Purpose: To obtain population-based estimates of direct healthcare costs for prostate cancer (PC) from diagnosis to death for health states in a state-transition model.
Method: PC patients, diagnosed in 1993, 1994, 1997, 1998, 2001, and 2002, and residing in three regions of Ontario, Canada, were selected from the Ontario Cancer Registry. We retrieved pathology reports to identify patient name, referring physician, and tumour information. With consent from referring physicians, we contacted patients and family of dead patients for consent to review charts. We visited physicians’ clinics and hospitals and reviewed charts to obtain data describing PC diagnosis, treatment, and outcome. We developed clinical criteria to allocate each patient’s observation time to 11 PC-specific Markov health states, including active surveillance, treatments, follow-up, recurrence, metastases, and death. We linked these data to health care administrative databases to calculate healthcare resource use and costs per health state, using previously developed costing methods. Mixed model multivariable regression determined predictors of costs. To assess model validity, we compared predicted costs estimated from the model with actual costs using the root mean square error and mean average error.
Result: The final sample numbered 829 patients (mean age = 67 years). Over 50% had T2 to T4 disease, and 5% were metastatic at diagnosis. The most costly primary treatment was radical prostatectomy ($4,702 per 100 days). The least costly health state was post-prostatectomy ($731 per 100 days). Costs before death and for hormone-refractory metastatic disease were high at $11,008 and $6,324 per 100 days, respectively. Costs increased with age (p<0.001), comorbidity (p<0.001), and advanced PC at diagnosis (p<0.05). Radical prostatectomy, metastatic disease, and final (pre-death) health states were significantly more costly than active surveillance (p<0.05), while post-prostatectomy and post-radiation therapy states cost significantly less (p<0.0001). The validity of the model was assessed; the root mean square error was $4,206 and mean average error was $1,873, relatively small compared with observed mean and median costs per 100 days, $4,344 and $2,338, respectively.
Conclusion: Combining chart reviews and administrative data is feasible to estimate mean adjusted costs and the effects of covariates on costs for state-transition models. However, this approach is very costly and time consuming. Administrative data alone may be sufficient for applications that do not require a high level of clinical detail.