To examine whether medical decisions regarding evaluation and management of musculoskeletal pain conditions varied systematically by characteristics of the patient or provider.
To examine whether medical decisions regarding evaluation and management of musculoskeletal pain conditions varied systematically by characteristics of the patient or provider.
We conducted a balanced factorial experiment among primary care physicians in the US. Physicians (n = 192) viewed 2 videos of different patients (actors) presenting with pain: undiagnosed sciatica symptoms or diagnosed knee osteoarthritis. Systematic variations in patient gender, socioeconomic status, and race and physician gender and experience (<20 versus ≥20 years in practice) permitted estimation of unconfounded effects. Analysis of variance was used to evaluate associations between patient or provider attributes and clinical decisions. Quality of decisions was defined based on the current recommendations of the American College of Rheumatology, American Pain Society, and clinical expert consensus.
Despite current recommendations, less than one-third of physicians would provide exercise advice (30.2% for osteoarthritis and 32.8% for sciatica). Physicians with fewer years in practice were more likely to provide advice on lifestyle changes, particularly exercise (P ≤ 0.01), and to prescribe nonsteroidal antiinflammatory drugs for pain relief, both of which were appropriate and consistent with current recommendations for care. Newer physicians ordered fewer tests, particularly basic laboratory investigations or urinalysis. Test ordering decreased as organizational emphasis on business or profits increased. Patient factors and physician gender had no consistent effects on pain evaluation or treatment.
Physician education on disease management recommendations regarding exercise and analgesics and implementation of quality measures may be useful, particularly for physicians with more years in practice.
The Institute of Medicine has called for more aggressive treatment of pain, including arthritis and musculoskeletal pain (). However, pain management is complicated by factors such as the potential toxicity of nonsteroidal antiinflammatory drugs (NSAIDs) or the risk of abuse and diversion of opioids (). Furthermore, the scientific understanding of pain pathology is incomplete; definitive standards for pain management applicable to all patients with a given set of symptoms or diagnosis do not exist, reflecting the inconsistency and sparseness of the evidence base. For example, in 2012, the American College of Rheumatology (ACR) provided “conditional recommendations” for the most commonly used therapies (e.g., NSAIDs) for knee osteoarthritis (OA) and no recommendations for intraarticular hyaluronates, duloxetine, or opioid analgesics. “Strong recommendations” were given only for physical exercise (cardiovascular and/or resistance exercise) and weight loss (for persons who are overweight) ().
Given limited clinical recommendations, it is not surprising that pain management decisions regarding opioids and other analgesics vary widely in practice ([4-8]). In particular, undertreatment of pain has been documented for women, African Americans, and Hispanics ([9-11]). Prescribing also reportedly varies by physician gender, specialty, and litigation fears ([12, 13]). Where there are strong recommendations, e.g., exercise for knee OA () or remaining active for low back pain (), there should be minimal variation in physicians' decision making. Particularly in such cases, understanding the sources of variation (e.g., provider experience, patient race) is important to assess and promote quality of care. Currently, most studies on variations in pain management focus on narcotic prescribing, yet attention to actions regarding patient education and advice is also essential given their beneficial effects on patient outcomes ([15-19]).
A difficulty with research attempting to identify extraneous factors in decision making is that numerous intricately related factors comprise patient presentations and physicians' backgrounds. Observational research inevitably produces confounded estimates of these various influences. For example, patient race/ethnicity and socioeconomic status (SES) are strongly correlated in the US ([20, 21]), and both have been associated with pain management ([7, 22]). The importance of disentangling their effects on pain care is that public health implications for policy changes and improvements in quality monitoring may be notably different if pain treatment decisions are influenced, for example, by educational level rather than race ([5, 21, 23]).
The objective of this analysis was to examine whether the quality of care regarding advice on exercise and weight loss, as well as decisions on imaging and analgesic prescribing, varied systematically by patient race, gender, or SES, or by the clinical experience, gender, or work environment of the physician. In light of the intricate relationships among these variables, we conducted a factorial experiment that provided unconfounded effects of 3 patient factors (race/ethnicity, SES, and gender) and 2 physician factors (gender and years of clinical experience) on the quality of clinical decisions regarding diagnosis and management of 2 common musculoskeletal conditions. Therefore, we were able to examine the distinct influences and unique interactions of these patient and physician factors on quality of care. Given the lack of clinical guidelines for pain management, quality of care is pragmatically defined herein as clinically appropriate actions, based on the current recommendations of the ACR and American Pain Society; similar approaches have been shown to provide acceptable measures of quality in health care ([24-26]).
The study design was a balanced factorial experiment, which permits the estimation of unconfounded main effects and interactions of patient and provider factors that varied by design. Six factors varied by design: patient gender, race/ethnicity (African American, Hispanic, or white), SES (lower or higher), and request for a particular pain medication (oxycodone or celecoxib; yes/no request), and physician gender and years of clinical experience (<20 or ≥20 years since medical school graduation). In previous analyses, we presented findings that patient requests for medication influenced physician prescribing decisions, to comply with the request ().
The experiment was conducted among practicing physicians identified from lists of licensed physicians in Illinois, Indiana, Massachusetts, Missouri, New Hampshire, and Rhode Island. To be eligible, physicians completed a medical residency program in internal medicine or family practice, were licensed as a primary care physician, and provided clinical care at least part time. Primary care physicians were of interest because, as the gateway to treatment and referrals, their decisions determine the course for patient care and outcomes. During recruitment, physicians were classified according to gender and year of graduation from medical school (1970–1991 versus 1992–2002, for a cut point of ∼20 years of clinical experience) and purposefully recruited until each combination needed for the factorial design was complete. Eligible physicians were telephoned and invited for in-person interviews to obtain written informed consent. Physicians were informed that the study was designed to examine how primary care doctors manage pain conditions. The study protocol was approved by the Institutional Review Board of the New England Research Institutes.
Physicians were randomized to watch 2 vignettes featuring 2 different patient presentations. One vignette presented an undiagnosed actor/patient discussing signs and symptoms strongly suggestive of sciatica (referred to as the “sciatica patient”). The other vignette presented an actor/patient with previously diagnosed OA of the knee requiring ongoing case management (referred to as the “OA patient”). Details on the symptoms presented in each case are listed in Table 1.
|Sciatica patient||Knee osteoarthritis patient|
|Inappropriate treatment||Inappropriate treatment|
|MRI for evaluationa||MRI for evaluationa|
|Narcotic medicationb||Narcotic medicationb|
To create the patient presentations, 6 professional actors/actresses were trained to portray a patient presenting to a primary care doctor. Scripts for the scenarios of interest were developed from tape-recorded role-playing sessions with clinically active physicians (including authors JNK and MAF). The scenario was repeatedly filmed, systematically changing the actor/patient's race (African American, Hispanic, white), gender, and SES (lower versus higher social class, e.g., a truck driver versus a sales representative for sciatica and a janitor versus a lawyer for OA, also expressed by clothing style), and whether he or she requested a particular pain medication. Each video simulated an initial consultation with a primary care physician and was 3–5 minutes in duration, reflecting the average length for initial history in an office consultation without physical examination and without physician interruptions for questioning (). Observation of the videos by other primary care providers confirmed their clinical authenticity.
Immediately after the physician subjects viewed each vignette, trained study staff members conducted a semistructured interview, recording responses using predefined categories and open-text fields. The data collected included a range of decisions as they relate to each clinical scenario, including the tests to be ordered, medication to be prescribed, referrals, and timeframe for the next visit. Advice to be given to the patient (including exercise and weight loss) was obtained by asking, “Would you advise the patient about lifestyle or behavior today?” with a followup question, “What would you advise for this patient today?”
There is no single gold standard for optimal treatment of these musculoskeletal conditions that could be used to strictly classify adherence to guidelines. The quality of decision making was assessed using published clinical practice recommendations of the ACR for OA of the knee () and of the American College of Physicians and American Pain Society for low back pain (). These clinical practice guidelines encapsulate a rigorous review of the published literature; therefore, they are generally congruent with practice guidelines. We chose to base our assessments on practice guidelines rather than quality measurement tools because the latter often are a blend of published evidence and expert opinion. A summary of the recommendations as they relate to the particular cases in this experiment is provided in Table 1. Physicians were not informed of the operational definition of quality of care used in the experiment.
After the interview, physicians completed a self-administered questionnaire on their personal background, medical training, practice setting, and use of clinical guidelines. Practice culture was measured by an abbreviated version of the standardized Medical Group Practice Culture Survey (), which was designed for use among primary care physicians and has been shown to have predictive value in small- and large-group practice environments. It includes 10 Likert-scaled items measuring 9 cultural dimensions; of particular interest in this analysis of clinical decision making were the dimensions of “business emphasis,” “information emphasis,” “quality emphasis,” and “organizational trust.” Satisfaction with career and current job were assessed using the global scales of the validated Physician Worklife Survey ([30, 31]).
Analysis of variance was used to test the effects of the design variables on a range of diagnostic and treatment decisions. The balanced factorial design produced 3 × 22 = 12 patient sociodemographic characteristic (race, gender, and SES) combinations. Each combination was portrayed twice to accommodate the drug request, yielding 24 distinct vignettes for each musculoskeletal condition. The 2 physician factors (gender and experience) created 4 strata. Within each stratum, 48 physician subjects were randomly assigned to view 1 of the 24 pairs of vignettes. In additional analyses, covariates (i.e., physician characteristics such as race/ethnicity, practice setting, and practice culture) were examined using logistic regression. A generalized coefficient of determination (R2) statistic was used to determine the proportion of variation in the outcome that was explained by the independent variables in the model. Due to the challenges of multiple testing, we emphasized consistency across the results.
The total sample of 192 physicians provided 80% power at a 0.05 level to detect an absolute difference in means of 0.2 SD, which is considered a small effect size. For example, for analysis of physician gender and advising patients regarding exercise, if the rate of giving exercise advice were 0.25, the study was powered to detect a difference in proportions between male and female physicians of 0.09, which results in a gender difference in rates of providing exercise advice ranging from 0.16 (0.25 − 0.09) to 0.34 (0.25 + 0.09).
Of the 192 physicians randomized in this experiment, most (79.7%) worked in a shared office practice (median 4 staff physicians), and the mean time at their current practice was 10.8 years. Twenty-two percent of physicians had been in practice for >30 years, whereas 20.8% had graduated within the past 10 years. The physicians' ages ranged from 35–74 years, with mean ± SD ages of 41.4 ± 4.5 years for physicians classified as having fewer years in practice and 57.7 ± 5.7 years for physicians classified as more experienced. Similar proportions reported being in family practice (49.5%) or internal medicine (44.8%); the remaining 5.7% were identified as general practitioners. Race/ethnicity varied, with 55.2% white, 24.5% Asian, 7.3% African American, and 5.7% of Hispanic ethnicity. Only 7.3% reported that they generally do not use clinical guidelines in the management of patients. Sixty-five percent of physicians reported that their knowledge of guidelines affected their decision making for the patient presentations at hand.
After viewing the video vignettes, 95% of the physicians reported the patient presentations viewed were typical of those seen in their everyday practice. The task of providing the most probable diagnosis for the sciatica patient was appropriately completed by most physicians: sciatica or a related condition (e.g., herniated disc) was listed as one of the possible diagnoses by 93.7% of physicians and was the most probable diagnosis of 64.1% of physicians.
Physician gender was not associated with decisions for either patient scenario and had no statistically significant interactions with patient gender. In contrast, physician length of time in practice was associated with the quality of clinical decisions. Newer physicians were more likely to give advice on lifestyle behaviors (P = 0.01), particularly regarding exercise habits (39.6% of newer physicians versus 26.0% of more experienced physicians for the sciatica patient or versus 20.8% of more experienced physicians for the OA patient). Results were similar for advice to lose weight and referrals for physical therapy (Figure 1). Additional analyses exploring the nature of the relationship indicated that there was a continuous linear inverse trend between the length of time in practice and the probability of giving such advice (P ≤ 0.01) (data not shown). In addition, lifestyle advice for knee OA was twice as likely to be provided by physicians who worked in organizations with greater emphasis on quality of care (odds ratio 2.35, 95% confidence interval 1.12–4.92; P = 0.01). Other work place culture values were not associated with lifestyle counseling.
In addition to advice and referrals, analgesic prescribing and test ordering were associated with physicians' years of experience and other provider characteristics (Figure 2). Newer physicians more commonly would prescribe NSAIDs for pain relief (68.8% versus 52.1%; P = 0.01 for the sciatica patient and 80.2% versus 67.7%; P = 0.02 for the OA patient). The total number of tests ordered was, on average, slightly lower among newer physicians (mean 1.9 versus 2.5 tests; P = 0.01 for the sciatica patient and mean 2.4 versus 2.9 tests; P = 0.07 for the OA patient), who less frequently would order basic laboratory work (e.g., complete blood count or metabolic panel, 9.4% versus 21.9%; P = 0.02) and urinalysis (4.2% versus 16.7%; P = 0.003), particularly for the sciatica patient. For the OA patient only, radiographs were more often ordered by newer physicians (85.4% versus 69.8%).
In addition to years since training, physicians who received their medical degree in the US (n = 127 [66.1%] of physician subjects) ordered fewer tests than did the foreign medical school graduates (mean ± SD 1.9 ± 1.4 versus 2.8 ± 1.8 tests; P = 0.003 for the sciatica patient), particularly basic laboratory work (8.1% versus 30.2%; P = 0.002). Test ordering also decreased as the organizational emphasis on business decisions and profit maximization increased (P = 0.02 for the OA patient); each unit increase in business emphasis score was associated with 70% lower odds of test ordering. Similarly, physicians practicing in for-profit organizations were less likely to order any tests/laboratory evaluations compared to physicians from nonprofit organizations (91.4% versus 100%; P = 0.01 for the OA patient) (data not shown).
Although magnetic resonance imaging (MRI) is considered inappropriate for these patient scenarios, 33.9% of physicians would order MRI for the sciatica patient and 13.5% of physicians would order MRI for the OA patient. For the OA patient only, MRI use was more frequent among physicians who generally do not use clinical guidelines in the management of their patients (21.4% versus 12.9%; P = 0.04). No associations were found between physician job or career satisfaction, payment structure (e.g., paid by salary or productivity), practice type (family practitioner, general practitioner, or internal medicine), practice size, or race/ethnicity and any of the decision outcomes (data not shown).
Overall, evaluation and treatment plans showed no consistent variation by patient factors (Table 2). For those with sciatica pain, muscle relaxants were more commonly prescribed to women and basic laboratory work was ordered more often for men. Patient SES was associated with the type of pain medication prescribed for the sciatica patient: narcotic medications (considered inappropriate for these scenarios) were more commonly chosen for patients of higher SES (68.7% versus 52.1%; P = 0.01), whereas NSAIDs, which are more appropriate, were more commonly mentioned for patients of lower SES (67.7% versus 53.1%; P = 0.03).
|Patient gender||Patient race/ethnicity||Patient SES|
|Avoid bending, lifting, etc.||28.1||30.2||0.8||23.4||35.9||28.1||0.3||30.2||28.1||0.8|
|Order tests/laboratory evaluations||64.6||77.1||0.05||78.1||71.9||62.5||0.14||72.9||68.8||0.5|
|No. of tests ordered, mean ± SD||2.2 ± 1.7||2.2 ± 1.6||0.7||2.0 ± 1.5||2.0 ± 1.1||2.7 ± 2.0||0.13||2.3 ± 1.7||2.2 ± 1.5||0.7|
|Basic laboratory workb||10.4||20.8||0.04||14.1||12.5||20.3||0.4||15.6||15.6||1.0|
|Avoid bending, lifting, etc.||12.5||7.3||0.3||15.6||7.8||6.3||0.2||10.4||9.4||0.8|
|Continue normal physical activity||10.4||17.7||0.16||12.5||20.3||9.4||0.2||15.6||12.5||0.6|
|Order tests/laboratory evaluations||94.8||94.8||1.0||96.9||87.6||100.0||0.01||93.8||95.8||0.5|
|No. of tests ordered, mean ± SD||2.7 ± 2.0||2.6 ± 1.9||0.04||2.8 ± 1.9||2.5 ± 2.1||2.6 ± 1.9||0.9||2.6 ± 1.9||2.6 ± 2.0||0.8|
|Basic laboratory workb||35.4||30.2||0.4||39.1||25.0||34.4||0.3||32.3||33.3||0.9|
This effect of patient SES was not observed for the OA patient. However, race was associated with a narcotic prescription for OA patients: narcotics were prescribed to white patients (46.9%) more than Hispanic (32.8%) or African American (26.6%) patients (P = 0.03). The only other racial/ethnic difference was testing for OA: 87.6% of Hispanic patients would receive any laboratory testing or imaging compared to 100% of white patients and 96.9% of African American patients (P = 0.01).
We evaluated the total contribution of patient (gender, race, SES, and medication request) and provider (gender and experience) design factors, together with personal (race, medical school training, guideline use, and job and career satisfaction) and organizational characteristics (profit/nonprofit, percent managed care, and practice culture emphasis on business, quality, information, and trust), in explaining variation in decision making. Results showed that the decision to provide exercise advice remained largely unexplained even after accounting for these various factors; 13.4% of the variation was explained by these factors for the OA patient and 19.9% for the sciatica patient. Similarly, 12.9% (OA) and 21.1% (sciatica) and 13.7% (OA) and 13.9% (sciatica) of the variation in decision making was explained by these factors for test ordering and for narcotic medication prescribing, respectively.
This health services factorial experiment found considerable variation in decision making for painful musculoskeletal conditions, yet little of this variation was explained by patient factors of race, SES, or gender. Rather, 2 factors that most consistently affected the quality of treatment decisions were 1) the physicians' years in practice, whereby newer physicians more often followed clinical practice recommendations to provide exercise advice or prescribed NSAIDs for pain relief, which was appropriate for the case patients, and 2) organizational values, whereby emphasis on quality of care was associated with providing lifestyle advice. These findings indicate that adherence to current clinical practice recommendations for patients with musculoskeletal pain is more likely to occur among newer physicians, regardless of patient characteristics, and that certain organizational practice cultures may influence pain care.
The possibility that newer physicians are more up to date on recommendations or guidelines from medical societies is consistent with prior studies ([32-34]). In our experiment, physicians' years in practice had effects on patient counseling, prescribing, and test ordering. Newer physicians were almost twice as likely to give lifestyle advice, particularly for exercise, to the OA patient. Currently, exercise (cardiovascular or resistance) is the only “strong recommendation” for knee OA treatment from the ACR (). This difference may represent a cohort effect during medical training. For low back pain, advice to “remain active” became accepted in the last 2 decades, replacing common earlier beliefs of need for bed rest ([14, 15]). Similarly, our finding that newer physicians more commonly prescribed NSAIDs supports the possibility that newer physicians may be more aware of and/or more inclined to follow recommendations; oral and topical NSAIDs were “conditionally recommended” for use for knee OA and considered as “first-line medication options” for low back pain ().
Of note, 1 in 3 physicians would order MRI for the sciatica patient. Use of MRI for either patient case contrasts with current evidence-based recommendations (). Overuse of MRI has been a prominent concern in the medical community, for reasons including 1) lack of evidence of improved outcomes, 2) identification of abnormalities that are uncorrelated with symptoms, and 3) the possibility of leading to additional unnecessary interventions and health care costs (). We found that MRI for knee OA was more frequently ordered by physicians who reported that they generally do not use clinical guidelines for patient management. Also, the total number of ordered tests was lower among newer physicians and among US medical school–trained versus foreign-trained physicians. A speculative explanation for these findings is that a greater concern for cost control and adherence to society recommendations leads to ordering of fewer tests. The inverse association between organizational cultures emphasizing business profits and test ordering also suggests that cost of care was considered. Interpretation of these results is problematic because the absolute number of tests ordered is not a definitive quality outcome measure. For example, we found that organizational culture emphasizing quality of care was not associated with test ordering, yet was associated with providing lifestyle advice. Such associations between practice culture and pain management decisions are similar to findings from other therapeutic areas ([35, 36]), but studies on pain care are lacking. Additional quantitative and qualitative research is needed to explore test ordering and the resulting implications on quality and costs of health care.
Prior research has suggested the importance of patient race and SES and provider gender in pain management, such as the receipt of narcotics (). In an earlier experiment using written patient vignettes among emergency room physicians, patient SES had a slight effect on narcotic prescriptions, whereas race/ethnicity had no effect (). In our experiment, narcotics were inappropriate for both cases, particularly given that the patients' occupations involved motor vehicle operation. Subjects with sciatica and lower SES had a lower likelihood of receiving a narcotic medication and had a greater likelihood of receiving NSAIDs than those with higher SES. The same SES effect was not observed for the OA patient; rather, a lower percentage of African Americans and Hispanics received narcotics compared to whites. Because the same physicians viewed both scenarios, it is possible that influence of race and SES depends on the type of pain presented, but additional research is warranted.
This experiment's use of video vignettes offers distinct advantages over observational studies and previous written vignette experiments: allowing manipulation of factors, strict standardization, inclusion of nonverbal cues, and inclusion of socioemotional components alongside complex symptom presentations ([38-40]). A unique experimental factor in this study was the patient's request for a particular pain medication. Indeed, the patient request influenced pain management, including receipt of that particular medication or others in its class (). Despite this influence, there were no consistent interactions with patient request and the factors of interest in the current analysis, and the current findings were robust to consideration of the patient request.
Although the experimental study design and standardized patient scenarios helped to maintain internal validity of this study, a limitation is that generalizability is hampered because actual interactions between the doctor and the patient were lacking. For decisions that are partly based on patient preferences, the quality of the doctor–patient communication is important. For example, for NSAID use, an interactive explanation of risks and benefits, tempered by patients' preferences and values, would allow an ideal scenario for evaluating quality of care. Also, the study did not simulate a physical examination or allow further questioning, which in the actual clinical setting may influence decisions. However, observational clinical studies are unable to clearly identify which are the key contributors to clinical decision making, and it is logistically difficult, burdensome, and often unethical to use real patients in such research. Therefore, clinical scenarios are commonly used by medical educators and health services researchers and in training efforts to improve quality of care ([41-49]).
In summary, this experiment found variation in the quality of musculoskeletal pain management decisions, particularly to provide exercise and other lifestyle advice, associated with physicians' years in practice and organizational cultural values. Generally, newer physicians had greater adherence to current recommendations. Unlike prior studies, characteristics of patients in terms of race, SES, and gender had no consistent effects. It is possible that racial distinctions have diminished over time, but additional research on physicians' cognitive processes during interactions with patients is warranted. Also, it remains uncertain whether results on musculoskeletal pain decision making apply to patient presentations where clearly defined clinical practice guidelines document right and wrong actions. These findings highlight key areas of clinical practice recommendations that primary care physicians, who often shape the patient's entire course of pain management, are prone to neglect, particularly regarding exercise advice, narcotic prescribing, and imaging.
Overall, the observed variations in decision making were still largely unexplained, even after accounting for all of the patient, provider, and organizational variables that were statistically significant in the multivariable models. The frequency at which physicians made recommendations not supported by current evidence-based recommendations supports prior calls for improved physician education or other interventions that may lead to changes in practice ([50-52]). Methods to more effectively disseminate current recommendations for diagnosis and management of pain conditions should be developed and tested to improve the quality of care for these common clinical problems.
All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Maserejian had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study conception and design. Maserejian, Fischer, Marceau, McKinlay, Katz.
Acquisition of data. Marceau, McKinlay.
Analysis and interpretation of data. Maserejian, Fischer, Trachtenberg, Yu, Katz.
Authors Maserejian, Trachtenberg, Yu, Marceau, and McKinlay are employees of New England Research Institutes, Inc.