Purpose. To demonstrate the systematic bias in projecting lifetime costs of diabetic and non-diabetic patients, and major systemic complications that is generated by ignoring empirical trends in the mean and variance of cost data over time, and to highlight methods to reduce that bias.

Methods. Population based cohorts were identified for the province of Ontario for the years 1994-2004. These cohorts include all incident cases of diabetes over the age of 35 (N=610,852), and a 1:2 matched non-diabetes cohort (N=1.22 million). Further subsets of all diabetic and non-diabetic patients who experienced a systemic complication (amputation, angina, blindness, heart failure, myocardial infarction, nephropathy, and stroke) were identified. For all cohorts, the annual total cost which includes hospitalization, outpatient services, public drug coverage, emergency room visits and home care services for up to 11 years of follow-up were included. For each cohort, the amount of systematic bias by not including the empirical time trends in mean and variance was estimated by comparing the present value of projected lifetime costs using trended and non-trended data. Further possible reduction in bias when the cost distributions are assumed log normal and Itô Calculus is investigated.

Results. The mean cost of care for an incident diabetic patient falls 1.17% per year. Failing to include this trend overestimates the discounted projected lifetime future costs by 12%. The standard deviation of costs also falls 2.07% per year and excluding this trend represents an increase in 52% in the size of the confidence intervals around these projected costs. The long term costs of complications falls 5.25% per year, and the standard deviation falls on average 4.19% per year, which translates into an overstatement of mean follow-up costs by 39% and confidence interval width by 86%. The time trend for the costs of systemic complications for non-diabetics falls at a faster rate than diabetic patients do, introducing even further systematic bias. Other advanced methods can reduce uncertainty.

Conclusions. When estimating the long term costs of a chronic disease or the follow-up costs of a systemic complication, considerable systematic bias may be generated when the trend in costs observed empirically are not included. For lifetime cost effectiveness analysis, there may be considerable bias by overestimating costs and the size of confidence intervals of projected costs.