The debate about how much to spend on health care is one that preoccupies many, if not all, countries around the world. In the UK, the NHS has enjoyed a period of unprecedented real growth of spending, following a commitment by the Prime Minister, Tony Blair, to raise the proportion of UK gross domestic product spent on health from the then 6.8% to the European national average of 8.0% [1]. This promise was of course made in the belief that increased spending would improve outcomes, although the relationship between healthcare expenditure and health outcomes is complex [2,3]. The study by Luce et al. published in this issue of Value in Health makes a valuable contribution to this debate by attempting to estimate the value of the health improvements that have come about from health-care expenditures in the United States in the last two decades of the 20th century [4]. It adds to a small but growing literature and is to be welcomed.

The estimates presented in the Luce et al. article reflect calculations that are unavoidably weakened by data limitations. Key assumptions, recognized by the authors, include: the uncertainty around the value we put on the units of health improvements; the uncertainty around the proportion of the improvements that are attributable to health-care expenditure, rather than other health improving social changes; and the assumed lags between spending and health-care benefits. In addition, there is the problem of splitting the attributed returns between investments in health care itself and investments in health-care research that underpin them. As similar exercises are underway to estimate the return on biomedical or health-related research [5], there are dangers of double counting as the returns in the form of health improvements are claimed by both those estimating the returns to research and those estimating the returns to healthcare services. All these details could be questioned and no doubt will be the focus of debate and refinement as others go on to undertake similar calculations. It is to be hoped that future analyses will be able to build on stronger data and to be tested against a wider range of alternative assumptions and methods. But to dismiss the estimates provided here, on the grounds that they are not perfect, would be churlish.