Are Nonpharmacologic Interventions for Chronic low Back Pain More Cost Effective Than Usual Care?

Proof of Concept Results From a Markov Model

Patricia M Herman, ND, PhD; Tara A Lavelle, PhD; Melony E Sorbero, PhD; Eric L Hurwitz, DC, PhD; Ian D Coulter, PhD


Spine. 2019;44(20):1456-1464. 

In This Article

Abstract and Introduction


Study Design: Markov model.

Objective: Examine the 1-year effectiveness and cost-effectiveness (societal and payer perspectives) of adding nonpharmacologic interventions for chronic low back pain (CLBP) to usual care using a decision analytic model-based approach.

Summary of Background Data: Treatment guidelines now recommend many safe and effective nonpharmacologic interventions for CLBP. However, little is known regarding their effectiveness in subpopulations (e.g., high-impact chronic pain patients), nor about their cost-effectiveness.

Methods: The model included four health states: high-impact chronic pain (substantial activity limitations); no pain; and two others without activity limitations, but with higher (moderate-impact) or lower (low-impact) pain. We estimated intervention-specific transition probabilities for these health states using individual patient-level data from 10 large randomized trials covering 17 nonpharmacologic therapies. The model was run for nine 6-week cycles to approximate a 1-year time horizon. Quality-adjusted life-year weights were based on six-dimensional health state short form scores; healthcare costs were based on 2003 to 2015 Medical Expenditure Panel Survey data; and lost productivity costs used in the societal perspective were based on reported absenteeism. Results were generated for two target populations: (1) a typical baseline mix of patients with CLBP (25% low-impact, 35% moderate-impact, and 40% high-impact chronic pain) and (2) high-impact chronic pain patients.

Results: From the societal perspective, all but two of the therapies were cost effective (<$50,000/quality-adjusted life-year) for a typical patient mix and most were cost saving. From the payer perspective fewer were cost saving, but the same number was cost-effective. Assuming all patients in the model have high-impact chronic pain increases the effectiveness and cost-effectiveness of most, but not all, therapies indicating that substantial benefits are possible in this subpopulation.

Conclusion: Modeling leverages the evidence produced from clinical trials to provide more information than is available in the published studies. We recommend modeling for all existing studies of nonpharmacologic interventions for CLBP.

Level of Evidence: 4


Many safe and effective nonpharmacologic interventions for chronic low back pain (CLBP) are recommended in treatment guidelines.[1–4] Recommended interventions include therapies such as acupuncture, mindfulness-based stress reduction, and yoga. Although these nonpharmacologic therapies have been available to the public for years, many are not part of the healthcare system, prescribed by physicians, nor generally covered by third-party payer plans. In addition, although all recommended therapies are effective, it is unknown whether some are more effective than others, especially for certain patient groups, and their impact on costs.

The effectiveness of each recommended nonpharmacologic therapy is supported by systematic reviews and meta-analyses over dozens of studies.[3,5–7] In a few cases, these studies directly compare two or more of the recommended therapies, but usually each therapy is studied as adjunct to usual care and compared to usual care alone. Network meta-analysis (aka multiple or mixed treatment comparisons) is one technique that goes beyond traditional pairwise meta-analysis to allow comparisons and ranking between therapies that were not directly compared in clinical trials.[8–11] However, the application of this technique requires that each therapy be directly compared to at least one comparator also used in another trial, and thus allowing indirect comparisons. Because of its common use, the logical common comparator for most therapies is usual care. However, the substantial variation across studies in what constitutes usual care and the lack of other common comparators lowers the reliability of network meta-analysis results.[8]

Decision analytic modeling is another method to synthesize evidence and compare across therapies.[12,13] Modeling allows consistent inputs to be used across interventions, the simulation of experiments such as head-to-head trials, and the inclusion of economic outcomes even when they were not included in the original studies.[12,14]

Because each therapy is studied as an addition to usual care, its likely incremental effects in other healthcare settings can be estimated by comparing to the version of usual care used in the trial.

The 2016 US National Pain Strategy[15] placed a focus on those with high-impact chronic pain defined as that "associated with substantial restriction of participation in work, social, and self-care activities for 6 months or more."[15] , p. 11 Studies have used different algorithms to identify those with high-impact chronic pain,[16–21] and to demonstrate their significantly higher healthcare costs, lower quality of life, worse mood, and increased absenteeism.[18–27] Nevertheless, studies of interventions for chronic pain still report results based on full samples (e.g., average change in symptoms or percent of patients who improve). To better understand chronic pain, better compare study results, and better target interventions, we need to move beyond simple duration of pain definitions (e.g., 3+ months) to identify meaningful chronic pain subtypes. Modeling can then be used to balance the baseline subtype mix of patients across studies and estimate differential treatment effects by subtype (e.g., chronic pain impact level).

Although several economic evaluations of nonpharmacologic therapies for CLBP have been published,[28–34] because economic outcomes are not generalizable across settings,[35] each study's results can only provide useful information about cost impacts in its country and setting. Modeling is used to adjust inputs and assumptions to adapt study results to other settings.[36]

This study presents the results of a proof-of-concept modeling effort that used individual patient-level data from 10 randomized studies and the categorization of these study's participants by chronic pain impact level to estimate the effectiveness and cost-effectiveness of adding recommended nonpharmacologic therapies for CLBP to usual care.