What Passed is Past? The Role of Recent Adverse Events in Physician Treatment Decisions

Overview: In this paper, I evaluate how within-physician information shocks – specifically, adverse clinical events - influence a physician’s subsequent treatment decisions. I find that recent patient deaths and surgical complications significantly influence a physician’s subsequent practice behavior, independent of patient clinical indication. Moreover, the influence of recent adverse events on physicians’ behaviors dissipates quickly, suggesting that physicians are initially overreacting to these salient, extreme outcomes. While the effect of an adverse event fades quickly, I find that the cumulative effect of adverse events is significant and economically meaningful. I conclude that physicians incorporate internal information into practice patterns using a variety of decision-making heuristics, with important implications for physician learning, resource use, and overall patient welfare.

Background: Prior research has examined how physicians allocate their scarce resources and make clinical decisions in complex informational environments. These decisions often have far-reaching clinical and economic consequences. I investigate the mechanics of physician decision-making in environments with noisy feedback where I hypothesize that physicians focus heavily on extreme outcomes when retrospectively evaluating the “correctness” of their decisions. This selective focus may be especially prominent in noisy systems because scarce cognitive resources lead physicians to search for salient signals, or because physicians have strong prior beliefs that extremes in noisy systems are less likely to be driven by randomness (and are thus more informative signals).

Methods: Using 100% inpatient EHR data from two academic hospitals (2015-2018), I empirically investigate this hypothesis in two clinical settings – namely, labor/delivery and the ICU - where the causal drivers of patient outcomes are unclear, negative outcomes are often extreme, and feedback is noisy. I estimate the effect of experiencing a prior adverse event on a physician’s subsequent treatment decisions separately in each of the two settings, conditional on a rich set of patient characteristics and time-invariant physician practice patterns. As a secondary analysis, I also look at the symmetries in response to both positive and negative “expected” and “unexpected” prior outcomes.

Results: I find that physicians heavily weight recent, extreme, and negative prior outcomes in deciding on the treatment for a patient. Specifically, in the labor/delivery setting, physicians who experience a patient complication in a vaginal delivery are more likely to switch to C-section in their next delivery. Conversely, physicians who experience a patient complication in a C-section are more likely to switch to vaginal in their next delivery. In the ICU setting, physicians are significantly more likely to send the next patient to the ICU if the previous patient dies, regardless of both the current and prior patient’s clinical indications. These behavioral effects fade rapidly over the subsequent few patients. Experiencing prior adverse events also leads to significant spillovers in a physician’s chosen level of treatment intensity for the next patient – as evidenced by increasing length of stay, number of tests ordered and total charges. Finally, there is evidence that physicians do not respond optimally to prior adverse events – they increase subsequent treatment intensity equally for “expected” and “unexpected” adverse events.