Input From Stakeholders

This topic was selected for review based on a nomination from the Centers for Medicare and Medicaid Services (CMS). The initial Key Questions for this Technology Assessment were developed with input from CMS staff. The Key Questions and scope were further developed with input from a group of stakeholders (Key Informants) convened for this report to provide diverse stakeholder perspectives and content and methodological expertise. The Key Informants consisted of experts in internal medicine, health services research, pain medicine, radiology, neurology, occupational medicine, and physical medicine and rehabilitation, as well as those representing the patient perspective. Key Informants disclosed financial and other conflicts of interest prior to participation. The AHRQ Task Order Officer and the investigators reviewed the disclosures and determined that the Key Informants had no conflicts of interest that precluded participation. A topic refinement document was then posted for public comment from December 17, 2013, through January 17, 2014 with the Key Questions and inclusion criteria. Based on public comments, we further revised the scope. The protocol for this Technology Assessment was finalized prior to initiation of the review, and was posted on the AHRQ Web site.

Literature Search Strategy

A research librarian conducted searches in Ovid MEDLINE, the Cochrane Central Register of Controlled Trials, the Cochrane Database of Systematic Reviews, and the National Guideline Clearinghouse from 2008 through October, 2014 (see Appendix A for full search strategy). We restricted search start dates to January 2008, as there are multiple recent systematic evidence reviews directly addressing the Key Questions in the current review, including a good-quality review conducted by the same investigators of the current review that was commissioned by the American Pain Society (APS) and conducted searches through July 2008.¹⁷ The APS review included all of the interventions addressed in the current review. We used the APS review and other systematic reviews³⁷ to identify studies published prior to 2008.

We also hand searched the reference lists of relevant studies and searched for unpublished studies in ClinicalTrials.gov. We did not solicit Scientific Information Packets for published and unpublished studies because the corticosteroid and local anesthetic drugs examined in this review are generic and the injections do not involve use of proprietary devices.

Literature searches will be updated while the draft report is posted for public comment and undergoing peer review to identify any new publications. Literature identified during the update search will be assessed by following the same process of dual review as all other studies considered for inclusion in the report. If any pertinent new literature is identified for inclusion in the report, it will be incorporated before the final submission of the report.

Study Selection

We developed criteria for inclusion and exclusion of articles based on the Key Questions and the populations, interventions, comparators, outcomes, timing, setting approach (Appendix B), in accordance with the AHRQ Methods Guide.³⁸ Articles were selected for full-text review if they were about epidural injections, facet joint injections, therapeutic medial branch injections, or sacroiliac injections with corticosteroids for radicular low back pain, spinal stenosis, or nonradicular low back pain, were relevant to a Key Question, and met the predefined inclusion criteria as described below. We excluded studies published only as conference abstracts, restricted inclusion to English-language articles, and excluded studies of nonhuman subjects. Studies had to report original data to be included.

To ensure accuracy, all excluded abstracts were dual reviewed. All citations deemed appropriate for inclusion by at least one of the reviewers were retrieved. Each full-text article was independently reviewed for eligibility for final inclusion by two team members. Discrepancies were resolved through discussion and consensus. A list of the included studies is available in Appendix C; excluded studies are shown Appendix D, with primary reasons for exclusion. Members of the review team were not involved in inclusion decisions for studies that they were authors on.

For epidural injections, we selected studies of adults undergoing epidural corticosteroid injections with radicular low back pain, spinal stenosis, nonradicular low back pain, or chronic postsurgical pain. We defined radiculopathy as presence of leg pain (typically worse than back pain), with or without sensory deficits or weakness, in a nerve root distribution. A number of studies used the term “sciatica,” which we classified as radiculopathy. We included epidural injections performed via any approach, including the transforaminal, interlaminar, or caudal techniques. We also included studies of injections performed via the transforaminal approach that targeted the affected nerve root but did not necessarily enter the epidural space (“periradicular” injections).

For facet joint and sacroiliac injections, we selected studies of adults undergoing corticosteroid injections in or around the facet or sacroiliac joints for nonradicular low back pain presumed to originate from the facet joints (facet injections) or sacroiliac joints (sacroiliac injections). We included injections into the joint (intra-articular) or around the joint (extra-articular [peri-articular]), as well as therapeutic medial branch blocks (injections at the site of the medial branch of the dorsal ramus nerves innervating the facet joints).

We excluded studies of patients younger than 18 years of age, pregnant women, and patients with back pain due to fracture, high-impact trauma, cancer, infection, or spondyloarthropathy. We excluded studies of noninjection ablative therapies such as intradiscal electrothermal therapy or radiofrequency denervation, other noninjection therapies such as nucleoplasty, and studies that involved injection of noncorticosteroid medications (such as ozone, antitumor necrosis factor medications, methylene blue, or clonidine) unless they were compared to a corticosteroid injection. Studies on the diagnostic accuracy of diagnostic blocks was outside the scope of the review, but we evaluated how use of diagnostic blocks to select patients impacted estimates of effectiveness.

We included studies of patients with symptoms of any duration prior to enrollment. We included studies that compared the injections of interest versus epidural nonsteroid injections, soft tissue injections, no injection, surgery, or noninjection, nonsurgical therapies. We classified epidural injections with local anesthetics or saline, soft tissue injections with local anesthetics or saline, and no injections as “placebo” interventions to distinguish them from “active” interventions such as epidural injections of other medications, other interventional procedures, surgery, or nonsurgical, noninterventional therapies. We also included studies that compared different injection techniques and corticosteroid doses.

Outcomes were pain, function, quality of life, opioid use, subsequent surgery, health care utilization, and harms, including bleeding, infection, neurological events, and systemic complications such as weight gain, diabetes, osteoporosis, and other endocrinological effects. We included outcomes measured 1 week or later after the injection.

We included randomized trials for all Key Questions. For harms, we also included large (sample size >1000 patients) treatment series of patients who underwent the injections of interest. We excluded case series and case reports. We reviewed reference lists of systematic reviews for potentially relevant references.

Data Extraction

We extracted the following information from included studies into evidence tables using Excel spreadsheets: study design, year, setting, country, sample size, inclusion and exclusion criteria (including age, sex, race, back pain condition, duration of pain, baseline pain, baseline function, prior therapies, imaging and diagnostic findings, and psychosocial factors), intervention characteristics (including type and dose of corticosteroid and local anesthetic, volume of injectate, number and frequency of injections, levels injected, injection approach, use of imaging guidance, and experience of the person performing the injection), characteristics of the control intervention, and results.

For studies of interventions, we calculated relative risks and associated 95 percent confidence intervals (CI) based on the information provided (sample sizes and incidence of outcomes of interest in each intervention group). We noted discrepancies between calculated and reported results when present.

Data extraction for each study was performed by two investigators. The first investigator extracted the data, and the second investigator independently reviewed the extracted data for accuracy and completeness.

Assessing Quality

We assessed quality (risk of bias) for each study using predefined criteria. We used the term “quality” rather than the alternate term “risk of bias;” both refer to internal validity. Randomized trials were evaluated with criteria and methods developed by the Cochrane Back Review Group.³⁹ These criteria were applied in conjunction with the approach recommended in the chapter, Assessing the Risk of Bias of Individual Studies When Comparing Medical Interventions,³⁸ in the AHRQ Methods Guide. Two investigators independently assessed the quality of each study. Discrepancies were resolved through discussion and consensus. Members of the review team who were authors on included studies were not involved in quality rating of those studies.

Individual studies were rated as having “poor,” “fair,” or “good” quality. We rated the quality of each randomized trial based on the methods used for randomization, allocation concealment, and blinding; the similarity of compared groups at baseline; whether attrition was adequately reported and acceptable; similarity in use of cointerventions; compliance to allocated treatments; the use of intent-to-treat analysis; and avoidance of selective outcomes reporting.^{39, 40}

Studies rated “good quality” are considered to have low risk of bias and their results are likely to be valid. Studies rated “fair quality” have some methodological shortcomings, but no flaw or combination of flaws judged likely to cause major bias. In some cases, the article did not report important information, making it difficult to assess its methods or potential limitations. The “fair-quality” (moderate risk of bias) category is broad and studies with this rating vary in their strengths and weaknesses; the results of some studies assessed to have moderate risk of bias are likely to be valid, while others may be only possibly valid. Studies rated “poor quality” (high risk of bias) have significant flaws that may invalidate the results. They have a serious or “fatal” flaw or combination of flaws in design, analysis, or reporting; large amounts of missing information (including publication of only preliminary results in a subgroup of patients randomized); or serious discrepancies in reporting. The results of these studies are at least as likely to reflect flaws in the study design as the differences between the compared interventions. We did not exclude poor quality studies a priori, but they were considered the least reliable when synthesizing the evidence, particularly when discrepancies between studies were present.

Treatment series of patients undergoing injection therapies were not formally rated because they already are known to have serious limitations due to the lack of a control group of patients who did not undergo injections.

Assessing Research Applicability

We recorded factors important for understanding the applicability of studies, such as whether the publication adequately described the study sample, the country in which the study was conducted, the characteristics of the patient sample (e.g., age, sex, race, type of back pain, imaging findings, duration or severity of pain, medical comorbidities, and psychosocial factors), the characteristics of the interventions used (e.g., specific corticosteroid, dose, technique, number or frequency of injections, and use of imaging guidance), and the magnitude of effects on clinical outcomes.⁴¹ We also recorded the funding source and role of the sponsor. We did not assign a rating of applicability (such as high or low) because applicability may differ based on the user of the report.

For interpreting the clinical importance of mean changes in outcome scores, we defined a minimum clinically important difference as an improvement in 15 points on a 0 to 100 pain scale, 10 points on the Oswestry Disability Index (ODI), and 5 points on the Roland Morris Disability Questionnaire (RDQ).⁴² These thresholds were recommended in a report from a panel of 36 experts in the low back pain field following a review of the evidence.⁴²

Evidence Synthesis and Rating the Body of Evidence

We constructed evidence tables summarizing study characteristics, results, and quality ratings for all included studies. We summarized evidence for each Key Question qualitatively using a hierarchy-of-evidence approach, where the best evidence was the focus of our synthesis for each Key Question.

We conducted meta-analyses to summarize data and obtain more precise estimates for outcomes and comparisons for which studies were homogeneous enough to provide a meaningful combined estimate.⁴³ Comparisons for which evidence was suitable for pooling were epidural corticosteroid injection versus placebo (epidural local anesthetic injection, epidural saline injection, soft tissue local anesthetic injection, soft tissue saline injection, needling with no injection, or no injection) for radiculopathy, epidural corticosteroid injection versus placebo for spinal stenosis, and transforaminal versus interlaminar epidural injections with corticosteroid for radiculopathy.

Outcomes were extracted and stratified by duration of followup as immediate (≤2 weeks), short term (2 weeks to ≤3 months), intermediate term (3 months to <1 year), and long term (>1 year). For each category, we used the longest duration data available. We analyzed two continuous outcomes: pain and function. For pain, we used pain as measured on a 0 to 10 or 0 to 100 visual analogue or numerical rating scale and converted all scores to 0 (no pain) to 100 (worst possible pain). We used data for leg pain when available; if leg pain was not specifically reported we used overall pain (leg pain is typically worse than back pain in patients with radiculopathy). For studies that assessed function using more than one measure, we prioritized the outcome used for pooling in the following order: the ODI (range 0 to 100), the RDQ (range 0 to 24), and other function scales. In the primary analyses, we combined weighted mean difference (WMD) for pain and standardized mean difference (SMD) for function. The mean difference was calculated using the change between the followup and baseline scores. We also conducted sensitivity analysis based on differences in scores at followup and estimates from analysis of covariance or other results that adjusted for other covariates if available; results were similar to the primary analyses and not reported further. When the standard deviation was missing, we imputed the missing value using the mean standard deviation from other studies in that analysis. For binary outcomes, we combined relative risks (RR) for pain and function “success” (e.g., ≥50% improvement in pain scores or ODI, or as otherwise defined in the trials), composite or overall measures of success (e.g., ≥50% improvement in pain and ≥50% function, or as otherwise defined in the trials), and rates of subsequent surgery. In addition, we pooled placebo response rates for the binary pain and function and mean difference in change scores for pain and function in the placebo group (as a measure of placebo response), stratified by the comparator used.

The studies were combined using the Dersimonian-Laird random effects method. We assessed the presence of statistical heterogeneity among the studies by using the standard Cochran’s chi-square test, and the magnitude of heterogeneity by using the I² statistic.⁴⁴ When statistical heterogeneity was present, we performed sensitivity analyses by conducting meta-analysis using the profile likelihood method.⁴⁵ For the primary analyses of epidural injections versus placebo, we pooled across approaches (transforaminal, interlaminar, or caudal), but also stratified the results by approach. We performed sensitivity analyses by excluding poor quality studies and outlier studies. When the number of studies was relatively large (8 to 10),⁴³ we performed additional subgroup analyses and meta-regression based on the specific corticosteroid used, the corticosteroid dose (converted into prednisolone equivalents, for transforaminal ≥50 mg vs. <50 mg; for interlaminar and caudal, ≥100 mg vs. <100 mg), the local anesthetic used, the specific comparator (epidural injection with local anesthetic, epidural injection with saline, soft tissue injection with local anesthetic, soft tissue injection with saline, needling without injection, or no injection), the volume of injected (for transforaminal, >3 ml vs. ≤3 ml; for interlaminar, ≥10 ml vs. <10 ml; for caudal, ≥20 ml vs. <20 ml), the duration of symptoms (restricted to acute [symptoms ≤4 weeks]), whether imaging correlation was required for patient enrollment, whether enrollment was restricted to patients with herniated disc on imaging, use of fluoroscopic guidance, whether the intervention was limited to a single injection, whether patients with prior surgery were excluded, overall quality rating (good, fair, poor), whether the person performing the injection was blinded, and blinding of outcomes assessors and patients. For transforaminal injections, we also stratified studies according to whether the injection clearly entered the epidural space or targeted the nerve root (“periradicular”) without clearly entering the epidural space. Similar analyses were performed for the analysis of transforaminal versus interlaminar injections when data allowed. A funnel plot was created and the Egger test performed for primary analyses across approaches (transforaminal, interlaminar, or caudal) with 10 studies to assess for small study effects.⁴⁶ All analyses were conducted using Stata/IC 13.0 (StataCorp LP, College Station, TX).

We assessed the strength of evidence (SOE) for each Key Question and outcome using the approach described in the AHRQ Methods Guide,³⁸ based on the overall quality of each body of evidence, based on the risk of bias (graded low, moderate, or high); the consistency of results across studies (graded consistent, inconsistent, or unable to determine when only one study was available); the directness of the evidence linking the intervention and health outcomes (graded direct or indirect); and the precision of the estimate of effect, based on the number and size of studies and confidence intervals for the estimates (graded precise or imprecise).

We graded the SOE for each Key Question using the four categories recommended in the AHRQ Methods Guide.³⁸ A “high” grade indicates high confidence that the evidence reflects the true effect and that further research is very unlikely to change our confidence in the estimate of effect. A “moderate” grade indicates moderate confidence that the evidence reflects the true effect and further research may change our confidence in the estimate of effect and may change the estimate. A “low” grade indicates low confidence that the evidence reflects the true effect and further research is likely to change the confidence in the estimate of effect and is likely to change the estimate. An “insufficient” grade indicates evidence either is unavailable or is too limited to permit any conclusion, due to the availability of only poor quality studies, extreme inconsistency, or extreme imprecision.

Peer Review and Public Commentary

Experts in low back pain and injection therapies, as well as individuals representing important stakeholder groups, have been invited to provide external peer review of this Technology Assessment. The AHRQ Task Order Officer will also provide comments and editorial review. To obtain public comment, the draft report was posted on the AHRQ Web site for 2 weeks. A disposition of comments report detailing the authors' responses to the peer and public review comments will be made available after AHRQ posts the final report on the public Web site.

Key Findings and Strength of Evidence Summary

The key findings of this review are summarized in the summary of evidence table (Table A) below.

Table A

Summary of evidence.

Strengths of our review are inclusion of additional trials compared with prior reviews, evaluation of epidural corticosteroid injections for back pain conditions other than radiculopathy, evaluation of continuous as well as dichotomous outcomes, evaluation of outcomes at defined time points, evaluation of the effectiveness of epidural corticosteroids versus other active interventions, conduction of additional analyses to evaluate effects of methodological limitations, patient characteristics, and injection techniques on findings.

Evidence was most robust for epidural corticosteroid injections in patients with radiculopathy. In trials of epidural corticosteroid injections versus placebo interventions, the only statistically significant effects were on mean improvement in pain at immediate-term (5 days to ≤2 week) followup (WMD −7.55 on a 0 to 100 scale, 95% CI −11.4 to −3.74), mean improvement in function at immediate-term (SMD −0.33, 95% CI −0.56 to −0.09) followup, and risk of surgery at short-term (>2 weeks to ≤3 months) followup (RR 0.62, 95% CI 0.41 to 0.92). However, the magnitude of the effect on pain was small (WMD −7.55 on 0 to 100 scale) and did not meet our predefined minimum clinically important differences of 15 points.⁴² The effect on immediate-term function was of only statistically significant when an outlier trial⁴⁷ was excluded. Differences were also small in the nonoutlier trials (5.1 and 7.6 points on the 0 to 100 ODI)^{48, 49} and 1.3 points on the 0 to 24 RDQ) and did not meet predefined thresholds for minimum clinically important differences (10 points for the ODI and 5 points for the RDQ).⁴² Differences were not present for either outcome at longer-term followup. There were also no differences at any time point between epidural corticosteroid injections and placebo interventions in likelihood of experiencing a successful pain, function, or composite outcome; or likelihood of undergoing surgery. Direct evidence from randomized trials on effects of performing epidural corticosteroid injections for radiculopathy using different approaches, different corticosteroids, or different doses was limited, but indicated no clear effects on outcomes. There were also no clear differential effects of the epidural approach used, different corticosteroids, different doses, use of imaging correlation, restriction to patients with herniated disc, duration of symptoms, or exclusion of patients with prior surgery, primarily based on meta-regression and subgroup analyses.

Although comparator interventions such as epidural local anesthetic injection, epidural saline injection, soft tissue injections, and no injection have traditionally been considered placebo interventions, it is possible that they may have some therapeutic effects.³⁵ However, placebo response rates were high in trials of epidural corticosteroid injections regardless of the comparator used and there were no clear differences in estimates of effectiveness based on the specific comparator, suggesting that observed improvements represent the natural history of radiculopathy or a general placebo response.

Trials of epidural corticosteroid injections for radiculopathy versus nonplacebo interventions did not clearly demonstrate effectiveness, but were limited by small numbers of trials for specific comparisons and methodological limitations, resulting in low or insufficient strength of evidence ratings.

Evidence was limited for epidural corticosteroid injections versus placebo interventions for spinal stenosis or nonradicular back pain, but showed no differences in outcomes related to pain or function. The evidence on epidural corticosteroid injections for spinal stenosis included a recent, well-conducted multicenter trial that was also the largest trial (n=386) to date in this population.⁵⁰ Although epidural injections could be performed by either the interlaminar or transforaminal approach in this trial, there was also no evidence of an effect when results were stratified according to the approach used. Another potential issue in interpreting this trial is that the corticosteroids and doses varied, which could have attenuated effects if certain corticosteroids or lower doses are associated with decreased effectiveness. For chronic postsurgical pain, evidence was very limited. No trial compared epidural corticosteroid injections versus placebo, and trials of epidural corticosteroid injections versus various other interventions found no clear differences.

Evidence was also limited for facet joint corticosteroid injections versus placebo interventions. Studies found no clear differences between various facet joint corticosteroid injections (intra-articular, extra-articular [peri-capsular], or medial branch) and placebo interventions. Although one trial found an intra-articular corticosteroid injection associated with better outcomes than a saline injection at 6 months, results are difficult to interpret because there were no differences at 1 month, the corticosteroid group received more cointerventions, and there was no difference in the likelihood of sustained improvement (improvement at 6 months in patients with improvement at 1 month).⁵¹ Trials of facet joint injections versus radiofrequency denervation were difficult to interpret because they reported inconsistent outcomes, evaluation of different types of injections (intra-articular or medial branch corticosteroid injection), and differential use of diagnostic blocks to select patients, depending on which intervention they were randomized to.^{52, 53}

There was insufficient evidence from one very small (n=24) trial to determine effects of peri-articular sacroiliac joint corticosteroid injections versus a placebo (local anesthetic) injection.⁵³

Methods for assessing harms in randomized trials were generally not well reported and data sparsely reported, but evidence from trials and observational studies were consistent in suggesting a low risk of serious harms following epidural injections such as neurological complications or infection. However, cases of serious neurological complications have been reported following lumbar epidural injections, and there was a recent outbreak of serious fungal infections due to contaminated methylprednisolone injections.^{17, 54, 55}

Findings in Relationship to What is Already Known

Our findings were consistent with a previous qualitative review¹⁷ conducted by our team and funded by the APS that found fair evidence that epidural corticosteroid injections for radiculopathy are more effective than placebo interventions for short-term symptom relief, but not for long-term symptom relief, and limited evidence of ineffectiveness for spinal stenosis and nonradicular low back pain. Unlike the current review, which found that short-term effects on pain did not meet the threshold for minimum clinically important differences, the APS review classified the magnitude of short-term benefit for pain relief as “moderate” (equivalent to a 10- to 20-point difference on a 100-point pain scale). However, it noted inconsistency between trials, did not perform meta-analysis, and based some conclusions on prior reviews with small numbers of studies. Our findings were also consistent with more recent qualitative and quantitative systematic reviews, despite variability in the studies included and methods used for data synthesis and meta-analysis.^{20, 21, 23, 37, 56, 57} Our review was also concordant with other reviews in finding limited evidence that lumbar epidural corticosteroid injections are associated with a low risk of serious harms.^{17, 20, 54} Like a prior review, we also found no evidence on effectiveness of multiple versus single injections.⁵⁸ Although one systematic review found an association between volume differences (higher volume relative to control interventions) and effectiveness of epidural corticosteroid injections, results may have been confounded by differences in epidural approaches (e.g., caudal injections typically used higher volumes) and inclusion of comparators that did not involve epidural injections (e.g., soft tissue injections or noninjection therapies).⁵⁹ We were unable to determine effects of differences in injectate volume on effectiveness, as only three trials that compared epidural corticosteroid injections versus epidural local anesthetic or saline injections reported a volume difference.^{47, 60, 61}

With regard to effects of control interventions on effect estimates, our findings were similar to a recent systematic review that found limited direct evidence showing no differences between epidural nonsteroid and nonepidural injections.³⁵ Although the prior systematic review found some evidence that epidural nonsteroid injections might be more effective than nonepidural injections, its conclusions were based on indirect comparisons that were highly discrepant with direct comparisons. We did not perform indirect comparisons, as the presence of such discrepancies may result in misleading findings.⁶² Although the prior APS review found some evidence that effects of epidural corticosteroid injections were more likely to be positive in studies that used epidural nonsteroid injection controls than in studies that used nonepidural injection controls, these findings were based on a qualitative evaluation based on counts of positive and negative studies.¹⁷

Our findings of limited evidence on facet joint corticosteroid injections versus placebo interventions, without clear demonstration of beneficial effects, were also consistent with prior systematic reviews.^{17, 63, 64}

Some systematic reviews reported more positive conclusions regarding the effectiveness of epidural corticosteroid and facet joint injections.^{24–26, 65–67} Differences between the methods used in these reviews and ours that may explain the discrepant findings included reliance on qualitative synthesis, inclusion of observational studies, categorization of improvement from baseline following an epidural corticosteroid injection as demonstrating effectiveness (even when there was no difference versus a placebo intervention), and failure to consider inconsistency between trials.

Applicability

Some issues could impact the applicability of our findings. Results are most applicable to patients with chronic back pain, as few trials enrolled patients with acute or subacute symptoms. Although studies were typically performed in the United States or Europe in specialty settings, they varied in how the injections were performed, including the use of imaging guidance, the specific epidural or facet injection technique used, the methods used to select patients (e.g., use of imaging, the imaging findings required for inclusion, or the use of diagnostic blocks), the types and doses of corticosteroid used, and the number and frequency of injections. Although we found no clear evidence of an association between these factors and estimates of treatment effectiveness, direct evidence from head-to-head trials on these factors was limited. The effectiveness of injection therapies is also likely to depend in part on the skill and experience of the person performing the injection, but we were unable to determine effects of provider experience on treatment effects, as studies did not report this information or reported it in a nonstandardized manner. Because most trials excluded patients with prior lumbar surgery, results may not be applicable to this patient population. The applicability of findings to patients with important medical and psychiatric comorbidities was also uncertain, and there was insufficient evidence to determine how effectiveness might vary based on the receipt of concomitant interventions such as supervised exercise therapy or cognitive behavioral therapy. Trials also differed in methods used to select patients for inclusion. For example, trials of radiculopathy and spinal stenosis differed in the clinical symptoms required for enrolment as well as in whether concordant imaging findings were required, and trials of presumed facet joint pain varied in whether a positive response to diagnostic blocks were required as well as methods for performing blocks (e.g., single or double block).

In order to facilitate interpretation of findings, we stratified outcomes by followup duration. We also compared observed effects on continuous outcomes to previously proposed thresholds for minimum clinically important changes. Although immediate-term effects of epidural corticosteroid injections versus placebo interventions on improvement in pain scores were statistically significant, they did not meet the predefined threshold for a minimum clinically important difference.⁴² We also evaluated effects of injection therapies on the likelihood of experiencing a clinically meaningful outcome, which might be more clinically interpretable than mean effects on continuous scales.⁶⁸ Although trials varied in how they defined clinically meaningful outcomes related to pain, function, or overall success, analyses were consistent in showing no effects at different time points.

Implications for Clinical and Policy Decisionmaking

Our review has implications for clinical and policy decisionmaking. Epidural corticosteroid injections are the most commonly used interventional procedure for low back pain, but evidence indicates that benefits are limited to pain relief and reduced risk of surgery shortly after the procedure in patients with radiculopathy, without long-term benefits on these outcomes and no effect on functional outcomes. Some clinical practice guidelines recommend epidural corticosteroid injections for short-term benefits in patients with persistent radicular symptoms, particularly for patients who are not candidates for or interested in undergoing surgery.²⁸ Factors that may influence decisions about performing epidural corticosteroid injections include how highly patients value short-term symptom improvement of small magnitude, preferences regarding alternative therapies (including surgery), the severity of symptoms, and costs and other burdens. Decisions should also consider the risk of serious harms with epidural injections, which have been reported in case series and other uncontrolled observational studies.⁶⁹

Potential strategies to enhance the effectiveness of epidural injections would be to perform them using techniques shown to be more effective, or to selectively perform injections in patients more likely to benefit. However, our review found no clear evidence of greater benefits based on technical factors such as the specific epidural technique used, use of fluoroscopic guidance, the specific corticosteroid, the dose, or the number or frequency of injections. Evidence on patient factors was also too limited to identify subgroups of patients more likely to benefit.

Other findings of our review that have implications for clinical and policy decisionmaking included limited evidence of no effectiveness of epidural corticosteroid injections for spinal stenosis or nonradicular low back pain, or for facet joint corticosteroid injections for presumed facet joint pain. Although prior guidelines found insufficient evidence to develop recommendations on use of epidural corticosteroid injections for spinal stenosis and nonradicular low back pain,²⁸ the strength of evidence has improved, particularly for spinal stenosis,⁵⁰ suggesting that re-evaluation may be appropriate. Guidelines are inconsistent with regard to use of facet joint corticosteroid injections,^{28, 70, 71} but recent trials have not provided additional evidence to support effectiveness.

Limitations of the Comparative Effectiveness Review Process

An important limitation of our review is that substantial statistical heterogeneity was present in several pooled analyses. To address this, we used the Dersimonian-Laird random effects model to pool studies. The Dersimonian-Laird random effects model may result in confidence intervals that are too narrow when heterogeneity is present, particularly when the number of studies is small.⁴⁵ Therefore, we repeated analyses using the profile likelihood method, which resulted in similar findings. Regardless of the method used, meta-analyses based on small numbers of trials can underestimate statistical heterogeneity and must be interpreted with caution.⁴⁵ We also stratified trials according to the epidural technique used, and further explored heterogeneity by performing additional analyses stratified based on the comparator used, exclusion of poor-quality and outlier studies, and meta-regression (when sufficient numbers of studies were available) on use of blinding, patient selection methods, and methods used to perform the injections (e.g., the corticosteroid or local anesthetic used and the dose). Although statistical heterogeneity remained present in some analyses, with some unexplained outlier trials, results were generally robust in sensitivity and stratified analyses.

Another limitation of our review is that we used indirect comparisons to supplement limited direct evidence on the effects of technical and patient factors on estimates. Although findings based on indirect comparisons were generally consistent with available evidence from head-to-head trials (e.g., showing no clear effects of different corticosteroids, different epidural approaches, or different doses), results based on indirect comparisons should also be interpreted with caution.⁷²

We excluded non-English language articles and did not search for studies published only as abstracts. We only formally assessed for publication bias using statistical and graphical methods to assess for small sample effects when there were at least 10 studies, as research indicates that such methods can be misleading with smaller numbers of studies.⁴⁶ We found no evidence of small sample effects based on analyses of short-term improvement in pain, short-term improvement in function, or long-term risk of surgery. Finally, we restricted evidence on serious harms to randomized controlled trials and large cohorts of patients undergoing injections, in order to be able to estimate rates of events. Although serious neurological events with epidural corticosteroid injections have been reported in case series and other uncontrolled studies, it is not possible to estimate the rates of events from such data.⁶⁹

Limitations of the Evidence Base

An important limitation of the evidence base is the small number of trials available on epidural corticosteroid injections for conditions other than radiculopathy and the small number on effectiveness of facet joint or sacroiliac corticosteroid injections. The lack of evidence made it difficult to reach strong conclusions regarding the effectiveness of these interventions versus placebo interventions or to evaluate effects of methodological, technical, or patient factors on outcomes. Although more trials were available for epidural corticosteroid injections for radiculopathy, there were fewer trials when results were stratified by specific outcomes and time points, and subgroup analyses were limited by the variability in techniques used on a number of factors.

In addition to the small number of trials, the evidence base was limited by methodological limitations in the available studies. Only eight trials were rated good quality. Of the 92 included trials, only 27 trials reported blinding of the person performing the injection, 57 trials blinding of patients, and 47 trials blinding of outcome assessors. Conclusions were generally not impacted by exclusion of poor-quality trials or assessments based on blinding status, but would be stronger if more high-quality trials were available.

Other limitations include the relatively limited number of trials that directly compared different injection techniques, corticosteroids, doses, and comparators. No trials directly compared use of imaging guidance versus no guidance, use of a single injection versus multiple injections, or effects of different injectate volumes. Few trials reported how effectiveness varied according to patient characteristics such as age, sex, race, medical or psychological comorbidities, duration of symptoms, imaging findings, cause of low back pain, use of concomitant therapies, or other factors.

Research Gaps

Research gaps limit the full understanding of the comparative effectiveness of low back injections. For radiculopathy, additional research could help determine whether patient characteristics such as severity or duration of symptoms, presence of specific imaging findings, or presence of psychiatric comorbidities are associated with responsiveness to injections. If such characteristics are identified, future trials could be designed to focus on more specific target populations that might experience greater benefits. Trials are also needed to understand whether injections may be more effective when given in the context of a more comprehensive approach that includes the delivery of concomitant treatments such as supervised exercise therapy or cognitive behavioral therapy. Additional research would also help confirm whether there are differences in outcomes associated with different epidural injection approaches, corticosteroids, doses, use of imaging guidance, and number and frequency of injections. Ideally such studies would include a placebo intervention group to aid in interpretability of findings.

For spinal stenosis and nonradicular low back pain, evidence was limited but indicated that epidural corticosteroid injections are not effective compared to placebo interventions. Because spinal stenosis is usually degenerative and the etiology of nonradicular low back pain may be difficult to determine, the rationale for performing epidural corticosteroid injections may not be as strong as for radiculopathy due to herniated disc. Additional research on the effectiveness of epidural corticosteroid injections for these conditions may only be warranted if specific subgroups of patients who have more of an inflammatory component can be identified. Limited evidence also indicates that facet joint corticosteroid injections are not effective compared with placebo interventions. The lack of effectiveness could be due to the ineffectiveness of the procedure or suboptimal accuracy methods for identifying patients with facet joint pain.¹⁷ A randomized trial found that use of dual or single diagnostic facet joint blocks to select patients for radiofrequency denervation (an intervention not included in this report) was associated with lower rates of successful outcomes than selection of patients without a diagnostic block.⁷³ Therefore, additional research on accurate methods for identifying patients with facet joint pain is needed to inform the design of future intervention studies.