6 The cost-effectiveness of the Dutch In Balance fall prevention intervention | 107 Sample size Based on the expected effect size of a 50% decrease in the number of falls in the intervention group compared to the control group, a minimum of 106 individuals per group was needed. This calculation assumed a statistical power of 0.80, a type II error rate (β) of 0.20, and a significance threshold (α) of 0.05. Taking into account an expected dropout rate of 20%, we aimed to include 256 participants. Due to the need for a sufficient number of participants to start a new In Balance training group in this study, we included a total of 264 participants. Data imputation To impute missing observations for costs and effects, we used Multiple Imputations by Chained Equations (MICE) with Predictive Mean Matching (PMM) (267, 268). PMM replaces missing values by sampling only from the distribution of the observed data and adds a random element to reflect uncertainty (267, 268). PMM is especially suitable for handling non-normally distributed data such as costs that are generally right skewed (267, 268). The number of imputations was increased until the loss of efficiency was less than 5%, resulting in 20 imputed datasets (267). The cost-effectiveness analyses as described below were performed on each imputed dataset, after which results were pooled using Rubin’s rules (241). Main analysis Cost and effect differences between the intervention and control group were estimated using Seemingly Unrelated Regression (SUR) analysis which preserves the correlation between costs and effects. Age and gender were included as confounders in all analyses (67), and physical activity was included in the analyses with falls and fall-related injuries as effect outcome (49). Other baseline characteristics were considered confounders if the estimate for the randomization group changed by at least 10% after adding that variable to the model. The confounders can be found in the footnote below the tables in the results section. We calculated ICERs by dividing the differences in costs by the differences in effects. To estimate the uncertainty surrounding the ICERs, we used bias-corrected accelerated bootstrapping with 5,000 replications (269). ICERs and their associated uncertainty were presented on the Cost-Effectiveness Plane (CE-plane). Moreover, the Incremental Net Monetary Benefit (INMB) approach was applied to estimate Cost-Effectiveness Acceptability Curves (CEAC). For falls and fall-related injuries, we used a willingness-to-pay (WTP) threshold of €10,000 per fall-related injury prevented, based on the average healthcare costs of €9,370 per fall after an ED visit (133). For QALYs, a WTP threshold was used of €50,000 per QALY gained, based on the currently applied range in the Netherlands (270).
RkJQdWJsaXNoZXIy MjY0ODMw