Payment systems and physician altruism under DRG reform: evidence from a multicenter artefactual field experiment in China

Payment mechanism reform is central to global efforts to balance cost containment and quality of care. We compared service choices under simulated Diagnosis-Related Group (DRG) and Fee-for-Service (FFS) payment across predefined severity levels and estimated the relative preference for experimental patient benefit and physician altruism. We conducted a multicenter artefactual field experiment with 252 practicing physicians from six tertiary public hospitals in three provinces of China. We summarized service quantity, patient benefit loss ratio (PBLR), and experimental profit, estimated preferences using mixed logit, and integrated 23 key-informant interviews through a joint display. FFS was associated with a significantly higher volume of services overall than DRG (mean 5.88 vs. 4.88; p < 0.001). Analysis by severity revealed a crossover effect, FFS incentivized overprovision for mild and moderate cases but yielded the model-based definition of optimality provision for severe cases. DRG reduced service volume but resulted in significant underprovision for severe cases ( p < 0.001). Severe-case PBLR was higher under DRG: mean 0.12 versus 0.10 and median 0.05 versus 0.04 ( p = 0.040), reflecting underprovision risk. Physician altruism, quantified using a mixed logit model, yielded a marginal rate of substitution of 5.499 (95% CI 4.532–6.466), in experimental profit tokens per patient-benefit token, indicating physicians forwent 5.499 units of experimental profit to generate one additional unit of patient benefit. Interviews confirmed that DRG-linked performance balances promoted under-provision and risk avoidance. Neither FFS nor DRG is universally superior. The experiment suggests severity-dependent trade-offs between service volume and model-defined patient benefit. Subject to confirmation in real-world implementation studies, these findings support evaluating severity-sensitive blended payment models and safeguards for complex cases.