Association of Self-reported Primary Care Physician Tolerance for Uncertainty With Variations in Resource Use and Patient Experience

Key Points Question Are variations in resource use and patient experience associated with physician self-reported tolerance for uncertainty? Findings This survey study of 217 physicians found that primary care physicians with a low tolerance for uncertainty were less likely than those with a high tolerance for uncertainty to order diagnostic tests. Primary care physicians with less tolerance for uncertainty had worse patient experience scores than those with a high tolerance for uncertainty, although associations were not monotonic. Meaning This study suggests that primary care physicians’ self-reported tolerance for uncertainty was associated with diagnostic test ordering and patient experience; however, because the causal links are not well defined, how tolerance for uncertainty could be potentially modified and what, if any, implications there are for health care quality and patient experience remain to be elucidated.


Introduction
Inappropriate variations in clinical practice are a known cause of poor quality care and safety, associated with poorer health outcomes, increased costs, disparities, and increased burden on the health system. 1,2For these reasons, a reduction in unwarranted clinical practice variations that cannot be explained by patient illness or patient preference is a central theme of quality improvement.Such variations in medical decision-making are often associated with clinician differences in both conscious and unconscious cognitive processing.p62) A clinician's tolerance for uncertainty is the balance between the positive and negative responses, which has been shown to be associated with the diagnostic process, with potential for diagnostic error that may effect patient outcomes. 8Findings on the association between tolerance for uncertainty and the use of health services have been mixed.On the one hand, low tolerance for uncertainty has been associated with increased tendencies to order tests, 9,10 failure to comply with evidence-based guidelines, 11 additional empirical treatment regimens, 12,13 withholding of negative genetic test results, 14 and fear of malpractice litigation and defensive practice. 15Conversely, some studies have shown that a lower tolerance for uncertainty may be associated with the lower use of tests, 16,17 which may be associated with an unconscious suppression of uncertainty, leading to premature closure of the diagnostic reasoning process-the single most common phenomenon in misdiagnosis. 18inician tolerance for uncertainty also has conceptual causal links to patient experience.
[27][28][29][30][31] Physicians who are intolerant of uncertainty are reluctant to disclose uncertainties to patients when making decisions, which can impede open, honest, and respectful communication. 32Inadequate management of uncertainty may cause unnecessary concern and distress to patients, risking undercutting the patient-physician relationship and decreasing trust. 33The inability to communicate uncertainty creates a false sense of certainty among patients, which can lead to substantial distrust when that certainty proves to be overstated.
To date, studies examining physicians' tolerance for uncertainty have been mainly small scale, with many inconsistent findings, to our knowledge.To address this gap, we sought to examine the association of physician tolerance for uncertainty with variations in resource use and patient experience among primary care physicians (PCPs) at a large multispecialty academic physician practice organization.

Study Design, Population, and Setting
We performed a survey study linking near-comprehensive physician survey data, patient experience survey data, and hospital billing data among PCPs at Massachusetts General Hospital, the largest pediatric PCPs, regardless of type of board certification.This data set was well positioned to answer this analytic question owing to the near-complete capture of the variation data, along with a high response rate (approximately 90%) to our physician survey in the setting of an incentive payment.

JAMA Network Open | Health Policy
The Mass General Brigham institutional review board approved this study.The Partners Human Research Committee determined that the project met the criteria for institutional review board exemption because the research was limited to the use of survey data, was not subject to US Food and Drug Administration regulations, and there was no more than minimal risk to study participants.
Completion of the survey was considered implied consent of participation.All data used were strictly anonymized; only a research coordinator had access to the file linking responses to identifiers.
Physicians were asked to consent to survey participation on the cover letter inviting them to participate.All methods for the survey are in compliance with the American Association for Public Opinion Research (AAPOR) reporting guideline for survey studies. 34

Survey Instrument and Variables
Information on physician tolerance for uncertainty was collected from the 2019 biennial Massachusetts General Physicians Organization (MGPO) survey.The main goals of the MGPO survey were to assess physician well-being, direct priorities regarding funding of practices, evaluate department and division chiefs, and better understand the functioning of the clinical enterprise.
Tolerance for uncertainty was measured using the single item, "I find the uncertainty involved in patient care disconcerting," adapted from the 15-item Physicians' Reaction to Uncertainty Scale, developed by Gerrity et al. 35 This single item has been shown to stratify tolerance for uncertainty among physicians, [36][37][38] and it is often used in surveys such as ours addressing multiple content areas within space constraints where use of the full scale is limited by length.The score for this single item ranges from 1 to 5, with 5 signifying greatest discomfort from uncertainty (or lowest tolerance for uncertainty) and 1 signifying least discomfort from uncertainty (or greatest tolerance for uncertainty).The extent to which participants agreed with the statement was reduced to 3 categories: low tolerance (strongly agree or moderately agree), medium tolerance (neither agree nor disagree), and high tolerance (moderately disagree or strongly disagree).The responses to this question along with some physician characteristics (gender and race and ethnicity) were extracted for merging with billing data on resource use and patient survey responses.Data on physician race were self-reported from categories including American Indian or Alaska Native, Native Hawaiian, Asian, Black, White, other, and prefer not to say.Data on physician ethnicity were self-reported from categories including Hispanic, non-Hispanic, and prefer not to say.We speculate that the "other" category may include physicians from parents of different races or physicians who self-identify as a race other than the provided categories, such as Middle Eastern and North African.Data on race and ethnicity were collected to assess equity among physicians at our organization.
The MGPO collects data on patient experience from a random sample of patients after outpatient visits through the Clinician and Group Consumer Assessment of Healthcare Providers and Systems Survey. 39For this analysis, we focused on the physician communication and overall physiclinician rating domains of this survey: "physician explained things in a way that is easy to understand" (MD Explain); "physician listened carefully to patient" (MD Listen); "physician showed

Sensitivity Analysis
Some of the data on patient experience came from years prior to the year when physician tolerance for uncertainty was measured (2019).Because tolerance for uncertainty improves with experience, one can argue that the physician level of tolerance in those years might have been different, potentially creating noise in our results.To test for such a possibility, we performed a sensitivity analysis by using only 2019 responses and specifying a single-level model (rather than hierarchical model).This sensitivity analysis did not change our main findings from the full model.

Statistical Analysis
Standard descriptive statistics were used to summarize data and compare the distributions among the 3 categories of physician tolerance for uncertainty.Because the degree of tolerance for uncertainty is ordinal in nature, we used trend-based tests to assess the association of physician tolerance for uncertainty with other variables.For categorical variables, we used the Cochran-Armitage test and the Cochran-Mantel-Haenszel test for row mean scores, as appropriate.For continuous variables, we used the Jonckheere-Terpstra test.For modeling and risk adjustment, we used a 2-stage hierarchical model with physicians as a random effect to account for clustering of patients within physicians.Binary outcomes were modeled using random-effect logistic regression, and count data outcomes were modeled using random-effect Poisson regression.For models of patient experience, we adjusted for patient age, patient-PCP gender (same vs different), patient-PCP race and ethnicity (same vs different), educational level, self-reported health status, length of relationship with physician, and visit year.For the rest of the models, we adjusted for demographic characteristics (age, gender, and race and ethnicity), socioeconomic variables (payer type and zip code income), comorbidity indicators, and physician class (pediatric vs adult).All P values were from 2-sided tests, and results were deemed statistically significant at P < .05.Regression results are reported as odds ratios (ORs) or rate ratios depending on the nature of the outcome measure.
Analyses were performed using SAS, version 9.4 (SAS Institute Inc).

Physician Tolerance for Uncertainty and Patient Characteristics
Physicians with a low tolerance for uncertainty tended to have a higher proportion of patients younger than 50 years than did physicians with a high tolerance for uncertainty (20 459

Physician Tolerance for Uncertainty and Diagnostic Test Ordering
In unadjusted comparisons, physicians with a low tolerance for uncertainty were less likely than physicians with a high tolerance to order complete blood cell counts Adjusting for case-mix factors, we found that physicians with a low tolerance for uncertainty were less likely than physicians with a high tolerance to order CBCs (OR, 0.66; 95% CI, 0.50-0.88),thyroid tests (OR, 0.67; 95% CI, 0.52-0.88),BMPs (OR, 0.78; 95% CI, 0.60-1.00),and LFTs (OR, 0.72; 95% CI, 0.53-0.99)(Table 3).Differences in test ordering tendency between physicians with a low tolerance for uncertaintly and physicians with a high tolerance for uncertainty were not statistically significant for ordering CBCs with differential, lipid tests, and high-cost imaging.Lastly, differences in test ordering tendencies between physicians with a medium tolerance for uncertainty and physicians with a high tolerance for uncertainty were not statistically significant.

Physician Tolerance for Uncertainty and Outpatient Visits
The median number of PCP visits per year were the same across all 3 groups of physicians (median, 1 [IQR, 1-2]).Similarly, the median number of visits to specialists were similar among the patients of the 3 physician groups (median, 1 [IQR, 0-3]) (Table 2).After adjusting for risk factors, we found no association between physician's level of tolerance for uncertainty and their patients' visits to PCP or specialty offices (Table 3).

Tolerance for Uncertainty and ED Visits
In unadjusted comparisons, we found that patients whose PCP had a lower tolerance for uncertainty were more likely to visit the ED than patients whose PCP had a medium or high tolerance for respectively; P < .001)(Table 2).After adjustment for risk factors, these differences were not statistically significant (Table 3).Finally, the likelihood of 30-day readmission was not associated with PCPs' tolerance for uncertainty in both adjusted and unadjusted models (Table 2 and Table 3).[88.3%];P = .10).

Physician Tolerance for Uncertainty and Patient Experience
Table 3 presents risk-adjusted parameter estimates for all our measured outcomes.Physicians with a higher tolerance for uncertainty were more likely than physicians with a medium tolerance for uncertainty to listen to patients carefully (MD Listen; OR, 0.65; 95% CI, 0.50-0.83)and to receive higher overall ratings (MD Rate; OR, 0.80; 95% CI, 0.66-0.98).These associations, however, were not demonstrated for physicians with a low tolerance for uncertainty compared with physicians with a high tolerance for uncertainty (MD Listen: OR, 0.79; 95% CI, 0.61-1.02;MD Rate: OR, 0.85; 95% CI, 0.68-1.06).

Discussion
This study adds new knowledge by examining associations between PCP tolerance for uncertainty and variations in resource use and patient experience.We found that PCPs who reported a lower tolerance for uncertainty ordered fewer diagnostic tests (specifically CBCs, thyroid tests, BMPs, and LFTs).Physicians who reported a lower tolerance for uncertainty also had worse patient-reported survey scores for listening carefully and for overall rating of patient experience, although this association was not monotonic and was less clearly established.We also found that physicians with a lower tolerance for uncertainty had fewer years since training and higher proportions of patients with a more vulnerable socioeconomic status than physicians with a high tolerance for uncertainty.Given that this research question requires detailed and complete information about physicians' clinical practice and associated patient experience, it is likely unanswerable using information from large national data sets.Although this is a single-center study, the opportunity to compare the degrees of physician tolerance for uncertainty with data on variations is rare, especially with a data set that has nearcomplete capture.
In contrast to prior research finding that a low tolerance for uncertainty was associated with increased test-ordering tendencies, 9,10 we found that physicians with a low tolerance for uncertainty were less likely to order diagnostic tests, risking premature closure on the diagnostic reasoning process and an increased chance of diagnostic error and downstream negative ramifications for the patient and the health care system. 18These prior results were derived in the ED setting, so we speculate that our results may differ because we studied PCPs.Both underuse and overuse of diagnostic testing can be associated with low-quality care.As such, the implications of our findings about the association between tolerance for uncertainty and overall health care quality are uncertain.
This study builds on earlier research evaluating the tolerance for uncertainty among health care professionals.Tolerance for uncertainty has been associated with various practice-related attitudes of medical students, with studies showing that students with a lower tolerance for uncertainty showed a more negative orientation toward patients' psychological problems, more Machiavellianism, a preference for a structured work environment, 40,41 more negative attitudes toward the underserved, 42 and bias against those who have alcohol use disorder. 43We found differences in some patient experience measures (listening and overall rating) but not others (explaining, respect, and time spent), and the demonstrated associations with listening and overall rating were not monotonic.Physicians with a medium tolerance for uncertainty had worse listening and overall rating scores than physicians with a high tolerance for uncertainty.These associations may be complex, however, because physicians with a low tolerance for uncertainty demonstrated a statistically insignificant trend toward lower ratings on only those 2 measures.Disclosing and discussing uncertainty have been recognized to be the 2 most challenging elements of risk communication. 44Although physicians often worry that admitting uncertainty will lead to loss of

JAMA Network Open | Health Policy
Physician Tolerance for Uncertainty and Variations in Resource Use and Patient Experience confidence, it has been suggested that appropriate expressions of uncertainty can lead to stronger physician-patient relationships. 45One study found that when PCPs used direct expressions of uncertainty, such as "I don't know" or "It's not clear," there were higher levels of positive talk, patient engagement, and patient satisfaction. 46It may be that physicians with a higher tolerance for uncertainty were able to communicate uncertainty better to patients, which was associated with higher overall ratings and a feeling of being listened to more carefully.By discussing uncertainty with patients, whether about diagnosis, prognosis, or treatment and management options, physicians may find a reduction in their stress and anxiety through sharing decision-making responsibility.A patient's values and preferences can often guide treatment choices when otherwise the best means of proceeding would be uncertain.
Our results suggest that responses to uncertainty may be associated with variations in resource use and patient experience.There is much evidence to suggest that tolerance for uncertainty is a state, not a trait, and therefore amenable to change through an educational and experiential process, 47 although there is likely an association with inherited personality traits and environmental influences that predispose individuals to specific psychological responses.Particular attention likely needs to be paid to those with less experience, with senior colleagues acting as role models to normalize the experience of uncertainty.Understanding and acknowledging uncertainty and acquiring proper coping strategies are now regarded as core clinical competencies for medical graduates and trainees in the UK, US, Australia, and much of Europe, [48][49][50][51][52][53][54]

Limitations
This study has several limitations.First, our results were subject to the inherent reporting biases that often occur in survey studies.However, because this was a retrospective analysis, any social desirability biases were minimized because survey participants were unaware of the specific hypothesis of this study and all data were collected in a deidentified and confidential manner.Owing to space constraints, we used a single-item self-reported measure to assess tolerance for uncertainty.
[38] However, conceptual definitions of uncertainty and tolerance vary, and different physicians may have interpreted the question differently.The variation data were collected separately without risk of reporting bias.Our response rate of 89.3% is a robust response, increasing the accuracy of the data and minimizing selection bias.Second, our results may not be generalizable beyond PCPs, although there is no reason to postulate that these findings would be unique to primary care.
Nevertheless, further studies are needed to confirm our findings in other hospital and academic settings.Third, as a single-center study, it is unclear the extent to which we can generalize these findings to other settings.However, we do think that, given the unusual lack of missing data in our survey, the internal validity of our results are better than a multisite survey with missing data.Fourth, we demonstrated improvements in some patient experience measures for physicians with a high tolerance for uncertainty relative to those with a medium tolerance for uncertainty, but these differences were not apparent for physicians with a low level of tolerance for uncertainty.These associations may be complex and nonmonotonic, or a larger data set with more statistical power might demonstrate a monotonic association.Fifth, although we demonstrated an association between lower self-reported tolerance for uncertainty and fewer tests ordered, we cannot know from these data how frequently these tests were ordered in guideline-adherent or appropriate ways.

Conclusions
Identifying and effectively managing inappropriate variations in clinical practice have proven to be difficult.This study supports the hypothesis that physicians' tolerance for uncertainty is associated with differences in resource use and patient experience and may, therefore, be associated with some of the variations, lending support to improving the management of uncertainty.Enhancing physician tolerance for uncertainty may help reduce unwarranted clinical practice variations and may also improve the patient experience by enhancing communication and satisfaction.By shifting the culture of medicine to acknowledge and openly discuss uncertainty-with colleagues and patientsempathetic, positive, and partnering relationships can be established that may bolster trust and increase patient engagement and comfort, improving communication, patient safety, and physician well-being.

JAMA Network Open | Health Policy Physician
Tolerance for Uncertainty and Variations in Resource Use and Patient Experience Physician Tolerance for Uncertainty and Variations in Resource Use and Patient Experience physician tolerance for uncertainty with patient age was overall similar when we excluded pediatric patients from the analysis.For the adult population, the proportion of patients younger than 50 years was 43.6% (11 536 of 26 470) among physicians with low tolerance for uncertainty and 34.0%(13 461 of 39 637) among physicians with high tolerance for uncertainty (P < .001).

Table 2 .
Unadjusted Association of Physician Tolerance for Uncertainty With Selected Outcomes a Trend-based tests are used to test association of outcomes with physician level of tolerance for uncertainty.For categorical outcomes with 2 levels, we used the Cochran-Armitage test, and for continuous outcomes, we used the Jonckheere-Terpstra test.bFor patient experience outcomes, except for MD Time, physicians with a medium tolerance for uncertainty have lower scores than physicians with a low tolerance for uncertainty.However, pairwise comparisons show that the difference in top score c Diagnostic tests and ED admissions are all specified as binary outcomes.We initially considered modeling them as count outcomes, but most patients had either 0 or 1 value.The proportions of patients with more than 1 test per year were 9.4% (11 093 of 117 644) for BMP, 6.0% (7100 of 117 644) for LFTs, 5.1% (5965 of 117 644) for CBC, 3.9% (4533 of 117 644) for lipid tests, 1.8% (2139 of 117 644) for high-cost imaging, and 1.6% (1895 of 117 644) for CBC with differential.Similarly, the proportion of patients with more than 1 visit was 6.4% (7493 of 117 644) for ED visits.

Table 3 .
Risk-Adjusted Association of Physician Tolerance for Uncertainty With Selected Outcomes a In addition to PCP level of tolerance, models also adjusted for the following variables (to save space, we have reported only estimates of tolerance for uncertainty): (1) patient experience: patient-PCP gender (same vs different), patient-PCP race (same vs different), visit year, patient age, length of relationship with PCP, self-reported health status, and educational level; (2) diagnostic tests: PCP class (adult vs pediatric), race, payer group, zip code median income, number of PCP visits, number of specialist visits, and presence of the following comorbidities: chronic obstructive pulmonary disease, chronic kidney disease, congenital heart failure, uncomplicated diabetes, complicated diabetes, mild liver disease, and peripheral vascular disease; (3) outpatient visits: PCP class (adult vs pediatrics), gender, age, race, payer group, zip code median income, and the presence of the following comorbidities: chronic obstructive pulmonary disease, chronic kidney disease, congenital heart failure, uncomplicated diabetes, complicated diabetes, mild liver disease, and peripheral vascular disease; (4) ED admissions: PCP class (adult vs pediatrics), gender, age, race, payer group, zip code median income, and the presence of the following comorbidities: chronic obstructive pulmonary disease, chronic kidney disease, congenital heart failure, uncomplicated diabetes, complicated diabtes, mild liver disease, and peripheral vascular disease.Physician Tolerance for Uncertainty and Variations in Resource Use and Patient Experience of 3768 [93.5%] vs 2159 of [90.0%] and 2166 of 2365 [91.6%], respectively; P = .001),and MD Rate (3231 of 3768 [85.8%] vs 1954 of 2398 [81.5%] and 1932 of 2365 [81.7%], respectively; P < .001).There was no significant difference in the scores for MD Respect (high, 3571 of 3768 [94.8%] vs medium, 2252 of 2398 [93.9%] vs low, 2237 of 2365 [94.6%];P = .63)and MD Time (high, 3372 of 3768 [89.5%] vs medium, 2105 of 2398 [87.8%] vs low, 2088 of 2365 a b These results are based on a 2-stage hierarchical model with random effect and random intercept for PCPs.Outcomes for patient experience, diagnostic tests, and admissions are based on a hierarchical logistic model, and their results are reported as odds ratios.Outcomes for outpatient visits are based on hierarchical Poisson models, and regression results are reported as rate ratio.JAMA Network Open | Health Policy JAMA Network Open.2022;5(9):e2229521.doi:10.1001/jamanetworkopen.2022.29521(Reprinted) September 1, 2022 7/13 Downloaded From: https://jamanetwork.com/ on 09/28/2023 Physician Tolerance for Uncertainty and Variations in Resource Use and Patient Experience Sixth, owing to the observational design our study, we were careful to test only for associations and do not draw conclusions about causality from our findings alone.We do not know if interventions to improve tolerance for uncertainty among physicians would be effective, nor if they would improve clinical quality and outcomes.However, our results at least raise the hypothesis that efforts to improve tolerance for uncertainty in medical training and practice may improve patient satisfaction and could increase ordering of tests.