The Importance of Placebo Effectsin Pain Treatment and ResearchJudith A. Turner, PhD; Richard A. Deyo, MD, MPH; John D. Loeser, MD; Michael Von Korff, ScD; Wilbert E. Fordyce, PhD

Objective.\p=m-\Toestimate the importance and implications of placebo effects inpain treatment and research from the existing literature, with emphasis on their mag-nitude and duration, the conditions influencing them, and proposed explanations.Data Sources.\p=m-\English-languagearticles and books identified through MED-

LINE (1980 through 1993) and PsycLIT (1967 through 1993) database searching,bibliography review, and expert consultation.Study Selection.\p=m-\Articleswere included if they pertained to the review objec-

tives.Results.\p=m-\Placeboresponse rates vary greatly and are frequently much higher

than the often-cited one third. Placebos have time-effect curves, and peak, cumu-lative, and carryover effects similar to those of active medications. As with medi-cation, surgery can produce substantial placebo effects, and this possibility is com-monly overlooked in case series reports on back surgery. Individuals are notconsistent in their placebo responses, and a placebo-responder personality has notbeen identified. Models advanced to explain placebo effects emphasize the role ofanxiety, expectations, and learning.

Conclusions.\p=m-\Placeboeffects influence patient outcomes after any treatment,including surgery, that the clinician and patient believe is effective. Placebo effectsplus disease natural history and regression to the mean can result in high rates ofgood outcomes, which may be misattributed to specific treatment effects. The truecauses of improvements in pain after treatment remain unknown in the absence ofindependently evaluated randomized controlled trials.

(JAMA. 1994;271:1609-1614)

TWO questions are ofmajor interest toclinicians and researchers with respectto pain treatments: What is the efficacyof a specific treatment (underwhat con-

From the Departments of Psychiatry and BehavioralSciences (Dr Turner), Rehabilitation Medicine (DrsTurner and Fordyce), Medicine (Dr Deyo), Health Ser-vices (Dr Deyo), Neurological Surgery (Dr Loeser), andAnesthesiology (Dr Loeser), and the MultidisciplinaryPain Center (Drs Turner and Loeser), University ofWashington, Seattle; the Health Services Research andDevelopment Field Program, Veterans Affairs MedicalCenter, Seattle (Dr Deyo); and the Center for HealthStudies, Group Health Cooperative of Puget Sound,Seattle (Dr Von Korff).Reprint requests to the Department of Psychiatry

and Behavioral Sciences RP-10, School of Medicine,University of Washington, Seattle, WA 98195 (DrTurner).

ditions and for what patients will it im¬prove certain dimensions of outcome),and why do patients improve with it(what is the mechanism)? There arethree general reasons for clinical im¬provement in a patient's condition.

1. Natural history and regression tothe mean. Most acute and some chronicpain problems resolve on their own ir¬respective of treatment.1 Many individu¬als have recurrent episodes ofpain suchas headache or low back pain, inter¬spersed with no or minimal pain. Pa¬tients with chronic conditions typicallyhave fluctuating symptoms and seekmedical care (and enroll in research stud¬ies) when symptoms are at their worst.Thus, the next change is likely to be an

improvement. This tendency ofextremesymptoms or findings to return towardthe individual's more typical state isknown as regression to the mean.2 Ap¬parent improvement may also reflectmeasurement error or random variationin patient symptoms over time.2

2. Specific effects of treatment, at¬tributable to the characteristic contentof the intervention.

3. Nonspecific effects of treatment,attributable to factors other than spe¬cific active components. These includephysician attention, interest, and con¬cern in a healing setting; patient andphysician expectations of treatment ef¬fects; the reputation, expense, and im-pressiveness ofthe treatment; and char¬acteristics of the setting that influencepatients to report improvement. Theterm placebo effect is often used syn¬onymously with nonspecific effects.It is helpful for clinicians to know the

contributions of each of these processesto treatment effects in order to make op¬timal treatment decisions. It is also es¬sential for investigators to understandthe extent to which placebo effects canaccount for improvements observed inclinical studies. The purpose of this ar¬ticle is to review the literature on placeboeffects, with an emphasis on their mag¬nitude and duration, the conditions influ¬encing them, proposed explanations, andtheir implications for pain treatment re¬search. Although this article pertains topain problems in general, low back painis used frequently as a specific examplebecause it is highly prevalent, costly, anda leading reason for seeking health care.

METHODSTo identify relevant English-language

articles for this review, the MEDLINE

by guest on 14 December 2010jama.ama-assn.orgDownloaded from

bibliographie database (1980 through1993) and the PsycLIT database (1967through 1993) were searched using theterm placebo effect. The MEDLINEsearch yielded 163 articles, and the Psyc¬LIT search yielded 41 articles (nine ofwhich were also found in MEDLINE).Additional articles and books were iden¬tified from personal files, bibliographyreviews, and requests to professionalswith expertise in the area. Three booksand 75 articles were read for this review.

RESULTSDefinitionsA placebo is an intervention designed

to simulate medical therapy, but not be¬lieved (by the investigator or clinician)to be a specific therapy for the targetcondition.3 It is used either for its psy¬chological effect or to eliminate observerbias in an experimental setting.3 Alter¬natively, it could be a treatment nowbelieved to be inefficacious, though be¬lieved efficacious at the time of use.3 Aplacebo effect is a change in a patient'sillness attributable to the symbolic im¬port of a treatment rather than a spe¬cific pharmacologie orphysiological prop¬erty.3 Thus, a placebo effect does notrequire a placebo. A placebo responserefers to any change in patient behavioror condition following the administra¬tion of a placebo.4 The literature doesnot always use these distinctions, andthere are many misconceptions, includ¬ing the following beliefs: (1) about onethird of patients will have a placebo re¬

sponse in any clinical trial; (2) placeboeffects are necessarily brief; (3) certainpersonality types are more likely to beplacebo responders; (4) placebo respond-ers had nothing wrong with them tobegin with; and (5) giving a placebo isthe same as doing nothing.Placebo Effects of MedicalTreatmentsThewidely accepted one-third placebo

response rate is based on the classicarticle by Beecher.5 This was a reviewof 15 studies of patients suffering a va¬

riety of conditions (postoperative pain,cough, angina pectords, headache, drug-induced mood changes, seasickness, anxi¬ety and tension, and common cold). Onaverage, symptoms were "satisfactorilyrelieved" by the placebo in 35% of pa¬tients in these studies, but the placeboresponse rate ranged from 15% to 58%.Wide variation in placebo response rateshas since been observed in other set¬tings aswell. The Table shows examplesof response rates after sham treatmentsfor painful conditions and in studies oftreatments initially considered effica¬cious but later shown to be no better

than placebo. The combined results forthe treatments reviewed by Roberts etal6 that were originally believed effica¬cious but later abandoned averaged 70%excellent or good outcomes, presumablyreflecting placebo and natural historyeffects. Sham treatments can also pro¬duce excellent results. For example, 64%ofpatients who underwent a sham tooth-grinding procedure for myofascial paindysfunction (temporomandibular disor¬der) reported total or near-total symp¬tom remission.9In sum, rates ofgood patient outcomes

after treatments that have no specifictherapeutic effects vary considerablyacross studies, but are strikingly high on

average. Even patients with a long his¬tory of back pain show clinically and sta¬tistically significant improvement withplacebo. Deyo et al10·11 found that scoreson measures of pain severity, pain fre¬quency, and functional status improved,on average, 20% to 40% after patientswith back pain received sham transcu-taneous electrical nerve stimulation plushot packs, despite the chronicity of theirpain (average duration, 4 years).Placebo Effects of Surgery

Beecher12 emphasized that surgerycould evoke a placebo effect and urgedcaution in interpreting the benefit ofnewoperations. Similarly, Spiro13<p42> wrotethat "skeptics have long noted that anoperation, particularly a new one, seemsto bring benefit for several years until itis reevaluated and then often abandoned."He noted that new operations are oftenassociated with a new diagnostic deviceyielding information that is interpretedas explaining a pain problem. Attemptsare then made to correct the problem byoperations or drugs. Spiro suggested thatthe experience ofsurgery and the symbolof the scarmust themselves be importantsources of pain relief.In the 1950s, there were two double-

blind randomized trials of internalmam¬mary artery ligation vs skin incision(with vessel exposure but no ligation)for patients with angina pectoris.7·8 Atthe time, internal mammary artery li¬gation was believed to help angina pec¬toris by increasing coronary artery bloodflow through increased collateral circu¬lation. As summarized in the Table, thesestudies demonstrated substantial andsustained improvement in angina afterskin incision alone.Placebo effects of back surgery are

suggested by Spangfort's14 review oflong-term outcomes of 2504 diskecto-mies for lumbar disk disease. Completerelief of sciatica was noted in 37% andcomplete relief of back pain in 43% ofpatients who had no disk herniation(negative surgical exploration). There

is no known therapeutic effect of sur¬gical exploration of the lumbar spine;changes in patient status were mostlikely attributable to placebo effect andnatural history.The success rates after sham or dis¬

credited procedures may be comparedto the success rates in spine surgerycase series. Across 74 studies of surgeryfor lumbar spinal stenosis, an average of64% of patients had good or excellentoutcomes.15 Similarly, 68% of patientshad good or excellent results among 47studies of lumbar spinal fusion.16 Thesefigures reflect outcomes reported atlong-term follow-up; the absence of ran¬domized controlled trials precludes theinterpretation of the outcomes as re¬

sulting from specific surgery effects as

opposed to placebo effects plus naturalhistory. However, the figures are simi¬lar to the average 70% excellent or goodoutcomes for several abandoned medi¬cal and surgical therapies.6In sum, nonspecific influences plus natu¬

ral history and regression to the mean

play an important role in pain relief aftersurgery. Important nonspecific influenceslikely include subjects' and surgeons' ex¬pectations of improvement. This situa¬tion with respect to low back surgery ishighlighted further by the weak associa¬tion between imaging test results andsymptoms,1·17"21 and between technical suc¬cess ofsurgery (eg, solid fusion) and symp¬tom improvement.22·23Pharmacokinetics of PlaceboResponsePlacebos have demonstrable time-ef¬

fect curves and peak, cumulative(greater effects with repeated adminis¬trations), and carryover effects after ces¬sation of treatment, which mimic thoseof active medications.24 When varyingdoses of analgesic followed by a placeboare administered, patients' placebo re¬

sponses correspond in degree of painrelief over time to their original dosageofanalgesic.25 Dose-response effects havealso been demonstrated; for example,two placebo capsuleswere shown to havemore pronounced effects than one.26Placebos are associated with side ef¬

fects, especially drowsiness, headaches,nervousness, insomnia, nausea, and con¬

stipation.27 Perceptual characteristics ofdrug preparations play a role in indi¬viduals' responses. Larger capsules tendto be viewed as stronger, yellow cap¬sules tend to be perceived as stimulantsor antidepressants, and white capsulestend to be perceived as analgesics ornarcotics.28 Injections may producelarger effects than do pills.26The duration of response to placebos

has not been studied extensively. Patientswith painful diabetic neuropathy who re-

by guest on 14 December 2010jama.ama-assn.orgDownloaded from

Placebo Effects in Studies of Medical and Surgical Treatments for Painful Conditions

Source Condition Treatment(s) % Improved CommentsRoberts et ale (review of treatments Herpes simplex virusoriginally believed efficacious but Infectionslater found to be ineffective)

Levamisole* 85% excellent or good Average across uncontrolledtrials that asserted the efficacyof levamisole

Photodynamic inactivatlontreatment*

85%-100% excellent or good Average across 5 uncontrolledtrials

Topical application oforganic solvents*

83% excellent or good Average across 5 uncontrolledtrials

Duodenal ulcers Gastric freezing* 98%-100% marked or

complete relief65% good/excellent

Results in initial studies

Average across 8 studies thatasserted the efficacy of gastricfreezing

Angina pectoris Internal mammary arteryligation

63% significant improvement, 34%decrease in nitroglycerine use

During first 6 mo after surgery

Skin incision only 56% significant improvement, 42%decrease in nitroglycerine use

Angina pectoris Internal mammary arteryligation

After surgery, 100% improved During year after surgery, 69%reported over 50%improvement in angina

Skin incision only After surgery, 100% improvedin exercise tolerance,nitroglycerine use, and angina

During year after surgery, 100%reported over 50%improvement in angina

Myofascial pain dysfunction Sham tooth-grinding(temporomandibulardisorder)

64% total or near-totalsymptom remission

*Treatment subsequently found to be no better than placebo in controlled trials.

ceived a placebo reported a decrease inpain intensity for the first 3 weeks, fol¬lowed by a partial return toward baselinelevels during the next 3 weeks.29 Clini¬cally significant improvement in anginasymptoms was maintained as long as 1year after sham surgery.8 These studiesdo not allow for a separation of placebofrom natural history effects.

Nocebo EffectsNonspecific influences of treatments

may produce adverse effects, sometimesreferred to as "nocebo effects." The over¬all incidence ofadverse events in healthyvolunteers during placebo administra¬tion was 19% in a review of 109 double-blind drug trials.30 Placebos can alsomake preexisting symptoms worse. Forexample, in a double-blind study31 of amagnetic device for which pain-reliev¬ing qualities were claimed, 13 of 58 painpatient subjects discontinued treatmentafter one or two treatments becausetheir pain was worse. Six months later,three of these patients believed thetreatment made their pain permanentlyworse. Placebos can also produce pain innormal subjects. Headaches were re¬

ported by 70% of students told that a

(nonexistent) electric current was pass¬ing through their heads.32Very little research has focused on

negative nonspecific influences in medi¬cine. One general practitioner randomlyassigned his patients who had symptomsbut no abnormal signs and in whom nodefinite diagnosis could be made to a posi¬tive or a negative encounter with him.33In the positive encounter, patients weregiven a diagnosis and told they would bebetter in a few days. In the negative en-

counter, the doctor told patients he wasnot certain what was the matter withthem. Two weeks later, 64% of the posi¬tive group, but only 39% of the negativegroup, reported that they had gotten bet¬ter (P=.001). The author speculated thatthese minor illnesses would be expectedto resolve spontaneously by 2 weeks inthemajority ofpatients, and that the 61%nonimprovement rate in the negative en¬counter group reflected adverse effectsof the encounter.

FACTORS INFLUENCINGPLACEBO RESPONSESPatient FactorsEfforts to identify personality, demo¬

graphic, and other characteristics thatpredict placebo responses have had littlesuccess.34 In fact, individuals tend not tobe consistent about showing placebo re¬

sponses across placebo administra¬tions.36·36 However, patient expectationsoftreatment effects clearly influence theirresponses. For example, when subjectswere given a pill containing only a mag¬net tomeasure stomach contractions, thecontractions increased, decreased, or didnot change according to the effects theywere told the pill would cause.37 In asth¬matic patients, isotonic saline producedincreases or decreases in airway resis¬tance according to what patients weretold to expect.38 Further, when patientswere given a true bronchodilator, its ef¬fects were about twice as great if pa¬tients were told it would produce thiseffect than if they were told it wouldproduce the opposite effect. The patient'spositive attitude toward the provider andtoward the treatment have been shownto predict improvement in studies ofpsy-

chiatric outpatients treated with placebo,psychotropic drugs, or psychotherapy.34There is also some evidence that highlyanxious subjects show the greatest pla¬cebo responses.34Highly compliant patients may have

better outcomes than noncompliant pa¬tients, even when complying with a pla¬cebo. In a randomized trial to evaluatethe efficacy of lipid-lowering drugs inthe therapy of coronary heart disease,patients in the placebo arm were di¬vided between those who were highlycompliant (took at least 80% of placebocapsules) and those who were less com¬

pliant (took less than 80% of capsules).39Even after controlling for 40 known or

suspected coronary risk factors, the pla¬cebo noncompliers had a 5-year mortal¬ity rate 57% higher than that of thecompilers.Wemay hypothesize that pla¬cebo effects had some effect on mortal¬ity, or that patient compliance relatedto other characteristics associated withmortality, but not assessed in the study.

Provider FactorsThe provider's warmth, friendliness,

interest, sympathy, empathy, prestige,and positive attitude toward the patientand toward the treatment are associatedwith positive effects of placebos as wellas ofactive treatments.34 The importanceof provider expectations was illustratedin a study of a new antihypertensivedrug.40 In the middle of this double-blindstudy, partners of the enthusiastic phy¬sician administering the drug broke thecode. Without tellinghimwhich pillswerethe placebo and which were the drug,they told him that the drug, though ef¬fective, appeared similar to existing

by guest on 14 December 2010jama.ama-assn.orgDownloaded from

drugs. Although less enthusiastic, theydecided to complete the study. The dif¬ference between the drug and placebowas maintained, but there was an im¬mediate andmarked increase in the bloodpressures of both groups.In a double-blind study of dental ex¬

tractions,41 placebo responseswere com¬

pared for patients in two groups: thosewhose clinicians knew they would ad¬minister a narcotic analgesic, a placebo,or a narcotic antagonist vs those whoseclinicians knew they would administeronly a placebo or narcotic antagonist.Placebo patients in the first group hadsignificantly less pain. Because the twoplacebo groups differed only in the cli¬nicians' knowledge of the range of pos¬sible treatments, this knowledge seemsto have resulted in subtle behaviors thatinfluenced patient responses.

EXPLANATIONS FORPLACEBO EFFECTSDecreased AnxietyStress and anxiety adversely affect sev¬

eral physiological processes and increasesymptom reporting. Placebos seem to bemost effective forhighly anxious subjects,and placebo effects are often attributedto anxiety reduction and associated de¬creased suffering.42 Placebos have beenshown to decrease anticipatory anxiety.43However, it is not clear whether anxietyreduction is a cause of the placebo effect,or a component of it.44ExpectationsThere are several possible explana¬

tions for how subject and researcher orclinician expectancies influence placeboeffects. A patient's expectation thattreatment will relieve symptoms mayreduce anxiety and thus amelioratesymptoms. Expectancy of improvementmay result in the patient's viewing thepain problem more positively and asmore controllable. Thus, patients maybe more likely to notice small improve¬ments, to disregard negative events, andto interpret ambiguous stimuli favor¬ably.45 Changes in appraisals and ex¬

pectanciesmay lead to beneficial behav¬ior changes. For example, a low backpain patient may resume physical andfunctional activities he or she hadavoided because of fear ofpain or harm.LearningTreatment may have a positive effect

because of its association with effectivetreatments the patient has had before(this learning process is referred to as

conditioning). Thus, inert or neutraldrugs, procedures, people, and placescan come to function as conditionedstimuli or discriminative stimuli for the

alleviation of symptoms, if they havebeen associated repeatedly with power¬ful unconditioned stimuli (eg, penicillin,nitroglycerine, analgesics) that reliablyrelieve symptoms.46 Further, neutralstimuli (eg, the physician, the physicalexamination, and medication prescrip¬tion) associated with the reduction ofunpleasant symptomsmay acquire posi¬tive conditioned properties for healingand anxiety reduction.Experiments have demonstrated that

placebo responses can be conditioned.47Furthermore, the direct experience ofconditioning appears to be more pow¬erful than expectancy formed throughverbal persuasion.48 Past treatment re¬sponses may influence a patient's re¬

sponses to subsequent treatments in a

positive or negative manner, dependingon the prior history. This raises one pos¬sible explanation ofwhy repeated backsurgery yields progressively poorer re¬sults and sometimes makes patientsworse rather than better.49

Endorphin EffectsIt has been suggested that placebo

responses may be mediated by endog¬enous opiate release in the central ner¬vous system.50 However, subsequentstudies have yielded contradictory re¬

sults, and the role of endogenous opiateprocesses is unclear at present.42,51·52ConclusionsThese models are not mutually exclu¬

sive, and each of these factors may playa role in placebo effects. In fact, Rob¬erts53 has argued persuasively for drop¬ping the term placebo altogether. Theunderstandingofplacebo effectsmaybeadvanced by studies undertaken to ex¬amine the variety ofpotential influencesother than specific treatment effects on

patient outcomes, including natural his¬tory, regression to the mean, patientexpectations, provider expectations,characteristics of the treatment situa¬tion that influence patients and physi¬cians to behave in certain ways (eg, toreport improvement), conditioning, andpsychophysiological states such as anxi¬ety and relaxation.53IMPLICATIONS FORRESEARCH DESIGNPlacebo effects are found with drugs,

medical treatments, surgery, biofeed¬back, psychotherapy, and even diagnos¬tic tests.46·54 Thus, placebo effects can playa role in all interactions between pro¬vider and patient. Only independentlyevaluated (ie, not by the treating clini¬cians, and preferably by observers un¬aware ofthe treatment assignment) ran¬domized controlled trials can establishan effect of a treatment above and be-

yond natural history of the condition andnonspecific effects. Random assignmentofpatients to treatment and control con¬ditions is essential to reduce systematicbias in group membership, which maylead to differential improvement attrib¬utable to differences in patient charac¬teristics rather than in the treatment.However, even in randomized controlledtrials, physician and patient know thereis a sham treatment and a real treat¬ment, and outcomes are influenced bytheir expectancies and beliefs aboutwhichtreatment the patient received. If eitheror both can guess (eg, by side effects)which treatment the patient received, orif one treatment is more credible, thismaybias the study results. To the extentthat the patient or clinician believes atreatment may be ineffective, the powerof nonspecific effects will be reduced orunderestimated.Therefore, the control treatment in a

trial should be as similar as possible tothe active treatment, to create similarexpectations. Patients receiving shamtherapy should have visit frequency, con¬tact, and support equivalent to that inthe active therapy condition. It can bedifficult to create placebo controls thatappear to be active treatments, but cre¬ative placebos have been devised (eg,sham transcutaneous electrical nervestimulation, the use of misplaced nee¬

dling as a control for acupuncture, theuse of subtherapeutic weight as a con¬trol for traction, and the use of massageas a control for spinal manipulation). Tri¬als in which control treatments mimicthe active intervention typically havefound less advantage of the active treat¬ment over the control than have trialswith obviously different types of therapyor with inert placebo controls.55,66A completely untreated group (eg,

waiting list) is not the same as a placebo-treated group. A waiting list conditioncontrols for the effects of the passage oftime, but not for patient expectations.However, an untreated group conditionin addition to a placebo group can helpdistinguish nonspecific effects from natu¬ral history. For chronic conditions, longbaselineswithmultiple measures oftheoutcome variable before treatment canreveal changes in the absence of treat¬ment and thereby help to estimate themagnitude of regression to the mean as

a source of within-patient change.57·58Ethical and practical factors make it

difficult to conduct surgery trials withsham controls. For situations in which itis not possible to have a sham surgerycontrol condition, randomized trials ofsurgery vs credible alternative nonsur-

gical therapies may be feasible, as hasbeen done with coronary artery bypasssurgery59 and diskectomy.60

by guest on 14 December 2010jama.ama-assn.orgDownloaded from

IMPLICATIONS FOR CLINICIANSThe administration of any treatment,

including surgery, has physiological andpsychological effects on the patient, andthese are interrelated. There are placeboeffects whenever the patient and the cli¬nician perceive the treatment as effec¬tive. These effects can be potent and canlead to erroneous claims of efficacy forany type of treatment. These effects arelikely to be strongest when the patient isanxious, the physician is perceived as hav¬ing great expertise, the patient and phy¬sician believe the treatment is powerful,and the treatment is impressive and ex¬

pensive. Placebo effects act synergisti-cally with active treatment effects andnatural history to influence patient out¬comes. Physicians should use these non¬

specific effects to their (and theirpatients')advantage. However, it is a gross error touse a placebo to assess whether a pa¬tient's pain or disease is "real," and todismiss or delegitimize the complaint onthe basis of a placebo response.Physicians who use inactive treat¬

ments in the hopes ofproducing positiveplacebo effects run several risks. Pa¬tients may feel deceived if they discoverthey have been treated with a placebo.The placebo can produce adverse reac¬tions. Failure to improve as expectedmay cause the patient to view his or herproblem as more serious, and the pa¬tient may consequently become moreconcerned about it.45 Failure to improvemay also increase the risk of not im¬proving with subsequent treatments.

Some patients with chronic pain mayfail to respond to a treatment that iseffective for other patients. This may beespecially likely, due to previous learn¬ing experiences, if the patient previouslyresponded poorly to different treat¬ments, including those that were, un¬known to the physician, no more thanplacebos. These patients may also beinfluenced by psychosocial, economic,and other factors that cause them tocontinue to fail additional treatments.

CONCLUSIONSIn most pain treatment and research

situations, nonspecific effects oftreatmentareunderestimated, andpatient improve¬ment is likely regardless of treatment.Nonspecific effects, natural history, andregression to the mean must be distin¬guished from specific effects when medi¬cal and surgical treatments are evalu¬ated. It cannot be assumed that a treat¬ment whose response rate is more thanone third is better than placebo. The ex¬tent to which patient outcomes after amedical or surgical treatment reflect non¬specific effects, regression to the mean,natural history, or specific treatment ef-

fects is unclear in the absence of random¬ized controlled trials with outcomes as¬sessed by persons blind to the patient'streatment. Our reviews15·16·61 of the pub¬lished literature on the treatment of lowback pain have repeatedly found that few,if any, of the articles suggest that out¬comes could be attributable to naturalhistory or nonspecific effects. These ef¬fects are likely to be substantial, may besustained over long periods of time, andmay explain some or all of the benefitsattributed to treatment. There are im¬portant implications for research and clini¬cal training in all areas of medicine. Thequality of the interaction between thephysician and patient can be extremelyinfluential in patient outcomes, and, insome (perhaps many) cases, patient andprovider expectations and interactionsmay be more important than specifictreatments.Analysis and interpretation ofplacebo-

related findings brings us to consider thenature of illness and disease and the re¬

lationships between body processes andthe environment. Confusion and uncer¬

tainty among physicians and other healthcare professionals about placebo effectssuggest an inadequate appreciation of theinteraction of body processes with pastexperience, anticipated events, and im¬mediate environmental influences. Fur¬ther, the body's capacity to modulatesymptoms and suffering involves morethansimply "psychological factors,"wherethose are seen as traits or personalitycharacteristics. Symptoms, illness, andtheir changes over time reflect complexinteractions between anatomical and neu-rophysiological processes, on the one hand,and cognitive-behavioral and environmen¬tal factors on the other. The findings re¬viewed herein support the thesis thatthese factors are inextricably intertwined.This project was supported by grant HS 06344

from the Agency for Health Care Policy andResearch (the Back Pain Outcome AssessmentTeam) and the Health Services Research and De¬velopment Field Program, Seattle (Wash) Veter¬ans Affairs Medical Center.

References1. Waddell G. A new clinical model for the treat-ment of low-back pain. Spine. 1987;12:632-644.2. Whitney CW, Von Korff M. Regression to themean in treated versus untreated chronic pain. Pain.1992;50:281-285.3. BrodyH. Placebo effect: an examination of Gr\l=u"\n-baum's definition. In: White L, Tursky B, SchwartzGE, eds. Placebo: Theory, Research, and Mecha-nisms. New York, NY: Guilford Press; 1985:37-58.4. Ross S, Buckalew LW. Placebo agentry: assess-ment of drug and placebo effects. In: White L,Tursky B, Schwartz GE, eds. Placebo: Theory, Re-search, and Mechanisms. New York, NY: GuilfordPress; 1985:67-82.5. BeecherHK. The powerful placebo. JAMA. 1955;159:1602-1606.6. Roberts AH, Kewman DG, Mercier L, Hovell M.The power of nonspecific effects in healing: impli-cations for psychosocial and biological treatments.

Clin Psychol Rev. 1993;13:375-391.7. Cobb LA, Thomas GI, Dillard DH, MerendinoKA, Bruce RA. An evaluation of internal-mammary-artery-ligation by a double-blind technic. N Engl JMed. 1959;260:1115-1118.8. Dimond EG, Kittle CF, Crockett JE. Comparisonof internal mammary ligation and sham operation forangina pectoris. Am J Cardiol. 1960;5:483-486.9. Goodman P, Green CS, Laskin DM. Response ofpatientswithmyofascial pain-dysfunction syndrometo mock equilibration. J Am Dent Assoc. 1976;92:755-758.10. Deyo RA, Walsh NE, Martin DC, SchoenfeldLS, Ramamurthy S. A controlled trial of transcu-taneous electrical nerve stimulation (TENS) andexercise for chronic low back pain. N Engl J Med.1990;322:1627-1634.11. Deyo RA. Practice variations, treatment fads,and rising disability: do we need a new clinicalresearch paradigm? Spine. 1993;18:2153-2162.12. Beecher HK. Surgery as placebo. JAMA. 1961;176:1102-1107.13. Spiro HM. Doctors, Patients, and Placebos. NewHaven, Conn: Yale University Press; 1986.14. Spangfort EV. The lumbar disc herniation: acomputer-aided analysis of 2504 operations. ActaOrthop Scand. 1972;142(suppl):1-95.15. Turner JA, Ersek M, Herron L, Deyo R. Sur-gery for lumbar spinal stenosis: attempted meta\x=req-\analysis of the literature. Spine. 1992;17:1-8.16. Turner JA, Ersek M, Herron L, et al. Patientoutcomes after lumbar spinal fusions. JAMA. 1992;268:907-911.17. FullenloveTM,Williams AJ. Comparative roent-gen findings in symptomatic and asymptomaticbacks. Radiology. 1957;68:572-574.18. Boden SD, Davis DO, Dina TS, Patronas NJ,Wiesel SW. Abnormal magnetic resonance scans ofthe lumbar spine in asymptomatic subjects: a pro-spective investigation. JBone Joint SurgAm. 1990;72:403-408.19. Magora A, Schwartz A. Relation between thelow back pain syndrome and x-ray findings. ScandJ Rehabil Med. 1976;8:115-125.20. SplithoffCA. Lumbosacraljunction: roentgeno-graphic comparison of patients with and withoutbackaches. JAMA. 1953;152:1610-1613.21. Wiesel SW, Tsourmas N, Feffer HL, Citrin CM,Patronas N. A study of computer-assisted tomogra-phy,I:theincidenceofpositiveCATscansinanasymp-tomatic group of patients. Spine. 1984;9:549-551.22. De Palma AF, Rothman RH. The nature ofpseudarthrosis. Clin Orthop. 1968;59:113-118.23. Flatley TJ, Derderian H. Closed loop instru-mentation of the lumbar spine. Clin Orthop. 1985;196:273-278.24. Lasagna L, Laties VG, DohanJL. Further stud-ies on the 'pharmacology' of placebo administra-tion. J Clin Invest. 1958;37:533-537.25. Laska E, Sunshine A. Anticipation ofanalgesia:a placebo effect. Headache. 1973;13:1-11.26. Blackwell B, Bloomfield SS, Buncher CR. Dem-onstration tomedical students ofplacebo responsesand non-drug factors. Lancet. 1972;1:1279-1282.27. Pogge RC. The toxic placebo, I: side and toxiceffects reported during the administration of pla-cebo medicine. Med Times. 1963;91:1-6.28. Buckalew LW, Coffield KE. An investigation ofdrug expectancy as a function of capsule color andsize and preparation form. J Clin Psychopharma-col. 1982;2:245-248.29. Max MB, Culname M, Schafer SC, et al. Ami-triptyline relieves diabetic neuropathy pain in pa-tients with normal or depressed mood. Neurology.1987;37:589-596.30. Rosenzweig P, Brohier S, Zipfel A. The placeboeffect in healthy volunteers: influence of experi-mental conditions on the adverse events profileduring phase I studies. Clin Pharmacol Ther. 1993;54:578-583.31. Long DM, Uematsu S, Kouba RB. Placebo re-

sponses to medical device therapy for pain. Ster-eotact FunctNeurosurg. 1989;53:149-156.32. Schweiger A, Parducci A. Nocebo: the psycho-logic induction ofpain.PavJBiol Sci. 1981;16:140-143.

by guest on 14 December 2010jama.ama-assn.orgDownloaded from

33. Thomas KB. General practice consultations: isthere any point in being positive? BMJ. 1987;294:1200-1202.34. Shapiro AK, Shapiro E. Patient-provider rela-tionships and the placebo effect. In: Matarazzo JD,Weiss SM, Herd JA, Miller NE, Weiss SM, eds.Behavioral Health:A Handbook ofHealth Enhance-ment and Disease Prevention. New York, NY:Wiley-Interscience; 1984:371-383.35. Lasagna L, Mosteller F, von Felsinger JM,Beecher HK. A study of the placebo response. AmJ Med. 1954;16:770-779.36. Liberman R. An experimental study of the pla-cebo response under three different situations ofpain. J Psychiatr Res. 1964;2:233-246.37. Sternbach RA. The effects of instructional setson autonomic responsivity. Psychophysiology. 1964;1:67-72.38. Luparello T, Leist N, Lourie CH, Sweet P. Theinteraction of psychologic stimuli and pharmaco-logic agents on airway reactivity in asthmatic sub-jects. Psychosom Med. 1970;32:509-513.39. Coronary Drug Project Research Group. In-fluence of adherence to treatment and response ofcholesterol onmortality in the coronary drug project.N Engl J Med. 1980;303:1038-1041.40. Shapiro AP, Myers T, Reiser MF, Ferris EB.Comparison of blood pressure response to Veriloidand to the doctor. Psychosom Med. 1954;16:478-488.41. Gracely RH, Dubner R, Deeter WR, WolskeePJ. Clinicians' expectations influence placebo an-

algesia. Lancet. 1985;1:43.42. Evans FJ. Expectancy, therapeutic instructions,

and the placebo response. In: White L, Tursky B,Schwartz GE, eds. Placebo: Theory, Research, andMechanisms. New York, NY: Guilford Press; 1985:215-228.43. Reiss S. Pavlovian conditioning and human fear:an expectancy model. Behav Ther. 1980;11:380-396.44. Wall PD. Pain and the placebo response. CibaFound Symp. 1993;174:187-211.45. Peck C, Coleman G. Implications of placebotheory for clinical research and practice in painmanagement. Theor Med. 1991;12:247-270.46. Wickramasekera I. A conditioned responsemodel of the placebo effect: predictions from themodel. In: White L, Tursky B, Schwartz GE, eds.Placebo: Theory, Research, and Mechanisms. NewYork, NY: Guilford Press; 1985:255-287.47. Voudouris NJ, Peck CL, Coleman G. Condi-tioned response models of placebo phenomena: fur-ther support. Pain. 1989;38:109-116.48. Voudouris NJ, Peck CL, Coleman G. The roleof conditioning and verbal expectancy in the pla-cebo response. Pain. 1990;43:121-128.49. Waddell G, Kummel EG, Lotto WN, GrahamJD, Hall H, McCulloch JA. Failed lumbar disc sur-gery and repeat surgery following industrial inju-ries. J Bone Joint Surg Am. 1979;61:201-207.50. Levine JD, Gordon NC, Fields HL. The mecha-nism of placebo analgesia. Lancet. 1978;2:654-657.51. GracelyRH. Verbal descriptormeasures ofpainclarify mechanisms of analgesia due to narcotics,brain stimulation and placebo. Anesth Prog. 1987;34:113-127.52. Grevert P,Goldstein A. Placebo analgesia, nalox-

one, and the role of endogenous opioids. In: WhiteL, Tursky B, Schwartz GE, eds. Placebo: Theory,Research, and Mechanisms. New York, NY: Guil-ford Press; 1985:332-350.53. Roberts AH. 'The powerful placebo' revisited:what power? what placebo? Presented at the Sev-enth World Congress of the International Associa-tion for the Study of Pain; August 23, 1993; Paris,France.54. Sox HC, Margulies I, Sox CH. Psychologicallymediated effects of diagnostic tests. Ann InternMed. 1981;95:680-685.55. Deyo RA. Clinical research methods in low backpain. Physical Med Rehabil. 1991;5:209-222.56. Thomson R. Side effects and placebo amplifi-cation. Br J Psychiatry. 1982;140:64-68.57. McDonald CJ, Mazzuca SA. How much of theplacebo 'effect' is really statistical regression? StatMed. 1983;2:417-427.58. Miller NE. Placebo factors in treatment: viewsofa psychologist. In: Shepherd M, Sartorius N, eds.Non-Specific Aspects of Treatment. Toronto, On-tario: Hans Huber Publishers; 1989:39-51.59. CASS Principal Investigators. Coronary Ar-tery Surgery Study (CASS): a randomized trial ofcoronary artery bypass surgery. Circulation. 1983;68:939-950.60. Weber H. Lumbar disc herniation: a controlled,prospective study with ten years of observation.Spine. 1983;8:131-140.61. Hoffman RM, Wheeler KJ, Deyo RA. Surgeryfor herniated lumbar discs: a literature synthesis.J Gen Intern Med. 1993;8:487-496.

by guest on 14 December 2010jama.ama-assn.orgDownloaded from