Effectiveness of Long-Term Opioid Therapy for Chronic Low Back Pain

Study Design and Participants

This retrospective cohort study included participants selected from the Pain Registry for Epidemiological, Clinical, and Interventional Studies and Innovation (PRECISION) from April 2016 through August 2022. Registry participants were screened and recruited from the 48 contiguous states and District of Columbia, primarily through social media advertising (eg, Facebook) that directed respondents to a Web landing page that included a link to the screening questionnaire (Appendix Figure 1). Both screenees and participants who were eventually enrolled in the registry were asked to use its digital research platform for electronic capture of self-reported data. However, screenees and registry participants also had the option of reporting data telephonically. Registry eligibility criteria included being 21 to 79 years of age at enrollment, having CLBP (≥3 months), having a physician (ie, allopathic or osteopathic physician) who provided usual care for CLBP, and sufficient English language proficiency to complete case report forms independently or with staff assistance. Screenees who reported being pregnant or residing in institutional facilities were excluded from the registry. The eligibility criteria were met in 4671 (55.0%) of the 8491 screening encounters during the study period and 1501 participants were enrolled. All study data were self-reported by registry participants at enrollment and quarterly encounters for up to 12 months. Data were not independently corroborated by physicians or medical records. Further information about PRECISION eligibility criteria and data collection is available at ClinicalTrials.gov.¹² The research was approved by the North Texas Regional Institutional Review Board (protocol 2015-169), and all participants provided informed consent before contributing data. This study is reported following the STROBE guidelines.¹³

• Download figure
• Open in new tab

Figure 1.

Participant flow through the study. Abbreviation: LTOT, long-term opioid therapy.

Measurement of Long-Term Opioid Therapy

Current opioid use was measured at registry enrollment by adapting the Minimum Dataset item recommended by NIH,¹⁴ which queries participants about their use of opioid painkillers for low back pain. The item provided prompts for specific medications, including generic and brand names of commonly used opioids. Current opioid use at registry enrollment may have entailed either new-onset use or unspecified longer-term use before enrollment. Long-term opioid therapy was defined as opioid use for >90 days,⁶ and established by reporting current opioid use at both registry enrollment and the 3-month encounter. Participants who did not report current opioid use at both registry enrollment and the 3-month encounter were considered opioid abstainers (ie, controls). Participants who reported discrepant current opioid use at these encounters were not eligible for inclusion in the study.

Participants also reported more comprehensive drug information at registry enrollment using open-ended items for name, dose, and daily frequency of administration for up to 12 medications for low back pain or other medical problems. The daily MME dosage for LTOT users was computed using data provided at registry enrollment for opioid class, dose, and daily frequency of administration, with conversion factors recommended by CDC.⁴

Propensity Score Matching

The LTOT users and abstainers were matched on propensity scores computed with a logistic regression model that included participant characteristics at registry enrollment pertaining to demography (age, gender, race), psychological factors (pain catastrophizing, pain self-efficacy), low back pain and related factors (duration of low back pain, presence of chronic widespread pain, work loss ≥1 month, receipt of disability or workers’ compensation benefits, law suits or legal claims, prior low back surgery, health-related quality of life), and clinical status (low back pain intensity, back-related disability, pain impact). These variables were measured using elements of the Minimum Dataset¹⁴ supplemented by the Pain Catastrophizing Scale,¹⁵ Pain Self-Efficacy Questionnaire,¹⁶ Patient-Reported Outcomes Measurement Information System (PROMIS),¹⁷ and other registry-specific instruments.¹² All factors entered into the model were dichotomous or categorical variables, except health-related quality of life, low back pain intensity, back-related disability, and pain impact. Long-term opioid therapy users and abstainers were matched within a caliper width of 0.001 to ensure that no standardized difference between treatment groups was >0.10, which may represent a threshold for a meaningful imbalance in a given covariate.¹⁸

Outcome Measures

Primary outcomes included low back pain intensity, back-related disability, and pain impact. A numerical rating scale from 0 to 10 measured average low back pain intensity within 7 days before each encounter. Back-related disability was measured with the Roland-Morris Disability Questionnaire, which yielded values that reflected the participant’s status on the encounter date and ranged from 0 (no disability) to 24 (greatest disability).¹⁹ Pain impact was measured using low back pain intensity and 8 items derived from PROMIS.^17,20 These included 4 items in each of 2 PROMIS scales involving physical function and pain interference with activities within 7 days before an encounter. Pain impact ranged from 8 to 50, with higher values representing worse outcomes. Secondary outcomes involved achievement of a minimally important change (MIC) in each primary outcome. These were defined as reductions of 1 point on the numerical rating scale for low back pain intensity,²¹ 2 points on the Roland-Morris Disability Questionnaire,²¹ and 7.5 points on the pain impact scale.²⁰ Outcome measures were collected at all completed quarterly encounters up to 12 months after registry enrollment. Participants who missed 2 consecutive quarterly encounters were withdrawn from the registry. However, available data for all withdrawn and in-progress participants before 12 months were retained and analyzed.

Statistical Analysis

Descriptive statistics were used to compare the baseline characteristics of participants in each treatment group at enrollment, including standardized differences to assess adequacy of propensity-score matching. The registry’s digital research platform and electronic data capture system precluded missing item responses on outcome measures at any completed encounter. However, participants may have reported ambiguous opioid names, doses, or daily frequencies of administration in open-ended fields, such as when they used pain-contingent (ie, “prn”) opioid dosing rather than fixed time-scheduled dosing.²² The resolution of ambiguous opioid names, doses, or daily frequencies of administration was generally performed by consensus among the investigators. This was facilitated by RxNorm, a naming system for generic and branded drugs developed by the National Library of Medicine to provide normalized names and unique identifiers for medicines,²³ which has been used in the development of the medications component of diagnostic decision support systems.²⁴ However, such resolution was not possible if participants reported prn dosing. In such cases, as in prior research involving opioid use among patients with chronic pain, multiple imputation was used to estimate missing drug data by incorporating key analytic variables and other variables potentially associated with the missing data.^25,26 Our imputation model for missing daily MME dosage included low back pain intensity, back-related disability, and pain impact at registry enrollment as key analytic variables, and age and gender as other variables potentially associated with the missing data. One hundred imputations were run to minimize statistical power falloff for small effect sizes.²⁷

In addition to overall analyses involving outcomes based on the LTOT groups established at 3 months using propensity-score matching, 4 other analyses were performed to assess continuity of opioid use and dose response. The first analysis measured the continuity of opioid use beyond the >90 days minimally needed to qualify for the LTOT group, and was based on the number of months of continuous opioid use, defined as 0, 3, 6, 9, or 12 months since enrollment. The second analysis measured dose response based on the total number of quarters during which opioids were used (not necessarily continuously used), ranging from 0 to 4 quarters over 12 months. The third analysis measured dose response based on daily MME dosages at registry enrollment. Daily MME dosages were classified as MMEs ≤ 30 (typical starting dosages for opioid-naïve patients), 30<MMEs < 50 (escalated opioid dosages for chronic pain), and MMEs ≥ 50 (dosages beyond which many patients may not experience benefit in pain or function).⁴ In each of these 3 analyses, analogous to intention-to-treat analyses in clinical trials, all participants were retained in their initial treatment group (LTOT or control) regardless of any subsequently reported treatment crossed over. Consequently, controls were classified as having no opioid exposure or dose in these 3 analyses. However, analogous to per-protocol analyses used in clinical trials to mitigate issues potentially pertaining to treatment crossover, the fourth analysis compared the subset of participants who were continuous opioid users versus continuous abstainers for all 12 months.

Generalized estimating equations (GEEs) were used to compare treatment groups on each primary outcome over 4 quarterly encounters (3, 6, 9, and 12 months) after LTOT and control group membership was established based on current opioid use and propensity-score matching. These analyses were performed using an autoregressive correlation matrix to compute maximum likelihood estimates for linear scale responses. To assess the possibility of misspecification of the GEE model or violation of its underlying assumptions,^28,29 sensitivity analyses were performed using linear mixed methods with an unstructured correlation matrix. Cox proportional hazards regression was used to measure treatment group differences in time to achieve a MIC for each primary outcome.

Statistical power was estimated with the GLIMMPSE software for repeated measures.³⁰ This was based on the primary low back pain intensity outcome because it is almost universally measured in CLBP trials. Forty-two high-quality trials of opioids versus placebo found a standardized difference of 0.69 favoring opioids on a 0 to 10 scale for low back pain intensity.¹⁰ The standard deviation of this scale was estimated to be 2.33.³¹ Assuming low back pain intensity base correlation = 0.6 with decay rate = 0.2 and inflation for 10% missing encounters over 12 months, 370 participants (185 in each treatment group) were needed for ≥0.90 statistical power. All data were managed and analyzed using the IBM SPSS Statistics Software (Version 29). Hypotheses were assessed at the 0.05 level of statistical significance, using 2-sided testing for all analyses (Appendix Table 1).