1 Introduction

Shoulder pain is one of the most common musculoskeletal complaints in primary care [1, 2]. It occurs in approximately 20–33% of the general population, affects almost one third of adults in a 1-month period, with a female predominance [1, 3, 4]. Many studies demonstrate that the prevalence of shoulder pain increases with age [1, 3,4,5]. Due to the decreased performance in activities of daily life, reduced health-related quality of life, increased use of health care resources, the socioeconomic burden of shoulder pain is substantial [1, 6, 7].

Several evidence-based clinical practice guidelines (CPGs) for the management of shoulder disorders have been published by specialty societies [8,9,10,11,12,13,14].These specialty society–specific CPGs may tend to have greater impact on specialty care for shoulder conditions when compared to the impact on primary care. Although orthopedic specialists who received comprehensive training on the management of shoulder conditions are critical to the treatment of shoulder pain, primary care physicians (PCPs) may play an increasing role as the primary prescriber for the patients with shoulder pain [15,16,17]. Given the challenges of providing appropriate care for shoulder pain in the ambulatory care settings, it is important to determine if there are differences in the content of the services ordered or provided by PCPs and non-primary care physicians (non-PCPs). To our knowledge, there are gaps in the literature regarding the potential disparity in the clinical management of patients with shoulder pain by PCPs and non-PCPs on a national level.

Therefore, the specific objective of our study was to identify the differences in the clinical management of shoulder pain by PCPs and non-PCPs in the United States by using the National Ambulatory Medical Care Survey database (NAMCS) from 2007 to 2019.

2 Materials and methods

2.1 Data source

The NAMCS is a national probability sample survey of visits to nonfederal, office-based physicians in the United States, which is conducted annually by the National Center for Health Statistics (NCHS), Centers for Disease Control and Prevention. The survey utilizes a multistage probability sample design to obtain an unbiased national representative estimation of the provision and use of ambulatory medical care services in the United States. The NAMCS collects the information including the patient demographic characteristics, insurance status, reasons for visit (RFV), physician diagnoses, physician specialty, screenings and procedures ordered, medications prescribed, and non-pharmacological therapeutic services recommended during the visit. The further details on the NAMCS sampling and weighting methods are publicly available on the NAMCS website, as well as processing and quality control procedures. Due to the utilization of publicly available and de-identified data from the NAMCS, this study was considered exempt from the institutional review board.

2.2 Study population

This study included the visits by adults (aged ≥ 18 years) if the primary diagnosis was consistent with a shoulder pain related disorder and one of the top 3 RFV was related to shoulder symptoms. Since NCHS changed the NAMCS coding system from the International Classification of Diseases, Ninth Edition, Clinical Modification (ICD-9-CM) codes to the International Classification of Diseases, Tenth Edition, Clinical Modification (ICD-10-CM) codes in 2016, the comprehensive shoulder-related diagnostic codes of the ICD-9-CM and ICD-10-CM were listed in Appendix 1. Meanwhile, the ambulatory care visits were identified using the RVF classification code number (19,400–19,405, 19,450–19,455) for which shoulder or arm symptoms were reported as one of the top 3 RFV. Due to the heterogeneity of treatment options, we excluded the visits with diagnoses of pregnancy, malignancy or neoplasm, infectious diseases, fracture or dislocation of shoulder. We also omitted the visits classified as “preventative care visit” and “pre-surgery or post-surgery visit”. Visits with missing PCP status were removed from the analysis, which represented 4.3% (97/2223) of the study period. The inclusion and exclusion strategies for identifying the sampled visits were presented in Fig. 1.

Fig. 1
figure 1

The inclusion and exclusion strategies for identifying the shoulder visits from NAMCS 2007–2019

Full size image

2.3 Study variables

The independent variable for the analysis was the primary care physician status (PCPs and non-PCPs). The dependent variables for the analysis included the prescription of pain medications, physical therapy (PT) referral, health education/counseling, and any diagnostic imaging. According to the Multum drug classification, pain medications were classified into NSAIDs (drug codes 061 and 278), narcotic analgesics (drug codes 060 and 191), skeletal muscle relaxants (drug codes 178 and 179), other pain medications (drug codes 059, 062 and 063). The health education/counseling included recommendations for diet/nutrition, exercise, injury prevention, stress management, tobacco use/exposure, weight reduction. Imaging modalities included x-ray, ultrasound, computed tomography (CT), and magnetic resonance imaging (MRI).

For the purpose of performing an adjusted analysis, this study included the data on patient sociodemographic characteristics such as age, sex (male, female), race (white, black, and other), payment type (private insurance, Medicare, Medicaid, worker's compensation, self-pay, and other). We also included clinical variables such as the chronicity of symptoms (new problem, chronic problem routine, chronic problem flare-up), presence of injury related to current visit, patient’s relationship with provider (established patient, new patient), number of medical comorbidities. Additional visit-level covariates included survey year and metropolitan statistical area (MSA) status. Since the data for 2017 had not been released by NCHS, this study was conducted over a recent 12-year period (2007–2016 and 2018–2019). The variables with missing rate higher than 20% were not included in our models because of the inability to make reliable estimates.

2.4 Statistical analysis

This was a secondary analysis of the data from the 2007–2019 NAMCS. To account for the stratified multistage sampling design employed by the NAMCS, all estimates were calculated using survey sample weights to produce national estimates.

Descriptive statistics (mean, standard deviation, frequencies, and percentages) were presented to assess the aforementioned characteristics of the sampled visits. The Rao-Scott Chi-square test was conducted to assess for categorical variables, the Student’s t-test was conducted for continuous variables. Multivariable logistic regression was employed to determine the association between previously defined visit characteristics and services provided by physicians in the ambulatory care for shoulder pain. To limit model overfitting, the variables included in the logistic regression model were the basic demographic characteristics (such as age, sex, race, payment type, MSA status, survey year) and the variables with statistically significant P values in baseline data. Results were presented as odds ratios (OR) with the corresponding 95% confidence intervals (CI). We conducted multiple imputations for missing data. As a sensitivity analysis, the visits with missing responses were removed from the analysis. When excluding these visits, the sensitivity analysis did not alter our results. So, the results presented in our final analyses included the imputed values for missing data. All statistical analyses were performed using the statistical computing language R software (version 4.1.2), with p values < 0.05 considered to be significant.

3 Results

From 2007 to 2019 (except 2017), an estimated 74.43 million ambulatory care visits by adults with shoulder pain were identified in our study, which represented an average of 6.2 million visits annually during this study period. We found that 20.59 million visits or 27.7% of all shoulder visits, were made to PCPs, with an average of 1.7 million visits annually made to PCPs for shoulder pain.

The descriptive statistics of the sampled visits are presented in Table 1. The average age within this study population was 57.1 (± 16.3) years, and females represented 48.1% of all shoulder visits. The majority of visits were made by white patients (82.9%). Approximately 91.2% of the visits occurred in the metropolitan statistical area (MSA). As for source of payment, 48.5% of all shoulder visits were covered by private insurance, followed by Medicare (30.6%). The most common reason for a visit was a new problem (48.1%), while routine follow up visits for chronic problems comprised 32.1% of all visits for shoulder pain. Established patients represented 77.4% of the population, and 31.0% were reported with the presence of injury related to current visit. The most common reported comorbidities included arthritis (52.4%) and hypertension (30.0%). About 16.7% of the visits were associated with greater than 2 comorbidities.

Table 1 Baseline characteristics of visits by adults with shoulder pain, stratified by primary care physician status, NAMCS 2007–2019
Full size table

Several differences in demographic and visit characteristics were quantified in our stratified analyses (Table 1). There was a higher proportion of visits made to PCPs for the reason new condition evaluation (61.8%) compared to non-PCPs (42.8%). About 97.1% of PCP visits were made by established patients, which was significantly higher than proportion of non-PCP visits (69.9%). Meanwhile, a greater proportion of PCP visits (28.9%) were associated with more than 2 comorbidities compared to non-PCP visits (12.1%). Between the two groups, no statistically significant differences were identified with regard to patient age, gender, or race.

An overview of the services ordered or provided by physicians for shoulder pain in ambulatory care visits is shown in Table 2. According to the pharmacologic treatments, the most common prescribed pain medication for shoulder pain was NSAIDs (27.9%). But no statistically significant difference was identified between the visits made to PCPs (28.2%) and non-PCPs (27.7%). Narcotic analgesics were prescribed in 16.2% of all shoulder visits, other pain medications were prescribed in 9.7%, and skeletal muscle relaxants were prescribed in 5.8%. Our results revealed a significant higher frequency of narcotic analgesics prescriptions by PCPs (22.3%) versus non-PCPs (13.8%). During the same period, the visits with skeletal muscle relaxants prescriptions accounted for 9.8% of PCP visits (versus 4.3% of non-PCP visits) and with other pain medications prescriptions accounted for 14.7% of PCP visits (versus 7.8% of non-PCP visits).

Table 2 Utilization of health services during visits for shoulder pain, stratified by primary care physician status, NAMCS 2007–2019
Full size table

According to the non-pharmacologic treatments, PT was prescribed in approximately 19.3% of all shoulder visits, while health education/counseling was provided in 18.8%. This study identified a significantly lower frequency of PT prescribed by PCPs (7.6%) compared to non-PCPs (23.8%). The health education/counseling was provided in approximately the similar frequency to PCP (20.7%) visits and non-PCP visits (18.1%).

Approximately 40% of all shoulder visits involved the use of any medical imaging: 28.2% received an x-ray, 15.3% received an MRI, 3.8% received an ultrasound, 1.0% received a CT scan. We observed differences in the use of MRI by the physician types. The PCP visits received far fewer MRI overall, compared with the visits made to non-PCPs (7.5% versus 18.3%). Due to the low rates of ordering ultrasound or CT, stratified analysis could not be conducted.

We performed a multivariable logistic regression analysis to identify the factors associated with the services provided in ambulatory shoulder care. After adjustment for the covariates, these differences in the management of shoulder pain by different physician specialty types (PCPs versus non-PCPs) remained (Table 3). We found that PCPs had significantly higher odds of prescribing narcotic analgesics (adjusted OR = 1.62, 95% CI: 1.04–2.51), skeletal muscle relaxants (adjusted OR = 2.71, 95% CI: 1.65–4.45), other pain medications (adjusted OR = 1.87, 95% CI: 1.13–3.07) for shoulder visits compared with non-PCPs. Meanwhile, there was a significantly lower odds of prescribing PT by PCPs compared with non-PCPs (adjusted OR = 0.34, 95% CI: 0.21–0.55). PCPs had decreased adjusted odds of ordering MRI for the visits with shoulder pain compared with non-PCPs (adjusted OR = 0.46, 95% CI: 0.25–0.84). The difference in ordering of any imaging (adjusted OR = 0.77, 95% CI: 0.5–1.18) between the PCPs (32.2%) and non-PCPS (43.0%) did not hold true in multivariable logistic regression analysis. By excluding the visits with missing data, the sensitivity analysis did not alter our results (Table 4).

Table 3 Adjusted and unadjusted odds ratios of primary care physician status on the likelihood of service utilization, NAMCS 2007–2019
Full size table
Table 4 Results of sensitivity analysis by removing the visits with missing data
Full size table

Our study observed several other factors associated with the services provided in ambulatory shoulder care. The established patients had significantly lower odds of receiving an MRI (adjusted OR = 0.59, 95% CI: 0.39–0.89), an x-ray (adjusted OR = 0.32,95%, 95% CI: 0.23–0.46), any image (adjusted OR = 0.32, 95% CI: 0.23–0.45), and PT utilization (adjusted OR = 0.58, 95% CI: 0.40–0.85) compared with new patients. Similarly, the routine follow up visits for chronic problems had lower odds of receiving an MRI (adjusted OR = 0.50, 95% CI: 0.32–0.77), an x-ray (adjusted OR = 0.49, 95% CI: 0.35–0.70), any image (adjusted OR = 0.51, 95% CI: 0.36–0.71), NSAIDs prescriptions (adjusted OR = 0.54, 95% CI: 0.30–0.98), and PT utilization (adjusted OR = 0.58, 95% CI 0.40–0.85) compared with the visits for new problems. The visits related to injury had higher odds of receiving an MRI (adjusted OR = 1.61, 95% CI: 1.09–2.38) compared with the visits unrelated to injury.

4 Discussion

From 2007–2019 (except 2017), an estimated 74.43 million ambulatory care visits were made by adults with shoulder pain, which imposed a significant burden to the public health care system in the United States. Nearly one-third of these shoulder visits were made to PCPs, accounting for an estimated 20.59 million visits during this study period. Since the current study methodology included only visits with a primary diagnosis of shoulder disorder and one of the top 3 RVF related to shoulder symptoms, the estimate in our study was lower than the previous published data from NAMCS between 2009 and 2016 [18]. Despite the increased role of primary care in treating shoulder pain [15,16,17], published data suggested that patients with shoulder pain at the primary care level had not received high quality care [17, 19,20,21,22]. Most of these studies were limited to specific populations [21,22,23], or relatively small sample sizes [19, 20, 24]. To our knowledge, this is the first study identified the potential disparity in the clinical management of shoulder pain by PCPs and non-PCPs in the United States based on national-level data [18, 25].

Shoulder diseases encompass a wide range of disorders, including shoulder osteoarthritis, rotator cuff pathology (ranging from tendonitis, and bursitis to rotator cuff tear arthropathy), adhesive capsulitis, sprains and strains of shoulder, acromioclavicular joint dysfunction, calcific tendonitis, and unspecified shoulder disorders. The heterogeneity in the etiology of shoulder complaints poses a significant clinical challenge to the primary care providers. In order to address the management of patients with shoulder disorders, the American Academy of Orthopaedic Surgeons (AAOS) has released and updated evidence-based clinical practice guidelines (CPGs) to guide the clinician's ability to diagnose and treat shoulder osteoarthritis and rotator cuff injuries [8, 9]. According to the published guidelines [8,9,10,11,12,13,14], the optimal decisions regarding the management of shoulder symptoms should be made based on a comprehensive assessment across the factors including the systemic medical condition of patient, demands of function and activity, socioeconomic status of patient, benefits and risks of treatment, patient and physician preference for the treatment technique. However, these CPGs might have different influences on the specialty care and primary care for shoulder pain.

In order to thoroughly describe the clinical practice patterns during ambulatory shoulder care visits, we focused on four types of services ordered or provided by physicians during the visits, including pain medications, PT referral, health education/counseling, and diagnostic imaging. After adjusting for the covariates, our study reported higher frequencies of narcotic analgesics, other pain medications prescribed, skeletal muscle relaxants by PCPs compared to non-PCPs. Meanwhile, PT and MRI were less frequently prescribed by PCPs compared to non-PCPs. According to recent guidelines [8,9,10,11,12,13,14], the main goals of treatment are to reduce pain, restore function, and improve quality of life. A multimodal approach involving pharmacological and non-pharmacological approaches should be considered in the management of shoulder pain [8,9,10, 13]. Decision-making process in assessing, diagnosing, and treating patients with shoulder pain is complex and challenging at primary care level [17]. Our results may reveal the higher reliance of pharmacological approaches, coupled with the potential under-utilization of PT during the ambulatory shoulder care provided by PCPs compared to non-PCPs in the United States.

While the prescribing practice of narcotic analgesics for the musculoskeletal pain has been well established [26, 27], the role of PCPs in the epidemic of narcotic analgesics during ambulatory shoulder care visits remains unclear. The increased public awareness of serious side effects associated with narcotic analgesics [28,29,30] has prompted a variety of guidelines to improve narcotic analgesics prescribing practices in the ambulatory care settings [31,32,33]. Under this circumstance, the significant higher frequency of narcotic analgesics prescriptions by PCPs (22.3%) versus non-PCPs (13.8%) might be the most concerning finding in this study. After adjusting for the covariates, PCPs remained an independent risk factor (adjusted OR = 1.62, 95% CI: 1.04–2.51) for the greater probability of prescribing narcotic analgesics prescriptions for shoulder pain. This finding was consistent with earlier studies using NAMCS data. Larochelle et al. reported patients visiting PCPs for chronic musculoskeletal pain had higher odds of receiving opioid prescriptions than patients visiting non-PCPs [26]. Rasu et al. identified that visits to PCPs were more likely to report use of opioids for non-malignant chronic pain compared with visits to non-PCPs [27]. The underlying reasons contributing to narcotic analgesics prescriptions for shoulder pain were multi-factorial and the significant differences in physician patterns of prescribing narcotic analgesics were difficult to interpret. Acuna et al. indicated familiarity with the prescription guidelines might influence patterns of opioid prescribing and reported 21% of the orthopedic shoulder and elbow providers were unaware of any type of prescription guideline [34]. The reported influential factors included previous opioid prescriptions (82%), patient age (48%), and duration of the symptoms (48%) [34]. Although multimodal programs or nonopioid individual modalities were recommended [9], we cannot comment on the appropriateness of narcotic analgesics prescribing practices in shoulder care due to the limitations inherent in the NAMCS database. But, an appraisal of our results makes it hard to ignore the significant differences we observed.

As we expected, the most common prescribed pain medication for shoulder pain was NSAIDs, with a frequency of 28.2% by PCPs and 27.7% by non-PCPs. Meanwhile, our study confirmed a statistically significant higher frequency of other pain medications prescriptions by PCPs (14.7%) versus non-PCPs (7.8%). Based on the drug classification of the database, other pain medications included miscellaneous analgesics, salicylates, and analgesic combinations. Due to the side effects of these medications such as gastrointestinal bleeding, cardiovascular events, and renal toxicity, appropriate caution must be exercised while prescribing for the vulnerable elderly patients [35, 36]. In general, non-narcotic medications administered were consistent with current practice guideline recommendations for the pharmacological management of shoulder pain [8, 9, 11].

It was unexpected to observe a statistically significant difference between the PCPs and non-PCPs in prescribing skeletal muscle relaxants. Although there is growing support for adjuvant drug use for the musculoskeletal complaints [37, 38], the prescription patterns of skeletal muscle relaxants for shoulder conditions remain unclear. Soprano et al. found that skeletal muscle relaxants use increased rapidly by using the 2005–2016 NAMCS data [38]. Our study reported a twofold higher frequency of skeletal muscle relaxants prescribed by PCPs compared to non-PCPs. Because of neurological side effects of skeletal muscle relaxants, such as drowsiness, dizziness, and confusion, physicians must constantly weigh in the risks and benefits of prescribing skeletal muscle relaxants in the elderly population [38, 39]. Spence et al. performed a conditional logistic regression analysis and found that older adults using skeletal muscle relaxants had an significantly increased risk of injury [40]. The difference reported here may signal gaps in skeletal muscle relaxants prescribing by PCPs and non-PCPs, which is important to determine a baseline for future evaluation and identify areas to improve the use of these adjuvant drugs.

Given the published literature, a typical management strategy for shoulder conditions involves a multimodal approach that depends not only on pharmacological but also non-pharmacological approaches [8,9,10, 13]. A successful outcome could be defined as a significant reduction of pain, restoration of function, and improved quality of life [13]. In terms of non-pharmacological approaches, PT and health education/counseling should be considered in the delivery of high-quality care to patients with shoulder pain. As an important complement to or replacement for pharmacological approaches, PT is usually recommended as the first-line treatment option for shoulder pain [9,10,11,12,13,14, 41]. Due to the distinct inclusion and exclude strategies, the estimate of PT use (19.3%) in our study was lower than the data reported by Song et al. (22.2%) [18]. However, our study revealed the significantly lower frequency of PT prescribed by PCPs (7.6%) compared to non-PCPs (23.8%), which remained after adjustment for the covariates. Since most insurance plans in many states only cover the cost of PT prescribed by a physician, the physician’s role in appropriate use of ambulatory PT is very important. Many factors are associated with physicians’ decisions to referral to PT, such as diagnosis, illness severity, PT supply, patient’s insurance status, physician experience and training [42,43,44]. The unobserved differences between PCPs and non-PCPs, such as physician experience or unmeasured illness severity, could partially but not fully explain the variation in physicians’ referral to PT for shoulder conditions. Such variation could be more likely to indicate a potential ambulatory under-use of PT for shoulder management by PCPs compared to non-PCPs. Our study fills the current literature gaps regarding the effect of physicians on the appropriate use of ambulatory PT for shoulder pain.

Additionally, the significantly lower frequency of MRI ordered by PCPs (7.5%) compared with non-PCPs (18.3%) might be consistent with the current consensus [45,46,47]. Although MRI can provide valuable information in lesions of articular and soft tissue [48], question to the diagnostic efficacy and reliability of MRI remains especially in the field of shoulder disorders [49]. The abnormal radiological findings may not be the source of clinical symptoms [49]. In primary or intermediate care for nontraumatic shoulder pain, Cuff et al. suggested MRI should not be routinely requested [45]. For atraumatic shoulder pain, Bradley et al. did not recommend MRI as a screening tool before a comprehensive clinical evaluation of the shoulder [46]. Therefore, the significantly lower frequency of MRI ordered by PCPs might be in line with the current consensus [45,46,47]. Limited by the available data in the NAMCS, we were unable to evaluate the rationality of the MRI ordered in the survey. It is beyond the scope of this study to assess the appropriateness of MRI ordered by PCPs or non-PCPs. However, physician experience and specialty training might be the drivers of this result, and additional research is needed [50].

Our study provided a broad overview of ambulatory management of shoulder pain by PCPs compared to non-PCPs in the United States. Although our analysis reflects a consistent difference in the services provided by PCPs versus non-PCPs for the patients with shoulder pain in ambulatory care settings, the underlying reasons contributing to these results are complex and warrant further investigation. Given the challenges of providing appropriate treatment for shoulder conditions, concerted efforts by clinicians, patients, and policymakers should be made to optimize effectiveness and efficiency of shoulder management in ambulatory care settings.

5 Conclusion

After adjusting for the covariates, this study identified compelling differences in the services ordered or provided by PCPs versus non-PCPs for shoulder pain in ambulatory care settings. PCPs were more likely to prescribe narcotic analgesics, skeletal muscle relaxants, and other pain medications for shoulder visits compared with non-PCPs. Meanwhile, PCPs were less likely to prescribe PT and MRI compared to non-PCPs. Results from this study can be valuable for clinicians and policy makers to prevent possible variations in the management of shoulder pain by PCPs and non-PCPs and may ultimately help improve the quality and reduce the cost of ambulatory care for patients with shoulder pain.

6 Limitations

This study has several limitations that should be acknowledged while interpreting the present results and conclusions. A primary limitation of this study is inherent to the nature of NAMCS data. While this was a retrospective cross-sectional study conducted with the visit-level data of NAMCS, the associations between the factors and treatment choices of shoulder pain should not be interpreted as causal relationships. Further prospective studies with longitudinal data analysis at the patient level should be conducted to draw the causal inferences. Second, NAMCS data does not capture the information on clinically important variables such as severity of the symptoms, phases of the disease, results of the examinations, or outcomes of the treatments. Therefore, the accuracy of diagnosis and appropriateness of treatment choices can’t be ascertained in the NAMCS database. Third, the analysis was limited to the visits with a primary diagnosis of shoulder disorder and one of the top 3 complaints related to shoulder symptoms, which consequently underestimated the shoulder-related visit prevalence in the sample population. While this inclusion and exclusion strategies might introduce selection bias, the services ordered or provided by physicians would be more directly linked to the visits seeking treatment for shoulder symptoms. Finally, interpretations about trends over time should be considered with caution since NAMCS changed the sampling strategies during the study period.