Keep it Simple? Predicting Primary Health Care Costs with Measures of Morbidity and Multimorbidity

In this paper we investigate the relationship between patients’ primary care costs (consultations, tests, drugs) and their age, gender, deprivation and alternative measures of their morbidity and multimorbidity. Such information is required in order to set capitation fees or budgets for general practices to cover their expenditure on providing primary care services. It is also useful to examine whether practices’ expenditure decisions vary equitably with patient characteristics. Electronic practice record keeping systems mean that there is very rich information on patient diagnoses. But the diagnostic information (with over 9000 possible diagnoses) is too detailed to be practicable for setting capitation fees or practice budgets. Some method of summarizing such information into more manageable measures of morbidity is required. We therefore compared the ability of eight measures of patient morbidity and multimorbidity to predict future primary care costs using data on 86,100 individuals in 174 English practices. The measures were derived from four morbidity descriptive systems (17 chronic diseases in the Quality and Outcomes Framework (QOF), 17 chronic diseases in the Charlson scheme, 114 Expanded Diagnosis Clusters (EDCs), and 68 Adjusted Clinical Groups (ACGs)). We found that, in general, for a given disease description system, counts of diseases and sets of disease dummy variables had similar explanatory power and that measures with more categories did better than those with fewer. The EDC measures performed best, followed by the QOF and ACG measures. The Charlson measures had the worst performance but still improved markedly on models containing only age, gender, deprivation and practice effects. Allowing for individual patient morbidity greatly reduced the association of age and cost. There was a pro-deprived bias in expenditure: after allowing for morbidity, patients in areas in the highest deprivation decile had costs which were 22% higher than those in the lowest deprivation decile. The predictive ability of the best performing morbidity and multimorbidity measures was very good for this type of individual level cross section data, with R2 ranging from 0.31 to 0.46. The statistical method of estimating the relationship between patient characteristics and costs was less important than the type of morbidity measure. Rankings of the morbidity and multimorbidity measures were broadly similar for generalised linear models with log link and Poisson errors and for OLS estimation. It would be currently feasible to combine the results from our study with the data on the number of patients with each QOF disease, which is available on all practices in England, to calculate budgets for general practices to cover their primary care costs.