Literature Search Strategy

We conducted electronic searches in Ovid^® MEDLINE^®, PsycINFO^®, Cochrane CENTRAL, and Cochrane Database of Systematic Reviews in April 2021 (see Appendix A for full strategies). The search reached back to 1990 for each database. This date corresponds to publication of the earliest clinical studies on the interventional procedures addressed in this review. Reference lists of relevant systematic reviews were screened for additional studies.

Inclusion and Exclusion Criteria and Study Selection

The criteria for inclusion and exclusion of studies for this review were based on the Key Question. The population was adults (≥18 years of age) undergoing one of the specified interventional procedures for pain. Details regarding the populations, interventions, comparators, outcomes, timing, and setting (PICOTS) are summarized in Table 1. Briefly, we included studies of populations undergoing the designated interventional procedures for the specified conditions. Pain could be of any duration, though for all of the conditions other than vertebral compression fracture, symptoms are typically subacute or chronic before interventional procedures are performed. Although the population of interest was patients eligible for Medicare, we did not restrict inclusion to studies of patients meeting Medicare eligibility criteria (e.g., based on age, presence of disability, or end-stage renal disease), in order to provide a more comprehensive synthesis of the available literature, and because findings in younger populations may be applicable to older populations. However, studies of patients eligible for Medicare were highlighted, when available.

We restricted inclusion to studies of the selected interventions administered as single therapy, in order to isolate the effects of the intervention. The exception was ozone therapy, for which we included trials of intradiscal ozone administered with an epidural corticosteroid, because there were no trials of ozone without corticosteroid. Comparisons were against a sham control, usual care, or no treatment. Sham controlled trials utilize a procedure that mimic the intervention of interest, without purported therapeutic benefit. Use of sham controls enables blinding of patients to the treatment received, avoiding placebo effects that could inflate estimates due to expectations of benefits. In usual care or no treatment trials, there is no attempt to blind patients to the treatment received. Although usual care or no treatment trials can overestimate benefits due to placebo effects, they are also more pragmatic, reflecting how treatments are administered in clinical practice.⁴⁶ In addition to placebo effects, other reasons that sham and usual care trials may conflict include any therapeutic effects of the sham interventions or differences between sham and usual care trials with regard to the populations evaluated, intervention methods, or other factors. For cooled and pulsed radiofrequency denervation, we also included comparisons against conventional radiofrequency denervation, since these interventions are considered modifications of conventional radiofrequency denervation. For ozone plus corticosteroid therapy, we modified the protocol to also include trials with a corticosteroid control arm (with or without local anesthetic), in order to assess the incremental effects of ozone therapy in addition to a corticosteroid.

Outcomes were pain, function, quality of life, health status, mood, medication use (including opioids), and harms. The review focused on randomized controlled trials (RCTs), given the susceptibility of observational and other non-randomized studies to bias and confounding, particularly for more subjective outcomes such as pain and function.^47,48 In the specific case of interventional pain treatments, there are a number of examples of non-randomized studies^49–56 that overestimated benefits compared with RCTs.^43,57 However, cohort studies were eligible for evaluation of benefits if no RCTs were available; large (n>500) observational studies were eligible for evaluation of rare, serious harms.⁵⁸

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.

All citations were reviewed by one investigator for potential inclusion and full text review. Excluded abstracts were reviewed by a second investigator to confirm the exclusion decision. Each full-text article was independently reviewed for eligibility by two team members. Disagreements were resolved by consensus.

Searches were updated for new publications while the draft report was posted for public comment. Literature identified during the update search was assessed using the process described above for the original search. Any new eligible literature identified in the update search was incorporated into the report prior to finalization.

Table 1

PICOTS (population, intervention, comparator, outcome, timing, setting).

Data Abstraction and Data Management

For studies meeting inclusion criteria, evidence tables were created to show data on study characteristics, outcomes, and applicability. If necessary, data were estimated from graphs provided in the studies. We also calculated standard deviations for baseline and followup data from standard errors or 95 percent confidence intervals (CIs) if necessary using standard formulas,⁵⁹ and calculated mean differences or relative risks with 95 percent CIs using online calculators.^60,61 Abstracted and calculated data were verified for accuracy and completeness by a second team member.

Effects on pain were abstracted as mean difference in pain intensity (continuous) and likelihood of experiencing improvement in pain (dichotomous) based on meeting a certain threshold (“pain response”). For pain and other outcomes evaluated as continuous outcomes, we abstracted adjusted mean differences at followup if available, as well as unadjusted differences in followup scores or change from baseline, and unadjusted differences in change from baseline. Pain intensity using a 0 to 100 scale was transformed to a 0 to 10 scale for ease of interpretation. For pain evaluated as a dichotomous outcomes (pain response), we abstracted (in descending order of prioritization) the proportion of patients experiencing improvement in pain intensity of at least 30 percent, at least 50 percent, or improvement in pain at an alternative threshold (e.g., ≥25%, or >2 point improvement on a 0 to 10 scale), or pain relief rated as moderate, good, or similar using a categorical scale.

Effects on function were based on the mean improvement in a functional scale (dichotomous) or the proportion of patients meeting a defined threshold of functional improvement (dichotomous, e.g. improvement in function of at least 30 percent or at least “moderate” improvement on a categorical scale). The most common functional outcomes were the Roland-Morris Disability Questionnaire (RDQ, 0 to 24 scale) or Oswestry Disability Index (ODI, 0 to 100 scale) for low back pain and the Migraine Disability Test (MIDAS) for migraine (based on number of disability days in the last 3 months for various activities; score >20 indicates severe disability); on each of these scales higher values indicate greater functional disability. Effects on health status (most commonly, the Short-form-36 Physical or Mental Component Summary Scales [0 to 100 scale, higher score indicate better health status]), generic quality of life (e.g., the EuroQOL 5-Dimension Questionnaire [EQ-5D, 0 to 1 scale, higher score indicates better quality of life]), condition-specific quality of life (e.g., the Quality of Life Questionnaire of the European Foundation for Osteoporosis [QUALEFFO, 0 to 100 scale, higher score indicates worse quality of life]), and mood (e.g., Beck Depression Inventory [BDI, 0 to 63 scale, higher value indicates more severe depression symptoms], Profile of Mood States [POMS, 0 to 200 scale, higher values indicates worse mood state], or others) were based on mean improvements in scales designed to assess these domains.

For pain, function, mood, and disease-specific quality of life (QUALEFFO), negative values for mean improvement indicate a better outcome; for health status and generic quality of life (EQ-5D), positive values indicate a better outcome. Effects on harms were based on the proportion of patients experiencing harms (mortality, serious adverse events, any adverse event, and incident vertebral fracture). The duration of pain symptoms at the time of study enrollment was classified as acute (<4 weeks), subacute (4 to 12 weeks), or chronic (>12 weeks). Outcomes were evaluated at predefined followup periods: ≥1 to ≤2 weeks (hereafter written as “1 to 2 weeks”), >2 weeks to ≥1 month (“2 to 4 weeks”), >1 to <6 months (“1 to 6 months”), ≥6 to <12 months (“6 to 12 months”), and ≥12 months (“12 months and longer”).

Study data was abstracted by one team member and all data were verified for accuracy and completeness by a second team member. A record of studies excluded at the full-text level with reasons for exclusion was maintained (Appendix B).

Quality (Risk of Bias) Assessment of Individual Studies

Methods from the AHRQ Methods Guide were used in concordance with the approach recommended in the chapter, Assessing the Risk of Bias of Individual Studies When Comparing Medical Interventions.^45,62 RCTs were assessed based on criteria established in the Cochrane Handbook for Systematic Reviews of Interventions (Chapter 8.5 Risk of Bias Tool),⁶³ and instruments tailored to observational studies were used for nonrandomized studies⁶⁴ (Appendix C). Each study was independently reviewed for risk of bias by two team members. Any disagreements were resolved by consensus. Based on the risk of bias assessment, individual included studies were rated as “good,” “fair,” or “poor” quality as described below:

Poor quality studies were not excluded a priori. When meta-analysis was conducted subgroup analysis was performed based on study quality. When meta-analysis was not conducted, results from poor quality studies were described but conclusions were based on higher (fair and good) quality studies when available.

Data Analysis and Synthesis

We constructed evidence tables showing study characteristics, results, and quality ratings for all included studies, and summary tables to highlight the main findings, organized by Key Question (intervention and condition).

Meta-analysis were limited to vertebroplasty, due to the small number of studies for other interventions, methodological limitations in the studies, and variability in the studies, including outcome measures assessed, timing of assessment, and patient characteristics.⁶⁵ For interventions other than vertebroplasty, evidence was synthesized qualitatively.

Subgroup analyses were conducted to evaluate how pain duration, volume, presence of bone marrow edema, and study quality impacted estimates. Details regarding meta-analysis methods are provided in Appendix D.

A sub-Key Question addressed how benefits and harms varied according to demographic (age, sex, race/ethnicity), clinical (type of pain, severity of pain, prior treatments, medical and psychiatric comorbidities), and technical factors (variations in techniques, intensity, frequency, dose, and number of treatments). Although planned techniques to assess these factors included sensitivity and stratified analyses, other than polymethyl methacrylate (PMMA) volume, evidence was too limited to apply these techniques. However, we evaluated findings from within-study subgroup analyses on these factors when available.

The magnitude of effects for pain and function were classified using used in other recent AHRQ reviews on pain conducted at our EPC.^66–70 A small effect was defined for pain as a mean between-group difference following treatment of 0.5 to 1.0 points on a 0- to 10-point numeric rating scale (NRS) or visual analogue scale (VAS) and for function as a standard mean deviation (SMD) of 0.2 to 0.5 or a mean difference of 5 to 10 points on the 0 to 100-point ODI, 1 to 2 points on the 0 to 24-point RDQ, or equivalent. A moderate effect was defined for pain as a mean difference of 1 to 2 points on a 0- to 10-point NRS or VAS and for function as an SMD of 0.5 to 0.8, or a mean difference of 10 to 20 points on the ODI, 2 to 5 points on the RDQ, or equivalent. Large/substantial effects were defined as greater than moderate. We applied similar thresholds to other outcomes measures.⁷¹ Small effects using this system may not meet proposed thresholds for clinically meaningful effects.⁷² However, there is variability in estimated minimum clinically important differences across studies, therapies for pain (included those recommended in guidelines) are often associated with effects below minimum clinically important difference thresholds,^44,73–76 and the clinical relevance of effects classified as small might vary for individual patients depending on preferences, baseline symptom severity, harms, cost, and other factors.^77,78 For some individuals, a small improvement in pain, function, or other outcomes using a treatment with low cost or no serious harms may be important.

Grading the Strength of Evidence

Regardless of whether evidence was synthesized quantitatively or qualitatively, the strength of evidence (SOE) was assessed as high, moderate, low, or insufficient, using the approach described in the AHRQ Methods Guide, based on study limitations, consistency, directness, precision, and reporting bias (Appendix E).⁴⁵ Based on input from the Technical Expert Panel, pain and function were classified as primary outcomes and the other outcomes were classified as secondary. When higher (fair- or good-quality) studies were available, poor-quality studies were not used to determine SOE. To ensure consistency and validity of the SOE evaluation, the initial assessment was made by one investigator and independently reviewed by at least one other investigator using the following criteria, with disagreements resolved by consensus.

Plain-language statements were used in the Abstract and Main Points to indicate the SOE. High SOE was described as “is associated with” or simply “reduces/increases;” moderate SOE was described as “probably;” and low SOE was described as “may”.

Assessing Applicability

Applicability to U.S. practice settings and the Medicare population (i.e., patients eligible for Medicare due to age 65 or greater or disability [including end-stage renal failure (ESRD)]) were assessed based on the AHRQ Methods Guide, using the PICOTS framework.⁴⁵ Applicability refers to the degree to which outcomes associated with the intervention are likely to be similar across patients and settings relevant to the care of the Medicare population based on the populations, interventions, comparisons, and outcomes synthesized across included studies. Factors that may affect applicability, which we identified a priori, include: (1) patient factors (e.g., age and disability status, medical and psychiatric comorbidities, symptom severity, duration and underlying pain condition); (2) technical factors (e.g., medications used [for procedures that involve medications], intensity or dose, number of treatments, frequency of treatments, duration of treatment, use of imaging guidance, technique utilized, and clinical background of person performing the procedure [e.g., anesthesia pain medicine, interventional radiology, or other]); (3) comparators (e.g., sham procedure, no treatment, or usual care); (4) outcomes (e.g., use of nonstandardized or unvalidated outcomes); and (5) settings (e.g., country). We used information regarding these factors to assess the extent to which interventions and results are likely most relevant to real-world clinical practice in typical U.S. settings that include the Medicare population and provided a qualitative summary of our assessment.

Peer Review and Public Commentary

Experts were invited to provide external peer review of this systematic review; AHRQ and an associate editor also provided comments. In addition, the draft report was posted on the AHRQ website for 4 weeks for public comment. Comments were reviewed, considered, and addressed as appropriate. Edits were made for clarity and accuracy; however, no changes were made to the evidence or to our conclusions.