In a computer simulation focusing on asymptomatic AAA management, physicians from three different specialties chose on average to operate sooner than recommended by current guidelines. The magnitude of this deviation from guideline-recommended behavior was similar for all three specialties. By selecting the 5.0% risk of perioperative mortality when the risk of spontaneous rupture was 1.6%, participants appeared on average to choose a 3.4% increased risk of perioperative mortality.
It is unclear why physicians in our study chose not to maximize perioperative survival. One possibility is the difference between our computer simulation and the more complex real-life clinical decision. For example, participants may have considered surgery inevitable and integrated the 5% perioperative mortality rate into their decision. Additionally, participants may have focused on years of life after surgery, choosing surgery earlier to “get it over with”19
. However, no participant identified either inevitability of surgery or years of life as relevant decision factors.
Most participants in our study identified “burst rate” as the most important decision factor. An asymmetric focus on the risk of rupture, instead of the 5% risk of perioperative death, may thus have played a role. In our previous work, lay adults found AAA rupture as upsetting as images of a person being forcefully abducted20
(Hemmerich, JA in press). Because AAA rupture is typically lethal, anxiety from waiting for the AAA to reach the optimal size for surgery may also have led to earlier operation. Aversion to uncertainty21
, and preferences among physicians and patients for action over inaction22
, are other possibilities. Although open AAA repair carries significant mortality risk, choosing surgery in our scenario may have appeared more certain than risking a potential rupture. Surgeons and anesthesiologists in our study may also have considered their clinical skills above average23
, leading them to undervalue the risk of perioperative death.
We also found a specialty-specific effect of recent experience. Surgeons exposed to the WWD condition chose surgery significantly sooner for a subsequent patient than did those exposed to either the PD or SO conditions. This tendency was consistent with our previous findings 8, 9
. Geriatricians exhibited this behavior to a lesser degree, and anesthesiologists exposed to the WWD condition did not behave differently from those exposed to the other two conditions. Why anesthesiologists and surgeons behaved differently with respect to recent experience is unclear. Self-reported experience with AAA management, age, and attitudes towards risk and anxiety did not differ between specialties. One possibility is divergent levels of regret24,25
Because of a greater sense of control over the outcome, surgeons may have felt more regret than anesthesiologists at ‘failing” to prevent a WWD. Experience with practice standards26
and quality improvement initiatives27
may also have made anesthesiologists more willing than surgeons to follow guidelines. Finally, anesthesiologists frequently begin their relationship with the patient after the decision to operate has been made. This greater degree of emotional distance may have insulated them from cognitive factors affecting operative decisions such as regret, anxiety, or uncertainty.
Our study has limitations. Because participants may have behaved differently in our simulation than in actual practice, the real-world relevance of our findings is unclear. Conflicts between maximizing survival vs. the financial donation, and unspecified decision factors such as quality of life, may also have influenced participants. To counter these possibilities, we took precautions to ensure that participants understood the scenario and decision algorithm, and were continuously provided with updated statistical information. Our debriefing experience suggested that participants were not confused about decisional factors. Rather, many noted (often ruefully) that they knew they had acted sub-optimally. In addition, participants choosing to maximize the financial donation (rather than perioperative survival) would have waited longer than 10 visits to operate28
. While we cannot say whether study subjects would have behaved differently in real life, discrepancies between physician practice and evidence-based guidelines are common2
. A second limitation is the composition of our specialty groups. Most participants were not experts at AAA management. However, in light of our previous work demonstrating that vascular surgeons are affected by negative prior experience9
, and evidence showing less guideline compliance with increased experience29
, it is not obvious that experts or those with more experience would have performed better. In addition, the geriatrician group had more women than the other two groups. While this discrepancy might have affected our results, female gender does not clearly predict better guideline compliance30
. Our scenario included no complex clinical details, required no advanced diagnostic assessment and continuously provided all relevant statistical data. To adhere to published guidelines, participants needed only to delay surgery until the risk of AAA rupture equaled the risk of perioperative death. That three separate specialty groups, studied at three different locations and times, did not do so suggests a decision preference worthy of further research.
In summary, in a realistic computer simulation of AAA management, we found two notable patterns of decision behavior. The first, shared by surgeons and anesthesiologists (and to a lesser degree geriatricians), was a preference to choose surgery when the risk of AAA rupture was lower than the risk of perioperative mortality. This preference differed from published guidelines. The second, primarily in surgeons, was a tendency to (over)react to a watchful waiting death by operating significantly sooner in a subsequent patient.
Our results raise important policy issues. That decision preferences may differ based on recent experience or specialty suggests differences in clinical training or baseline characteristics that may distort physician decision-making. That physicians with different responsibilities in multispecialty care may decide differently raises issues about how such decisions should be made, what patients should be told, and by whom. Without understanding how non-statistical factors influence decision behavior, policies targeted at increasing healthcare decision consistency through guideline dissemination may be ineffective. Identifying these factors under rigorous experimental conditions represents a first step towards bridging the gap between guideline recommendations and non-guideline-based physician behavior.