Clinical Epidemiology (Nov 2022)
Validation of Algorithms to Identify Acute Myocardial Infarction, Stroke, and Cardiovascular Death in German Health Insurance Data
Abstract
Katharina Platzbecker,1 Annemarie Voss,1 Jonas Reinold,1 Anne Elbrecht,2 Wolfgang Biewener,2 Daniel Prieto-Alhambra,3,4 Annika M Jödicke,3 Tania Schink1 1Department of Clinical Epidemiology, Leibniz Institute for Prevention Research and Epidemiology – BIPS, Bremen, Germany; 2Department of Epidemiological Methods and Etiological Research, Leibniz Institute for Prevention Research and Epidemiology – BIPS, Bremen, Germany; 3Pharmaco- and Device Epidemiology, Centre for Statistics in Medicine, Nuffield Department of Orthopaedics, Rheumatology and Musculoskeletal Sciences, University of Oxford, Oxford, UK; 4Department of Medical Informatics, Erasmus University Medical Center, Rotterdam, NetherlandsCorrespondence: Tania Schink, Department of Clinical Epidemiology, Leibniz Institute for Prevention Research and Epidemiology – BIPS, Achterstrasse 30, Bremen, 28359, Germany, Tel +4942121856865, Email [email protected]: Validation of outcomes allows measurement of and correction for potential misclassification and targeted adjustment of algorithms for case definition. The purpose of our study was to validate algorithms for identifying cases of acute myocardial infarction (AMI), stroke, and cardiovascular (CV) death using patient profiles, ie, chronological tabular summaries of relevant available information on a patient, extracted from pseudonymized German claims data.Patients and Methods: Based on the German Pharmacoepidemiological Research Database (GePaRD), 250 cases were randomly selected (50% males) for each outcome between 2016 and 2017 based on the inclusion criteria age ≥ 50 years and continuous insurance ≥ 1 year and applying the following algorithms: hospitalization with a main diagnosis of AMI (ICD-10-GM codes I21.- and I22.-) or stroke (I63, I61, I64) or death with a hospitalization in the 60 days before with a main diagnosis of CV disease. Patient profiles were built including (i) age and sex, (ii) hospitalizations incl. diagnoses, procedures, discharge reasons, (iii) outpatient diagnoses incl. diagnostic certainty, physician specialty, (iv) outpatient encounters, and (v) outpatient dispensings. Using adjudication criteria based on clinical guidelines and risk factors, two trained physicians independently classified cases as “certain”, “probable”, “unlikely” or “not assessable”. Positive predictive values (PPVs) were calculated as percentage of confirmed cases among all assessable cases.Results: For AMI, the overall PPV was 97.6% [95% confidence interval 94.8– 99.1]. The PPV for any stroke was 94.8% [91.3– 97.2] and higher for ischemic (98.3% [95.0– 99.6]) than for hemorrhagic stroke (86.5% [76.5– 93.3]). The PPV for CV death was 79.9% [74.4– 84.4]. It increased to 91.7% [87.2– 95.0] after excluding 32 cases with data insufficient for a decision.Conclusion: Algorithms based on hospital diagnoses can identify AMI, stroke, and CV death from German claims data with high PPV. This was the first study to show that German claims data contain information suitable for outcome validation.Keywords: claims data, algorithm validation, patient profiles, positive predictive value
