Trial design

The trial was a randomised, double blind, placebo controlled study of two parallel groups (fig 1). Crossover was precluded because of possible carry over effects from homoeopathy. We recruited participants over six weeks from the middle of February so that the prospectively defined stopping time was before the start of the local pollen season.

Fig 1

Recruitment of patients and their progress through the trial

• Download figure
• Open in new tab
• Download powerpoint

At the start of the qualification period the doctor assessed each patient's history, allergy status, and nasal obstruction. The principal allergen determining the prescription was then chosen on the basis of the largest skin test weal concordant with the allergy history. In seven cases in which the prescriber had difficulty in determining the principal allergen, telephone advice was given by a doctor experienced in prescribing homoeopathic immunotherapy. The first of three phials of placebo corresponding to the principal allergen was then administered by the doctor on to the patient's tongue. Patients were unaware that the phials contained placebo, although the researchers were not blinded.

The following two weeks served jointly as the qualification and baseline period. At the second visit, qualifying patients were randomised by a restricted technique of permuted blocks of two,7 generated from random number tables and stratified for the indicated allergen. A double blinded prescription was immediately dispensed. The trial ended with a follow up visit four weeks later.

Allergy and respiratory tests

Operators were trained to test sensitivity to house dust mite, cat fur, dog hair, tree pollens, grass pollens, and cladosporium by skin prick testing with preloaded lancets (Phazets, Pharmacia, Milton Keynes) and to aspergillus, feathers, and house dust using a needle and allergen solution (Bencard, Welwyn Garden City). Negative and positive histamine controls were used. A weal reaction of 3 mm or more in its greatest diameter after 15 minutes was taken as positive. To confirm the diagnosis, allergen specific serum IgE was measured by radioimmunoassay according to the manufacturer's instructions (Pharmacia). Samples were tested in batches to avoid interassay variability.

Patients were taught to use a Youlten nasal inspiratory peak flow meter (Clement Clark, Harlow),8 which is an objective, sensitive, valid, and reliable indicator of nasal obstruction.9-12 This measure decreases after nasal challenge in allergic rhinitis. 11 13 In clinical trials increases of 13 l/min after low dose desensitisation14 and 18.5 l/min after topical steroids15 have been correlated with clinical improvements.

Study diaries

At the same time each morning and evening patients recorded three successive nasal inspiratory peak flow measurements. Before making these measurements patients oriented themselves by noting each morning (on a 0 to 4 integer scale) how their symptoms had interfered with their sleep and, each night, rating blocked, runny, or itchy nose symptoms, sneezing, and any eye and chest symptoms. Patients then recorded their daily overall visual analogue scale score. To allow comparison with our previous trials1-3 the identical wording was used: “Overall today I felt …” on a scale of 0–100 mm, where 0 is fine and 100 is terrible. Visual analogue scale scores are a recommended measure of the severity of rhinitis.16 Adverse events, including initial aggravations of symptoms as observed in our previous rhinitis trial,2 were documented by the patients, clarified by the doctor, and recorded on standard adverse experience report forms. Any use of conventional drugs was also noted.

Medication preparation and administration

Using original standard allergen material from the Pasteur Institute in Paris, a homoeopathic laboratory (Boiron, Lyons, France) prepared the drugs according to the French homoeopathic pharmacopoeia through 30 stages of 1 in 99 serial agitated dilutions to produce a 30c dilution, as reported previously.3 Each treatment consisted of three identical phials containing 1 g of lactose-sucrose globules that had been impregnated with either a 30c homoeopathic dilution of the principal allergen or placebo. The three phials constituted a split single dose that was to be taken equally spaced over 24 hours to cover any diurnal variation in the patient's sensitivity to treatment and to ensure compliance. Only one dose was taken. The placebo dilution consisted of the same batch of diluent identically diluted and vibrated but without the starting allergen. The treatments were indistinguishable in packaging, taste, and smell. Random samples of drug phials were checked by independent laboratories for the presence of extraneous house dust mite allergen (der p1) by enzyme linked immunosorbent assay (ELISA)17 (University of Virginia, Charlottesville) and for antiallergy drugs by gas chromatography-mass spectrometry (MD800, Fisons, Manchester). No such contamination was found.

The coded drug packages were sent to the pharmacy department of Glasgow Royal Infirmary where, to augment blinding, each one was recoded with a unique number according to the randomisation schedule and then delivered to the pharmacy department of Northwick Park Hospital for distribution to each centre. The codes were held by both the French laboratory and a hussier de justice (notary) and remained unbroken until the analyses were completed.

Analysis

The prestudy power calculation was based on the results of the hay fever trial,2 from which a mean difference of 15 mm between the groups on visual analogue scale scores and a corresponding standard deviation of 29 were obtained. With a choice of 5% significance and 80% power, we estimated that 60 patients would be required in each group to avoid false negative results. No interim analysis was carried out. The predefined main measures of outcome were the changes from baseline in mean visual analogue scale scores and nasal inspiratory peak flow (use of mean values supported by Wihl and Malm18) over the third and fourth weeks after randomisation, when initial aggravations would be likely to be over and any treatment effects evident. Predefined secondary measures of outcome were differences between the groups in reports of adverse events, including initial aggravations of symptoms and use of drugs for rhinitis. Intention to treat analysis was used. Variables with normal distributions (nasal inspiratory peak flow and visual analogue scale) were analysed by using two tailed, two sample t tests and confidence intervals. χ² tests were applied to categorical comparisons and proportions, but if any cell in a contingency table was less than 5, Fisher's exact test was used. The independent statistician verified the coded data entry, confirmed all analyses, and then carried out a repeated measures analysis of covariance on the changes from baseline over the four weeks after randomisation for each variable. The possible terms for inclusion in the model were treatment, time into study, and treatment-time interaction; the baseline score was included as a potential covariate.

Overview

To summarise our results, the original data19 from all four trials were pooled and analysed by an independent worker using Cochrane Collaboration meta-analysis software (Revman 3.0). All available visual analogue scale scores from every randomised patient in the four trials were used on an intention to treat basis, with each patient acting as his or her own control. The four trials had been designed as a series to address the hypothesis that homoeopathy is a placebo response. Each trial studied atopic inhalant allergies and assessed subjective effects in the same way over the third and fourth weeks after randomisation. They had also all used homoeopathic immunotherapy at 30c potency. The studies should therefore have a good degree of clinical homogeneity. Statistical heterogeneity was assumed to be present when the P value for heterogeneity was less than 0.10, and therefore the random effects model was used.20 The weighted mean difference was used as an estimate of the treatment effect (average change on homoeopathy minus average change on placebo).21 To aid presentation the overall daily graph was plotted with smoothed values by using simple robust non-linear procedures.22 As the main objective measure varied across the studies we compared them in a simple overview.

Participants

Fifty one patients successfully completed qualification screening and were randomised (fig 1). Because of the exacting screening, strict qualification criteria, and the prospectively defined requirement to stop enrolment before the pollen season, we did not recruit the number of patients that the power calculation had estimated we required.

At baseline, clinical characteristics (table 1) and main measures of outcome (table 2) were similar in both groups. One patient (homoeopathy group) was subsequently lost to follow up despite repeated postal reminders.

Fig 2

Mean (SE) weekly change from baseline in nasal inspiratory peak flow

• Download figure
• Open in new tab
• Download powerpoint

Nasal inspiratory peak flow

We found a clear objective difference between the effects of placebo and homoeopathy on nasal airflow (table 2, fig 2). Patients in the homoeopathy group had an overall improvement from baseline averaging 21% compared with 2% in the placebo group over the third and fourth weeks after randomisation. This difference was significant (P=0.0001) and was confirmed by repeated measures analysis of covariance. The improvement was consistent across all recruitment centres.

Visual analogue scale

Subjectively, both treatments resulted in improvement, with no significant difference between them (table 2). This change began during the single blind placebo run-in period, decreasing the baseline value by the time of randomisation. Subsequently both groups showed further improvement. The individual symptom scores echoed these trends. Centre by centre analysis showed that in all but one of the centres the homoeopathy group improved more than the placebo group. This anomalous result seemed to be related to the increased frequency of aggravations and later entry in that centre.

Fig 3

Overview and pooled analysis of four trials of homoeopathic immunotherapy

• Download figure
• Open in new tab
• Download powerpoint

Other measures

Non-respiratory adverse events were minor, and no difference was found between the groups. Initial aggravations of rhinitis symptoms were provoked more by homoeopathy than by placebo. By 48 hours after randomisation seven (29%) patients in the homoeopathy group reported a worsening of rhinitis, two with wheeze, compared with two (7%) patients in the placebo group, neither of whom had wheezing (P=0.04, Fisher's exact test). By 14 days, 11 (46%) patients in the homoeopathy group had reported adverse events, 10 of whom had rhinitis related aggravations, compared with seven (26%) in the placebo group, five of whom had rhinitis related aggravations (χ²=3.28, P=0.07). In general, most aggravations were short lived, averaging four days, and all had resolved by day 16. Aggravations of rhinitis in patients who received homoeopathy seemed to point to a good outcome. Initial deterioration was followed by subjective improvement and a corresponding improvement in nasal inspiratory peak flow. Only one patient in each group resorted to conventional rhinitis drugs, and both took them for less than four days.

Overview

Figure 3 compares the underlying patterns from the four trials based on daily visual analogue scale data from all 253 randomised patients and the main objective measures. The subjective changes measured by the visual analogue scale show a therapeutic response from homoeopathy across the four trials. There was a lesser and more variable response in the placebo groups, which probably accounts for the evidence of some statistical heterogeneity (χ²=9.04, df=3, P=0.03). The pooled estimate of the treatment effect (middle column, fig 3) showed an average improvement in visual analogue scores in the homoeopathy group compared with the placebo group (improvement 11.1 mm, 95% confidence interval 3.3 to 18.8; P<0.01). Thus, overall, the trend of the individual trial results and pooled data point towards homoeopathy differing from placebo. The trends in the objective measures all follow the same direction as the subjective measures (fig 3), again indicating a difference between a homoeopathic dilution and placebo.

Figure 4 shows the average response to homoeopathic immunotherapy over time in the four trials. A distinct separation emerges between the action of homoeopathy and placebo. On average, over the last two weeks after randomisation, patients who received homoeopathy had a 28% improvement compared with 3% among those in the placebo group. There was a mean reduction of the visual analogue scale score of 10.9 mm in the homoeopathy group compared with 1.1 mm in the placebo group (95% confidence interval for difference 4.2 to 15.4, P=0.0007; two sided, two sample t test).

Validity of results

Like any other therapy, homoeopathy requires rigorous scientific testing, and one study is insufficient evidence. Some perspective may be gleaned by viewing the results of this trial in the context of the series of which it is part. Except for the subjective measure in this fourth trial, the subjective and objective results show a trend across these four trials clearly pointing to homoeopathy being different from placebo. If the results were due to chance then some trends in favour of placebo would be expected. So does homoeopathy work or are our results due to some other factor?

Recent attempts to resolve the controversy surrounding homoeopathy have centred on the 180 or so controlled trials to date. A criteria based review in 1991 found that the evidence was positive but not conclusive.29 In a 1997 update, other workers concluded that 73% of the existing trial data supported homoeopathy being more effective than placebo, with the pooled odds ratio from a criteria based meta-analysis of 89 trials suggesting homoeopathy showed around twice the overall mean effect of placebo. The difference was significant and proved robust in sensitivity analyses that included correction for publication bias.30 A third working group, independently set up by the European Commission, selected 17 comparisons in 2001 patients for a meta-analysis. The pooled P value was highly significant, and the group commented that “it is likely that among the tested homeopathic approaches some had an added effect over nothing or placebo.”31 Are these findings “meta-errors” or, however implausible, does something tangible lie at the core of homoeopathy?

To interpret these findings as arguing for homoeopathy having an effect may now be more plausible than our previous hypothesis of serial false positive results. 3 32 For now, we conclude that this study has failed to confirm our original hypothesis that homoeopathy is a placebo.