Study characteristics
Key trial characteristics are provided in Supplementary Table S1, with more detailed descriptions in Supplementary Box S2. Of 27 included trials, 12 were long term,11,13,14,17,23,25,27,29,30,33,35,36 two were intermediate term,24,34 and 13 were short term.12,15,16,18–22,26,28,31,32,37 All but two short-term crossover trials employed parallel designs.19,32
Nineteen trials enrolled participants with OA,11,13,15,16,20–22,24–31,33–35,37 seven enrolled participants with chronic low back pain,12,14,17–19,32,36 and one enrolled a mixed population.23 Across all included trials, participants were 57.5% female, and had a mean age of 57.3 years. Most trials (n = 24/27) included both opioid-naïve and opioid-experienced participants (often with limitations on maximum dosage or frequency of use), while two did not specify,12,22 and one required opioid experience.19 Oxycodone was used in 13 of 27 trials (48%).11,14,16,20,25–27,31–34,36,37 Other opioids included buprenorphine,13,19,28 oxymorphone,21,26 hydromorphone,30,35 transdermal fentanyl,24,29 morphine,15,18 and the remaining used either ADL5747 or ADL5859 (δ-opioid receptor agonists),31 cebranopadol,17 codeine/acetaminophen,22 propoxyphene,12 or ‘general opioid therapy’,23 which meant the opioid was selected according to prescriber/participant preferences. Most trials (n = 24, 89%) compared with placebo. The remaining three compared with a) the prescriber’s choice of a variety of non-opioid medications (simultaneously available to those in the opioid arm);23 b) tolterodine (a potentially side effect-producing ‘active placebo’);32 and c) acetaminophen.22 Characteristics of the excluded EERW trials included in the sensitivity analysis are in Supplementary Box S3.
Risk of bias
Risk of bias for all included trials is summarised in Figure 2 and risk of bias for individual analyses is provided in Supplementary Figure S1. Detailed risk of bias assessments can be found in Supplementary Box S4. A common issue across included trials was unclear reporting, specifically regarding the blinding of outcome assessors.
Visual inspection of the funnel plot for the responder analysis (Figure 3) suggests smaller trials that are less favourable to opioids are missing, indicating probable publication bias in favour of opioids. Similar opioid-favouring publication bias is also evident in the assessment of pain scores (see Supplementary Figure S2).
Analgesic efficacy over time
Primary outcome: number of responders
The analgesic efficacy of opioids compared with control differed significantly between short-, intermediate-, and long-term trials (χ² = 6.04, degrees of freedom [df] = 2, P = 0.05). Opioids displayed superior pain relief only in short-term trials, with long-term trials trending towards fewer of the participants reporting meaningful pain relief in the opioid arm (Table 1 and Figure 4).
Table 1. Summary of findings for primary and secondary outcomes by treatment duration. Population: patients with OA and chronic low back pain; intervention: opioid; comparison: placebo or opioid-minimised pain management strategies The sensitivity analysis, comparing the excluded EERW trials with the study’s included trials, was limited to nine trials with ≥12 weeks duration as only one EERW trial had a shorter duration. These EERW trials had an average opioid titration period of 5.6 weeks before randomisation to opioid withdrawal, and clinically significant pain relief was significantly more likely in the opioid group in EERW trials (risk ratio 1.29, 95% CI = 1.22 to 1.37), versus traditional RCTs (χ² = 8.76, df = 1, P = 0.003). A forest plot of this analysis is provided in Figure 5.
Secondary outcome: pain on a 100-point scale
Although there was a trend towards diminishing opioid benefit with time (Table 1 and Supplementary Figure S3), that trend was not statistically significant (χ² = 3.01, df = 2, P = 0.22), in part because differences in pain between groups were small at all timepoints. For all treatment durations, 95% CIs ruled out a mean difference of ≥8 on a 100-point scale.