This paper examines whether the responsiveness of health care utilization to cost-sharing varies by health status and the implications of such heterogeneity. First, we show theoretically that if health care utilization of those in poor health is less responsive to cost sharing, this, combined with the skewness of health expenditures in health status, leads to overestimates of the effect of cost sharing. This bias is exacerbated when elasticities are generalized to populations with greater expenditure skewness. Second, we show empirically that cost-sharing responsiveness does differ by health status using data from the Medicare Current Beneficiary Survey. Medicare beneficiaries are stratified into health status groups based on activity of daily living (ADL) impairments and self-reported health status. Separately, for each of the health status groups, we estimate the effect of Medigap insurance on Part B utilization using a two-part expenditure model. We find that the change in expenditures associated with Medigap is smaller for those in poorer health. For example, when stratified using ADLs, Medigap insurance increases expenditures for 'healthy' groups by 36.4%, while the increase for the 'sick' group is 12.7%. Results are qualitatively the same for different forms of supplemental insurance and different methods of health status stratification. We develop a test to demonstrate that adjusting our results for selection bias would result in estimates of greater heterogeneity. Our results imply that a lowerbound estimate of the bias from neglecting heterogeneity is about 2-7%.