Key Findings and Strength of Evidence

The key findings or this review, including strength of evidence ratings are summarized in the summary of evidence table (Table 30); the factors used to determine the overall strength of evidence are summarized in Appendix H. This report updates a previous review that we conducted^{15, 16} for the American College of Physicians and American Pain Society, expanding upon it with new evidence and evaluation of several additional interventions (e.g., tai chi, taping, electrical muscle stimulation). This report also incorporates evidence on new drugs within previously reviewed classes (e.g., the antidepressant duloxetine and the antiseizure medication pregabalin). Because of the large number of interventions addressed in this review, reviewing all of the primary literature was not feasible. Therefore, we used relevant, well-conducted systematic reviews when available, including updates of systematic reviews included in our prior report, and supplemented prior reviews with additional trials that were published subsequent to the reviews or not included for other reasons. All conclusions are based on the totality of evidence (i.e., studies included in systematic reviews plus additional primary studies). Across interventions, pain intensity was the most commonly reported outcome, followed by back-specific function (typically measured using the Roland-Morris Disability Questionnaire [RDQ] or the Oswestry Disability Index [ODI]). When present, observed benefits were generally in the small (5 to 10 points on a 100-point visual analogue scale [VAS] or equivalent or standardized mean difference [SMD] of 0.2 to 0.5) to moderate (10 to 20 points, or SMD of 0.5 to 0.8) range for pain. Effects on function were typically smaller than effects on pain; in some cases there were positive effects on pain but not on function, and fewer studies measured function than pain. Other outcomes (such as quality of life, mood, work, analgesic use, or utilization of resources) were generally reported inconsistently and data were too sparse to reach reliable conclusions.

Table 30

Summary of evidence.

New evidence affected conclusions for several classes of medications. For acetaminophen, the prior review concluded that acetaminophen was effective for acute low back pain, primarily based on indirect evidence from trials of acetaminophen for other conditions and trials of acetaminophen versus other analgesics. However, a recent, well-conducted trial—the first placebo-controlled trial in patients with acute low back pain—found acetaminophen to be no more than effective than placebo (strength of evidence [SOE]: low).⁴³ For antidepressant drugs, no studies in the prior review evaluated drugs in the serotonin norepinephrine reuptake inhibitor class. Evidence from several trials indicates that duloxetine is more effective than placebo for pain and function in patients with chronic low back pain (SOE: moderate).^152-154 However, effects were small (less than 1 point on a 0 to 10 scale) and all trials were funded by the manufacturer of duloxetine and led by the same researcher. For antiseizure medications, new evidence is available on pregabalin for radicular low back pain, but the studies had methodological shortcomings and were too inconsistent to reliably estimate effects (SOE: insufficient).^{160, 163} The prior review found no studies on the effects of benzodiazepines for radiculopathy; one recent trial found that benzodiazepines were no more effective than placebo in for this condition (SOE: low).¹³⁵ The trial also found that for some outcomes, such as return to work, benzodiazepines were associated with worse outcomes versus placebo.

Main conclusions regarding the benefits and harms of pharmacological therapies for low back pain were otherwise relatively unchanged from the prior review. One area in which conclusions did change was related to effectiveness of tricyclic antidepressants. In our prior review, tricyclic antidepressants were found to be associated with small beneficial effects for chronic low back pain. However, evidence reviewed for this report suggests that tricyclic antidepressants are not effective versus placebo (4 trials; SMD −0.10, 95% CI −0.51 to 0.31; I²=32%) (SOE: moderate).¹³⁹ As noted above, duloxetine, a serotonin norepinephrine reuptake inhibitor that is not associated with the anticholinergic and cardiac side effects of tricyclics, is now available as a potential alternative antidepressant. Skeletal muscle relaxants appear to be effective for short-term pain relief in patients with acute low back pain, but are also associated with an increased risk of central nervous system adverse events (in particular, sedation) (SOE: moderate). Systemic corticosteroids do not appear to be effective versus placebo for either radicular or nonradicular low back pain (SOE: moderate) and evidence on the effectiveness of benzodiazepines versus placebo for nonradicular low back pain remains sparse (SOE: insufficient).³⁴

Evidence on the effectiveness of opioids for low back pain remains limited to short-term trials showing modest effects versus placebo on short-term pain and function⁶⁹ (SOE: moderate). Almost all trials of opioids enrolled patients with chronic low back pain, and no trial focused on patients with radicular symptoms. There remain no clear differences among different long-acting opioids or among long- versus short-acting opioids. Findings regarding the increased risk of opioids versus placebo for harms such as constipation, nausea, sedation, and dry mouth are also unchanged. Most trials of opioids used an enriched enrollment and withdrawal design; evidence from studies of chronic pain in general (not restricted to low back pain) suggest that estimates of harms are larger in trials that did not use such a design.⁹² Trials of opioids for low back pain were not designed to assess risk of serious adverse events such as overdose, abuse or addiction, or accidental injuries, due to their relatively small samples and short duration of followup. In addition, trials of opioids typically excluded patients with risk factors for overdose, abuse, or addiction. Observational studies have shown an association between use of opioids for chronic pain and serious harms such as overdose that appears to be dose-dependent.⁵⁹⁹ However, such studies did not meet inclusion criteria for this report because they were not restricted to patients with low back pain.

Relatively few studies directly compared the effectiveness of different medications for low back pain, or the effectiveness of a combination of medications versus one of the component medications of the combination alone. We found no clear differences between opioids versus nonsteroidal anti-inflammatory drugs (NSAIDs) in pain relief or function, and no clear differences between benzodiazepines versus skeletal muscle relaxants. As described above, there were no clear differences between acetaminophen versus NSAIDs in patients with chronic low back pain.

Serious harms were generally not observed in trials of nonopioid medications, though harms were generally not reported well. Like trials of opioids, trials of nonopioid medications were not designed to assess risk of serious, uncommon harms (e.g., liver toxicity with acetaminophen, bleeding with NSAIDs, fracture or infection with corticosteroids, or abuse or addiction with benzodiazepines).

The current report addresses several nonpharmacological therapies not addressed in the prior APS/ACP review. Evidence on taping (using techniques to increase skin tension) did not clearly show beneficial effects versus sham taping comparisons, though findings were limited by methodological shortcomings and inconsistency (SOE: insufficient to low). There was insufficient evidence to determine the effects of electrical muscle stimulation, due to methodological shortcomings in the trials and imprecision (SOE: insufficient). Two trials found that tai chi was more effective versus wait list control for pain intensity and function²⁸⁵ (SOE: low); effects appeared similar to those observed for other types of exercise and related interventions.

As in the APS/ACP review, we found little evidence to support the use of most passive physical modalities for low back pain. An exception was superficial heat, which was found to be more effective than a nonheated control for acute or subacute low back pain (SOE: moderate). There remains insufficient evidence to determine effects of superficial cold. There also remains insufficient evidence to determine the effectiveness of percutaneous electrical nerve stimulation, interferential therapy, short-wave diathermy, traction and lumbar supports versus sham or no treatment. Although evidence on effectiveness of ultrasound and transcutaneous electrical nerve stimulation (TENS) was previously classified as insufficient, additional evidence now supports the findings that ultrasound is not effective versus sham ultrasound⁴⁹⁷ and that TENS is not effective versus sham TENS,⁵¹² though the strength of evidence remains low due to methodological limitations in the trials and imprecision. Based on three trials,^{550, 552, 553} low-level laser therapy was more effective than sham laser for pain, though methods for assessing pain and duration of followup varied; there was insufficient evidence from one trial to determine effects on function. Evidence to compare effects of one physical modality versus another, or a physical modality versus another active intervention, was generally too limited to reach reliable conclusions.

As in the APS/ACP review, we found evidence that psychological therapies (progressive relaxation, electromyography (EMG) biofeedback, operant therapy, combined psychological therapy [e.g., cognitive-behavioral therapy]) are associated with lower pain intensity (effects small to moderate) versus wait list control; effects of function were observed for progressive relaxation and combined psychological therapy only (SOE: low).³¹³ Multidisciplinary rehabilitation (consisting at a minimum of exercise therapy plus psychological therapy, with some coordination) was associated with moderately lower pain intensity versus usual care, with smaller effects on function and no clear effect on return to work (SOE: moderate).³⁴⁸ Psychological therapies and multidisciplinary rehabilitation were primarily evaluated for chronic low back pain, with insufficient evidence to determine effects in patients with acute low back pain or in those with radicular symptoms. Unlike the prior review, a stratified analysis reported in a systematic review found no association between the intensity of multidisciplinary rehabilitation and estimates of effectiveness,³¹³ though head-to-head comparisons of different intensities of multidisciplinary rehabilitation are not available. In head-to-head comparisons, there were no clear differences between psychological therapies versus exercise therapy, though multidisciplinary rehabilitation was moderately more effective than physical therapy not administered as part of a multidisciplinary program.

Our findings regarding the effectiveness of massage, acupuncture, and manipulation were generally consistent with the APS/ACP review in showing some beneficial, primarily short-term effects. These interventions were primarily evaluated for chronic low back pain, with few trials of patients with acute low back pain or specifically with radicular symptoms. Evidence was generally stronger for acupuncture^{364, 365} and spinal manipulation^{437, 438} than for massage,⁴⁰⁵ which was evaluated in fewer trials, though the strength of evidence varied depending on the specific comparison evaluated (no SOE was rated above moderate). For all of these therapies, the evidence was characterized by marked heterogeneity in the interventions evaluated as well as in the intensity and number of sessions. Although some evidence suggested that massage is more effective versus other interventions considered active, it was not possible to draw strong conclusions due to methodological limitations and imprecision.⁴⁰⁵ Although acupuncture was more effective than sham acupuncture for chronic low back pain,³⁶⁵ sham acupuncture techniques varied among trials (superficial needling at acupuncture points, superficial needling at nonacupuncture points, nonpenetrating needles or pressure at acupuncture points) and there was inconsistency, with some trials showing no differences between acupuncture versus sham and effects were primarily observed immediately after treatment, with limited evidence of no effects at longer-term followup. Spinal manipulation was no more effective than sham manipulation for chronic low back pain, but manipulation was as effective as other interventions thought to be effective.⁴³⁸ Therefore, there remains some uncertainty regarding the specific effects of these interventions, versus nonspecific effects related to needling, mobilization or manipulation, or other aspects of administering these treatments (e.g., attentional or placebo effects). Head-to-head trials that directly compared different massage, acupuncture, or spinal manipulation techniques generally found no clear differences.

Findings regarding the effectiveness of exercise therapies and related interventions were also consistent with the APS/ACP review. Most trials evaluated patients with chronic nonradicular low back pain. For yoga, newer trials strengthen conclusions regarding effectiveness, particularly for yoga versus educational interventions (SOE: moderate). Evidence on motor control exercises, which were not addressed in the APS/ACP review, was generally consistent with evidence for other types of exercise in showing small to moderate effects (SOE: low). Head-to-head trials of exercise programs generally found no clear differences in estimates of effectiveness.

Harms were not well-reported in trials of nonpharmacological therapies, though serious adverse events appear rare. For physical modalities, harms when reported were mostly related to superficial effects at the application site. Severe neurological complications were not reported in trials of lumbar spinal manipulation and serious infections, bleeding, or other complications were not reported in trials of acupuncture.

Findings in Relationship to What Is Already Known

Our findings are generally consistent with prior systematic reviews on noninvasive treatments for low back pain, in part because our report builds upon a prior review and used previously published, high-quality systematic reviews to inform its findings. However, our findings were also generally consistent with other recent systematic reviews that were not used in this report that found NSAIDs and opioids associated with small to moderate effects versus placebo for chronic low back pain, and tricyclic antidepressants associated with small effects that were not statistically significant.^{600, 601} Like other reviews, we only found evidence supporting short-term benefits of opioids. Although another review found no differences between opioids versus placebo for low back pain, searches were conducted through 2005 and its findings were based on only four trials, with a pooled estimate that slightly favored opioids (SMD −0.18, 95% CI −0.49 to 0.11).⁶⁰² As in other reviews, we found no randomized trials to determine long-term effectiveness of opioids for low back pain. In a recent review that we conducted on opioids for chronic pain in general, we also found no cohort studies to determine the long-term effectiveness of opioids versus no opioid therapy.⁶⁰³ In that review, we found insufficient evidence from randomized trials to determine the risk of serious harms associated with opioids, due to small samples, inadequate length of followup, poorly standardized methods for assessing harms, and suboptimal harms reporting. In addition, trials typically excluded patients at higher risk for abuse or overdose, though evidence^{604, 605} indicates that such patients are more likely to be prescribed opioids in clinical practice than people without risk factors.^{600, 606}

Our finding that acetaminophen is not effective for acute low back pain is based on a recent, well-conducted randomized controlled trial (RCT)⁴³ and differs from our prior review, which concluded that there was good evidence of moderate effects. However, the prior conclusion was based on indirect evidence of acetaminophen for other pain conditions and effects of acetaminophen versus NSAIDs, which showed few differences. Another systematic review, noting the absence of placebo-controlled trials at the time and imprecision and methodological shortcomings in the available studies, rated the same evidence as insufficient.⁶⁰⁷ Like our review, a recent systematic review found that acetaminophen was ineffective for low back pain, primarily based on the results of the new trial.⁶⁰⁸

Our prior report and other previous systematic reviews^{137, 138} found tricyclic antidepressants associated with small beneficial effects for low back pain. However, the evidence reviewed for this report suggests that they are not effective versus placebo for pain relief (4 trials; SMD −0.10, 95% CI −0.51 to 0.31; I²=32%) or function.¹³⁹ One potential reason for the discrepancy between this finding and prior reviews are that some of the prior reviews did not conduct a meta-analysis.^{136, 138} One positive review¹³⁷ did base findings on a meta-analysis. However, the study in the meta-analysis that reported the largest effect in favor of antidepressants did not report being randomized,⁶⁰⁹ it did not include relevant studies that were in the more current review,^{141, 146, 149} and it did not report methods for data imputation for two trials included in the meta-analysis.^{140, 148}

Our findings regarding the small to moderate effectiveness of the antidepressant duloxetine are consistent with its recent approval by the US Food and Drug Administration for chronic musculoskeletal pain, including chronic low back pain.⁶¹⁰ Our conclusions were based on trials of duloxetine versus placebo. Although a systematic review found no differences between duloxetine versus other oral medications, its findings were based on a network analysis based on indirect comparisons.⁶¹¹

For nonpharmacological treatments, our findings are also generally consistent with other systematic reviews. Like other reviews, we found some evidence to support use of complementary and alternative medicine therapies such as acupuncture, spinal manipulation, and massage.^{600, 612-616} Although acupuncture was no more effective than sham acupuncture in some trials, other reviews have also found that the overall evidence (including pooled estimates) suggest beneficial effects on pain.^{364, 365} As in prior reviews, we found no clear evidence to support one specific type of massage, manipulation, or acupuncture over another.^{362, 404, 426, 617}

Findings regarding the effectiveness of exercise are similar to our prior review and other reviews.^618-620 Our findings are also consistent with more specific reviews that focused on specific types of exercise such as aquatic exercise,⁶²¹ sling exercise,⁶²² walking,⁶²³ stability exercises,⁶²⁴ and modifying patterns of movement.⁶²⁵ Additional evidence published since our prior review strengthens conclusions that yoga is effective for low back pain,^{626, 627} a finding consistent with other recent systematic reviews, and newer evidence supports the effectiveness of motor control exercises. Evidence on tai chi was previously unavailable, but recent randomized trials support its effectiveness. As in our prior review, evidence does not clearly demonstrate that one type of exercise therapy is superior to another. This is consistent with other systematic reviews that have evaluated specific exercise therapy comparisons (e.g., McKenzie versus other exercise methods).¹⁸³

Our findings that psychological therapies and multidisciplinary rehabilitation were both effective are consistent with our prior review and other reviews.⁶²⁸ Other reviews that focused on related interventions such as functional restoration or cognitive-behaviorally based physical therapy (in which the literature overlaps with that on multidisciplinary rehabilitation) have also reached positive conclusions.^{606, 629, 630} Although there was insufficient evidence to determine which patients are most likely to benefit from psychological therapies and multidisciplinary rehabilitation, a recent randomized trial⁶³¹ found that a stratified approach in which patients are assessed for risk factors for chronicity, and higher-risk patients receive more intensive cognitive-behavioral based physical therapy, is more effective than usual care without a stratified approach, suggesting that these therapies may be most effective in higher-risk people. Unlike our prior report, which found that higher-intensity multidisciplinary rehabilitation appeared more effective than lower-intensity programs, a stratified analysis based on currently available evidence³⁴⁸ indicated no clear difference in effects.

Like our prior review, we found that for most physical modalities, evidence was too weak to determine effectiveness. Although we previously found insufficient evidence to conclude that ultrasound and TENS are not effective, albeit with low strength of evidence. A recent assessment of TENS came to a similar conclusion.⁶³²

As in other reviews, we found that evidence the effectiveness of therapies for radicular low back pain was quite limited.^{435, 633} Like other reviews, including our prior report, we found that systematic corticosteroids are not effective for radicular low back pain.^{633, 634} Although duloxetine and other serotonin norepinephrine reuptake inhibitors and antiseizure medications such as gabapentin and pregabalin are increasingly being prescribed for low back pain, particularly when associated with radicular symptoms, evidence on the effectiveness of nonduloxetine serotonin norepinephrine reuptake inhibitors is not available and results of trials of pregabalin and gabapentin have been inconsistent or have not shown clear effects. Although a network meta-analysis has been performed on various treatments for radicular low back pain, the most commonly evaluated treatments were surgical and interventional, findings for noninvasive therapies were primarily based on indirect comparisons, and many estimates were imprecise.⁶³⁵

Applicability

A number of issues could impact the applicability of our findings. Some studies did not specifically enroll patients with acute, subacute, or chronic low back pain, but rather enrolled mixed populations or did not clearly describe the duration of symptoms, which could make it difficult to apply findings if benefits differ according to duration of symptoms. Relatively few studies enrolled patients specifically with radicular symptoms, and many studies did not specifically describe whether patients with radicular symptoms were excluded. Therefore, the degree to which it is possible to extrapolate evidence from studies of patients with primarily nonradicular symptoms to patients with radicular symptoms is uncertain. In addition, studies that focused on radicular pain evaluated clinically diverse populations. Some studies required imaging findings of disc herniation (typically involving younger individuals) or spinal stenosis (typically affecting older individuals), while others did not require imaging confirmation of radicular symptoms. Among studies of patients with nonradicular symptoms, most studies did not attempt to evaluate whether effectiveness varied in subgroups of patients defined by clinical, demographic, imaging, or other characteristics. It is not possible to determine whether effectiveness varies among groups with nonradicular pain based on these factors. For example, most trials of antidepressants excluded patients with depression or only included a small minority of such patients,¹³⁹ such that it is unclear whether antidepressants might have additional effects on mood in patients with low back pain and depression. Across interventions, few studies enrolled any or many older adults. Although trials of motor control exercise (MCE) generally selected patients on the basis of tests showing deficits in motor control, specific testing methods and criteria for inclusion varied, and it is unclear whether effects of MCE vary according to findings on motor control tests.¹⁸⁶

For nonpharmacological treatments, the applicability of our findings is affected by the variability among trials in the interventions and comparators evaluated. For example, trials of acupuncture varied in the sites in which needles were applied, the length of acupuncture sessions, the number of sessions, and the time period over which the sessions were performed.^{364, 365} In trials that evaluated “usual care” comparators, the components of usual care were often not well described or standardized, making it difficult to apply findings to clinical practice. Other factors that could impact the applicability of our findings regarding nonpharmacological interventions includes differences related to the setting in which the intervention was performed (e.g., United States versus another country, specialist versus primary care setting) or due to the training or skill of the person performing the intervention. For acupuncture, for example, some evidence suggests that patient expectations have an important influence on the effectiveness of treatment,^{636, 637} such that results from countries in which acupuncture is widely practiced may not be applicable to settings in which it is considered an alternative practice. Another factor that could impact the effectiveness of interventions is the use of cointerventions, which varied across trials and was frequently not reported well. We separately analyzed comparisons that specifically involved the use of one intervention plus another intervention versus the other intervention alone. For example, spinal manipulation plus another intervention was more effective than the other intervention alone, suggesting potential additive effects.

To help interpret the results of the trials, we categorized the magnitude of effects for pain and function using the system in the APS/ACP review. Based on these categories, beneficial effects when present were in the small or moderate range. However, effects that we classified as small (e.g., 5-10 points on a 0 to 100 scale for pain or function) are below some proposed thresholds for minimum clinically important differences (e.g., 15 points on a 0 to 100 VAS for pain, 2 points on a 0 to 10 NRS for pain or function, 5 points on the RDQ, and 10 points on the ODI; or a 30% change from baseline).²⁸ Nonetheless, our classification system provides some objective bench marks for assessing magnitude of effects, including the smaller effects typically observed in low back pain trials. We also evaluated the proportion of patients who experienced a clinically important improvement in pain or function (e.g., 50% improvement in pain or on the RDQ). However, many studies did not report such dichotomous outcomes, and among those that did, definitions for clinically important improvements varied. A factor that complicates interpretation of findings is that the magnitude of effects might vary depending on the baseline severity, with some evidence suggesting that treatment may be more effective in people with higher baseline symptom severity.⁶³⁸ Also, the clinical relevance of the same absolute improvement in an outcome measure might differ in individual patients depending on the baseline score (e.g., a 1 point change on a 0- to 10-point NRS for pain might differ for someone with a baseline pain score of 4 versus 9). Most trials enrolled patients with pain symptoms of at least moderate intensity (e.g., >5 on a 0- to 10-point NRS for pain). When present, most beneficial effects were observed at shorter-term (e.g., <3 months) followup, with outcomes often assessed only through the end of the active treatment period, such that it was difficult to determine whether there were sustained benefits. When evidence on longer-term followup was available, effects were typically attenuated or no longer observed. Understanding long-term outcomes is particularly critical for chronic low back pain, given the persistent nature of symptoms.

Implications for Clinical and Policy Decisionmaking

Our findings have implications for clinical and policy decisionmaking. A number of pharmacological and nonpharmacological therapies are supported by some evidence of effectiveness in patients with acute (Tables 31, 32, 33, and 34) or chronic low back pain (Tables 35, 36, 37, and 38). Although clinical practice guidelines recommend acetaminophen as a first-line pharmacological therapy for acute and chronic low back pain,^{14, 639} new evidence⁴³ that acetaminophen is ineffective for acute low back pain call into question its appropriateness as a recommended therapy, though findings are based on a single trial and other factors such as low cost, favorable side-effect profile, and effectiveness for other acute pain conditions could also impact decisions regarding its use.⁶⁴⁰ Although tricyclic antidepressants have long been recommended as a secondary treatment option for chronic low back pain, duloxetine has specifically been approved by the US Food and Drug Administration for this condition and appears to be more effective than tricyclic antidepressants as well as associated with a more favorable safety profile, which could impact the selection of drugs within the antidepressant class.

Table 31

Pharmacological therapies versus placebo for acute low back pain.

Table 32

Pharmacological therapies versus active comparators for acute low back pain.

Table 33

Nonpharmacological treatments versus sham, no treatment, or usual care for acute or subacute low back pain.

Table 34

Nonpharmacological treatments versus active comparators for acute or subacute low back pain.

Table 35

Pharmacological therapies versus placebo for chronic low back pain.

Table 36

Pharmacological therapies versus active comparators for chronic low back pain.

Table 37

Nonpharmacological treatments versus sham, no treatment, or usual care for chronic low back pain.

Table 38

Nonpharmacological treatments versus active comparators for chronic low back pain.

The use of opioids for chronic pain has become an area of increasing concern, due to uncertain long-term effectiveness and marked increases in the number of accidental overdoses, as well as other harms related to their abuse potential.⁶⁰³ Patients with low back pain are frequently prescribed opioids and account for a high proportion of the patients prescribed opioids. Decisions regarding the appropriate use of opioids for low back pain must weigh short-term, relatively modest benefits against potential harms. Guidelines recommend risk assessment, careful patient selection, and close monitoring and followup in patients prescribed opioids.⁶⁴¹

The continued paucity of evidence to determine effective treatments for radicular low back pain (Tables 39, 40, and 41) necessitates that most decisions are based on extrapolation of evidence on the effectiveness of treatments for nonradicular low back pain or other nonback-related neuropathic pain conditions. This could explain why antiseizure medications such as gabapentin and pregabalin are being prescribed more for radicular low back pain, despite the lack of evidence showing that they are effective. Systemic corticosteroids continue to be used for treatment of radicular back pain, despite trials showing that they are ineffective, presumably based upon their known anti-inflammatory properties and use in epidural injections.

Table 39

Pharmacological therapies versus placebo for radicular low back pain.

Table 40

Nonpharmacological treatments versus sham, no treatment, or usual care for radicular low back pain.

Table 41

Nonpharmacological treatments versus active comparators for radicular low back pain.

Our review support clinical practice guidelines that found insufficient evidence to recommend most physical modalities, other than superficial heat. However, these therapies are still commonly used in clinical practice. Among nonpharmacological therapies that were found to be effective, there was insufficient evidence to determine which patients are most likely to benefit from specific therapies. However, a recent trial which found that a stratified approach (in which patients are assessed for risk factors for chronicity, and higher-risk patients receive more intensive cognitive-behavioral based physical therapy) is more effective than usual care without a stratified approach suggests that psychologically-based therapies and multidisciplinary rehabilitation may be most effective in higher-risk people.⁶³¹ Other factors that may impact decisions regarding which nonpharmacological therapies to use include cost, availability, and patient preferences. Some evidence suggests that greater patient expectations of benefit from a particular treatment are associated with greater benefits,^{636, 637} suggesting that patient preferences should be considered in the selection of therapies. Potential barriers to use of some nonpharmacological therapies include variability in health insurance coverage (e.g., for complementary and alternative medicine therapies)⁶⁴² and nonavailability depending on locale or other factors (e.g., multidisciplinary rehabilitation).⁶⁴³

Limitations of the Review Process

We included previously published systematic reviews. The reliability of systematic reviews depends on the rigor with which they are conducted.⁶⁴⁴ Therefore, we focused on higher-quality reviews. For a number of interventions addressed in this report, more than one higher-quality systematic review exists. In addition to quality, we therefore also selected systematic reviews for inclusion based on the closest relevance match based on the Key Questions and scope and how recently searches were conducted.²⁴ If two or more reviews were similar on these criteria, we prioritized inclusion of updates of reviews that were in the prior APS/ACP review. In some cases, the highest-quality systematic reviews did not provide all of the information necessary to assess findings. We only included such reviews if we could address any methodological limitations through review and assessment of the primary studies. Otherwise, such reviews were excluded.

We did not conduct meta-analyses or update meta-analyses included in prior systematic reviews. However, for comparisons without a meta-analysis, we synthesized results qualitatively, using the methods in the AHRQ methods guide. For comparisons for which pooled results were available from prior systematic reviews, we evaluated the consistency of results from new trials against the pooled estimates.

Other limitations of the review process are that we excluded non-English language articles and did not search for studies published only as abstracts. However, some systematic reviews included non-English language articles and abstracts, which did not materially impact conclusions. We were unable to assess for publication bias using graphical or statistical methods to detect small sample effects, methodological limitations in the trials, heterogeneity in the interventions, populations, and outcomes addressed, and small numbers of trials for many comparisons. However, based on searches of reference lists, clinical trials registries, and peer review suggestions, we did not find evidence to suggest that unpublished trials would impact conclusions.

There are other noninvasive interventions for low back pain that we did not address, including herbal medicines,⁶⁴⁵ educational interventions,^{646, 647} advice to remain active,^{646, 648} mattresses, shoe insoles,⁶⁴⁹ and others.^{650, 651} We also did not include comparisons of noninvasive therapies versus surgery or interventional procedures, which were outside the scope of this review. We also excluded pain treatment trials that were not restricted to patients with low back pain. The applicability of such trials would depend in part on the proportion of patients with low back pain and what other pain conditions were present.

Limitations of the Evidence Base

The evidence base had a number of important limitations. As noted previously, evidence on the effectiveness of interventions for radicular low back pain was sparse. Most trials of nonpharmacological treatments focused on patients with chronic low back pain, with insufficient evidence to determine effects for acute low back pain, with the exception of superficial heat. This could be due in part because the natural history or acute low back pain is characterized by rapid improvement, such that nonpharmacological therapies are typically reserved for patients who do not improve in the initial period. A number of interventions were evaluated in small numbers of trials or in trials that primarily had important methodological limitations, precluding strong preclusions. In addition, there were relatively few head-to-head trials of different interventions, making it difficult to compare the effectiveness of one type of therapy versus another, particularly for comparisons of nonpharmacological versus pharmacological therapies.

Another limitation of the evidence base is that studies were frequently short term and often only evaluated patients at the end of a course of therapy, making it difficult to determine whether sustained benefits are present. In addition, many trials reported mean changes in outcome measures (typically pain and function), but did not report dichotomized outcomes (e.g., ≥30% or ≥50% pain relief or functional improvement). Assessment of outcomes based on continuous as well as dichotomized outcomes would provide a more complete assessment of treatment effects.⁶⁵²

Some limitations of the evidence were particularly relevant for trials of nonpharmacological interventions. Studies of nonpharmacological interventions were typically characterized by marked heterogeneity in the specific intervention techniques evaluated, as well as in the duration and intensity of treatments, which could attenuate treatment benefits if suboptimal treatment techniques or intensity of therapy was evaluated. In addition, a number of nonpharmacological therapies (e.g., psychological therapies, exercise therapy, massage, and spinal manipulation) are difficult to blind effectively. Therefore, observed benefits could be due in part to placebo, attentional, or other nonspecific effects, and results are susceptible to performance and other biases, though it is not possible to reliably quantify the extent of such effects. Finally, trials of nonpharmacological therapies did not report harms well; this could be in part because serious harms are not expected with most of these treatments.

Research Gaps

A number of research gaps limit the full understanding of the effectiveness, comparative effectiveness, and harms of therapies for low back pain.^653-656 More research is needed to determine effective treatments for low back pain with radicular symptoms and in understudied populations such as older adults. Trials should be designed to not just evaluate patients immediately after they have completed therapy, but for at least several months after the completion of therapy, in order to help understand whether beneficial effects are sustained. Studies that use more pragmatic designs (e.g., more flexible dosing or use of cointerventions) might help improve patient recruitment and reduce attrition, which is high in long-term trials of low back pain. For nonpharmacological treatments, research to identify optimal treatment techniques and regimens (including intensity and duration of treatments) would be very helpful for defining more standardized interventions to be evaluated in trials.

Studies are needed to determine the long-term effectiveness and harms of opioids for chronic low back pain, including higher-risk patients like those commonly encountered in clinical practice. Studies that compare opioids versus nonopioid therapies that address the psychosocial factors often associated with chronic pain (e.g., interdisciplinary rehabilitation, exercise therapy, or psychological therapies) are needed. Observational studies that are designed to assess serious long-term harms provide some evidence regarding risks of opioids for chronic pain in general, but data specifically in patients with low back pain are lacking.⁶⁰³ Although observational studies based on administrative databases can be helpful for evaluating risks of opioids, such studies are often limited in their ability to address important potential confounders, and should ideally be supplemented by clinical data, such as in well-designed clinical registries. For systemic corticosteroids, the largest trial to date was recently completed and should help further characterize the effectiveness (or lack thereof) of this treatment.⁶⁵⁷

More research is needed to help understand whether nonradicular low back pain can be reliably classified into clinically meaningful subgroups, and which patients or subgroups are most likely to benefit from specific therapies.^658-662 Trials are also needed to confirm whether effects of risk-stratified approaches are reproducible in the United States,^{663, 664} and to optimize their implementation.⁶⁶⁵ More research is needed to better understand whether combination therapy with different pharmacological or nonpharmacological treatments is associated with incremental benefits versus individual components of the combination therapy, and which combinations and sequences of therapy are the most effective.

Pain relief was the most commonly assessed outcome in trials of treatment for low back pain, followed by back-specific function. Trials should more consistently assess other outcomes related to return to work, quality of life, and health care utilization, in order to provide a more complete picture of treatment effects. Studies that evaluate the effectiveness of interventions for preventing future episodes of low back pain would also be very helpful, as low back pain can be a recurrent, episodic condition and these patients are likely to account for a high proportion of resources. In addition, trials should evaluate the effectiveness of interventions for preventing the transition from acute to chronic low back pain. In order to provide balanced assessments of low back pain interventions, trials should more consistently and rigorously evaluate and report harms. Trials should routinely collect information on known or common harms associated with a particular intervention, and use more open-ended methods to identify unexpected or uncommon harms.