Epidural Injections for Spinal Pain: A Systematic Review and Meta-analysis Evaluating the “Control” Injections in Randomized Controlled Trials

SPINAL pain is a leading cause of disability in the industrialized world. The lifetime prevalence for low back pain ranges between 50 and 80%¹ ; for neck pain, the estimates are between 50 and 67%.^2,3  Compounding the high disease burden is the absence of any reliably effective treatment.

More than one third of back pain cases can be classified as predominantly “neuropathic.”⁴  The distinction between nociceptive and neuropathic spinal pain has significant treatment implications in that the latter may be more amenable to therapy. A cornerstone of conservative treatment for radiculopathy is epidural steroid injections (ESI), which have been used for more than 50 yr.⁵  In the United States, ESI are the most commonly performed intervention for pain.⁶ 

Despite their frequent use, the question of whether ESI afford long-term benefit is mired in controversy. Recent reviews demonstrate a glaring lack of consensus. Most experts concede that ESI provide at least short-term palliation in well-selected patients, but the results are divided as to whether they confer long-term benefit.^7–26  In one review that evaluated the effect physician specialty has on conclusions regarding efficacy, 15 of 23 systematic or evidence-based review articles concluded that ESI are effective, with those reviews performed by interventionalists being approximately three times more likely to be positive compared with reviews conducted by noninterventionalist physicians.²⁷ 

Challenges in evaluating ESI studies include disparities in selection criteria, injection parameters, and criteria for success. It is generally acknowledged that patients with shorter duration of symptoms, radicular symptomatology, lesser disease burden, and the absence of coexisting psychosocial pathology, fare better with therapeutic interventions.^9,16,28–31  But what is not commonly appreciated is the impact the “control” injection has on outcomes.

Two main types of “control” injections are used in ESI randomized controlled trials (RCTs): epidural saline or local anesthetic [epidural nonsteroid injection (ENSI)]; and intramuscular or ligamentous injections (nonepidural). In evaluating the literature, most experts fail to discern the differences, considering them as “equivalent placebos.” However, the potential benefit of corticosteroids for a chronic condition devoid of an inflammatory component is minimal. In addition, recent studies suggest that radiculopathy may also result from chemical irritation of nerve roots due to inflammatory cytokines released from herniated discs.^32,33  Hence, ENSI may provide significant pain relief by several mechanisms: diluting inflammatory cytokines; lysing scar tissue; enhancing blood flow to ischemic nerve roots³⁴ ; suppressing ectopic discharges from injured nerves³⁵ ; and “unwinding” central sensitization. However, few investigators have entertained this possibility.^{9,28,31,36,37}  If ENSI provide benefit, then the proportion of controlled studies evaluating ESI in which the results are positive should be less when the control group received ENSI than for nonepidural injections, because the former constitutes a comparative-effectiveness study. The purpose of this study is to examine whether epidural injections of noncorticosteroid mixtures constitute a treatment or true placebo in patients with spinal pain. This was done by comparing between-group differences for pain outcomes from RCTs in which the “control” injectate was administered epidurally (ENSI), with those in which it was injected into the soft tissue (nonepidural injection).

This quantitative systematic review and meta-analysis followed recent methodological guidelines.^38,39  A search of PubMed and EMBASE databases using the terms “epidural steroid,” “epidural injection,” “caudal,” “segmental nerve block,” “nerve root block,” and “transforaminal injection” was performed without publication date or language restriction in April 2009, and repeated in October 2012. Besides online language tools, in-person translation services were provided through the Johns Hopkins Hospital international services translation program. Search limitations included RCTs and adults older than 18 yr of age. Additional studies were identified through hand searches of ESI review reference lists. Figure 1 and table 1 show further details of the search strategy. Among 690 potentially eligible studies, 244 duplicate references were excluded leaving 446 studies for evaluation, with another 356 deemed ineligible after initial abstract screening. This left 90 articles for final review.

All authors performed study selection by consensus. Eligible studies included only RCTs with: (1) patients with back or neck pain with or without radiculopathy; (2) a treatment group receiving epidural injections of corticosteroids or another analgesic; (3) a control group receiving either an epidural injection devoid of treatment drug or a nonepidural injection; and (4) short-term outcome data up to 12 weeks after the initial injection (if the injection scheme was open-ended) or after the final injection in an injection series. On the basis of these criteria, 47 studies were further excluded. The remaining 43 studies comprised the systematic review. For inclusion in the meta-analysis, studies had to present numeric pain data including SD.

Data extraction was performed independently by two authors (Drs. Bicket and Gupta), and included patient characteristics, control and treatment injections, rating scores for pain, pain symptoms, and disability scores. Variables including number and percentage of patients, and mean with SD, were extracted, calculated from primary data, or estimated from figures. When not given, SD was calculated using standard errors (SE) and 95% CI.

Study risk of bias was assessed using a Cochrane risk of bias tool^‖ and secondarily using Jadad⁴⁰  methodological quality scale, whereas an ESI technical quality scale evaluated stringency of selection criteria (table 2). The ESI technical quality scale was developed by the investigators after a review of clinical studies evaluating factors associated with treatment outcomes for ESI and back pain in general.^{29,30,41–60}  The questions chosen were designed to identify those factors most likely to be associated with treatment response, and to address relevant methodological concerns not reflected in the methodological assessments (e.g., avoidance of cointerventions). This scale was then reviewed with slight modifications by two disinterested Pain Medicine Program Directors at nonstudy institutions, underwent test–retest reliability assessments (>95%) by three study investigators, and is consistent with other rating scales used to evaluate technical quality and clinical relevance for procedural interventions.⁶¹  Although its design suggests possible face, content, and construct validity,^62–64  the scale was not formally validated in a clinical trial. All bias and technical ratings were performed by two of three authors independently (Drs. Brown, Gupta, and Bicket). In the event of disparate ratings, a third author (Dr. Cohen) adjudicated the results. Low methodological quality studies had at least one high likelihood of bias Cochrane risk domain or fulfilled less than three Jadad criteria, whereas low technical quality studies scored less than 4 points on the ESI technical quality scale.

Categorical pain ratings were transformed into a dichotomous “positive response” variable, with “positive response,” “success,” “relief of pain,” and “50% or more reduction in pain” representing positive responses. Visual and numerical pain ratings were transformed into a continuous 0–10 rating scale and, when presented, analyzed by body site (global, leg, back). When global pain ratings were not available for aggregate analysis, leg pain ratings were used in their place.^{37,52,65–69}  Baseline and comparison data were the most recent data points available before and after the first injection (or injection series), respectively. All comparison data were observed within 12 weeks of the initial injection (if the injection scheme was open-ended) or after the final injection in the first injection series. Data on intramuscular steroids and intramuscular saline/local anesthetic were combined as a comparison group for nonepidural injections, a decision consistent with RCTs and systematic reviews demonstrating a lack of efficacy of parenteral steroids for sciatica.^70,71 

The principal summary measures were positive response (dichotomous) and pain score reduction on an 11-point rating scale (continuous). Effect size of dichotomous data was calculated as relative risk (RR), which represents the risk of a positive response for pain relief in the ESI treatment group divided by risk of a positive response in the control group. Effect size of continuous data was calculated as mean difference (MD), which represents the difference in pain score reduction between the two groups.

Random effects models were examined. Heterogeneity was measured by I² which assessed variability among studies not attributable to chance alone. Significant heterogeneity was present with I² values of more than 50%. To assess for small study effects and possible publication bias, a funnel plot was analyzed when more than 10 studies were present. Indirect comparisons of aggregate data were performed by calculating differences in pertinent treatment outcomes using the formulas log(RR_AB) − log(RR_AC) = log(RR_BC); MD_AB − MD_AC = MD_BC; and SE_AB² + SE_AC² = SE_BC^2. 72  Quality analysis was performed excluding each group of low-quality studies for both methodological and technical scores, and body site analysis was performed by substituting back pain for leg pain data. Calculations were done using Microsoft Excel 2011 (Microsoft Corp., Redmond, WA), RevMan Version 5.1.7. (The Cochrane Collaboration, The Nordic Cochrane Centre, Copenhagen, Denmark, 2011), and Stata 12 (StataCorp LP, College Station, TX). Statistical significance for all tests was set at a P value of 0.05 or less.

In the systematic review, 43 studies provided data for ESI treatment and control groups representing 3,641 patients^{37,52,65–69,73–108}  (fig. 2). Table 3 summarizes study design, patient population, injection groups, outcome measures, and results. Sample sizes ranged from 22 to 228 patients. Injections varied by location, route, frequency, volume, and steroid and local anesthetic content. The number of injections was one in 9 studies, two in 5 studies, three in 4 studies, and a variable number in the remaining 25 studies. Five studies reported on cervical injections, 25 studies on lumbar injections, and 13 on caudal injections. Twenty-eight studies were of both high methodological and technical quality (fig. 3 and table 4).

Three studies directly compared epidural nonsteroid with nonepidural injections.^67,81,83  Both injections were examined in one study of high methodological and technical quality with 130 patients using four “control” groups. In this study⁸¹  the proportion of patients with 50% or more pain relief was not significantly different among all comparison groups: 7% for transforaminal local anesthetic, 19% for transforaminal saline, 21% for intramuscular steroids, and 13% for intramuscular saline. No difference in pain score reduction was found among the four groups. Two other studies were of high technical quality and low methodological quality. In a study by Iversen et al.⁶⁷  evaluating 116 patients using two “control” groups, there was no significant difference in pain score reduction between epidural and intramuscular saline. In an earlier study by Klenerman et al.⁸³  using three “control” groups in 63 patients, neither reduction in pain scores nor positive response as judged by a physician (sham nonepidural dry needling 83%, epidural local anesthetic 69%, and epidural saline 69%) significantly differed among treatments. Of note, none of these three studies were designed to detect a difference between two “control” groups, and the latter two studies^67,83  used excessively high injectate volumes (≥20 ml) that diluted the steroid dose, resulting in no differences being observed between the steroid and any control group.

Among studies included in the systematic review, 22.9% (8 of 35) of studies comparing ESI with ENSIs demonstrated benefit for the treatment, which was less than the 58.3% (7 of 12) reporting a positive effect when nonepidural injections were used as the control. When low-quality studies were excluded, these numbers changed only slightly, to 27.3% (6 of 22) and 50.0% (2 of 4), respectively.

For the positive response outcome, 166 patients from two studies provided data for the direct meta-analysis of ENSIs and nonepidural injections (fig. 4). For the indirect meta-analysis, 1,512 patients from 23 studies provided data comparing ESI versus ENSIs and 663 patients from seven studies provided data comparing ESI versus nonepidural injections. The indirect meta-analysis revealed a greater than two-fold increased likelihood for a positive response to ENSI, compared with nonepidural injection (RR [95% CI], 2.17 [1.87–2.53]). Differences between epidural nonsteroid and nonepidural injections for the direct meta-analysis were not significant (RR [95% CI], 0.90 [0.60–1.33]). Table 5 presents other effect estimates for positive response. The absolute benefit favoring epidural nonsteroid over nonepidural injections is actually greater (risk difference [95% CI], 0.27 [0.15–0.39]) than the difference between ESI and epidural nonsteroid (0.04 [−0.01 to 0.10]). Heterogeneity was 0% for the direct comparison and 31–33% for the two groups used in indirect comparisons. When studies of low methodological or technical quality were excluded, no significant changes in outcomes or heterogeneity were noted for any comparisons involving a positive response, which is consistent with previous reviews.⁵⁴ 

For the pain score reduction outcome, 201 patients from two studies provided data for the direct meta-analysis (fig. 5). For the indirect meta-analysis, 1,936 patients from 22 studies provided data comparing ESI versus ENSIs and 619 patients in four studies provided data comparing ESI versus nonepidural injections. Differences between epidural nonsteroid and nonepidural injections were nonsignificant in the direct meta-analysis (MD [95% CI], 0.22 [−0.50 to 0.94]). For the indirect meta-analysis, a small, nonsignificant difference favoring ENSIs over nonepidural injections was noted (MD [95% CI], −0.15 [−0.55 to 0.25]). Heterogeneity was 0% for the direct comparison and 60–72% for the two groups used in indirect comparisons. When studies of low methodological or technical quality were excluded, no significant changes in outcome or heterogeneity were noted for all comparisons involving pain score reduction.

Estimates of both positive response and pain score reduction did not change significantly with either exclusion of low technical and methodological studies or substitution of back pain scores for leg pain scores for the seven studies stratifying pain by body site. The only comparison group with 10 or more studies was the ESI versus epidural nonsteroid comparison. Examination of funnel plots for studies comparing ESI and ENSIs for both primary outcomes revealed small study effects for only the pain score reduction outcome. The Egger test confirmed the presence of possible publication bias (P < 0.001). Post hoc sensitivity analysis to identify and correct for funnel plot asymmetry arising from publication bias included the “trim and fill” method. This method estimated eight studies were needed to account for possible publication bias, and provided a corrected pain score reduction estimate that was not significant.

The findings in this systematic review and meta-analysis comparing epidural nonsteroid and nonepidural injections are mixed, with only one study of high quality directly comparing these treatments. Although no difference in direct outcomes between the two “control” injections was demonstrated, the larger number of studies providing indirect comparisons suggests ENSIs may provide greater benefit for spinal pain than nonepidural injections. This conclusion is based on the significant but small difference found between the two treatments when examining the positive response outcome. For this outcome, the benefit favoring epidural nonsteroid over nonepidural injections is actually greater (risk difference [95% CI], 0.27 [0.15–0.39]) than the difference between ESI and epidural nonsteroid, suggesting that, at least in the short term, most of the benefit of epidural injections may derive from the solution itself, rather than the steroid. ENSIs also showed a nonsignificant trend toward greater relief when examining pain score reduction with indirect comparisons. A single binary outcome measure may represent a better reflection of global perceived effect than reduction in pain score, which is only one of many core domain outcome measures,¹⁰⁹  and in most studies analyzed signified only the pain rating at a single cross-section in time.

Although several review articles^9,31,36  and clinical trials^67,81  have alluded to the possibility of a therapeutic effect for epidural nonsteroid solutions, this assertion has never been systematically examined. In addition to the evidence presented here, several other randomized studies indirectly bolster this assertion. Randomized, double-blind studies comparing high doses of steroid with lower doses in which the steroid was replaced by saline^110,111  or local anesthestic¹¹²  have consistently failed to demonstrate any significant differences between treatment groups. A systematic review by Rabinovitch et al.¹¹³  found a statistically significant benefit for larger epidural injectate volumes irrespective of the contents, suggesting that the beneficial effect of nonsteroid solutions may counterbalance dilution of steroids.

There are several possible explanations for our findings. The most likely is that nonsteroid solutions injected epidurally may provide benefit comparable with that of steroids via a host of different mechanisms, to include the suppression of ectopic discharges from inflamed nerves, enhancing blood flow to ischemic nerve roots, lysis of iatrogenic and inflammatory adhesions, the washout of proinflammatory cytokines, and reversing peripheral and central sensitization.^9,31–34  A second possible reason involves the placebo effect. Epidural injections, especially those administered via the transforaminal approach, often elicit a reproduction of radicular symptoms,^114,115  which is not observed with soft-tissue injections, and may undermine the effectiveness of blinding. When this occurs, it may lead to a greater placebo response with ENSIs, as patients mistakenly believe they received epidural steroids.¹¹⁶ 

The implications of these findings are widespread and protean. Investigators designing clinical trials, and specialty organizations, patient advocate groups, and third-party payers evaluating studies, should consider these results when evaluating the efficacy of ESI. Specifically, trials that include an epidural nonsteroid “placebo” group may be less likely to demonstrate a difference in pain outcomes compared with nonepidural injections. In high-risk patients (e.g., patients with previous surgery) and procedures (e.g., cervical and thoracic transforaminal ESI) in which the inadvertent intravascular injection of depo-steroids can have catastrophic consequences such as paralysis and death,^117–119  physicians might consider removing steroids from the injectate and using nonsteroid solutions as a first-line treatment. Using nonsteroid solutions may also reduce the risk of rare but potentially fatal complications such as meningitis, which has recently been attributed to a contaminated steroid batch.^# On the basis of these results and the results of other clinical trials demonstrating no differences between high- and low-dose ESI^{110–112,120}  the dose of steroids may be considerably reduced or even eliminated in high-risk patient populations. Examples of these patients might include individuals at high risk for avascular necrosis,¹²¹  and those with diabetes,¹²²  at high risk for infection,¹²³  poor wound healing, or in whom the temporary suppression of the adrenocortical axis could adversely affect outcomes (e.g., those scheduled for major surgery).¹²⁴ 

These results should be interpreted in the context of some limitations. First, direct comparisons of the different types of controlled injections were only present in a limited number of low-quality randomized clinical trials. Although no standard guidelines exist regarding the minimum number of studies needed to perform a meta-analysis, analyses of limited trials do exist¹²⁵  and generally agree with longer-term results.¹²⁶  Indirect comparisons do not qualify at the same level of evidence as randomized comparisons, because they represent mere observations of trials, may be subject to bias, and may inaccurately estimate treatment effect.¹²⁷  Yet, none of the three studies that directly compared the different types of “control” injections were designed or powered to detect a difference between nonsteroid groups (which would require significantly more patients than a study designed to detect a difference between ESI and a true placebo),^67,81,83  and two used excessively high injection volumes that resulted in a failure to detect a difference between the diluted steroid treatment (ESI) and any control group.^67,83  As the authors of the most robust study noted,⁸¹  detecting a difference between two treatments (or control groups) with similar effect sizes would require between 1,000 and 2,000 patients, which is not practical. Consequently, indirect analyses evaluating numerous, well-designed studies may provide a better likelihood of detecting a difference between nonepidural injections and ENSIs in this context. Second, some analyses exhibited substantial heterogeneity, which is likely attributable to differences in methods or outcome assessments. Although the conversion of different pain-rating scales may result in increased heterogeneity and greater difficulty in detecting differences in outcomes, previous studies have consistently determined that there is a high correlation between pain-rating scales,^128–130  and scores derived from different scales are often combined in meta-analyses, including those evaluating ESI.¹³¹  With regard to differences in treatment parameters (e.g., region, number of injections, dose and type of steroid), recent reviews have concluded that minor variations in practice are likely to have no significant effect on outcome.^27,132  For example, increasing the depo-steroid dose of more than 40 mg appears to provide no added benefit, and there is little evidence that a series of ESI results in better outcomes than a single injection, or tailoring the number of injections to patient response.^{27,110–112,120,132,133}  However, the conglomeration of these different factors (e.g., injection type and number, dose, volume) may have a cumulative effect, and hence limit the generalization of the meta-analyses. Third, publication bias may be present for studies that compared ESI and ENSIs, with modeling suggesting a nonsignificant outcome favoring ESI when a correction for small study effects was performed. Fourth, our technical rating scale remains formally unvalidated. If detecting a difference between a placebo and control requires between 50 and 150 patients, identifying outcome difference for different variables (e.g., fluoroscopy vs. no fluoroscopy, disability vs. no disability) would require exponentially more patients, and be logistically challenging. Fifth, inherent to any meta-analysis are the biases contained in the included studies. Finally, to enhance generalization, we elected to include studies with follow-up periods varying from a few days to up to 3 months. Hence, this efficacy analysis was not designed to assess the long-term benefits of ESI or controlled injections.

In conclusion, the evidence comparing epidural nonsteroid with nonepidural injections is limited but suggests that ENSIs may not constitute a true placebo treatment. In light of these findings, opportunities exist for clinicians and investigators to modify their approach to these procedures, such as reducing^{110–112,120}  or even in some cases eliminating, the steroid component of epidural injections in high-risk scenarios, and performing high-quality RCTs that directly compare epidural nonsteroid and nonepidural injections.