User's guide to the Chiropractic Literature-IB: how to use an article about therapy

Abstract

IntroductionIn the first article in this series, we described a patient who presented to a chiropractic clinic with a complaint of long-standing neck pain.1Busse JW, Guyatt GH, Bhandari M, Cassidy JD. User's guide to the chiropractic literature-IA: how to use an article about therapy. J Manipulative Physiol Ther 2003;26:330-7Google Scholar The patient had expressed concern surrounding the risks of cervical manipulation. The clinician made a tentative diagnosis of mechanical neck pain. In order to provide the patient with the most appropriate choices of therapy and taking into consideration his concerns about the risks of cervical manipulation, the clinician initiated a literature search to obtain the most recent high-quality information on treatment of chronic mechanical neck pain with spinal manipulation and identified 2 relevant clinical trials.2Jordan A. Bendix T. Nielsen H. Hansen F.R. Host D. Winkel A. Intensive training, physiotherapy, or manipulation for patients with chronic neck pain. A prospective, single-blinded, randomized clinical trial.Spine. 1998; 23: 311-318Google Scholar, 3Bronfort G. Evans R. Nelson B. Aker P.D. Goldsmith C.H. Vernon H. A randomized clinical trial of exercise and spinal manipulation for patients with chronic neck pain.Spine. 2001; 26: 788-797Google Scholar In the first article of the series, we addressed the validity of the 2 studies. The current article, the second in a proposed 4-part series designed to help chiropractors effectively use the published literature in everyday practice, provides a strategy for interpreting the results. The User's Guides to the Medical Literature has provided much of the material in this series.4Guyatt G, Rennie D. User's guides to the medical literature. A manual for evidence-based clinical practice. Chicago: AMA Press; 2002. p. 55-79Google ScholarDiscussionInterpreting study resultsHow large was the treatment effect?Investigators conducting randomized clinical trials will often report the proportion of patients who experience adverse events or outcomes. Examples of these dichotomous outcomes (yes-or-no outcomes that either happen or do not happen) include onset of pain, onset of neurological signs, and death. Patients either do or do not have an event, and the investigators report the proportion of patients who develop such events. Consider, for example, a study in which 20% of the control group but only 5% of the treatment group developed headaches. How might these results be expressed? One way would be as the absolute difference (known as the absolute risk reduction or risk difference) between the proportion who had headaches in the control group (χ) and the proportion who had headaches in the treatment group (γ), or χ − γ = 0.20 − 0.05 = 0.15. Another way to express the impact of treatment would be as a relative risk: the risk of events among patients receiving the new treatment compared with that among controls, or γ / χ = 0.05/0.20 = 0.25. The most commonly reported measure of dichotomous treatment effects is the complement of this relative risk, known as relative risk reduction (RRR). This measure is expressed as a percent: (1 − γ / χ) × 100 = (1 − 0.25) × 100 = 75%. A relative risk reduction of 75% means that the new treatment reduced the risk of developing headaches by 75% compared with the risk among control patients; the greater the relative risk reduction, the more effective the therapy. Investigators may calculate the relative risk over a period of time, as in a survival analysis; this is called a hazard ratio.Investigators may use other methods to describe the size of a treatment effect, such as an odds ratio. However, regardless of what measure is used, point estimates of effect can be misleading unless accompanied by a measure of precision, such as confidence intervals (CIs). The reporting of a significant P value provides more limited information than a CI. The determination of a P value is calculated on the assumption that the finding is due to chance—the null hypothesis. If the likelihood that the observed differences between groups are due to chance is less than what the investigators decide is acceptable, the result is reported as being statistically significant; usually this threshold is set at 5%, or α = .05, so a P value of ≤.05 is considered significant. This threshold is somewhat arbitrary, and if one wishes to be more confident that chance cannot explain a particular difference (which might be the case, for instance, if an experimental treatment were associated with substantial side effects), one can choose a lower threshold P value. For example, a recent pilot study investigating the effect of chiropractic manipulation on chronic pediatric asthma declared, a priori, that only P values less than .01 (α ≤ .01) would be considered significant for their primary outcome measures.5Bronfort G. Evans R.L. Kubic P. Filkin P. Chronic pediatric asthma and chiropractic spinal manipulation a prospective clinical series and randomized clinical pilot study.J Manipulative Physiol Ther. 2001; 24: 369-377Google ScholarP values can be misleading if investigators have performed multiple comparisons. If n independent associations are examined for statistical significance by way of P values, with α = .05, the probability that at least 1 of them will be statistically significant by chance alone is 1 − (1 − .05) n, assuming all the individual null hypotheses are true (ie, none of the associations are actually statistically significant). In other words, the greater the number of independent associations that are tested for significance by obtaining P values, the greater the chance that a significant association will be found that is not actually real—a false-positive result. For example, if an investigator performs 20 comparisons, with α = .05 (a false-positive rate of 5%), there is a 64% chance (1 − 0.9520 = 0.64) that a significant association will be found, even if all associations are actually false. Some authors have attempted to correct for this by increasing the P value for each additional comparison being made (ie, Bonferroni correction); however, this tactic runs the very real risk of also obscuring “real” findings—a false-negative result.6Savitz D.A. Olshan A.F. Multiple comparisons and related issues in the interpretation of epidemiological data.Am J Epidemiol. 1995; 142: 904-908Google ScholarHow precise was the estimate of the treatment effect?The true treatment effect of any given intervention could only be calculated if all individuals on earth were randomized to receive either the intervention or a placebo. While this is not possible, we can investigate the best estimate of the true treatment effect through rigorously conducted clinical trials. This estimate is called a point estimate to remind us that, although the true value hopefully lies close to it, it is unlikely to be precisely correct. Investigators can convey the range within which the true effect likely lies by the statistical strategy of calculating CIs.7Altman D.G. Gore S.M. Gardner M.J. Pocock S.J. Statistical guidelines for contributors to medical journals.in: Gardner M.J. Altman D.G. Statistics with confidence. Confidence intervals and statistical guidelines. British Medical Journal, London1989: 83-100Google Scholar Investigators usually use the 95% CI, which can be considered as defining the range that includes the true treatment effect 95% of the time. The true treatment effect will lie beyond these extremes only 5% of the time, a property of the CI that is closely related to the conventional level of statistical significance of P < .05. The treatment effect can be either the improvement noted for a given outcome measure (pain, range of motion) or the decrease in an adverse event. In the case of a decrease in an adverse event, one useful way of expressing the magnitude of the treatment effect is, as we have already pointed out, with the relative risk reduction. The following example illustrates the use of CIs.If a trial investigating an experimental treatment for neck pain randomized 200 patients, 100 to a treatment group and 100 to a control group, and if there were 20 adverse events (ie, headaches) in the control group and 15 in the treatment group, the authors would calculate a point estimate of the relative risk reduction of 25% (χ = 20/100 or 0.20, γ = 15/100 or 0.15, and 1 − γ / χ = [1 − 0.75] × 100 = 25%). You might guess, however, that the true relative risk reduction may well be much smaller or much greater than 25%, which is based on a difference between groups of just 5 headaches. In fact, you may surmise that the treatment might provide no benefit (a relative risk reduction of 0%) or might even cause harm (a negative relative risk reduction). As it turns out, when one carries out the statistical analysis, either of these possibilities may be correct, as these results are consistent with both a relative risk reduction of −38% (patients given the treatment under investigation might be 38% more likely to develop a headache than control patients) and a relative risk reduction of nearly 59% (patients given the new treatment might be almost 60% less likely to develop a headache than control patients) (Fig 1; calculations not shown). In other words, the 95% CI for this relative risk reduction is −38% to 59%, and the trial has not really helped us to decide whether to offer the new treatment.What if the trial enrolled not 100 but 1000 patients per group and the rates of headache onset were the same as before, so that there were 200 headaches in the control group (χ = 200/1000 = 0.20) and 150 headaches in the treatment group (γ = 150/1000 = 0.15)? The point estimate of the relative risk reduction is again 25% (1 − γ / χ = 1 − [0.15/0.20] × 100 = 25%). In this larger trial, you might be much more confident that the true reduction in risk is much closer to 25%, and this would be correct. The entire 95% CI for the relative risk reduction for this set of results is on the positive side of zero and ranges from 9% to 41% (Fig 1; calculations not shown).These examples demonstrate that the larger the sample size of a trial, the larger the number of outcome events and the greater our confidence that the true relative risk reduction (or any other measure of efficacy) is close to what we have observed. In the second example above, the lowest plausible value for the relative risk reduction was 9% and the highest value was 41%. The point estimate, in this case 25%, is the one value most likely to represent the true relative risk reduction. Values farther and farther from the point estimate are less and less consistent with the observed relative risk reduction. By the time that one crosses the upper or lower boundary of the 95% CI (9%-41%), the values are extremely unlikely to represent the true relative risk reduction, given the point estimate (that is, the observed relative risk reduction). Figure 1 represents the CIs around the point estimate of a relative risk reduction of 25% in these 2 examples, with a risk reduction of zero representing no treatment effect. In both scenarios, the point estimate of the relative risk reduction is 25%, but the CI is far narrower in the second scenario, due to the much larger sample size.It is evident that the larger the sample size, the narrower the CI. When is the sample size large enough to allow for confidence in a conclusion of treatment effect? In a positive study, a study in which the authors conclude that the treatment is effective, it is most useful to determine the lower boundary of the CI. In the second example, this lower boundary was 9%. If this relative risk reduction (the lowest relative risk reduction that is consistent with the study results) is still clinically important (that is, if it is large enough for the chiropractor to recommend the treatment to the patient), then the investigators have enrolled a sufficient number of patients. If, on the other hand, a relative risk reduction of 9% is not considered clinically important, then the study cannot be considered definitive, even if the results are statistically significant (that is, if they exclude a risk reduction of zero). The CI also helps us to interpret a negative study, one in which the authors have concluded that the experimental treatment is no better than the control group therapy.8Detsky A.S. Sackett D.L. When was a “negative” trial big enough? How many patients you needed depends on what you found.Arch Intern Med. 1985; 145: 709-712Google Scholar All one needs to do is to examine the upper boundary of the CI. If the relative risk reduction at this upper boundary would, if true, be clinically important, then the study has failed to exclude an important treatment effect.Consider how this approach would apply to the first scenario presented in this section; the study with 100 patients in each group. This study does not exclude the possibility of harm (indeed, it is consistent with a 38% increase in relative risk), and the study would be considered negative in that it failed to show a convincing treatment effect (Fig 1). Recall, however, that the upper boundary of the CI was a relative risk reduction of 59%. If this large relative risk reduction was actually the true treatment effect, the benefits of the intervention would be substantial. However, as the CI crosses 0, we cannot conclude if the treatment is better, or worse, than the control group. The results of such a study are considered equivocal, and it is likely that there would be a recommendation by the authors for a repeat trial with a larger sample size to narrow the CI and better determine the true effect size.Not all randomized trials have dichotomous outcomes, nor should they. For example, the authors of a randomized, controlled trial on the effect of cervical manipulation on lateral epicondylalgia9Vicenzino B. Collins D. Wright A. The initial effects of a cervical spine manipulative physiotherapy treatment on the pain and dysfunction of lateral epicondylalgia.Pain. 1996; 68: 69-74Google Scholar reported differences in an upper limb tension test (measured in degrees), pain-free grip strength (measured in Newtons), pressure pain threshold (measured in kilopascals), and pain and function at 24 hours posttreatment (according to a visual analog scale) in both the treatment and the control groups. Each of these measures is a continuous variable. The mean difference in subject pain rating was 19% (1.9/10; P < .01) in favor of the patients treated with cervical manipulation, as compared to both placebo and control groups. Here, too, one should look for the 95% CI for this difference in 24-hour pain rating and consider the implications. The lower boundary of the 95% CI was 14% and the upper boundary was 24%. Thus, even the lower boundary of the CI favors the treatment group, and the difference is still clinically important. Having determined the magnitude and precision of the treatment effect, clinicians can turn to the final question of how to apply the results of the study to their patients and their clinical practice.Results of the trials on management of chronic neck painThe study by Jordan et al2Jordan A. Bendix T. Nielsen H. Hansen F.R. Host D. Winkel A. Intensive training, physiotherapy, or manipulation for patients with chronic neck pain. A prospective, single-blinded, randomized clinical trial.Spine. 1998; 23: 311-318Google Scholar noted significant improvement, on the order of approximately 50% from baseline measures, for self-reported pain and disability in all 3 treatment groups (intensive training, physiotherapy, manipulation) (Table 1). Differences between groups in the degree of improvement in any of these outcome measures were not significant. These improvements in all 3 groups were sustained at 4- and 12-month follow-up; however, the improvements in self-perceived pain or disability were not significantly different between groups at any time. Patients in each group reported high levels of satisfaction, with no significant differences between treatment arms. There were no differences between groups for physician-assessed outcome. Self-reported medication usage decreased for all groups, with no significant differences between groups. In terms of physical measurements, all groups demonstrated significant increases in maximal isometric strength in extension (P < .05) but with no differences between groups. All groups demonstrated significant improvement in relative isometric endurance, with a significantly greater increase in the intensive training group (P = .03). Adverse effects were described for 2 subjects (2%): 1 individual who developed headaches in the intensive training group and 1 individual in the manipulation group who experienced “worsening of her condition” and was subsequently diagnosed with fibromyalgia. All groups improved in all outcomes; there were no differences in any outcomes between groups at any point in time and no appreciable trends (Table 1).Table 1Results recorded in the study by Jordan et al2OutcomeGroupBaselineCompletion4-Month follow-up12-Month follow-upPain levelIntensive training12(10–15)6(3–9)4(3–10)6(4–9)Physiotherapy12(10–15)6(3–8)4(3–10)8(6–11)Chiropractic13(10–15)6(4–7)6(5–8)6(6–8)Disability scaleIntensive training8(7–10)5(4–7)5(3–7)5(4–7)Physiotherapy9(8–11)4(3–6)5(3–8)6(4–7)Chiropractic8(7–10)4(4–5)6(4–7)5(3–6)Patients’ perceived effectIntensive training2(1–4)3(1–4)3(1–4)Physiotherapy2(1–4)3(1–4)3(1–4)Chiropractic2(1–5)3(1–5)3(1–4)Doctors’ global assessmentIntensive training2(1–4)Physiotherapy2(1–4)Chiropractic2(1–4)Patients’ using medicationIntensive training50%38%40%21%Physiotherapy45%22%24%27%Chiropractic60%18%40%26%Range of movement in extensionIntensive training60(58–64)64(60–64)Physiotherapy62(58–62)64(60–68)Chiropractic60(60–64)62(60–68)Maximum isometric strength in extensionIntensive training19(16–22)21(19–26)Physiotherapy15(13–21)20(19–28)Chiropractic17(15–22)21(18–28)Maximum isometric strength in flexionIntensive training15(13–18)14(14–18)Physiotherapy16(13–19)17(13–20)Chiropractic13(12–17)15(12–17)Isometric enduranceIntensive training60(50–70)120(110–150)Physiotherapy70(50–80)110(80–130)Chiropractic60(40–70)90(70–110)Values are expressed as medians and 90% CIs, except for Patients' perceived effects and Doctors' global assessment, which are expressed as medians and ranges, and Patients using medication, which is expressed as a percentage. Open table in a new tab The study by Bronfort et al3Bronfort G. Evans R. Nelson B. Aker P.D. Goldsmith C.H. Vernon H. A randomized clinical trial of exercise and spinal manipulation for patients with chronic neck pain.Spine. 2001; 26: 788-797Google Scholar also noted large and important improvement in all 3 treatment groups (manipulation alone, manipulation with exercise, exercise alone) (Table 2). On completion of the trial, in terms of patient-rated outcomes (pain, disability, quality of life, satisfaction, use of medication), the only significant difference between treatments was in patient satisfaction, which was significantly greater for the group receiving manipulation with exercise versus manipulation alone (P = .03). On 12-month follow-up, a significant improvement in patient-rated outcomes was observed for the groups who had received manipulation with exercise or exercise alone versus manipulation alone (P = .01). Between the 2 groups who received exercise, the group that received exercise in combination with manipulation demonstrated significantly higher patient satisfaction versus the group who received exercise alone (P < .01). In terms of observer-rated outcomes, the spinal manipulation with exercise group demonstrated greater gains in all measures of strength, range of motion, and endurance than the group receiving manipulation only (P < .05). The spinal manipulation with exercise group demonstrated more improvement in flexion endurance and in flexion and rotation endurance than the exercise group (P < .03). The exercise group demonstrated greater gains in extension strength and flexion-extension range of motion than the group receiving manipulation alone (P < .05). Bronfort et al3Bronfort G. Evans R. Nelson B. Aker P.D. Goldsmith C.H. Vernon H. A randomized clinical trial of exercise and spinal manipulation for patients with chronic neck pain.Spine. 2001; 26: 788-797Google Scholar did report on adverse effects (neck pain, headaches, radicular pain, thoracic pain) but found that all these side effects, which occurred in 12% of subjects (23 of 191), were self-limited and there were no significant differences in adverse events between treatment groups.Table 2Results recorded in the study by Bronfort et al3OutcomeGroupBaselineWeek 5Completion3-Month follow-up6-Month follow-up12-Month follow-upPainSMT/exercise56(41-71)33(15-51)24(6-42)30(9-51)30(9-51)31(8-54)MedX57(42-72)33(16-50)24(5-43)25(5-45)30(8-52)30(10-50)SMT57(44-70)39(22-56)31(9-53)37(15-59)36(12-60)37(15-59)Neck Disability IndexSMT/exercise26(17-35)19(9-28)14(5-23)14(3-25)15(4-26)16(5-27)MedX27(17-37)17(7-27)12(2-22)14(2-26)15(2-28)16(3-29)SMT28(18-38)20(8-32)16(4-28)19(6-32)18(5-31)20(7-33)SF-36SMT/exercise72(60-84)77(63-91)82(70-94)78(63-93)79(66-92)77(63-91)MedX69(56-82)76(65-87)81(69-93)80(67-93)80(68-92)78(64-92)SMT69(52-86)73(56-90)79(63-95)75(57-93)74(56-92)74(56-92)Medication useSMT/exercise100(42-158)87(39-135)81(38-124)80(37-123)84(39-129)81(35-127)MedX95(44-146)87(40-134)92(44-140)78(36-120)82(36-128)79(36-122)SMT91(38-144)93(43-143)88(40-136)92(45-139)88(41-135)93(45-141)Patient-rated improvementSMT/exercise83(36-130)77(35-119)78(36-120)76(33-119)79(33-125)MedX93(44-142)86(36-136)78(29-127)78(29-127)78(27-129)SMT91(37-145)99(51-147)92(48-136)93(47-139)92(47-137)Patient-rated satisfactionSMT/exercise76(37-115)76(34-118)73(31-115)69(29-109)68(27-109)MedX87(46-128)89(46-132)86(42-130)90(48-132)87(41-133)SMT103(54-152)97(48-146)92(46-138)94(47-141)99(54-144)Static flexion enduranceSMT/exercise165(117-213)MedX66(16-116)SMT74(29-119)Static extension enduranceSMT/exercise285(185-388)MedX160(55-265)SMT146(51-241)Dynamic flexion enduranceSMT/exercise89(74-105)MedX29(13-46)SMT21(5-6)Dynamic extension enduranceSMT/exercise79(59-99)MedX70(50-90)SMT47(28-67)Strength in flexionSMT/exercise8(7-10)MedX6(5-8)SMT4(3-6)Strength in extensionSMT/exercise8(6-10)MedX8(6-10)SMT2(1-4)Strength in rotationSMT/exercise4(3-5)MedX2(1-3)SMT1(−1-3)Flexion/extension ROMSMT/exercise8(5-11)MedX7(4-10)SMT2(−1-4)Rotation ROMSMT/exercise11(9-14)MedX8(5-11)SMT6(3-8)Side bending ROMSMT/exercise8(5-10)MedX5(2-8)SMT2(0-5)Values are expressed as mean values and 95% CIs, except for measures of strength, endurance, and range of motion (ROM), which are expressed as mean change from baseline and 95% CIs. Open table in a new tab Applicability/generalizabilityCan the results be applied to my patient?Often, the patient whom you must treat is somewhat different from those enrolled in a trial. If the patient would have been eligible for the study, that is, if the patient meets all of the inclusion criteria and none of the exclusion criteria, then you can apply the results to your patient's care with considerable confidence. Even here, however, there is a limitation: treatments are not uniformly effective. Typically, some patients respond extremely well, while others derive no benefit. Conventional randomized trials report point estimates of the mean treatment effects; thus, the clinician will likely be exposing some patients to the cost and risks of the treatment without benefit. Additionally, whenever there is clinical skill involved in carrying out the treatment under consideration, the chiropractor must ask if his or her individual level of skill with the treatment is likely to be comparable with that of the chiropractor(s) who provided the care in the reported trial.10Triano JJ. When it comes to spinal manipulation, skill still counts. J Am Chiropr Assoc 1999;36:28-31Google ScholarA final issue arises when your patient shares the features of a subgroup of patients in the reported trial. In assessing the results of a trial (especially when the treatment does not appear to have been efficacious for the average patient), the investigators may have examined subgroups of patients with different stages of an illness, different comorbid conditions, or other differences in potential prognostic factors at the time of entry into the trial. Often these subgroup analyses were not planned ahead of time, and the data are dredged in an attempt to find an effect. Investigators may sometimes overinterpret these data-dependent analyses as demonstrating that the treatment really has a different effect in a subgroup of patients; for instance, they may suggest that patients who were older or sicker benefited substantially more or less than did other subgroups of patients in the trial.One should be critical of subgroup analyses.11Assmann S.F. Pocock S.J. Enos L.E. Kasten L.E. Subgroup analysis and other (mis)uses of baseline data in clinical trials.Lancet. 2000; 355: 1064-1069Google Scholar There are, however, circumstances in which subgroup analysis is appropriate and able to provide important, valid information. Obvious examples are analyses of data collected through the Framingham study or the Nurses Health Study. Both of these large cohort studies were initiated many years ago, and it is entirely reasonable to reanalyze subgroup data in light of new information available that allows us to recognize the potential importance of different questions. Guidelines exist for evaluating the validity of subgroup analysis and emphasize looking for statistical significance of the difference in effect between the subgroups, the magnitude of the effect, whether the hypothesis preceded the analysis, and if other studies or compelling indirect evidence supported the hypothesis.12Oxman A.D. Guyatt G.H. A consumer's guide to subgroup analyses.Ann Intern Med. 1992; 116: 78-84Google Scholar While applying these guidelines will seldom permit definitive conclusions regarding the validity (or lack thereof) of subgroup analysis, they provide a useful tool for determining the likelihood that such analyses may be valid.Were all clinically important outcomes considered?Treatments are indicated when they provide important benefits. The demonstration that a manual therapy increases the range of motion of a joint does not necessarily mean that this treatment should be adopted for routine use, particularly if there is no evidence that an increased range of motion results in important functional improvement. What is required is evidence that the treatment improves outcomes that are important to patients, such as reducing pain or disability or improving function. Another long-neglected outcome is that of the resource implications of alternative treatment strategies. Few randomized trials measure either direct costs, such as drug or program expenses and health care worker salaries, or indirect costs, such as the patient's loss of income due to complications. Indeed, this is the case in the trials on chronic neck pain identified for this report.2Jordan A. Bendix T. Nielsen H. Hansen F.R. Host D. Winkel A. Intensive training, physiotherapy, or manipulation for patients with chronic neck pain. A prospective, single-blinded, randomized clinical trial.Spine. 1998; 23: 311-318Google Scholar, 3Bronfort G. Evans R. Nelson B. Aker P.D. Goldsmith C.H. Vernon H. A randomized clinical trial of exercise and spinal manipulation for patients with chronic neck pain.Spine. 2001; 26: 788-797Google Scholar The increasing constraints on resources that health care systems face mandate careful economic analysis, particularly of resource-intensive interventions.Are the likely treatment benefits worth the potential harm and costs?If the study's results are relatively strong (that is, unlikely to be biased or misleading) and if they are applicable to the patient, the next question concerns whether the probable treatment benefits are worth the effort that the chiropractor and the patient must put into the enterprise. In regard to cervical manipulation, most agree that a small, but so far undetermined, risk exists.13Rothwell D.E.M. Bondy S.J. Williams J.I. Chiropractic manipulation and stroke a population-based case-control study.Stroke. 2001; 32: 1054-1060Google Scholar, 14Haldeman S. Carey P. Townsend M. Papadopoulos C. Arterial dissections following cervical manipulation the chiropractic experience.CMAJ. 2001; 165: 905-906Google Scholar However, no matter how small the risk, any therapy cannot be justified if its benefits are not well established and important to patients or if there exists other readily available treatments with lesser risk and with equivocal, or greater, efficacy. Alternately, if a therapy, such as cervical manipulation, provides a real but minimal benefit, the onus on providers is to demonstrate that the benefits are sufficient to justify the risk.15Shekelle P.G. What role for chiropractic in health care?.N Engl J Med. 1998; 339: 1074-1075Google ScholarApplication to practiceThe effort involved in searching and analyzing the literature on manipulation and chronic neck pain has proved valuable in this instance. Two recent and very relevant trials used methods that were sufficiently strong that they provide useful information.2Jordan A. Bendix T. Nielsen H. Hansen F.R. Host D. Winkel A. Intensive training, physiotherapy, or manipulation for patients with chronic neck pain. A prospective, single-blinded, randomized clinical trial.Spine. 1998; 23: 311-318Google Scholar, 3Bronfort G. Evans R. Nelson B. Aker P.D. Goldsmith C.H. Vernon H. A randomized clinical trial of exercise and spinal manipulation for patients with chronic neck pain.Spine. 2001; 26: 788-797Google Scholar Randomization ensured that prognostic factors tha