated the positive predictive value (PPV) and sensitivity and provided results stratified by age, sex and severity. Using survey data from the GPs, PPV of AD diagnosis was 89.8% (95% CI: 84.7-94.9). Using the dataset linkage, PPV was 74.8 (95% CI: 73.1-76.4) for algorithm A1 (AD diagnoses), and 72.3 (95% CI: 70.7-73.9) for algorithm A3 (diagnosed or treated patients without previous conditions); sensitivity was 71.4 (95% CI: 69.6-73.0) and 83.3 (95% CI: 81.8-84.6) for algorithms A1 (AD diagnoses) and A3, respectively. Stratified results did not differ by age, but PPV and sensitivity estimates decreased amongst men and severe patients, respectively. PPV estimates differed depending on the gold standard. The development of algorithms integrating diagnoses and treatment of dementia improved the AD case ascertainment. PPV and sensitivity estimates were high and indicated that AD codes recorded in a large primary care database were sufficiently accurate for research purposes.