Red flags to screen for vertebral fracture in patients presenting with low‐back pain

Notes

Editorial note

See https://www.cochranelibrary.com/cdsr/doi/10.1002/14651858.CD014461.pub2/full for a more recent review that covers this topic and has superseded this review.

Abstract

Background

Low‐back pain (LBP) is a common condition seen in primary care. A principal aim during a clinical examination is to identify patients with a higher likelihood of underlying serious pathology, such as vertebral fracture, who may require additional investigation and specific treatment. All 'evidence‐based' clinical practice guidelines recommend the use of red flags to screen for serious causes of back pain. However, it remains unclear if the diagnostic accuracy of red flags is sufficient to support this recommendation.

Objectives

To assess the diagnostic accuracy of red flags obtained in a clinical history or physical examination to screen for vertebral fracture in patients presenting with LBP.

Search methods

Electronic databases were searched for primary studies between the earliest date and 7 March 2012. Forward and backward citation searching of eligible studies was also conducted.

Selection criteria

Studies were considered if they compared the results of any aspect of the history or test conducted in the physical examination of patients presenting for LBP or examination of the lumbar spine, with a reference standard (diagnostic imaging). The selection criteria were independently applied by two review authors.

Data collection and analysis

Three review authors independently conducted 'Risk of bias' assessment and data extraction. Risk of bias was assessed using the 11‐item QUADAS tool. Characteristics of studies, patients, index tests and reference standards were extracted. Where available, raw data were used to calculate sensitivity and specificity with 95% confidence intervals (CI). Due to the heterogeneity of studies and tests, statistical pooling was not appropriate and the analysis for the review was descriptive only. Likelihood ratios for each test were calculated and used as an indication of clinical usefulness.

Main results

Eight studies set in primary (four), secondary (one) and tertiary care (accident and emergency = three) were included in the review. Overall, the risk of bias of studies was moderate with high risk of selection and verification bias the predominant flaws. Reporting of index and reference tests was poor. The prevalence of vertebral fracture in accident and emergency settings ranged from 6.5% to 11% and in primary care from 0.7% to 4.5%. There were 29 groups of index tests investigated however, only two featured in more than two studies. Descriptive analyses revealed that three red flags in primary care were potentially useful with meaningful positive likelihood ratios (LR+) but mostly imprecise estimates (significant trauma, older age, corticosteroid use; LR+ point estimate ranging 3.42 to 12.85, 3.69 to 9.39, 3.97 to 48.50 respectively). One red flag in tertiary care appeared informative (contusion/abrasion; LR+ 31.09, 95% CI 18.25 to 52.96). The results of combined tests appeared more informative than individual red flags with LR+ estimates generally greater in magnitude and precision.

Authors' conclusions

The available evidence does not support the use of many red flags to specifically screen for vertebral fracture in patients presenting for LBP. Based on evidence from single studies, few individual red flags appear informative as most have poor diagnostic accuracy as indicated by imprecise estimates of likelihood ratios. When combinations of red flags were used the performance appeared to improve. From the limited evidence, the findings give rise to a weak recommendation that a combination of a small subset of red flags may be useful to screen for vertebral fracture. It should also be noted that many red flags have high false positive rates; and if acted upon uncritically there would be consequences for the cost of management and outcomes of patients with LBP. Further research should focus on appropriate sets of red flags and adequate reporting of both index and reference tests.

Plain language summary

Physician use of red flags to screen for fractured vertebrae for patients with new back pain

This review describes the understanding of a common practice for checking for spinal injuries when patients come to a family practice doctor, back pain clinic or emergency room with new back pain. Doctors usually ask a few questions and examine the back to check for the possibility of a spinal fracture. The reason for this check for fractures is that the treatment is different for common back pain and fractures. Fractures are usually diagnosed with an x‐ray, then treated with rest, a back brace and pain relievers. Common back pain is treated with exercise, chiropractic manipulation, and pain relievers; x‐rays, computed tomography (CT) and magnetic resonance imaging scans are not useful for diagnosis. Fractures are rare, being the cause of back pain in the range of 1% to 4.5% of new back pain visits to family doctors.

Eight studies including several thousand patients described 29 different questions and physical exam tests that have been used to look for spinal fractures. Most of the 29 were not accurate. The best four questions asked about use of steroids (which can cause weak bones), the patient’s age (age above 74 increases the risk of fractures) and recent trauma such as a fall. Using a combination of the best questions appears to improve the accuracy. For example, women above age 74 are more likely to have a fracture when they come to the physician complaining of back pain. In the emergency room, the best indication of a spinal fracture was a bruise or scrape on the painful area of the back.

Fractures are rare and generally do not require emergency treatment, even if red flags exist clinicians and patients can watch and wait. During the waiting period, patients should avoid treatments like exercise and manipulation that are not recommended for spinal fractures.

The worst effects of low quality red flag screening are overtreatment and undertreatment. If the tests are not accurate, patients without a fracture may get an x‐ray or CT scan that they don’t need—unnecessary exposure to x‐rays, extra worry for the patient and extra cost. At the other extreme (and much less common), it might be possible to miss a real fracture, and cause the patient to have extra time without the best treatment.

Most of the studies were of low or moderate quality, so more research is needed to identify the best combination of questions and examination methods.

Summary of findings

Summary of findings 1. Results of potentially useful red flags.

Review question: What is the accuracy of red flags to screen for vertebral fracture in patients presenting with low‐back pain or for lumbar examination Patient population: Patients with low‐back pain or requiring examination of the lumbar spine when presenting to care in primary, secondary or tertiary settings Index tests: All relevant features taken during a history or physical examination Target condition: Vertebral fracture Reference standard: Diagnostic imaging (MRI, CT, X‐ray, bone scan) Included studies: Prospective cohorts (4), Retrospective chart reviews (4) Main limitations:Small number of studies included; large heterogeneity between studies and index tests precluded pooling of results; descriptive analysis presented;inadequate reporting of methods
Results: Single red flags
Population	Reference standard	Risk of bias	Sensitivity (95% CI) or range of estimates ^	Specificity (95% CI) or range of estimates ^	False negative rate in a population of 1000 (95% CI)*	False positive rate in a population of 1000 (95% CI)*	LR+ (95% CI)**	LR‐ (95% CI)**
Primary care: 1 study Index test: Age > 70 years	Follow‐up with diagnostic imaging	Moderate (7/11)	0.50 (0.16 to 0.84)	0.96 (0.94 to 0.97)	25 (8 to 42)	38 (29 to 57)	11.19 (5.33 ‐ 23.51)	0.52 (0.26 ‐ 1.05)
Primary care: 2 studies Index test: Age > 74 years	X‐ray, follow‐up with diagnostic imaging	Moderate (6‐7/11)	range: 0.25 to 0.59	range: 0.84 to 0.97	38 (18 to 49)**	29 (19 to 38)**	9.39 (2.69 ‐ 32.75)	0.77 (0.52 ‐ 1.15)
Primary care: 3 studies Index test: Significant trauma	X‐ray, follow‐up with diagnostic imaging	High to moderate (4‐7/11)	range: 0.25 to 0.65	range: 0.90 to 0.98	38 (18 to 49)**	19 (19 to 38)**	10.03 (2.87 ‐ 35.13)	0.77 (0.52 ‐ 1.15)
Primary care: 2 studies Index test: History of corticosteroid use	X‐ray, follow‐up with diagnostic imaging	Moderate (6‐7/11)	range: 0.00 to 0.25	range: 0.99	38 (18 to 49)**	10 (0 to 10)**	48.50 (11.46 ‐ 204.98)	0.75 (0.51 ‐ 1.13)
Tertiary care:1 study Index test: Contusion/abrasion	X‐ray	Moderate (6/11)	0.85 (0.70 to 0.94)	0.97 (0.95 to 0.98)	17 (7 to33)	27 (18 to 45)	31.09 (18.25 ‐ 52.96)	0.15 (0.07 ‐ 0.32)
Results: Combined red flags
Primary care:2 studies Index test: Female and age > 64 years	X‐ray, follow‐up with diagnostic imaging	Moderate (6‐7/11)	range:0.59 to 0.63	range: 0.84 to 0.97	19 (5 to 38)**	38 (29 to 57)**	14.58 (8.0 ‐ 26.61)	0.39 (0.16 ‐ 0.96)
Primary care:2 studies Index test: Female and age > 74 years	X‐ray, follow‐up with diagnostic imaging	Moderate (6‐7/11)	range: 0.25 to 0.45	range: 0.89 to 0.98	38 (18 to 49)**	19 (10 to 19)**	16.17 (4.47 ‐ 58.43)	0.76 (0.51 ‐ 1.14)
Primary care: 1 study Index test: 2 of 4 red flags +ve (Henschke)	Follow‐up with diagnostic imaging	Moderate (7/11)	0.63 (0.24 to 0.91)	0.96 (0.95 to 0.97)	19 (5 to 38)	38 (29 to 48)	15.48 (8.45 ‐ 28.36)	0.39 (0.16 ‐ 0.96)
Secondary care: 1 study Index test: 3 of 5 features +ve (Roman)	X‐ray or CT	Moderate (8/11)	0.76 (0.60 to 0.89)	0.69 (0.66 to 0.71)	12 (6 to 20)	295 (276 to 323)	2.45 (2.02 ‐ 2.97)	0.34 (0.19 ‐ 0.61)
Secondary care: 1 study Index test: 4 of 5 features +ve (Roman)	X‐ray or CT	Moderate (8/11)	0.37 (0.22 to 0.54)	0.96 (0.95 to 0.97)	32 (23 to 39)	38 (29 to 48)	9.62 (5.88 ‐ 15.73)	0.66 (0.52 ‐ 0.84)
Tertiary care:1 study Index test:Trauma & neurological signs	X‐ray	High (5/11)	0.29 (0.04 to 0.71)	0.98 (0.93 to 1)	79 (32 to 106)	18 (0 to 63)	14.43 (2.38 ‐ 87.64)	0.73 (0.46 ‐ 1.15

^Sensitively and specificity are the point estimate and 95% CI when only one study investigated an index test and the point estimate range when multiple studies investigated the same test

* Based on maximum prevalence: 5% primary care and secondary care, 11% tertiary care

** Based on results of study with lowest risk of bias

Henschke combination: Age >70 years, significant trauma, prolonged corticosteroid use, altered sensation from the trunk down

Roman combination: Age >52 years, no leg pain present, BMI ≤ 22, does not exercise regularly, female gender

Background

It is widely agreed that low‐back pain (LBP) can be seriously disabling and imposes an enormous social and economic burden on the community. A common theme presented in clinical practice guidelines is that LBP should be managed in the primary care setting because it is generally benign and the few cases of serious disease can be readily detected with a focused clinical assessment and subsequent diagnostic studies in those suspected of having serious disease (Koes 2010). In the primary care setting, between 1% and 5% of all patients who present with LBP will have a serious spinal pathology which requires further assessment and specific treatment (Chou 2007; Henschke 2009). The most common of these serious spinal pathologies, which may initially manifest as LBP, is vertebral fracture, followed by malignancy, infection, and inflammatory disease.

The identification of serious pathologies is one of the primary purposes of the clinical assessment of LBP patients. Clinical guidelines recommend the awareness of "red flags" as the ideal method to accomplish this purpose (Koes 2010). Red flags are features from the patient's medical history and physical examination which are thought to be associated with a higher risk of serious pathology. The presence of a red flag should alert clinicians to the need for further examination and in most cases, specific management (Waddell 2004). As most clinical guidelines explicitly recommend against the use of routine radiography and diagnostic imaging for patients with LBP, it is important to determine whether red flags can be used to aid a clinician's judgment to screen for serious spinal pathologies.

Target condition being diagnosed

Vertebral fractures are an important indicator of osteoporosis (Grigoryan 2003) and can cause significant pain, deformity, depressive mood, and functional impairments (Edmond 2005; Melton 2003). It is estimated that in Europe, one in eight individuals older than 50 years of age will have a vertebral fracture (Woolf 2003) and worldwide, there were 1.4 million vertebral fractures diagnosed in the year 2000 (Johnell 2006). In patients presenting to primary care with acute LBP, the prevalence of vertebral fractures has been estimated to be between 0.5% (Suarez‐Almazor 1997) and 4% (Chou 2007).

Identification of vertebral fracture is important, not least because it is associated with an increased risk of mortality (Cooper 1993), but that it is also an independent predictor of future fracture at both the spine and the hip (Cauley 2007; Hasserius 2005). According to pooled estimates, the presence of a prevalent (existing or first) vertebral fracture confers a four‐ to five‐fold increased risk of further vertebral fractures (Ensrud 2000), and a three‐fold risk of hip fracture (Delmas 2001; Ensrud 1999).

Despite the burden of vertebral fracture, less than one third of individuals with vertebral fractures receive medical attention and even fewer are treated (Freedman 2008). It has been suggested that only 30% of vertebral fractures are diagnosed in clinical practice because the presentation is similar to that of acute non‐specific LBP and adequate screening tests are not available (Grigoryan 2003; Papaioannou 2002). In addition, there appears to be agreement amongst leading clinicians that spinal manipulative therapy (which is a common treatment for acute LBP endorsed in clinical practice guidelines) is contraindicated in individuals with osteoporosis or vertebral fractures (Maitland 2001; Waddell 2004). Accurate identification of vertebral fracture is clearly essential not only in clinical trials and epidemiological studies, but also for the clinical management of patients with osteoporosis or acute LBP (Ferrar 2005; Hancock 2008).

Index test(s)

As a first step to identifying vertebral fracture in patients presenting with acute LBP, clinical practice guidelines generally recommend assessing for the following red flags: a recent history of trauma; prolonged use of corticosteroids; age greater than 50 years; and obvious structural deformity (Koes 2010). The inclusion of these features in the guidelines has often been poorly justified, for example, by reference to previous guidelines (van Tulder 2004) or unpublished data (Bigos 1994), and less often to primary diagnostic studies (NHMRC 2003). Despite their inclusion in the guidelines, the usefulness of screening for red flags continues to be debated (Underwood 2009) and there remains very little information on their diagnostic accuracy and how best to use them in clinical practice.

The low prevalence of vertebral fractures and other serious spinal pathologies in patients with LBP makes it difficult to develop screening tools which are both easy to apply and accurate. While the guidelines usually suggest individual red flags and leave their interpretation up to the clinicians, arguably more effective screening would be possible if diagnostic accuracy data were available on red flags used alone or in combination with each other. Ideally, an effective series of red flag questions for vertebral fracture would involve readily assessed features from the patient’s history and physical examination, avoiding invasive and potentially harmful tests, to identify all patients who require further assessment.

In 2008, a systematic review of 12 studies evaluated a total of 51 clinical features used to screen patients with LBP for vertebral fracture (Henschke 2008). The review found that five clinical features were useful to raise or lower the probability of vertebral fracture: age > 50 years (positive likelihood ratio [LR+] = 2.2, negative likelihood ratio [LR‐] = 0.34), female gender (LR+ = 2.3, LR‐ = 0.67), major trauma (LR+ = 12.8, LR‐ = 0.37), pain and tenderness (LR+ = 6.7, LR‐ = 0.44), and a painful, distracting injury (LR+ 1.7, LR‐ = 0.78). However, the review noted that the available studies were generally of poor methodological quality and very few studies had been carried out in the primary care setting; which is unfortunate as this is the setting for which guidelines suggest red flags should be used. Most studies evaluated these red flags as isolated diagnostic tests (i.e. not in combination with each other), and most red flags were only evaluated by one study.

Alternative test(s)

In the absence of robust information about the diagnostic accuracy of clinical red flags, clinicians are left with the prospect of routine diagnostic imaging of all patients to identify vertebral fracture. Given the substantial number of patients with LBP who seek care, this would prove to be very expensive and unnecessarily subject many patients to harmful radiation (Chou 2009).

Assessment of standard radiographs remains the "gold standard" for diagnosing vertebral fractures, however, there is no consensus in the literature regarding the definition of vertebral fractures (Grigoryan 2003). Numerous quantitative and semi‐quantitative methods have been developed to assess vertebral fracture from conventional lateral radiographs, but the results are highly dependent on the experience of the observer (Ferrar 2008). For therapeutic trials and epidemiological studies, Genant's semi‐quantitative assessment used by a trained and experienced observer is the preferred method, based on its good reproducibility and ability to differentiate fractures from other deformities. Intra‐observer agreement for this method is 97% (kappa = 0.89) for experienced observers and 93% (kappa = 0.73) for inexperienced observers (Genant 1993).

More recently, quantitative methods of vertebral morphometry have been applied to lateral spine images from dual‐energy X‐ray absorptiometry (DXA) with the intention to improve diagnosis and treatment without increasing cost and exposure to radiation (Damiano 2006; Samelson 2011). The preferred term for vertebral morphometry evaluation using DXA is now "vertebral fracture assessment" (VFA). Image acquisition requires only a few minutes and radiation exposure is only three micro‐Sieverts (mSv), compared to 600 mSv for a lateral radiograph of the thoracic and lumbar spine (Vokes 2006). Numerous sources of error still exist when using VFA (e.g. positioning the measurement points on the vertebrae, anatomic variants, and deformities related to degenerative disease), so standard radiographs are still recommended to confirm the presence of the abnormality and to determine whether it is a fracture or a deformity. VFA has shown good agreement with quantitative morphometry of digitised radiographs (94.8%, kappa = 0.70) (Rea 2000), and consensus qualitative radiograph assessment by two experts (kappa = 0.71) (Ferrar 2000). Lower levels of inter‐observer agreement have been found with VFA (kappa = 0.56 to 0.69) than with Genant’s semi‐quantitative assessment (kappa = 0.60 to 0.86) of standard radiographs (Damiano 2006; Schousboe 2006).

Overall, the preferred method to confirm the presence of vertebral fracture is Genant's semi‐quantitative assessment by a trained and experienced observer. Vertebral fracture assessment can be used to separate normal vertebrae from doubtful or fractured vertebrae, which should then be examined by an expert.

Rationale

We aimed to update a previous systematic review (Henschke 2008) in light of the recent publication of primary diagnostic studies on this topic and the recent development of methods recommended by the Cochrane Diagnostic Test Accuracy Working Group (Deeks 2009). The protocol for this review (Henschke 2010b) was based upon the first Diagnostic Test Accuracy review published with the Cochrane Back Review Group (van der Windt 2010).

The review was performed concurrently with another Cochrane review on the diagnostic test accuracy of red flags to screen for malignancy in patients with LBP (Henschke 2010a), in order to provide evidence on the diagnostic accuracy of red flags to identify serious spinal pathologies presenting as LBP. The results aim to provide researchers and clinicians with a clearer definition of which red flags are useful to screen for vertebral fracture, and identify in which situations it is appropriate to use them in the management of LBP. The potential consequences of a missed or late diagnosis of vertebral fracture necessitate the need for accurate diagnostic information on red flag screening questions.

Objectives

This review aims to provide information on the diagnostic accuracy of tests used to screen for vertebral fracture in patients presenting with LBP or lumbar spine examination. Specifically, the review aims to assess the performance of information collected during the history taking or physical assessment (potential red flags). Such information will assist clinicians in making decisions about the appropriate management course in patients with LBP.

Investigation of sources of heterogeneity

Where feasible, the review will assess the influence of sources of heterogeneity on the diagnostic accuracy of red flags used to screen for vertebral fracture. In particular the influences of study design (e.g. consecutive series or case‐control), healthcare setting (e.g. primary or secondary care), and aspects of study methods as reflected in the items of the QUality Assessment of Diagnostic Accuracy Studies (QUADAS) checklist (Appendix 1), will be assessed.

Methods

Criteria for considering studies for this review

Types of studies

We considered primary diagnostic studies if they compared the results of history taking or physical examination, to identify vertebral fracture in patients with LBP, to those of a reference standard. Cohort and cross‐sectional studies of a consecutive series of patients, as well as case‐control studies, that present sufficient data to allow estimates of diagnostic accuracy (such as sensitivity and specificity) were considered for eligibility in this review. Studies reported in abstracts or conference proceedings and full journal publications, in all languages were considered. Appropriate translation was sought when required.

Participants

Studies evaluating adult patients presenting to care for treatment of LBP or for lumbar spine examination were eligible. Studies with a substantial proportion of recruited patients (> 10%) already diagnosed with vertebral fracture as the likely cause of LBP were excluded. Patients presenting to primary, secondary and tertiary care settings were included in this review. Results of studies carried out in different settings are clearly labelled in text and tables.

Index tests

Studies investigating any aspect of the history taking or physical examination of LBP were considered eligible for inclusion. Such information included demographic characteristics (e.g. age, gender), clinical and medical history (e.g. pain intensity, previous history of falls), and physical examination results (e.g. tenderness, reduced range of motion). Studies where diagnostic accuracy of individual red flags were evaluated in isolation or as part of a combination were included, however, combination red flags required clear description of which tests were involved. "Clinical diagnosis" or "global clinician judgement" (involving some unknown combination of history and physical examination) were excluded as a clinical judgement based on undefined parameters is difficult to interpret or generalise.

Target conditions

Studies that investigated patients presenting with LBP or for lumbar spine examination, and reported the prevalence of vertebral fracture were included.

Reference standards

Studies were included if red flags were compared with diagnostic imaging procedures such as plain radiographs, computed tomography (CT), magnetic resonance imaging (MRI), and bone scans to confirm the presence of vertebral fracture. Clinical follow‐up was also considered suitable if suspected fracture was confirmed with imaging.

Search methods for identification of studies

Electronic searches

The search strategy utilised was developed in collaboration with a medical information specialist. Relevant computerised databases, including MEDLINE, OLDMEDLINE (PudMed), EMBASE (EMBASE.com), and CINAHL (Ebsco) were searched for eligible studies from the earliest year possible until 7th March 2012. The search strategy for MEDLINE, EMBASE and CINAHL are presented in Appendix 2; Appendix 3; and Appendix 4. The previous systematic review on the diagnostic performances of red flags for vertebral fracture was used as a point of reference (Henschke 2008). All publications included in that review are indexed in MEDLINE, and were identified by the proposed search strategy. The strategy used combinations of searches related to the patient population, history taking, physical examination, and the target condition.

Searching other resources

Reference lists of all included publications were checked and forward citation search was conducted on relevant studies. All included studies were subjected to a forward citation check using Web of Science. The electronic search was composed to identify relevant (systematic) reviews in MEDLINE and Medion (www.mediondatabase.nl), from which reference lists were also checked.

Data collection and analysis

Selection of studies

Two review authors (CW and NH) independently applied the selection criteria to all citations (titles and abstracts) identified by the search strategy and excluded citations that were clearly not relevant. Full publications were retrieved for any citation that potentially met the inclusion criteria. Final selection was based on a review of full publications. Disagreements were resolved by consensus or by consulting other review authors in cases of persisting disagreement. The selection criteria and the QUADAS criteria were piloted on selected diagnostic studies to ensure consistency of interpretation among review authors.

Data extraction and management

Two review authors (CW and NH) independently performed data extraction. This included data on the characteristics of studies and participants, types of index tests and reference standards and their descriptions, relevant aspects of the study methods and data for the assessment of diagnostic accuracy (true positive, false positive, true negative and false negative counts for all index tests).

Characteristics of studies (and participants) included details on: setting (type: primary care, secondary care or tertiary care, and country); reason for presentation; enrolment procedures (consecutive or non‐consecutive); number of participants (including number eligible for the study, number enrolled in the study, number receiving the index test and reference standard, number for whom results are reported in the 2 x 2 table, reasons for withdrawal); patient demographics (age, gender) and duration and history of LBP.

Test characteristics included the type of index test; methods of execution; experience and expertise of the assessors; type of reference standard; and where relevant, cut‐off points for diagnosing vertebral fracture (e.g. quantitative radiographic measures).

Assessment of methodological quality

Review authors CW, NH and CM independently assessed the risk of bias of each study using the QUADAS checklist (Whiting 2003). The 11‐item version of the QUADAS recommended by The Cochrane Diagnostic Test Accuracy Working Group (Deeks 2009) was used. The 11 items were considered individually for each study, without the application of weights or the use of a summary score to select studies with certain risk of bias in the analysis. A guideline (Appendix 1) was used to classify each item as "yes" (adequately addressed); "no" (inadequately addressed); or "unclear" (inadequate detail presented to allow a judgment to be made). Disagreements were resolved by consensus.

If a review author was involved in a primary study, then this author was not involved in the risk of bias rating and data extraction for that study.

Statistical analysis and data synthesis

The prevalence of vertebral fracture in each study is presented along with measures of diagnostic accuracy for each test reported. Indices of diagnostic performance were extracted or derived from data presented in each primary study for each red flag or combination of red flags. Diagnostic 2 x 2 tables were generated to calculate sensitivities and specificities with their 95% confidence intervals and are presented in forest plots. In one study (Roman 2010), 2 x 2 tables were reconstructed using information from other relevant parameters (sensitivity, specificity, and predictive values).

Due to the heterogeneity of tests, study settings and methodology, pooling of results was not possible.

The magnitude and precision (95% CI) of likelihood ratios (LRs) for individual studies were inspected to provide a study‐specific judgement of the informativeness of tests. The lowest criterion for a meaningful result for a LR+ was that the lower bound of the 95% CI did not include one. It is common place to consider LR+ between two to five to provide a small increase in post‐test probability, five to 10 a moderate increase and above 10 a large increase. This criteria detailed in Additional Table 2 was used to guide interpretation about informativeness of each test in the review.

1. Criteria for study‐specific judgement of the informativeness of tests based on LR+.

Test result	Interpretation
Upper bound of confidence interval is less than 2	Uninformative test
Lower bound of confidence interval spans 1 but upper bound >2	Indeterminate result because we have an imprecise estimate of the test accuracy
Point estimate in the range 2‐5 and lower bound of confidence interval is above 1.0	Small increase in likelihood of fracture
Point estimate in the range 5‐10 and lower bound of confidence interval is above 1.0	Moderate increase in likelihood of fracture
Point estimate > 10 and lower bound of confidence interval is above 1.0	Large increase in likelihood of fracture

LR+: positive likelihood ratios

All statistics were calculated using Review Manager 5.1.

Investigations of heterogeneity

Factors contributing to heterogeneity could not be rigorously investigated to determine their influence on the diagnostic accuracy of index tests due to the small number of studies that reported similar tests.

Results

Results of the search

Eleven thousand, two hundred and thirty‐six unique titles were identified by the electronic searches with a further 221 identified by handsearching (Figure 1). Only one relevant systematic review was found (written by the current authors) Henschke 2008. After screening titles and abstracts, 60 full‐text articles were retrieved and examined in detail for eligibility. Fifteen studies were considered eligible for inclusion. Forward citation searches were performed on the reference lists of all included studies with no additional studies meeting the inclusion criteria. A further seven studies were excluded as these included patients presenting to accident and emergency (A&E) departments for blunt trauma or multi‐trauma, not low‐back pain (LBP). Overall, the exclusion of studies was mainly the result of irrelevant populations (e.g. reason for presentation was not LBP), reference standards or index test and inadequate methodology (e.g. not diagnostic studies or case series).

1.

Study flow diagram

Description of Studies

Characteristics of the eight included studies (e.g. study design, setting, population, target condition, index test and reference standards) are detailed in the Characteristics of included studies tables.

Of the eight studies included in the review, four were conducted in primary care (Deyo 1986; Henschke 2009; Scavone 1981; van den Bosch 2004) accounting for a total of 4671 participants. The prevalence of fracture ranged from 0.7% to 4.5%. Two studies (Deyo 1986; Henschke 2009) employed prospective sampling of consecutive patients presenting with LBP. The remaining two studies were retrospective chart reviews of patients referred for lumbar imaging, of which one included all patients with imaging studies (Scavone 1981) and the other a random sample of those with imaging for LBP (van den Bosch 2004). There were 20 different red flags identified from the four studies. One study provided data on a combination of four red flags (Henschke 2009). Three studies from primary care described x‐ray as the reference standard (Deyo 1986; Scavone 1981; van den Bosch 2004). For the remaining study (Henschke 2009), the authors confirmed that imaging studies were conducted for suspected serious pathology upon specialist consultation, in the long‐term follow‐up of the study.

There was one study set in secondary care (Roman 2010) which examined the charts of all patients consulted in the spinal surgery clinic (n = 1448) with lumbar disorders. The prevalence of fracture in this study was 3% and was confirmed by standard radiograph or computed tomography (CT). Eight individual tests and five combinations of tests were investigated.

The three studies that took place in tertiary care (an A&E setting) had a combined pool of participants of 1259. The prevalence of fracture ranged for 6.5% to 11%. Only one study enrolled patients presenting with (acute) lumbar pain (Gibson 1992). The remaining two studies enrolled all patients requiring lumbar imaging. One did this prospectively (Reinus 1998) and the other by retrospective chart review (Patrick 1983). There were 12 index tests described in the three A&E studies. A history of trauma (all three studies) and tests of abnormal neurology were the most commonly investigated red flags. The reference standards used were plain radiograph or CT scan.

The specific criterion for a radiological diagnosis of fracture was not well defined in any setting.

Methodological quality of included studies

The individual results of the 'Risk of bias' assessment for included studies are displayed in Figure 2. Most studies included a representative spectrum of patients (63%). The main reason for failing this criterion was enrolling only patients with imaging studies. Since it is possible for patients who present to care to have a positive result to a red flag and not be referred for imaging this was considered as a potential selection bias. Most studies utilised an acceptable reference standard (88%), avoided incorporation of the index tests and reference standards (88%), interpreted index tests when blind to the reference standard (75%) and assessed relevant clinical tests (100%). An acceptable time frame between tests was rarely reported (38%) nor were explanations for withdrawals (38%). No studies specifically reported blinding of index tests when the reference standard was assessed and 75% of studies reported vague or uninterpretable results. Partial verification bias and differential verification bias were only avoided by 38% of studies respectively.

2.

Methodological quality summary: review authors' judgements about each methodological quality item for each included study.

Findings

There were 29 different groups of similar index tests (e.g. trauma = significant trauma, history of trauma, major trauma, direct trauma, or fall; age = age with six different cut‐offs) reported across the eight studies. Sixteen of these groups featured in more than one study and only two similar groups of red flags were reported in more than two studies (trauma = five and sensation change = three). A further 10 tests were different combinations of red flags. Few tests descriptions were reported adequately for clinical use. Calculating pooled estimates of test accuracy for the current review was inappropriate given the heterogeneity of tests reported across the studies and study setting.

Results for all tests included are detailed in the Additional Tables (Additional Table 3; Additional Table 4; Additional Table 5; Additional Table 6; Additional Table 7). Table 1 describes the performance of red flags that were considered potentially useful. The findings presented are from descriptive analyses only.

2. Results of all red flags taken during a history in primary care.

Study	Index test	Reference standard	Sample size	Prevalence	LR+	LR‐
Deyo 1986	Age > 50 years	X‐ray	621	0.045	2.16 (1.58 ‐ 2.95)	0.34 (0.12 ‐ 0.92)
Henschke 2009	Age > 50years	Follow‐up/imaging	1178	0.007	1.84 (1.07 ‐ 3.17)	0.57 (0.23 ‐ 1.39)
van den Bosch 2004	Age > 54 years	X‐ray	2100	0.041	1.72 (1.54 ‐ 1.91)	0.33 (0.20 ‐ 0.53)
Henschke 2009	Age > 54years	Follow‐up/imaging	1178	0.007	2.57 (1.49 ‐ 4.44)	0.50 (0.2 ‐ 1.21)
van den Bosch 2004	Age > 64 years	X‐ray	2100	0.041	2.46 (2.16 ‐ 2.8)	0.32 (0.21 ‐ 0.48)
Henschke 2009	Age > 64 years	Follow‐up/imaging	1178	0.007	7.13 (4.04 ‐ 12.59)	0.41 (0.17 ‐ 1.01)
Henschke 2009	Age > 70 years	Follow‐up/imaging	1178	0.007	11.19 (5.33 ‐ 23.51)	0.52 (0.26 ‐ 1.05)
van den Bosch 2004	Age > 74 years	X‐ray	2100	0.041	3.69 (3.00 ‐ 4.53)	0.49 (0.38 ‐ 0.63)
Henschke 2009	Age > 74 years	Follow‐up/imaging	1178	0.007	9.39 (2.69 ‐ 32.75)	0.77 (0.52 ‐ 1.15)
van den Bosch 2004	Female Gender	X‐ray	2100	0.041	1.26 (1.10 ‐ 1.45)	0.65 (0.46 ‐ 0.92)
van den Bosch 2004	Female > 54 years	X‐ray	2100	0.041	2.01 (1.68 ‐ 2.40)	0.54 (0.41 ‐ 0.72)
Henschke 2009	Female > 54 years	Follow‐up/imaging	1178	0.007	5.39 (3.08 ‐ 9.43)	0.42 (0.17 ‐ 1.04)
van den Bosch 2004	Female > 64 years	X‐ray	2100	0.041	2.75 (2.26 ‐ 3.35)	0.52 (0.40 ‐ 0.68)
Henschke 2009	Female > 64 years	Follow‐up/imaging	1178	0.007	14.59 (8.00 ‐ 26.61)	0.39 (0.16 ‐ 0.96)
van den Bosch 2004	Female > 74 years	X‐ray	2100	0.041	4.14 (3.17 ‐ 5.44)	0.62 (0.51 ‐ 0.75)
Henschke 2009	Female > 74 years	Follow‐up/imaging	1178	0.007	16.17 (4.47 ‐ 58.43)	0.76 (0.51 ‐ 1.14)
Deyo 1986	Significant trauma	X‐ray	621	0.045	3.42 (1.57 ‐ 7.45)	0.72 (0.49 ‐ 1.06)
Henschke 2009	Significant trauma	Follow‐up/imaging	1178	0.007	10.03 (2.87 ‐ 35.13)	0.77 (0.52 ‐ 1.15)
Scavone 1981	Significant trauma	X‐ray	871	0.030	12.85 (8.58 ‐ 19.24)	0.36 (0.22 ‐ 0.62)
Deyo 1986	Corticosteroid use	X‐ray	621	0.045	3.97 (0.20 ‐ 79.15)	0.98 (0.89 ‐ 1.07)
Henschke 2009	Prolonged Corticosteroid use	Follow‐up/imaging	1178	0.007	48.50 (11.46 ‐ 204.98)	0.75 (0.51 ‐ 1.13)

LR+: positive likelihood ratios
LR‐: negative likelihood ratios

3. Results of all red flags taken during a physical examination in primary care.

Study	Index test	Reference standard	Sample size	Prevalence	LR+	LR‐
Henschke 2009	Altered sensation (from trunk down)	Follow‐up/imaging	1178	0.007	3.32 (0.22 ‐ 50.86)	0.96 (0.82 ‐ 1.13)
Scavone 1981	Sensation change	X‐ray	871	0.030	2.21 ( 1.14 ‐ 4.27)	0.83 (0.66 ‐ 1.05)
Scavone 1981	Motor deficit	X‐ray	871	0.030	2.19 (1.06 ‐ 4.54)	0.86 (0.70 ‐ 1.06)
Scavone 1981	DTR abnormality	X‐ray	871	0.030	1.08 (0.37 ‐ 3.18)	0.99 (0.86 ‐ 1.14)
Scavone 1981	Tenderness	X‐ray	871	0.030	0.70 (0.25 ‐ 1.97)	1.11 (0.87 ‐ 1.41)
Scavone 1981	Spasm	X‐ray	871	0.030	1.25 (0.42 ‐ 3.70)	0.98 (0.85 ‐ 1.12)
Scavone 1981	Sciatica	X‐ray	871	0.030	0.42 (0.06 ‐ 2.91)	1.06 (0.98 ‐ 1.15)
Scavone 1981	Hip/leg pain	X‐ray	871	0.030	0.21 (0.01 ‐ 3.35)	1.08 ( 1.02 ‐ 1.14)

DTR: deep tendon reflex
LR+: positive likelihood ratios
LR‐: negative likelihood ratios

4. Results of all red flags in secondary care.

Study	Index test	Reference standard	Sample size	Prevalence	LR+	LR‐
Roman 2010	Age > 52 years	X‐ray or CT	1448	0.026	1.52 (1.42 ‐ 1.68)	0.14 (0.04 ‐ 0.53)
Roman 2010	Female	X‐ray or CT	1448	0.026	1.51 (1.35 ‐ 1.71)	0.26 (0.10 ‐ 0.65)
Roman 2010	Concomitant osteoarthritis	X‐ray or CT	1448	0.026	1.05 (0.76 ‐ 1.45)	0.95 (0.69 ‐ 1.32)
Roman 2010	No regular exercise	X‐ray or CT	1448	0.026	1.47 (1.25 ‐ 1.75)	0.42 (0.21 ‐ 0.81)
Roman 2010	Absense of leg or buttock pain	X‐ray or CT	1448	0.026	2.24 (1.38 ‐ 3.64)	0.80 (0.64 ‐ 0.99)
Roman 2010	BMI < 23	X‐ray or CT	1448	0.026	2.22 (1.44 ‐ 3.42)	0.76 (0.59 ‐ 0.97)
Roman 2010	Descreased pain on sitting	X‐ray or CT	1448	0.026	1.56 (0.94 ‐ 2.59)	0.87 (0.71 ‐ 1.07)
Roman 2010	No gait abnormality	X‐ray or CT	1448	0.026	0.85 (0.68 ‐ 1.08)	1.49 (0.95 ‐ 2.34)

BMI: body mass index
CT: computed tomography
LR+: positive likelihood ratios
LR‐: negative likelihood ratios

5. Results of red flags in tertiary care.

Study	Index test	Reference standard	Sample Size	Prevalence	LR+	LR‐
Gibson 1992	History of direct trauma	X‐ray	225	0.06	1.93 (1.48 ‐ 2.52)	0.121 (0.01 ‐ 1.79)
Patrick 1983	Presenting history of trauma	X‐ray	552	0.07	1.77 (1.48 ‐ 2.13)	0.36 (0.20 ‐ 0.68)
Reinus 1998	History of trauma (fall)	X‐ray	482	0.11	0.18 (0.07 ‐ 0.48)	1.54 (1.38 ‐ 1.71)
Patrick 1983	Sensory deficit	X‐ray	552	0.07	1.42 (0.19 ‐ 10.95)	0.99 (0.94 ‐ 1.04)
Patrick 1983	Motor deficit	X‐ray	552	0.07	1.39 (0.08 ‐ 25.38)	0.98 (0.96 ‐ 1.03)
Patrick 1983	DTR abnormality	X‐ray	552	0.07	1.54 (0.49 ‐ 4.87)	0.97 (0.89 ‐ 1.06)
Gibson 1992	Neurological signs and/or SLR < 40 degrees	X‐ray	225	0.06	2.40 (0.67 ‐ 8.70)	0.81 (0.51 ‐ 1.30)
Reinus 1998	Any motor or sensory dysfunction*	X‐ray	482	0.11	0.69 (0.22 ‐ 2.17)	1.03 (0.96 ‐ 1.10)
Patrick 1983	Positive SLR	X‐ray	552	0.07	1.02 (0.51 ‐ 2.05)	1.00 (0.86 ‐ 1.16)
Patrick 1983	Tenderness	X‐ray	552	0.07	1.76 (1.42 ‐ 2.19)	0.47 (0.28 ‐ 0.78)
Patrick 1983	Spasm	X‐ray	552	0.07	1.47 (0.83 ‐ 2.60)	0.90 (0.75 ‐ 1.09)
Patrick 1983	Contusion/abrasion	X‐ray	552	0.07	31.09 (18.25 ‐ 52.96)	0.15 (0.07 ‐ 0.32)

* Weakness, numbness, or paraesthesias present on testing or questioning
DTR: deep tendon reflex
LR+: positive likelihood ratios
LR‐: negative likelihood ratios
SLR: Straight leg raise

6. Results of combined red flags in primary, secondary and tertiary care.

Study	Index test	Reference standard	Sample Size	Prevalence	LR+	LR‐
Gibson 1992	Trauma & neurological signs	X‐ray	225	0.06	14.43 (2.38 ‐ 87.64)	0.73 (0.46 ‐ 1.15)
Patrick 1983	Multiple findings	X‐ray	552	0.07	2.01 (1.35 ‐ 2.99)	0.73 (0.56 ‐ 0.96)
Henschke 2009	1 of 4 red flags +ve	Follow‐up/imaging	1178	0.007	1.75 (1.34 ‐ 2.29)	0.25 (0.04 ‐ 1.57)
Henschke 2009	2 of 4 red flags +ve	Follow‐up/imaging	1178	0.007	15.48 (8.45 ‐ 28.36)	0.39 (0.16 ‐ 0.96)
Henschke 2009	3 of 4 red flags +ve	Follow‐up/imaging	1178	0.007	906.11 (50.37 ‐ 16299.11)	0.61 (0.36 ‐ 1.03)
Roman 2010	1 of 5 positive +ve	X‐ray or CT	1448	0.026	1.04 (0.98 ‐ 1.10)	0.43 (0.06 ‐ 2.98)
Roman 2010	2 of 5 positive +ve	X‐ray or CT	1448	0.026	1.43 (1.32 ‐ 1.56)	0.16 (0.04 ‐ 0.60)
Roman 2010	3 of 5 positive +ve	X‐ray or CT	1448	0.026	2.45 (2.02 ‐ 2.97)	0.34 (0.19 ‐ 0.61)
Roman 2010	4 of 5 positive +ve	X‐ray or CT	1448	0.026	9.62 (5.88 ‐ 15.73)	0.66 (0.52 ‐ 0.84)
Roman 2010	5 of 5 positive +ve	X‐ray or CT	1448	0.026	7.63 (0.91 ‐ 63.67)	0.98 (0.93 ‐ 1.03)

CT: computed tomography
LR+: positive likelihood ratios
LR‐: negative likelihood ratios

Primary care

Twenty different red flags were reported in four primary care studies (Additional Table 3 and Additional Table 4). 'Significant trauma' was the most common red flag reported (Deyo 1986; Henschke 2009; Scavone 1981) in studies where the prevalence of fracture ranged from 0.7% to 4.5%. The sensitivity of ‘trauma’ as a test ranged from 0.25 (95% confidence interval (CI) 0.03 to 0.65) to 0.65 (95% CI 0.44 to 0.83), the specificity from 0.90 (95% CI 0.86 to 0.93) to 0.98 (95% CI 0.96 to 0.98) and the positive likelihood ratios (LR+) from 3.42 (95% CI 1.57 to 7.45) to 12.85 (95% CI 8.58 to 19.24).

'Older age' at five different cut‐offs was reported by three studies (Deyo 1986; Henschke 2009; van den Bosch 2004). Using raw data from Henschke 2009 allowed four age cut‐offs to be assessed in two studies. Comparing the LR+ at different cut‐offs indicated the most informative cut‐off was 'age greater than 74 years' (9.39, 95% CI 2.69 to 32.75 and 3.69, 95% CI 3.00 to 4.53;Henschke 2009; van den Bosch 2004 respectively). This was a small‐to‐moderate increase in the likelihood of fracture. Combining age and gender revealed more favourable results with increases in likelihood ratios indicating a higher suspicion (moderate‐to‐large) of fracture (e.g. the LR+ for age > 64 increased two‐fold; 7.13, 95% CI 4.04 to 12.59 to 14.59, 95% CI 8.00 to 26.61 when combined with female gender).

Two studies investigated corticosteroid use (Deyo 1986; Henschke 2009). The accuracy was similarly poor in terms of sensitivity (0.00, 95% CI 0.00 to 0.23 and 0.25 95% CI 0.03 to 0.65), however, the test was highly specific (0.99, 95% CI 0.98 to 1.00). This test yielded very different but still imprecise LR+ between the studies (3.97, 95% CI 0.20 to 79.15 and 48.50, 95% CI 11.46 to 204.98 respectively).

One primary care study Henschke 2009 investigated a combination of four red flags (age, gender, corticosteroid use and history of trauma). The presence of at least two positive red flags was highly specific (0.96; 95% CI 0.95 to 0.97) while maintaining reasonable sensitivity (0.63 95% CI 0.24 to 0.91). Any two of the four features resulted in a large increase in the suspicion of fracture (LR+ 15.48, 95% CI 8.45 to 28.36). While a positive test for three red flags yielded a large LR+ (906.11) the confidence interval of this estimate (50.37 to 16,299.10) indicated very sparse data and suggests caution in interpreting this result (Additional Table 7).

Secondary care

One study (Roman 2010) from secondary care investigated eight potential red flags (Additional Table 5). The sensitivity and specificity of the eight tests investigated varied greatly however, the largest LR+ were obtained from the tests; BMI < 23 (2.22, 95% CI 1.44 to 3.42) or the absence of leg/buttock pain (2.24, 95% CI 1.38 to 3.64). These reflected a small increase in the likelihood of fracture. The authors also reported diagnostic accuracy of a combination of tests. Four positive tests (out of five) yielded the best LR+ (9.62, 95% CI 5.88 to 15.73) representing a moderate increase in the likelihood of fracture (Additional Table 7).

Tertiary care (A&E)

The 12 red flags investigated in the three studies set in tertiary care (A&E) revealed varying rates of accuracy (Additional Table 6). For the most common red flag (a history of trauma) sensitivity ranged from 0.07 (95% CI 0.02 to 0.18) to 1.00 (95% CI 0.59 to 1.00) and a specificity from 0.51 (95% CI 0.41 to 0.62) to 0.60 (95% CI 0.56 to 0.65). The LR+ was also inconsistent ranging from 0.18 (95% CI 0.07 to 0.48 ) to 1.93 (95% CI 1.45 to 2.52). One red flag from the physical examination (presence of contusion or abrasion) in an A&E study (Patrick 1983) was superior to all other tests being reasonably sensitive (0.85, 95% CI 0.70 to 0.94) and highly specific (0.97, 95% CI 0.95 to 0.98). The LR+ was similarly high (31.09 95% CI 18.25 to 52.96).

Discussion

Summary of main results

This review aimed to summarise the evidence for the diagnostic accuracy of tests to screen for vertebral fracture in patients presenting with low‐back pain (LBP). Index tests were defined as ‘red flags’ obtained during the history taking or physical examination. The included studies were set in primary, secondary and tertiary care.

Our review noted that from only eight studies 29 different groups of red flags to screen for vertebral fracture have been investigated. Evidence from single studies suggests that many of these red flags are uninformative; the 95% CI of the positive likelihood ratios (LR+) spans 1.0 and most have high rates of false positives. From the five red flags most commonly recommended in guidelines for fracture (osteoporosis, history of trauma, corticosteroid use, older age and female gender), osteoporosis did not feature in any study. Three of the remaining four, trauma, older age and prolonged use of corticosteroids, appeared informative but when used in isolation had modest diagnostic accuracy, considering they are required to detect a condition that has a low prevalence. Combinations of red flags also seemed more informative than red flags used alone. For example, each of the primary care studies reported meaningful LR+ for ‘older age’ with the LR+ increasing from around two for someone in their 50s to around nine for someone in their 70s. Given the low prevalence of fracture, arguably these LRs are too low to move across a decision threshold. However, when female gender and older age were combined the LR+ was higher (e.g.16.17 for age >74 combined with female gender). Similarly, large LR+ were reported for a decision rule based upon four red flags (Henschke 2009: two of four positive, LR+ 15.48, 95% CI 8.45 – 28.36). The findings of this review support the use of a smaller set of red flags to screen for fracture and reveal that certain combinations of red flags are preferable to individual tests. However, these recommendations are limited by the strength of the available evidence; imprecise estimates and sometimes moderate risk of bias are factors of consideration. As there is ambiguity surrounding which red flags should be used to detect certain serious pathology presenting as LBP, caution is required not to dismiss red flags that are appropriate to identify other serious pathologies (e.g. infection, cancer). Because serious pathology presenting as LBP is rare in primary care and many red flags proposed for detecting fracture have high false positive rates, the consequences of uncritical use, in terms of patient outcomes and cost of management from unnecessary ancillary testing, should be considered carefully.

Population and setting

A key objective for all clinicians managing presentations of LBP is to identify underlying serious pathology when it is present. This review included studies from primary, secondary and tertiary care settings. However, it appears that the recommendation to use most red flags to identify vertebral fracture as a cause of LBP, in any of these settings, is not well supported by the available evidence; either because they have poor diagnostic accuracy or are of unknown value. While the review criteria allowed inclusion of patient populations with varying presentations of LBP (e.g. duration of symptoms) few studies were identified. As only a smaller subset of studies reported similar tests and further heterogeneity was invoked by different study methodology or poor reporting of methods, data pooling was not appropriate.

Findings of risk of bias were consistent across setting. The exclusion of patients without imaging posed a risk of selection bias in three of eight studies. Verification bias, commonly introduced by a failure to compare index tests and reference standards in all participants, may have overestimated the accuracy of the tests investigated (Lijmer 1999).

Reference Standard

Adequate reference standards according to the methodological quality criteria (Appendix 1) were observed in most studies (seven out of eight). Plain radiographs were the most common reference standard however, due to poor reporting of how the method was used (e.g. detail about the image view, experience of the observer and definition of vertebral fracture), the influence of these characteristics could not be determined between studies or settings. Likewise, the influence of different methods of imaging on our results could not be investigated. Only one study used CT scan as the reference standard and no study used MRI. One study in primary care used a long‐term clinical follow‐up in conjunction with imaging and specialist review for suspected cases of serious disease. While this is the guideline‐recommended clinical pathway, it is possible that there were missed (undiagnosed) cases of fracture, which would affect the estimates of diagnostic accuracy in this study.

Index Tests

For adequate use in practice, red flags when present should indicate the need for further investigation of serious disease. Conversely, when absent, red flags would reduce the suspicion of serious disease and remove the need for further testing. For managing LBP, these details should improve the efficiency of diagnostic test ordering by reducing unnecessary routine imaging. Based on the small number of studies included in this review, the use of many red flags to screen for fracture is not supported by the available evidence. Because the pre‐test probability is low and most tests have modest LR+, few would raise the index of suspicion of having fracture (rule fracture in) much above the reported prevalence. The sensitivity of index tests was also generally low (below 40%), meaning when fracture is present there is a high rate of missed diagnoses. The implications of undiagnosed fracture are most apparent in populations where there appears to be a higher risk of subsequent fracture (e.g. osteoporosis) Lindsay 2001.The negative likelihood ratios (LR‐) of most tests were not meaningful, despite generally high specificity (above 70%), suggesting that using red flags to decrease the suspicion of fracture (rule fracture out) in practice is also probably not useful. Arguably, the consequences of this in a condition with such a low prevalence are minimal. Importantly, there was a high rate of false positives among the red flags investigated. This finding presents the possibility that the indiscriminate use of red flags to indicate the need for imaging to confirm fracture may contribute to high rates of imaging and costs of LBP management.

While the application of results from individual studies requires caution, there were three red flags that stood out as potentially useful in the primary care setting. These tests (prolonged use of corticosteroids, significant trauma and age > 74 ) taken from a patient’s presenting history, all showed a range of LR+ that appeared meaningful (point estimate: 48.50; 3.42 to 12.85; 3.69 to 9.39 respectively; (Table 1). However, the LR+ calculated from some studies also lacked precision (e.g. corticosteroids 48.50, 95% CI 11.46 to 204.98; trauma 10.03, 95% CI 2.87 to 35.13; age > 74, 9.39, 95% CI 2.69 to 32.75) and probably reflects the low incidence of fracture observed in these studies. Regardless, caution about the application of these results is required.

In the tertiary care environment, opportunity costs of wasted resources may be more apparent than other settings. For this reason, the accuracy of tests to appropriately indicate the need for diagnostic imaging or not may have greater consequences. Unfortunately, in tertiary care, few tests were useful to implicate vertebral fracture or rule it out. 'Contusion/abrasion' was the only informative red flag (LR+ 31.09, 95% CI 18.25 to 52.96; LR‐ 0.15 95% CI 0.07 to 0.32). The effect of age on other red flags is a worthy consideration. For example, assessing the severity of trauma with respect to age may be more relevant where minor trauma in elderly patients may result in significant injury but not in younger patients (Henschke 2009).

Surprisingly, only two of the included studies thoroughly investigated combinations of red flags even though this would most likely reflect how these tools are used in clinical practice. Overall, combination of red flags appeared to produce more favourable results to implicate fracture. For instance Gibson 1992 combined trauma and neurological signs which yielded a LR+ of 14.43 (95% CI 2.38 to 87.64) compared to the best LR+ of 1.93 (95% CI 1.48 to 2.52) for trauma and 2.40 (95% CI 0.67 to 8.70) for neurological signs individually. The combination of female gender and older age improved the detection of fracture at a number of cut‐offs, in two studies. The remaining combinations were reported in only one study. While the precision of some combinations was questionable, the magnitude of many LR+ warrants consideration.

Strengths and weaknesses of the review

This review was modelled on a previous review conducted by contributing authors, and the Cochrane review by van der Windt and colleagues (van der Windt 2010). We used a diagnostic filter in the electronic search and identified all of the papers detected in the original review. Several other eligible publications were also identified and all languages were considered. van der Windt 2010 et al investigated the use of a diagnostic filter and found this did not contribute to missing any relevant citations.

A limitation of the review is the paucity of diagnostic research in the back pain field. There was a large range of tests investigated by individual studies however, the number of studies in which a given index test was investigated was small. Additionally, poor reporting of methods, differing study designs and lack of specific details of the index tests investigated meant that statistical pooling for diagnostic accuracy was not appropriate. For this reason, the analysis was only descriptive and we were not able to investigate the influences of potential sources of heterogeneity on the accuracy of red flags.

The 'Risk of bias' assessment in this review was aided by clear guidelines set a priori (Appendix 1). The majority of studies reported potential biases poorly. Subsequently, many items were scored as “unclear”. Moreover, since most studies were not designed specifically for diagnostic accuracy, the quality of the methods are further brought into question.

Applicability of findings to the review question

Routine imaging is not recommended for LBP but instead is reserved for those cases where serious pathology is suspected (Koes 2010). Despite the low prevalence of serious disease presenting as LBP, in primary care the use of imaging remains high (Williams 2010). Our review potentially explains that pattern of practice as we found a large range of red flags, many with high false positive rates. If practitioners refer any patient with a positive response to any of the 29 red flags identified in this review it is understandable that many patients are referred for further investigations. Since vertebral fracture in primary care is rare, the main implications of this are excessive costs of care, inappropriate diagnostic labelling of benign back pain (Chou 2007) and unnecessary radiation exposure (Flynn 2011), rather then missed diagnoses of fracture. Since there are also potential consequences for not using red flags, for instance increased referrals for imaging because of higher levels of uncertainty, a period of close clinical follow‐up instead of immediate referral for imaging may be useful, in cases where positive red flags indicate vertebral fracture. This approach would maintain caution about interventions that are contraindicated for fracture. A similar view has been proposed by others, whereby patients with suspected vertebral fracture but no neurological symptoms would be subjected to an observation period prior to imaging (Chou 2011).

Many proposed red flags for vertebral fracture are uninformative; others are of unknown value because of imprecision in the estimates of diagnostic accuracy. Of the red flags that were informative most yielded only small increases in the likelihood of fracture when used alone. Because fracture is uncommon, tests that provide at least moderate increases in the likelihood of fracture are required. Combinations of red flags appeared more useful with larger likelihood ratios observed compared to individual tests. These results support the view that recommendations about imaging referral for suspected fracture in patients presenting with LBP should be based on a smaller set of red flags and using certain combinations of red flags are preferable. Considering the quality of evidence available and the descriptive analysis employed from single studies, the strength of these recommendations is weak but follows high false positive rates of many red flags with subsequent consequences for the cost of management and potential adverse outcomes for patients with LBP.

Authors' conclusions

Implications for practice.

The available evidence suggests that the use of many red flags to guide decisions about the need for further investigation of suspected vertebral fracture is unfounded. Most red flags neither increase the likelihood of fracture enough when present, nor decrease its likelihood when absent. The descriptive analyses of single studies revealed three red flags with promising results in primary care (older age, significant trauma and corticosteroid use) and one in tertiary care (contusion/abrasion). Combinations of red flags also appeared more informative to assist clinical decision‐making than using individual tests. Unfortunately clear implications for practice regarding which red flags should be used remains challenging given limitations in the evidence available. Since vertebral fracture is rare in patients presenting with LBP and many red flags have high false positive rates, the current set of recommendations for screening of fracture in clinical practice should be reviewed.

Implications for research.

The lack of primary diagnostic research investigating the accuracy of red flags to detect fracture in patients presenting with LBP is concerning. There is little consistency in the literature regarding which red flags are investigated. The specific details of index tests are also poorly reported. Moreover, many red flags are investigated as individual tests yet these would be rarely applied in this way in a clinical situation. It would be more appropriate to examine multi factorial diagnostic models as considered in clinical reasoning. Consideration should also be given to which red flags are appropriate for fracture in a clinical context and subsequently investigated. Future studies should examine and validate well‐defined sets of index tests that have clinical relevance. As the prevalence of fracture in primary care is low, study design will need to consider large samples to produce robust estimates of diagnostic accuracy.

In the absence of validated screening tests, with appropriate diagnostic accuracy to detect vertebral fracture, enquiry into alternate models of management is warranted to overcome the undesirable effects of unnecessary imaging. For instance, determining the costs, benefits and consequences associated with managing patients presenting with suspicion of fracture by close clinical follow‐up rather than immediate referral may help provide a better direction for future guideline recommendations.

What's new

Date	Event	Description
5 January 2024	Amended	Amended to add Editorial Note: review superseded.

History

Protocol first published: Issue 8, 2010
Review first published: Issue 1, 2013

Date	Event	Description
19 January 2011	Amended	Contact details updated.
3 August 2010	Amended	references updated

Appendices

Appendix 1. Guide to scoring QUADAS Quality Assessment items

Item and Guide to classification

1. Was the spectrum of patients representative of the patients who will receive the test in practice? Is it a selective sample of patients? Classify as ‘yes’ if a consecutive series of patients or a random sample has been selected. Information should be given about setting, in inclusion and exclusion criteria, and preferably number of patients eligible and excluded. If a mixed population of primary and secondary care patients is used: the number of participants from each setting is presented. Classify as ‘no’ if healthy controls are used. Also, score ‘no’ if non‐response is high and selective, or there is clear evidence of selective sampling. Also, score ‘no’ if a population is selected that is otherwise unsuitable, for example, > 10% patients are known to have other specific causes of LBP (severe OA, malignancies, etc). Classify as ‘unclear’ if insufficient information is given on the setting, selection criteria, or selection procedure to make a judgment.

2. Is the reference standard likely to classify the target condition correctly? Classify as ‘yes’ if one of: 1) plain radiography; 2) magnetic resonance imaging (MRI); 3) computed tomography (CT); or 4) other imaging tests such as bone scan; is used as a reference standard. Classify as ‘no’ if you seriously question the methods used, if consensus among observers, or an unknown combination of the clinical assessment (“clinical judgment”) is used as reference standard. Classify as ‘unclear’ if insufficient information is given on the reference standard to make an adequate assessment.

3. Is the time period between the reference standard and the index test short enough to be reasonably sure that the target condition did not change between the two tests? Classify as ‘yes’ if the time period between clinical assessment and the reference standard is one week or less. Classify as ‘no’ if the time period between clinical assessment and the reference standard is longer than one week. Classify as ‘unclear’ if there is insufficient information on the time period between index tests and reference standard.

4. Did the whole sample or a random selection of the sample receive verification using a reference standard of diagnosis? Classify as ‘yes’ if it is clear that all patients who received the index test went on to receive a reference standard, even if the reference standard is not the same for all patients. Classify as ‘no’ if not all patients who received the index test received verification by a reference standard. Classify as ‘unclear’ if insufficient information is provided to assess this item.

5. Did patients receive the same reference standard regardless of the index test result? Classify as ‘yes’ if it is clear that all patients receiving the index test are subjected to the same reference standard. Classify as ‘no’ if different reference standards are used. Classify as ‘unclear’ if insufficient information is provided to assess this item.

6. Was the reference standard independent of the index test (i.e. the index test did not form part of the reference standard)? Classify as ‘yes’ if the index test is not part of the reference standard. Classify as ‘no’ if the index test is clearly part of the reference standard. Classify as ‘unclear’ if insufficient information is provided to assess this item.

7. Were the reference standard results interpreted without knowledge of the results of the index test? Classify as ‘yes’ if the results of the reference standard are interpreted blind to the results of the index tests. Also, classify as ‘yes’ if the sequence of testing is always the same (i.e. the reference standard is always performed first, followed by the index test) and consequently, the reference standard is interpreted blind of the index test. Classify as ‘no’ if the assessor is aware of the results of the index test. Classify as ‘unclear’ if insufficient information is given on independent or blind assessment of the index test.

8. Were the index test results interpreted without knowledge of the results of the reference standard? Classify as ‘yes’ if the results of the index test are interpreted blind to the results of the reference test. Also, classify as ‘yes’ if the sequence of testing is always the same (i.e. the index test is always performed first, followed by the reference standard), and consequently, the index test is interpreted blind of the reference standard. Classify as ‘no’ if the assessor is aware of the results of the reference standard. Classify as ‘unclear’ if insufficient information is given on independent or blind assessment of the reference standard.

9. Were the same clinical data available when the index test results were interpreted as would be available when the test is used in practice? Classify as ‘yes’ if clinical data (i.e. patient history, other physical tests) would normally be available when the test results are interpreted and similar data are available in the study. Also, classify as ‘yes’ if clinical data would normally not be available when the test results are interpreted and these data are also not available in the study. Classify as ‘no’ if this is not the case, e.g. if other test results are available that cannot be regarded as part of routine care. Classify as ‘unclear’ if the paper does not explain which clinical information was available at the time of assessment.

10. Were uninterpretable / intermediate test results reported? Classify as ‘yes’ if all test results are reported for all patients, including uninterpretable, indeterminate, or intermediate results. Also, classify as ‘yes’ if the authors do not report any uninterpretable, indeterminate, or intermediate results AND the results are reported for all patients who were described as having been entered into the study. Classify as ‘no’ if you think that such results occurred, but have not been reported. Classify as ‘unclear’ if it is unclear whether all results have been reported.

11. Were withdrawals from the study explained? Classify as ‘yes’ if it is clear what happens to all patients who entered the study (all patients are accounted for, preferably in a flow chart). Also, classify as ‘yes’ if the authors do not report any withdrawals AND if the results are available for all patients who were reported to have been entered in the study. Classify as ‘no’ if it is clear that not all patients who were entered completed the study (received both index test and reference standard), and not all patients are accounted for. Classify as ‘unclear’ when the paper does not clearly describe whether or not all patients completed all tests, and are included in the analysis.

Appendix 2. MEDLINE search strategy

1 Index test: clinical red flags

"Medical History Taking"[mesh] OR history[tw] OR "red flag"[tw] OR "red flags" OR Physical examination[mesh] OR "physical examination"[tw] OR "function test"[tw] OR "physical test"[tw] OR ((clinical[tw] OR clinically[tw]) AND (diagnosis[tw] OR sign[tw] OR signs[tw] OR significance[tw] OR symptom*[tw] OR parameter*[tw] OR assessment[tw] OR finding*[tw] OR evaluat*[tw] OR indication*[tw] OR examination*[tw])) OR (ra[sh] OR ri[sh])

OR "Wounds and Injuries"[mesh] OR trauma[tw] OR injury[tw] OR "Accidental Falls"[mesh]

2. Population:low‐back pain and anatomical location

(back pain[mesh] OR sciatica[mesh] OR "back ache"[tw] OR backache[tw] OR "back pain"[tw] OR dorsalgia[tw] OR lumbago[tw] OR sciatica[tw] OR Pain[mesh] OR pain[tw] OR ache*[tw] OR aching[tw] OR complaint*[tw] OR dysfunction*[tw] OR disabil*[tw] OR neuralgia[tw]) AND (Back[mesh] OR spine[mesh] OR back[ti] OR lowback[tw] OR lumbar[tw] OR lumba*[tw] OR lumbo*[tw] OR sciatic*[tw] OR ischia*[tw] OR sacroilia*[tw] OR spine[tw] OR spinal[tw] OR radicular[tw] OR "nerve root"[tw] OR "nerve roots"[tw] OR disk[tw] OR disc[tw] OR disks[tw] OR discs[tw] OR vertebra*[tw] OR intervertebra*[tw] OR Sacroiliac‐joint[mesh] OR Lumbar vertebrae[mesh])

3. Target condition: vertebral fracture

Fractures, Bone[mesh] OR Fractures, stress[mesh] OR Fractures, Spontaneous[mesh] OR Fractures, Compression [mesh] OR Fractures, Closed [mesh] OR fracture*[tw] OR Spinal Injuries [mesh] OR Spinal Diseases [mesh] OR Lumbar vertebrae [mesh]

4. Exclusion criteria: children, case reports, animal studies

(exp Child [mesh] OR exp Infant [mesh]) NOT ((exp Child [mesh] OR exp Infant [mesh]) AND (exp Adult [mesh] OR Adolescent [mesh])) OR (Animals [mesh] NOT (Animals [mesh] AND Humans [mesh])) OR “case report”[ti]

Search combination

1 AND 2 AND 3 NOT 4

Appendix 3. EMBASE search strategy

1 Index test: clinical red flags

'medical history taking'/exp OR 'history'/de OR history OR 'red flag' OR 'red flags' OR 'physical examination'/exp OR 'physical examination' OR 'function test'/de OR 'function test' OR 'physical test' OR (clinical OR clinically AND ('diagnosis'/de OR sign OR signs OR significance OR symptom$ OR parameter$ OR assessment OR finding$ OR evaluat$ OR indication$ OR examination$)) OR 'radiography'/exp OR 'radionuclide'/exp AND [humans]/lim

2. Population:low‐back pain and anatomical location

back AND 'pain'/exp OR 'back pain' OR 'low back' AND 'pain'/exp OR 'low back pain' OR 'sciatica'/exp OR sciatica OR backache OR coccyx OR coccydynia OR dorsalgia OR 'lumbar pain' OR spondylosis OR lumbago AND [humans]/lim

3. Target condition: vertebral fracture

'fractures, bone'/exp OR 'fractures, stress'/exp OR 'fractures, spontaneous'/exp OR 'fractures, compression'/exp OR 'fractures, closed'/exp OR fracture$ OR 'spinal injuries'/exp OR 'spinal diseases'/exp OR 'wounds and injuries'/exp OR trauma$ OR injury AND [humans]/lim

4. Exclusion criteria: case reports, animal studies

'case report' AND [humans]/lim

Search combination

1 AND 2 AND 3 NOT 4

Appendix 4. CINAHL search strategy

1 Index test: clinical red flags

MH "Patient History Taking" or TX history or TX "red flag" or MM “Physical examination” or TX "physical examination" or TX "physical test" or TX clinical* or MH "Diagnostic Tests, Routine" and (TX diagnosis or TX sign or TX signs or TX significance or TX symptom* or TX parameter* or TX assessment or TX finding* or TX evaluat* or TX indication* or TX examination*)

2. Population:low‐back pain and anatomical location

MH “Back Pain” or MH “Low back pain” or TX “back pain” or TX “low back pain” or MM Sciatica or TX sciatica or TX Backache or TX Coccyx or TX Coccydynia or TX Dorsalgia or TX lumbar pain or TX spondylosis TX lumbago

3. Target condition: vertebral fracture

MH "Fractures, Bone" or MH "Fractures, stress" or MH "Fractures, Spontaneous" or MH "Fractures, Compression" or MH "Fractures, Closed" or MH "Fractures, Vertebral Compression" or TX fracture* or MH "Spinal Injuries" or MH "Spinal Diseases" or MH "Wounds and Injuries” or TX trauma or TX injury

Search combination

1 and 2 and 3

Data

Presented below are all the data for all of the tests entered into the review.

Tests. Data tables by test.

Test	No. of studies	No. of participants
1 Age > 50 years	2	1483
2 Age > 54 years	2	3179
3 Age > 52 years	1	1449
4 Age >64 years	2	3179
5 Age > 70	1	1172
6 Age >74 years	2	3179
7 Gender	2	3455
8 Female > 54 years	2	3179
9 Female > 64 years	2	3182
10 Female > 74 years	2	3179
11 Trauma	6	3496
13 Corticosteroid use	2	1483
14 OA	1	1448
15 No regular exercise	1	1448
16 Sensation change	3	2595
17 Motor deficit	2	1424
18 DTR abnormality	2	1423
19 Neurological signs	2	590
20 SLR	1	552
21 Tenderness	2	1423
22 Spasm	2	1423
23 Contusion/abrasion	1	552
24 Sciatica	1	871
25 Hip/leg pain	1	871
26 Absence of buttock/leg pain	1	1448
27 BMI< 23	1	1448
28 Decreased pain on sitting	1	1448
29 No gait abnormality	1	1448
30 Trauma and neurological signs	1	108
31 Multiple findings	1	552
32 Henschke rule 1 sign positive	1	1172
33 Henschke rule 2 positive signs	1	1172
34 Henschke rule 3 positive signs	1	1172
35 1 of 5 positive ‐ Roman Cluster	1	1449
36 2 of 5 positive ‐ Roman Cluster	1	1449
37 3 of 5 positive ‐ Roman Cluster	1	1448
38 4 of 5 positive ‐ Roman Cluster	1	1448
39 5 of 5 positive ‐ Roman Cluster	1	1448

1. Test.

Age > 50 years

2. Test.

Age > 54 years

3. Test.

Age > 52 years

4. Test.

Age >64 years

5. Test.

Age > 70

6. Test.

Age >74 years

7. Test.

Gender

8. Test.

Female > 54 years

9. Test.

Female > 64 years

10. Test.

Female > 74 years

11. Test.

Trauma

13. Test.

Corticosteroid use

14. Test.

OA

15. Test.

No regular exercise

16. Test.

Sensation change

17. Test.

Motor deficit

18. Test.

DTR abnormality

19. Test.

Neurological signs

20. Test.

SLR

21. Test.

Tenderness

22. Test.

Spasm

23. Test.

Contusion/abrasion

24. Test.

Sciatica

25. Test.

Hip/leg pain

26. Test.

Absence of buttock/leg pain

27. Test.

BMI< 23

28. Test.

Decreased pain on sitting

29. Test.

No gait abnormality

30. Test.

Trauma and neurological signs

31. Test.

Multiple findings

32. Test.

Henschke rule 1 sign positive

33. Test.

Henschke rule 2 positive signs

34. Test.

Henschke rule 3 positive signs

35. Test.

1 of 5 positive ‐ Roman Cluster

36. Test.

2 of 5 positive ‐ Roman Cluster

37. Test.

3 of 5 positive ‐ Roman Cluster

38. Test.

4 of 5 positive ‐ Roman Cluster

39. Test.

5 of 5 positive ‐ Roman Cluster

Characteristics of studies

Characteristics of included studies [ordered by study ID]

Deyo 1986.

Study characteristics
Clinical features and settings	Primary care, USA ‐ Sampling method unclear ‐ Only patients with main complaint of back pain considered ‐ Selective criteria used for ordering x‐ray
Participants	621 patients presenting with LBP, 311 received imaging Exclusions: maximal pain above T12, evidence of urinary tract disease, women of age less than 45 who did not practice contraception and whose last menstrual period occurred greater than 10 days previous, and participants of a concurrent study.
Study design	Prospective cohort
Target condition and reference standard(s)	Fracture or malignancy: X‐ray ‐ anteroposterior and lateral lumbar views
Index and comparator tests	Significant trauma, Age > 50 years, Corticosteroid use
Follow‐up	All radiology reports obtained and medical records reviewed
Notes	Prevalence of vertebral fracture 4.5%
Methodological quality
Item	Authors' judgement	Support for judgement
Representative spectrum? All tests	Yes	Consecutive series of patients with LBP
Acceptable reference standard? All tests	Yes	X‐ray ‐ anteroposterior and lateral lumbar views
Acceptable delay between tests? All tests	Yes	84% of reference test obtained on the day of the index test or within 6 days thereafter
Partial verification avoided? All tests	No	Only 311 of 621 received the x‐ray reference test
Differential verification avoided? All tests	Unclear	Unclear from text
Incorporation avoided? All tests	Yes	X‐ray not part of index tests
Reference standard results blinded? All tests	No	Radiologists were aware of the index tests
Index test results blinded? All tests	Yes	Index tests performed before reference standard
Relevant clinical information? All tests	Yes	Index tests available in usual care
Uninterpretable results reported? All tests	No	Possible fracture in patients not receiving the reference test
Withdrawals explained? All tests	Unclear	Unclear from text

Gibson 1992.

Study characteristics
Clinical features and settings	Accident and Emergency Department, UK ‐ Consecutive sample over 6‐month period
Participants	225 patients presenting with pain in the lumbar region of less than 48 hours duration 108 (48%) had radiographs
Study design	Prospective cohort
Target condition and reference standard(s)	Vertebral fracture: Plain radiograph ‐ not further defined
Index and comparator tests	Trauma, trauma and neurological signs, neurological signs
Follow‐up
Notes	Prevalence of vertebral fracture was 6.5%
Methodological quality
Item	Authors' judgement	Support for judgement
Representative spectrum? All tests	Yes	Consecutive series of patients with pain in the lumbar region ‐ number excluded not reported
Acceptable reference standard? All tests	Yes	Plain radiography
Acceptable delay between tests? All tests	Unclear	No time frame specified
Partial verification avoided? All tests	No	Only 48% of sample received reference standard
Differential verification avoided? All tests	Unclear	Unclear from text
Incorporation avoided? All tests	Yes	Questionnaire (index test) not part of reference standard
Reference standard results blinded? All tests	No	Treating doctor involved in radiological reporting
Index test results blinded? All tests	Yes	Index tests performed prior to reference standard
Relevant clinical information? All tests	Yes	Index tests available in usual care
Uninterpretable results reported? All tests	No	More than half of the patients did not receive imaging
Withdrawals explained? All tests	Unclear	Unclear from results

Henschke 2009.

Study characteristics
Clinical features and settings	Primary care, Australia Consecutive sample from general practice, physiotherapy or chiropractic
Participants	1,172 participants presenting with acute LBP, mean age 43.97 years, 53.4% male, 72.6% from physiotherapy, 59.4% less than 1 week since onset
Study design	Prospective cohort
Target condition and reference standard(s)	Serious spinal pathology including vertebral fracture: 12‐month clinical follow‐up with suspected cases confirmed by imaging studies and specialist review
Index and comparator tests	25 red flags ‐ 5 specific to vertebral fracture; age > 70, significant trauma, prolonged use of corticosteroids, altered sensation, clinician judgement
Follow‐up
Notes	Prevalence of fracture was 0.68%
Methodological quality
Item	Authors' judgement	Support for judgement
Representative spectrum? All tests	Yes	Consecutive sample of patients with LBP with clear inclusion criteria
Acceptable reference standard? All tests	No	Long‐term follow‐up of all patients with only those suspected of serious pathology referred for specialist appointment and imaging (i.e., asymptomatic fracture could go undetected)
Acceptable delay between tests? All tests	Unclear	Unclear from text
Partial verification avoided? All tests	Yes	All patients had long‐term follow‐up
Differential verification avoided? All tests	Yes	All patients received the reference standard
Incorporation avoided? All tests	Unclear	Unclear from text
Reference standard results blinded? All tests	Unclear	Unclear from text
Index test results blinded? All tests	Yes	Index test completed prior to reference standard
Relevant clinical information? All tests	Yes	Index tests available in usual care
Uninterpretable results reported? All tests	Yes	All results reported
Withdrawals explained? All tests	Yes	All participants completed follow‐up

Patrick 1983.

Study characteristics
Clinical features and settings	Accident and Emergency Department and University Hospital Medical Centre, USA ‐ Consecutive sample of patients with imaging studies ordered over three‐month period
Participants	552 patients with lumbar spine x‐ray referral, 54% male, age range 6 to 95 years, 99% complained of LBP
Study design	Retrospective chart review
Target condition and reference standard(s)	Lumbar spine bony injury: X‐ray (not further defined)
Index and comparator tests	Trauma, tenderness, LBP with radiation and/or hip pain, spasm, Sensory deficit, motor deficit, tendon reflex abnormality, positive straight leg raise, contusion/abrasion, multiple findings (distinction between fracture and other bony injury unclear for some index tests)
Follow‐up
Notes	Prevalence of fracture was 7.2%
Methodological quality
Item	Authors' judgement	Support for judgement
Representative spectrum? All tests	Unclear	Consecutive sample of with lumbar spine imaging requested ‐ reason for presentation not described
Acceptable reference standard? All tests	Yes	X‐ray
Acceptable delay between tests? All tests	Yes	At time of admission
Partial verification avoided? All tests	Yes	All had reference standard
Differential verification avoided? All tests	Yes	X‐ray
Incorporation avoided? All tests	Yes	X‐ray not part of clinical examination
Reference standard results blinded? All tests	Unclear	Unclear from text
Index test results blinded? All tests	Unclear	Unclear from text
Relevant clinical information? All tests	Yes	Index tests available in usual care
Uninterpretable results reported? All tests	No	Distinction between fracture and other bony injury was unclear for some index tests
Withdrawals explained? All tests	Unclear	Unclear from text

Reinus 1998.

Study characteristics
Clinical features and settings	Accident and Emergency Department, USA ‐ Consecutive sample of all patients receiving lumbosacral radiographs in a 14‐month period
Participants	482 patients, 35% male, age range 17‐98 years, 92% with back pain
Study design	Prospective cohort
Target condition and reference standard(s)	Suspected clinical diagnosis including fracture and spondylosis: Lumbosacral AP, lateral, bi‐lateral posterior oblique and coned down radiological views
Index and comparator tests	Trauma, neurological deficit
Follow‐up
Notes	Prevalence of fracture was 11% (24 intermediate age, 21 chronic, 10 acute)
Methodological quality
Item	Authors' judgement	Support for judgement
Representative spectrum? All tests	Unclear	Consecutive sample of patients with lumbosacral imaging ‐ reason for presentation not described
Acceptable reference standard? All tests	Yes	Lumbosacral AP, lateral, bi‐lateral posterior oblique and coned down radiological views
Acceptable delay between tests? All tests	Unclear	Not time specified
Partial verification avoided? All tests	Unclear	Unclear from text
Differential verification avoided? All tests	Unclear	Unclear from text
Incorporation avoided? All tests	Yes	Radiological diagnosis not part of clinical examination
Reference standard results blinded? All tests	No	Radiologists aware of clinical history
Index test results blinded? All tests	Yes	Clinical examination prior to radiograph
Relevant clinical information? All tests	Yes	Index tests available in usual care
Uninterpretable results reported? All tests	No	Ambiguous reporting of index test in patients with confirmed intermediate and chronic fracture
Withdrawals explained? All tests	Yes	All results reported

Roman 2010.

Study characteristics
Clinical features and settings	University Hospital Spine Clinic, USA ‐Consecutive sample in 5 year period
Participants	1448 patients with lumbar‐related disorders, 41% male
Study design	Retrospective chart review
Target condition and reference standard(s)	Compression fracture or wedge deformity: Standard radiograph or CT assessing sagittal alignment, vertebral body compression and spinal canal dimensions
Index and comparator tests	Leg or buttock pain, gender, age, BMI, gait abnormality, regular exercise, sitting pain, osteoarthritis, multiple signs
Follow‐up
Notes	Prevalence of fracture was 3%
Methodological quality
Item	Authors' judgement	Support for judgement
Representative spectrum? All tests	Yes	Consecutive sample of patients with lumbar disorders
Acceptable reference standard? All tests	Yes	Radiograph or CT
Acceptable delay between tests? All tests	Yes	Part of standard clinical assessment
Partial verification avoided? All tests	Yes	Part of standard clinical assessment
Differential verification avoided? All tests	No	X‐ray or CT
Incorporation avoided? All tests	Yes	Imaging was not part of clinical examination
Reference standard results blinded? All tests	Unclear	Unclear from text
Index test results blinded? All tests	Unclear	Unclear from text
Relevant clinical information? All tests	Yes	Index tests available in usual care
Uninterpretable results reported? All tests	Yes	Results for all participants were reported
Withdrawals explained? All tests	Yes	Results for all participants were reported

Scavone 1981.

Study characteristics
Clinical features and settings	University teaching hospital medical centre, USA ‐ Sampling: all patients with lumbar spine x‐ray
Participants	871 patients in 12‐month period, 176 were inpatients and 695 outpatients
Study design	Retrospective chart review
Target condition and reference standard(s)	Abnormal radiological finding including fracture: AP and lateral x‐ray views
Index and comparator tests	Major trauma, minor trauma, tenderness, LBP with radiation, hip/leg pain, muscle spasm, neurological deficits, sciatica, abnormal physical examination
Follow‐up
Notes	Prevalence of fracture was 3%
Methodological quality
Item	Authors' judgement	Support for judgement
Representative spectrum? All tests	Unclear	Selection criteria not clear
Acceptable reference standard? All tests	Yes	X‐ray: AP and lateral views
Acceptable delay between tests? All tests	Unclear	No time frame specified
Partial verification avoided? All tests	Unclear	Unclear from text
Differential verification avoided? All tests	Unclear	Unclear from text
Incorporation avoided? All tests	Yes	X‐ray not part of clinical examination
Reference standard results blinded? All tests	Unclear	Unclear from text
Index test results blinded? All tests	Yes	Radiologists not aware of clinical symptoms
Relevant clinical information? All tests	Yes	Index tests available in usual care
Uninterpretable results reported? All tests	No	Reporting of fracture prevalence was ambiguous
Withdrawals explained? All tests	No	History for 57 persons could not be obtained

van den Bosch 2004.

Study characteristics
Clinical features and settings	University hospital medical centre, UK ‐ Sampling: random sample of 2100 lumbar spine radiology reports from 6269 patients referred for imaging by general practitioners
Participants	2007 patients with full radiographic and demographic details, 42% were men, mean age was 49.9 for men and 56.7 for women
Study design	Retrospective chart review
Target condition and reference standard(s)	Serious radiological findings including fracture X‐ray (not further defined)
Index and comparator tests	Age, gender
Follow‐up
Notes	Prevalence of vertebral fracture was 4.1%
Methodological quality
Item	Authors' judgement	Support for judgement
Representative spectrum? All tests	Yes	Random sample of patients with LBP referred for imaging by general practitioners
Acceptable reference standard? All tests	Yes	X‐ray
Acceptable delay between tests? All tests	Unclear	Unclear from text
Partial verification avoided? All tests	No	X‐ray reports of 93 patients not available
Differential verification avoided? All tests	Yes	X‐ray only
Incorporation avoided? All tests	Yes	Index test not part of reference test
Reference standard results blinded? All tests	Unclear	Unclear from text
Index test results blinded? All tests	Yes	Index test performed prior to reference standard
Relevant clinical information? All tests	Yes	Index tests available in usual care
Uninterpretable results reported? All tests	No	93 patients with no records available
Withdrawals explained? All tests	No	93 patients with no records available

AP: anterior‐posterior
BMI: body mass index
CT: computed tomography
LBP: low‐back pain

Characteristics of excluded studies [ordered by study ID]

Study	Reason for exclusion
Durham 1995	Patients presented with blunt trauma not LBP.
Frankel 1994	Patients presented with blunt trauma not LBP.
Gestring 2002	Patients presented with blunt trauma not LBP.
Holmes 2003	Patients presented with blunt trauma not LBP.
Hsu 2003	Patients presented with blunt and multi‐trauma not LBP.
Samuels 1993	Patients presented with blunt trauma not LBP.
Terregino 1995	Patients presented with blunt trauma not LBP.

LBP: low‐back pain

Differences between protocol and review

Due to the limited number of index tests evaluated in studies and the heterogeneity in study setting, meta‐analyses were not performed.

Contributions of authors

All review authors contributed to discussions regarding the conception and design of the study. All review authors read and approved the final manuscript.

Sources of support

Internal sources

• Vrije Universiteit, EMGO+ Institute for Health and Care Research, Netherlands

• University of Sydney, Australia

• ErasmusMC, University Medical Centre, Netherlands

• The George Institute for Global Health, Australia

External sources

• Dutch Health Care Insurance Board, Netherlands

• National Health and Medical Research Council, Australia

Declarations of interest

No conflicts of interest are declared.

Edited (no change to conclusions)