Randomized trial vs. observational study of acupuncture for migraine found that patient characteristics differed but outcomes were similar

Journal of Clinical Epidemiology 60 (2007) 280e287

Randomized trial vs. observational study of acupuncture for migrainefound that patient characteristics differed but outcomes were similar

Klaus Lindea,*, Andrea Strenga, Andrea Hoppea, Wolfgang Weidenhammera,Stefan Wagenpfeilb, Dieter Melcharta,c

aCentre for Complementary Medicine Research, Department of Internal Medicine II, Technische Universit€at M€unchen, Munich, GermanybInstitute of Medical Statistics and Epidemiology, Technische Universit€at M€unchen, Munich, Germany

cDivision of Complementary Medicine, Department of Internal Medicine, University Hospital Zurich, Zurich, Switzerland

Accepted 6 June 2006

Abstract

Objective: To compare patient characteristics and outcomes between a randomized controlled trial (RCT) and an observational study(OS) of acupuncture treatment in patients with migraine.

Study Design and Setting: Within the framework of a research program for acupuncture for chronic pain, both an RCT and an OS ofroutine care were performed in patients with migraine. One hundred forty-five patients received acupuncture in the RCT and 732 in the OS.All patients completed a pain questionnaire, including instruments evaluating disability, quality of life, and depression, at baseline and at6-month follow-up.

Results: At baseline, patients participating in the RCT had a significantly longer history of complaints and more headache days butbetter quality of life and fewer depressive symptoms than patients participating in the OS. In unadjusted analyses, improvements after6 months were more pronounced in the OS than in the RCT. After adjustment for potential confounders, the differences decreased but werestill significant for several outcomes.

Conclusion: Participants of the RCT were probably not representative of migraine patients undergoing acupuncture treatment in routinecare under current regulations in Germany. However, changes after treatment were similar in the RCT and the OS of routine care. � 2007Elsevier Inc. All rights reserved.

Keywords: Acupuncture; Migraine; Randomized controlled trial; Observational study; External validity; Generalisability

1. Introduction

Acupuncture is one of the most widespread complemen-tary therapies for the interval treatment of migraine,although the evidence for its effectiveness is not convincing[1]. Throughout the 1990s, the costs of acupuncture therapyadministered by physicians were partially covered by theGerman health insurance funds on an informal basis. Underincreasing pressure to cut health care costs, however, theFederal Committee of Physicians and Health Insurers(Bundesausschuss der €Arzte und Krankenkassen) decidedin October 2000 that the scientific evidence supportingacupuncture was insufficient to justify routine reimburse-ment. Nevertheless, it recommended special research andreimbursement programs (‘‘Modellvorhaben’’) in order to

* Corresponding author. Tel.: þ49-89-726697-15; fax: þ49-89-

726697-22.

E-mail address: [email protected] (K. Linde).

0895-4356/07/$ e see front matter � 2007 Elsevier Inc. All rights reserved.

doi: 10.1016/j.jclinepi.2006.06.014

better determine the role of acupuncture in the treatmentof chronic pain within the framework of the statutory healthcare system in Germany. The authors of this article coordi-nated such a research program funded by a group of statu-tory sickness funds (‘‘Ersatzkassen’’) that cover a quarter(about 20 million individuals) of the total German popula-tion [2]. The results of this and two similar programs [3,4]will be the basis for a new decision on whether acupuncturewill become part of routine reimbursement in Germany.

The scientific core part of our program consisted of ran-domized controlled trials (RCTs) investigating whetheracupuncture is more effective than no treatment and a shamintervention. However, due to the legal framework of such‘‘Modellvorhaben’’ it was also necessary to perform obser-vational studies (OSs). The results of the OS can providerelevant additional information as the RCTs might haveonly limited external validity for the following reasons:(1) It seemed likely a priori that patients giving consentto a randomized trial would not be representative of those

281K. Linde et al. / Journal of Clinical Epidemiology 60 (2007) 280e287

seeking acupuncture under conditions of routine practice.(2) Well-established guidelines for RCTs in migraine re-commend rather narrow inclusion criteria [5]. In routinepractice many patients have less well-defined illness andcomorbidities. (3) Some standardization of the treatmentprotocol seemed adequate in the randomized trials to war-rant reproducibility and transparency. However, in routinepractice acupuncture is applied in a highly variable manner.(4) In the RCT we tried to minimize cointerventions. Inroutine practice an unknown proportion of patients receivea variety of cointerventions. (5) The cognitive and emo-tional framework (information, uncertainty about groupallocation, standardization of communication and interven-tion, etc.) in an RCT might differ considerably from routinepractice. It is unknown whether these factors have animpact on outcome in patients with migraine.

In the analysis presented in this article, we aim to com-pare characteristics and outcomes of migraine patientsreceiving acupuncture treatment in the RCT and in theOS in routine care. The main results of the RCT and ofthe OS have been published elsewhere [6,7].

2. Methods

2.1. The randomized controlled trial

The RCT was a three-armed multicenter 28-week studycomparing treatment with real acupuncture to treatmentwith minimal acupuncture (a sham intervention with super-ficial needling of nonacupuncture points) and to no treat-ment (waiting list) in patients suffering from migraine.Inclusion criteria were a diagnosis of migraine with orwithout aura, according to the criteria of the InternationalHeadache Society [8]; two to eight migraine attacks permonth during the previous 3 months and during the base-line period; age 18e65 years; duration of migraine at least12 months; completion of a baseline headache diary; andwritten informed consent. The main exclusion criteria wereas follows: interval headaches or additional tension-typeheadache on more than 10 days per month, inability to dis-tinguish between migraine attacks and additional tension-type headache, secondary headaches, onset of headachesafter age of 50 years, use of analgesics on more than 10days per month, and prophylactic headache treatment withdrugs during the previous 4 weeks.

After a baseline period of 4 weeks, patients werecentrally randomized. One hundred forty-five patients wereallocated to receive acupuncture. The acupuncture interven-tion had been developed in a consensus process withacupuncture experts and societies and was provided by 30well-trained and experienced physicians in 18 centers. Acu-puncture treatment consisted of 12 sessions of 30 minutesduration, each administered over a period of 8 weeks (pref-erably two sessions in each of the first 4 weeks, followed byone session per week in the remaining 4 weeks). All patients

were treated at standardized ‘‘basic’’ points unless explicitreasons for doing otherwise were given. Additional pointscould be chosen individually. Patients were allowed to treatacute headaches as needed, in accordance with the guide-lines of the German Migraine and Headache Society [9].These treatments had to be documented in the headachediary.

All patients completed headache diaries in the 4 weeksbefore randomization (baseline phase), the 12 weeks afterrandomization, and in weeks 21e24 after randomization.In addition, patients were asked to complete a modified ver-sion of the pain questionnaire of the German Society for theStudy of Pain (DGSS) [10] before treatment, after 12weeks, and after 24 weeks. The questionnaire includesquestions on sociodemographic characteristics, numericalrating scales for pain intensity, questions on workdays lost,and global assessments, as well as the following validatedscales: (1) the German version of the Pain Disability Index(PDI) [11], (2) a scale for assessing affective and sensoricaspects of pain (Schmerzempfindungs-Skala SES) [12],(3) the depression scale ADS (Allgemeine Depressionsska-la, the German version of the Center of EpidemiologicalStudies Depression scale CES-D) [13], and (4) the Germanversion of the Short Form-36 scale for assessing health-related quality of life (SF-36) [14].

One hundred forty-four of the 145 patients randomizedto acupuncture attended at least one treatment session, 138(96% of those starting) completed all 12 sessions, and 131(91%) provided data at 24 weeks after randomization.

2.2. The observational study

The OS was embedded in a nationwide reimbursementprogram of a group of German statutory sickness funds.All patients suffering from chronic low back pain, osteoar-thritis pain, or chronic headache for more than 6 monthsand insured by one of the participating statutory sicknessfunds were eligible for acupuncture treatment (paying5e10 V per session out of pocket with approximately25 V per session payed by the statutory sickness funds),if basic data on patient history, treatment, and outcomeswere documented by the physicians. About half a millionpatients and 10,000 physicians participated in the programbetween August 2001 and July 2003. In 36.3% (178,500) ofpatients, headache was reported as the main diagnosis.

Physicians accredited for the program were asked to par-ticipate in an OS collecting more details on a sample ofpatients. To avoid major selection effects and minimizeworkload, physicians had to screen and recruit up to sixpatients, who had been born in a month randomly assignedto the physician by the coordinating study center, for inclu-sion in the OS. General inclusion criteria were age at least18 years, knowledge of German, and the ability to under-stand and complete the questionnaires. An additional selec-tion criterion for the analysis presented here was anexclusive diagnosis of migraine (with or without aura).

282 K. Linde et al. / Journal of Clinical Epidemiology 60 (2007) 280e287

Acupuncture could be provided for up to a maximum of 15treatment sessions; the duration of each session had to be atleast 30 minutes. All other aspects of treatment were at thediscretion of the treating physician. Other treatments (med-ication, physiotherapy, etc.) could be prescribed as needed.

Physicians were asked to document details of thepatients’ history and treatments performed and prescribed.Patients filled out the pain questionnaire of the GermanSociety for the Study of Pain before the first acupuncturesession, after the final session, and 6 months after the firstsession.

Baseline data were collected for 3,822 headache pa-tients. A complete set of questionnaires was available for2,438 patients (64%) treated by 1,615 physicians. Of these,732 patients suffered exclusively from migraine. On aver-age, patients were treated with acupuncture in 8.7 sessions.Five percent of patients stopped the treatment prematurelyor attended fewer than five sessions. Two percent were pre-scribed conventional interval drug treatment, and relaxationwas recommended as an interval treatment to 13%. In 75%of patients, no interval treatment other than acupuncturewas documented.

2.3. Analyses

For the analyses presented below, the data of the 145 mi-graine patients randomized to acupuncture in the RCT andthe 732 migraine patients in the OS were pooled to createa single data set. These data included sociodemographiccharacteristics and data from the pain questionnaire. Since,for logistical reasons, a headache diary could not be used inthe OS, it was not possible to compare frequency data di-rectly. However, we included the number of days with head-ache derived from the headache diary in the RCT and theanswer of patients to the question, ‘‘how many days didyou have pain in the last 4 weeks’’ (asked in the baselineand 6-month questionnaire in the OS), as a crude indicatorof headache frequency. We predefined the Physical Compo-nent Scale of the SF-36 and the PDI as primary outcomes ofinterest. The timing of outcome measurement after treat-ment differed in the RCT (4 weeks after completion of12 sessions of acupuncture within 8 weeks) and the OS(directly after the last acupuncture session; variable durationof treatment). Therefore, the outcome measurement afterapproximately 6 months (24 weeks after randomization inthe RCT and 6 months after inclusion in the OS) was usedas the primary parameter for comparing treatment effects.

Data from the RCT and the OS were compared usingStudent’s t-test, Fisher’s exact test, and the chi-square test.To compare changes from baseline after acupuncture and at6 months (absolute values), analysis of covariance (adjust-ing for potential confounders) was also done. Covariateswere chosen either a priori (age, gender, education, dura-tion of pain, baseline value for the outcome tested, mentalhealth, and baseline depressive symptoms) or based on ob-served baseline differences (P ! 0.1; additional variables

were single household, baseline values for the PDI, andphysical health). If there were significant pairwise correla-tions between variables, interaction terms were also in-cluded. Variables for the final model were selected usingbackward and forward variable selection procedures. Allanalyses were performed using SPSS 11.5 (SPSS, Chicago,IL, USA). P-values given are two sided and subject to a sig-nificance level of 0.05.

3. Results

3.1. Baseline comparability

Patients randomized to acupuncture in the RCT and pa-tients receiving acupuncture in the OS were similar regard-ing age, body mass index, and average and maximal painduring the previous 4 weeks as well as regarding affectiveand sensoric aspects of pain (see Table 1). Patients in theRCT were slightly less often female, lived more often ina single household, and tended to have lower ratings onthe PDI. Highly significant differences were observed foreducation (higher in the RCT), duration since onset of dis-ease (longer in the RCT), physical and mental health ac-cording to the SF-36, and depressive symptoms (all worse

Table 1

Comparison of baseline characteristics of patients receiving

acupuncture in the randomized trial and in the observational study

(values are means 6 standard deviations or percentages)

Randomized

trial, n 5 145

Observational

study, n 5 732 P-value

Female 89% 94% 0.04

Age (years) 43.3 6 11.8 43.8 6 12.5 0.64

Body mass index 23.9 6 4.1 24.0 6 4.0 0.79

Average pain during

prior 4 weeks (NRS)

5.6 6 1.6 5.8 6 2.0 0.17

Maximal pain during prior

4 weeks (NRS)

8.2 6 1.5 8.1 6 1.8 0.55

Secondary school completed 36% 23% 0.002

Single household 22% 14% 0.01

Headaches for more

than 10 years

75% 56% !0.001

PDI 32.8 6 15.5 36.0 6 18.2 0.05

Pain sensation

affective (SES)

55.4 6 9.0 55.6 6 9.2 0.83

Pain sensation

sensoric (SES)

54.7 6 8.7 55.8 6 9.4 0.19

Depressive symptoms (ADS) 50.3 6 8.6 53.2 6 9.1 0.001

SF-36 Physical

Component Scalea41.6 6 7.7 38.7 6 8.4 !0.001

SF-36 Mental

Component Scalea47.6 6 10.1 43.7 6 11.3 !0.001

Number of headache

days in the prior 4 weeksb8.3 6 3.4 6.2 6 3.8 !0.001

NRS 5 numerical rating scale from 0 5 no pain to 10 5 worst imagin-

able pain.a Higher values indicate better state.b Outcome measured with different methods: in the randomized trial

derived from a headache diary, in the observational study from a single

question (headache days in the last 4 weeks).

283K. Linde et al. / Journal of Clinical Epidemiology 60 (2007) 280e287

in the OS). Significantly better values were found for RCTpatients in all subdimensions of the SF-36 except bodilypain (Table 2). Comparison of the number of headache daysreported by the RCT patients in their baseline headache di-aries and by the OS patients when queried about the previ-ous 4 weeks showed significantly more headache days forthe RCT. The histograms indicate that there were no pa-tients with fewer than three headache days in the RCT(as defined by the RCT inclusion criterion of at least twomigraine attacks during the baseline period), while a consid-erable number of patients reported zero to two headachedays in the OS (see Fig. 1).

3.2. Predictors of outcome

In the regression analyses (both if analyses were per-formed for the RCT and the OS separately and if they werepooled), the baseline value of an outcome was found to bethe most important factor predicting change after 6 months(Table 3). Introduction of further factors into the modelsessentially did not increase the coefficient of determination.The final model included age, duration of migraine, base-line scores of the PDI, the Physical Component Scale andMental Component Scale of the SF-36, the interaction termfor age and the Physical Component Scale, and the respec-tive baseline values for secondary outcomes (Table 4).

3.3. Comparison of results

Comparison of unadjusted mean values for changes frombaseline to 6 months yielded significantly better results inOS patients for all predefined outcomes except the PDIand headache days (Table 4). After adjustment for baselinevalues or the selected covariates, the differences were nolonger significant for the Physical Component Scale andMental Component Scale of the SF-36, and for depressivesymptoms. However, the difference in headache days be-came highly significant. This is due to the fact that the re-gression analysis adjusted for the considerable baselinedifferences (higher number of headache days in the RCT;see also Table 1).

Table 2

Comparison of baseline values for SF-36 subdimensions of patients

receiving acupuncture in the randomized trial and in the observational

study (standardized values; values are means 6 standard deviations)

Randomized

trial, n 5 145

Observational

study, n 5 732 P-value

Physical functioning 47.2 6 10.8 43.6 6 13.1 0.002

Physical role function 37.6 6 14.9 32.2 6 14.6 !0.001

Bodily pain 32.4 6 7.2 31.3 6 7.9 0.12

General health perception 47.1 6 10.6 43.4 6 11.3 !0.001

Vitality 42.5 6 9.7 39.6 6 10.7 0.003

Social function 40.0 6 11.5 35.8 6 12.4 !0.001

Emotional role function 45.2 6 13.9 39.1 6 17.7 !0.001

Mental well-being 46.7 6 10.7 42.7 6 11.9 !0.001

Differences between RCT and OS patients were morepronounced (indicating better outcomes in OS patients) ifoutcomes after treatment were analyzed instead of out-comes at 6 months. However, these results are difficult tointerpret as the timing of outcome measurements differedin RCT and OS patients (see Section 2).

4. Discussion

4.1. Summary of main findings

The results of our analysis indicate that (1) patients par-ticipating in the RCT tended to have higher education, longerduration of disease, more headache days but better quality oflife than patients participating in the OS, (2) improvements inclinical outcome measures observed in the OS under condi-tions of routine care were similar to or slightly more pro-nounced than those reported in the RCT. There were nostatistically significant differences for the two main outcomemeasures (Physical Component Scale of the SF-36 and PDI)after adjustment for baseline differences.

4.2. Why a comparison of data from RCTs and OSsis of particular relevance for complementary therapies

RCTs are considered the best method to investigate theeffectiveness of an intervention. This principle clearly alsoapplies to complementary therapies such as acupuncture.However, RCTs might not provide all answers that are rel-evant to decision makers. Despite the lack of unequivocalevidence, acupuncture is widely used in many countries.In Germany, it is estimated that 20,000e50,000 physiciansand an unknown number of non-medically qualified pro-viders use acupuncture [15]. Within the reimbursement pro-gram described above, more than 10,000 physicians treatedover a million patients since 2001. Other German statutorysickness funds run similar programs; we estimate that thetotal number of patients suffering from chronic pain treatedsince 2001 within all of these programs is more than3 million.

New drugs are (or should be) introduced after a long andexpensive research process in routine care only if they haveproven efficacy. Instead, it is typical for complementarytherapies to undergo rigorous evaluation only after theiruse has become so widespread that they cannot be ignoredany longer. The widespread use is associated with highlyvariable practice styles, strong beliefs, and distinct prefer-ences among providers and patients. In this situation, itseems necessary to verify whether the findings of an RCTin a well-defined population investigating one specificintervention are compatible with those obtained from theobservation of routine care. The analyses presented in thispaper suggest that the outcomes regarding effectiveness inour RCT have reasonable external validity despite the sig-nificant differences in patient characteristics.

284 K. Linde et al. / Journal of Clinical Epidemiology 60 (2007) 280e287

Randomized trial (diary)1815129630

num

ber o

f pat

ient

s

30

20

10

0

Observational study (single question)1815129630

num

ber o

f pat

ient

s

100

80

60

40

20

0

Fig. 1. Histogram for the variable number of headache days at baseline. The number of headache days was derived from a headache diary in the randomized

trial and from a single question covering the preceding 4 weeks in the observational study.

4.3. Strengths and limitations

To the best of our knowledge, this is the first time thatcharacteristics and outcomes of patients receiving an inter-val treatment of migraine in experimental conditions andunder conditions of routine practice have been comparedsystematically. The large sample size allowed detection ofsubtle differences and the follow-up at 6 months allowedcomparison of long-term effects. However, the question-naires used as well as the specific framework in whichthe OS was performed have led to some methodologicalshortcomings, which have to be taken into account when in-terpreting our data: (a) The questionnaire used to measureoutcomes was a generic instrument, which is not optimalfor headache patients, although it is recommended by theGerman Society for the Study of Pain for chronic pain ingeneral, including headaches. Scales asking patients to re-port some sort of average level for the previous weeks such

as the PDI or the SF-36 are difficult to rate for migrainepatients who have severe complaints during attacks andno complaints between attacks. Still these instruments aresensitive to change and differentiate between differentheadache diagnoses [7]. (b) It was logistically impossibleto use headache diaries in the OS; therefore, we have onlybasic data on headache frequency. (c) While the follow-uprate in the RCT was high (over 90%), this is not true forthe OS. Therefore, the improvements in the OS could beoverestimated due to selection bias. Further analyses ofour data (comparing patients with and without follow-upin the OS) suggest that the majority of data loss is not dueto dropout of patients unsatisfied with the treatment effect,but due to organizational problems [16]. As the coordinatingresearch center did not have access to names and addressesof patients, the local organization of the OS was completelyin the hands of the participating physicians. A relevant mi-nority of these physicians either did not comply with the

Table 3

Regression coefficients (P-values; adjusted r2; number of valid observations) from linear regression for differences between values at baseline and at 6

months entering only the baseline value as an independent variable or several variables (age; duration; baseline values of the SF-36 Physical Component

Scale, the SF-36 Mental Component Scale, and PDI; and the interaction term for age and the SF-36 Physical Component Scale) in the modela

Baseline values only Complex model

Global assessment n.a. 0.17 (0.002; 0.21; 760)

Average pain in the prior 4 weeks (NRS) 0.47 (!0.001; 0.25; 824) 0.50 (!0.001; 0.24; 749)

Maximal pain in the prior4 weeks (NRS) 0.29 (!0.001; 0.08; 828) 0.34 (!0.001; 0.10; 753)

PDI 0.57 (!0.001; 0.33; 803) 0.60 (!0.001; 0.35; 753)

Pain sensation affective (SES) 0.32 (!0.001; 0.10; 772) 0.35 (!0.001; 0.11; 712)

Pain sensation sensoric (SES) 0.42 (!0.001; 0.18; 758) 0.45 (!0.001; 0.20; 701)

Depressive symptoms (ADS) 0.46 (!0.001; 0.21; 693) 0.48 (!0.001; 0.23; 646)

SF-36 Physical Component Scale 0.48 (!0.001; 0.23; 762) 0.55 (!0.001; 0.29; 738)

SF-36 Mental Component Scale 0.61 (!0.001; 0.37; 762) 0.62 (!0.001; 0.38; 738)

Difference in headache days per 4 weeksb 0.52 (!0.001; 0.27; 781) 0.56 (!0.001; 0.31; 717)

n.a. 5 not applicable; NRS 5 numerical rating scale from 0 5 no pain to 10 5 worst imaginable pain.a Data of patients receiving acupuncture in the randomized trial and in the observational study were pooled for the analysis.b Outcome measured with different methods: in the randomized trial derived from a headache diary, in the observational study from a single question

(headache days in the last 4 weeks).

Table 4

Global assessment of effecti trial and the observational study:

results from unadjusted (t-te

es Adjusted for several covariatesa

, Mean (95% CI) P-value RCT, Mean (95% CI) OS, Mean (95% CI) P-value

Global assessmentb 2.6 (2.4e2.7) 2.3 (2.2e2.7) 0.013

D Average pain in the prior

4 weeks (NRS)

4 (2.3e2.6) 0.035 2.0 (1.6e2.5) 2.5 (2.3e2.6) 0.069

D Maximal pain in the prior

4 weeks (NRS)

0 (2.8e3.2) !0.001 2.1 (1.5e2.6) 3.0 (2.8e3.2) 0.002

D PDI 1 (16.9e19.4) 0.562 16.7 (13.7e19.7) 18.1 (16.8e19.4) 0.395

D Pain sensation affective (S 7 (8.0e9.4) 0.021 6.7 (5.0e8.5) 8.7 (7.9e9.5) 0.050

D Pain sensation sensoric (S 0 (5.3e6.7) 0.005 3.5 (1.8e5.2) 6.0 (5.2e6.8) 0.011

D Depressive symptoms (AD 9 (5.2e6.6) 0.623 5.2 (3.6e6.8) 6.0 (5.3e6.7) 0.368

D SF-36 Physical Compone 1 (6.4e7.7) 0.520 6.6 (5.2e8.0) 7.2 (6.6e7.8) 0.472

D SF-36 Mental Component 7 (4.1e5.3) 0.508 3.7 (2.3e5.2) 4.8 (4.2e5.5) 0.179

D Headache days per 4 wee 9 (2.7e3.2) 0.012 1.9 (1.2e2.6) 3.1 (2.8e3.4) 0.003

n 5 number of valid obse CI 5 95% confidence interval; D 5 change from baseline (positive values indicate

improvement).a Age; duration; baseline tion term for age and SF-36 Physical Component Scale; plus baseline value of the

respective outcome.b On a scale from 1 5 vc Outcome measured wi nal study from a single question (headache days in the last 4 weeks).

28

5K

.L

ind

eet

al.

/Jo

urn

al

of

Clin

ical

Ep

idem

iolo

gy

60

(20

07

)2

80

e2

87

veness/change from baseline to month 6 in patients receiving acupuncture in the randomized

st) and adjusted analyses (analysis of covariance)

Unadjusted

RCT OS Adjusted for baseline valu

n Mean (95% CI) n Mean (95% CI) P-value RCT, Mean (95% CI) OS

128 2.6 (2.4e2.8) 715 2.3 (2.2e2.4) 0.011 n.a.

128 1.9 (1.5e2.2) 697 2.5 (2.3e2.7) 0.022 2.0 (1.5e2.4) 2.

129 2.0 (1.6e2.4) 698 3.0 (2.8e3.2) 0.001 2.0 (1.5e2.5) 3.

129 15.5 (12.9e18.1) 675 18.5 (16.9e20.0) 0.125 17.2 (14.3e20.1) 18.

ES) 126 6.4 (5.0e7.9) 647 8.7 (7.9e9.5) 0.022 6.5 (4.8e8.2) 8.

ES) 124 2.8 (1.2e4.5) 635 6.1 (5.3e6.9) !0.001 3.4 (1.7e5.0) 6.

S) 114 4.2 (2.5e5.9) 580 6.1 (5.3e6.9) 0.049 5.5 (3.9e7.0) 5.

nt Scale 121 5.2 (3.9e6.5) 642 7.3 (6.6e8.0) 0.018 6.6 (5.1e8.0) 7.

Scale 121 2.4 (1.0e3.9) 642 5.0 (4.2e5.8) 0.010 4.1 (2.7e5.6) 4.

ksc 131 3.1 (2.4e3.8) 651 2.7 (2.4e3.1) 0.422 2.0 (1.3e2.7) 2.

rvations; NRS 5 numerical rating scale from 0 5 no pain to 10 5 worst imaginable pain; 95%

values of SF-36 Physical Component Scale, SF-36 Mental Component Scale, and PDI; interac

ery good effect to 5 5 insufficient effect.

th different methods: in the randomized trial derived from a headache diary, in the observatio

286 K. Linde et al. / Journal of Clinical Epidemiology 60 (2007) 280e287

protocol or failed to organize data collection for follow-upvisits. However, although our data suggest that most of thedata loss is due to organizational failure, they also suggestthat in fact some selection bias did occur [7]. (d) In the OSstudy, physicians prescribed or recommended other thera-pies in addition to acupuncture in about a quarter of the pa-tients, while in the RCT use of cointerventions was restrictedto continuation of treatment(s) already used for acute attacksin the past. Therefore, improvements observed in the OSmight be due partly to effective cointerventions.

4.4. Interpretation of findings

Compared to the OS data, the RCT data have been pro-duced under much more controlled and standardized condi-tions, within a trial performed according to commonstandards (ICH-GCP) and recommendations for migrainestudies [4]. The proportion of patients with at least 50%attack reduction (when interpreting all patients droppingout or with missing data as nonresponders) was 54% after12 weeks and 44% at 24 weeks [6]. Compared to the no-treatment waiting list group, patients receiving acupuncturein the RCT did significantly better in all relevant outcomemeasures. However, minimal acupuncture (superficial nee-dling at nonacupuncture points) was as effective as acu-puncture with response rates of 53% after 12 weeks and48% after 24 weeks. These rates are in the same range asthose observed in migraine drug trials and higher than thosecommonly observed in placebo groups of drug trials[17e19].

While there is evidence from RCTs that drugs like pro-pranolol, metoprolol, or flunarizine are more effective thanmedical placebo, studies investigating how well migraineprophylaxis with drugs works in everyday practice are morethan scarce [18,20]. Dropout rates in RCTs of drugs suggestthat compliance and patient acceptance are suboptimal andRCTs with long-term follow-up are almost nonexistent. Forexample, although propranolol has been investigated inmore than 50 trials, there are very few trials with more than1-month follow-up after completion of treatment, whichhave acceptable follow-up rates [18]. There is an urgentneed for studies investigating the effectiveness of drugsfor migraine prophylaxis under routine conditions invol-ving patients with comorbidities, clear preferences, etc.It might be that under conditions of routine care and ina population of patients with a positive attitude toward acu-puncture and a critical attitude toward interval treatmentwith drugs like beta-blockers or calcium antagonists, acu-puncture is the preferable option in spite of its limited oreven absent point-specific effects. Several randomized trialscomparing acupuncture and drug treatment (under experi-mental conditions) found similar improvements and fewerside effects with acupuncture [21e23].

It is easier to interpret our comparison of characteristicsof patients participating in the OS and those participating inthe RCT than to compare the outcomes. Some differences

in patient characteristics are remarkable. Although RCT pa-tients had longer disease duration and more headache days,they had a better quality of life and fewer depressive symp-toms than patients participating in the OS. This suggeststhat RCT patients were better compensated.

4.5. Conclusions

In summary, participants in our RCT were probably notrepresentative of migraine patients who undergo acupuncturetreatment in Germany under conditions of routine practiceand current regulations. Our finding that improvements afteracupuncture treatment, observed under conditions of routinepractice, are slightly better than those observed under exper-imental conditions, must be interpreted with great cautiondue to methodological limitations. Despite the statisticallysignificant differences in outcomes, the results of the RCTand the OS seem consistent on a clinical level. The resultsof our RCTand other RCTs suggest that nonspecific needlingand acupuncture treatment as a whole might be more effec-tive than a medical placebo for preventing migraine attacksbut that the location of needling is of minor or no relevance.In our opinion both OSs and RCTs are needed to guide deci-sions for practice [24], particularly if experimental condi-tions and conditions of routine care use of an interventiondiffer in a relevant mannerdwhich is quite often the case.

Acknowledgments

Funding was provided by the following statutory sick-ness funds: Deutsche Angestellten-Krankenkasse, Ham-burg; Barmer Ersatzkasse, Wuppertal; Kaufm€annischeKrankenkasse, Hannover; Hamburg-M€unchener Kranken-kasse, Hamburg; Hanseatische Krankenkasse, Hamburg;Gm€under Ersatzkasse, Schw€abisch Gm€und; HZK Kranken-kasse f€ur Bau- und Holzberufe, Hamburg; Br€uhler Ersatz-kasse, Solingen; Krankenkasse Eintracht Heusenstamm,Heusenstamm; and Buchdrucker Krankenkasse, Hannover.

References

[1] Melchart D, Linde K, Fischer P, White A, Allais G, Vickers A, et al.

Acupuncture for recurrent headaches: a systematic review of random-

ized controlled trials. Cephalalgia 1999;19:779e86.

[2] Melchart D, Streng A, Hoppe A, J€urgens S, Weidenhammer W,

Linde K. Akupunktur bei chronischen Schmerzen. Ergebnisse aus

dem Modellvorhaben der Ersatzkassen. Dtsch €Arztebl 2006;103:

A187e95.

[3] Diener HC, Kr€amer J, Wippermann F, Molsberger A, Sch€afer H,

Victor N, et al. Die gerac Akupunktur StudiendGerman acupuncture

trials. Z €Arztl Fortbild Qual Gesundhwes 2004;98:468e71.

[4] Witt C, Brinkhaus B, Jena S, Selim D, Lobel S, Willich SN. Modell-

vorhaben Akupunktur der Techniker Krankenkasse und der dem

Modellvorhaben beigetretenen Krankenkassen. Z €Arztl Fortbild Qual

Gesundhwes 2004;98:473e5.

[5] International Headache Society Clinical Trials Subcommittee.

Guideline for controlled trials of drugs in migraine: second edition.

Cephalalgia 2000;20:765e86.

287K. Linde et al. / Journal of Clinical Epidemiology 60 (2007) 280e287

[6] Linde K, Streng A, J€urgens S, Hoppe A, Brinkhaus B, Witt C, et al.

Acupuncture for patients with migraine. A randomized controlled

trial. JAMA 2005;293:2118e25.

[7] Melchart D, Weidenhammer W, Streng A, Hoppe A, Pfaffenrath V,

Linde K. Acupuncture for chronic headachesdan epidemiological

study. Headache 2006;46:632e41.

[8] International Headache Society. ICH-10 guide for headaches. Ceph-

alalgia 1997;17(Suppl 19):1e82.

[9] Diener HC, Brune K, Gerber WD, Gobel H, Pfaffenrath V. Behand-

lung der Migr€aneattacke und Migr€aneprophylaxe. Dtsch €Arztebl

1997;94:C-2277e83.

[10] Nagel B, Gerbershagen HU, Lindena G, Pfingsten M. Entwicklung

und empirische Uberpr€ufung des Deutschen Schmerzfragebogens

der DGSS. Schmerz 2002;16:263e70.

[11] Dillmann U, Nilges P, Saile E, Gerbershagen HU. Behinderung-

seinsch€atzung bei chronischen Schmerzpatienten. Schmerz 1994;8:

100e10.

[12] Geissner E, Schulte A. Die Schmerzempfindungsskala (SES). Wein-

heim: Beltz; 1991.

[13] Hautzinger M, Bailer M. Allgemeine Depressionsskala (ADS). Die

deutsche Version des CES-D. Weinheim: Beltz; 1991.

[14] Bullinger M, Kirchberger I. SF-36 Fragebogen zum Gesundheitszu-

stand. Gottingen: Hogrefe; 1998.

[15] Marstedt G, Moebus S. Inanspruchnahme alternativer Methoden in

der Medizin. In: Robert-Koch-Institut, editor, Gesundheitsberichter-

stattung des Bundes, Heft 9. Berlin: Verlag Robert-Koch-Institut;

2002. p. 7.

[16] Weidenhammer W, Streng A, J€urgens S, Hoppe A, Linde K,

Melchart D. Aufbau, Realisierung und Daten€ubersicht einer

Beobachtungsstudie im Rahmen des Programms zur Evaluation der

Patientenversorgung mit Akupunktur (PEP-AK) des Modellvorha-

bens der Ersatzkassen. Gesundheitswesen 2005;67:264e73.

[17] Van der Kuy PHM, Lohmann JJHM. A quantification of the placebo

response in migraine prophylaxis. Cephalalgia 2002;22:209e21.

[18] Linde K, Rossnagel K. Propranolol for migraine prophylaxis.

[Cochrane Review]. The Cochrane Library 2004.

[19] Kaptchuk TJ, Goldman P, Stone DA, Stason WB. Do medical de-

vices have enhanced placebo effects? J Clin Epidemiol 2000;53:

786e92.

[20] Gray RN, Goslin RE, McCrory DC, Eberlein K, Tulsky J, Hasselblad

V. Drug treatments for theprevention of migraine headache. Technical

review 2.3. Durham, NC, Duke University, Center for Clinical Health

Policy and Research, for the Agency for Health Care Policy and Re-

search, Rockville, MD: 1999.

[21] Allais G, De Lorenzo C, Quirico P, Airola G, Tolardo G, Mana O,

et al. Acupuncture in the prophylactic treatment of migraine without

aura: a comparison with flunarizine. Headache 2002;42:855e61.

[22] Hesse J, Mogelvang B, Simonsen H. Acupuncture versus metoprolol

in migraine prophylaxis: a randomized trial of trigger point inactiva-

tion. J Intern Med 1994;235:451e6.

[23] Diener HC, Kronfeld K, Boewing G, Lungenhausen M, Maier C,

Molsberger A, et al. Efficacy of acupuncture for the prophylaxis of

migraine: a multicentre randomised controlled clinical trial. Lancet

Neurol 2006;5:310e6.

[24] Linde K, Scholz M, Melchart D, Willich SN. Should systematic

reviews include non-randomized and uncontrolled studies? The case

of acupuncture for chronic headache. J Clin Epidemiol 2002;55:

77e85.