Abstract
Purpose: Primary care physicians (PCPs) often face a complex intersection of patient expectations, evidence, and policy that influences their care recommendations for acute low back pain (aLBP). The purpose of this study was to elucidate patterns of PCP orders for patients with aLBP, identify the most common patterns, and describe patient clinical and demographic characteristics associated with patterns of aLBP care.
Methods: This prospective cohort study included 9574 aLBP patients presenting to 1 of 77 primary care practices in 4 geographic locations in the United States. We performed a cluster analysis of PCP orders extracted from electronic health records within the first 21 days of an initial visit for aLBP.
Results: 1401 (15%) patients did not receive a PCP order related to back pain within the first 21 days of their initial visit. These patients predominantly had aLBP without leg pain, less back-related disability, and were at low-risk for persistent disability. Of the remaining 8146 patients, we found 4 distinct order patterns: combined nonpharmacologic and first-line medication (44%); second-line medication (39%); imaging (10%); and specialty referral (7%). Among all patients, 29% received solely 1 order from their PCP. PCPs more often combined different guideline concordant and discordant orders. Patients with higher self-reported disability and psychological distress were more likely to receive guideline discordant care.
Conclusion: Guideline discordant orders such as steroids and NSAIDS are often combined with guideline recommended orders such as physical therapy. Further defining patient, clinician, and health care setting characteristics associated with discordant care would inform targeted efforts for deimplementation initiatives.
Introduction
Chronic low back pain (cLBP) is a leading cause of disability and health care utilization.1,2 Effective initial management of acute low back pain (aLBP) may prevent cLBP3,4 and downstream health care costs.5,6 Primary care physicians (PCPs) are often the initial clinician for patients with aLBP;7 back pain is the second most common symptomatic reason for a primary care visit.8 Thus, PCPs are uniquely positioned to influence the trajectory of patients with aLBP in terms of recovery and subsequent health care utilization.
Clinical practice guidelines by the American College of Physicians recommend nonpharmacologic approaches such as massage and spinal manipulation for initial management for aLBP (recommendation strength: strong).9 Medications such as nonsteroidal anti-inflammatories (NSAIDs) and muscle relaxants are considered a first-line treatment for aLBP only if the patient expresses preference for a medication (recommendation strength: strong).9,10 Pharmacologic treatments such as opioids are associated with a greater risk profile and are not recommended for aLBP. In the absence of progressive neurologic symptoms or other red flags, radiographic and magnetic resonance imaging should be delayed until after a trial of nonpharmacologic and/or first-line pharmacologic treatment. While there is no guideline recommendation for when to refer to a medical specialist, this referral is often associated with utilization of invasive procedures (eg, epidural injections) that are not recommended.11
Although the practice patterns of PCPs for low back pain have been described previously,11⇓–13 few studies capture the complexity of PCP orders relative to current clinical practice guidelines. More commonly, previous literature focuses on single interventions rather than how treatments are combined in real-world practice. For example, imaging,14,15 opioid prescribing,16 physical therapy (PT),17 and other nonpharmacologic treatments18 have been compared against themselves in primary care settings using claims databases. However, claims analyses typically examine new episodes of care which are likely a heterogeneous group of acute, subacute, and chronic back pain. Moreover, claims data does not provide a comprehensive view of PCP recommendations. PCPs may recommend several concurrent treatments (eg, opioid prescription and order for PT), diagnostic tests, or consultations (eg, a radiograph and referral to a specialist) and the patient self-selects the services they use. Last, claims data provide limited insight on patient-related clinical characteristics that may be influencing PCP decision making. Given the high prevalence of back pain in primary care settings, there is a need to describe the combinations of treatments PCPs recommend to patients with aLBP.
Recently, in a large sample of patients with aLBP initially seen in primary care, we found associations between the amount of guideline discordant treatments (eg, opioids) or procedures (eg, imaging) provided and the risk of developing cLBP.19 However, we did not explore combinations of orders that may be both guideline concordant and discordant simultaneously. Clarifying the degree to which nonconcordant treatments are recommended as the sole intervention or in combination with other treatments is important to understanding the degree to which current clinical practice guidelines have been implemented in routine primary care practice. Moreover, identifying any distinguishing characteristics in populations receiving certain forms of care will guide implementation efforts to improve care. The aims of this study were to: 1) describe guideline concordant and discordant health care orders initiated by PCPs for patients with aLBP, 2) identify similar patterns of orders, and 3) describe patient clinical and demographic characteristics associated with order patterns.
Methods
Study Design
The Targeted Interventions to Prevent Chronic Low Back Pain in High‐Risk Patients (TARGET) protocol and primary results articles have been published elsewhere.19⇓–21 Briefly, 9547 aLBP patients presenting to 1 of 77 primary care practices in 4 geographic locations (Baltimore, MD; Boston, MA; Pittsburgh, PA; Salt Lake City, UT) were screened using the STarT Back Tool22 and stratified as being at low, moderate, or high risk of developing chronic low back pain. Regardless of risk stratification, all patients were followed over the next twelve months as part of the inception cohort study. High-risk patients (n = 2300) were additionally enrolled in a pragmatic, multi-site, randomized controlled trial that compared usual care to referral to a stratified approach to care using psychologically informed physical therapy (ie, PT that is combined with cognitive behavioral strategies). Patients were enrolled between May 2016 and June 2018. Four institutional review boards approved the trial. The current study included all patients enrolled in the trial and cohort.
Sample
Patients were eligible for enrollment if they were 18 years of age or older and presented to the clinic with a primary complaint of acute axial LBP or LBP with associated leg pain as determined by ICD-9 or ICD-10-CM diagnosis codes. Patients with the signs or symptoms associated with serious pathology (eg, vertebral fracture, cancer) were excluded. To ensure acuity of low back pain, a 2-item acute/chronic LBP screening questionnaire was created by adapting the National Institute of Health Research Standards definition for chronic LBP.23 Patients were asked: (1) how long has your low back pain interfered with your ability to do regular daily activities, and, (2) in the last 6 months, how often has low back pain interfered with your ability to do regular activities. If the patient answered “more than 3 months” for the first question, and “half or more than half the days” for the second question, they were considered chronic and excluded from the study.
Data Collection
Primary care orders and patient characteristics were extracted from their respective field in the patient’s electronic medical record. Clinical characteristics were extracted from the electronic medical record except for disability and risk stratification which were collected at index visit.
Primary Care Orders
Initial primary care orders were defined as occurring within the first 21 days of an index visit. Orders were categorized into 5 main groups, described below. A detailed breakdown of orders (eg, medications included as steroids, pain management clinicians) is provided in Appendix A.
Non-Pharmacologic
Behavioral health (eg, psychology, psychiatry), chiropractic, mind-body therapies (eg, acupuncture, massage), pain management (eg, pain clinic, physical medicine and rehabilitation), PT, social work, wellness coaching.
First-Line Pharmacologic. Acetaminophen, muscle relaxant, NSAID, and topical NSAID.
Second-Line Pharmacologic. Antidepressant, benzodiazepine, opioid, steroid.
Specialty Referral. Neurology, spine/orthopedic surgery.
Imaging. Computed tomography (CT), magnetic resonance imaging (MRI), or plain radiography (radiograph).
Patient and Clinical Characteristics
Patient and clinical characteristics are described as follows:
Patient Characteristics
Demographics, health insurance (commercial, Medicare, Medicaid and self-pay/other), smoking status (yes/no), body mass index, geographic location of the clinics.
Clinical Characteristics
Back pain diagnosis (axial back pain vs back and leg pain), self-reported LBP functional disability, and risk stratification. Back pain diagnosis was identified via ICD-9 or ICD-10 diagnostic codes present at the index encounter. Self-reported LBP functional disability was measured using the Oswestry Disability Index (ODI),24 a self-report scale ranging from 0 to 100 with lower scores indicating less disability. Risk stratification was assessed using the Keele STarT Back Tool22 which assigns a risk of developing cLBP based on patient response to 4 symptom-based and 5 psychological-based items. Scores range from 0 to 9. Patients are characterized as low-risk (total score ≤3), medium‐risk (total score ≥4 and psychological score ≤3), or high risk (total score ≥4 and psychological score ≥4). https://startback.hfac.keele.ac.uk/training/resources/startback-online/.
Analysis
Individual PCP orders and common combinations of orders relevant to low back pain guidelines were examined and are described as frequencies.
Next, a cluster analysis was used to identify distinct groups of patients based on patterns or similarities of particular variables of interest. In this analysis, different types of PCP orders were the variables used to cluster patients. When performing a cluster analysis on a large data set with dichotomous variables (ie, received an order or did not), halving the data into a “training” set and “validation” set can guide the interpretation of the clusters.25 The “training” set is first analyzed and used to independently determine the number of clusters. The “validation” set is then analyzed using the same clustering method and number of clusters from the training set. In this analysis, comparisons of order frequencies as well as clinical and demographic data of the participants in each cluster were examined to evaluate similarity between the clusters from each data set.
Patients who received no orders within the first 21 days were extracted as a separate group leaving those who had at least 1 early primary care order within 21 days of index visit included in the cluster analysis. These patients were split into 2 equal samples using simple random sampling. One half of the data, the “training” dataset, was used to determine the number of clusters. The clustering method was average linkage hierarchical agglomerative clustering and the Jaccard index was used as a measure of distance between patients.25 The research team reviewed the different models of clusters with the aim of identifying the most parsimonious model with robust numbers per cluster and distinct order patterns among clusters. Once the number of clusters was selected, the other half of the data were used to validate the original clusters.
Clusters were named according to the predominant category (eg, Non-Pharmacological & first-Line Pharmacological) of PCP orders within the cluster. Categories of PCP orders are presented as frequencies within each cluster. The categories were broken down into specific types of orders within each cluster. In addition, common combinations of individual orders in the clusters are described as frequencies. Finally, we compared patient and clinical characteristics among clusters in the training and validation sets to determine if characteristics were associated with clusters. Patient, clinician, and clinical characteristics within each cluster are described as means or frequencies. Analysis was performed in SAS version 9.4.
Results
Of the full sample (9574), 1401 (15%) patients received no PCP order within the first 21 days (Table 1). These patients were predominantly stratified as low-risk (60%), had axial back pain only (89%), and low disability (mean ODI = 23). In the remaining 8146 patients, PCPs solely placed 1 order in 29% of initial visits. When comparing solely nonpharmacologic versus pharmacologic orders, 8% of patients received only nonpharmacologic orders whereas 33% received only pharmacologic (first or second line) orders. There was wide variability in the orders made by PCPs and their concordance with the guidelines. When considering guideline concordant care, 34% of patients received an order for PT and, of them, 11% got an additional NSAID order and 13% received an order for a muscle relaxant. When considering orders that were partially concordant with the guidelines, 13% of patients received a combination of a concordant nonpharmacologic order with a nonconcordant second-line pharmacologic order. However, 14% of patients received solely nonconcordant second-line pharmacologic (8%), imaging (4%), or specialty (2%) orders.
Primary Care Physician Orders for 9574 Patients with Acute Low Back Pain Within the First 21 Days of Initial Encounter
The training cluster data set (n = 4073) yielded 4 different models ranging from 2 to 5 clusters. The 4-cluster model was selected based on distinctiveness of physician orders (Table 2). This model was used in the validation data set and compared against the training data set to assess and confirm similarity (Appendix B). The proportion of order frequency as well as patient demographic and clinical characteristics were evaluated to determine consistency between the 2 data sets. The 4 clusters were: 1) Non-Pharm and first-Line Pharm (44%); 2) second-Line Pharm (36%); 3) Imaging (13%); and 4) Specialty (7%).
Cluster Analysis of Primary Care Physician Orders Within the First 21 Days from Initial Visit in 4073 Patients with Acute Low Back Pain
The Non-Pharm and first-Line Pharm cluster (n = 1810) was the youngest (mean age = 48) and had the largest proportion of Black patients (22%) of the clusters. They mostly had axial back pain (81%) and less disability (mean ODI = 31). In this cluster, 39% and 20% of patients received solely first-line pharmacologic or nonpharmacologic orders, respectively (Table 3). The most common combination was a nonpharmacologic order with a first-line pharmacologic order (25%). Frequent orders in this cluster were PT (n = 934; 52%), muscle relaxants (n = 940; 52%), and NSAIDs (n = 933; 52%) (Table 4).
Common Combinations of Primary Care Physician Orders Within the First 21 Days from Initial Visit in 4073 Patients with Acute Low Back Pain
Specific Primary Care Physician Orders Within the First 21 Days from Initial Visit in 4073 Patients with Acute Low Back Pain
Characteristics of 1401 Patients with Acute Low Back Pain Who Did Not Receive a Physician Order and 4073 Patients Clustered by Predominant Physician Orders Within 21 Days of Initial Visit
The second-Line Pharm cluster (n = 1470) had the highest proportion of white (88%), privately insured (53%) patients, and the highest disability (mean ODI = 40) (Table 5). In this cluster, 71% of patients received 1 (27%) or more than 1 (44%) second- and first-line pharmacologic orders. The most frequent orders in this cluster were steroids (n = 913; 62%), muscle relaxants (n = 747; 51%), and opioids (n = 655; 45%).
The Imaging cluster (n = 525) was the oldest (mean age = 56) and had the highest proportion of females (62%). Imaging orders were predominantly for plain radiographs (n = 434; 83%) and were combined with second-line pharm (n = 118; 22%) or nonpharm (n = 107; 20%) orders. The Specialty cluster was comparatively small (n = 268) but contained the largest proportion of patients who were stratified as high risk (38%), had back and leg pain (41%) and had pain greater than 3 months (23%).
Discussion
We determined PCP patterns of care for 9574 adults with aLBP presenting to 77 primary care clinics in 4 geographic regions. We found that PCPs often combined multiple orders rather than a single order, but that there was wide variation in the order type and combination. This frequently resulted in PCP orders being partially concordant with current guidelines (ie, referring to a recommended treatment) while also prescribing a second-line or nonrecommended medication or imaging for initial care of aLBP.
Our findings have implications for trials that have “usual care” comparators.26 Without fully understanding what is included in usual care, estimates for comparative treatment effects may be confounded. This is, nonetheless, the reality of low back primary pain care where patient complexity is common and there are multiple choices for evidence-based treatments. And, while the recommendations from the guidelines are “strong,” the quality of the evidence for the different treatments varies (eg, low quality evidence for massage, spinal manipulation and acupuncture). Various factors may influence PCP orders such as patient comorbidities, patient preferences, availability of resources, long wait-lists for nonpharmacologic clinicians, payer types, and clinician beliefs about the evidence base for the guidelines.27 We identified some distinguishing patient-characteristics that may explain some of the variation in PCP orders.
Low-Moderate Risk Patterns
PCPs appeared to recognize and provide guideline concordant care to patients who had low disability and were stratified by the STarT Back score as low-risk. A common combination for patients with low to moderate risk of developing cLBP was a nonpharmacologic order with a first-line medication. This practice is congruent with guidance that recommends nonpharmacologic treatments first and, if requested by the patient, in combination with an NSAID or muscle relaxant.
High-Risk Patterns
Patients who were stratified as high risk of developing disability and had longer duration of symptoms tended to be in the clusters that received orders associated with greater risk or cost such as opioids, steroids, imaging orders, and to a lesser extent, specialty referral. It is conceivable that PCPs seeing patients with more severe presentation recommended invasive or aggressive treatments that were not guideline concordant. Thus, guideline implementation or deimplementation interventions targeting clinicians may need to focus on this patient subgroup. However, it is also conceivable that patients who self-report higher disability and psychological distress perceive a stronger need for advanced care and place pressure on clinicians to receive it.28⇓⇓–31 This pattern of patient presentation has been described in the emergency medicine department where patients with nonspecific low back pain receive high-rates of imaging and opioid medications.32⇓–34 However, clinical trials do not support the premise that pain medications are more effective among patients with severely intense or disabling aLBP compared with nonpharmacologic approaches.35,36 Our previous study suggests that opioids and nonopioid medications are associated with a higher transition to cLBP.19
The STarT Back tool stratifies patients into high-risk if they respond positively to statements about pain related psychological distress as these patients have been shown to have worse outcomes.22 With this in mind, using integrated behavioral health or referring to other clinician (eg, rehabilitation) who can help patients process and cope with their pain may be more useful than medications and referrals. We say this while acknowledging that in the embedded trial (the TARGET trial) psychologically informed physical therapy did not improve outcomes among high-risk patients.20 Yet, a program evaluation looking at implementation at 1 of the sites found implementation was challenging at multiple levels – screening in primary care, PCP referrals, scheduling visits, and attendance for psychologically informed physical therapy.37 It may be that pain medication prescribing is common because it is convenient, routine, and has established systems for payment or coverage.38 Previous work suggests that PCPs may select more aggressive care such as pain medication due to pressure from patients to receive immediate pain relief, concerns of patient dissatisfaction, and fear of litigation from patients.39 Indeed, public health campaigns lead by clinicians to educate patients about the risks of overdiagnosing and overtreating LBP have been initiated in primary care and, while still in infancy, have faced resistance from patients.40 The continued variability in practice and the persistence of guideline discordant care warrants research into implementation at multiple levels.
Our study has several limitations. Embedded within this observational cohort was a randomized trial on a PT intervention. Thus, nonpharmacologic orders, particularly PT, may have been inflated and our findings may not be generalizable to other systems that do not have strongly integrated PT care options. Low order rates of non-PT nonpharmacologic approaches may not be captured in orders (eg, clinician recommendations for yoga, mindfulness based stress reduction occurring outside the medical system). Second, our medication orders were not linked to ICD-9 or ICD-10 in the electronic health record systems. Therefore, it is possible that certain orders such as antidepressant medications were related to other patient conditions. Our data cannot provide an explanatory model for why certain guideline discordant orders may have been chosen based on patient comorbidities (eg, imaging in patients with osteoporosis or history of malignancy). Last, a cluster analysis is helpful for identifying homogeneous groups of people based on a variable or variables of interest (in our case PCP orders). Similarities between generated clusters across other patient level variables suggests that there is not absolute distinction between who receives which PCP orders.
Conclusion
Primary care physicians navigate a complex interaction between policy guidelines, patient preference, and resource availability when treating aLBP. Future research that measures clinical effectiveness and downstream utilization of guideline based combinations of treatments may be more pragmatic. In addition, implementation efforts targeted at patients and clinicians to increase use of nonpharmacologic treatment and prevent discordant care remains warranted.
Acknowledgments
The authors acknowledge the TARGET trial team.
Appendices.Appendix A
Breakdown of Specific Medications Within Each Category
Appendix B
Comparison of Training Data and Validation Data Clusters in the 4-Cluster Model
Notes
This article was externally peer reviewed.
Funding: The TARGET Trial and Inception Cohort were funded by the Patient Centered Outcomes Research Institute (PCORI contract # NCT02647658). ClinicalTrials.gov Identifier: NCT02647658.
Conflict of interest: The authors have no conflicts of interest to declare.
This is the Ahead of Print version of the article.
To see this article online, please go to: http://jabfm.org/content/00/00/000.full.
- Received for publication March 28, 2023.
- Revision received May 31, 2023.
- Revision received June 13, 2023.
- Accepted for publication June 20, 2023.
References
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.