Article Text

Download PDFPDF

Harms and benefits of opioids for management of non-surgical acute and chronic low back pain: a systematic review
  1. Hannah-Rose Tucker1,
  2. Katie Scaff1,
  3. Tess McCloud1,
  4. Kimberly Carlomagno1,
  5. Kathleen Daly1,
  6. Alessandra Garcia1,
  7. Chad E Cook2
  1. 1 Doctor of Physical Therapy, Duke University School of Medicine, Durham, North Carolina, USA
  2. 2 Physical Therapy, Duke University, Durham, North Carolina, USA
  1. Correspondence to Hannah-Rose Tucker, Doctoral Student, Duke University Division of Physical Therapy, Durham NC 27708, USA; hannah.rose.tucker{at}


Background Consequences of prescription opioid use involve harms, addiction, tolerance and death. Despite routine prescription, opioids are not recommended for initial intervention by any major multidisciplinary low back pain (LBP) guideline.

Objective Our primary purpose was to improve overall understanding of the harms and benefits associated with oral opioid interventions prescribed for treatment of acute or chronic back pain. Our second goal was to evaluate pain intensity and to compare and contrast these data with the harms. Our last objective was to evaluate conflicts of interest among the study authors and the findings.

Design/data/eligibility criteria Studies incorporating oral prescription opioid management of non-surgical LBP were evaluated. After systematic assessment, no studies that met inclusion included participants with specifically acute LBP. Therefore, extracted data reflects only populations with subacute and chronic LBP. Data on reported harms, severe harms, pain outcomes and withdrawal rates were extracted and meta-analyses were completed for opioid versus placebo trials and opioids versus non-opioid trials.

Results Fourteen studies met inclusion/exclusion requirements. All trials involved short-term management with limited follow-up. A high percentage of harms were identified across most studies. Opioids were not shown to be superior to other medications, and only showed superiority to placebo comparators (at cost of additional harms).

Conclusion This review identified trends of higher harms rates and higher percentages of severe harms in opioid arms for the management of subacute and chronic LBP. The majority of trials that demonstrated benefits with opioids also had potential conflicts of interest. Lastly, non-opioid medications demonstrated statistically significant pain improvement compared with opioids. We feel that the results of the trial are supportive of current LBP guidelines and do not condone the initial use of opioids in management of subacute or chronic LBP.

Trial registration number CRD42017070914

  • back
  • lower back
  • pharmacology
  • spine

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

What is already known?

  • Opioids are commonly prescribed for chronic, non-cancer pain, which has contributed to the current opioid epidemic in the USA.

  • Consensus from multiple guidelines state that opioids should be used sparingly or not at all in the conservative management of low back pain (LBP).

What are the new findings?

  • Higher harms and higher severe harms are associated with the use of opioids when compared with a placebo and non-opioid.

  • Seventy per cent of the studies that favoured the use of opioid for the treatment of LBP in this review demonstrated conflicts of interest that call to question the results of those trials.


In the late 1990s, state medical boards lessened the controls on laws regarding the prescription of opioids for the management of non-cancer pain, consequently, increasing the number of prescriptions to this population.1 This action, in conjunction with rigorous pharmaceutical and clinician marketing as well as the belief that pain is/was undertreated, contributed to what is considered a crisis, especially in North America.1 Today, opioids are commonly prescribed for chronic non-cancer pain.2 According to a report issued by the Substance Abuse and Mental Health Services Administration, in 2010 alone, misuse of prescription drugs led to 1.2 million emergency department visits.3 Remarkably, in the last 20 years, there have been over 100 000 deaths in the USA that resulted from the prescription of opioids.4

According to a study completed by the Global Burden of Disease, low back pain (LBP) was one of two leading causes of disability in 2015 across the globe.5 This disability is highly prevalent and recurrent in nature; the lifetime prevalence is estimated at 84% and approximately 55% of patients will have at least 10 episodes in their lifetime.6 Despite consensus from multiple guidelines that opioids should be used sparingly or not at all in the conservative management of LBP, opioids are commonly prescribed for treatment of acute and chronic conditions.7 In a previous observational trial,8 nearly 20% of the 26 014 eligible patients diagnosed with LBP received long-term opioids in their management process.

A number of narrative reviews and suggested guidelines have evaluated the effect of opioids on LBP versus many different comparators.9–11 One review12 sought to identify the relationship between opioid dose and treatment discontinuation due to adverse events in the management of LBP. However, this study failed to stratify the severity of harms classified between groups. Furthermore, this review did not analyse conflicts of interest (COI) within the study by authors who were employed by drug manufacturers and included studies in which the comparator was an opioid or same class of drug. One review13 looked at long-term trials as well as short-term trials using the following parameters: acute (<4 weeks), subacute (4–12 weeks) and chronic (≥12 weeks). This study explored the value of opioids for non-surgical management of LBP and has found modest effects favouring opioids for chronic LBP.13 The review also incorporated the overview of multiple other pharmaceutical interventions for LBP, was limited to chronic conditions and failed to comprehensively outline detailed specifics of each of the studies. Furthermore, the review did not juxtapose the benefits of opioids on resolution of pain in context of the influence of harms, but in full disclosure, reported that the trials were not designed to assess serious harms. Lastly, the review failed to evaluate potential COI among the studies and the potential these may have had on the results. Consequently, the purposes of this systematic review were to address three primary areas. Our primary purpose was to improve the overall understanding of the harms and benefits associated with oral opioid interventions commonly prescribed for the treatment of acute or chronic LBP. Our second goal was to evaluate pain intensity and to compare and contrast analgesic benefits with the harms. Our third objective was to evaluate COI among the study authors and the findings.


Search strategy

This systematic review was written according to the guidelines set forth in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement for reporting systematic reviews and meta-analyses of studies.14 Following official study registration with PROSPERO in July 2017, a systematic search was conducted in MEDLINE, Embase and Web of Science databases. In September 2018, a second, identical systematic search was conducted in MEDLINE, Embase and Web of Science databases again to ensure that our study encompassed the most relevant literature. To identify appropriate articles, the systematic search was performed in collaboration with a university biomedical librarian trained in systematic review methods. The full search strategy is provided in online supplementary appendix 1.

Supplementary file 1

Study selection

The approach was to find studies comparing oral opioid interventions for acute/subacute and chronic LBP with any other alternative pain modulating management methods including pharmacological treatments as well as conservative treatments. For the studies with pharmacological comparators, we accepted pain modulating methods such as placebo and non-steroidal anti-inflammatory drugs (NSAIDs). We included only randomised controlled trials (RCTs) that reported harms and pain intensity outcomes related to opioid for non-specific acute/subacute or chronic LBP; and that were published in English after the year 2000 in order to capture the effect of opioid prescription post-1990’s regulations. One of our goals was to evaluate pain outcomes (clinical analgesia) and to compare and contrast the benefits of opioids management versus the harms received. For this reason, we chose to include RCTs, as the design enables comparatively evaluated pain.

Subjects were required to be 18 years of age or older and could be previous opioid users or opioid naïve. Exclusion criteria were radicular pain (ie, motor, reflex or sensation deficits), prior surgical intervention, cancer, deformity, instability and infection. We also excluded any trials using oral opioids combined with a second opioid or another active drug. Inclusion and exclusion criteria are available in online supplementary appendix 2.

Supplementary file 2

Selection process

Following the initial database search, duplicates were removed from the initial search results by two independent reviewers (H-RT, KS). Studies were systematically assessed for eligibility based on: 1) title, 2) abstract and 3) full text. Titles and abstracts reviews were conducted by two members of the team (H-RT, KS and TMC, KC respectively), using a third person to resolve disagreement (KD). Lastly, full texts were reviewed by all five members of the team to ensure accuracy and efficiency. RCTs that met the inclusion criteria were then selected for review and quality assessment. Additionally, four group members completed a thorough hand search of the citations of our RCTs and relevant articles. If studies remained unresolved, the senior authors (H-RT, KS) met to discuss inclusion. In October 2018, one author (H-RT) re-assessed and re-evaluated every paper of the initial database search results for clarity. Following the second, updated database search performed in September 2018, two independent reviewers (H-RT, KS) systematically assessed titles, abstracts and full texts, using a third person (CEC) to resolve disagreement.

Interventions and comparators

The details of the opioid interventions (type, dose, morphine equivalent) from each study were extracted and tabulated by one member the team (H-RT). One team member (H-RT) used a standard conversion table to calculate morphine equivalents when needed. The non-opioid comparator specifics were also extracted and tabulated. If a study had more than one comparator only the non-opioid comparator data were reported. For this systematic review, we did not include comparisons between opioid treatments. One study15 had three groups (two opioids and one placebo) and one of the comparisons was between two opioids, so only the data for the comparison with the placebo was captured. Another study16 also had three groups (two opioids and one placebo), but that study compared both opioids with the placebo, therefore we reported that study as A and B. A third study17 had three groups (one opioid, one placebo and one non-opioid analgesic) with treatments that fit our inclusion criteria, therefore, data were extracted for all three groups. The method by which each study accounted for prior opioid use was also extracted (online supplementary appendix 3).

Supplementary file 3

Data extraction and quality assessment

Two independent reviewers (H-RT, KS) extracted data using a standardised data extraction form. The following data were extracted from each of the RCTs and tabulated: study location, patient demographics, study length, duration of follow-up, interventions and comparators, morphine equivalents, pain intensity, harms, serious harms and withdrawal due to harms. Additional data were extracted pertaining to the selection of the participants, inclusion and exclusion criteria and previous opioid use.

Pain intensity was chosen due to the reasoning behind the use and/or prescription of the opioid interventions being pain reduction, because the majority of the reported data being related to pain, and because our study aimed to discuss the relationship of pain intensity (clinical analgesia) and harms. The pain intensity measures used in the trials were numerical rating scale (NRS) score (range 0–10) and visual analogue scale (VAS) score (range 0–100). Harms were defined as a reflection of the negative outcomes of a treatment18 and were extracted and reported from trial phase separately when applicable.

Severe harms were defined as any untoward medical occurrence that at any dose results in death, is life-threatening, requires inpatient hospitalisation or causes prolongation of existing hospitalisation.19 One investigator (TMC) determined the severity of the harms using the previously proposed categorisation of harms by Carlesso.19 These reports were then verified by the other investigators (H-RT, KS, KC, KD). The classification of serious harms was based on the initial categorisation by the authors of the studies, as we did not deem any additional harms as serious or change the classification of those already categorised as serious. For many studies, withdrawal symptoms were not included in the total harms count, in which case, authors (H-RT and AG) extracted instances of withdrawal symptoms and added the numbers to each studies’ harms count. Harms were categorised into the body systems in which they originate or effect (table 2).

Each of the chosen articles were evaluated for bias using the Cochrane Risk of Bias tool.20 The tool is used to assess risk of bias for RCTs. Bias is assessed as a judgement (high, low or unclear) for individual elements from seven domains: sequence generation and allocation concealment (both within the domain of selection bias or allocation bias), blinding of participants and personnel (performance bias), blinding of outcome assessors (detection bias), incomplete outcome data (attrition bias), selective reporting (reporting bias) and an auxiliary domain: ‘other bias’. Each domain was evaluated for each study collectively by H-RT and KS. For each item, the reviewers reached a definitive decision of high, low or unclear in a consensus meeting in which the quality of each article across rating categories was discussed and scoring consensus reached.

Two independent authors (KS and H-RT) evaluated COI in the studies by analysing COI statements of the authors and by independently exploring the research profile of each investigator. We assigned a ‘yes/no’ determination if authors were employed, consulted for or received an honorarium from the opioid manufacturer, or if the study was funded by the pharmaceutical company that manufactured the prescription opioid. These data were collected and reported in the ‘other bias’ domain of the Cochrane tool according to recommendations in the updated guideline for systematics reviews in the Cochrane back and neck group.21

Data synthesis and analysis

Pain intensity outcome was converted to a common 0–100 scale (0, no pain to 100, the worst pain). We expressed pooled effects of continuous variables (pain intensity) as mean differences (MD). We used risk ratios (RR) with 95% CI to calculate categorical variables (harms, serious harms and withdrawing from study due to harms).

For the measurement of effect sizes, we defined three levels: small effect size (MD <10% of the scale; RR <1.25, medium effect size (MD 10%–20% of the scale; RR 1.25–2.0) or large effect size (MD >20% of the scale; RR >2.0).22 The effect was considered clinically important when the magnitude of the effect size was at least medium (>10% of the scale).22

Pain intensity data were extracted for short term (follow-up evaluations <3 months), intermediate term (≥3 months, <12 months) and long term (≥12 months). In the case of multiple comparisons from a single study, we adopted the Cochrane Handbook20 recommendations and divided the number of subjects in the common arm by the number of comparisons. Meta-analysis and risk of bias graphs were carried out using RevMan review management software (V.5.1). Pooled effects were calculated using a random effects model (AG).

To evaluate the overall quality of the evidence, we used the grading recommendations assessment, development and evaluation criteria.23 The quality of evidence was downgraded a level for each of four factors: 1) limitation in the design and implementation (≥25% of trials, weighted by a sample size, had high risk of bias; 2) inconsistency of results (≥25% of trials, weighted by sample size, had results large heterogeneity (I2 >50); 3) imprecision (sample size <400 for each outcome) and 4) publication bias (assessed using funnel plot analysis). We did not assess the indirectness criterion (inability to generalise) in this review because we included the same population and outcome. We downgraded single RCTs for inconsistency and imprecision.

Two reviewers (H-RT and AG) judged whether the four aforementioned factors were present for each outcome. The quality of evidence was defined as ‘high quality’ (further research is unlikely to change confidence in the estimate of effect and there are no known or suspected reporting biases: all domains are fulfilled), ‘moderate quality’ (further research is likely to have an important effect on confidence in the estimate of effect and might change the estimate: one of the domains is not fulfilled), ‘low quality’ (further research is likely to have an important effect on confidence in the estimate of effect and is likely to change the estimate) and ‘very low quality’ (little confidence in the effect estimate).21 24


Search results

The initial database search returned 3388 articles from the included databases. Following the removal of duplicates, 2581 articles remained. From these, 570 titles were selected by two researchers to determine eligibility for inclusion. From the titles, 47 abstracts were chosen. From the abstracts, 11 full-text articles15 16 25–33 were selected for their inclusion criteria and relevance to the topic of prescription of opioids for the treatment of chronic LBP and retrieved for quality assessment in this review. While the inclusion criteria stated that any other non-opioid comparator, including non-pharmacological conservative options, would be included, the database search only provided studies with pharmacological interventions or placebos as comparators. The hand search, completed by all team members, resulted in one additional article.34 The updated search returned 497 articles following duplicate removal. Of these, none of the titles were selected by researchers to determine eligibility for inclusion. The re-evaluation process of the initial database search, complete by one author (H-RT), resulted in two additional studies,17 35 allowing for the conclusion of the study search and the final number of trials being 1415–17 25–35 (figure 1).

Figure 1

Flow of systematic review (adapted from Preferred Reporting Items for Systematic Reviews and Meta-Analyses). LBP, low back pain; RCT, randomised controlled trial.

Patient demographics and study characteristics

A total of 14 trials of opioid analgesics were included in this review. The total number of subjects at the start of each study (including the open-label periods) varied with the lowest number being 139 participants and the highest being 957. Following the completion of the double-blind period, the number of participants ranged from 67 to 636. Only seven studies16 17 25 26 31 33 34 followed up with patients postcompletion. Patient demographics and study characteristics are presented in table 1. All of the included trials were completed during or after the year 2000 up to the year 2015. Ten studies15 26–34 were completed at multidisciplinary and/or educational sites across the USA; one trial16 was performed in the USA, Canada and Australia at multidisciplinary sites; one study25 was completed at the Medical College of India; one trial17 was completed at active investigational sites across Germany and one study35 was completed at hospitals in Beijing, China. Patient demographics were relatively consistent across the 14 studies such that patients were over the age of 18 years and had suffered from moderate-to-severe non-malignant LBP for at least 3 months, with the exception of one study35 which stated that patients had suffered from moderate-to-severe non-malignant LBP for at least 4 weeks. Follow-up times ranged from 4 days to 3 months. None of trials evaluated long-term follow-up (>12 weeks).

Table 1

Participant, location, duration, comparison and morphine equivalents

Interventions and comparators

The following oral opioid medications were administered across the 14 studies: tramadol, tapentadol, morphine, oxymorphone extended release (ER), oxycodone, hydromorphone ER and hydrocodone ER. The most commonly used opioid within these studies was tramadol ranging from 50 to 300 mg. The comparator that was used among the chosen studies was relatively consistent, as 11 studies used unidentified weight-matched placebo pills as the comparator intervention. Two studies25 35 used flupiritine as the comparator, which is a non-opioid analgesic. O’Donnell et al 30 used celecoxib, which is a commonly prescribed anti-inflammatory. Three of the 11 studies with a placebo comparator used a 1:1:1 ratio,16 17 34 meaning that there were three arms to the studies and participants in each arm only received one of the three treatment options; there was no combination of the treatments. Bunyak et al 16 had two comparators, 100–250 mg of tapentadol and 20–50 mg of oxycodone; Vorsanger et al 34 used the same comparator, tramadol, but at two different dosages, 200 and 300 mg and Uberall et al 17 had two different comparators, 400 mg of lupirtine and 200 mg of tramadol. Because of the 1:1:1 randomisation, these studies were each split into ‘A’ and ‘B’ groups in order to collect and analyse data from each arm. Morphine equivalents (ME) were not consistently reported. Doses were given as low as 5 mg ME and up to 320 mg ME. One study15 indicated that doses were not to exceed 660 mg ME. For the meta-analysis, each of the 14 studies were categorised as either ‘opioid versus placebo’ or ‘opioid versus non-opioid’. Details on interventions and comparators are also present in table 1.

Bias assessment

In addition to high variability and inconsistency in reporting doses of oral opioid medications across studies, each of the 14 studies15–17 25–35 were determined to have a high risk of bias in the categories of incomplete outcome data and other sources of bias (figure 2). The high prevalence of incomplete outcome data throughout the studies was a result of large percentages of participants withdrawing from the study. Notable reasons for withdrawal from the studies were harms, lack of efficacy and protocol violation. All 14 studies15–17 25–35 were found to have a high risk of bias under other sources of bias due to the reporting of harms. Common reasons were failing to assess whether the harms were related to the study or failing to have an unbiased third party establish the relationship between the harm and the study.

Figure 2

Risk of bias graph indicating the percentage of bias (low, unclear, high) in each study (A) and risk of bias graph breaking down each category of bias assessed in each study per the guidelines of the Cochrane Risk of Bias tool (B).

Our third objective involved evaluating the potential COI of the studies selected. Twelve of the 14 studies15–17 27–35 included were identified to have potential COI, as these were either funded by the pharmaceutical manufacturer of the opioid or involved investigators who were consultants, employees or received honoraria from the pharmaceutical manufacturer.

Pain intensity results

There was some consistency regarding outcome measures across the chosen studies, as seven studies16 17 27 30 31 33 35 used the NRS-11, six studies15 26 28 29 32 34 used 100 mm VAS and one study25 used both to collect outcome data (table 1). The meta-analysis for pain intensity revealed that there was very low-quality evidence of a small, but not clinically important effect of opioid analgesic for reducing pain at short term (<3 months) compared with placebo (MD −8.98; 95% CI −11.71 to −6.25; 13 trials, n=3071) (figure 3). We downgraded the evidence quality due to limitation in the design (≥25% of trials had high risk of bias); inconsistency (I2 >50%) and publication bias. In order to explore heterogeneity, we temporarily removed two somewhat outlier studies15 28 from the analysis, resulting in a moderate heterogeneity across the 11 remaining studies. It is worth noting that there are often a variety of characteristics (eg, clinical differences, methodological issues)36 that range across studies and can impact the treatment effects.20 We observed that studies which presented higher effect sizes (eg, Hale et al,15 Hale et al,28 Katz et al 29 and Schnitzer et al 32) influenced a higher heterogeneity. When these four studies were temporarily removed from the analysis, we observed a moderate heterogeneity across the nine remaining studies. When compared with the non-opioid intervention, the meta-analysis for pain intensity revealed that there was moderate quality evidence of small effect of the non-opioid for reducing pain at short term (<3 months) compared with the opioid (MD 2.16; 95% CI 0.60 to 3.71; three trials, n=635 (figure 4). Evidence quality was downgraded due to limitation in the design (≥25% of trials had high risk of bias). One of the four studies that used a non-opioid analgesic as a comparator did not provide the mean and SD for pain intensity for the 3-month follow-up period time point and the authors did not respond to efforts to retrieve these data.30

Figure 3

Forest plot for effects of opioid analgesics on short-term pain intensity (comparison opioid vs placebo).

Figure 4

Forest plot for effects of opioid analgesics on short-term pain intensity (comparison opioid vs non-opioid).

Harms results

A meta-analysis was performed for three sets of data: harms, serious harms and treatment discontinuation. Although all 14 studies15–17 25–35 reported harms, the quantity and type of harms that were reported by each study varied vastly, and not all studies differentiated which harms were treatment-related. There was a very low-quality evidence of a medium effect of opioid presenting higher harms at short term compared with placebo (RR 1.42; 95% CI 1.24 to 1.63; 13 trials, n=4048) (figure 5A). We downgraded the evidence quality to very low due to limitation in design (≥25% of trials had risk of bias), inconsistency (I2 >50) and publication bias. Heterogeneity (I2) across opioid versus placebo studies was 74%, but reduced to 24% after removing Hale et al 15 and Chu et al 26 suggesting harms were more homogenous across the 10 remaining studies. We hypothesise that statistical heterogeneity manifests itself in the observed highest harms compared with others. There was a moderate-quality evidence of a medium effect of opioids presenting higher harms at short term compared with the non-opioid intervention (RR 1.50; 95% CI 1.31 to 1.71; four trials, n=1445) (figure 6A). We downgraded the evidence quality to moderate due to limitation in design (≥25% of trials had high risk of bias).

Figure 5

Forest plot for harms (A), serious harms (B) and withdrawing from study due to harms (C) on short term (comparison opioid vs placebo).

Figure 6

Forest plot for harms (A) and for withdrawing due to harms (B) between opioid and non-opioid on short term (comparison opioid vs non-opioid).

Serious harms that led to patients dropping from participation in the trial were highly variable across studies. Researchers in many of the studies failed to specify which serious harms occurred or whether the events were related to the study medication. There was a moderate-quality evidence of a large effect that there were significantly more severe harms in opioid groups versus placebo groups (RR 2.22; 95% CI 1.19 to 4.14; eight trials, n=2558) (figure 5B). The evidence quality was downgraded to moderate quality due to limitation in design (≥25% of trials had high risk of bias). Two of the four studies that used non-opioid analgesics as a comparator did not provide the mean and SD for participants experiencing a severe harm for the 3-month follow-up period time point and the authors did not respond to efforts to retrieve these data.25 30 The remaining two studies that used non-opioid analgesics as a comparator stated that no severe adverse events occurred, therefore, there was no data to be extracted.17 35

There was moderate evidence of a large effect that withdrawal from the study occurred more in the studies using an opioid intervention than in studies using non-opioid interventions (RR 8.28; 95% CI 3.00 to 22.80; three trials, n=1235) (figure 6B). The evidence quality was downgraded due to limitation in design (≥25% of trials had high risk of bias). However, there is very low-quality evidence of a medium effect that there is no difference in the rate of withdrawal between the opioid intervention and placebo (RR 1.43; 95% CI 0.75 to 2.72; 13 trials, n=4048) (figure 5C). The evidence quality was downgraded due to limitation and design (≥25% of trials had risk of bias), inconsistency (I2 >50) and publication bias. Heterogeneity across treatment discontinuation was high at 83%. Interestingly, two studies16 31 found much higher discontinuation in the opioid arms than other studies did. Without these studies, heterogeneity was 0%, but there was also no difference in the RR between opioid and placebo groups. These studies were the only ones to detect difference between groups.

A final meta-analysis was conducted for harms by category considering the variability of the harms, which is reflected in table 2. This harms classification system was adopted from review authors who used the WHO system to assemble harms data into classes by organ system.37 The occurrences of harms ranged in both severity and body system and were divided into seven categories: cardiorespiratory system disorders, dermatology system disorders, gastrointestinal system disorders, genitourinary system disorders, musculoskeletal disorders, nervous system disorders and withdrawal syndrome. The least prevalent harms overall were found in the cardiorespiratory system disorders category, with the exception of erectile dysfunction, urinary tract infection, arthralgia, hypoacusis and sedation. The most prevalent harms overall were headaches and somnolence (nervous system disorders), nausea/vomiting and constipation (gastrointestinal system disorders) and pruritus (dermatology system disorders).

Table 2

Meta-analysis for harms by category


Our goal is to expansively explore each study that has assessed the benefit of opioids for LBP by analysing small details that may influence clinical analgesic outcomes in reported studies. The current opioid epidemic in the USA makes this systematic review both timely and critical for 1) determining risks and benefits of LBP-related opioid interventions, and 2) identifying key questions for future research. While literature exists evaluating the efficacy of opioid analgesics in the management of LBP, our review looked at modern opioid dispensing methods with studies post-2000 to 2018 making this review timely while also taking into consideration the most up-to-date evidence with future clinical implications. Furthermore, our harms stratification by Carlesso’s19 levels allows our study to focus on comparing serious harms between groups in addition to quantifying treatment discontinuation. While we did not investigate the effect of opioid dose on treatment effect, we did evaluate drop outs related to harms or serious adverse events. Furthermore, no study, to our knowledge, has assessed COI as it relates to the drug manufacturers, authorship and the correlated findings. We reviewed the full texts of 50 total RCTs and identified 14 studies15–17 25–35 that corresponded with our inclusion and exclusion criteria. We noticed three major findings in our review. First, while the opioid group, when compared with a placebo comparator, did experience a significant reduction in pain intensity, the non-opioid comparator experienced a significant reduction in pain intensity when compared with the opioid. Second, the opioid arm of studies experienced significantly greater harms than both placebo and non-opioid comparators. Third, a majority (9 of 10) of the studies that reported significant results in favour of the opioid demonstrated COI.

Opioids and harms

Studies that demonstrated a significant reduction in pain with an opioid showed a high prevalence of harms in the open-label stages. In the double-blind period of the opioid versus placebo studies, the incidence of harms ranged from 103 to as high as 3554 occurrences. The majority of the harms experienced were in the gastrointestinal system disorders or nervous system disorders categories, both of which contain the side effects that are often associated with opioids such as dizziness, nausea, vomiting and constipation.38

One prevalent harm was withdrawal syndrome, which is defined as symptoms of varying degree manifested by discontinuation or dose reduction of a psychoactive drug taken repeatedly leading to dependence.37 This symptom that often accompanies the use of an opioid had 917 occurrences across studies. Furthermore, none of the studies followed up long-term, meaning that the withdrawal effects that were accounted for were not further assessed nor were they an accurate depiction of the elongated effects of opioid use. So, one must consider the possibility of increased withdrawal effects over a longer period of time. The meta-analysis regarding harms for the opioid versus placebo comparison revealed that significantly greater harms were associated with the use of opioids compared with the placebo arm. More importantly, the meta-analysis of harms for the opioid versus non-opioid comparison also exhibited greater harms associated with the use of the opioid compared with the non-opioid option. This result calls to question the reasoning behind the prescription of opioids for the treatment of subacute or chronic LBP when the non-opioid options, such as NSAIDs, show significantly fewer harms. This point is further supported in an updated clinical practice guideline, which states that for chronic LBP, NSAIDs should be used along with other conservative approaches such as exercise therapy and psychosocial interventions.39

Unsurprisingly, the opioid group significantly demonstrated an increased number of severe harms compared with the placebo intervention, which led to withdrawal from the studies. For the opioid versus placebo comparison, the meta-analysis did not demonstrate a significant difference for withdrawal rates due to harms between opioids and placebo. However, for the opioid versus non-opioid comparison, the meta-analysis did demonstrate a significant difference for withdrawal rates due to harms, meaning that when compared with a non-opioid, opioids cause a significantly larger number of study withdrawals secondary to harms. This result further supports the initial use of non-opioid analgesic rather than opioid prescribing to minimise risk of harms. It is worth mentioning that we did not qualitatively observe a distinct pattern between trials that used more intense dosing and withdrawal rates.

This lack of significant results for the opioid versus placebo comparison could be contributed to a lack of consistency with reporting harms and withdrawal rates, which highlights the importance and the need for a more comprehensive way for the reporting of harms and withdrawal rates across studies.

Opioids and pain intensity outcomes

Across the majority of the studies, there was a significant improvement in pain intensity scores compared with the placebo favouring the use of opioids. However, there was a wide variation in dosage and in morphine equivalents across studies, as morphine equivalents ranged from 5 to 320 mg, with one study15 simply stating that they did not exceed 660 mg. Furthermore, the frequency of the doses varied markedly, such that there was no pattern seen at all between studies. This raises the argument that there is a lack of standardisation regarding the strength and dosing of opioids. For example, Hale et al 27 reported outcome scores that favoured opioids, but had one of the highest morphine equivalents of all 14 studies (320 mg). Moreover, the meta-analysis for the opioid versus non-opioid comparison demonstrated a significant difference in favour of the non-opioid option in regard to pain intensity. Therefore, non-opioid analgesics present with significantly fewer harms when compared with the opioid option, and provide equivalent pain relief.

Potential conflicts of interest

One last trend that was identified within our review is worth discussing. It is well known that COI can influence the outcome of trials involving spinal surgery40 and pharmaceutical agents.41 In 1215–17 27–35 of the 14 studies (85.7%), pharmaceutical funding from companies that manufactured and distributed the respective opioids was present and was identified as the primary supporter of the trial. In these studies, senior investigators were identified as affiliates of the pharmaceutical manufacturers. We found through published COI statements within the papers that the majority of the investigators held roles as full-time employees, consultants or received an honorarium from the pharmaceutical companies that manufactured the opioids that were investigated (online supplementary appendix 4). Nine15–17 27–29 31 32 34 of these 12 (75%) studies with COI reported pain outcomes favouring the opioid over the placebo comparator with the exception of the study by Uberall et al,17 which reported pain outcomes favouring the opioid over placebo and non-opioid comparators. Collectively, these findings call to question the objectivity of the researchers in their reporting data and in their declared support of the use of opioids for the treatment of subacute or chronic LBP.

Supplementary file 4

Additional findings

In addition to the key findings found in reporting outcome scores and harms, other noteworthy trends were evident. There was dissimilarity among the 14 studies regarding prior opioid use. Three trials15 27 31 required previous use in their inclusion criteria; two studies29 35 excluded patients who were experienced, three studies17 32 34 held a washout period weeks before trial initiation and three studies16 27 33 indicated a dosage requirement that prior users were required to reach before starting the study. There were two studies25 30 that failed to indicate pretrial opioid use. One27 of the three studies15 27 31 that required patients to have prior use of opioids also prescribed the highest morphine equivalent to a portion of patients during the double-blind phase, leading one to question if tolerance was the reason behind such a high dosage. None of the studies provided rationale for their choice in dosage of the opioids. Considering the fact that with the exception of a few minor discrepancies, the demographics and baseline pain intensities of the patients in all 14 studies were comparable, we found the variation in morphine equivalents is perplexing.


Possible limitations of our review include the fact that our literature search was limited to RCTs. We understand that RCTs are not the preferred design for evaluating harms. But, systematic reviews and/or meta-analyses of RCTs have proven to be the preferred method of presenting complete, transparent and reproducible data. Furthermore, RCTs rarely, and poorly, assess and identify harms, which leads to the misunderstanding that the given interventions are safe, which is a fallacy that aligns with the purposes of our study.42 Worth mentioning is the lack of studies including participants with acute LBP. While our intent was to evaluate harms associated with opioid use for both acute and chronic LBP populations, this sample was not represented following a systematic assessment of studies meeting inclusion criteria. We feel this is a limitation of our review as we were unable to capture harms and benefits associated with oral opioid interventions commonly prescribed for the treatment of acute LBP. Aforementioned is this review’s inability to capture the harms and benefits of non-pharmacological pain-modulating methods in comparison to opioid interventions for management of LBP. While the primary aim of this study was to compare opioids with alternative pain management methods, the search results did not provide any RCTs that had non-pharmacological approaches in addition to meeting other inclusion criteria. This necessitates future research implementing non-pharmacological conservative options as the non-opioid comparator being investigated to determine clinical implications. The search was also limited to articles published in the year 2000 and forward and the exclusive use of English language. We also did not include grey literature. Furthermore, the hand search that was completed was only performed on the references of the chosen articles and could have been more comprehensive and representative of the current literature that was not captured in our initial search. Lastly, the homogeneity of these data does not extrapolate to external validity.


Based on the findings from the 14 studies15–17 25–35 that were selected for their inclusion criteria, we feel that one should use caution before administering opioids for the management of subacute or chronic LBP. Higher incidences of harms are present with the use of opioids and pain outcomes do not appear to be superior to comparators such as non-steroidal anti-inflammatory agents. And, while only three studies with low quality reported outcomes that favoured the non-opioid comparators, pain outcomes for the opioid intervention do not appear to be superior to comparators such as non-steroidal anti-inflammatory agents. Furthermore, we feel that there is a risk that findings may be influenced by COI. Moreover, the research lacks long-term follow-up data, which would provide the necessary information to assess the withdrawal effect that has led to this epidemic.


The authors would like to acknowledge the assistance of Leila Ledbetter, biomedical librarian.



  • Contributors CEC generated the study idea. AG led and executed meta-analyses and GRADE. H-RT served as lead author and managed overall organisation and planning, led writing and aided GRADE execution. KS extracted and reported outcome data and researched and reported conflicts of interest. TMC extracted harms and described trends. KC was key in writing methodology and aiding with overall editing. KD tabulated all data and formatted all tables. All authors contributed to the data extraction and writing. All authors approved the final version of the manuscript.

  • Funding CEC is funded by the NIH/DOD/VA UG3/UH3 collaborative AT-17-001 and by the DOD W81XWH-13-PRORP-TRA.

  • Competing interests CEC receives royalties from Book sales and educational material. CEC is a paid editor for JOSPT. All other authors report no competing interests.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement Any unpublished data such as meta-analyses graphs or harms tables can be retrieved from Hannah-Rose Tucker via email at or

  • Patient consent for publication Not required.