This paper investigates the predictive power of several risk attitude measures on a series of medical practices. We elicit risk preferences on a sample of 1500 French general practitioners (GPs) using two different classes of tools: scales, which measure GPs' own perception of their willingness to take risks between 0 and 10; and lotteries, which require GPs to choose between a safe and a risky option in a series of hypothetical situations. In addition to a daily life risk scale that measures a general risk attitude, risk taking is measured in different domains for each tool: financial matters, GPs' own health, and patients' health. We take advantage of the rare opportunity to combine these multiple risk attitude measures with a series of self-reported or administratively recorded medical practices. We successively test the predictive power of our seven risk attitude measures on eleven medical practices affecting the GPs' own health or their patients' health. We find that domain-specific measures are far better predictors than the general risk attitude measure. Neither of the two classes of tools (scales or lotteries) seems to perform indisputably better than the other, except when we concentrate on the only non-declarative practice (prescription of biological tests), for which the classic money-lottery test works well. From a public health perspective, appropriate measures of willingness to take risks may be used to make a quick, but efficient, profiling of GPs and target them with personalized communications, or interventions, aimed at improving practices.