We study a situation where physicians differing in their degree of altruism exert a diagnostic effort before deciding whether to test patients to determine the most appropriate treatment. The diagnostic effort generates an imperfect private signal of the patient’s type, while the test is perfect. In the absence of correcting transfers, physicians exert insufficient diagnostic effort and may rely excessively on testing. We show that the first-best allocation (where the degree of altruism is observable) can be decentralized by a payment scheme composed of i) a pay-for-performance (P4P) part based on the proportion of correctly treated patients to ensure the provision of the optimal diagnostic effort, and of ii) a capitation part to ensure both the optimal testing decision and the participation of physicians. When physicians’ degree of altruism is non-observable, the optimal contract is pooling rather than separating, an instance of non-responsiveness. Its uniform P4P component induces more altruistic physicians to exert a larger diagnostic effort. To promote optimal testing decisions, the capitation component should be adjusted based on diagnostic test costs.