In this cost-effectiveness analysis, we compared current practice in trauma triage with hypothetical interventions that would modify physician decision making. We found that even an expensive intervention to change the cognitive biases of individual physicians would be economically reasonable compared with either allowing current practice to continue or another top-down intervention to change physicians’ decisional thresholds. However, we also found that given different conditions, such as trauma centers that could deliver less expensive inpatient care, an intervention to change decisional thresholds might be the better alternative. Therefore, the intervention choice should be customized to the characteristics of the trauma system.
The clinical uncertainty involved in trauma triage results in two different types of errors in physician decision making: under-triage (the admission of patients with moderate to severe injuries to non-trauma centers) and over-triage (the transfer of patients with minor injuries to trauma centers). Under-triage denies patients the benefits of care at trauma centers, while over-triage increases the costs of care without a commensurate improvement in outcomes. MacKenzie et al.
have shown that care provided at trauma centers to patients with moderate to severe injuries is cost-effective compared with care provided at non-trauma centers.30
However, this analysis does not account for the degree to which rates of over-triage affect the costs and clinical outcomes of regionalization. Moreover, since quality improvement interventions have different effects on under- and over-triage, the optimal strategy for improving regionalization is similarly unclear.
As a precursor to developing a quality improvement program in trauma triage, we performed a thought experiment to identify the best target for intervention at the physician level. We used principles of behavioral science to categorize different cognitive processes that might affect physician compliance with clinical practice guidelines,6,7
and limited the analysis to interventions that targeted a single cognitive process. Our experiment produced several key observations. First, we found that an intervention that could change the perceptual sensitivity of individual physicians, such as a program to recalibrate physicians’ cognitive biases, would be cost-effective compared to either current practice or an intervention to change decisional thresholds. Improving physicians’ ability to discriminate between patients who do and do not meet the reference standard for transfer would cost $62,799 per QALY gained. By comparison, an intervention to improve influenza vaccination rates among the elderly costs $49,000 per QALY gained; placing automatic cardiac debrillators in public places costs $57,000 per QALY gained, and dialysis costs $129,000 per QALY gained.30
Second, we found that such an intervention could cost up to $800 per patient and would meet societal standards for cost-effective care. To put that number into context, non-trauma centers in Pennsylvania hospitalize approximately 25,000 trauma patients each year.5
Consequently, the state of Pennsylvania could spend up to $20 million helping physicians at non-trauma centers adhere to clinical practice guidelines and still provide cost-effective trauma care. Even a moderate additional investment in quality improvement could therefore have a profound impact on public health.
Third, we found that the most cost-effective target for a quality improvement intervention varied based on the relative effectiveness and costs of care at trauma centers compared with non-trauma centers. For example, if trauma centers could provide less expensive care, an intervention to change decisional thresholds would surpass the alternatives. As the costs of over-triage decrease, the relative benefit associated with reducing under-triage increases. In other words, the need for nuanced discrimination among patients diminishes. Under these conditions, we speculate that excluding the physician from the decision making process entirely might offer the best method of improving regionalization in trauma. For example, a mid-level provider could possibly use a computerized decision support tool, which screens patients based on a set of clinical predictors, to triage patients. Consequently, identifying the best quality improvement intervention for a region depends on the characteristics of the individual trauma system, precluding a single national or state-wide strategy to reduce variability in trauma triage.
This paper has several limitations. First, estimates of triage patterns derive from an analysis of Pennsylvania discharge data. These values may not be generalizable to other regions of the country. However, varying these parameters in sensitivity analyses produced similar results, suggesting the robustness of our conclusions. Second, our assessment of costs associated with regionalization did not include the effect of increasing the number of patients at trauma centers. Given that hospitals routinely operate close to system capacity, raising the census may have implications for patient outcomes. Consequently, the implications of a large-scale redistribution of patients would require a demonstration project to characterize adequately. Third, we tested hypothetical instead of actual interventions. Although a series of quality improvement interventions exist in trauma, their effect on physician decision making is unknown. Sketching the problem broadly as barriers to regionalization eliminated some of the uncertainty involved in our thought experiment. Finally, we did not account for all the potential consequences of injury, including costs that might arise from disabilities such as post-traumatic stress disorder. Based on current evidence that trauma centers do not influence functional outcomes after trauma, we hypothesize that further specification of our model would not have significantly altered our analysis.
In conclusion, investing in an intervention to improve adherence to clinical practice guidelines for the triage of trauma patients would yield significant value even if the intervention were expensive. Our cost-effectiveness analysis highlights that reducing under-triage may require a more nuanced and context-dependent approach than has been previously recognized.