- Email: [email protected]
Pain patients in a randomized trial did not show a significant effect of a positive consultation
Journal of Clinical Epidemiology 58 (2005) 708–713
Pain patients in a randomized trial did not show a significant effect of a positive consultation Paul Knipschilda,*, Arnoud Arntzb b
a Department of General Practice, Maastricht University, Postbox 616, 6200 MD Maastricht, The Netherlands Department of Medical, Clinical and Experimental Psychology, Maastricht University, Postbox 616, 6200 MD Maastricht, The Netherlands
Accepted 3 January 2005
Abstract Objective: To examine whether telling symptomatic patients that they will soon be better shortens the duration of their illness. Study Design and Setting: We repeated an earlier trial in general practice on this subject. All our patients were reassured that they probably had no serious underlying disease. We randomized 128 patients into two groups. Half of them got a clear diagnosis and were told that they would soon be better. The others heard from their general practitioners (GPs) that they probably had no serious underlying disease but that the GP did not know exactly what was wrong; these patients were advised to come back later if necessary. For 78 patients, study protocol was followed exactly. Result: Our trial could not show an effect. The Kaplan–Meier curves of patients with a positive and negative consultation ran nearly parallel. The hazard ratio in the total group was 0.97 (95% precision interval PI ⫽ 0.69–1.41) and in the optimal subgroup 0.91 (95% PI ⫽ 0.57–1.46). We still cannot rule out a modest placebo effect. Conclusion: Our finding contrasts with that of the earlier trial. We think that symptomatic patients mainly want to know that they do not have a serious problem. Often, they will miss the extra information that they get—and then they certainly cannot benefit from it. 쑖 2005 Elsevier Inc. All rights reserved. Keywords: Placebo effect; Physician–patient relationship; General practice; Pain; Recovery; Co-interventions
1. Introduction
1.2. The doctor–patient relationship
1.1. The placebo effect
This brings us to another, more psychological definition of a placebo effect: it is the effect of co-interventions in a treatment study connected to the doctor–patient relationship. We found a recent review on this subject [6], and an update of the same review [7]. The authors concluded that there was evidence for a moderate placebo effect, but they added that much of the literature is methodologically flawed. A recent study by de Craen and others [8] on the effect of promotion of a pain medication in chronic pain patients could not show any positive placebo effect. For the moment, it seems doubtful whether certain aspects of the doctor– patient relationship are helpful in getting patients better.
The placebo effect is a recurrent subject of discussion in medical science and health care. Quite often, however, it is mistakenly thought to be the effect of a treatment in patients with a certain disease who get a placebo, such as an inactive pill [1]. This is not strictly the placebo effect, but rather a combination of effects caused by random error and bias, as well as what happens after the spontaneous course of the disease, and all sorts of co-interventions, including psychological effects [2]. A definition of the placebo effect being the difference in effect between the placebo group and the spontaneous course in a randomized clinical trial comes closer to what we think it is [3]. In 2001, two Danish researchers wrote a comprehensive review on this subject; to their own surprise, they could not find evidence for the existence of such an effect, except perhaps for pain [4,5].
* Corresponding author. Tel.: ⫹31-43-3882188; fax: ⫹31-43-3619344. E-mail address: [email protected] (P. Knipschild). 0895-4356/05/$ – see front matter 쑖 2005 Elsevier Inc. All rights reserved. doi: 10.1016/j.jclinepi.2005.01.005
1.3. Placebo effects in a general practice Placebo effects, if defined in this psychological manner, can occur in specialist, second line medicine, but they are probably easier to show in first-line medicine, be it alternative or conventional. We know of two other examples in the literature where a paternalistic attitude of the doctor seemed helpful [9,10]. The outstanding example, however, commonly cited (and involving no placebo), comes from general
P. Knipschild, A. Arntz / Journal of Clinical Epidemiology 58 (2005) 708–713
practice. The now retired general practitioner (GP) Bruce Thomas used to see all sorts of symptomatic patients for consultation at the academic general practice surgery in Southampton. He wondered if he could behave like a colleague of his, who claimed (and still claims) to be a walking placebo [11]. Thomas did a trial with 200 patients who came to the surgery with all kinds of symptomatic problems. For all the patients, he took a history and did a physical examination. Next, he excluded the patients who probably had a serious underlying disease. Thereafter, he randomly told half of the remaining patients that they did not have a serious disease, and he actually gave them a firm diagnosis and reassured them that they would soon be better. The other half were similarly told by Thomas that they did not have a serious disease, but he gave no firm diagnosis, saying instead that he did not know what exactly was the matter. He advised the patients in this latter group to come back to the surgery if their symptoms were not over in a week or so. It turned out that the first group of patients, the ones who got a positive consultation, improved much more quickly. Most of these patients (64%) were better in 2 weeks; in the second group, only 39% were better in 2 weeks [12,13]. Thomas’s trial was published in 1987. We could not find a successful replication and therefore we started one ourselves, reported here.
2. Methods and patients 2.1. Changes: number of GPs Compared to the earlier trial, we made a few important changes. First, the English study had been done with one GP, Thomas himself, who was a notably charismatic personality. In our trial, we involved many more doctors, in total 17. Among them were eight GPs who provided five or more patients. All GPs had their practices in Maastricht and surrounding areas in the Netherlands and Belgium. 2.2. Changes: restriction to adults with pain We restricted the patients group to adults (⭓18 years) who came to the surgery with certain pain complaints. All children, for whom we would need a special informed consent, were excluded from our trial, in contrast to Thomas’s trial. Apart from the consent problem, we expected a placebo effect to be rather small in small children. We excluded patients who were seen at home or were talked to on the phone. We restricted our population further to the following symptomatic diagnoses: headache, sore throat, abdominal pain, and pain related to movement; Thomas had allowed several other symptomatic diagnoses. Like Thomas, we excluded all patients whom we suspected of having an underlying disease (e.g., sinusitis, appendicitis, rheumatoid arthritis, ankylosing spondylitis).
709
2.3. Changes: new patients or episodes In contrast to Thomas’s trial, we included only patients who had their present pain for not more than 1 week before they went to see their GP. More importantly, we aimed at including only new patients with certain, localized pain or patients with a clearly new episode of the same pain. Patients who had had the same problem within the last 3 months were excluded. In this respect, Thomas was much looser in his admission criteria; he included everybody. We think that a placebo effect can be more easily detected among new patients than among patients who have suffered from their complaints for a long time. 2.4. Randomization At the surgery, the pain-complaint patients were orally informed of the research project. After history taking and a thorough physical examination, the project was explained to them by their own GP. Thereafter, he rang the first author or his assistant for the randomization of the patient. They always had a randomization list of advice 1 (specific consultation) or advice 2 (negative consultation) with them and could be phoned on their portable phone. Next, the patients were told that they did not have a serious underlying disease and then the GP gave them advice 1 or 2. After a GP had expressed his willingness to include patients for the trial, the study protocol was followed at the GP’s surgery. The doctors were also asked to tape-record their advices. Advice 1: You probably do not have a serious underlying disease. I will tell you precisely what the matter is with you (followed by a clear explanation). You will be better within a week or so. Advice 2: You probably do not have a serious underlying disease. But I do not know what precisely is the matter with you. If you are not better in a week or so, come back to the surgery. The randomization scheme was stratified according to movement pain or other pain, so that both conditions, with a different pain duration, were represented in both advice 1 and advice 2 groups. The patients answered a few demographic questions and gave written informed consent. They rated on a 0–100% Visual Analog Scale how contented they had been with the consultation, from not contented at all to very contented. The GP stated afterwards how convincingly he had given the advice to his individual patients. 2.5. Follow-up After the consultation, all patients were phoned weekly (through a maximum of 100 days) by the first author or his assistant, to see if their pain had gone. The first author and his assistant were both blinded for the intervention. We told all patients in our written patient information and later on by phone that our main interest was in the variation of the length of the pain. Not earlier than at the end of 2003
710
P. Knipschild, A. Arntz / Journal of Clinical Epidemiology 58 (2005) 708–713
all patients learned that we had a hidden, second goal: to see if advice 1, compared to 2, gave a shorter pain duration. If patients wanted a prescription for their problem, all GPs advised them to take acetaminophen (paracetamol) or an NSAID. We did not record the actual consumption of pain medication. 2.6. Check of conversation The tape-recorded doctor–patient conversations were rated afterwards by the authors, still blinded for the advice given. They had to judge which advice had been given to the patients, and also had to rate the degree of conviction. This was done on a 0–100% Visual Analog Scale, from not convincing at all to completely convincing. The advice was considered convincing enough if both authors had scored the advice ⬎50%.
consultation day. One patient (with a back problem) was included and rightly so, but he was lost to follow-up after 4 weeks when he moved out of our region. This left us with 128 patients who fulfilled the admission criteria and could be followed up. For 102 patients, this was the first episode; 17 had had the same problem more than a year ago, and 9 (also) in the last year, but not in the last 3 months. After more than 100 days of follow-up, all but 8 of the 128 patients were better (the exceptions being 5 patients with advice 1 and 3 patients with advice 2). For 16 patients, the advice had not been tape-recorded: 8 with advice 1 and 8 with advice 2. In another 34 cases, the GP had not given the patient the correct advice in a convincing way (advice 1 unconvincing 9 times; advice 2, no less than 25 times), as evaluated by the authors. This left us with 78 patients in whom everything had gone according to the protocol, including the tape-recording.
2.7. Ethical approval The trial as outlined above was approved by the Medical Ethical Committee of Maastricht University and the Academic Medical Hospital in Maastricht. 2.8. Needed number of patients In the earlier trial in England, 200 patients were included. Because we considered our trial to be more efficient, we reasoned that we might do with fewer participants; however, we fell victim to Lasagna’s law (“you always end with far too few patients in a trial”). At α ⫽ .05 two-sided, β ⫽ 0.20, and proportions of 64% and 39% (taken from the trial by Thomas), we would need at least 2 times 63 patients for statistical purposes. 2.9. Statistics The main analyses to be performed were Kaplan–Meier curves of the time until the complete disappearance of pain, for all patients and separately for the patients in whom everything had gone according to the protocol, including the tape-recordings. These two main analyses include P-values according to log-rank testing. We also used Cox regression analysis to get hazard ratios with their 95% precision intervals (95% PI).
3. Results 3.1. Actual number of patients It took us much more time than we expected to get enough patients. Actually, after more than 2 years, we stopped recruitment in the early summer of 2003, when we had 157 patients randomized. Of these, however, we had to exclude 28 patients at the first telephone interview because we found them to be chronic rather than new pain patients, or because their pain had been present for more than a week at the
3.2. Data on illness duration Table 1 gives the division of the patients according to the reason for seeing their GPs, separately for patients with advice 1 and 2. It turns out that there were 64 patients each for advice 1 and advice 2, respectively, and that these two groups had approximately the same distribution over the various pain sites. The pain duration was nearly twice as high for movement pain than for other pain (median 18 and 10 days, respectively). The proportion of movement pain compared to other pain was close to 60%–40%, alike for patients with advice 1 and advice 2. Figure 1A shows the Kaplan–Meier curves for these 128 patients. We started the duration of the pain at the first complaint day, but beginning at the consultation day (median 3 days later) did not change the picture. There was hardly any difference between advice 1 and 2. After 14 days, the number of patients who reported being better was 55% among patients with advice 1 and 59% with advice 2. After 28 days, this was 81% and 78%, respectively. The accompanying log-rank test gives P ⫽ .75. The Cox regression hazard ratio (advice 1 to advice 2) was 0.97, with a 95% PI from 0.69 to 1.41.
3.3. Data of the 78 optimal patients In 78 patients, the procedure was followed entirely as intended under the study protocol. We found the GPs to have fewer shortcomings when we told them on the phone to give advice 1. The proportion of patients in this subgroup was 47 of 64 (73%) for advice 1 and 31 of 64 (48%) for advice 2. Quite often, the GPs seemed to feel a little uncomfortable about giving advice 2. They appeared to be hesitant to tell their patients that they did not know what exactly was the matter and that the patients had to come back if not better in a week or so. This was confirmed by what we heard at the tape-recordings of their advices to their patients.
P. Knipschild, A. Arntz / Journal of Clinical Epidemiology 58 (2005) 708–713
711
Table 1 Two advices given to the full group (128 patients) and to the subgroup of 78 in whom everything had gone according to our study protocol, by site, with mean duration of paina Patients given Advice 1, no. (%)
Patients given Advice 2, no. (%)
Median duration, daysb
Total
Subgroup
Total
Subgroup
Total
Subgroup
Spine Arm Leg Breast Movement
18 7 10 2 37
(28) (11) (16) (3) (58)
12 6 8 1 27
(26) (13) (17) (2) (58)
15 9 8 7 39
(23) (14) (13) (11) (61)
7 6 4 3 20
(23) (19) (13) (10) (65)
18 18 24.5 12 18
17 13 15 12 17
Throat Head Abdomen Other pain
18 7 2 27
(28) (11) (3) (42)
15 4 1 20
(32) (9) (2) (42)
18 6 1 25
(28) (9) (2) (39)
8 2 1 11
(26) (6) (3) (35)
11 8 10 10
11 8 8 10
All patients
64
13
13
47
64
31
a
Advice 1: “You probably do not have a serious underlying disease. I will tell you precisely what the matter is with you [followed by a clear explanation]. You will be better within a week or so.” Advice 2: “You probably do not have a serious underlying disease. But I do not know what precisely is the matter with you. If you are not better in a week or so, come back to the surgery.” b Starting with the first day of pain.
Figure 1B shows the Kaplan–Meier curves for our subgroup of 78 patients in whom everything had gone according to protocol, again separately for patients with advice 1 and 2. The two curves now differ a little: patients with advice 1 compared to advice 2 seemed to have a slightly better prognosis. After 14 days, the proportion of patients who were already better was 57% for advice 1 and 58% for advice 2; after 28 days, it was 80% for advice 1 and 73% for advice 2. The log-rank test, over the whole range of the pain duration, was not statistically significant (P ⫽ .48). The Cox regression hazard ratio (advice 1/advice 2) was now 0.91, with a 95% PI from 0.57 to 1.46.
3.4. Other data We asked the GPs for every participating patient how firmly they had given their advices to their patients. It turned out that only in 4 of the 128 cases was this percentage considered ⬍50%; in 6 cases it was between 50% and 60%. The median for advice 1 was 81%; for advice 2 it was 79%. The patients’ answers to the question of how contented they had been with their GP-consultations were also very affirmative. Only 1 patient of the 128 had a score ⬍50%, 10 times the patients had scored 50% (indifferent), and almost all others were very contented, with nearly everybody
Fig. 1. The cumulative proportion of patients in time who got better. (A) Total group (N ⫽ 128). (B) Subgroup (n ⫽ 78). Data are given separately for patients with advice 1 (⫹) and advice 2 (⫺).
712
P. Knipschild, A. Arntz / Journal of Clinical Epidemiology 58 (2005) 708–713
scoring ⬎75%. We did not find a difference between advice 1 and 2 (median 90% and 88%, respectively).
Doctors tend to give their opinion on placebos and placebo effects without knowing much of the scientific literature on the subject—and in fact the number of published trials on placebo effects is limited and often these trials seem to be flawed. Therefore, we do not know too well how important certain context variables, related to the doctor–patient relationship, are in helping patients to cope with their health problems. A presumed important doctor-related effect is the expressed belief by the doctor that a firm diagnosis of a symptomatic problem will make the patient better sooner. Of the latter, the best example that we know of is Thomas’s trial, published in 1987 [12]. It concludes that a positive consultation appears to shorten the problem duration substantially.
same persons asked the patients weekly whether their pain had gone away. Often, however, the person who told the GP differed from the one who asked the patient. Even when the same person was involved, that person was not necessarily aware of the earlier advice to the GP when he or she phoned the patient. The time period between the two made it highly unlikely that the earlier advice to the GP was remembered at the later time. In contrast with Thomas’s trial, ours did not show a shorter duration of problems when the GP gave advice 1 compared to advice 2. The Kaplan–Meier curves for advice 1 and 2 run nearly parallel, both in the larger group of 2 times 64 patients and also in the more optimal subgroup of 47 and 31 patients. Nevertheless, the 95% PI of the hazard ratio in both comparisons is rather broad, and so with our trial we cannot rule out a modest, but important effect of giving advice 1 compared to 2 on the duration of the problem. We suggest that others do a third trial elsewhere, if possible a larger trial than ours, to decide more definitely what the effects are of a positive compared to a neutral or a negative consultation.
4.2. How negative was the negative consultation?
4.5. Charisma and 17 doctors
Thomas himself asked the first author of this article to replicate his study. We did a new trial, but concentrated on incident cases of certain pain complaints among adults, thereby probably improving the quality of the trial. It took us a long time to get GPs in our area interested in our trial, however. Most doctors in Maastricht and its surrounding area are used to giving their symptomatic patients positive consultations. Many of them were not willing to tell some of their patients for research purposes that they did not themselves know precisely what the matter was and that the patients must come back if the problems did not go away. Some of those who wanted to help us with the trial told us afterwards that they had great difficulty in giving patients advice 2. Quite often, when we asked the GPs to give a negative consultation, they used instead a rather neutral formulation. Even at our own Department of General Practice at Maastricht University there was some opposition to the trial.
Two other aspects might help to explain our data. Thomas was a rather charismatic GP who may have had a placebo effect influence on his patients. It did not take him too long to find 200 patients for his trial. On the other hand, in our trial no less than 17 GPs had to help us. It is possible that they made their own selection of patients whom they considered fit to participate in our trial. One never knows how representative these patients are of all patients.
4. Discussion 4.1. Thomas’s and other trials
4.3. Too few patients At the start of our trial, we aimed at 200 patients (100 with advice 1 and 100 with advice 2), just as in Thomas’s trial. In the end, we were glad that we had 128, twice 64, which is approximately the minimal number of needed patients at a power of 80%. If we exclude the patients with no tape-recording or with a neutral or incorrect advice, we end up with only 78 participants.
4.6. Not remembering what has been said Even more importantly, sometimes doctors and patients appeared to communicate with each other on different wave lengths. When we asked some patients later about what their GP had said to them, often they said that they had found the consultation quite reassuring but they did not tell us about the type of advice given. To summarize the literature on this lack of communication in one sentence, it seems that 40%–80% of medical information provided by health care professionals is forgotten immediately by the patient and that half of the information that is remembered, is incorrectly remembered [14]. Thus, it may be disappointing but true that many patients in our trial did not understand or remember correctly the advice that their GPs had given them—and when you cannot remember what the doctor has said to you, you cannot benefit from it.
4.4. Blinding
Acknowledgments
In theory, there may have been a problem of blinding. The first author or his assistant had to tell the GP which advice, 1 or 2, he was to give a certain patient. Later, the
We are grateful to the patients in the project and the many general practitioners who helped us. Among them were eight who provided us with five or more patients: Canio Gilio,
P. Knipschild, A. Arntz / Journal of Clinical Epidemiology 58 (2005) 708–713
Sjoerd Hobma, Goele Janssen, Paul Olbrechts, Herman van Rens, Bert Schilte, Rudi Thijs, and Frans Vissers. Our research assistants, Marion Gijbels, Hilde Sielhorst, Bernadette Zinsen, were important for data collection and analysis. Harry Crebolder and Gerben ter Riet gave useful advice, especially in the first year of the project. After we had written the first draft of this article, it benefited from comments by Iain Chalmers, Jos Kleijnen, and Hubert Schouten.
References [1] Beecher HK. The powerful placebo. J Am Med Assoc 1955;159: 1602–6. [2] Kienle GS, Kiene H. The powerful placebo effect: fact or fiction? J Clin Epidemiol 1997;50:1311–8. [3] Hro´bjartsson A. The uncontrollable placebo effect. Eur J Clin Pharmacol 1996;50:345–8. [4] Hro´bjartsson A, Gøtzsche PC. Is the placebo powerless? An analysis of clinical trials comparing placebo to no treatment. N Engl J Med 2001;344:1594–602 [Erratum in: N Engl J Med 2001;345:304]. [5] Hro´bjartsson A. What are the main methodological problems in the estimation of placebo effects? J Clin Epidemiol 2002;55:430–5.
713
[6] Di Blasi Z, Harkness E, Ernst E, Georgiou A, Kleijnen J. Influence of context effects on health outcomes: a systematic review. Lancet 2001;357:757–62. [7] Di Blasi Z, Kleijnen J. Context effects: powerful therapies or methodological bias? Eval Health Prof 2003;26:166–79. [8] Craen AJM de, Lampe-Schoenmaeckers AJEM, Kraal JW, Tijssen JGP, Kleijnen J. Impact of experimentally-induced expectancy on the analgesic efficacy of tramadolin chronic pain patients: a 2 × 2 factorial, randomized, placebo-controlled, double-blind trial. J Pain Symptom Manage 2001;21:210–7. [9] LeBaron S, Reyher J, Stack JM. Paternalistic vs egalitarian physician styles: the treatment of patients in crisis. J Family Pract 1985;21: 56–62. [10] Savage R, Armstrong D. Effect of a general practitioner’s consulting style on patients’ satisfaction: a controlled study. BMJ 1990;301:968– 70. [Erratum in: BMJ 1990;301:1316]. [11] Lewith GT. Every doctor a walking placebo. In: Lewith GT, Aldridge D, editors. Clinical research methodology for complementary therapies. Chapter 4. London: Hodder and Stoughton; 1993:38–45. [12] Thomas KB. General practice consultations: is there any point in being positive? Br Med J (Clin Res Ed) 1987;294:1200–2. [13] Thomas KB. The placebo in general practice. Lancet 1994;344: 1066–7. [14] Kessels RPC. Patients’ memory for medical information. J R Soc Med 2003;96:219–22.
Pain patients in a randomized trial did not show a significant effect of a positive consultation Paul Knipschilda,*, Arnoud Arntzb b
a Department of General Practice, Maastricht University, Postbox 616, 6200 MD Maastricht, The Netherlands Department of Medical, Clinical and Experimental Psychology, Maastricht University, Postbox 616, 6200 MD Maastricht, The Netherlands
Accepted 3 January 2005
Abstract Objective: To examine whether telling symptomatic patients that they will soon be better shortens the duration of their illness. Study Design and Setting: We repeated an earlier trial in general practice on this subject. All our patients were reassured that they probably had no serious underlying disease. We randomized 128 patients into two groups. Half of them got a clear diagnosis and were told that they would soon be better. The others heard from their general practitioners (GPs) that they probably had no serious underlying disease but that the GP did not know exactly what was wrong; these patients were advised to come back later if necessary. For 78 patients, study protocol was followed exactly. Result: Our trial could not show an effect. The Kaplan–Meier curves of patients with a positive and negative consultation ran nearly parallel. The hazard ratio in the total group was 0.97 (95% precision interval PI ⫽ 0.69–1.41) and in the optimal subgroup 0.91 (95% PI ⫽ 0.57–1.46). We still cannot rule out a modest placebo effect. Conclusion: Our finding contrasts with that of the earlier trial. We think that symptomatic patients mainly want to know that they do not have a serious problem. Often, they will miss the extra information that they get—and then they certainly cannot benefit from it. 쑖 2005 Elsevier Inc. All rights reserved. Keywords: Placebo effect; Physician–patient relationship; General practice; Pain; Recovery; Co-interventions
1. Introduction
1.2. The doctor–patient relationship
1.1. The placebo effect
This brings us to another, more psychological definition of a placebo effect: it is the effect of co-interventions in a treatment study connected to the doctor–patient relationship. We found a recent review on this subject [6], and an update of the same review [7]. The authors concluded that there was evidence for a moderate placebo effect, but they added that much of the literature is methodologically flawed. A recent study by de Craen and others [8] on the effect of promotion of a pain medication in chronic pain patients could not show any positive placebo effect. For the moment, it seems doubtful whether certain aspects of the doctor– patient relationship are helpful in getting patients better.
The placebo effect is a recurrent subject of discussion in medical science and health care. Quite often, however, it is mistakenly thought to be the effect of a treatment in patients with a certain disease who get a placebo, such as an inactive pill [1]. This is not strictly the placebo effect, but rather a combination of effects caused by random error and bias, as well as what happens after the spontaneous course of the disease, and all sorts of co-interventions, including psychological effects [2]. A definition of the placebo effect being the difference in effect between the placebo group and the spontaneous course in a randomized clinical trial comes closer to what we think it is [3]. In 2001, two Danish researchers wrote a comprehensive review on this subject; to their own surprise, they could not find evidence for the existence of such an effect, except perhaps for pain [4,5].
* Corresponding author. Tel.: ⫹31-43-3882188; fax: ⫹31-43-3619344. E-mail address: [email protected] (P. Knipschild). 0895-4356/05/$ – see front matter 쑖 2005 Elsevier Inc. All rights reserved. doi: 10.1016/j.jclinepi.2005.01.005
1.3. Placebo effects in a general practice Placebo effects, if defined in this psychological manner, can occur in specialist, second line medicine, but they are probably easier to show in first-line medicine, be it alternative or conventional. We know of two other examples in the literature where a paternalistic attitude of the doctor seemed helpful [9,10]. The outstanding example, however, commonly cited (and involving no placebo), comes from general
P. Knipschild, A. Arntz / Journal of Clinical Epidemiology 58 (2005) 708–713
practice. The now retired general practitioner (GP) Bruce Thomas used to see all sorts of symptomatic patients for consultation at the academic general practice surgery in Southampton. He wondered if he could behave like a colleague of his, who claimed (and still claims) to be a walking placebo [11]. Thomas did a trial with 200 patients who came to the surgery with all kinds of symptomatic problems. For all the patients, he took a history and did a physical examination. Next, he excluded the patients who probably had a serious underlying disease. Thereafter, he randomly told half of the remaining patients that they did not have a serious disease, and he actually gave them a firm diagnosis and reassured them that they would soon be better. The other half were similarly told by Thomas that they did not have a serious disease, but he gave no firm diagnosis, saying instead that he did not know what exactly was the matter. He advised the patients in this latter group to come back to the surgery if their symptoms were not over in a week or so. It turned out that the first group of patients, the ones who got a positive consultation, improved much more quickly. Most of these patients (64%) were better in 2 weeks; in the second group, only 39% were better in 2 weeks [12,13]. Thomas’s trial was published in 1987. We could not find a successful replication and therefore we started one ourselves, reported here.
2. Methods and patients 2.1. Changes: number of GPs Compared to the earlier trial, we made a few important changes. First, the English study had been done with one GP, Thomas himself, who was a notably charismatic personality. In our trial, we involved many more doctors, in total 17. Among them were eight GPs who provided five or more patients. All GPs had their practices in Maastricht and surrounding areas in the Netherlands and Belgium. 2.2. Changes: restriction to adults with pain We restricted the patients group to adults (⭓18 years) who came to the surgery with certain pain complaints. All children, for whom we would need a special informed consent, were excluded from our trial, in contrast to Thomas’s trial. Apart from the consent problem, we expected a placebo effect to be rather small in small children. We excluded patients who were seen at home or were talked to on the phone. We restricted our population further to the following symptomatic diagnoses: headache, sore throat, abdominal pain, and pain related to movement; Thomas had allowed several other symptomatic diagnoses. Like Thomas, we excluded all patients whom we suspected of having an underlying disease (e.g., sinusitis, appendicitis, rheumatoid arthritis, ankylosing spondylitis).
709
2.3. Changes: new patients or episodes In contrast to Thomas’s trial, we included only patients who had their present pain for not more than 1 week before they went to see their GP. More importantly, we aimed at including only new patients with certain, localized pain or patients with a clearly new episode of the same pain. Patients who had had the same problem within the last 3 months were excluded. In this respect, Thomas was much looser in his admission criteria; he included everybody. We think that a placebo effect can be more easily detected among new patients than among patients who have suffered from their complaints for a long time. 2.4. Randomization At the surgery, the pain-complaint patients were orally informed of the research project. After history taking and a thorough physical examination, the project was explained to them by their own GP. Thereafter, he rang the first author or his assistant for the randomization of the patient. They always had a randomization list of advice 1 (specific consultation) or advice 2 (negative consultation) with them and could be phoned on their portable phone. Next, the patients were told that they did not have a serious underlying disease and then the GP gave them advice 1 or 2. After a GP had expressed his willingness to include patients for the trial, the study protocol was followed at the GP’s surgery. The doctors were also asked to tape-record their advices. Advice 1: You probably do not have a serious underlying disease. I will tell you precisely what the matter is with you (followed by a clear explanation). You will be better within a week or so. Advice 2: You probably do not have a serious underlying disease. But I do not know what precisely is the matter with you. If you are not better in a week or so, come back to the surgery. The randomization scheme was stratified according to movement pain or other pain, so that both conditions, with a different pain duration, were represented in both advice 1 and advice 2 groups. The patients answered a few demographic questions and gave written informed consent. They rated on a 0–100% Visual Analog Scale how contented they had been with the consultation, from not contented at all to very contented. The GP stated afterwards how convincingly he had given the advice to his individual patients. 2.5. Follow-up After the consultation, all patients were phoned weekly (through a maximum of 100 days) by the first author or his assistant, to see if their pain had gone. The first author and his assistant were both blinded for the intervention. We told all patients in our written patient information and later on by phone that our main interest was in the variation of the length of the pain. Not earlier than at the end of 2003
710
P. Knipschild, A. Arntz / Journal of Clinical Epidemiology 58 (2005) 708–713
all patients learned that we had a hidden, second goal: to see if advice 1, compared to 2, gave a shorter pain duration. If patients wanted a prescription for their problem, all GPs advised them to take acetaminophen (paracetamol) or an NSAID. We did not record the actual consumption of pain medication. 2.6. Check of conversation The tape-recorded doctor–patient conversations were rated afterwards by the authors, still blinded for the advice given. They had to judge which advice had been given to the patients, and also had to rate the degree of conviction. This was done on a 0–100% Visual Analog Scale, from not convincing at all to completely convincing. The advice was considered convincing enough if both authors had scored the advice ⬎50%.
consultation day. One patient (with a back problem) was included and rightly so, but he was lost to follow-up after 4 weeks when he moved out of our region. This left us with 128 patients who fulfilled the admission criteria and could be followed up. For 102 patients, this was the first episode; 17 had had the same problem more than a year ago, and 9 (also) in the last year, but not in the last 3 months. After more than 100 days of follow-up, all but 8 of the 128 patients were better (the exceptions being 5 patients with advice 1 and 3 patients with advice 2). For 16 patients, the advice had not been tape-recorded: 8 with advice 1 and 8 with advice 2. In another 34 cases, the GP had not given the patient the correct advice in a convincing way (advice 1 unconvincing 9 times; advice 2, no less than 25 times), as evaluated by the authors. This left us with 78 patients in whom everything had gone according to the protocol, including the tape-recording.
2.7. Ethical approval The trial as outlined above was approved by the Medical Ethical Committee of Maastricht University and the Academic Medical Hospital in Maastricht. 2.8. Needed number of patients In the earlier trial in England, 200 patients were included. Because we considered our trial to be more efficient, we reasoned that we might do with fewer participants; however, we fell victim to Lasagna’s law (“you always end with far too few patients in a trial”). At α ⫽ .05 two-sided, β ⫽ 0.20, and proportions of 64% and 39% (taken from the trial by Thomas), we would need at least 2 times 63 patients for statistical purposes. 2.9. Statistics The main analyses to be performed were Kaplan–Meier curves of the time until the complete disappearance of pain, for all patients and separately for the patients in whom everything had gone according to the protocol, including the tape-recordings. These two main analyses include P-values according to log-rank testing. We also used Cox regression analysis to get hazard ratios with their 95% precision intervals (95% PI).
3. Results 3.1. Actual number of patients It took us much more time than we expected to get enough patients. Actually, after more than 2 years, we stopped recruitment in the early summer of 2003, when we had 157 patients randomized. Of these, however, we had to exclude 28 patients at the first telephone interview because we found them to be chronic rather than new pain patients, or because their pain had been present for more than a week at the
3.2. Data on illness duration Table 1 gives the division of the patients according to the reason for seeing their GPs, separately for patients with advice 1 and 2. It turns out that there were 64 patients each for advice 1 and advice 2, respectively, and that these two groups had approximately the same distribution over the various pain sites. The pain duration was nearly twice as high for movement pain than for other pain (median 18 and 10 days, respectively). The proportion of movement pain compared to other pain was close to 60%–40%, alike for patients with advice 1 and advice 2. Figure 1A shows the Kaplan–Meier curves for these 128 patients. We started the duration of the pain at the first complaint day, but beginning at the consultation day (median 3 days later) did not change the picture. There was hardly any difference between advice 1 and 2. After 14 days, the number of patients who reported being better was 55% among patients with advice 1 and 59% with advice 2. After 28 days, this was 81% and 78%, respectively. The accompanying log-rank test gives P ⫽ .75. The Cox regression hazard ratio (advice 1 to advice 2) was 0.97, with a 95% PI from 0.69 to 1.41.
3.3. Data of the 78 optimal patients In 78 patients, the procedure was followed entirely as intended under the study protocol. We found the GPs to have fewer shortcomings when we told them on the phone to give advice 1. The proportion of patients in this subgroup was 47 of 64 (73%) for advice 1 and 31 of 64 (48%) for advice 2. Quite often, the GPs seemed to feel a little uncomfortable about giving advice 2. They appeared to be hesitant to tell their patients that they did not know what exactly was the matter and that the patients had to come back if not better in a week or so. This was confirmed by what we heard at the tape-recordings of their advices to their patients.
P. Knipschild, A. Arntz / Journal of Clinical Epidemiology 58 (2005) 708–713
711
Table 1 Two advices given to the full group (128 patients) and to the subgroup of 78 in whom everything had gone according to our study protocol, by site, with mean duration of paina Patients given Advice 1, no. (%)
Patients given Advice 2, no. (%)
Median duration, daysb
Total
Subgroup
Total
Subgroup
Total
Subgroup
Spine Arm Leg Breast Movement
18 7 10 2 37
(28) (11) (16) (3) (58)
12 6 8 1 27
(26) (13) (17) (2) (58)
15 9 8 7 39
(23) (14) (13) (11) (61)
7 6 4 3 20
(23) (19) (13) (10) (65)
18 18 24.5 12 18
17 13 15 12 17
Throat Head Abdomen Other pain
18 7 2 27
(28) (11) (3) (42)
15 4 1 20
(32) (9) (2) (42)
18 6 1 25
(28) (9) (2) (39)
8 2 1 11
(26) (6) (3) (35)
11 8 10 10
11 8 8 10
All patients
64
13
13
47
64
31
a
Advice 1: “You probably do not have a serious underlying disease. I will tell you precisely what the matter is with you [followed by a clear explanation]. You will be better within a week or so.” Advice 2: “You probably do not have a serious underlying disease. But I do not know what precisely is the matter with you. If you are not better in a week or so, come back to the surgery.” b Starting with the first day of pain.
Figure 1B shows the Kaplan–Meier curves for our subgroup of 78 patients in whom everything had gone according to protocol, again separately for patients with advice 1 and 2. The two curves now differ a little: patients with advice 1 compared to advice 2 seemed to have a slightly better prognosis. After 14 days, the proportion of patients who were already better was 57% for advice 1 and 58% for advice 2; after 28 days, it was 80% for advice 1 and 73% for advice 2. The log-rank test, over the whole range of the pain duration, was not statistically significant (P ⫽ .48). The Cox regression hazard ratio (advice 1/advice 2) was now 0.91, with a 95% PI from 0.57 to 1.46.
3.4. Other data We asked the GPs for every participating patient how firmly they had given their advices to their patients. It turned out that only in 4 of the 128 cases was this percentage considered ⬍50%; in 6 cases it was between 50% and 60%. The median for advice 1 was 81%; for advice 2 it was 79%. The patients’ answers to the question of how contented they had been with their GP-consultations were also very affirmative. Only 1 patient of the 128 had a score ⬍50%, 10 times the patients had scored 50% (indifferent), and almost all others were very contented, with nearly everybody
Fig. 1. The cumulative proportion of patients in time who got better. (A) Total group (N ⫽ 128). (B) Subgroup (n ⫽ 78). Data are given separately for patients with advice 1 (⫹) and advice 2 (⫺).
712
P. Knipschild, A. Arntz / Journal of Clinical Epidemiology 58 (2005) 708–713
scoring ⬎75%. We did not find a difference between advice 1 and 2 (median 90% and 88%, respectively).
Doctors tend to give their opinion on placebos and placebo effects without knowing much of the scientific literature on the subject—and in fact the number of published trials on placebo effects is limited and often these trials seem to be flawed. Therefore, we do not know too well how important certain context variables, related to the doctor–patient relationship, are in helping patients to cope with their health problems. A presumed important doctor-related effect is the expressed belief by the doctor that a firm diagnosis of a symptomatic problem will make the patient better sooner. Of the latter, the best example that we know of is Thomas’s trial, published in 1987 [12]. It concludes that a positive consultation appears to shorten the problem duration substantially.
same persons asked the patients weekly whether their pain had gone away. Often, however, the person who told the GP differed from the one who asked the patient. Even when the same person was involved, that person was not necessarily aware of the earlier advice to the GP when he or she phoned the patient. The time period between the two made it highly unlikely that the earlier advice to the GP was remembered at the later time. In contrast with Thomas’s trial, ours did not show a shorter duration of problems when the GP gave advice 1 compared to advice 2. The Kaplan–Meier curves for advice 1 and 2 run nearly parallel, both in the larger group of 2 times 64 patients and also in the more optimal subgroup of 47 and 31 patients. Nevertheless, the 95% PI of the hazard ratio in both comparisons is rather broad, and so with our trial we cannot rule out a modest, but important effect of giving advice 1 compared to 2 on the duration of the problem. We suggest that others do a third trial elsewhere, if possible a larger trial than ours, to decide more definitely what the effects are of a positive compared to a neutral or a negative consultation.
4.2. How negative was the negative consultation?
4.5. Charisma and 17 doctors
Thomas himself asked the first author of this article to replicate his study. We did a new trial, but concentrated on incident cases of certain pain complaints among adults, thereby probably improving the quality of the trial. It took us a long time to get GPs in our area interested in our trial, however. Most doctors in Maastricht and its surrounding area are used to giving their symptomatic patients positive consultations. Many of them were not willing to tell some of their patients for research purposes that they did not themselves know precisely what the matter was and that the patients must come back if the problems did not go away. Some of those who wanted to help us with the trial told us afterwards that they had great difficulty in giving patients advice 2. Quite often, when we asked the GPs to give a negative consultation, they used instead a rather neutral formulation. Even at our own Department of General Practice at Maastricht University there was some opposition to the trial.
Two other aspects might help to explain our data. Thomas was a rather charismatic GP who may have had a placebo effect influence on his patients. It did not take him too long to find 200 patients for his trial. On the other hand, in our trial no less than 17 GPs had to help us. It is possible that they made their own selection of patients whom they considered fit to participate in our trial. One never knows how representative these patients are of all patients.
4. Discussion 4.1. Thomas’s and other trials
4.3. Too few patients At the start of our trial, we aimed at 200 patients (100 with advice 1 and 100 with advice 2), just as in Thomas’s trial. In the end, we were glad that we had 128, twice 64, which is approximately the minimal number of needed patients at a power of 80%. If we exclude the patients with no tape-recording or with a neutral or incorrect advice, we end up with only 78 participants.
4.6. Not remembering what has been said Even more importantly, sometimes doctors and patients appeared to communicate with each other on different wave lengths. When we asked some patients later about what their GP had said to them, often they said that they had found the consultation quite reassuring but they did not tell us about the type of advice given. To summarize the literature on this lack of communication in one sentence, it seems that 40%–80% of medical information provided by health care professionals is forgotten immediately by the patient and that half of the information that is remembered, is incorrectly remembered [14]. Thus, it may be disappointing but true that many patients in our trial did not understand or remember correctly the advice that their GPs had given them—and when you cannot remember what the doctor has said to you, you cannot benefit from it.
4.4. Blinding
Acknowledgments
In theory, there may have been a problem of blinding. The first author or his assistant had to tell the GP which advice, 1 or 2, he was to give a certain patient. Later, the
We are grateful to the patients in the project and the many general practitioners who helped us. Among them were eight who provided us with five or more patients: Canio Gilio,
P. Knipschild, A. Arntz / Journal of Clinical Epidemiology 58 (2005) 708–713
Sjoerd Hobma, Goele Janssen, Paul Olbrechts, Herman van Rens, Bert Schilte, Rudi Thijs, and Frans Vissers. Our research assistants, Marion Gijbels, Hilde Sielhorst, Bernadette Zinsen, were important for data collection and analysis. Harry Crebolder and Gerben ter Riet gave useful advice, especially in the first year of the project. After we had written the first draft of this article, it benefited from comments by Iain Chalmers, Jos Kleijnen, and Hubert Schouten.
References [1] Beecher HK. The powerful placebo. J Am Med Assoc 1955;159: 1602–6. [2] Kienle GS, Kiene H. The powerful placebo effect: fact or fiction? J Clin Epidemiol 1997;50:1311–8. [3] Hro´bjartsson A. The uncontrollable placebo effect. Eur J Clin Pharmacol 1996;50:345–8. [4] Hro´bjartsson A, Gøtzsche PC. Is the placebo powerless? An analysis of clinical trials comparing placebo to no treatment. N Engl J Med 2001;344:1594–602 [Erratum in: N Engl J Med 2001;345:304]. [5] Hro´bjartsson A. What are the main methodological problems in the estimation of placebo effects? J Clin Epidemiol 2002;55:430–5.
713
[6] Di Blasi Z, Harkness E, Ernst E, Georgiou A, Kleijnen J. Influence of context effects on health outcomes: a systematic review. Lancet 2001;357:757–62. [7] Di Blasi Z, Kleijnen J. Context effects: powerful therapies or methodological bias? Eval Health Prof 2003;26:166–79. [8] Craen AJM de, Lampe-Schoenmaeckers AJEM, Kraal JW, Tijssen JGP, Kleijnen J. Impact of experimentally-induced expectancy on the analgesic efficacy of tramadolin chronic pain patients: a 2 × 2 factorial, randomized, placebo-controlled, double-blind trial. J Pain Symptom Manage 2001;21:210–7. [9] LeBaron S, Reyher J, Stack JM. Paternalistic vs egalitarian physician styles: the treatment of patients in crisis. J Family Pract 1985;21: 56–62. [10] Savage R, Armstrong D. Effect of a general practitioner’s consulting style on patients’ satisfaction: a controlled study. BMJ 1990;301:968– 70. [Erratum in: BMJ 1990;301:1316]. [11] Lewith GT. Every doctor a walking placebo. In: Lewith GT, Aldridge D, editors. Clinical research methodology for complementary therapies. Chapter 4. London: Hodder and Stoughton; 1993:38–45. [12] Thomas KB. General practice consultations: is there any point in being positive? Br Med J (Clin Res Ed) 1987;294:1200–2. [13] Thomas KB. The placebo in general practice. Lancet 1994;344: 1066–7. [14] Kessels RPC. Patients’ memory for medical information. J R Soc Med 2003;96:219–22.