The association between early MRI and length of disability in acute lower back pain: a systematic review and narrative synthesis

Background Clinical guideline recommendations are against early magnetic resonance imaging (eMRI) within the first 4 to 6 weeks of conservative management of acute low back pain (LBP) without “clinical suspicion” of serious underlying conditions (red flags). There is some limited evidence that a significant proportion of patients with LBP receive eMRI non- indicated by clinical guidelines, which could be associated with increased length of disability (LOD). The aim of this systematic review was to investigate whether eMRI for acute LBP without red flags is associated with increased LOD. The LOD was defined as the number of disability days (absence from work). Methods Medline, EMBASE, and CINAHL bibliographic databases were searched from inception until June 5, 2021. Two reviewers independently assessed the methodological quality of included studies using the Newcastle–Ottawa scale and extracted data for the review. The search identified 324 records, in which seven studies met the inclusion criteria. Three of the included studies used the same study population. Owing to between-study heterogeneity, a narrative synthesis of results was used. Results All included studies were of good methodological quality and consistently reported that patients with acute LBP without red flags who received eMRI had increased LOD compared to those who did not receive eMRI. Three retrospective cohort studies reported that the eMRI groups had a higher mean LOD than the no eMRI groups ranging from 9.4 days (95% CI 8.5, 10.2) to 13.7 days (95% CI 13.0, 14.5) at the end of 1-year follow-up period. The remaining studies reported that the eMRI groups had a higher hazard ratio of work disability ranging between 1.75 (95% CI 1.23, 2.50) and 3.57 (95% CI 2.33, 5.56) as compared to the no eMRI groups. Conclusion eMRI is associated with increased LOD in patients with acute LBP without red flags. Identifying reasons for performing non-indicated eMRI and addressing them with quality improvement interventions may improve adherence to clinical guidelines and improve disability outcomes among patients with LBP. Supplementary Information The online version contains supplementary material available at 10.1186/s12891-021-04863-9.

It is commonly observed that MRI findings of agerelated degenerative changes are prevalent in people without LBP [14,15]. In addition, a recent study found no relationship between MRI changes in the lumbar spine and pain intensity, health-related quality of life, and depressive and anxiety symptoms among patients with LBP [16]. Furthermore, a systematic review and metaanalysis of imaging strategies for LBP showed that lumbar imaging does not improve clinical outcomes in acute LBP cases without suspected serious underlying conditions [17].
Clinical guidelines for the management of acute nonspecific LBP recommend that imaging, specifically MRI, should not be performed in the first month of conservative management unless red flags (e.g., fracture, tumor, infection, and neurological deficit) are suspected [18][19][20][21]. Despite this, eMRI scanning for patients with acute LBP is common (27.7%; 95% confidence interval (CI) 21.3, 35.1) [22] and was found to be associated with increased LOD, more healthcare utilization, and higher medical costs [10][11][12]. For instance, Mahmud et al., found that eMRI was associated with increased LOD by 102 days (unadjusted 115 vs. 13 days in eMRI and no eMRI groups, respectively) [12], whereas Graves et al., reported that eMRI was associated with an unadjusted 120-day increase in LOD [10]. Undertaking eMRI has been hypothesized to lead healthcare providers to overinterpret the findings and carry out additional and possibly unnecessary interventions, such as surgery, epidural steroid injections physiotherapy, osteopathy, and hospital admission [23][24][25] and thus lead to an increased LOD [12].
With multiple studies showing an independent association between eMRI and increased LOD among patients with acute LBP, it becomes necessary to synthesize the evidence from those studies. To our knowledge, only one systematic review has assessed the relationship between imaging, including MRI, and absence from work in patients with acute LBP [25]. However, that systematic review did not employ a specific timing for MRI scanning for LBP and included only two studies examining the relationship between eMRI and LOD in LBP cases and synthesized the findings using unadjusted LOD estimates between the eMRI and MRI groups. We are aware of more than two studies reporting on this relationship. Therefore, the aim of this systematic review is to summarize the findings of epidemiologic studies examining the relationship between eMRI and LOD in patients with acute LBP without "clinical suspicion" of serious underlying conditions (hereafter referred to as red flags).

Search strategy
The protocol for this review was registered with the International Prospective Register of Systematic Reviews (PROSPERO) under registration number CRD42021259296 (available from https:// www. crd. york. ac. uk/ prosp ero/ displ ay_ record. php? RecordID = 259,296). Reporting of this systematic review was guided by the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement (Supplemental file S1) [26]. We searched Medline, EMBASE, and CINAHL bibliographic databases from their inception until June 5, 2021, using medical subject heading (MeSH) or Emtree and free-text terms on LBP, MRI, and work disability (Supplemental file S2). In addition, reference lists of all relevant papers were searched, and citations of included studies were tracked using the Web of Science Citation Index. No restrictions on language, study design, or time of publication were applied.

Types of studies
All epidemiologic study designs examining the association between eMRI and LOD in patients with acute LBP were considered for inclusion.

Types of participants
Patients with a medical diagnosis of acute LBP, occupational LBP or non-specific LBP were included. Studies including patients with chronic or complicated LBP (e.g., severe injuries, multiple traumas, infection, autoimmune disease, or cancer) were not considered for inclusion in the review.

Types of exposures
The exposure was eMRI defined as an MRI of the lumbar spine for LBP within the first 4 to 6 weeks of the first recorded medical visit for the current LBP episode.

Types of outcome measures
The main outcome was the measure of association between eMRI and LOD whether it was reported as odds ratios, relative risk, or mean difference in LOD between the eMRI group and the no eMRI group. The LOD was defined as the number of disability days (absence from work) due to the current episode of LBP [7,[11][12][13].

Study selection process
All retrieved records were imported to Covidence webbased application and duplicate records were removed. Initially, titles and abstracts of all records were screened, then full text of relevant papers were reviewed for eligibility for inclusion in the review. The study selection process was conducted independently by two reviewers, and any disagreements were resolved by discussion with a third reviewer. The reasons for study exclusions made during the second stage were reported in the PRISMA flow diagram (Fig. 1).

Quality assessment
Methodological quality assessment of included studies was conducted independently by two reviewers using the Newcastle-Ottawa scale for cohort studies (Supplemental file 3) [27] and any disagreements were resolved by discussion with a third reviewer. Where there was a conflict of interest or potential reviewer bias, the reviewer in question was not involved in the quality assessment. This tool assesses the quality of the sample selection process, comparability of cohorts, and the assessment of outcome. Each study can be given a maximum of one star for each element within the sample selection process and the outcome and a maximum of two stars can be given for the comparability section. The sample selection section evaluated the: (1) representativeness of the exposed cohort (representative of populations presenting with acute Reports assessed for eligibility (n = 16) Reports excluded: Included patients with chronic LBP (n = 1). No definition of eMRI was provided and no measure of association was reported (n = 1). Did not examine the association between eMRI and LOD (n = 7).

Identification of studies via databases and registers
Identification Screening Included Fig. 1 The PRISMA flow diagram of studies in the review LBP without red flags and exposed to eMRI scanning), (2) selection of non-exposed cohort, (3) ascertainment of the exposure and (4) demonstration that the outcome was not present at the start of the study. The comparability section evaluated: (1) whether a study adjusted for the most important factors deliberately and (2) whether a study adjusted for other important risk factors. The outcome section evaluated: (1) the method used to assess the outcome, (2) whether the follow-up period was long enough for outcomes to occur and (3) loss to follow up rate. To summarize the risk of bias in each study, we converted the Newcastle-Ottawa scales to the Agency for Healthcare Research and Quality (AHRQ) standards using the following recommended thresholds

Data extraction
The following data were extracted: study aim, source of funding, source of data, methods of data collection, study design, setting, follow up duration, population, number of participants, demographics, definition of LBP, definition of eMRI, definition of LOD, outcomes of association between eMRI and LOD, strengths and limitations, and conclusion. Data extraction was undertaken independently by two reviewers. Any disagreements were resolved by unanimity after involving a third reviewer.
Where there was a conflict of interest or potential reviewer bias, the reviewer in question was not involved in the data extraction. Contacting authors for any missing data was considered. However, all required data was presented in the included papers.

Data analysis
Meta-analysis was considered but owing to betweenstudy heterogeneity in measures of association between eMRI and LOD reported in included studies, formal pooling of the results was not feasible. Therefore, a narrative synthesis of results was conducted. Narrative synthesis was presented as reported in the original study and no additional analysis/synthesis were conducted.

Study characteristics
The characteristics of the included studies are presented in Table 1. All included studies were conducted in the United States (US) and used workers' compensation (WC) administrative databases. Six studies used a retrospective cohort study design [6,7,[11][12][13]37], and one study used a prospective cohort study design [10]. Three studies by Shraim and colleagues [6,7,13] used the same sample to examine the relationship between different individual-level variables (including eMRI) with neighborhood and state-level variables and LOD in LBP cases (see Table 3 for list of variables included in each study). The sample size ranged from 98 to 59,360 with a total number of 64,232 LBP cases in all studies. The proportions of males ranged from 69 to 73%. The mean age of participants ranged between 39.4 and 41.4 years [6,7,11,13,37]. The median age was 34 years in one study [12], and one study included individuals aged 16-61 years but no summary measure of age was provided [10]. All studies included cases with uncomplicated LBP identified using ICD-9 codes [6,7,11,13,37], nature of injury codes [10], or combinations of body part and nature of injury codes [11]. eMRI was defined as lumbar MRI within 30 days [6,7,11,13,37] or 6 weeks of seeking medical care [10]. The LOD was defined as the total number of days of continuous paid indemnity (lost wage replacement for temporary total or temporary partial lost days) and truncated at either 1-year [6,7,11,13,37] or 2-year of follow-up periods [11,37].

Quality assessment
None of the studies examined in the quality assessment stage were excluded. All included studies were of good methodological quality. Six studies scored nine stars [6,7,[11][12][13]37] and one scored eight stars due to 30% loss to follow up [10] ( Table 2). The score given to the representativeness of the exposed cohort was based on the study population which may differ in Number of days of continuous paid indemnity (lost wage replacement for temporary total or temporary) followed by more than 7-day period without indemnity payments, and truncated at the end of the 1-year follow-up period Abbreviations: WC Workers' compensation, SD Standard deviation, ICD International classification of disease, LBP Low back pain, eMRI Early magnetic resonance imaging, LOD Length of disability characteristics of the general population. One of the current review authors (MS) is an author in three of the included studies, therefore, MS was not involved in the quality assessment and any subsequent data extraction of Shraim and colleagues' studies [6,7,13]. A total of 2 out of the 7 included studies had reviewer disagreement in relation to the outcome score of Newcastle-Ottawa scale. This disagreement was resolved by referring the two studies in question to a third reviewer (BA).

The association between eMRI and LOD
All included studies investigated the association between eMRI and LOD. The studies used multivariable analyses and adjusted for potential confounders. The main variables that all studies consistently adjusted for were age and gender (Table 3) (not reported), and 79.6 (105.1) days in the no eMRI group, respectively. One of these studies did not report the SD for the LOD [12]. Two studies [11,37] [11]. Two studies [10,11] reported that eMRI groups with LBP and radiculopathy had a higher HR of increased LOD than the no eMRI groups with LBP and radiculopathy by 2.08 (95% CI 1.67, 2.63) [10] and 3.57 (95% CI 2.33, 5.56) [11]. One study controlled for potential MRI indication bias using the propensity of belonging to the eMRI group, computed based on demographic and severity indicators with adjustment for potential residual confounding of covariates [37]. This study reported that low-propensity eMRI subgroup had a higher HR of increased LOD than the low-propensity no eMRI subgroup and high-propensity no eMRI subgroup by 3.0 (95% CI 2.6, 3.4) and 2.9 (95% CI 2.3, 3.5), respectively [37].

Discussion
This systematic review examined the relationship between eMRI for LBP without red flags and LOD. All included studies showed that subjects who received eMRI for LBP had an increased LOD than those who did not receive eMRI. The findings of our systematic review are consistent with the findings of a previous systematic review of two studies which concluded that patients with acute non-specific LBP who received MRI had a higher LOD as compared to the no MRI group [25]. The current systematic review included 7 studies from 5 unique study populations and added further evidence that eMRI is associated with increased LOD in patients with LBP without red flags even after accounting for several factors associated with LOD in this population. The three studies by Shraim and colleagues used the same study population and found that eMRI was associated with increased LOD in patients presenting with acute LBP without red flags after accounting for neighborhood socio-economic characteristics and state-level variables, including WC policy characteristics [6,7,13]. One study by Graves and colleagues also showed that eMRI was associated with increased LOD in patients with LBP without red flags after accounting for baseline functional disability, pain severity, quality of life, catastrophizing, work-fear avoidance, job accommodation, previous LBP status, job satisfaction, industry, physical demands at work, and type of first medical visit [10]. Shraim 2017 [7] **** ** *** 9 Shraim 2015 [6] **** ** *** 9 Webster 2013 [11] **** ** *** 9 Graves 2012 [10] **** ** ** 8 Webster 2010 [37] **** ** *** 9 Mahmud 2000 [12] **** ** *** 9   Despite recommendations of clinical practice guidelines against eMRI scanning for acute LBP without red flags, significant proportions of patient with LBP receive eMRI [10][11][12][13]37]. The exact reasons for this are not clear. Previous studies hypothesized that lack of adherence to clinical guidelines could be explained by several factors, including patient's demand for diagnostic imaging, patient reassurance by diagnostic findings, concerns about litigation especially in WC settings, physicians' inadequate awareness about the natural history of acute LBP, and inertia of previous experience, or outcome expectancy [11,[37][38][39].
This review used a comprehensive search strategy and searched key bibliographical databases and the grey literature to identify relevant studies. The included studies consisted of large samples of LBP cases and used WC administrative data which captures complete information on medical bills, treatment, interventions, and duration of work disability.
This review has some limitations that should be noted. First, the current review included a small number of studies (7 studies from 5 study populations). Second, the included studies in this review used WC databases as the primary source of data. This data does not provide information on some predictors of LOD, such as level of functional disability, work accommodation, nature of job, fear-avoidance, and other comorbidities, including psychiatric conditions. However, this is unlikely to influence the findings unless the distribution of those predictors differs significantly between the eMRI and no MRI groups. In addition, the study by Graves et al., found that eMRI group had an increased HR of LOD than the eMRI group even after controlling for baseline pain, Roland-Morris disability questionnaire scores, pain intensity, quality of life (role physical, physical functioning, and mental health scores), catastrophizing, work-fear avoidance, offered job accommodation for disability, previous LBP status, job satisfaction, industry, physical demands at work, and type of first medical visit [10]. Third, the included studies measured LOD using wage replacement data. This may underestimate the observed association between eMRI and increased LOD because termination of wage replacement does not necessarily translate to complete recovery or return to work. Fourth, all included studies were conducted in the US, which may limit the generalizability of the findings to other countries that have different healthcare systems. However, these studies have good methodological quality and reported consistent findings related to the review question. Fifth, formal pooling of the results using meta-analysis was not feasible owing to between-study heterogeneity.
More research is needed to uncover the exact reasons for ordering the non-indicated eMRI for acute LBP without red flags. This information is useful for developing interventions and strategies to improve adherence to clinical guidelines' recommendations about the management of patients presenting with acute LBP.

Conclusions
eMRI is associated with an increased LOD in patients with acute LBP without red flags. Further research is needed to fully understand the reasons for the use of non-indicated eMRI for patients presenting with LBP. Developing healthcare interventions to enhance adherence to clinical guidelines may improve disability outcomes among patients with LBP.