Boxplots depict interquartile range (IQR) with whiskers indicating values within 1.5 times the IQR. Individual points represent outliers. Lines within boxes denote median values; medians were equal to the 75th percentile in panel A.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Shalowitz DI, Schorge JO. Suggestibility of Oncologists’ Clinical Estimates. JAMA Oncol. 2015;1(2):251–253. doi:10.1001/jamaoncol.2015.62
Quantitative estimates, including prognosis and likelihood of clinical outcomes, are an integral part of oncologic counseling and treatment. In general, estimates should be grounded in the best available data, adjusted as necessary for individual patients’ circumstances. However, it is well established that judgments are often biased by unrelated or uninformed numerical anchors (eg, guesses).1,2 There is good reason to suspect that physicians’ clinical assessments may be influenced by cognitive shortcuts, or heuristics. We therefore sought to determine whether oncologists’ clinical estimates might be biased by extraneous information.
Members of the New England and Mid-Atlantic Associations of Gynecologic Oncologists received emails containing an electronic link to 1 of 2 survey forms. Each survey included 2 clinical scenarios. Respondents were not compensated for their participation.
Scenario 1 described a 77-year-old woman with platinum-resistant serous ovarian carcinoma who expressed a desire to live either an additional 2 months or 30 months (randomized). Respondents indicated whether they believed that the patient would live as long as desired and provided their estimate of the patient’s life expectancy. Scenario 2 described a patient who had recently had cytoreductive surgery for serous ovarian carcinoma with symptoms clinically ambiguous for pulmonary embolism. Respondents were told that a medical student believed that the patient had either a 1% or 95% likelihood of pulmonary embolism (randomized). Respondents indicated whether they believed that the medical student’s estimate was too high or too low, provided their own estimate of the likelihood of pulmonary embolism, and indicated whether they would order diagnostic imaging or initiate therapeutic anticoagulation (vs expectant management alone) for the patient described. Means were calculated for each group and tested for significant difference using t tests for prognostic estimates and 2-proportion z tests for ordering decisions. The partners institutional review board exempted this study from final review.
Responses were obtained from 58 staff gynecologic oncologists, for a response rate of 34%. In scenario 1 (Figure, A), the mean (SD) estimated life expectancy of the patient who desired to live at least 2 additional months was 11.5 (4.9) months (median, 12.0 months) vs 15.4 (7.8) months (median, 18.0 months) for the patients who desired to live at least 30 additional months (P = .02).
In scenario 2 (Figure, B), when the medical student believed there was a 1% chance of pulmonary embolism (PE), the oncologists’ mean estimate was 18% (21%) (median, 10%). This increased to 33% (25%) (median, 25%) when the medical student believed there was a 95% chance of PE (P = .02). There was no significant difference in ordering diagnostic imaging or therapeutic anticoagulation between the 2 groups (65% vs 48%; P = .21). Information on nonresponders was not available. However, respondents were evenly distributed over time since completing training; this variable did not consistently affect estimates.
Oncologists’ clinical judgments, including estimates of life expectancy, seem to be susceptible to influence by extraneous information consistent with anchoring bias, a heuristic very well characterized in cognitive psychology.3 Quantitative estimates are likely to be disproportionately influenced by initially presented values, irrespective of relevance or reliability. This study was limited by low response rate and small sample size. Nevertheless, the results suggest that extensive training and availability of data may not protect clinicians from nonrational clinical decision-making. Rather, whenever there is clinical uncertainty, there is potential for cognitive bias.4
Furthermore, awareness alone is not sufficient to avoid bias. Existing data suggest that the anchoring effect is robust and refractory to many possible corrective strategies. However, one approach that has shown some promise is to “consider the opposite.”5 Oncologists should attempt to identify possible anchors during counseling and clinical care, and correct underestimates or overestimates by invoking available evidence. It may not be possible to completely debias clinical encounters, but physicians should consider critical monitoring of their decision-making processes to be an essential part of high-quality cancer care.6
Corresponding Author: David I. Shalowitz, MD, Perelman Center for Advanced Medicine, 3400 Civic Center Blvd, Third Floor W, Philadelphia, PA 19104 (firstname.lastname@example.org).
Published Online: March 12, 2015. doi:10.1001/jamaoncol.2015.62.
Author Contributions: Dr Shalowitz had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Shalowitz.
Acquisition, analysis, or interpretation of data: Both authors.
Drafting of the manuscript: Both authors.
Critical revision of the manuscript for important intellectual content: Both authors.
Statistical analysis: Shalowitz.
Study supervision: Schorge.
Conflict of Interest Disclosures: None reported.
Additional Information: The text of the scenarios is available on request.
Additional Contributions: Brian Zikmund-Fisher, PhD (University of Michigan, Ann Arbor), helped with critical revisions of this manuscript. He was not compensated for his assistance.
Create a personal account or sign in to: