Surgeons Have an Implicit Preference for Specific Disease Over Nonspecific Illness

Abstract

Background

Many symptoms are not associated with a specific, measurable pathophysiology. Such nonspecific illnesses may carry relative social stigma that biases humans in favor of specific diseases. Such a bias could lead musculoskeletal surgeons to diagnose a specific disease in the absence of a specific, measurable pathology, resulting in potential overdiagnosis and overtreatment.

Questions/purposes

(1) What factors are associated with surgeon implicit preference for specific disease over nonspecific illness? (2) What factors are associated with surgeon explicit preference for specific disease over nonspecific illness? (3) Is there a relationship between surgeon implicit and explicit preferences for specific disease over nonspecific illness?

Methods

One hundred three members of the Science of Variation Group participated in a survey-based experiment that included an Implicit Associations Test (IAT) to assess implicit preferences for specific, measurable musculoskeletal pathophysiology (specific disease) compared with symptoms that are not associated with a specific, measurable pathophysiology (nonspecific illness), and a set of four simple, face valid numerical ratings of explicit preferences. The Science of Variation Group is an international collaborative of mostly United States and European (85% [88 of 103] in this study), mostly academic (83% [85 of 103]), and mostly fracture and upper extremity surgeons (83% [86 of 103]), among whom approximately 200 surgeons complete at least one survey per year. The human themes addressed in this study are likely relatively consistent across these variations. Although concerns have been raised about the validity and utility of the IAT, we believe this was the right tool, given that the timed delays in association that form the basis of the measurement likely represent bias and social stigma regarding nonspecific illness. Both measures were scaled from -150, which represents a preference for nonspecific illness, to 150, which represents a preference for specific disease. The magnitude of associations can be assessed relative to the standard deviation or interquartile range. We used multivariable linear regression to identify surgeon factors associated with surgeon implicit and explicit preference for specific disease or nonspecific illness. We measured the relationship between surgeon implicit and explicit preferences for specific disease or nonspecific illness using Spearman correlation.

Results

Overall, there was a notable implicit bias in favor of specific diseases over nonspecific illness (median [IQR] 70 [54 to 88]; considered notable because the mean value is above zero [neutral] by more than twice the magnitude of the IQR), with a modestly greater association in the hand and wrist subspecialty. We found no clinically important explicit preference between specific disease and nonspecific illness (median 8 [-15 to 37]; p = 0.02). There was no correlation between explicit preference and implicit bias regarding specific disease and nonspecific illness (Spearman correlation coefficient -0.13; p = 0.20).

Conclusion

Given that our study found an implicit bias among musculoskeletal specialists toward specific diseases over nonspecific illness, future research might address the degree to which this bias may account, in part, for patterns of use of low-yield diagnostic testing and the use of diagnostic labels that imply specific pathophysiology when none is detectable.

Clinical Relevance

Patients and clinicians might limit overtesting, overdiagnosis, and overtreatment by anticipating an implicit preference for a specific disease and intentionally anchoring on nonspecific illness until a specific pathophysiology accounting for symptoms is identified, and also by using nonspecific illness descriptions until objective, verifiable pathophysiology is identified.

Introduction

Clinicians are trained to think systematically and to try to link symptoms to pathophysiology as represented in the concept of differential diagnosis. This process can be efficient and accurate among patients with a recognizable pattern of symptoms and signs that are characteristic of a specific pathophysiology. However, this approach may not work as well for patients with symptoms and signs that are not characteristic of a specific, common pathophysiology (nonspecific illness). Nonspecific illness is characterized by discomfort and incapability that is not accounted for by objective, measurable impairment and pathophysiology (physical examination findings, imaging results, or laboratory values) [2, 21, 22, 26]. Some examples include fibromyalgia, chronic fatigue syndrome, radial tunnel syndrome, complex regional pain syndrome, and repetitive strain injury [3]. Many symptoms brought to medical attention are never associated with a specific pathophysiology [16, 23]. Nonspecific illness, particularly nonspecific pain, is common in musculoskeletal specialty care [15, 26]. Nonspecific illness is associated with feelings of distress (symptoms of anxiety and depression) [24, 31] and tends to be associated with higher healthcare use [4, 8] and worse quality of life than illnesses with measurable pathophysiology [1]. Nonspecific illness may signal opportunities for interventions to bolster mental and social health [30]. We may also be in the habit of applying diagnoses that specify pathophysiology rather than descriptive diagnoses of nonspecific illness; for example, lumbar strain rather than low back pain, or tendinitis instead of idiopathic, activity-related forearm pain. These habits and tendencies may be partly innate, partly conditioned by training and observation of other clinicians, partly related to interactions with patients, and partly reinforced by diagnostic coding systems that reinforce attempts at specific categorization of symptoms in pathophysiologic terms. Although patients may feel some sense of validation and direction when they receive a specific diagnosis that implies an underlying pathophysiology, there are potential harms if this is done in the absence of objective, verifiable pathophysiology [10]. Additionally, some patients feel frustrated by nonspecific diagnoses, although they were accepted as specific diagnoses overall in one study [26]. Specific diagnoses applied to nonspecific symptoms can cause psychologic harm through false hope, reinforcement of an illness identity, and threats to cherished roles, among other factors. There is also the potential for iatrogenic and financial harm through unhelpful tests and treatments [23].

A tendency to describe illnesses characterized by no detectable pathophysiology using anatomic or pathophysiologic terms could reflect an unconscious preference for the biomedical paradigm in which all symptoms are reduced to specific, measurable pathophysiology. However, little is known about whether clinicians exhibit this unconscious preference for specific, identifiable diseases. In a pilot study, we created an Implicit Association Test (IAT) that could only be administered from a single device necessitating a relatively small cohort [21]. That study identified implicit and explicit preferences of musculoskeletal specialists for specific pathophysiology over nonspecific illness [21]. We developed a new internet-based version of the IAT that allowed us to study a much larger cohort of practicing surgeons. Using this IAT, we measured the implicit and explicit attitudes of musculoskeletal surgeons regarding specific and nonspecific illness. Although some have concerns about the validity and utility of the IAT [5, 29], we believe this was the right tool given that the timed delays in association that form the basis of the measurement likely represent bias and social stigma regarding nonspecific illness. If an implicit bias toward specific disease is confirmed, independent of explicit preference, it would represent a potential target for strategies and processes that can bolster effective, evidence-based, comprehensive care.

Using a novel IAT to identify implicit and explicit preferences of musculoskeletal surgeons for specific diseases or nonspecific illness, we sought to answer the following questions: (1) What factors are associated with surgeon implicit preference for specific disease over nonspecific illness? (2) What factors are associated with surgeon explicit preference for specific disease over nonspecific illness? (3) Is there a relationship between surgeon implicit and explicit preferences for specific disease over nonspecific illness?

Patients and Methods

Study Design, Setting, and Participants

We conducted a survey-based experiment among the members of the Science of Variation Group, an international collaboration of musculoskeletal surgeons. We welcome new members and collaborators. Among the approximately 200 members who participate in at least one survey-based experiment per year, 103 completed this survey. Most were men (92% [95]), and most had a role in supervising trainees (83% [85]) (Table 1).

Table 1.

Surgeon demographics

Variable	Total (n = 103)
Gender
Women	8 (8)
Practice location
United States	51 (53)
Europe	34 (35)
Other	15 (15)
Years in practice
0 to 5	19 (20)
6 to 10	23 (24)
11 to 20	27 (28)
More than 20	30 (31)
Subspecialty
Hand and wrist	40 (41)
Trauma	31 (32)
Shoulder and elbow	13 (13)
General orthopaedics or other	17 (17)
Supervises trainees	83 (85)

Data presented as % (n).

The survey was administered using Qualtrics (Qualtrics). In addition to demographic, training, and experience factors, the survey included an IAT constructed using IATgen software (IATgen) [7]. The IAT included terms referring to nonspecific musculoskeletal illnesses, such as fibromyalgia and repetitive strain injury, and terms referring to specific orthopaedic diagnoses, such as meniscus tear and scoliosis (Supplemental Table 1; http://links.lww.com/CORR/B261). It also included positive terms such as “excellent” and negative terms such as “unimportant.” During the IAT, participants were shown a word in the middle of their computer screen with two words in the upper right and left corners. They were instructed to use computer keys to associate the center word with one of the words in the upper corners. The IAT consisted of practice and test rounds. In the first practice round, participants were shown positive and negative terms in the center and were asked to click to associate the term as positive (left) or negative (right). In the next practice round, participants completed the same task with diagnoses in the middle and clicking right to designate the diagnoses as nonspecific or left to designate it as specific. In the test round, the middle words were diagnoses and participants were asked to associate them with nonspecific or positive in one corner and specific or negative in the other. The final rounds presented specific or positive words in the left upper corner and nonspecific or negative in the right upper corner for association with diagnostic terms in the middle. The Science of Variation Group member database was used to collect surgeon characteristics, including gender, practice location, years in practice, subspecialty, and wither a surgeon supervises trainees.

Primary and Secondary Study Outcomes

Our primary outcome measure was the degree of implicit preference for specific versus nonspecific illness, as measured by a participant’s IAT D-score. IAT D-scores that were transformed onto a scale ranging from -150 to 150 were computed as a difference in mean response latency between nonspecific and specific diagnosis items divided by the participant’s response latency standard deviation. Higher IAT D-scores indicate greater tendency toward specific diseases, and lower IAT D-scores indicate greater tendency toward nonspecific illnesses. Using the analogy of an effect size, we considered scores to be notable if the median value was above zero (neutral) by more than twice the magnitude of the interquartile range.

Our secondary outcome measure was explicit preference. Explicit preference was evaluated after completing the IAT using four items that we feel quantify a surgeon’s relative comfort with the biopsychosocial paradigm in which symptoms do not correspond precisely with pathophysiology, and for some illnesses there is no measurable pathophysiology. These seem to have face validity and represent a preference for specific pathophysiology along with a sense that symptoms directly relate to pathophysiology. The results might also demonstrate an explicit preference for conceptualizing symptoms or symptom intensity as not completely accounted for by pathophysiology. These cannot be externally validated because there are no other similar measures for comparison or correlation. Items included “I am doing the most benefit to my patients when I can fix something that is physically wrong,” “I feel uncomfortable when treating patients with no clearly defined illness,” “My patients with a clear diagnosis are typically suffering more than my patients with an unclear diagnosis,” and “Variation in patients’ pain and physical limitation is best explained by pathophysiology, like the grade of osteoarthritis, for example.” Participants responded to each item on a scale from 0 (completely disagree) to 10 (completely agree), with final scores scaled from -150 to 150. In accordance with the IAT D-scores, a higher explicit preference score indicates more tendency toward specific diseases, and a lower score indicates more tendency toward nonspecific illnesses.

Ethical Approval

We obtained approval for this study from the institutional review board at the University of Texas at Austin, Austin, TX, USA (protocol number 2020-05-0040).

Statistical Analysis

Descriptive statistics were collected from all participants. Discrete variables are reported as percentage (number). We used the mean when the distribution of the variable was parametric and the median when the variable was nonparametric. We did not do an a priori power analysis because we had a fixed pool of respondents from which to draw. When analyzing relationships between implicit or explicit bias scores and binary variables, we used Mann-Whitney tests. We used Kruskal-Wallis tests to analyze relationships between implicit or explicit bias scores and nonbinary variables. In the multivariable analysis, the response variable was an explicit preference for specific disease over nonspecific illness. The explanatory variables were practice location (continent) and whether a surgeon supervises trainees. Given the limited number of participants, we felt it was advisable to develop a parsimonious multivariable model by preselecting included variables in a bivariate analysis to help limit the potential for overfitting. We explored factors such as gender, practice location, years in practice, subspecialty, and role in supervising trainees in bivariable testing and advanced those with p < 0.10 to a multivariable linear regression analysis. Practice location and role in supervising trainees were advanced and considered in that model. Regression coefficients, 95% confidence intervals (CIs), standard errors, p values, partial r-squared values, and adjusted r-squared values are reported.

Results

Factors Associated With the Degree of Implicit Preference for Specific Disease

We found that only the hand and wrist subspecialty was associated with surgeon implicit preference for specific disease compared with nonspecific illness overall (Supplemental Table 2; http://links.lww.com/CORR/B262). Overall, there was a notable implicit bias for specific diseases over nonspecific illness (median [IQR] 70 [54 to 88]; p < 0.001) (Fig. 1).

Fig. 1.

This histogram of surgeon scores for implicit preference demonstrates a notable preference for specific disease over nonspecific illness, scored on a scale from -150 to 150.

Factors Associated With the Degree of Explicit Preference for Specific Disease

Accounting for potential confounding variables including role in supervising trainees, the multivariable analysis found that a greater explicit preference toward specific disease was independently associated with not practicing in the United States (regression coefficient -36 [95% CI -60 to -12]; p = 0.003) (Table 2). We found no clinically important explicit preference between specific disease and nonspecific illness (median 8 [-15 to 37]; p = 0.02) (Fig. 2).

Table 2.

Multivariable linear regression analysis of surgeon factors associated with explicit preference for specific disease over nonspecific illness^a

Variable	Regression coefficient (95% CI)	Partial r2	p value
Practice location
Others	Reference value
United States	-36 (-60 to -12)	0.09	0.003
Europe	-23 (-48 to 2)	0.034	0.07
Supervising trainees
No	Reference value
Yes	16 (-6 to 38)	0.02	0.15

^aThe adjusted r-squared value for the model was 0.092, which reflects the amount of variation in explicit preference accounted for by the entire model.

Partial r-squared is a measure of the variation in explicit preference accounted for by that variable; a partial r-squared of 0.09 means 9% is accounted for.

Fig. 2.

This histogram for surgeon scores for explicit preference shows, overall, no preference for specific disease over nonspecific illness, scored on a scale from -150 to 150.

Correlation Between Implicit Bias and Explicit Preference

There was no correlation between explicit preference and implicit bias toward a specific disease over nonspecific illness (Spearman correlation coefficient -0.13; p = 0.20) (Fig. 3).

Fig. 3.

This scatter plot shows no correlation between implicit and explicit surgeon preference for the specific disease over nonspecific illness. There is a wide range of explicit preference at each level of implicit preference.

Discussion

Musculoskeletal illnesses are often nonspecific, meaning that the symptoms cannot be clearly associated with a discrete pathophysiology. However, nonspecific illness could be associated with social stigma that may bias people in favor of specific diseases. Strategies to label a nonspecific illness with a specific term (such as fibromyalgia or pronator syndrome) may reflect, in part, an unconscious bias of humans toward the biomedical paradigm in which all symptoms can be reduced to specific, measurable pathophysiology. A bias in favor of the biomedical paradigm and diagnosis of specific pathophysiology might contribute to overtesting and overdiagnosis when a specific pathophysiology is unlikely. Using an IAT, we found that although musculoskeletal specialists overall do not have an explicit preference for specific pathophysiology over nonspecific illness, they have a notable implicit bias for specific disease, with no correlation between the two and no meaningful associations identified.

Limitations

First, we created the IAT used in this study. Although it might be possible to create an IAT with greater rigor, we find that IATs are straightforward to construct and interpret, and we believe the results of this experiment are unlikely to change much with modifications to the IAT. Second, there are some considerations regarding interpretation of an IAT [5, 29]. First, there is evidence that an IAT measures salience (noticeability or prominence) in addition to implicit associations. Second, IATs have moderate reliability. The somewhat limited reliability may be due, in part, to participants attempting to choose more socially acceptable responses. Third, the findings could be specific to the words and diseases included in the IAT and may not be generalizable to all nonspecific illnesses and specific diseases. This seems unlikely given that the terms we chose reflect a larger group of representative diseases and illnesses. Nevertheless, additional evidence using other terms may help test this assumption. Fourth, these results could apply best to the participating surgeons or to surgeons with similar characteristics (White, men, and academic). The reader can decide whether these are human traits that likely apply to surgeons similar to them, which seems probable. In any case, awareness that an international group of mostly academic surgeons has a bias for specific pathophysiology informs areas of debate in the profession. Fifth, the explicit preference questions addressed unclear diagnosis and no measurable pathophysiology, rather than directly addressing specific illness labels. This was intentional because we did not want to measure participants’ perceptions of specific illnesses but rather their feelings about the circumstances that characterize those illnesses. Lastly, the implicit and explicit measures used different measurement tools, which might influence correlation. On the other hand, the use of different measurement strategies limits the potential for common method variance, which can overestimate the correlation [19]. Previous studies have indicated this could result in a lower correlation between implicit biases and explicit preferences [11, 25].

Factors Associated With the Degree of Implicit Preference for Specific Disease

The finding that greater implicit surgeon preference toward specific disease was modestly associated with the hand and wrist subspecialty might be spurious or specific to the participant sample, or it might reflect aspects of what attracts people to the specialty. The finding of an implicit bias toward specific diseases and away from nonspecific illness reflects aspects of both the patient and clinician experiences of care, as documented in prior studies. For instance, a review of patient-clinician interactions for patients with illness characterized by daily pain, often without a specific pathophysiology, found that patient priorities of feeling understood and legitimized contrasted with a clinician focus on diagnosis and treatment [12]. The clinician’s drive to identify a specific associated pathophysiology can be frustrating and fruitless. For instance, a survey study of one common nonspecific illness (fibromyalgia) identified low physician confidence regarding disambiguating symptoms, applying a diagnosis, and formulating a treatment plan [9]. That is, the availability of a specific diagnosis (fibromyalgia) for a nonspecific illness (widespread, daily, unexplained pain with associated incapability) did not make the process of care easier or more comfortable. Another difficulty is that the communication strategies for helping a patient feel understood and legitimized in the context of a nonspecific illness are not obvious or straightforward. For instance, a systematic review of communication strategies for diagnostic uncertainty (an alternative framing of nonspecific illness) in primary care found two common communication tactics: empathy and exclusion of serious diagnoses [9]. Empathy was more positively regarded by patients, but the overall results were mixed. Most specialists know from experience that effective communication strategies do not eliminate disappointing visits for people with nonspecific illness who may feel their only hope is to find the problem and fix it [6, 13, 14]. Some qualitative evidence suggests that people may not achieve the meaning they desire from applying a diagnostic label [2, 14]. Overall, patients receiving upper extremity specialty care have comparable experiences, regardless of whether they receive a specific or nonspecific diagnosis [26]. The bias toward specific pathophysiology and away from nonspecific illness may result from aspects of clinician discomfort and perhaps some unpleasant experiences of a relatively small group of patients who have a notable aversion to a nonspecific diagnosis. In a systematic review of qualitative studies of people receiving care for low back pain, people hoped clinicians would pay as much attention to them as an individual as they did to diagnosis and treatment [18]. It might be helpful for others to confirm the identified bias of clinicians toward specific disease with various methods and in varied settings.

Factors Associated With the Degree of Explicit Preference for Specific Disease

The finding of no explicit preference for specific disease suggests that specialists perceive themselves as open to all types of illness. It could also be a limitation of our measure of explicit preferences, but that seems less likely, given that the questions were straightforward and unambiguous. The association of explicit preference toward specific disease with surgeons practicing outside of the United States might be spurious or specific to the sample of participants, or it could reflect cultural variations. For instance, explicit preferences could be due to variations in social desirability bias, which may vary by region. In IAT studies of racial preference, a social desirability bias was observed in the gap between a measured implicit preference toward White individuals and a rated explicit preference that is either neutral or in favor of non-White races [17, 20].

Correlation Between Implicit Bias and Explicit Preference

The finding that explicit preference and implicit bias did not correlate is consistent with studies of implicit and explicit biases regarding mental illness [27, 28] and racial prejudice [25]. In our view, the gap between clinician implicit preference for specific illness (IAT) and clinician-rated (explicit) lack of a preference, and the lack of correlation between implicit and explicit preferences has face validity considering the evidence of clinician discomfort with nonspecific diagnoses reviewed above. This gap may represent a clinician blind spot that may be associated with low-value and potentially harmful test and treatment strategies. Clinician awareness of this overall implicit preference for specific diagnoses, independent of explicit, expressed preferences, is a necessary first step. As a next step, clinicians can work to develop effective communication strategies that prioritize the patient’s goals of feeling understood and legitimized independent of a specific diagnosis. At the same time, clinicians can develop checklists and other debiasing strategies that can help limit the use of low-yield, low-value, potentially harmful tests and treatments in the setting of nonspecific illness. These approaches merit further study.

Conclusion

We documented an implicit bias for specific diseases that clinicians might be unaware of, given the lack of correlation with explicitly rated preferences. In part because of this implicit bias, clinicians may find it relatively difficult to navigate communication strategies and maintain an open mind in their encounters with people experiencing a nonspecific illness. Among other factors such as billing strategies and worries about missed diagnoses, the measured implicit bias toward specific diseases over nonspecific illness that was measured in this experiment may account, in part, for patterns of use of low-yield, potentially harmful diagnostic testing in a search of specific pathophysiology and the use of diagnostic labels that imply specific pathophysiology when none is detectable. Patients and clinicians can anticipate and account for an implicit preference for specific disease by anchoring on nonspecific illness until a specific pathophysiology accounting for symptoms is identified, and by using nonspecific illness descriptions in the meantime. This may be a difficult, culture-changing endeavor, but given the prevalence of symptoms that cannot be associated with specific pathophysiology, it is worth investigating.

Supplementary Material

abjs-482-648-s001.docx ^{(22.5KB, docx)}

abjs-482-648-s002.docx ^{(25KB, docx)}

abjs-482-648-s003.docx ^{(215.2KB, docx)}