5 × 365 days; (3) medicine for temporary use = frequency × 0.5 × recommended duration; (4) medicine for incidental use = 10% from the number of units in case of chronic use and (5) for participants who dropped out before the second home visit, the number of units was estimated based on half the number of days until drop out. In the second, third and fifth assumption, it was unknown how long the participant had been taking a medication on the time point of assessment. Therefore, 0.5 × the expected Temsirolimus manufacturer total duration was believed to be the overall best estimated duration.

Information on recommended duration of medications was obtained from the pharmaceutical guidelines published by the Dutch Health Insurance Board (CVZ) [33]. The prices per medication were obtained from the Royal Dutch Society of Pharmacy [34]. Costs of healthcare devices, aids and adaptations were estimated by asking retail prices from three suppliers in The Netherlands. For each product, the average price was used. All costs

were expressed in 2007 Euros. Statistical methods Baseline characteristics were estimated for the intervention and usual care groups. The economic evaluation was performed according to the intention-to-treat principle. The incremental cost-effectiveness ratios were calculated (differences in costs divided by differences in effects between the intervention and usual care groups). Imputation of missing values PFT�� manufacturer was done using the Multivariate Imputation by Chained Equations algorithm [35]. The imputation model, which was used to estimate the imputed values, included the Talazoparib in vivo variables group many randomisation, age, sex, education level, Mini-Mental State Examination, number of chronic diseases and score on the fall risk profile. According to the variables in the imputation

model, imputed values were based on linear, logistic or polytomous regression estimates. Imputation of cost variables was done before multiplying volumes by cost prices. For medication, the total costs were imputed. Five imputed datasets were created. The quality of the imputations depends on the amount of missing data. When this does not exceed 50%, as in our study (approximately 10%), five imputations are enough to get valid cost estimates [36]. The analyses were done in each dataset and presented are the pooled results of the five imputed datasets as described below. Arithmetic mean (standard deviation, SD) costs were computed for both groups. Means and differences in costs and effects were estimated in each imputed dataset and results were combined by using Rubin’s rules [37]. Mean difference between groups and the associated bias-corrected and accelerated confidence intervals were calculated using bootstrapping techniques.