Imaina Widagdo, Mhairi Kerr, Lisa Kalisch Ellett, Clement Schlegel, Elham Sadeqzadeh, Alvin Wang, Allison Louise Clarke, Nicole Pratt
Purpose: Identifying patient health conditions in observational studies is essential for accurately measuring healthcare practices and planning effective health policy interventions. This analysis evaluates the validity of the Rx-Risk Index, a tool that uses medication dispensing data to identify patient comorbidities and measure overall health. We examined an updated version of the Rx-Risk Index, reflecting changes in treatment practices, to assess its validity as a tool for identifying specific health conditions and as a measure of overall health to aid in risk adjustment in observational studies.
Patients and Methods: We conducted a validation study using two Australian linked health datasets, the Person-Level Integrated Data Asset (PLIDA) and the National Health Data Hub (NHDH), from 2010 to 2018, focusing on individuals aged 65 years or older. The sensitivity, specificity, PPV/NPV, Cohen's kappa, and F1 scores were used to assess agreement between Rx-Risk Index conditions and two reference standards: patient self-reported conditions and hospital diagnosis. The Rx-Risk Index's predictive validity for one-year mortality was also evaluated using logistic regression, with model fit assessed by AIC and c-statistic.
Results: Data were analysed from 3,959 individuals in PLIDA and 157,709 individuals in NHDH. The Rx-Risk Index showed high sensitivity (���75%) for diabetes, chronic airways disease, hyperlipidemia, and epilepsy against both self-reported conditions and hospital diagnoses. However, hyperlipidemia and hypertension showed lower specificity (<70%). High PPVs (���78%) were observed for diabetes and renal failure. The agreement between the Rx-Risk Index and self-reported conditions was stronger (Cohen's kappa: 0.41-0.81 for 7 conditions) than between Rx-Risk Index and ICD10-AM diagnoses (kappa: 0.73 for one condition). The Rx-Risk Index was a strong predictor of one-year mortality, with c-statistic of 0.820 (95% CI: 0.817-0.825).
Conclusion: Selected Rx-Risk Index conditions are reasonable proxies for identifying specific conditions, particularly those requiring pharmacological management. The Rx-Risk Index was a strong predictor of one-year mortality, suggesting it is a valid measure of overall health. This study demonstrates the Rx-Risk Index's potential to enhance disease classification and risk adjustment in observational studies, supporting informed decision-making in health policy planning.
