Predicting healthcare costs with diagnoses recorded in primary and secondary care: an analysis of linked records

Abstract

Most risk-adjustment models rely on diagnoses recorded during previous contacts in the same care setting to predict service use and cost. When diagnostic information from multiple settings has been used, studies have not examined how diagnoses recorded in different care settings influence model performance. Using a single set of diagnostic indicators recorded in primary or secondary care can incentivise case-finding and treatment outside hospital, but may reduce model fit if secondary care diagnosis indicates higher levels of severity. Using linked primary and secondary care records for 12.8 million patients in England, we used 205 chronic conditions recorded in primary care to complement those recorded during recent hospital admissions. We examined predictions of hospital use and cost for different population groups and considered the related incentives and implications for efficiency and fairness. Most patients (56 %) had at least one condition ever recorded in primary care, while only 15 % had at least one recorded in secondary care in the previous two years. Adding diagnoses recorded only in primary care as a separate additional set of predictors improved the model fit for total costs, planned and unplanned costs, elective and emergency admissions, outpatient visits, and emergency department attendances. Using a single set of diagnoses recorded in either setting did not improve model fit, except for outpatient visits. Including primary care diagnoses reduced under and over-compensation and increased the predicted service needs of younger patients in less deprived areas and older patients in more deprived areas.