Will This Patient Develop Persistent Disabling Low Back Pain?

A 48-year-old woman is evaluated in clinic with a 3-day history of low back pain without leg pain. She has no previous history of cancer and no weight loss, anorexia, or night sweats. Her physical examination reveals mild paralumbar tenderness with normal strength, sensation, and lower extremity reflexes. She has not worked for 3 days due to the back pain. She does not recall any specific work-related injury. She rates the pain as 8 out of 10 and reports little improvement with over-the-counter acetaminophen.

As her physician, you suspect acute nonspecific low back pain. You encourage her to remain active and prescribe nonsteroidal anti-inflammatory drugs. The patient states she is worried about her ability to return to work. She is avoiding many of her usual activities and has stopped doing her daily 2-mile walk due to the pain and fear of making her back worse. She also has a history of chronic depression. Will this patient develop chronic disabling low back pain?

Low back pain is extremely common and costly. It is the fifth most frequent reason for office visits in the United States, accounting for approximately 2% of all visits.^{1 - 2} Among patients who see a health care professional, studies show that pain and function typically improve substantially in the first month.³ Indeed, most patients with acute low back pain do not go on to develop chronic disabling symptoms. Common imaging changes (eg, degenerative disk disease or bulging disks) are only weakly correlated with presence of symptoms,^{4 - 5} so most primary care patients with low back pain (approximately 85%) have pain that is termed nonspecific.^{4 ,6} Patients who develop chronic disabling low back pain account for a markedly disproportionate share of the costs associated with low back pain.^{7 - 8}

An earlier Rational Clinical Examination systematic review⁹ and published guidelines¹⁰ on the history and physical examination for acute low back pain suggest an approach that (1) assesses for risk factors (red flags), suggesting a serious, specific underlying condition (such as cancer, infection, or compression fracture); (2) determines the presence and degree of neurological compromise; and (3) identifies findings associated with prolonged or delayed recovery.¹¹ The latter have come to be known as yellow flags. In the absence of yellow flags, clinicians can provide patients with informed reassurance of quick recovery; however, their presence affects the frequency and intensity of follow-up and choice of interventions. This article does not update the earlier Rational Clinical Examination systematic review, which focused on the value of the history and physical examination for identifying specific conditions, but instead we synthesized the current literature on the yellow flags that might identify patients at risk for chronic disabling low back pain.

Assessment of yellow flags usually focuses on psychosocial factors, including presence of preexisting psychological conditions, maladaptive coping strategies (such as avoiding usual or recommended activities because of fears that they will harm the back or hinder recovery), lower socioeconomic or educational status, poor job satisfaction, higher physical work demands, poor general health or functional level, tobacco use, obesity, receipt of workers' compensation or disability/sick leave, and unresolved litigation or compensation issues related to the back pain.^{12 - 13} Sex and age have also frequently been evaluated as predictors of worse outcomes. The presence of Waddell signs on physical examination may indicate a nonorganic or psychological component to low back pain (Table 1).¹⁴ Some of these items and others have been combined in risk assessment instruments.^{9 ,15 - 16}

We searched MEDLINE using Ovid (1966-January 2010) and EMBASE (1974-February 2010) for relevant studies (eTable 1) and also reviewed reference lists of retrieved articles. We included English-language studies of adult patients with fewer than 8 weeks of low back pain that prospectively evaluated the prognostic accuracy of individual risk factors or risk prediction instruments for persistent, disabling low back pain (eMethods). We assessed study quality using 8 criteria for assessment of diagnostic¹⁷ or prognostic¹⁸ studies (eTable 2). For studies of risk assessment instruments, we also evaluated whether the study evaluated diagnostic test performance in a population other than the one used to derive the instrument (external validation).¹⁹

We categorized potential yellow flags into 1 of 16 risk factor domains (eTable 3). These domains were based on potential yellow flags commonly evaluated in prognosis studies or found to be predictive in other systematic reviews. We excluded potential yellow flags that could not be categorized into 1 of these domains.

For each study, we used the diagti procedure in Stata version 10 (StataCorp, College Station, Texas) to calculate sensitivities, specificities, and likelihood ratios (LRs)²⁰ with 95% confidence intervals (CIs), based on outcomes at 3 to 6 months and at 1 year. In our review, the positive LR is the odds of developing chronic disabling low back pain among patients with the risk factor present.²¹ The negative LR is the odds of developing chronic disabling low back pain among patients without the risk factor. For baseline pain, impairment in function, and maladaptive pain coping behaviors, which were measured using various numerical or ordinal scales, we categorized results as high (eg, score ≥7 on a 0-10 scale), moderate (eg, score 4-6), or low (eg, score <4). We dichotomized other risk factors that were measured using numerical or ordinal scales.

Because of variability in the thresholds, definitions of risk factors, populations, and outcomes assessed, we summarized results with median LRs and total range. The total range, rather than the interquartile range, was chosen because several prognostic factors were reported in a few studies and because the summary range highlights the greater uncertainty we have in the estimates. To evaluate the usefulness of individual risk factors and risk prediction instruments, we considered the median LR value, the variability of estimates across studies, and whether the LR range crossed 1.

We performed stratified analyses according to whether studies evaluated workers' compensation or other populations to determine if and how results differed. We also performed stratified analyses of studies that focused on outcomes related to return to work vs nonwork outcomes, whether they did or did not meet various quality criteria, longer (>1 year) vs shorter duration of follow-up, and less than 4 weeks vs 4 to 8 weeks' duration of low back pain symptoms.

Our search yielded 11 841 citations (eFigure). A total of 197 articles potentially met initial inclusion criteria based on review of titles and abstracts. After review of the full-text articles, 20 studies of 10 842 patients met inclusion criteria (eTable 4). Fourteen studies (reported in 16 articles^{22 - 37} ) evaluated individual risk factors and 10 studies evaluated risk prediction instruments.^{15 - 16 ,34 ,38 - 44} In 4 cases, data from the same population were used to evaluate both individual risk factors,^{23 ,26 - 27 ,34 ,36} as well as a risk prediction instrument.^{34 ,40 - 42} We obtained unpublished data to calculate LRs for 6 studies.^{23 ,29 ,33 ,35 ,37 ,42} Duration of follow-up ranged between 3 months and 2 years.

The degree to which studies met the quality criteria varied (eTable 5). No study met all criteria. The most common methodological shortcomings were failure to describe assessment of outcomes blinded to assessment of potential yellow flags (only 1 study met this criterion^{36 - 37} ) and failure to assess interventions received following initial evaluation (4 studies met this criterion^{22 ,31 - 32 ,39} ).

One-third of the studies evaluated patients in workers' compensation settings^{15 ,25 ,35 - 37 ,41 ,43 - 44} and two-thirds in clinical care settings (primary care,^{16 ,22 - 24 ,26 - 27 ,29 ,31 - 34 ,39 - 40 ,42} physical therapy,²⁸ or specialty clinics^{30 ,38} ). Most of the studies used work disability status as the primary outcome,^{15 - 16 ,23 - 25 ,34 - 44} with the other studies evaluating pain,^{30 ,32 - 33} function,^{26 - 27 ,29} overall satisfaction with condition,²² or mixed outcomes.^{28 ,31 ,34 ,39}

The probability of a poor outcome depends on how poor outcomes are defined, the patient setting, and duration of follow-up. In studies conducted in primary care settings that focused on work absenteeism or compensation status, the median (range) proportion of patients with a poor outcome was 11% (2%-20%) at 3 to 6 months (4 studies^{16 ,39 ,42 ,45} ), as well as at 1 year (11% [9%-13%]; 2 studies^{24 ,42} ). In studies conducted in primary care settings that focused on pain, functional status, or mixed outcomes, the median (range) proportion of patients with a poor outcome at 3 to 6 months was 26% (2%-48%) for 6 studies^{31 - 34 ,39 ,45} and at 1 year was 21% (7%-42%) for 6 studies.^{22 ,27 ,29 ,31 - 32 ,34 ,46} Including studies conducted in workers' compensation or referral settings, the median (range) proportion of patients with worse work-related outcomes at 3 to 6 months was 19% (2%-42%) for 11 studies^{15 - 16 ,23 ,25 ,35 - 36 ,38 - 39 ,42 - 45} and at 1 year was 13% (9%-18%) for 4 studies.^{23 - 24 ,36 ,42} In studies that focused on pain, functional status, or mixed outcomes (9 studies^{26 ,28 ,30 - 34 ,39 ,45} ), the median (range) proportion of patients with a poor outcome at 3 to 6 months was 35% (2%-48%); all of the studies reporting these outcomes at 1 year were conducted in primary care settings.

Demographic and Work-Related Features. Table 2 and eTable 6 show the demographic characteristics and work-related features for predicting chronic disabling low back pain. Age, sex, education level, smoking status (described as either “ever smoked” or “current smoker”), and overweight (defined by body mass index, calculated as weight in kilograms divided by height in meters squared) consistently failed to predict worse outcomes, with LRs approaching 1 in most or all studies. Receiving compensation at baseline was associated with slightly increased likelihood of worse outcomes at 1 year (median [range] LR for receiving compensation, 1.4 [1.2-1.8]). Higher work dissatisfaction and higher physical work demands did not predict worse outcomes at 3 months, but did at 1 year (median [range] positive LR: 1.5 [1.3-1.8] and 1.4 [1.2-1.7], respectively).

Health Status at Onset of Back Pain. Table 3 and eTable 7 show the accuracy of general health, psychiatric comorbidities, and prior low back pain episodes for predicting chronic disabling low back pain. Quiz Ref IDWorse general health status^{22 - 23 ,25 ,29 - 30 ,34 ,37} before the onset of pain was associated with worse outcomes at 3 to 6 months (median [range] LR: 1.6 [1.1-1.7] for lower general health status; 0.73 [0.66-0.88] for better health status) and at 1 year (median [range]: positive LR, 1.8 [1.1-2.0]; negative LR, 0.85 [0.56-0.99]). The presence of higher scores for current psychiatric comorbidity (measured using various scales) had an effect somewhat stronger than poorer overall general health at 3 to 6 months (median [range] LR: 1.9 [1.4-2.1] for higher psychiatric comorbidity; 0.69 [0.55-0.85] for lower psychiatric comorbidity) and at 1 year (median [range]: positive LR, 2.2 [1.9-2.3]; negative LR, 0.85 [0.55-0.93]).^{22 - 23 ,25 ,29 ,35 - 37} A history of previous or recurrent episodes of low back pain was not useful for predicting worse outcomes at 3 to 6 months or 1 year, with LRs approaching 1.

Signs and Symptoms. Table 4 and eTable 8 show the accuracy of signs (nonorganic signs and baseline function) and symptoms (baseline pain, radiculopathy, and fear avoidance or pain coping behavior) for predicting chronic disabling low back pain.

Baseline Pain. High pain intensity predicted worse outcomes at 3 to 6 months (median [range] LR, 1.7 [1.1-3.7]), but was a less useful predictor at 1 year (median [range] LR, 1.3 [1.2-2.0]). Low pain intensity was associated with a broad range of LRs at both 3 to 6 months (median [range] LR, 0.70 [0.07-0.86]) and at 1 year (median [range] LR, 0.33 [0.08-0.97]).

Radiculopathy. The presence of radiculopathy or leg pain slightly increased the odds of worse outcomes at 3 to 6 months (median [range] LR, 1.4 [1.1-1.7]) and at 1 year (1.4 [1.2-2.4]); however, its absence had a median LR slightly less than 1 at 3 to 6 months (0.63 [0.52-0.93]) and at 1 year (0.82 [0.54-0.94]).

Fear Avoidance or Pain Coping Behavior. Quiz Ref IDMaladaptive pain coping behaviors include fear avoidance (avoidance of work, movement, or other activities due to fear that they will damage or worsen the back) and catastrophizing (pain coping characterized by excessively negative thoughts and statements about the future⁴⁷ ). Patients with high maladaptive coping behaviors, which were measured by several scales, such as the Fear-avoidance Beliefs Questionnaire,⁴⁸ were more likely to have worse outcomes at 3 to 6 months (median [range] LR, 2.2 [1.5-4.9]) and at 1 year (2.5 [2.2-2.8]). Patients in the low category were less likely to have worse outcomes at 3 to 6 months (median [range] LR, 0.46 [0.30-0.73]) and at 1 year (0.39 [0.38-0.40]).

Nonorganic Signs. Nonorganic signs refer to findings that suggest a strong psychological component to pain, or intentionally false or exaggerated pain symptoms. Higher somatization scores predict failure to return to work at 3 months (LR, 2.5; 95% CI, 1.8-3.4).²³ Similarly, higher somatization scores or more generalized pain at baseline increase the likelihood of a worse outcome at 1 year (median [range] LR, 3.0 [1.7-4.6]), but the usefulness of lower scores was variable (median [range] LR, 0.71 [0.31-0.76]). No study reported LRs for Waddell signs, possibly the best known of the nonorganic signs.

Baseline Function. Baseline functional impairment was measured with the Roland Morris Disability Questionnaire,^{23 ,35 - 37 ,49} the Oswestry Disability Index,^{25 ,50} and various ordinal scales.^{29 - 30} Results using the various measures could be categorized into 3 levels of functional impairment (none or weak, moderate, and severe or extreme). Baseline functional impairment showed increasing likelihood of poor outcomes from the highest functional impairment at 3 to 6 months (median [range] LR, 1.4 [1.3-3.5]) and at 1 year (2.1 [1.2-2.7]) to the lowest functional impairment at 3 to 6 months (0.53 [0.18-1.07]) and at 1 year (0.40 [0.10-0.52]). The Roland Morris Disability Questionnaire was the most frequently used measure of function.^{23 ,33 ,35 - 37}

Stratified Analyses. Conclusions appeared similar in subgroups of studies stratified according to whether they evaluated a workers' compensation or non−workers' compensation population, whether they evaluated return-to-work vs a nonwork functional outcome, and whether they evaluated patients with acute or subacute low back pain. There was no pattern suggesting that results of studies meeting various quality criteria differed from those that did not meet the criteria. In the 1 study that reported LR estimates for 2-year outcomes,²³ results were similar to estimates based on 1-year outcomes.

eTable 9 shows the accuracy of risk prediction instruments for predicting chronic disabling low back pain.

Vermont Disability Prediction Questionnaire. In a derivation study,³⁸ an instrument developed at the Vermont Rehabilitation Engineering Center was highly useful for predicting nonreturn to work at 6 months (positive LR, 8.3; 95% CI, 5.2-13.2; and negative LR, 0.17; 95% CI, 0.10-0.31), based on a cutoff score of 0.5 (0-1 scale). A modified version of this instrument, the 11-item Vermont Disability Prediction Questionnaire (VDPQ), was evaluated in 1 derivation study¹⁵ and 1 validation study.⁴³ The derivation study found higher scores on the VDPQ increased the likelihood of not returning to work at 3 months (LR, 5.7; 95% CI, 3.9-8.5 at a cutoff of 0.48; and LR, 7.1; 95% CI, 3.7-13.4 at a cutoff of 0.65)¹⁵ ; however, in the validation study,⁴³ the VDPQ was not as useful, with the exception of very high (≥0.76) scores (LR, 7.4; 95% CI, 2.8-19.7). Several items in the VDPQ did not predict outcomes in other studies (previous back problems and physician visits for back pain) or have not been studied well.

Acute Low Back Pain Screening Questionnaire. The 21-item, patient self-administered Acute Low Back Pain Screening Questionnaire (ALBPSQ)⁵¹ was evaluated in 2 studies.^{16 ,42} The ALBPSQ increased the likelihood of more than 30 days off work through 6 months at various cutoffs (LRs, 2.2-3.9),¹⁶ and the results were validated using a Norwegian version (LR, 4.8; 95% CI, 2.0-11.4 for a cutoff score of at least 105; and LR, 7.7; 95% CI, 2.8-21.3 for a cutoff score of at least 112).⁴² In this same study,⁴² the ALBPSQ was less useful for predicting days off work through 12 months (LRs, 2.1-2.3 at various cutoffs, with CIs approaching or crossing 1).

Other Risk Prediction Instruments. Three other risk prediction instruments were evaluated in derivation studies. Deyo and Diehl³⁹ found that the presence of negative responses to all 3 items of an instrument (previous episodes, feel sick all the time, and education level) to be more useful for predicting no improvement in Sickness Impact Profile scores (LR, 0.10; 95% CI, 0.01-0.71) than 1 to 3 positive responses (LRs, 0.72-2.93). Fulton-Kehoe et al⁴¹ evaluated a different 3-item instrument (each scored as 1 point: pain interference with activities ≥5 on a 0-10 scale, not working for pay in last week, and presence of leg pain) and found a score of zero (LR, 0.01; 95% CI, 0.00-0.10) to be more useful than scores of 1 to 3 (LRs, 0.29-3.42) for predicting receipt of wage compensation at 1 year. Thomas et al³⁴ evaluated a 6-item instrument (each scored as 1 point: female sex, dissatisfaction with employment situation, history of low back pain, radiating leg pain, widespread pain, ≥2 restrictions in spinal movement) and found both high (5 or 6) and low (0-2) scores predicted the likelihood of persistent disabling low back pain through 12 months (LR, 5.3; 95% CI, 2.6-10.8 and LR, 0.15; 95% CI, 0.05-0.46; respectively).

A validation study that evaluated failure to return to work at 2 years found that a clinical prediction algorithm for return to work was a somewhat weaker predictor of failure to return to work compared with the other instruments (positive LR, 2.0; 95% CI, 1.6-2.4; and negative LR, 0.41; 95% CI, 0.28-0.60).⁴⁰ A study of an 8-factor model found patients scoring in higher percentiles on a predictive score had progressively greater likelihood of receiving benefits for more than 3 months, but a pragmatic method of administering and scoring was not provided.⁴⁴

Studies differed in how they defined or analyzed risk factors in different domains and in how they defined worse low back pain outcomes, which could introduce clinical heterogeneity. In addition, many studies on risk of developing persistent low back pain were excluded because we could not calculate the LRs. Several risk factors were only evaluated in a relatively small number of studies, making it difficult to reach strong conclusions. Quiz Ref IDNone of the studies we reviewed evaluated different potential etiologies of nonradicular pain (eg, piriformis syndrome, facet joint pain, myofascial pain, sacroiliac pain, or diskogenic pain); therefore, outcomes from these more specific diagnoses may or may not be associated with disability. However, no reliable methods for identifying pain caused by these different potential etiologies are available, and many experts would lump all of these conditions and others as nonspecific low back pain. In addition, we did not have individual patient data and could not assess collinearity between individual risk factors. The use of multivariate regression was variable across studies, and studies that performed multivariate regression usually did not specifically address collinearity.

The patient had decreased baseline function and described behaviors consistent with fear avoidance. She scored 20 on the Roland Morris Disability Questionnaire, which is associated with an increased likelihood of persistent disabling low back pain at 3 to 6 months (median [range] LR, 1.4 [1.3-3.5]) and at 1 year (median [range] LR, 2.1 [1.2-2.7]). Based on a median pretest probability of 11% for work absenteeism in acute low back pain primary care populations (pretest odds: 0.11/[1 − 0.11] = 0.12), the posttest probability increases slightly to 15% at 3 to 6 months (posttest odds: 0.12 × 1.4 = 0.17; posttest probability: 0.17/[1 + 0.17] = 0.15), but is higher at 20% at 1 year (posttest odds: 0.12 × 2.1 = 0.25; posttest probability: 0.25/[1 + 0.25] = 0.20). She scored 94 on the ALBPSQ risk prediction instrument, which predicts a higher likelihood of persistent disabling low back pain at 3 to 6 months (LR, 3.4; 95% CI, 2.0-5.7; posttest probability, 29%), but a similar likelihood at 1 year (LR, 2.1; 95% CI, 1.1-3.9; posttest probability, 20%).⁴² In a population with a lower pretest probability of a poor outcome at 1 year (eg, 5%), the posttest probability will be lower (10%).

The patient is counseled that usual activities will not hurt her back, that she is likely to recover from this episode, and that she should remain active and return to work as soon as possible. It is also explained to her that extensive treatments or additional diagnostic tests are not necessary at this time. She is given a prescription for a nonsteroidal anti-inflammatory drug and set up for a follow-up appointment at 4 weeks, at which time she is referred for supervised physical therapy or behavioral therapy, aimed at addressing fear avoidance behaviors if her symptoms are persistent.

Quiz Ref IDA systematic approach for primary care patients with low back pain that includes an assessment for high levels of maladaptive pain coping behaviors, presence of nonorganic signs, high levels of baseline functional impairment, low general health status, and psychiatric comorbidities can increase the likelihood of correctly predicting the development of persistent disabling low back pain through 1 year. Low levels of fear avoidance and low baseline functional impairment are the most useful items for predicting likelihood of recovery. Variables related to the work environment, baseline pain, and presence of radiculopathy are less useful for predicting worse outcomes, and a history of prior low back pain episodes and demographic variables (age, sex, smoking status, weight, and educational level) are not useful. Although Waddell's signs are often used to assess for psychological components to back pain, they were designed to assess current pain rather than to predict future disability and have not been studied for that purpose.

Because individual risk factors are relatively weak, risk prediction instruments could be more helpful than individual yellow flags for predicting outcomes. Quiz Ref IDThe items and domains in well-validated instruments would also help clinicians understand the types of factors that independently affect outcomes in patients with low back pain, but there is insufficient evidence to recommend the routine use of any instrument. No instrument has been extensively validated, some validation studies show LRs similar to estimates for individual predictors, and most instruments include individual items that are not predictive (such as sex, education level, or previous low back pain episodes). More research is needed to understand the clinical usefulness of risk prediction instruments for identifying high-risk patients and the optimal strategies to decrease the likelihood of chronic disabling back pain.⁵² For now, clinicians should refer to clinical practice guidelines for recommended approaches to management of acute low back pain without features suggesting a serious underlying condition.¹⁰