work-related activities consisting of nurse, pharmacist, and/or patient medication dispensing, preparation, administration and/or storage. Projects consisted of time and cost differences related to 1) three proton pump inhibitor dosage...
morework-related activities consisting of nurse, pharmacist, and/or patient medication dispensing, preparation, administration and/or storage. Projects consisted of time and cost differences related to 1) three proton pump inhibitor dosage forms and seven administration methods, and 2) seven recombinant human growth hormone administration methods/devices. Performance-based time data were then used to determine personnel/patient opportunity time and supply costs associated with different forms of medications and delivery devices. Simulations were developed and used to hold independent variables constant so only observed differences between medications and/ or device types could be assessed. Statistical and micro-economic cost analyses were conducted specific to each type of medication and/or device. RESULTS: Processes and results show two detailed examples as case studies of how simulation-based research may be used to assess health care processes at the micro level. The advantages of isolating processes of interest from the day-to-day complexity of patient care are demonstrated. Simulations may also represent an efficient assessment alternative of health care processes at the micro level with potential for projection to the macro level as compared to live, direct observation, cost-intensive, patient-centered care practice evaluations. CONCLUSIONS: Simulation-based time-and-motion and activity-based cost analyses allowed detailed micro-level time, workload, and supply evaluations that may be projected to the macro level. Professional schools' simulation laboratories offer appropriate settings for such studies. OBJECTIVES: Demonstrate a technique to characterize the economic burden of illness over time, correcting for censoring bias and controlling for difference in baseline characteristics between comparison groups. A sample of patients with diagnosis of disease A in 2004-2008 were extracted from MarketScan® databases and followed to death, disenrollment, or December 31, 2008 (cases). The first diagnosis date was the index date. Enrollees without disease A were extracted as controls. Their index dates were assigned based on the distribution of index dates of cases. METHODS: First, Kaplan-Meier estimates for the probability of remaining in the data were calculated by month and disease status. Failure event was death or disenrollment. Censoring event was termination of MarketScan contract or end of study period. Next, total health care costs were estimated using generalized linear models (GLM) on the subsample of survivors/enrollees in each month, controlling for disease status, patient demographic and clinical characteristics. Adjusted costs were calculated by month and disease status using the regression estimates and average characteristics. Estimated total costs during the whole follow-up period were the sum of probability of remaining in the data multiplied by regression-adjusted costs in each month. RESULTS: At the end of year 1, 40.5% of cases deceased or disenrolled from insurance compared to 13.6% of controls (p < 0.001); at the end of year 4 these figures were 76.4% and 30.5%, respectively (p < 0.001). GLM results indicated significantly higher cost among the group with disease A in each month during follow-up. The adjusted costs based on average characteristics was $86,592 for cases vs. $6,178 for controls in the first year and $151,077 vs. $21,890 in the first 4 years. CONCLUSIONS: In studies with variable-length follow-up, an estimator combining the survivor probability and regression-adjusted cost is more robust to censoring bias, and better depicts the economic burden of illness over time.