Methods: The design of our prostate cancer policy model includes: I) estimating healthcare costs for all men diagnosed with prostate cancer in Ontario in 1992-2002 (Cohort A, n=37,933). II) estimating out of pocket costs and productivity losses for 1,500 men in three regions of Ontario (urban, suburban, rural, Cohort B). Total healthcare costs for Cohort A are identified by deterministically linking databases describing hospitalization, physician services, same day surgery, physician visits, drugs, radiation, long-term care, and home care. Net and attributable healthcare costs are estimated by matching prostate cancer cases with population-derived controls using propensity methods. Observation time for each case and control is allocated into 1 of 5 costing periods using a period hierarchy. Statistical models estimate the effects of demographic and disease-related factors on log-transformed net and attributable costs for each costing period. Chart review of cohort B cases will allow allocation of observation time into more precisely defined Markov health states. Out of pocket costs and productivity losses for survivors, obtained by survey, will be mapped onto existing health states. Longitudinal costs for Cohort B will be also estimated from linked administrative data.

Results: Observation time is available for 35,933 Cohort A patients for five costing periods: I) 90 days prior to diagnosis (100% of cohort); II) 180 days post diagnosis (97%); III) 180 days prior to death (20%); IV) 365 days prior to period III (18%); and V) continuing care (between II and IV) (89%). Total direct costs, excluding radiation, home care and long term care costs, per 100 patient days are: I) ($CAD 2004) $1073; II) $3292; III) $9233; IV) $3247; V) $1173. Propensity matching and calculation of net costs are underway. Data for approximately 30% of cohort B patients have been gathered.

Discussion: Obtaining primary data from a wide variety of primary sources to populate decision analytic policy models is labour-intensive but feasible, and may improve the evidentiary basis of the policies supported by the model.