Clinician approaches to new spine pain cases in primary care: balance of opioid prescribing and early linkage to exercise therapy and spinal manipulation

Abstract

Rationale, Aims and Objectives:

Spine pain (SP) is common and often disabling. Clinical practice guidelines (CPGs) discourage opioid treatment and outline the value of varied nonpharmacologic therapies (NPTs). This study elucidates the amount of variability in primary-care clinicians’ (PCPs’) prescribing of opioids and in their cases’ receipt of the two most common NPTs (exercise therapy and spinal manipulation).

Method:

The design was a retrospective cohort study examining variation in the treatment of PCPs’ new SP cases, classified by receipt of a) prescription of an opioid at the initial visit; b) exercise therapy and/or spinal manipulation within 30 days of initial visit. The study was set in the primary care clinics at military treatment facilities (MTFs) of the US Military Health System (MHS) in the period between October 2011 and September 2016.

Results:

The majority of cases did not receive a study treatment (66.3%); 19.6% of cases received only NPT within 30 days of initial visit; 11.5% were prescribed only an opioid at the initial visit with receipt of both NPT and opioid during early treatment rare (2.6%). Exercise therapy within 30 days exhibited more than a 2-fold difference in interquartile percentile rates (IQR) (median provision 15.8%, IQR 9.8%, 22.1%). The other treatments exhibited even greater variation; specifically, spinal manipulation (median 8.5%, IQR 3.3%, 15.8%), and opioid at initial visit (median 10.3%, IQR 4.4%, 18.2%). The availability of physical therapists and doctors of chiropractic had significant association with several clinical provision rates.

Conclusion:

Among providers of spine care for a sample of Army soldiers, there was substantial variation in the early provision of exercise therapy, spinal manipulation, and opioid prescriptions. The magnitude of the case-mix adjusted variation and its association with facility availability of providers suggests that quality of care initiatives may help reduce this variation.

BACKGROUND:

Spine pain (SP) is one of the most common reasons for primary care visits [1], and neck and back pain accounted for the highest amount of U.S. health care spending among all conditions with an estimated $134.5 billion expenditure in 2016 [2]. In the United States Armed Forces, back problems have been consistently associated with the highest percentage of medical encounters [3]. However, the clinical guidance for treatment of back pain is equivocal and there is concern that too often back pain is primarily treated with prescription opioids [4] despite the finding that treatment with opioid compared to non-opioid medication treatment has not been associated with improvement in functional outcomes [5]. Treatment options for clinicians seeing patients presenting with back pain include pharmacologic approaches, particularly opioid versus non-opioid medication and nonpharmacologic therapies (NPTs) including physical therapy and chiropractic care. Clinical practice guidelines (CPGs) discourage opioid treatment and outline the value of watchful waiting and varied NPTs. These guidelines give different weights of recommendation for prescribing NPT in acute versus chronic low back pain (LBP), and the bulk of studies have measured the efficacy of NPTs for chronic presentations [6].

For acute LBP, the guidelines of the American College of Physicians note no evidence of benefit with exercise therapy and only low strength of evidence for spinal manipulation [6]. An expert review of acute LBP treatment described no positive effect with exercise therapy [7], and a recent overview of recommendations for acute/subacute LBP gave physical therapy a “C” (marginally support a recommendation for use) and spinal manipulation a “B” (moderately support a recommendation for use) [8]. Additionally, a meta-analysis found exercise therapy was not associated with improvement in acute LBP [9]. CPGs for neck pain are equally guarded, suggesting “clinicians may consider range of motion exercise with manipulation or mobilization” [10].

On the other hand, some studies have found benefit with NPT. One study found that by comparison with usual care, patients who first self-refer to physical therapists (PTs) or doctors of chiropractic (DCs) have better outcomes [11]. A study of U.S. Army soldiers found early treatments for LBP improved military readiness outcomes [12]. Also, a recent evaluation of almost one million outpatient visits for acute LBP found that the minority (11%) of patients referred to prompt (less than 15 days after presentation) physical therapy incurred significantly lower healthcare costs [13]. Similarly, a large study of early physical therapy (PT) for acute low back pain found significant reductions in subsequent use of lumbosacral injections, lumbar surgery, and frequent physician office visits for low back pain [14]. A clinic created in the US military to promptly evaluate acute low back pain (with opportunity for medication receipt) by physical therapists found improved outcomes, including less medication and imaging [15]. Despite uncertainty regarding the value of exercise therapy and/or spinal manipulation for acute LBP, many clinicians and patients have embraced these treatments [16] and the majority of commercial health plans and Medicare will pay for physical therapy and chiropractic care [17].

The specific objectives of this paper were to examine variation among primary care clinicians in 1) linkage of SP patients to exercise therapy and/or spinal manipulation within the first 30 days of initial visit, and 2) opioid prescribing at initial visit. This study is centered in the United States Military Health System (MHS), a coordinated system providing healthcare for a young, relatively healthy population. Back pain is common in the active duty military population with an incidence rate of 40.5 per 1000 persons-per years [18]. We focus on clinician variation in early patient treatment as a strategy to identify potential areas for quality improvement.

METHODS:

Data Sources

Data used in this study included Military Health System (MHS) eligibility, ambulatory care, and prescription records for the SUPIC sample as reported in the Military Data Repository (MDR). Records across these data sources were linked by encrypted identifiers. Data were obtained under a data use agreement with the Defense Health Agency/Department of Defense and the Brandeis University and Uniformed Services University determined the study was exempt from ongoing IRB review.

Study Design

Unlike prior research that has focused on individuals as the unit of analysis, this study aggregated primary-care providers’ (PCPs’) rates of recommending different treatments to cases and explored factors associated with variation in their decision to recommend these treatments during index visits. We identified predictors related with greater likelihood of exercise therapy and/or spinal manipulation early in treatment or lower likelihood of prescribing opioid at the initial visit.

Participants

We examined SP care delivered to Army soldiers returning from combat deployments for whom the authors have constructed longitudinal healthcare records as part of the Substance Use and Psychological Injury Combat (SUPIC) study [19]. Soldiers tend to be young individuals without chronic disease who, nonetheless, experience frequent acute injuries associated with occupational duties [3][20]). Eligibility criteria for study included having a SP diagnosis in the principal position (in category ‘205’ in AHRQ’s clinical classification system [CCS]) on an ambulatory encounter record from a primary care clinic at a Military Treatment Facility (MTF). Eligible SP cases had to be new, as defined by having no SP diagnosis in any position on an MHS claim within the 90 days prior to the visit. New SP cases first diagnosed in non-primary care settings (e.g., emergency department, specialty clinic, civilian settings rather than MTF, or in hospital setting) were excluded. Notably, soldiers could represent multiple SP cases when their initial encounters were separated by more than 90 days.

Individual clinicians analyzed in this study consisted of providers serving in primary care clinics (based on Medical Expense and Performance Reporting [MEPR2]), who were HIPAA classified as physicians [doctor of medicine (MD) or doctor of osteopathy (DO)], advanced practice nurses (APN) or physician assistants (PA). We restricted the analysis to these HIPAA classifications because they had authority to diagnose, prescribe opioids, and refer to rehabilitation services of interest, such as provided by physical therapists or doctors of chiropractic. Other HIPAA classifications were excluded because they lacked authority to diagnose and/or prescribe opioids. To assure the reliability of aggregated rates of NPT and opioid utilization, we also required study clinicians to initiate care for 10 or more SP cases during the window of October 1, 2010 to September 30, 2016 (hereafter, FY2011-FY2016).

Measurement

For each primary care clinician, four provision rates were generated indicating the proportion of various treatments for new SP cases: a) prescription of an opioid at the initial visit; b) receipt of exercise therapy within 30 days of initial visit; c) receipt of spinal manipulation within 30 days of initial visit; d) receipt of either exercise therapy or spinal manipulation within 30 days of initial visit. Opioid medications included codeine, fentanyl, hydrocodone, hydromorphone, meperidine, morphine, oxycodone, oxymorphone and tramadol. Therapies were defined based on Current Procedural Terminology (CPT) Code on the ambulatory record.

Following Song et al. [21], clinicians’ treatment rates were adjusted for patient sex and specific mental health and pain diagnoses, which would clinically influence the appropriate treatment of a case. These diagnoses had to be current, present as principal diagnoses on ambulatory records (MTF and civilian care) within the 90 days prior to and including the initial SP visit. Following adjustment, factors were explored which could predict variation in clinician treatment rates such as the clinician’s HIPAA specialty (MD/DO, APN, PA), the fiscal year of initial visit, and two contextual measures of a facility’s availability to provide treatments of interest. These availability measures were defined as a) physical therapist full time equivalents (PT FTEs) per 100,000 persons served and b) doctor of chiropractic (DC) full time equivalents (DC FTEs) per 100,000 persons served. Both measures were derived from monthly workforce data maintained in the DMHRSi (Defense Medical Human Resources System-Internet) together with MHS data for each study military treatment facility (MTF) and unique patient Data Medical Information System [DMIS] identifiers. After being annualized, these MTF-based FTE measures were grouped into quartiles (labeled as low, middle, high, and very high). For PT FTE, the quartile cutoffs were 2.43, 5.00, and 9.67 PTs per 100,000 persons served; for DC FTE, the quartile cutoffs were 0.25, 0.55, and 1.37 DCs per 100,000 persons served.

Statistical Analysis

We created the distributions of patient characteristics of the SP cases and summarized unadjusted clinician rates for initial visit opioid prescribing, patient use of exercise therapy or spinal manipulation within 30 days, patient use of both treatments, and use of neither. To address potential bias due to case mix, clinician provision rates were then adjusted for sex and mental health and pain diagnoses. This adjustment was performed by using logistic regression to generate clinician expected rates, dividing observed rate by corresponding expected rate, and then creating each clinician’s observed over expected rate of provision, and then multiplying the quotient by an observed overall rate. To provide perspective on the distribution of clinicians’ adjusted rates, we calculated summary values, specifically the mean and 5^th, 25^th, median, 75^th, and 95^th percentiles. These summary values were stratified by explanatory factors of interest (type of clinician, fiscal year, MTF availability of PT and availability of DC).

As a final component of the analyses, we used hierarchical linear models to test three related hypotheses:

H1: Clinician rates of immediate prescription of opioid treatment, provision of therapeutic exercise, and spinal manipulation within 30 days will be associated with the HIPAA classification of clinician and the year of the index spine pain diagnosis.

H2: Clinician rates of immediate prescription of opioid treatment, provision of therapeutic exercise, and spinal manipulation within 30 days will be associated with facility capacity of physical therapists and doctors of chiropractic as represented by the full-time equivalents per 100,000 users served.

H3: Even after considering the factors tested in Hypotheses 1 and 2, significant unexplained variation in clinician use rates will remain.

Conclusions concerning the first two hypotheses were based on the significance of the corresponding factor in the hierarchical linear model. The conclusion for the last hypothesis was based on the between-provider proportion of unexplained variance in each hierarchical model.

FINDINGS

Our analytic sample after exclusions consisted of 4,302 unique clinicians and 279,017 cases of newly diagnosed SP cases (see Appendix Study Flow Diagram). Interestingly, MDs/DOs represented the largest group of clinicians (43.4%), but they initiated only 33.1% of the cases. PAs represented almost as many clinicians (41.1%), but they were responsible for over half the cases (51.7%). Advanced practice nurses (APNs) accounted for the small remaining balance of both clinicians (15.5%) and cases (14.2%) (Table 1). Reflecting the demographics of Army soldiers, overwhelmingly the SP cases represented males (86.3%) within the enlisted ranks (83.3%), who were between 25 and 40 years of age (59.1%). Prior 90 days comorbidities experienced by more than 3% of cases included pain in limb or joint (18.3%), pain from injury (6.3%), pain in back [diagnoses not included in CCS 205] (4.3%), headache pain (4.0%), depression (3.9%), anxiety (3.9%), PTSD (3.7%), and mental health issue due to stress (6.2%).

Table 1.

Characteristics of Sample

Cases by Specialty	Count of Cases	Percent	Count of Clinicians	Percent
Physician (MD/DO)	92503	33.1	1868	43.4
Physician Asst	146996	52.7	1769	41.1
Advanced Practice Nurses	39518	14.2	665	15.5
Total	279017	100.0	4302	100.0
Characteristics of study spine pain cases	Cases	Percent
Sex from eligibility record Female	38,295	13.7
Male	240,722	86.3
Race/ethnicity from eligibility record
American Indian/Amer Native, NH	2,364	0.9
Asian or Pacific Islander, NH	28,507	10.2
Black, NH	63,210	22.7
Hispanic	34,797	12.5
White, NH	145,576	52.2
Other, NH	4,563	1.6
Measured at return from deployment
Age Group
18–20	13,675	4.9
21–24	59,422	21.3
25–29	67,266	24.1
30–34	52,536	18.8
35–39	45,294	16.2
40–44	26,921	9.7
45+	13,898	5.0
Component
Active duty	258,198	92.5
National Guard	12,039	4.3
Reserve	8,780	3.2
Rank
Junior enlisted	101,899	36.5
Senior enlisted	127,686	45.8
Junior officer	22,523	8.1
Senior officer	16,012	5.7
Warrant officer	10,897	3.9
Number of prior deployments
None, first deployment	170,352	61.1
One prior	85,448	30.6
2 or more prior	23,217	8.3
Fiscal year of deployment return
2011	42,989	15.4
2012	52,668	18.9
2013	53,104	19.0
2014	52,018	18.6
2015	48,534	17.4
2016	29,704	10.7
Clinical service characteristics in 90 days before index visit *
Specialty mental health service	54,444	19.5
Major outpatient surgery	21,072	7.5
Emergency department visit	23,966	8.6
Hospital admission	4,055	1.4
Opioid prescription received	39,519	14.2
Acute stress/adjustment disorder	17,154	6.1
Anxiety disorder (other than PTSD)	10,873	3.9
Depression disorder	10,821	3.9
Posttraumatic stress disorders (PTSD)	10,481	3.7
Limb pain condition	51,184	18.3
Injury pain condition	17,455	6.3
Headache pain	11,088	4.0
Privileged provider availability at MTF the year of index visit (FTE per 100,000 beneficiaries)
Physical therapist FTE
Low (0 – 2.43)	68606	24.6
Middle (2.43+ - 5.00)	70580	25.3
High (5.00+ - 9.67)	68665	24.7
Very High (9.67+)	70752	25.4
Doctor of Chiropractic FTE
Low (.00 - .25)	67091	24.1
Middle(.25+-.55)	70803	25.4
High (.55+ - 1.37)	69870	25.1
Very High (1.37+)	70839	25.4

Note:

^*all other mental health, substance use clinical disorders and pain comorbidities had less than 3% prevalence.

Table 2 presents the observed mix of care for SP cases at the initial visit or within 30 days stratified by type of clinician. The majority of cases did not receive a study treatment (66.3%); 19.6% of cases received only NPT within 30 days of initial visit; 11.5% were prescribed only an opioid at the initial visit. Use of both NPT and opioid during early treatment was rare (only 2.6%). Broken down by the type of clinician, we found that APNs were slightly more likely than others to immediately prescribe an opioid or an opioid with therapy, and PAs were slightly less likely to do so, but the differences were not large.

Table 2.

Summary Distribution of Observed Utilization Patterns (n=279,019 episodes)

Type of Clinician	No NPT & No Opioid	Opioid, No NPT	NPT No Opioid	NPT & Opioid
MD/DO	64.6	12.2	20.5	2.7
Physician Asst	67.6	10.8	19.1	2.5
Advanced Practice Nurses	65.2	12.6	19.2	2.9

Opioid defined as opioid fill linked to index visit

NPT = nonpharmacologic therapy. defined here specifically to be exercise therapy and/or spinal manipulation based on procedure codes observed on visits within 30 days after index visit

Variation in Adjusted Clinician Rates

In our analysis of adjusted clinician rates for initial visit opioid prescription and provision of NPT within 30 days, we describe our estimated distributions based on the mean and the 5th, 25th, 50th, 75th, and 95th percentiles in rates, stratified by different factors (Table 3). Our analyses exhibited substantial variations in these rates within each clinician type. Exercise therapy (beginning within 30 days) exhibited at least 2-fold differences in interquartile percentile rates, from approximately 9% for the 25^th percentile to over 20% for the 75^th percentile (Table 3A). The other treatments exhibited even greater variation. Specifically, spinal manipulation (Table 3B) increased from approximately 2.5%−3.7% for the 25^th percentile to 14.1%−16.5% for the 75^th percentile, and opioid at initial visit increased from approximately 3.3%−5.0% for the 25^th percentile to 16.7%−18.6% for the 75^th percentile (Table 3C).

Table 3.

Variation in Clinician Practice Patterns: A. Adjusted percent of cases which received exercise therapy within 30 days of index date

Explanatory Factor	Mean	5th Percentile	25th Percentile	Median	75th Percentile	95th Percentile
Overall	15.8%	0.0%	9.8%	15.8%	22.1%	26.2%
Type of provider
MD/DO	15.2%	0.0%	8.4%	14.0%	20.8%	33.0%
Physician Assistant	15.0%	2.1%	8.9%	13.7%	19.9%	32.2%
Nurse Practitioner	16.0%	0.0%	8.4%	15.2%	21.6%	35.1%
Fiscal Year of index visit
2011	13.5%	0.0%	0.0%	9.4%	19.6%	43.0%
2012	12.5%	0.0%	0.0%	8.5%	18.6%	41.0%
2013	13.9%	0.0%	0.0%	10.2%	20.8%	43.9%
2014	15.0%	0.0%	0.0%	11.9%	22.1%	45.1%
2015	16.7%	0.0%	0.0%	13.8%	24.7%	48.6%
2016	19.7%	0.0%	0.0%	16.6%	28.3%	54.9%
Physical therapist FTE / 100,000 beneficiaries
Low (0–2.43)	11.8%	0.0%	5.9%	10.5%	16.8%	25.9%
Medium (2.4301–5.00)	14.5%	1.0%	8.2%	13.6%	19.7%	31.2%
High (5.0001–9.67)	16.1%	1.4%	9.6%	14.8%	20.9%	33.5%
Very High (9.6701 +)	16.1%	0.0%	9.1%	14.8%	21.7%	34.6%
Doctor of Chiropractic FTE / 100,000 beneficiaries
Low (.00 – .25)	14.0%	0.0%	7.2%	12.6%	19.2%	32.4%
Middle (.25–.55)	13.6%	0.0%	7.3%	12.3%	18.8%	30.1%
High (.55 – 1.37)	15.8%	1.8%	8.9%	14.6%	21.1%	33.5%
Very High (1.37+)	16.9%	3.5%	10.2%	15.8%	22.1%	34.4%
B. Adjusted percent of cases which received spinal manipulation within 30 days of index date
Explanatory Factor	Mean	5th Percentile	25th Percentile	Median	75th Percentile	95th Percentile
Overall	10.2%	0.0%	3.3%	8.5%	15.8%	22.9%
Type of provider
MD/DO	11.8%	0.0%	3.0%	8.5%	16.5%	35.9%
Physician Assistant	10.3%	0.0%	3.7%	8.1%	14.7%	28.7%
Nurse Practitioner	9.5%	0.0%	2.5%	7.3%	14.1%	27.6%
Fiscal Year of index visit
2011	10.7%	0.0%	0.0%	3.4%	14.6%	45.1%
2012	9.4%	0.0%	0.0%	1.7%	13.4%	38.3%
2013	10.6%	0.0%	0.0%	3.0%	15.2%	42.0%
2014	11.1%	0.0%	0.0%	4.3%	17.1%	41.1%
2015	11.9%	0.0%	0.0%	5.3%	18.0%	43.3%
2016	11.7%	0.0%	0.0%	3.8%	17.6%	47.4%
Physical therapist FTE / 100,000 beneficiaries
Low (0–2.43)	6.6%	0.0%	0.0%	5.0%	9.0%	20.4%
Medium (2.4301–5.00)	9.0%	0.0%	2.3%	6.5%	12.8%	28.1%
High (5.001–9.67)	13.7%	0.0%	5.9%	11.4%	19.2%	33.6%
Very High (9.671 +)	11.2%	0.0%	3.1%	8.6%	16.1%	32.9%
Doctor of Chiropractic FTE / 100,000 beneficiaries
Low (.00 – .25)	9.2%	0.0%	1.6%	6.8%	12.2%	27.8%
Middle (.25–.55)	7.0%	0.0%	0.0%	5.1%	9.8%	23.0%
High (.55 – 1.37)	11.8%	0.0%	5.2%	9.9%	16.2%	30.8%
Very High (1.37+)	14.1%	0.0%	5.6%	11.5%	20.1%	35.9%
C. Estimated percent of cases received opioid prescription at index visit
Explanatory Factor	Mean	5th Percentile	25th Percentile	Median	75th Percentile	95th Percentile
Overall	11.6%	0.0%	4.4%	10.3%	18.2%	24.2%
Type of provider
MD/DO	11.6%	0.0%	4.1%	9.1%	16.3%	32.8%
Physician Assistant	10.9%	0.0%	3.3%	7.7%	14.7%	34.2%
Nurse Practitioner	13.4%	0.0%	5.1%	10.5%	18.6%	36.0%
Fiscal Year of index visit
2011	13.2%	0.0%	0.0%	7.7%	19.4%	45.8%
2012	13.6%	0.0%	0.0%	7.0%	19.5%	52.2%
2013	13.1%	0.0%	0.0%	7.0%	19.0%	49.9%
2014	12.3%	0.0%	0.0%	6.0%	17.9%	45.9%
2015	10.7%	0.0%	0.0%	4.5%	15.4%	43.0%
2016	9.7%	0.0%	0.0%	0.0%	13.0%	43.7%
Physical therapist FTE / 100,000 beneficiaries
Low (0–2.43)	11.5%	0.0%	3.6%	9.3%	16.5%	32.7%
Medium (2.4301–5.00)	11.4%	0.0%	3.9%	8.6%	16.2%	32.3%
High (5.001–9.67)	11.3%	0.0%	3.8%	8.0%	15.4%	35.4%
Very High (9.671 +)	11.9%	0.0%	3.7%	9.1%	16.8%	33.7%
Doctor of Chiropractic FTE / 100,000 beneficiaries
Low (.00 – .25)	12.6%	0.0%	4.4%	10.0%	18.2%	35.5%
Middle (.25–.55)	10.6%	0.0%	3.2%	8.1%	14.9%	31.5%
High (.55 – 1.37)	12.1%	0.0%	4.3%	9.2%	16.8%	36.3%
Very High (1.37+)	11.1%	0.0%	3.6%	8.2%	15.3%	31.7%

Note: Overall and type of provider analysis based on sample size of 4,302 clinicians (in italics). All other analysis based on physician-fiscal year data (n=15,542); hence percentile values may reflect more outliers than provider analysis. Adjustment percent of cases is based on regression models using gender and medically based indicators as adjustors.

Mean reliability for provider level estimates of exercise therapy is .739

Mean reliability for provider estimates of spinal manipulation is .666

Mean reliability for provider estimates of opioid at first visit is .815

Variation over Time

Examining trends over time (Table 3), our models found that clinician utilization rates changed significantly between 2012 and 2016 with decreases in opioid prescribing and increases in cases’ receipt of NPT. Mean opioid prescribing rates in 2011 were 13.2% and 9.7% in 2016; mean rate of receipt of spinal manipulation in 2011 was 10.7% and in 2016 was 11.7%; mean receipt of exercise therapy in 2011 was 13.5% and in 2016 was 19.7%.

In bivariate analyses, we found a pronounced shift over time in the proportion of spine cases presenting in facilities with low versus very high availability of physical therapists and doctors of chiropractic (Table 4). While in FY2011 41.9% of back pain cases were seen in facilities with low PT availability (0 −2.43 FTEs per 100,000 annual patients), only 8.6% were seen in such facilities in FY2016 and more than one-third (36.5%) of spine pain cases were seen in facilities classified as very high PT availability. A similar but less pronounced shift over the same timeframe was observed in the availability of DCs in facilities: a sharp decline in the proportion seen in low-capacity facilities and a concomitant increase in the proportion seen in very high-capacity facilities. We note that the availability rate of PTs was much higher than that of DCs.

Table 4.

Distribution of Spine Pain Cases by Facility-level Capacity of Physical Therapists and Doctors of Chiropractic (FTE per 100,000 beneficiaries) by fiscal year

	fy2011	fy2012	fy2013	fy2014	fy2015	fy2016
Case count by year	42,989	52,668	53,104	52,018	48,534	29,704
Physical therapy FTE / 100,000 beneficiaries
Low (0–2.43)	41.9%	44.1%	24.0%	15.9%	7.9%	8.6%
Medium (2.43 −5.00)	23.2%	19.3%	34.3%	31.4%	20.5%	20.0%
High (5.01–9.67)	20.7%	19.6%	20.1%	22.2%	34.7%	34.9%
Very High (9.671+)	14.1%	16.8%	21.5%	30.3%	36.7%	36.5%
Doctors of Chiropractor FTE / 100,000 beneficiaries
Low (.00–.25)	35.4%	33.2%	23.7%	21.9%	13.5%	13.1%
Medium (.25–.55)	31.2%	33.0%	17.2%	17.2%	28.0%	28.2%
High (.55– 1.37)	21.6%	15.8%	36.1%	33.1%	20.2%	20.4%
Very high (1.37+ )	11.7%	17.9%	22.9%	27.7%	38.1%	38.1%

Factors Associated with Adjusted Clinician Rates

We conducted hierarchical linear models (with cases nested inside primary care clinicians) to identify factors correlated with clinician provision rates of treatments to test our three hypotheses. The hierarchical models showed that only clinician early provision rates comparing APNs to physicians differed significantly although the regression coefficients indicate the differences are relatively small (Table 5): compared with MD/DOs, APNs had higher rates for initial opioid prescriptions and lower likelihood of cases receiving spinal manipulation within 30 days. In contrast, cases seen by PAs did not differ significantly from physicians.

Table 5.

Association of factors with adjusted clinician therapy and opioid prescribing rates

	Model Dependent Variables: clinician choice of therapy
a. Independent variables	Exercise Therapy within 30 days			Spinal Manipulation within 30 days			Opioid prescrip. at initial SP visit
	Estimate	Std Err	T value	Estimate	Std Err	T value	Estimate	Std Err	T value
Clinician type
Physician (reference group)
Physician Assistant	0.0012	0.0035	0.34	−0.0056	0.0033	−1.71	−0.0073	0.0035	−2.11
Adv Practice Nurse	0.0032	0.0045	0.70	−0.0180	0.0043	−4.16	0.0138	0.0046	3.02
Physical Therapist availability group									.
Very high (reference)
High	−0.0014	0.0043	0.33	0.0199	0.0038	5.20	0.0202	0.0041	4.88
Medium	−0.0161	0.0050	−3.24	0.0041	0.0044	0.94	0.0237	0.0047	4.99
Low	−0.0338	0.0048	−6.97	−0.0075	0.0043	−1.73	0.0153	0.0047	3.29
Doctors of Chiropractor availability group
Very High (reference)
High	−0.0015	0.0046	−0.33	−0.0199	0.0041	−4.87	−0.0008	0.0044	−0.19
Medium	−0.0085	0.0052	−1.62	−0.0534	0.0046	−10.64	−0.0084	0.0049	−1.69
Low	−0.0182	0.0042	−4.30	−0.0405	0.0038	−11.65	0.019.	0.0041	4.55
b. Independent variables	Exercise Therapy within 30 days			Spinal Manip. within 30 days			Opioid prescrip. at initial BP visit
	Estimate	Std Err	T value	Estimate	Std Err	T value	Estimate	Std Err	T value
Fiscal Year
2011	−0.0598	0.0052	−11.51	−0.0111	0.0045	−2.49	0.0357	0.0048	7.40
2012	−0.0712	0.0050	−14.15	−0.0238	0.0043	−5.50	0.0408	0.0047	8.72
2013	−0.0564	0.0050	−11.34	−0.0127	0.0043	−2.99	0.0357	0.0046	7.74
2014	−0.0469	0.0050	−9.43	−0.0076	0.0042	1.79	0.0269	0.0046	5.86
2015	−0.0312	0.0050	−6.25	−0.0014	0.0042	0.34	0.0112	0.0046	2.43
2016 (reference group)

Bolded results were significant at the p<0.01 level

Notes: Two model specifications were estimated, with different independent variables: (a) clinician type, physical therapist availability group, and chiropractor availability group; (b) fiscal year. The rates estimated were adjusted for age group, and indicator variables for mental health and pain morbidities and indicators for invasive surgery, hospitalization, emergency department visit, and opioid prescription fill identified in the 90 days prior to the index visit.

Our analyses of the association between PT treatment availability and clinician use rates found some significant and interesting results. Compared with clinicians at facilities with very high PT availability, clinicians at facilities with lower PT availability had lower rates of exercise therapy and clinicians at facilities with lower PT availability had higher rates of opioid at initial visit. Analyses of the relationships between DC availability and clinician provision rates of treatment also identified significant associations. Compared with clinicians at facilities with very high DC availability, clinicians at facilities with low DC availability had lower rates of exercise therapy and higher rates of opioid prescription (at initial visit), and clinicians at all groups of three facilities with lower DC availability had lower rates of spinal/chiropractic manipulation.

Proportions of unexplained variation attributable to clinicians versus facilities

Our hierarchical models showed that, after adjustment for covariates, there still remained substantial clinician and facility variation among the outcomes (See Appendix Table 1, intra-class correlation values). With respect to the remaining unexplained variation of exercise therapy, 4.6% was attributable at the clinician level and 1.1% at the facility level. For spinal manipulation, the corresponding values were 7.1% at the clinician level and 2.1% at the facility level, and for an opioid prescription at initial visit, the overall contribution was the highest, 10.0% at the clinician level and 1.0% at the facility level.

DISCUSSION

Appropriate treatment for spine pain is vital for military readiness because it is a common disability; hence an understanding of the variation in treatment at the clinician level may be foundational for quality evaluation. In this study, based on outpatient visits for new episodes of spine pain, we have found substantial variation in both opioid prescribing rates and in receipt of exercise therapy and/or spinal manipulation. Regarding our hypotheses, we found only minimal differences in the clinical approach of MDs relative to APNs and PAs, and over time, we found substantial decrease in opioid prescribing and substantial increase in use of exercise therapy but only modest increase in use of spinal manipulation. Our analyses supported the hypothesis that availability of PTs and DCs would have a large impact on the use of NPTs, and we found larger unexplained variation at the clinician level than at the facility level.

Patients who see clinicians who are more likely to prescribe opioids have been shown to be at somewhat higher risk of long-term opioid receipt [22] and may have lower job performance and productivity [23]. Despite the attempt of expert guidelines to steer clinicians away from prescribing opioids for acute LBP [6], there are many reasons to expect variation. First, there will be different intensities of pain and different patient expectations of medication pain relief. Second, there may be other variables particular to differing contexts [24]. Third, there are differing clinician views toward the risks of opioid medication [25]; [26]; [27]. Finally, different clinicians’ personalities may drive different practice approaches. A study conducted with veterans in the Veterans Administration medical system using latent class analysis identified three treatment approaches among primary care physicians addressing chronic pain: 1) multimodal (aggressive) 14%; 2) low action (38%), and psychosocial/non-opioid (48%). These choices were influenced by the availability of resources, opioid-related beliefs, and practice characteristics [26].

Although expert guidelines do not recommend exercise therapy for acute LBP, several factors might weigh toward physical therapy referral. First, it should be noted that the guidelines are based on very few studies, and they have low strength of evidence (SOE); guidelines with low SOE have been shown to produce clinician hesitancy [28]. Second, there are studies which show benefit from early physical therapy [29] and studies that show better outcomes for patients who begin their treatment with physical therapy rather than with a medication-prescribing clinician [11]. Third, patient preferences for NPT are strong and can move prescribers to physical therapy referrals. Not surprisingly, patients who have seen physical therapists in the past are more likely to return to physical therapy [30]. As one example of patient characteristics driving physical therapy referrals, tobacco users with low back pain have been found less likely to receive a PT referral [31]. Fourth, referring clinicians are often part of larger practices that include physical therapists who may provide them with confidence that early physical therapy has merit.

We found greater variation in clinicians’ practice patterns than in facility rates. This is noteworthy as a Veterans Administration study found wide facility-level variation in utilization of different pain-related modalities, including initiation of long-term opioid use and use of NPT [32]; this study did not control for provider level variation.

We have found that facilities with higher NPT staffing ratios provided more NPT services. This is supported by findings in the province of Quebec, Canada of the use of early physiotherapy services. That study concluded that variation in provision of physiotherapy service did not appear to be related to demand for care, but rather was driven by wait times and indicators of insufficient provision of service [33].

We also reported an inverse association of the availability of PTs and chiropractors with the tendency to prescribe opioids at the first SP visit. A cross-sectional study of younger (disabled) Medicare beneficiaries also reported that the per-capita supply of DCs and spending on chiropractic manipulative therapy were inversely correlated with the percentage of younger Medicare beneficiaries receiving at least one opioid as well as 6 or more opioid prescription fills [34].

In this study, we focused on exercise therapy and spinal manipulation as the most utilized NPTs, but the MHS has added more modalities (acupuncture, stress management/relaxation, mindfulness meditation) [35]. As noted, the MHS has also employed more PTs and chiropractors [36].

Referral for spinal manipulation may take two paths. The bulk of spinal manipulation in our study is provided by physical therapists and is thus often part of a regimen that also includes exercise therapy. The second path, referral to a chiropractor, is less common in the US Army where chiropractic staffing is far below that of physical therapy [36]. This mirrors civilian practice where referrals from physicians to chiropractors are far less common than patient self-referral to chiropractors and some physicians have negative attitudes toward chiropractic care [37]. Complicating care paths in a military population, it should be noted that service members may delay or forgo care for pain because it may be perceived to interfere with their military career or take them away from a mission [38][39].

Understanding the amount of variation in treatment approaches is a pillar of quality improvement [40]. In this study we have evaluated the amount of variation that can be ascribed to individual clinicians and the amount of variation associated with differing treatment facilities. Given the significant risks [41],[42], and uncertain benefits [43] associated with opioid treatments it behooves clinicians and health care systems to analyze the value of treating acute pain with opioids. Understanding that each circumstance must be evaluated individually, various initiatives have been undertaken to improve clinician/patient communication and decision-making around the issue of opioid receipt [44][45][46][47].

There has been little attention to diminishing variation around the volume of referrals to physical therapy and an argument has been made that current practice patterns that allow same-day opioid receipt but create some barriers to NPT receipt is counterproductive [48]. There have been several initiatives to better coordinate prescribing-clinician approaches with NPT-treating clinicians. An orthopedic performance improvement initiative in the U.S. Navy involved audit and feedback of all orthopedic consults leading to a 27% decrease in orthopedic referrals with a 32% increase in physical therapy referrals leading to a 33%-dollar savings [49].

In future analyses we intend to examine the relationships between clinicians’ treatment approach and later outcomes. For instance, some studies have used a quasi-random approach to measure outcomes for patients seen by emergency department clinicians who prescribe opioids at higher or lower opioid rates [50][22][23] and we plan to use a similar approach to study outcomes resulting from opioid treatment of acute SP. Given the large investment in NPTs for acute SP and the uncertain benefit associated with these treatment modalities, we plan to evaluate outcomes for patients seen by clinicians with higher or lower NPT referral rates.

Limitations: There are several limitations to this study. First, we focused exclusively on primary care encounters for spine pain within MTFs, although some Army soldiers who are in communities without a local MTF will receive their care in a civilian medical clinic. While findings cannot be generalized to those settings, most Army soldiers receive most of their care in MTFs. We demonstrated that NPT rates and opioid prescribing rates changed slightly over time, thus, these findings may not generalize to current practices in the MHS. Finally, these findings from an Army sample may not generalize to providers who predominantly serve military members from other services.

CONCLUSIONS

In summary, wide variation in clinicians’ referrals for exercise therapy and spinal manipulation are not surprising given CPGs which suggest that “clinicians should reassure patients that acute or subacute low back pain usually improves over time, regardless of treatment [4].” These recommendations, based on a few studies and low-quality evidence, should be weighed against the findings of a recent large study of patients at 77 primary care practices in 4 regions across the United States; this study found the overall transition rate from acute to chronic low back pain at six months was 32% [51]. Clearly, a search for effective treatment of acute spine pain is important.

Appendix

APPENDIX Table 1.

Random-effects Intra-class Correlation Results from Mixed-Effects Multi-level Regression Models

Model statistics	Models 1 – 4 outcome measures: Probability of Receipt of Service
	Exercise therapy within 30 days	Spinal Manipulation within 30 days	Either Exercise T. or spinal manipulation within 30 days	Opioid at intake encounter
MTF-level (n=121)
ICC Estimate	.011	.034	.021	.010
Std. Error	.003	.006	.004	.003
95% confidence interval	.007 – .018	.024 - .048	.014 – .031	.006- .017
Provider-level (n=5,961)
ICC Estimate	.046	.096	.071	.10
Std. Error	.003	.006	.004	.003
95% confidence interval	.041 - .052	.085-.109	.063 - .079	.098 - .110

Note: Sample size of index episodes n=279,012. Covariates included in the models (not reported here) were sex, age group, and indicator variables for morbidity identified in prior 90 days to index visit: invasive surgery, hospitalization, emergency department visit, and opioid prescription fill.

Appendix Figure 1. Study Flow Diagram.

Note that providers in this study may have had other spine condition cases during the period for non-SUPIC (Army active duty and Reserve component) patients (e.g., other than Army service branch, dependent beneficiary, retired beneficiary). Abbreviations: APN = advance practice nurse, DO = doctor of osteopathy, FY = federal fiscal year starting October 1, MD = medical doctor, MHS = Military Health System, MTF = military treatment facility, PA = physician assistant