Numerous clinical trials have investigated the effectiveness of interventions for low back pain (LBP). Despite the magnitude of research in this area, the evidence remains somewhat inconclusive for many of the more common interventions. One explanation offered for the uncertain nature of the evidence relates to the heterogeneous inclusion criteria used in many of the investigations. Simply speaking, many researchers fail to recognize that all patients with LBP are not the same, thus it is not reasonable to expect that all patients will benefit from identical interventions. Matching patients with interventions for which they are likely to benefit has been advocated as a method to improve the power of investigations.^1,2

For decades, clinicians have attempted to assign patients to specific interventions based on their clinical presentation. Robin McKenzie, a New Zealand physical therapist, was one of the first to describe a classification approach that linked clinical presentation to interventions utilizing a patient’s directional preference³. In the approach advocated by McKenzie³ patients are requested to perform a series of movements and positions during the course of their examination. A directional preference is identified based on the direction of movement that improves the patient’s pain pattern. Essentially, if patients’ symptoms improve from a particular movement (i.e. flexion, extension, or side glide), they are treated with a series of movements and/or positions in that specific direction (directional preference). Researchers have since demonstrated the efficacy of this approach when patients are provided with interventions that match their directional preference.^4-6

In one investigation, researchers classified patients according to their directional preference and reported superior outcomes in the group treated in accordance with their directional preference when compared with patients treated with interventions irrespective of their classification.⁶ Other researchers have proposed a broader classification criteria that bases physical therapy interventions on the patient’s clinical signs and symptoms.⁴ This approach is comparable to that of the McKenzie method, as patient outcomes using the classification system are superior to other methods of assigning interventions that do not directly consider the patient’s clinical presentation.^7,8

Another means of selecting interventions based on a patient’s clinical presentation is through the use of clinical prediction rules (CPRs). CPRs have been developed as both a means of assisting clinicians in the diagnostic process and in the selection of optimum interventions.⁹ CPRs estimate the probability of a specific diagnostic outcome and may predict the prognosis from select interventions. Simply stated, CPRs may guide the clinician in the accurate diagnosis and selection of interventions.

CPRs have been in existence for a few decades in the area of musculoskeletal medicine. Examples include the: 1) Ottawa ankle rules^10,11 (Table 1) designed to predict the need for ankle radiographs following injury and 2) the CPR for identifying deep vein thrombosis in the lower extremity (Table 2).¹² In addition to improving diagnostic utility, CPRs have been developed to assist clinicians in selecting optimum interventions such as spinal manipulation and stabilization.^13-15

The purpose of this module is to briefly review the evolutionary history of CPRs, discuss the methodological standards for developing CPRs, review the hierarchy of evidence for applying CPRs in clinical practice, and to describe the spinal manipulation and stabilization CPRs currently in use for the management of patients with LBP.

Development of Clinical Prediction Rules

The development of a CPR requires a series of three steps (Table 3) that includes derivation, validation, and impact analysis.⁹ Table 4 illustrates a hierarchical classification system used to describe the level of evidence that may be applied to CPRs based on their evolutionary stage of development. The levels of evidence in this classification range from 1 to 4, with Level 1 considered the highest based on the CPR having undergone all three steps of development. An example of a Level 1 CPR is the Ottawa Ankle Rule illustrated in Table 1. To date there are no CPRs for selecting physical therapy interventions that have reached a Level 1 on the evidence hierarchy. 

Step One: Derivation. Step 1 involves derivation of the CPR using rigorous methodological standards as described in Table 5. Outcomes or predictors of interest are first identified and will typically include items (predictors of interest) from the history, examination, and diagnostic testing. An example of predictors may include symptom location, time passed since onset, or the results of special testing such as the straight leg raise. Researchers then must examine a group of patients and determine which of the predictors of interest are present in the individuals who have responded favorably to the intervention for which the CPRs are being investigated. Statistical analysis is then used to determine which of the predictors (present in the group who responded favorably to the intervention) are most accurate for use in the CPRs.⁹ Technically at this stage, the rules are classified as Level 4 evidence on the hierarchical system illustrated in Table 4, as they have not been validated. At this stage the CPR is not ready to be applied in a clinical setting with confidence; however, clinically important information may still be identified. Clinicians familiar with the CPR may alter practice patterns and in fact pay more attention to variables with good predictive power and place less emphasis on those variables that failed to show predictive power.⁹ At this time if the rule looks promising, investigators may move forward into the validation phase. 

Step Two: Validation. Once researchers determine that a CPR has been derived from sound methodological standards (Table 5) and the results are not merely due to chance, the next step of validation is indicated. In Step 2, predictors used in the derivation stage must be validated in a different population. A key consideration in the validation process is to make sure the CPR performs well in a variety of settings, among different clinicians, and is not idiosyncratic to a single population. In other words, one may not assume that the results of a study conducted at one clinic by a single examiner would necessarily be valid in another setting or by a different examiner. For example, the patient population at a specific clinical setting that tends to see a sedentary non-athletic population may be more prone to spinal deconditioning, perhaps making them more likely to benefit from a conditioning program. A validation study that has been conducted in different settings among different clinicians would then be classified as a Level 2 CPR on the hierarchical system. A Level 2 CPR may be applied with confidence in various settings. 

Step Three: Impact Analysis. Once CPRs are validated, an impact analysis must be performed to ascertain whether the rule has changed clinician behavior, improved patient outcomes, or reduced costs. For a CPR to progress to Level 1 of the evidence hierarchy, an impact analysis is necessary. Implementation of CPRs requires clinician time and effort; thus incorporation into clinical practice may be challenging. Part of this challenge may lie in the ability or desire of clinicians to practice outside of their normal daily routine or educational dogma. For example, clinicians may find it necessary to refer to publications or other written documentation to recall specific CPRs, requiring additional time they might not necessarily have during a patient’s visit. Moreover, clinicians may in fact find themselves withholding interventions that they routinely used for years. A CPR that is not incorporated into routine clinical practice with improved outcomes must be questioned, regardless of the accuracy or methodological standards used in the preceding derivation and validation steps. 

Interpreting the Results of CPR Investigations

Equally important to the methodological standards of development and the level of evidence associated with a CPR is the predictive power. While it is beyond the scope of this manuscript to discuss the detailed statistical calculations for interpreting the results, a brief overview is indicated.

The predictive power of a CPR is often presented with diagnostic accuracy statistics. Each item of the history and physical examination is a diagnostic test that either increases or decreases the probability of a diagnosis or may predict the prognosis from an intervention. The question at hand when considering a CPR is determining what patient characteristics modify the pretest probability, thus yielding a new posttest probability. For example, the authors in the manipulation derivation study cite the pretest probability of success from manipulation as 45%.¹⁴ With this being considered, individuals who met the CPRs and were treated with manipulation had a 95% posttest probability of success after analysis.¹⁴ Therefore, patients who met the CPR have a 95% probability of a good outcome from a manipulation intervention compared with a 45% probability of a good outcome if the CPR was not met.

Statistical analysis provides a numeric index of the posttest examination probability of a specific condition or outcome following an intervention. More common diagnostic accuracy statistics include calculations of sensitivity that estimate the true positive rate, and specificity that estimates the true negative rate of a condition. The change from pretest to posttest probability is often determined by the likelihood ratio (LR) calculation. LRs provide an indication as to how much the results of a test (posttest) or fulfillment of CPRs will raise or lower the pretest likelihood of a condition or outcome being present. A positive LR of 1 indicates that the posttest probability is equal to the pretest probability; thus the results are equivocal. A positive LR greater than 1 increases the probability that the target disorder or desired outcome is present, thus the higher the positive LR the greater the increase in prognosis or accuracy from pretest to posttest.¹⁶ Likelihood ratios greater than 10 have been proposed to significantly alter the probability that the target disorder or desired outcome is present. For example, when applying the manipulation CPR (Table 6), individuals who meet 4 of the 5 criteria will have a positive LR of 24.38, which would significantly alter the probability of a good outcome.¹⁴ Readers who desire an in-depth discussion of this topic are referred to the article:¹⁶ “Users guide to the medical literature, III: How to use an article about a diagnostic test” where a detailed summary for interpreting values of diagnostic accuracy is presented.

Interventions for the Lumbar Spine

Numerous interventions have been recommended for patients with LBP, including specific exercises, massage, and thermal and electrical modalities, to name a few. Of these interventions, spinal stabilization and manipulation have received much attention in the literature. Spinal manipulation and stabilization are described in the Guide to Physical Therapist Practice as common interventions used for patients with LBP.¹⁷ Researchers have developed CPRs to describe factors that may identify which patients are likely to derive a favorable benefit from either spinal manipulation or stabilization.

Spinal Manipulation. Manipulation is an intervention used in the management of patients with LBP. While there is sufficient evidence that manipulation is efficacious in the population with LBP,^13,18,19 there is evidence that manipulation may not be better than other interventions.^20,21 The disparity in the outcomes ascribed to manipulation may lie in the methodology used in many investigations. Researchers in many investigations have not attempted to purposefully select patients to identify those more likely to benefit from manipulation, thus patients are often provided with interventions irrespective of their clinical presentation.

A CPR has been developed and validated to identify patients with LBP who are likely to benefit from spinal manipulation (Table 6).^13,14 The CPR for spinal manipulation includes the following 5 predictor variables: 1) duration of symptoms less than 16 days, 2) one hip with internal rotation greater than 35 degrees, 3) hypomobility during spring testing of the lumbar spine, 4) absence of symptoms below the knee and 5) a Fear Avoidance Beliefs Questionnaire²² Work Subscale score less than 19.¹⁴ In the derivation study, when 4 of the 5 predictor variables listed were present, patients were likely to improve from manipulation (+LR = 24.38), whereas having 2 or less predictor variables resulted in a failure to improve. Moreover, patients who met the CPR in the derivation study had a 95% chance of a good prognosis from a manipulation intervention compared with a 45% chance of a good prognosis if the CPR was not met.¹⁴ A validation study using the same CPRs in patients with LBP found that individuals with at least 4 of the predictor variables experienced greater improvement in pain and a greater reduction in disability at 1 week, 4 weeks, and 6 months compared with those presenting with fewer than 4 of the predictor variables.¹³

Researchers have since identified factors likely to predict non-success from spinal manipulation.²³ These factors included: 1) longer symptom duration, 2) the presence of symptoms distal to the low back, 3) reduced hip rotation range of motion, 4) little discrepancy in hip internal rotation range of motion when comparing sides, and 5) a negative Gaenslen’s test for the sacroiliac joint. Patients exhibiting several of these signs during the examination were found to derive minimal benefit from manipulation.²³

Spinal Stabilization. Several research trials have been conducted to determine the efficacy of spinal stabilization interventions with conflicting results.^24-29 Specific strengthening exercises of the multifidus musculature have been found to be efficacious in patients with LBP.^25,26 One investigation found that patients with spondylolitic disorders had better outcomes when placed on a program that included specific stabilization exercises designed to recruit the multifidus and transversus abdominis when compared with patients treated with non-specific exercise.²⁸ Researchers in another investigation reported no difference in outcomes between a group that received conventional physical therapy and a group that received specific spinal stabilization exercise (multifidus and transversus abdominis) when the groups were randomly assigned.²⁴ A third study reported that an exercise intervention was not effective in acute patients but may be helpful in those with chronic LBP.²⁹ Due to the conflicting nature of the research results, one may conclude that stabilization may be beneficial for select groups with low back pain, but not for all patients.

Prior to the derivation of the spinal stabilization CPR, researchers proposed a classification system that identified patients likely to respond to stabilization.⁴ The classification for stabilization was inclusive of characteristics such as hypermobility and relief with immobilization. A CPR has since been developed to determine prognostic factors associated with a favorable outcome following spinal stabilization.¹⁵ In the investigation where the spinal stabilization CPR was developed, participants were provided with an 8-week stabilization program. This program targeted both the deep and local stabilizers. Specifically, the investigators had participants perform exercises for the rectus abdominus, transversus abdominis, internal obliques, erector spinae, multifidus, and quadratus lumborum.¹⁵ Four predictor variables were found to be associated with a favorable prognosis (Table 7). Participants with 3 or more of the predictor variables had a greater likelihood of deriving a benefit from stabilization (+LR = 4.0). However, in the same investigation greater predictive power was found for identifying patients not likely to benefit from a stabilization program as CPRs may predict a desired outcome (good or bad). The four factors predictive for failure are listed in Table 8. Subjects who possessed at least 3 of these “negative” factors were not likely to benefit from a stabilization intervention (+LR = 18.8).  

Discussion

CPRs have been developed in the areas of lumbar spine manipulation and stabilization. The CPRs for manipulation have been validated and may be used for widespread clinical practice, therefore corresponding to a Level 2 on the evidence hierarchy illustrated in Table 2. Impact analysis has not yet been conducted; therefore, it is uncertain whether the manipulation CPR has influenced clinician behavior. Additionally, when considering spinal manipulation, the techniques used for the CPR were non-specific to the lumbar and pelvic region. Research on patient outcomes and recent research using the CPRs suggests that the specific technique itself may not be an important predictor of outcome.^30-32 When considering previous research one may conclude that the accurate identification of patients likely to respond to the intervention is seemingly more important than the specific technique used.

The lumbar stabilization CPR illustrated in Table 7 has not yet been validated in a second population; therefore, the rule corresponds to a Level 4 on the evidence hierarchy. When making clinical decisions strictly based on the evidence hierarchy, the lumbar stabilization CPRs may not necessarily be recommended for widespread use. Moreover, the study used to derive the stabilization CPRs did not include a control group, so one may not with certainty state that stabilization is better than other interventions — or perhaps no treatment at all. When conducting outcomes-based research, the results of one intervention are often compared to those of another. From this comparison one may infer that a particular intervention is superior to another. Although there was no control group in the stabilization CPR investigation, one may infer that the extent of improvement among individuals who receive a stabilization program may be associated with the established CPRs. While the level of evidence may not be high according to the hierarchical values listed in Table 4, one must understand that research is a constantly evolving process and one cannot conclude that an individual study represents the terminal stage of investigation. CPRs traditionally go through an evolutionary stage of development as listed in Table 3 and a derivation study is often a precursor to validation studies. Although the strict levels of evidence suggest that a Level 4 CPR may not be appropriate for widespread use, the CPRs established for stabilization are clinically plausible and possess face validity. Lastly, the stabilization CPR represents common clinical decisions, as it is not uncommon to provide stabilization interventions for patients with segmental hypermobility and positive tests of instability.

Conclusion

While there is no substitute for clinical experience and intuition, there may be times where experience or intuition are misleading or perhaps guided by bias. CPRs provide an efficient and inexpensive means of selecting optimum interventions or establishing diagnoses. The values of CPRs lie not only in their ability to identify those likely to respond to an intervention, but also in identifying those who may need a different approach.

CPRs are an attempt to move beyond the educational dogma and intuitive guesswork routinely incorporated into clinical decisions. While the rules may complement existing knowledge, they are not an absolute tool that should replace sound clinical judgment. Impact analysis investigations are still required to honestly evaluate CPRs and shed light on their value to reduce costs and change clinician behavior.