763Validity of algorithms for identifying five chronic conditions in MedicineInsight, Australian national primary care data

Abstract Background MedicineInsight is a database containing de-identified electronic health records (EHRs) from over 700 Australian general practices. To support the trust placed in analyses of MedicineInsight data, additional evidence regarding the accuracy of the data is needed. Methods This study measures the validity of algorithms available in MedicineInsight that identify patients with depression, anxiety, asthma, type 2 diabetes and osteoporosis. Fifty practices met eligibility criteria regarding patient load and location, five were randomly selected and four agreed to participate. Within each practice, 250 patients aged ≥ 40 years were randomly selected. This age restriction increased the prevalence of the evaluated conditions, thereby optimising statistical power. Trained staff review the full EHR for these patients, including progress notes and correspondence, which are not available in MedicineInsight because they may contain identifiable information. Results With data collection almost complete, the target sample size will not be attainable. Power calculations indicate the current sample of 479 should provide adequate precision. For each condition of interest, the sensitivity, specificity, positive predictive value and negative predictive value of the algorithm is calculated. The full EHR review is the gold standard against which the algorithms are benchmarked. Conclusions The findings will indicate whether these algorithms demonstrate adequate accuracy to be used for research and decision-making. Key messages This additional understanding regarding the accuracy of MedicineInsight data will facilitate the interpretation of analyses of MedicineInsight data and guide improvements to the algorithms.