Predicting hospital costs for patients receiving renal replacement therapy to inform an economic evaluation.

OBJECTIVE: To develop a model to predict annual hospital costs for patients with established renal failure, taking into account the effect of patient and treatment characteristics of potential relevance for conducting an economic evaluation, such as age, comorbidities and time on treatment. The analysis focuses on factors leading to variations in inpatient and outpatient costs and excludes fixed costs associated with dialysis, transplant surgery and high cost drugs.
METHODS: Annual costs of inpatient and outpatient hospital episodes for patients starting renal replacement therapy in England were obtained from a large retrospective dataset. Multiple imputation was performed to estimate missing costs due to administrative censoring. Two-part models were developed using logistic regression to first predict the probability of incurring any hospital costs before fitting generalised linear models to estimate the level of cost in patients with positive costs. Separate models were developed to predict inpatient and outpatient costs for each treatment modality.
RESULTS: Data on hospital costs were available for 15,869 incident dialysis patients and 4511 incident transplant patients. The two-part models showed a decreasing trend in costs with increasing number of years on treatment, with the exception of dialysis outpatient costs. Age did not have a consistent effect on hospital costs; however, comorbidities such as diabetes and peripheral vascular disease were strong predictors of higher hospital costs in all four models.
CONCLUSION: Analysis of patient-level data can result in a deeper understanding of factors associated with variations in hospital costs and can improve the accuracy with which costs are estimated in the context of economic evaluations.