Studies on the effect of ageing on health care expenditure (HCE) have revealed the importance of controlling for time-to-death (TTD). These studies, however, are subject to possible endogeneity if HCE influences the remaining life expectancy. This paper introduces a 10-year observation period on monthly HCE, socioeconomic characteristics and survivor status to first predict TTD and then use the predicted values as an instrument in the regression for HCE. While exogeneity of TTD has to be rejected, core results concerning the role of TTD rather than age as a determinant of HCE (the 'red herring' hypothesis) are confirmed.