Dynamic modelling of single-case (N-of-1) data: Challenges and novel applications

Introduction: Single-case studies are increasingly recognised as a valid and efficient mechanism for making individualized evidence-based treatment decisions. Statistical analyses of N-of-1 data require accurate modelling of the outcome variable while accounting for its distribution, time-related trend and error structures (e.g. autocorrelation) as well as reporting readily usable effect sizes for clinical decision making. A substancial number of statistical approaches have been documented but no consensus exist on which method is most appropriate for which kind of design and data.

Methods: We discuss, from a statistical perspective, the limitations and advantages of N-of-1 studies. We describe several regression methods for the analysis of N-of-1 data, borrowing ideas from longitudinal and event history methodologies which explicitly incorporate the role of time and the dependence of future on past. The aims include identifying predictors of response, describing adaptive changes over time, or predicting future behaviour given prior history.

Results: The methods are applied to data from two N-of-1 observational studies of physical activity (PA) during retirement transition and weight loss maintenance and one N-of-1 randomized clinical trial related to PA and Type 2 diabetes. The studies span several outcome types: dichotomous (PA or no PA), continuous (weight) and count (number of PA bouts). Our approach is shown to be adaptable to different types of outcomes, flexible, powerful and capable with dealing with the different challenges inherent to N-of-1 modelling.

Conclusions: Dynamic modelling has the potential to expand access of N-of-1 researchers to robust and user-friendly statistical methods.