We invited to participate in our study all 42 general practices in the London borough of Tower Hamlets and additionally two practices in the neighbouring borough of Newham, which served large Bangladeshi populations using the Royal London Hospital, where the specialist nurses were based. All 44 practices consented to take part. We randomised practices to intervention and control groups using a minimisation programme,16
stratifying by partnership size, training practice status, hospital admission rate for asthma, employment of practice nurse, and whether the practice nurse was trained in asthma care ().
Stratified randomisation and characteristics of 44 participating general practices allocated to nurse led intervention for acute asthma or standard guidelines for asthma
Patients were eligible for inclusion if they had asthma diagnosed by a doctor, were aged 4-60 years, and had been admitted to or attended the accident and emergency department at the Royal London Hospital or the general practitioner out of hours service with acute asthma. We identified eligible participants prospectively by monitoring attendance at these services. We also wrote to patients who had been admitted to or had attended the accident and emergency department with asthma in the previous two years. We invited eligible patients to attend a specialist nurse run asthma clinic at the Royal London Hospital, where we sought consent to participate according to the randomisation status of their practice and access to medical records.
Practices and participants
The two specialist nurses were accredited by the National Respiratory Training Centre in Stratford, east London. They intervened at the levels of the general practice and the patient (see bmj.com
General practices randomised to the intervention group received two one hour visits by the specialist nurses at the start of the study to discuss guidelines for managing patients with acute asthma. We used a behaviour change model, incorporating discussion of relevant research evidence.17
We provided a computer template or stamp to prompt patient review, including identification as a patient with high risk asthma, inhaler technique and peak expiratory flow assessment, and self management advice.
Participants registered with the intervention practices were reviewed for asthma control and drugs by the specialist nurses at the nurse run clinic immediately after recruitment. They discussed a self management plan. Patients with sufficient understanding were provided with a peak flow meter, a supply of rescue oral corticosteroids for future use, and a written plan produced by the National Asthma Campaign with standard thresholds for peak flow and symptoms. Instructions were to double their dose of inhaled corticosteroid if their peak expiratory flow rate was between 70% and 80% of best or they were waking at night with symptoms, to contact their doctor to arrange a course of oral corticosteroids if their peak expiratory flow rate was between 50% and 70% or breathlessness was increasing, and to contact their doctor urgently if their peak expiratory flow rate was below 50% of best or if symptoms continued to worsen. Most South Asian participants were Bangladeshis speaking Sylheti—a dialect with no written form; they received a plan written in English, explained through a bilingual advocate. Participants with insufficient comprehension of guided self management were educated about their drugs and encouraged to contact their general practice should their asthma control worsen. Nurses reinforced advice with a face to face or telephone consultation.
The control group received a single visit from the nurses to discuss standard guidelines for asthma. Participants registered with control practices were checked for inhaler technique in the nurse run clinic immediately after recruitment. Drugs were unaltered. Participants otherwise continued with usual care.
Prespecified primary outcomes were the percentage of participants attending for unscheduled asthma care and the time to first attendance for unscheduled asthma care in the year after intervention. Unscheduled care was defined as a consultation in primary, secondary, or out of hours care, where a participant presented with symptoms or signs related to acute asthma, including wheeze, cough, or breathlessness. We included acute presentations for cough requiring for example an increase in asthma drugs or antibiotics, and we logged these separately.
Secondary outcomes were rates of attendance for unscheduled care and review, self management behaviour, and quality of life, assessed by generic (EQ-5D) and respiratory specific (AQ20 and north of England) scales.6,18-20
Consultations for non-acute asthma were classified as reviews or other consultations.
Researchers blinded to the randomisation status of the general practice photocopied written and computerised general practice records. We obtained medical records for the 13 participants who had moved out of the study area. With blinding retained, we checked the hospital administration records for admissions and accident and emergency attendances that may have been missing from the general practice records. To retain blinding during data extraction, GF removed from the records any letters from the specialist nurse. CG extracted data outside the practice setting. Completeness and accuracy of extraction was validated by another blinded researcher (GSF), who checked 10 sets of records, chosen using random numbers. Of 165 consultations assessed, there were five discrepancies (3%) involving misclassification of unscheduled care.
Two researchers blinded to randomisation status interviewed participants in person at baseline and by telephone at two, six, nine, and 12 months after recruitment to gather data on personal characteristics, quality of life, and self management of asthma. Participants self identified their ethnicity. We adapted outcome scales into Sylheti; validation of this audiotaped version was by back translation using lay and expert panels and comparison of randomised sequential interviews of bilingual respondents.21
Statistical power and analyses
Audit in a pilot general practice showed that 90% of high risk patients had unscheduled care over one year. Allowing for clustering, our study was powered (80%) to detect a 15% decrease (90% to 75%) in the percentage of patients requiring unscheduled care, at the 5% significance level; we considered a 10-15% change to be clinically important. Based on previous studies, we used 0.05 as an estimate of intracluster correlation. After taking into account drop outs and unequal numbers of patients in practices, we estimated 145 patients were needed in each group. Sample size for time to first attendance could not be estimated due to inadequate data.
Before breaking the coded allocation of practices, we carried out main and prespecified subgroup analyses. Analyses for primary outcomes were by ethnicity (South Asian (Bangladeshi, Indian, Pakistani), white, other), after exclusion of patients with both asthma and chronic obstructive pulmonary disease noted in the medical records, those recruited retrospectively and prospectively, and children and adults. For secondary outcomes we carried out the main and subgroup analyses by ethnicity.
For unscheduled care, review, and quality of life, we fitted generalised estimating equations to individual level data in STATA, fitting different equations depending on the type of outcome; binary, time to event, or number of events. In each case we specified the model to take account of the clustering by practice. For time to unscheduled care and time to review we fitted proportional hazards models assuming an underlying Weibull distribution for the hazard. We used model diagnostics to choose appropriate models. Analyses were by intention to treat. Participants who changed general practice during the study were analysed in the group to which they were allocated at recruitment.
We allowed for clustering by practice except when the intracluster correlation was negative (implying that individuals within clusters were more divergent for the outcome than the general population); in this case we attributed this to sampling error and present results with and without clustering. Covariates were number of exacerbations or reviews at baseline and time since last exacerbation, as appropriate. Time to first contact for unscheduled care and review are represented by Kaplan-Meier plots.
For self management we analysed data at the level of the episode (exacerbation). Self management behaviour was reported for each exacerbation. We initially fitted models accounting for clustering within individuals, which was much stronger than clustering within practice. Because these models showed wide and inconclusive confidence intervals we did not pursue further analyses adjusting for clustering at both individual and practice level.