Resilience characterized and quantified from physical activity data: A tutorial in R

• Published in: Current Issues in Sport Science 2023-02, Vol.8 (2), p.54
• Main Authors: Baretta, Dario, Koch, Sarah, Cobo, Inés, Castano-Vinyals, Gemma, De Cid, Rafael, Carreras, Anna, Buekers, Joren, Garcia-Aymerich, Judith, Inauen, Jennifer, Chevance, Guillaume
• Format: Article
• Language: English
• Subjects: ecological momentary assessment; mood; physical acitivity; sedentary behavior; stress
• ISSN: 2414-6641; 2414-6641
• DOI: 10.36950/2023.2ciss054

Rationale Consistent physical activity (PA) is key for health and well-being; however, various events (e.g., pandemic outbreaks, acute diseases) can act as stressors for PA, and thus may be followed by a decrease in PA levels. The process of recovering from such stressors and bouncing back to the previous state of PA can be referred to as resilience. Quantifying resilience is fundamental to assess the impact of stressors on daily PA and to identify factors and strategies to foster adaptive capacities. However, PA time series are typically characterized by evident daily fluctuations that mask the underlying trajectory and, therefore, prevent from detecting the point in time when the system consistently recovers from the stressor. Objective: In this paper, we present a methodological approach to identify the recovery point and ultimately quantify the resilience process from PA data using the area under the curve (AUC). Methods: As use case to illustrate the methodology, we quantified resilience in step count for eight participants following the stressor represented by the start of the first COVID-19 lockdown (March 15, 2020). Step count time series were collected in Barcelona, as part of the COVICAT study, in the period between October 1, 2019 and September 30, 2020. Steps were collected daily using wrist-worn devices. The methodology is implemented in R and a tutorial is available on Open Science Framework. Results: We applied the following analytical steps to each participant’s time series. 1) We calculated the pre-stressor baseline level for step count as the median step count before the first COVID-19 lockdown. 2) We fitted multiple growth models (i.e., linear, quadratic, generalized additive) to each participant’s post stressor time series and identified the best model using the Root Mean Squared Error (RMSE). 3) We used the fitted values from the selected model to identify the point in time when the participants recovered from the stressor (bouncing back to the baseline level). 4) We quantified the AUC and, ultimately, resilience as the cumulative difference between baseline level and each fitted value before the recovery point. Further resilience features were extracted to capture the different aspects of the process and will be presented. Conclusions: Our contribution offers a methodological guide to quantify resilience from PA data and proposes a ready-to-go toolbox that can be easily applied by interested researchers and foster further scientific i