Does long-term warfarin affect the quality of life of older people?

312 In our study we stated that each pair of wards had identical bed numbers, skill mix, nurse staffing establishments and patients with similar dependency levels. The design and layout of each pair of wards were almost identical. Terry Haines and Keith Hill are of course right to point out that other confounding differences between the wards may have existed. A statistical debate on correct methodology for analysing results appears to be a common feature in most falls trials, if not perhaps all studies. In this study we received three separate expert opinions from statisticians on how best to analyse the results, and therefore we fully appreciate and agree with the comments on the difficulties and inconsistencies encountered with statistical analysis. We do however feel that the method we chose to present the data is valid and perhaps more realistic than using falls per year as in reality patients do not stay in hospital for a year. Randomising individuals rather than wards reduces bias, but does introduce new difficulties, including consent issues given the prevalence of dementia and acute confusion in fallers, and the likelihood of falls occurring soon after admission. It also presents difficulties in maintaining a control group, as staff may not restrict what they view as helpful interventions only to designated patients, or if extra time and attention is given to intervention patients it may reduce staff time and attention to control patients. Balancing the difficulties of both clustered and individualised RCTs in inpatient falls prevention appears to be a key issue for future research, and may require a variety of study designs.