To help ensure comparability of the intervention and usual care nursing homes for rates of antimicrobial prescriptions at baseline, we paired nursing homes within each province or state by size (number of occupied beds) and by proportion of residents with indwelling catheters. One member of each pair was randomised to the intervention and the other to usual care. A statistician independent of the study team used a random numbers table to assign the intervention to nursing homes (odd or even) corresponding to the number selected. We measured outcomes over 12 months. The nursing homes served as the unit of allocation, intervention, and analysis.
A research coordinator contacted nursing homes in southern Ontario and Idaho about the study. Only eligible for our study were free standing, community based nursing homes with 100 or more residents and no stated policy for diagnosis or treatment of urinary tract infections. To reduce the potential for selection bias, all residents in study nursing homes were eligible for participation. We contacted a total of 56 nursing homes, 36 in the Hamilton region, Ontario and 20 in the Boise region, Idaho. Eligible nursing homes had to agree to refrain from introducing new management strategies for antimicrobial use or clinical pathways for urinary tract infection during the 12 months of the study. To enhance representativeness of nursing homes in the community, we excluded nursing homes directly associated with tertiary care centres. Of the 56 homes approached, 24 were randomised, of which 16 were from Ontario and eight from Idaho. Nursing homes were enrolled from September 2001 to February 2002, with the last follow-up in March 2003. Participating and non-participating homes were of similar bed size (mean (SD) 183 (64.7) beds v 168 (73.4) beds, P = 0.40).
Intervention nursing homes
We introduced the nurses and physicians of the intervention nursing homes to the diagnostic and therapeutic algorithms. Before data collection, one of two study investigators presented six case scenarios lasting a total of 30 minutes to small groups of between 10 and 15 registered nurses or registered nursing assistants. Participation was active, and nurses were asked to decide whether to order antibiotics and urine cultures and to justify their answers using the algorithms. We videotaped a reconstruction of the small group sessions and distributed the video to the nursing homes for viewing by existing and new staff over the course of the study. We sent the algorithms, along with written explanatory material, to all the physicians who cared for the nursing home residents. One of three investigators met once individually with the physicians who cared for 80% or more of residents in each nursing home. The algorithms were explained to them using the six case scenarios, printed on pocket cards and distributed to the physicians and nursing staff at the start of the study, and mounted as large posters at all nursing stations. The physicians and nurses were asked to use the algorithms when assessing residents for fever or suspected urinary tract infection. We asked the nurses to complete a one page log of presenting symptoms and signs for every resident in whom urinary tract infection was suspected, as a reminder to use the algorithms. One member of staff in each nursing home was assigned the role of reminding nurses to use the algorithms. The intervention homes were allowed a four week training period before data collection. We visited the nursing homes every three months to address any questions that the staff had and to carry out audits of the records to check that antimicrobial prescriptions for suspected urinary tract infection had not been missed.
Usual care nursing homes
Nurses and physicians in the usual care nursing homes were notified about the study and were informed about how data were going to be collected. No other interventions were applied to these homes.
The main outcome was the number of prescriptions for antimicrobials. We considered all antimicrobials given for one particular indication to be one course and antimicrobials prescribed for a second indication during the same period or prescribed after one week for the same indication to be separate courses. Other outcomes included number of urine cultures ordered, admissions to hospital, and deaths. Each facility's infection control practitioner used standardised data collection forms to collect data on antimicrobials prescribed and urine cultures sent.
Although allocation was concealed, given the nature of the intervention the nursing home staff could not be blinded to the intervention. Pharmacies affiliated with the study (the source of confirmation of antimicrobial prescriptions) were, however, blinded. To verify accuracy of data recorded at the nursing home, we carried out onsite audits of the charts records of the nursing home residents and obtained records from the pharmacies of antimicrobials prescribed.
The unit of analysis was the nursing home. We used paired t tests, weighted by size of nursing home (number of beds) to analyse the following within pair differences in matched pairs of nursing homes: rates of antimicrobials prescribed for suspected urinary tract infections, proportions of antimicrobials prescribed for urinary tract infections, total rates of antimicrobials prescribed, urine cultures obtained, admissions to hospital, and mortality. To assess the effect of the intervention over time, we used linear regression to model the difference in antimicrobial rates for suspected urinary tract infection between the study homes by study month. Analyses were carried out using SAS version 8.2.
Sample size calculation
We determined that we would need 142 prescriptions for antimicrobials for suspected urinary tract infection (71 in each arm) to have 80% power to detect a 20% reduction in prescriptions at an α of 0.05, assuming a 30% baseline rate of prescriptions. To adjust for the effect of within cluster dependency, we calculated the intracluster correlation coefficient (variance for urinary antimicrobial prescriptions between homes divided by the sum of variance between and within the homes) and found this to be 0.04 using data from an Ontario long term care facility study.4
The variance inflation factor was 11,20
such that we required 1562 prescriptions for suspected urinary tract infection. Since these represent about 30% of all antimicrobial prescriptions,12
we increased the sample size to 5206 prescriptions to assess whether a reduction in prescriptions for antimicrobials for suspected urinary tract infection could also reduce overall use of antimicrobials. On the basis of prescribing rates from a large cohort study,6
we estimated that we would need to follow 20 (10 pairs) nursing homes for 12 months. Since we did not account for matching in the sample size calculation, which would improve efficiency, these figures were conservative. We recruited another four homes to maintain the target sample size in case of withdrawals from the study.