P139: How available is availability bias? Examining factors that influence diagnostic error

Jonathan Sherbino, Sandra Monteiro, Jonathan S. Ilgen, Emily M. Hayden, Elizabeth Howey, Geoffrey R. NormanCanadian 2018 · DOI

Introduction

Diagnostic error represents a significant challenge in modern healthcare, often leading to adverse patient outcomes and increased costs. Within the medical literature, cognitive biases are frequently cited as the primary drivers of these errors. Among these, "availability bias"—the tendency to judge the probability of a diagnosis based on the ease with which similar recent examples come to mind—is often treated as a ubiquitous threat to clinical reasoning. However, despite its prominence in theory, empirical evidence linking availability bias to actual diagnostic errors in clinical practice remains sparse. This study investigates the real-world influence of availability bias across different levels of medical expertise to determine if clinical experience serves as a protective factor.

Methodology

The researchers implemented a two-phase experimental design to induce and subsequently measure the effects of availability bias among 111 participants from McMaster University, the University of Washington, and Harvard Medical School. The cohort was divided into three groups based on expertise: Emergency Medicine staff (n=40), residents (n=39), and medical students (n=32).

Phase 1: The Priming Intervention

Participants were first asked to verify a diagnosis for a specific clinical case. This case served as a "prime" and was manipulated into three conditions:

  1. Representative and Similar: The prime was representative of a specific diagnosis (Diagnosis A) and shared features with a future test case.
  2. Representative but Dissimilar: The prime was representative of Diagnosis A but lacked similarity to the test case.
  3. Alternative Diagnosis: The prime was representative of a completely different diagnosis (Diagnosis B).

Phase 2: The Diagnostic Test

Participants then diagnosed eight written clinical cases. These cases were carefully constructed so that two different diagnoses (A and B) were equally plausible based on the provided information. The goal was to observe whether the "priming" from Phase 1 would bias the participants toward choosing one diagnosis over the other.

Key Findings

The Expertise Buffer

The study’s most striking result was that availability bias had virtually no impact on experienced clinicians. Both Emergency Medicine staff and residents maintained an exceptionally high diagnostic accuracy of 98% across all experimental conditions. For these groups, the recently seen "prime" case did not statistically alter their diagnostic path, suggesting that clinical expertise provides a robust shield against this specific cognitive trap.

The Novice Susceptibility

In contrast, medical students demonstrated a clear susceptibility to the experimental conditions, with an overall accuracy of 85%. However, the nature of this susceptibility was unexpected. While students were more easily influenced by recent examples, the "prime" did not always lead to error.

In the "no prime" baseline condition, students achieved an accuracy of 73%. When provided with a representative category prime, their accuracy significantly improved to 94% ($p < 0.002$). This suggests that for novices, recent examples may actually serve as helpful pattern-recognition aids rather than purely distracting biases.

Statistical Significance

The difference in accuracy between expertise levels was highly significant ($F = 35.6, p < 0.0001$), reinforcing the conclusion that the transition from student to resident involves developing a form of reasoning that is resilient to simple availability effects.

Discussion and Implications

These results challenge the prevailing narrative that availability bias is a universal source of error for all physicians. Instead, the data suggests that as clinicians gain expertise, they move beyond simple pattern matching to more resilient forms of reasoning that are not easily swayed by recent experience.

For medical education, the finding that priming actually improved student performance is noteworthy. It suggests that "availability" can be a double-edged sword: while it may lead to bias in some contexts, it also facilitates learning and pattern recognition in the early stages of medical training.

Limitations and Future Directions

While this study suggests experts are resistant to availability bias in a controlled setting, the authors note several limitations:

  • Case Format: The study utilized written clinical cases. It remains possible that the high-pressure, multi-sensory, and interrupted environment of a real-world Emergency Department could trigger biases that a written test does not.
  • Other Biases: Resistance to availability bias does not imply immunity to all cognitive errors. Further research is needed to identify which other factors contribute to diagnostic errors in experienced staff.
  • Educational Transition: Future studies should investigate the specific point in training where clinicians develop this "bias-resistant" reasoning to better inform medical curricula.

Summary updated

BibTeX
@inproceedings{sherbino2018p,
  author = {Jonathan Sherbino and Sandra Monteiro and Jonathan S. Ilgen and Emily M. Hayden and Elizabeth Howey and Geoffrey R. Norman},
  title = {{P139: How available is availability bias? Examining factors that influence diagnostic error}},
  booktitle = {Canadian},
  year = {2018},
  doi = {10.1017/cem.2018.337},
}
Figure