House dust mite control measures in the management of asthma: meta-analysis

British Medical Journal, Oct 24, 1998 by Peter C. Gotzsche, Cecilia Hammarquist, Michael Burr, Abi Berger

Introduction

The major allergen in house dust is derived from mites, and a recent review concluded that the environmental control of allergens should be an integral part of the management of sensitised patients.[1] Some of the evidence in the review, however, was derived from observational studies. Since clinical trials have shown equivocal results of the effectiveness of measures to reduce exposure to mite antigen, we decided to synthesise the findings of all clinical trials.

Methods

Our objective was to determine whether patients with asthma who were sensitised to house dust mites benefited from measures designed to reduce their exposure to mite antigen in the home. All randomised trials in any language performed at any time that compared chemical (acaricidal) or physical measures (such as vacuum cleaning, heating, barrier methods, or air filtration systems) to control mites and analysed their effects on patients with bronchial asthma as compared with an untreated control group were eligible for inclusion in the meta-analysis. Asthma had to have been diagnosed by a doctor and sensitisation to mites had to have been assessed by skin tests, bronchial provocation tests, or serum assays for specific IgE anti-bodies.

Search strategy

We searched the Asthma and Wheez* databases set up by the Cochrane Airways Group which contain records from the Cumulative Index to Nursing and Allied Health Literature, Medline, and Embase. Mite* in the title, abstract, or keyword (descriptor) field was combined with random*, trial*, placebo, double-blind, double blind, single-blind, single blind, comparative study, or controlled study in all fields. Primary authors were contacted to obtain additional information if necessary. CH searched issues of Respiration (1980-96) and MB searched Clinical and Experimental Allergy (1980-96) by hand.

Extraction of data

Two of the authors (CH and MB) selected the trials for inclusion. Two (PCG and CH) extracted data on the following outcomes: subjective wellbeing, improvement in asthma symptoms, use of drugs to control asthma, number of days of sick leave taken from school or work, number of unscheduled visits made to a doctor or hospital, forced expiratory volume in 1 second, peak expiratory flow rate, provocative concentration that causes a 20% fall in forced expiratory volume in 1 second, and results of skin prick testing. Ambiguities were resolved by discussion.

Statistical methods

Review Manager software was used to analyse the data.[2] If P [is less than] 0.10 in the test for heterogeneity a random effects analysis was carried out. Since the results from crossover trials were usually reported as if they had come from a parallel group trial we used tine data accordingly and assumed that no carryover effect had occurred. Continuous data were often presented on different scales in different studies (for example, peak expiratory flow rate was given either as absolute values or as a per cent of predicted values). Because of this, we calculated the standardised mean difference in our analysis of these data. With this method, the difference in effect is divided by the standard deviation of the measurements. Since data on wellbeing and improvements in asthma symptoms were closely related we summarised categorical data as the number of patients whose asthma improved; we summarised continuous data in the category of asthma symptoms. In general, the provocative concentration that causes a 20% fall in the forced expiratory volume in 1 second had been analysed after logarithmic transformation because the data were highly skewed. If the mean values and standard deviations had been converted from the logarithmic to the arithmetic scale we reconverted them.[3] We excluded data on the provocative concentration from one study which had not used logarithmic transformation (see appendix 1 on the website).

In studies in which the use of several anti-asthma treatments had been reported we used the data on bronchodilators. In studies in which data were recorded at several points in time we used the longest observation period during which patients were still on randomised treatment.

We did not adjust for baseline differences since inequalities occurring despite randomisation would be expected to cancel each other out over a number of trials. Furthermore, baseline values were not always available. If we had made adjustments when possible we would have risked biasing the review since investigators may be inclined to show baseline differences and adjust for them when this procedure favours the experimental treatment. Bias occurring during the analysis of data is common and almost always favours the new treatment.[4]

Results

Trials included in the analysis

Altogether, 458 references were identified; half of these were irrelevant and the other half were retrieved so that the full study could be examined. Eighteen of the 229 studies met the inclusion criteria.[5-22] Another four trials were retrieved from MB's personal archive.[23-26] The reference lists of the 229 articles retrieved were also searched but no further appropriate studies were found. One of the papers included in the analysis[11] reported on a trial with three arms; this was treated as two separate trials in the meta-analysis. Thus, most of the analyses below refer to 23 trials. (A list of the excluded trials which were not evidently irrelevant and the reasons for their exclusion appear in appendix 2 on the website.)