Skip to main page content

• HOME
• Current
• Past issues
• Contact us
• Subscribe
• Sign up for e-Alerts
• Help

PHYS THER
Vol. 89, No. 7, July 2009, pp. 698-704
DOI: 10.2522/ptj.20080351

This Article Abstract Full Text (PDF) Discussion Podcast All Versions of this Article: ptj.20080351v1 89/7/698    most recent Submit a response Read responses to this article Alert me when this article is cited Alert me when Rapid Responses are posted Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Hancock, M. Articles by Maher, C. G. Search for Related Content PubMed PubMed Citation Articles by Hancock, M. Articles by Maher, C. G. Related Collections Evidence-Based Practice Perspectives Classification Clinical Prediction Rules Social Bookmarking                      What's this?

Perspectives

A Guide to Interpretation of Studies Investigating Subgroups of Responders to Physical Therapy Interventions

Mark Hancock, Robert D. Herbert and Christopher G. Maher

M. Hancock, PT, PhD, is Lecturer, Back Pain Research Group, The University of Sydney, Sydney, New South Wales, Australia. Mailing address: Faculty of Health Sciences, The University of Sydney, PO Box 170, Lidcombe 1825, New South Wales, Australia.
R.D. Herbert, PT, PhD, is Senior Research Fellow, The George Institute for International Health, The University of Sydney.
C.G. Maher, PT, PhD, is Director, The George Institute for International Health, The University of Sydney.

Address all correspondence to Dr Hancock at: M.Hancock{at}usyd.edu.au

Submitted November 3, 2008; Accepted April 6, 2009

 
Many researchers and clinicians believe the effectiveness of existing physical therapy interventions can be improved by targeting the provision of specific interventions at patients who respond best to that treatment. Although this approach has the potential to improve outcomes for some patients, it needs to be implemented carefully because some methods used to identify subgroups can produce biased or misleading results. The aim of this article is to assist readers in assessing the validity and generalizability of studies designed to identify subgroups of responders to physical therapy interventions. The key messages are that subgroups should be identified using high-quality randomized controlled trials, the investigation should be limited to a relatively small number of potential subgroups for which there is a plausible rationale, subgroup effects should be investigated by formally analyzing statistical interactions, and findings of subgroups should be subject to external validation.

Top Abstract Introduction Study Design Predictor Variables Analysis Strategy Sample Size Subgroups Where the Mean... Validation Summary Appendix References

 
Identifying patients who respond best to certain interventions has the potential to improve outcomes for existing and new physical therapy interventions. This is particularly the case in conditions, such as back pain, that are considered heterogeneous. It is unlikely that any one intervention for heterogeneous conditions will be highly effective for all patients. Some researchers¹ have reported that interventions previously shown to have little effect when provided to heterogeneous populations can be more effective when provided to selected subgroups of patients.

Characteristics that identify subgroups of patients who respond differently to a specific intervention are called treatment effect modifiers^2,3 or effect moderators. A patient's status for a given effect modifier provides information on how much additional benefit the patient is likely to gain from a specific intervention. For example, for a patient with an acute stroke, the type of stroke (ischemic or hemorrhagic) is a powerful effect modifier for response to anticoagulant therapy. The treatment has been shown to be highly effective for people whose stroke was caused by a clot^4,5 but could be harmful or even fatal for those with a hemorrhagic stroke.

Effect modifiers must be differentiated from prognostic factors. Prognostic factors are characteristics that identify patients who recover at different rates or have different outcomes. Prognostic factors are useful in providing patients with a more accurate prognosis, but they do not provide any information about which patients will respond best to a specific intervention. An example of a prognostic factor for recovery from low back pain is initial pain intensity. Patients with lower initial pain intensity typically have been shown to recover more quickly than patients with higher pain intensity, regardless of intervention.^6,7 Although it may be tempting to presume that low pain intensity also predicts response to treatment, this assumption does not logically follow. It is quite possible that a particular intervention is most effective when applied to people with more intense pain, or that its effectiveness does not depend on pain intensity.

It is important that studies investigating subgroups of patients who respond best to particular treatments use appropriate design and analysis strategies. In some studies, single predictors are investigated, whereas other studies investigate combinations of predictors. Regardless of whether a single predictor or combinations of predictors are used, the design and analysis must match the aim of identifying specific predictors of response to the intervention. In particular, it is important to avoid using a design and analysis that are suitable for identifying prognostic factors (ie, predictors of outcome) when the intention is to identify specific predictors of subgroups of patients who respond best to the particular intervention.

Clinical prediction rules (CPRs) originally were used to quantify the usefulness of clusters of patient characteristics (eg, history and physical examination findings) for diagnosis and prognosis. However, CPRs also can be used to identify patients who respond best to certain interventions. Publications reporting on this sort of CPR have become more numerous in the physical therapy literature,^8–12 especially in back pain research.

A large body of literature has described the stages of development of CPRs.^13–15 The 3 main stages described in the literature are derivation, validation, and impact analysis.^13–15 In the derivation stage, studies are conducted to determine whether particular variables predict the outcome of interest. In the validation stage, variables found to be informative in the derivation study are tested in a new setting with new patients to assess internal and external validity. Finally, impact analysis studies investigate whether the use of the CPR in clinical practice results in better outcomes for patients.

Adherence to generic guidelines for development and validation of CPRs is important but does not obviate the need to use study designs and analyses appropriate for a specific research question. This important point was emphasized in an instructional article on CPRs by Beattie and Nelson in which the authors stated that "it is important to note that different research methods are needed in the development phase of a clinical prediction rule based upon its proposed use."^13(p159) Studies investigating diagnosis typically use cross-sectional designs, studies of prognosis use longitudinal designs (often single-arm studies), and studies of effect modification use controlled (2-arm) trials. It has become common for researchers to use study designs and analyses that are appropriate for identifying prognostic factors (ie, nonspecific predictors of outcome) but to interpret the results as if they had identified subgroups of responders to treatment.^9–12 Clinical prediction rules can be used to identify subgroups of patients who respond best to certain interventions, but only if the appropriate design is used.

Research into subgroups of responders using CPRs has the potential to advance the science and practice of physical therapy and to improve outcomes for patients, but it also can produce biased or misleading results if not conducted and interpreted correctly. The aim of this article is to assist readers in interpreting the validity of studies investigating subgroups of responders to physical therapy interventions by considering key issues in study design and analysis. The focus will be on design and analysis of CPRs of subgroup effects using randomized controlled trials (RCTs). The article is structured so that each key issue is discussed and then summarized. The Appendix lists the 6 key points.

Top Abstract Introduction Study Design Predictor Variables Analysis Strategy Sample Size Subgroups Where the Mean... Validation Summary Appendix References

 
When the aim of a study is to identify factors associated with the effect of an intervention, the study must use an experimental design that can estimate the effect of that intervention. As the effect of treatment is assessed by comparing outcomes in patients receiving the treatment with outcomes in a control group who do not receive the treatment, the study must have 2 groups. Preferably, the study should be a well-conducted RCT. Similarly, if the aim of the study is to identify patient characteristics that would help a clinician choose between 2 interventions (eg, spinal manipulation or exercise) for a specific patient, the appropriate design is an RCT with patients randomly assigned to receive spinal manipulation or exercise. It is important to realize that only RCTs can provide a rigorous test of whether a subgroup characteristic is associated with response to intervention. Single-arm trials cannot provide rigorous estimates of the effect of treatment because there is no control group not receiving the treatment; thus, single-arm trials cannot identify factors associated with the effect of treatment.

It is important to use controlled trials in all stages of the development of a CPR that aims to identify subgroups of responders to treatment. In the derivation stage of a CPR, a wide range of predictor variables may be examined to develop the CPR. In the validation stage, the CPR is tested for internal and external validity in new samples of patients. As both the development and validation studies assess the effect of treatment, both require controlled trials.

We are aware of one example that appears to justify the use of single-armed trials to identify treatment effect modifiers. Flynn and colleagues¹¹ used a single-armed trial to demonstrate that a particular clinical presentation was associated with a particularly good outcome in patients receiving manipulation for low back pain. Subsequently, Childs and colleagues⁸ validated the CPR with a randomized trial. In our opinion, there is no reason to expect that factors found in single-arm trials to be predictive of outcome will subsequently be found in 2-arm trials to be predictive of response to treatment.

Where researchers believe there is a rationale for why a CPR derived in a single-arm study also may predict response to an intervention, the CPR should be investigated in a controlled trial before any suggestion about the role in predicting response to treatment is made. When a CPR is developed in a single-arm trial, any subsequent evaluation in a controlled study should be considered a derivation study and not a validation study. The reasoning is that the single-arm trial derived a CPR for prognosis, whereas the controlled trial derived a CPR for response to treatment. The analyses in the 2 studies are quite different; therefore, replication clearly has not occurred. We also would caution that use of the term "validation study" may be misleading, as it suggests a single study can validate a CPR. A series of studies usually is required to validate any CPR, regardless of whether it investigates diagnosis, prognosis, or response to treatment. Validation studies could include replication in similar clinical scenarios to show that the original result was not just due to chance (internal validation) through to testing in different scenarios to establish how generalizable the CPR is (external validation).

Several previous studies that have claimed to identify subgroups of responders to intervention used single-arm trials with no control group.^9–12 As these studies did not include a control group, they could not estimate the effect of treatment (that is, they could not estimate the difference between treated and control groups) and, therefore, could not identify factors that modify the effect of a intervention. Studies that do not contrast outcomes between treatment and control groups can identify nonspecific predictors of outcome (or prognostic factors), but they cannot identify treatment effect modifiers.

The following hypothetical example illustrates why a single-arm study design cannot validly identify subgroups of responders to treatment. A single-arm study investigated 100 patients with ankle fractures who received treatment with passive accessory mobilization of the ankle. The outcome measure was pain, as assessed with a 0 to 10 visual analog scale (VAS), at 6 weeks after cast removal. Two features (nondisplaced fracture and nonsurgical management) predicted a good outcome at 6 weeks and were included in a CPR. Patients who met the CPR had a 6-week mean VAS score of 2/10, whereas patients who were negative on the CPR had a mean VAS score of 5/10 at 6 weeks after cast removal. Is it, therefore, appropriate to conclude that the CPR identifies patients with ankle fractures who respond best to passive accessory mobilization?

Let us now imagine that the same study was a placebo controlled trial. Patients in the control group who met the CPR had a mean 6-week VAS score of 4/10. Patients in the control group who were negative on the CPR had a mean 6-week VAS score of 7/10. Therefore, the patients who met the CPR and received passive accessory mobilization (2/10) were, on average, 2 points better than those who met the rule and did not receive treatment (4/10) (Table). The patients who were negative on the CPR and received physical therapy (5/10) also were, on average, 2 points better than those who were negative on the CPR and did not receive physical therapy (7/10) (Table). That is, while those patients who met the CPR had better outcomes, they did not have a greater response to the treatment (compared with the control group) than those patients who did not meet the CPR (Table). The single-arm trial demonstrates that in a group of patients with ankle fractures who receive passive accessory mobilization, the outcome (regardless of treatment) is better in those who meet the CPR. The single-arm trial does not demonstrate that the CPR identifies a subgroup of patients who respond best to passive accessory mobilization.

View this table: [in this window] [in a new window]

Table. Hypothetical Example: Mean Pain Score (Out of a Maximum Score of 10) in Patients With Ankle Fractures Treated With Passive Accessory Mobilizationa

A number of studies have used RCTs to evaluate whether a certain patient profile is associated with response to treatment.^8,16–19 For example, Childs and colleagues⁸ conducted an RCT that demonstrated patient characteristics predicted response to spinal manipulation and exercise compared with exercise alone. Their data provide a direct test of the hypothesis that certain subgroups respond better than others to manipulation. Underwood and colleagues’ analysis of the UK BEAM trial revealed that age, work status, age of leaving school, pain and disability, quality of life, and beliefs were prognostic factors for recovery from low back pain, but the same factors did not predict response to manipulation, exercise, or both treatments combined.¹⁹

Up to this point we have argued that CPRs that identify predictors of response to treatment must use 2-armed (controlled) trials. Ideally, we would like a CPR to be tested in a high-quality randomized trial. We know from surveys that trials of physical therapy²⁰ vary enormously in methodological quality. This is of concern because trials of low quality are associated with exaggerated treatment effects.²¹

Key point 1: Treatment effect modifiers should be investigated in an RCT before any conclusion regarding their role in predicting response to treatment is drawn.

Top Abstract Introduction Study Design Predictor Variables Analysis Strategy Sample Size Subgroups Where the Mean... Validation Summary Appendix References

 
Even when predictors of response to treatment are investigated in high-quality randomized trials, they are still prone to spurious findings.^22–24 A major reason for this is that the analyses often are conducted post hoc and involve investigation of a large number of variables with no plausible rationale for being predictors of response to intervention. Many authors^22–24 have warned about the risks associated with multiple subgroup analyses in controlled trials. With a critical P value of .05, the chance of a falsely significant finding is as high as 5% for each predictor investigated. Consequently, when a large number of predictors are investigated, it usually is likely that one or more predictors will incorrectly be found to be a statistically significant predictor of response to intervention. The most appropriate way to deal with this problem is to define a limited number of plausible predictor variables prior to the conduct of the trial. Ideally, these predictors should be specified explicitly in a trial protocol that is published when the trial is registered. In a derivation study, it is reasonable to investigate a somewhat wider range of variables, but the results then need prospective validation in a future trial prior to being recommended for clinical practice.

The study by Childs et al⁸ is a good example of an RCT in which the risk of spurious findings was limited by making the subgroup analysis the primary analysis of the study. In that study, only one predictor variable (positive or negative on the CPR) was investigated.

Key point 2: Studies investigating treatment effect modifiers should limit the analyses to a small number of plausible predictors (subgroups) that are nominated prior to the conduct of the trial.

Top Abstract Introduction Study Design Predictor Variables Analysis Strategy Sample Size Subgroups Where the Mean... Validation Summary Appendix References

 
A common^25–30 and seemingly logical approach to the evaluation of treatment effect modifiers is to examine the effectiveness of treatment compared with a control condition in subgroups of patients from a trial. Researchers who analyze their data in this way often find statistically significant effects of treatment in some subgroups but not others, and they may claim that this provides evidence that some subgroups respond differently than others. An example of this approach is the low back pain study by Gudavalli et al.²⁷ These authors reported a statistically significant effect in favor of "flexion-distraction" treatment compared with active exercise in patients with radiculopathy (P=.05) but not in patients without radiculopathy (P=.13). They concluded that people with radiculopathy respond best to flexion-distraction treatment. However, this conclusion was based on a flawed analysis.^31–33 Simulation studies show that even when there is no subgroup effect, researchers can expect to find significant effects of treatment in one subgroup only (P<.05) in 7% to 64% of tests.³¹

The correct analysis involves a test of interaction. That is, the correct analysis involves demonstrating that the effect of flexion-distraction compared with active exercise for one subgroup (patients with radiculopathy) is greater than the effect of flexion-distraction compared with active exercise in the other subgroup (patients without radiculopathy). The size of the interaction tells us how much more benefit (compared with the control condition) patients in the subgroup received from treatment compared with those not in the subgroup. Simulation studies show that when there is no subgroup effect, researchers can expect to find statistically significant interaction effects (P<.05) in 5% of tests, as is expected. That is, tests of interactions are much less prone than tests within subgroups to false positive findings.³¹

Key point 3: Identification of treatment effect modifiers should be based on tests of interactions.

Top Abstract Introduction Study Design Predictor Variables Analysis Strategy Sample Size Subgroups Where the Mean... Validation Summary Appendix References

 
It is important to note that a test of interaction requires a significantly larger sample size to achieve the same level of statistical power or statistical precision than a test of the overall effect. When about half of the participants are in each subgroup, 4 times as many participants are required for a test of interaction than would be required for a test of a main effect of the same size.³¹ When the subgroups are not equal in size, as is usually the case, even greater sample sizes are required. Most randomized trials are powered only for tests of the main effect of treatment, so they have insufficient power to detect an interaction effect. Readers of reports of CPRs can inspect the confidence intervals around estimates of interaction effects to ascertain how much uncertainty is associated with a particular CPR. If a study is underpowered for testing an interaction, it will have wide confidence intervals and will risk incorrectly finding the effect modifier to be uninformative. A potential solution to the problem of inadequate power for testing interactions is to combine data from similar RCTs using meta-analysis techniques.³⁴

Key point 4: Studies investigating treatment effect modification require significantly larger sample sizes than studies of main effect. Convincing evidence of treatment effect modification requires precise estimates of interactions, as evidenced by narrow confidence intervals.

Top Abstract Introduction Study Design Predictor Variables Analysis Strategy Sample Size Subgroups Where the Mean... Validation Summary Appendix References

 
When a treatment has a moderate or large mean effect across all patients included in a trial, the researchers may be less inclined to investigate the presence of subgroups of responders. It is when treatments have small effects that researchers often will want to look for subgroups, but this is the context in which subgroup effects are least plausible. Where there are moderate or large effects, it is possible that a proportion of patients could receive a large effect while other patients receive no benefit from the treatment. However, where the mean effect is close to zero, the only way that a proportion of patients can receive a large effect is if the treatment is actually harmful (compared with the control condition) for other patients. Although, theoretically, this is possible, it would seem unlikely in most situations. Thus, in most situations, evidence of subgroup effects should be treated with caution if the main effect of treatment is small.

Key point 5: Evidence of effect modification should be treated cautiously when the main (pooled) effect of treatment is close to zero.

Top Abstract Introduction Study Design Predictor Variables Analysis Strategy Sample Size Subgroups Where the Mean... Validation Summary Appendix References

 
When researchers identify treatment effect modifiers using a particular set of data (for instance, data from a particular randomized trial), there is always the risk that the result is applicable only to the specific population from which the sample was drawn. In general, effect modifiers will be less predictive when they are applied to data other than the sample on which they were identified. This is particularly the case when CPRs are developed to investigate the ability of combinations of variables to predict response to an intervention. For this reason, it is important to externally validate effect modifiers, including CPRs, before recommending them for use in clinical practice.^13–15 External validation involves testing the validity of the effect modifier on other populations. We can differentiate "narrow" external validation, which involves testing in settings and in patients very similar to those of the original trial, and "broad" external validation, which involves testing in different settings and on different types of patients. An effect modifier that has been broadly validated is one that is widely applicable. Generalizability is important if an effect modifier is to be recommended for use in clinical practice to select patients who will respond best to a specific intervention, where many things, including the patients, settings, co-interventions, and clinicians, will be different from the original trial in which the CPR was developed.

Key point 6: Effect modifiers should be externally validated before they are incorporated into guidelines for clinical practice.

Top Abstract Introduction Study Design Predictor Variables Analysis Strategy Sample Size Subgroups Where the Mean... Validation Summary Appendix References

 
This article has identified a number of features that enhance the validity of studies designed to identify subgroups of patients who respond best to intervention. The Appendix provides a list of the key principles to be considered when interpreting studies investigating subgroups of responders to physical therapy interventions.³⁵

Top Abstract Introduction Study Design Predictor Variables Analysis Strategy Sample Size Subgroups Where the Mean... Validation Summary Appendix References

 

View larger version (40K): [in this window] [in a new window]

Appendix Key Principles for Assessing Points Determining the Validity of Trials Reporting Subgroups of Responders to Interventions

 
All authors provided concept/idea/project design and writing and reviewed the final manuscript.

Dr Herbert's and Dr Maher's fellowships are funded by the National Health and Medical Research Council of Australia.

Top Abstract Introduction Study Design Predictor Variables Analysis Strategy Sample Size Subgroups Where the Mean... Validation Summary Appendix References

 

1. Brennan GP, Fritz JM, Hunter SJ, et al. Identifying subgroups of patients with acute/subacute "nonspecific" low back pain: results of a randomized clinical trial. Spine. 2006;31:623–631.[CrossRef][Web of Science][Medline]
2. Kopec JA, Esdaile JM. Functional disability scales for back pain. Spine. 1995;20:1943–1949.[Web of Science][Medline]
3. Pocock SJ, Collier TJ, Dandreo KJ, et al. Issues in the reporting of epidemiological studies: a survey of recent practice. BMJ. 2004;329:883.[Abstract/Free Full Text]
4. Kwiatkowski TG, Libman RB, Frankel M, et al; National Institute of Neurological Disorders and Stroke Recombinant Tissue Plasminogen Activator Stroke Study Group. Effects of tissue plasminogen activator for acute ischemic stroke at one year. N Engl J Med. 1999;340:1781–178[Abstract/Free Full Text]
5. National Institute of Neurological Disorders and Stroke Recombinant Tissue Plasminogen Activator Stroke Study Group. Tissue plasminogen activator for acute ischemic stroke. N Engl J Med. 1995;333:1581–158[Abstract/Free Full Text]
6. Croft PR, Dunn KM, Raspe H. Course and prognosis of back pain in primary care: the epidemiological perspective. Pain. 2006;122:1–3.[CrossRef][Web of Science][Medline]
7. Hancock MJ, Maher CG, Latimer J, et al. Can rate of recovery be predicted in patients with acute low back? Development of a clinical prediction rule. Eur J Pain. 2009;13:51–55.[CrossRef][Web of Science][Medline]
8. Childs JD, Fritz JM, Flynn TW, et al. A clinical prediction rule to identify patients with low back pain most likely to benefit from spinal manipulation: a validation study. Ann Intern Med. 2004;141:920–928.[Abstract/Free Full Text]
9. Cleland JA, Childs JD, Fritz JM, et al. Development of a clinical prediction rule for guiding treatment of a subgroup of patients with neck pain: use of thoracic spine manipulation, exercise, and patient education. Phys Ther. 2007;87:9–23.[Abstract/Free Full Text]
10. Fernandez-de-las-Penas C, Cleland JA, Cuadrado ML, Pareja JA. Predictor variables for identifying patients with chronic tension-type headache who are likely to achieve short-term success with muscle trigger point therapy. Cephalalgia. 2008;28:264–275.[Abstract/Free Full Text]
11. Flynn T, Fritz J, Whitman J, et al. A clinical prediction rule for classifying patients with low back pain who demonstrate short-term improvement with spinal manipulation. Spine. 2002;27:2835–2843.[CrossRef][Web of Science][Medline]
12. Iverson CA, Sutlive TG, Crowell MS, et al. Lumbopelvic manipulation for the treatment of patients with patellofemoral pain syndrome: development of a clinical prediction rule. J Orthop Sports Phys Ther. 2008;38:297–309.[CrossRef][Web of Science][Medline]
13. Beattie P, Nelson RM. Clinical prediction rules: what are they and what do they tell us? Aust J Physiother. 2006;52:157–163.[Web of Science][Medline]
14. Laupacis A, Sekar N, Stiell IG. Clinical prediction rules: a review and suggested modifications of methodological standards. JAMA. 1997;277:488–494.[Abstract/Free Full Text]
15. McGinn TG, Guyatt GH, Wyer PC, et al; Evidence-Based Medicine Working Group. Users’ guides to the medical literature, XXII: how to use articles about clinical decision rules. JAMA. 2000;284:79–84.[Abstract/Free Full Text]
16. Kalauokalani D, Cherkin DC, Sherman KJ, et al. Lessons from a trial of acupuncture and massage for low back pain: patient expectations and treatment effects. Spine. 2001;26:1418–1424.[CrossRef][Web of Science][Medline]
17. Klaber Moffett JA, Carr J, Howarth E. High fear-avoiders of physical activity benefit from an exercise program for patients with back pain. Spine. 2004;29:1167–1172.[CrossRef][Web of Science][Medline]
18. Stewart MJ, Maher CG, Refshauge KM, et al. Randomized controlled trial of exercise for chronic whiplash-associated disorders. Pain. 2007;128:59–68.[CrossRef][Web of Science][Medline]
19. Underwood MR, Morton V, Farrin A; Team UBT. Do baseline characteristics predict response to treatment for low back pain? Secondary analysis of the UK BEAM dataset [ISRCTN32683578]. Rheumatology (Oxford). 2007;46:1297–1302.[CrossRef][Medline]
20. Moseley AM, Herbert RD, Sherrington C, Maher CG. Evidence for physiotherapy practice: a survey of the Physiotherapy Evidence Database (PEDro). Aust J Physiother. 2002;48:43–49.[Web of Science][Medline]
21. Schulz KF, Chalmers I, Hayes RJ, Altman DG. Empirical evidence of bias: dimensions of methodological quality associated with estimates of treatment effects in controlled trials. JAMA. 1995;273:408–412.[Abstract/Free Full Text]
22. Brookes ST, Whitley E, Peters TJ, et al. Subgroup analyses in randomised controlled trials: quantifying the risks of false-positives and false-negatives. Health Technol Assess. 2001;5:1–56.[Medline]
23. Moye LA, Deswal A. Trials within trials: confirmatory subgroup analyses in controlled clinical experiments. Control Clin Trials. 2001;22:605–619.[CrossRef][Web of Science][Medline]
24. Yusuf S, Wittes J, Probstfield J, Tyroler HA. Analysis and interpretation of treatment effects in subgroups of patients in randomized clinical trials. JAMA. 1991;266:93–98.[Abstract/Free Full Text]
25. Callaghan MJ, Selfe J, McHenry A, Oldham JA. Effects of patellar taping on knee joint proprioception in patients with patellofemoral pain syndrome. Man Ther. 2008;13:192–199.[CrossRef][Web of Science][Medline]
26. Clegg DO, Reda DJ, Harris CL, et al. Glucosamine, chondroitin sulfate, and the two in combination for painful knee osteoarthritis. N Engl J Med. 2006;354:795–808.[Abstract/Free Full Text]
27. Gudavalli MR, Cambron JA, McGregor M, et al. A randomized clinical trial and subgroup analysis to compare flexion-distraction with active exercise for chronic low back pain. Eur Spine J. 2006;15:1070–1082.[CrossRef][Web of Science][Medline]
28. Pearson AM, Blood EA, Frymoyer JW, et al. SPORT lumbar intervertebral disk herniation and back pain: does treatment, location, or morphology matter? Spine. 2008;33:428–435.[CrossRef][Web of Science][Medline]
29. Skargren EI, Carlsson PG, Oberg BE. One-year follow-up comparison of the cost and effectiveness of chiropractic and physiotherapy as primary management for back pain: subgroup analysis, recurrence, and additional health care utilization. Spine. 1998;23:1875–1883.[CrossRef][Web of Science][Medline]
30. Yip YB, Tse H-MS, Wu KK. An experimental study comparing the effects of combined transcutaneous acupoint electrical stimulation and electromagnetic millimeter waves for spinal pain in Hong Kong. Comp Ther Clin Pract. 2007;13:4–14.[CrossRef]
31. Brookes ST, Whitely E, Egger M, et al. Subgroup analyses in randomized trials: risks of subgroup-specific analyses: power and sample size for the interaction test. J Clin Epidemiol. 2004;57:229–236.[CrossRef][Web of Science][Medline]
32. Klebanoff MA. Subgroup analysis in obstetrics clinical trials. Am J Obstet Gynecol. 2007;197:119–122.[CrossRef][Medline]
33. Lagakos SW. The challenge of subgroup analyses: reporting without distorting. N Engl J Med. 2006;354:1667–1669.[Free Full Text]
34. Schellingerhout JM, Verhagen AP, Heymans MW, et al. Which subgroups of patients with non-specific neck pain are more likely to benefit from spinal manipulation therapy, physiotherapy, or usual care? Pain. 2008;139:670–680.[CrossRef][Web of Science]
35. Beneck GJ, Kulig K, Landel RF, Powers CM. The relationship between lumbar segmental motion and pain response produced by a posterior-to-anterior force in persons with nonspecific low back pain. J Orthop Sports Phys Ther. 2005;35:203–209.[Web of Science][Medline]

CiteULike    Complore    Connotea    Del.icio.us    Digg    Reddit    Technorati    What's this?

This article has been cited by other articles:

				T. R. Stanton, M. J. Hancock, C. G. Maher, and B. W. Koes Author Response Physical Therapy, June 1, 2010; 90(6): 858 - 859. [Full Text] [PDF]

				D. F. Kallmes, J. G. Jarvik, R. H. Osborne, B. A. Comstock, M. P. Staples, P. J. Heagerty, J. A. Turner, and R. Buchbinder Clinical Utility of Vertebroplasty: Elevating the Evidence Radiology, June 1, 2010; 255(3): 675 - 680. [Full Text] [PDF]

				T. R. Stanton, M. J. Hancock, C. G. Maher, and B. W. Koes Critical Appraisal of Clinical Prediction Rules That Aim to Optimize Treatment Selection for Musculoskeletal Conditions Physical Therapy, June 1, 2010; 90(6): 843 - 854. [Abstract] [Full Text] [PDF]

				L. O.P. Costa, C. G. Maher, J. Latimer, P. W. Hodges, R. D. Herbert, K. M. Refshauge, J. H. McAuley, and M. D. Jennings Author Response Physical Therapy, February 1, 2010; 90(2): 307 - 308. [Full Text] [PDF]

				W. Clark, S. Lyon, J. Burnes, M. O. Baerlocher, P. L. Munk, D. M. Liu, J. C. Lotz, A. Grey, M. Bolland, R. Buchbinder, et al. Trials of Vertebroplasty for Vertebral Fractures N. Engl. J. Med., November 19, 2009; 361(21): 2097 - 2100. [Full Text] [PDF]

				S. C. Allison On "A guide to interpretation of studies..." Hancock M, et al. Phys Ther. 2009;89:698-704. Physical Therapy, October 1, 2009; 89(10): 1098 - 1099. [Full Text] [PDF]

				M. Hancock, R. Herbert, and C. G. Maher Author Response Physical Therapy, October 1, 2009; 89(10): 1099 - 1100. [Full Text] [PDF]

This Article Abstract Full Text (PDF) Discussion Podcast All Versions of this Article: ptj.20080351v1 89/7/698    most recent Submit a response Read responses to this article Alert me when this article is cited Alert me when Rapid Responses are posted Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Hancock, M. Articles by Maher, C. G. Search for Related Content PubMed PubMed Citation Articles by Hancock, M. Articles by Maher, C. G. Related Collections Evidence-Based Practice Perspectives Classification Clinical Prediction Rules Social Bookmarking                      What's this?