Economic Evaluation: A Randomized Pragmatic Trial of a Primary Care-Based Cognitive Behavioral Intervention for Adults Receiving Long-Term Opioids for Chronic Pain

Abstract

Background

Chronic pain is prevalent and costly; cost-effective non-pharmacological approaches that reduce pain and improve patient functioning are needed.

Objective

Report the incremental cost-effectiveness ratio (ICER),compared to usual care, of cognitive behavioral therapy (CBT) aimed at improving functioning and pain among patients with chronic pain on long-term opioid treatment.

Design

Economic evaluation conducted alongside a pragmatic cluster randomized trial

Subjects

Adults with chronic pain on long-term opioid treatment (N=814)

Intervention

A CBT intervention teaching pain self-management skills in 12 weekly, 90-minute groups delivered by an interdisciplinary team (behaviorists, nurses) with additional support from physical therapists, and pharmacists.

Outcome Measures

Cost per quality adjusted life year (QALY) gained, and cost per additional responder (≥ 30% improvement on standard scale assessment of Pain, Enjoyment, General Activity and Sleep). Costs were estimated as-delivered, and replication.

Results

Per patient intervention replication costs were $2,145 ($2,574 as-delivered). Those costs were completely offset by lower medical care costs; inclusive of the intervention, total medical care over follow-up was $1,841 lower for intervention patients. Intervention group patients also had greater QALY and responder gains than did controls. Supplemental analyses using pain-related medical care costs revealed incremental cost-effectiveness ratios (ICERs) of $35,000, and $53,000 per QALY (for replication, and as-delivered intervention costs, respectively); the ICER when excluding patients with outlier follow-up costs was $106,000

Limitations

Limited to one-year follow-up; identification of pain-related utilization potentially incomplete

Conclusion

The intervention was the optimal choice at commonly accepted levels of willingness-to-pay for QALY gains; this finding was robust to sensitivity analyses.

Introduction

Chronic pain is common and exacts a high cost on patients and the healthcare system, with costs of lost productivity overshadowing healthcare costs.(1) Indeed, chronic pain costs are greater than those of many common chronic diseases (i.e. heart disease, cancer or diabetes), in part due to its high prevalence.(1) Opioids have been the principal clinical tool for pain treatment, however due to their attendant societal and clinical problems,(2-4) alternative treatments are needed. Recent work has focused on multidisciplinary approaches to pain management.(5) However, analyses of effects and cost-effectiveness of such approaches in patients on long-term opioid therapy is limited. In this paper we report an economic analysis conducted alongside a pragmatic cluster randomized trial (Pain Program for Active Coping & Training (PPACT); ClinicalTrials.gov: NCT02113592) that evaluated the effectiveness, compared to usual care, of a multidisciplinary cognitive behavioral intervention aimed primarily at improving functional status, and secondarily at reducing opioid and benzodiazepine use among chronic pain patients. Our objectives were to 1) estimate the intervention’s cost, 2) report follow-up healthcare costs, and 3) to estimate the incremental cost-effectiveness of the intervention (compared to usual care).

Methods

We followed state-of-the-art guidelines(6) for reporting of cost-effectiveness analyses.

Trial design:

The study was conducted at three sites (Kaiser Permanente regions; Northwest, Southeast, and Hawaii) and approved by their institutional review boards.

Subjects:

Primary care clinician clusters (n=106; 850 patients) were randomized to treatment or control (usual care); we included 814 patients with complete outcome data in this economic evaluation. Full details of the parent trial are available elsewhere.(7, 8) In summary: subjects were aged 18 and over; health plan members for at least 6 months prior to recruitment; had at least one non-cancer pain diagnosis in the 12 months prior; and were long-term opioid users [defined as (i) having at least 2 dispenses of long-acting opioids in the 6 months prior or (ii) at least a cumulative 90-day supply of short-acting opioids during any 4-month period within the 6 months prior. Eligibility screening required pain interference with general activity ≥4 on an 11-point scale. Patients in both groups were allowed to continue opioids during the trial.

Intervention:

The intervention consisted of 1) comprehensive intake evaluation (included physical therapy (PT) evaluation and pharmacist medication review), 2) cognitive behavioral therapy (CBT)-based pain coping skills training and yoga-based adapted movement practice provided in 12 weekly group sessions (delivered by behavioral specialist and nurse; including individual catch-up sessions if needed), and 3) primary care provider (PCP) consultation and patient outreach. This primary care-based multidisciplinary intervention, directed at helping patients develop pain self-management skills, included behavioral specialists, nurses, physical therapists, and pharmacists.

Usual care:

Patients randomized to the control group continued to receive the usual care provided by their PCP, including pharmacologic and nonpharmacologic treatments, without restriction.(7)

Patients in both groups were followed for 12-months post-randomization.

Outcomes, assessed at baseline and quarterly over 12 months for both treatment and control patients, included pain impact (4-item assessment of Pain, Enjoyment, General Activity, Sleep (PEGS);(7, 9) a variant of the PEG that includes a 4^th item on pain interference with sleep) and, secondarily, pain-related disability (24-item Roland Morris Disability Questionnaire; RMDQ). (10-14) As was done in the clinical trial analysis,(7) patients with a 30% or greater PEGS improvement from baseline were defined as ‘treatment responders’.(15)

Clinical trial results:

Intervention patients showed larger improvements on pain impact (PEGS: −0.434 point (95% CI, −0.690 to −0.178 point) and pain-related disability (RMQD: −0.060 point (CI, −0.084 to −0.035 point). Opioid use did not differ significantly between groups.(7)

Economic outcomes:

We report 2 incremental cost-effectiveness ratio (ICER) outcomes established a priori; 1) cost per quality adjusted life year (QALY) gained, and 2) cost per additional treatment responder. A lower ICER reflects more favorable value. While we could find no guidance on threshold values for ICER on additional treatment responder, commonly accepted thresholds for willingness-to-pay for (WTP) per additional QALY gained range from $100,000 to $150,000 in the US.(16-18)

Utility estimation:

The importance of a pragmatic trial (versus an explanatory trial) is to prioritize ‘real world’ application to support decisions regarding intervention adoption in standard care.(19) As a pragmatic trial, the study purposively limited study assessment length at each timepoint. To minimize intrusion in clinical care, we used the clinical trial’s findings of PEGS and RMDQ, assessed at baseline, and 3, 6, 9, and 12 months, to estimate utilities at those same timepoints. To perform this estimation, we obtained patients’ information on utilities by administering a one-time survey to participants after their 12-month study completion (response n=516) which included the EuroQol-5 Dimension (EQ-5D-5L)(20, 21) and concurrent PEGS and RMDQ assessments. We then used direct utility mapping (a common approach in economic evaluation) on those survey data.(22) Specifically, EQ5D-5L utility values (US scoring)(23) for each patient/quarterly timepoint were predicted from the post-trial survey data, using regularized linear regression; predictor variables included PEGS, RMDQ, age, sex, baseline diagnoses (anxiety, depression), site, and randomized group.(24). Regularized regression seeks to minimize the sum of squared residuals while imposing a penalty to limit model complexity, which helps to reduce the potential of overfitting and improve the accuracy of out of sample predictions.(25) We used the lasso method, penalizing the absolute size of coefficients.(26, 27) The predicted values we used are based on the ordinary least squares coefficients from the set of predictors selected by the lasso to alleviate bias induced by regularization.(28) The EQ5D utility data (collected from the post-study survey) did not appear multimodal nor were there any obvious utility “gaps”, minimizing bias concerns(29) with basic regression modeling. The distribution of the residuals did not substantially depart from normality. A small proportion (10%) of patient’s quarterly measures were missing RMDQ values; for those timepoints utility estimation was done without the RMDQ.

Intervention Costs:

Our analysis takes the payer (health plan) perspective. Intervention costs include both labor and non-labor inputs (Table 2). Examples of labor tasks were patient identification, intake, and scheduling, intervention training and delivery, and patient charting. A tracking system was used to estimate intervention staff costs, wherein staff reported time spent on tasks in support of the intervention. Recognizing that these costs include the cost of doing the research, adjustments subtracted out research protocol-driven costs to arrive at an ‘as-delivered’ cost. We also estimated the replication costs (i.e. the cost that might be incurred by a health plan adopting our intervention without major modification). After discussing principles of cost allocation, adjustments were arrived with consensus from study staff, investigators, and health plan employees who currently administer similar programs. Thus, for both labor and non-labor intervention inputs, we estimated cost per participant in two ways: 1) as-delivered (adjusted to exclude research protocol driven costs); and 2) replication cost (further adjusted to reflect a referral-based adoption in other health plans). Wage rates come from study personnel salary; to fully allocate the labor costs we added a fringe benefit of 30% and overhead of 20%.(30, 31) Other estimates are actual paid amounts to vendors in the trial. To reflect the opportunity cost of the group sessions, we assigned the session costs regardless of patient’s attendance. Coaching calls (weekly outreach to intervention patients to assist them in applying coping skills into their daily lives) and rescue sessions (one-on-one, typically telephonic sessions for missed group visits) were costed as variable (per unit of use).

Table 2 -.

Resource components and costs

		As delivered	Replication
	Subcomponents and description
Patient Selection
	Population Identification/Program Manager1	$3.30	$2.97
	PCP approval/referral 2	$6.95	$17.70
	Patient outreach (initial mailing, phone calls to explain
	intervention/orient, notification and scheduling) 3	$266.70	$56.03
Intervention
	Comprehensive Intake (3 separate visits)	$275.79	$275.79
	Pharmacist review and input to PCP	$115.71	$115.71
	PCP/Interventionist Consult at Intake	$38.75	$38.75
	PCP/Patient Outreach at intake	$9.35	$9.35
	12 Group Sessions (2 hours each)	$606.92	$606.92
	PT Mid-program visit	$92.63	$92.63
	Post-program visit	$82.26	$82.26
	Charting/Administrative work 4	$342.82	$205.69
	Study materials (handouts, MP3 player and recordings, yogaDVD)	$46.91	$46.91
	Coaching Calls (telephone call between sessions) Average call= 15 min/ 10 participants	$224.07	$224.07
	Rescue Session (Missing Session Makeup)	$14.10	$14.10
	DVD of Sessions - Filming time 5	$3.75	$0.00
	Production cost 5	$8.66	$0.00
	Interactions between interventionist and PCP duringintervention period	$14.42	$14.42
	PCP/Interventionist Consult at end of intervention	$43.68	$43.68
	PCP/Patient Outreach at end of intervention	$9.35	$9.35
	Intervention Supervision	$28.45	$28.45
	Meeting room space group sessions	$232.96	$232.96
Data Collection 6
	Quarterly Administration of PEGS & RMDQ	$2.12	$0.00
	Email message	$6.36	$0.00
	Staff Call	$38.06	$0.00
	Data Collection Overhead	$10.09	$0.00
Training
	Base Training 7	$15.02	$11.57
	Turn Over 8	$11.70	$8.78
	Ongoing training and monitoring 9	$23.43	$7.03
Total		$2,574.28	$2,145.10
	Variable cost of coaching calls (n=1,472)	$65.91	$65.91
	Variable cost of rescue sessions (n=486)	$23.62	$12.56

¹As delivered includes initial data pull and population identification; Replication includes Program Manager review

²For replication costs increased physician time was added for patient referral and identification

³Intital and orientation phone contacts were removed from replication costs

⁴Charting in Replication was reduced to 30% to reflect clinical practice

⁵DVD filming and production costs excluded from Replication since DVD available from trial

⁶Data collection is excluded from Replication to reflect clinical practice

⁷Base training is depreciated over a 10 year period, therefore .1 is allocated to year 1

⁸Replication cost include 75% of staff turnover

⁹Replication costs include 30% of ongoing training and monitoring

Healthcare utilization and costs:

Healthcare utilization information came from electronic virtual data warehouses (VDW) maintained by each site. These data warehouses compile comprehensive information from electronic medical records, hospital discharge abstracts, claims, outpatient pharmacy and other health plan administrative data sources; data were standardized across plans.(32) We divided healthcare utilization into ambulatory, telephone or email, hospital stays, and pharmacy, as well as pain-related procedures (injections, surgery and imaging). Administrative, diagnostic, and billing codes used to identify those utilization categories are available on request. The costs of medical care encounters were estimated using the standardized relative resource cost algorithm (SRRCA).(33) The SRRCA adapts the 15 payment systems the Centers for Medicare & Medicaid Services uses to reimburse providers for covered services into costing modules (e.g., inpatient, outpatient, hospital ambulatory) to calculate encounter level cost estimates from VDW data.

We report total costs, and, separately, pain-related costs. Total costs included all encounters over follow-up. Pain-related medical care encounters were identified using diagnostic(34) and procedure coding. Pain-related medications were identified using dispensings from outpatient pharmacy records for oral and transdermal opioids, and non-opioid pain-related medications (nonsteroidal anti-inflammatory drugs (NSAIDs), sedative hypnotics, anticonvulsants, antianxiety agents, and antidepressants).(35)

Statistical Analysis:

We calculated the ICER as the ratio of extra cost to the extra effect (i.e., ΔC/ΔE). We used net benefit regression methods,(36-38) with 1000 bootstrapped samples, adjusted for baseline utility and cost to estimate the intervention’s probability of being cost-effective for each outcome, illustrated using a cost-effectiveness acceptability curve (CEAC).(39) A CEAC displays a range of willingness-to-pay for a unit outcome gain on its x-axis; the y-axis shows each group’s probability of being cost-effective along the range of willingness-to-pay. We also created a cost-effectiveness plane,(39) recording each bootstrap sample’s incremental cost and incremental effect estimates. By convention, we reported the percent of bootstrap estimates divided into quadrants. Two of these quadrants are definitive for cost-effectiveness; SE quadrant (intervention was less costly and more effective; always cost-effective); and NW (intervention was more costly and less effective; never cost-effective). Estimates falling into the other quadrants may be cost-effective, depending on a decision-maker’s willingness-to-pay for a unit of effectiveness; NE (intervention was more costly and more effective); and SW (intervention was less costly and less effective). Additionally, as described above, we report costs associated with pain-related care. Costs and counts of healthcare utilization were compared using generalized linear regression (for costs identity link, normal distribution; for utilization, log-link, negative binomial distribution). We investigated the distribution of baseline and follow-up costs. Mean baseline total costs were higher in the control group ($11,766) than for treatment group ($9,985). We adjusted all analyses to account for this difference; baseline utility was additionally adjusted in the net-benefit regression. To understand the effect of outliers, we identified baseline cost outliers as patients with costs above the 95^th percentile of baseline costs; follow-up cost outliers were patients with costs above the 95^th percentile of follow-up costs.

Our base-case was the cost per QALY gained and included intervention replication costs with total follow-up medical care cost. We also present 6 supplemental ICER analyses that varied the outcome and/or cost considerations on the base-case, incremental cost : 1) per additional treatment responder; 2) per QALY gained using as-delivered intervention cost; 3) per QALY gained using pain-related medical care costs; 4) per QALY gained using as-delivered intervention costs and pain-related medical care costs; 5) per QALY gained with baseline cost outliers excluded; and 6) per QALY gained with follow-up cost outliers excluded.

The intra-class correlation coefficient on PCP-clustering was very small (.006) and the likelihood ratio test comparing the mixed model with a random effect for PCP and linear regression model without that random effect was not significant (p=.41). This suggests that the independence of observations assumption was not violated by using simpler models that do not account for PCP clustering. SAS [version 9.4] and STATA [version 15] were used for all analyses.

Results

Baseline characteristics of the population are shown in Table 1. Per-patient costs of the intervention and its delivery (Table 2) were estimated at $2,574 and $2,145 for as-delivered and replication, respectively. The group sessions ($607) were the largest contributor to the overall intervention cost. In the replication cost scenario, we assumed the program would be referral based (versus a centralized identification process as was done in this trial); this decreased the patient selection costs by more than $200, compared to as-delivered.

Table 1.

Baseline Characteristics of Patients in the PPACT Study

	N (%)
Characteristic	All Patients	Intervention Patients	Usual Care Patients
Patients (n=814 [Total], n=412[Intervention] and n=402 [Usual Care]) a
Age, median (IQR), y	60.4 (12.0)	61.6 (11.7)	59.2 (12.2)
Sex
Women	550 (67.6)	270 (65.5)	280 (69.7)
Men	264 (32.4)	142 (34.5)	122 (30.4)
Race
White	623 (76.5)	318 (77.2)	305 (75.9)
Black or African American	106 (13.0)	51 (12.4)	55 (13.7)
Other	85 (10.4)	43 (10.4)	42 (10.5)
Hispanic ethnicity	28 (3.4)	16 (3.8)	12 (3.0)
Receive disability benefitsb	205 (25.2)	98 (23.8)	107 (26.2)
Current smokingb	132 (16.3)	61 (14.8)	71 (17.8)
Body mass index (BMI)c, mean (SD)	32.8 (8.9)	32.8 (9.1)	32.8 (8.7)
Alcohol abuse (history and/or current)b	36 (4.4)	18 (4.4)	18 (4.5)
Drug abuse (history and/or current)b	41 (5.0)	23 (5.6)	18 (4.5)
Chronic co-morbiditiesb
Diabetes	193 (23.7)	108 (26.2)	85 (21.1)
Cardiovascular disorder	190 (23.3)	98 (23.8)	92 (22.9)
Hypertension	392 (48.2)	208 (50.5)	184 (45.8)
Chronic pulmonary disease	167 (20.5)	79 (19.2)	88 (21.9)
Two or more of above chronic co-morbidities	290 (35.6)	142 (34.5)	148 (36.8)
Mental health co-morbiditiesb
Anxiety	177 (21.7)	80 (19.4)	97 (24.1)
Depression	286 (35.1)	145 (35.2)	141 (35.1)
Other mental health diagnoses	40 (4.9)	18 (4.4)	22 (5.5)
Number of non-malignant chronic pain typesd, median (IQR)	4.0 (2.0)	4.0 (2.0)	4.0 (2.0)
Prioritized Patient Characteristics
Average daily morphine milligram equivalents (MME)b median (IQR)	29.75 (46.9)	28.8 (40.4)	30.9 (59.1)
Average daily MME ≤90b	150 (18.4)	66 (16.0)	84 (21.0)
Benzodiazepine receipt b	214 (26.3)	105 (25.7)	108 (26.8)
High utilizer of primary care services (≥12 contacts in 3-month period)e	40 (4.9)	18 (4.4)	22 (5.5)

^aUnless otherwise stated.

^bAssessed for 180 days prior to randomization.

^cBMI for patient n=806 [Total], n=408 [Intervention] and n=398 [Usual Care]

^dAssessed for 360 days prior to randomization. Pain types include: 1) back and neck pain; 2) limb/extremity pain, joint pain fibromyalgia; 4) general and widespread pain; 5) headache; 6) orofacial, ear, and temporomandibular disorder pain; 7) abdominal and bowel pain; 8) urogenital, pelvic, and menstrual pain; 9) musculoskeletal chest pain; 10) neuropathy; 11) other painful conditions.

^eAssessed for 90 days prior to randomization; primary care services include in-person, phone and email encounters.

Healthcare costs were right skewed with unadjusted, (6 month) baseline mean total costs of $10,865 in the overall sample (median=$6,837, 95^th percentile=$34,672). Corresponding (12 month) follow-up mean total healthcare costs were $23,844 (median, $13,352, 95^th percentile=$74,166). We used the 95^th percentile of the overall sample’s baseline costs ($34,672) to determine baseline-cost outliers; a similar proportion (4.9%) of both the treatment and control group were considered baseline-cost outliers. Using the 95^th percentile of the overall sample’s follow-up costs ($74,166) to determine follow-up-cost outliers; we found 2.9% of the treatment and 7.0% of the control group were follow-up-cost outliers.

Table 3 shows that, adjusted for baseline utilization and excluding study-related visits, treatment and control patients used similar levels of total ambulatory visits (treatment=20.85 visits; 95% CI 19.4, 22.41, control=21.59; 95% CI 20.07, 23.22). But the treatment group had, on average, more PT visits (1.86; 95% CI 1.48, 1.11) than did controls (1.11, 95% CI 0.88, 1.41), while nearly every other resource use category was lower for the treatment group. For example, the treatment group had fewer pain-related imaging tests (treatment=1.55; 95% CI 1.35, 1.77, control=1.72; 95% CI 1.51, 1.96); mental health visits (treatment=0.73; 95% CI 0.56, 0.96, control=0.98; 95% CI 0.75, 1.29) and hospital days (treatment=0.86; 95% CI 0.59, 1.26, control=1.43; 95% CI 0.98, 2.09). Correspondingly, the adjusted mean total costs of care (exclusive of intervention costs) were lower for the treatment group by over $4,000 ($21,372; 95% CI $18,114, $24,630) than controls ($25,648; 95% CI $22,350, $28,946). The largest contributors to the difference in total cost were hospital stays (treatment=$4,546; 95% CI $1,936, $7,156, control=$7,891; 95% CI $5249, $10533) and pharmacy (treatment=$10,373; 95% CI $9,153, $11,592, control=$12,024; 95% CI $10,790, $13,258). When considering only pain-related utilization, the difference in overall adjusted cost was not as great as for total cost, but still favored the treatment group by about $2,000 (treatment=$9,395; 95% CI $7,216, $11,574, control=$11,343; 95% CI $9,137, $13,549).

Table 3.

Follow-up Utilization and Costs, adjusted for baseline utilization and cost

	Total Utilization				Pain Related Utilization
		Treatment		Control		Treatment		Control
	Mean	(95% CI)	Mean	(95% CI)	Mean	(95% CI)	Mean	(95% CI)
Encounters with pain-related procedures1
Pain-related injections	NA		NA		0.58	(0.49, 0.68)	0.58	(0.49, 0.69)
Pain-related surgeries	NA		NA		0.06	(0.04, 0.1)	0.09	(0.06, 0.14)
Pain-related imaging	NA		NA		1.55	(1.35, 1.77)	1.72	(1.51, 1.96)
Ambulatory visits
Primary care	4.91	(4.59, 5.24)	4.82	(4.51, 5.16)	3.03	(2.83, 3.26)	2.89	(2.69,3.11)
Physical therapy3	1.86	(1.48, 2.34)	1.11	(0.88, 1.41)	1.62	(1.27, 2.07)	0.9	(0.7, 1.17)
Pain clinic	0.5	(0.39, 0.65)	0.72	(0.57, 0.92)	0.46	(0.36, 0.59)	0.65	(0.51,0.83)
Other specialty medical care4	8.19	(7.5,8.94)	8.99	(8.23, 9.82)	2.99	(2.67, 3.34)	3.17	(2.83,3.54)
Mental health	0.73	(0.56, 0.96)	0.98	(0.75, 1.29)	0.04	(0.02, 0.07)	0.06	(0.04,0.09)
Complementary and alternative medicine (CAM)5	0.21	(0.11, 0.43)	0.23	(0.11, 0.46)	0.17	(0.08, 0.35)	0.22	(0.11,0.45)
ED, Urgent care, and Observation Beds	1.13	(0.98, 1.31)	1.35	(1.17, 1.55)	0.57	(0.48, 0.68)	0.72	(0.61,0.85)
Hospital ambulatory visits	0.58	(0.47, 0.73)	0.72	(0.58, 0.9)	0.25	(0.19, 0.34)	0.24	(0.17,0.32)
Same day surgeries	0.26	(0.2, 0.33)	0.25	(0.2, 0.32)	0.13	(0.1, 0.18)	0.16	(0.12,0.21)
Home health	1.36	(0.98, 1.9)	1.3	(0.93, 1.82)	0.62	(0.43, 0.89)	0.56	(0.39,0.81)
Total Ambulatory visits	20.85	(19.4, 22.41)	21.59	(20.07, 23.22)	10.42	(9.62, 11.28)	10.11	(9.33, 10.96)
Telephone or email encounters	24.04	(22.57, 25.6)	23.18	(21.75, 24.71)	2.78	(2.51, 3.08)	2.81	(2.53,3.11)
Hospital stays 6	0.22	(0.17, 0.28)	0.28	(0.22, 0.36)	0.07	(0.05, 0.1)	0.09	(0.07,0.13)
Hospital days6	0.86	(0.59, 1.26)	1.43	(0.98, 2.09)	0.22	(0.13, 0.39)	0.48	(0.28,0.82)
Total Pharmacy Dispenses	50.45	(48.53, 52.44)	52.22	(50.21, 54.3)	NA		NA
Opioid Pain Dispenses	NA		NA		13.38	(12.76, 14.02)	13.85	(13.21, 14.52)
Non-Opioid Pain Dispenses	NA		NA		11.12	(10.4, 11.89)	11.81	(11.04, 12.63)
	Total Cost				Pain Related Cost
	Treatment		Control		Treatment		Control
	Mean	(95% CI)	Mean	(95% CI)	Mean	(95% CI)	Mean	(95% CI)
Ambulatory visits
Primary care	$805	($749, $860)	$807	($751, $863)	$548	($509, $588)	$525	($486, $565)
Physical therapy3	$245	($196, $293)	$145	($95, $194)	$202	($155, $249)	$122	($75, $170)
Pain clinic	$130	($98, $162)	$150	($118, $183)	$123	($93, $153)	$141	($111, $172)
Other specialty medical care4	$1508	($1324, $1693)	$1772	($1585, $1959)	$551	($477, $626)	$582	($507, $657)
Mental health	$224	($169, $280)	$224	($167, $280)	$11	($5, $17)	$17	($11, $24)
Complementary and alternative medicine (CAM)5	$84	($36, $132)	$82	($34, $131)	$64	($25, $103)	$82	($42, $121)
ED, Urgent care, and Observation Beds	$1235	($893, $1577)	$979	($633, $1325)	$992	($693, $1292)	$685	($381, $988)
Hospital ambulatory visits	$827	($483, $1172)	$665	($316, $1013)	$309	($205, $413)	$301	($196, $407)
Same day surgeries	$417	($307, $528)	$431	($318, $543)	$200	($125, $275)	$264	($188, $340)
Home health	$355	($247, $463)	$318	($209, $428)	$159	($100, $217)	$131	($71, $190)
Total Ambulatory visit cost	$5869	($5206, $6532)	$5535	($4864, $6206)	$3175	($2787, $3563)	$2835	($2442, $3228)
Telephone or email encounters	$392	($365, $419)	$395	($368, $423)	$51	($45, $57)	$53	($47, $58)
Hospital stays 6	$4546	($1936, $7156)	$7891	($5249, $10533)	$1937	($−59, $3933)	$3588	($1567, $5608)
Total Pharmacy Dispenses	$10373	($9153, $11592)	$12024	($10790, $13258)	NA		NA
Opioid Dispenses	NA		NA		$1389	($1292, $1486)	$1452	($1353, $1550)
NonOpioid Pain Dispenses	NA		NA		$2987	($2704, $3271)	$3268	($2981, $3555)
Total Cost	$21372	($18114, $24630)	$25648	($22350, $28946)	$9395	($7216, $11574)	$11343	($9137, $13549)

^1.Pain-related procedures are identified by CPT, HCPCS and ICD-9-PCS or ICD-10-PCS codes. Pain-related surgeries or injections are also included in ambulatory and hospital stays categories (double counted; do not sum).

^2.In-person health care encounters with a pain-related ICD-9-CM or ICD-10-CM diagnostic code

^3.Physical therapy includes Physical Therapy, Occupational Therapy and Physiatry visits

^4.Specialty medical care includes in-person visits to any non-Primary care department that is not included in the table

^5.Complementary and Alternative Medicine includes Acupuncture, Chiropractic, Holistic Health, Naturopathy, and Osteopathy

^6.Pain-related hospitalizations had a primary (or principal) pain-related diagnostic code NA = not applicable

Table 4 shows that total healthcare cost over follow-up, including the replication cost of the intervention and its delivery, was lower for the treatment group by $1,841 (treatment=$23,665, control=$25,506), while pain-related costs (including intervention cost) were higher in the treatment group by $821 (treatment=$11,853, control=$11,032).

Table 4:

Incremental Cost Effectiveness

Base Case, cost per quality adjusted life year (QALY) using intervention replication cost, total medical care cost
	Cost	Incremental cost	QALY	Incremental QALY	ICER
Control group	$25,506		0.5459
Treatment group	$23,665	−1841	0.5695	0.0236	Intervention Dominant
Cost-effectiveness plane quadrant results: SE 80%; NW 0%; NE 20%; SW 0%

Dominant means the intervention had lower costs and greater effect

Supplemental analysis difference from base-case shown in italics

Cost-effectiveness plane quadrant results legend:

SE=intervention was less costly and more effective; always cost-effective

NW=intervention was more costly and less effective; never cost-effective

NE=intervention was more costly and more effective; cost-effective depending on willingness-to-pay for a unit of effectiveness

SW=intervention was less costly and less effective; cost-effective depending on willingness-to-pay for a unit of effectiveness

All analyses adjusted for baseline cost and baseline outcome measure

The treatment group had lower costs and greater QALY gains than did control patients (Table 4). This is illustrated in the cost-effectiveness plane (Figure 1, Panel A), which, as described in the methods, is a scatter plot wherein each point represents the incremental cost and outcome for each bootstrap replication. Of the observations, 80% (SE quadrant) indicated the intervention had greater QALY gains and lower total costs.; 20% (NE quadrant) had greater QALY gains and higher total costs. Figure 1 (Panel B) also shows the cost-effectiveness acceptability curve, allowing decision makers to choose their own willingness-to-pay for a unit outcome gain. For example, at a willingness-to-pay of $50,000 per QALY gain, the intervention had a 90% probability of being cost-effective.

Figure 1.

Cost-effectiveness Plane and Cost-effectiveness Acceptability Curve Base Case (cost per QALY gained, intervention replication cost, total healthcare cost)

Panel A: Cost Effectiveness Plane

Panel B: Cost Effectiveness Acceptability Curve

The intervention group also had more treatment responders and lower costs than controls (Supplemental Digital Content 1, ICER); 62% of the bootstrap replicates showed lower costs and better outcomes (Supplemental Digital Content 2, Cost-effectiveness plane). At a willingness-to-pay of $50,000 per treatment responder, the intervention had a 95% chance of being cost-effective (Supplemental Digital Content 3, CEAC). We found similar results for Supplemental Analysis 2 on the cost per additional QALY using intervention as-delivered costs. Including only pain-related costs increased the ICER on QALY gains compared to the base-case: Supplemental Analysis 3 had an ICER of $35,000 with a probability of cost-effectiveness over 50% at a willingness-to-pay of $45,000 per additional QALY. Supplemental Analysis 4 further included as-delivered intervention costs and had an ICER of $53,000 with a 50% probability of cost-effectiveness at a willingness-to-pay of $60,000 per additional QALY. Excluding baseline cost outliers (Supplemental Analysis 5) led to nearly identical follow-up costs between the groups with a probability of being cost-effective over 50% at a willingness-to-pay near $0 (i.e., intervention costs are completely offset). Excluding follow-up cost outliers (Supplemental Analysis 6) led to increased costs in the treatment group of $2,282 with a probability of being cost-effective over 50% at a willingness-to-pay of $105,000.

Discussion

We found the treatment group had lower total follow-up costs (including the cost of the intervention and its delivery) as well as greater outcome (QALY and responder) gain. The majority of the intervention was delivered in a group-format, making it potentially more affordable than individually delivered CBT interventions. Recent CBT-based electronically-delivered interventions(40, 41) may provide even greater value-for-money, given the potentially lower intervention-related costs of such approaches.

We found few multidisciplinary pain management studies that reported the cost per additional responder as an outcome.(5) Our study can help establish a benchmark; in this case, an intervention that has an 90% chance of being cost-effective at a willingness-to-pay of $50,000 per additional QALY had a 95% chance of being cost-effective at a willingness-to-pay of $50,000 per additional treatment responder.

There is evidence that the intervention had cost and utilization impacts beyond pain-related care. For example, the cost difference between treatment and control patients (exclusive of intervention costs; Table 4) is greater for total care costs (treatment group approximately $4,000 lower) than that for pain-related costs (treatment approximately $2,000 lower). This magnification of resource use when considering total healthcare cost may reflect the multifactorial nature of pain and its impact on other conditions.(42) We used diagnostic and procedure coding to identify healthcare encounters that were pain-related. We don’t know the test performance characteristics of those codes, so there may be some encounters that we missed (false negatives) as well as some false positives. Additionally, healthcare encounters, particularly those in primary care, are likely to involve treatment of several health conditions during the same visit. This joint production during visits makes it difficult to disentangle resource use aimed only at pain-related care. Similarly, coping with pain has many effects on patient’s well-being and can impact non-pain conditions(43) our methods of identifying pain-related resource use would have missed those effects.

We found that pain-related costs ($9,395 intervention; $11,343 control) were about half of the total costs ($21,372 intervention; $25,648 control); this pattern was seen across overall ambulatory visits, pharmacy, and inpatient stays. The lower pain-related costs meant the cost-offset available for the intervention’s cost was less; this was at least partly responsible for the observation that pain-related ICERs were less favorable than total cost ICERs.

We examined healthcare service use by category (Table 3) and found similar use of total ambulatory visits in both groups (treatment=20.85 visits; 95% CI 19.4, 22.41, control=21.59; 95% CI 20.07, 23.22). Treatment group patients used, on average, more PT services (1.86; 95% CI 1.48, 1.11) than did controls (1.11, 95% CI 0.88, 1.41), a result consistent with the intervention which included PT visits, and may have led patients to seek these non-intervention PT visits. While the confidence intervals between groups are overlapping for all categories in Table 3, these results yield some insights into potential mechanisms by which the intervention might lead to lower costs. For example, mental health, pain clinic visits, pharmacy dispensings, and inpatient days, were, on average, lower for intervention patients.

We noted cost differences during baseline (possibly due to cost outliers among this high-utilizing population), so we adjusted for baseline costs in all analyses. When we further considered the potential effects of outliers by removing patients with baseline outliers with costs above the 95^th percentile, the treatment and control patients had nearly identical mean follow-up costs (including both intervention and medical care costs). The cost per QALY gained in that baseline outlier analysis was $374, suggesting that even after further accounting for the potential effect of baseline outliers, the intervention represents good value-for-money at very low levels of willingness-to-pay for a QALY gain. When we excluded patients with costs above the 95^th percentile of follow-up costs from the whole sample, the ICER was just above $100,000 per QALY, often considered a limit threshold for cost-effectiveness. But the intervention may have caused patients to decrease follow-up healthcare utilization overall, particularly for the highest utilizers. By removing the influence of these patients with the highest follow-up costs (more prevalent in the control group (7.0%) vs. treatment (2.9%)), we may have removed some of the intervention’s effect on decreasing follow-up costs, leading to the markedly higher ICER in that analysis. However, it is also possible that at least some of those high-cost control patients at follow-up experienced severe healthcare events that the intervention could not have influenced.

Conclusion:

Intervention group patients had lower costs and better outcomes. We found the intervention to the optimal choice at commonly-accepted levels of willingness-to-pay; this finding was robust to most sensitivity analyses. Strengths of our study include that it was conducted alongside a pragmatic, randomized trial,(7) within integrated healthcare systems allowing a thorough assessment of costs from the payer perspective. There were few exclusion criteria, making the findings widely applicable. We don’t know the durability of treatment effects beyond the clinical trial’s 12-month follow-up. But if the impact of the intervention persists over time, the intervention would presumably show better value-for-money in the long run. The findings apply most directly to integrated delivery systems however, the intervention components are widely available outside these systems.