Authors: Krone, Boessen, Bijlsma, van Stokkum, Clabbers, & Pasman, (2020)
The statistical analysis of data collected from N-of-1 trials is often perceived to be a key challenge for the widespread adoption of these types of designs. Krone et al’s paper, ‘The Possibilities of The Use of N-of-1 and Do-It-Yourself Trials in Nutritional Research’, investigates different statistical approaches for aggregating the results from multiple N-of-1 studies. In aggregated N-of-1 studies the data from several N-of-1 studies are ‘pooled’. Aggregated N-of-1 studies are valuable because they can be used to provide an estimate of the intervention effect at both the individual- and population-level. Krone et al. aim to answer two key questions (1) is better to use meta-analysis or linear mixed modelling approaches to aggregate data from several N-of-1 studies? (2) is linear mixed modelling using a Bayesian framework more accurate than linear mixed modelling using a freqentist approach? Through some creative simulation modelling, Krone et al. show that both frequentist and Bayesian linear mixed modelling approaches are more accurate than meta-analysis. Furthermore, they show that when credible priors are available, linear mixed modelling using a Bayesian framework has the advantage of more power over a frequentist approach, which is particularly useful in analyses of a small number of participants. Frequentist and Bayesian linear mixed modelling approaches appear to be equivalent when no priors are available. The findings reported in this paper support recommendations by others (e.g. Zucker et al., 1997) to use linear mixed modelling as the most suitable way to aggregate N-of-1 data across participants.
Click here to view the article.
Full citation: Krone, T., Boessen, R., Bijlsma, S., van Stokkum, R., Clabbers, N.D. & Pasman, WJ. (2020). The possibilities of the use of N-of-1 and do-it-yourself trials in nutritional research. PLoS ONE. 15(5):e0232680. https://doi.org/10.1371/journal.pone.0232680