A recent study published in the journal NPJ Exercise Medicine and Health explored the relationship between dynamic complexity in daily step counts and abrupt changes in physical activity among adults. The research focused on identifying early warning signs, particularly for decreases in activity, to inform adaptive intervention strategies.
Physical activity is crucial for overall health, yet maintaining consistent activity levels can be challenging. Individuals often experience periods of reduced activity or complete disengagement. Recognizing and intervening during these vulnerable times is essential for developing effective support mechanisms that encourage long-term behavioral changes. Mobile sensing technologies, combined with complex systems theory, view behavioral activity as a dynamic process influenced by various factors that can lead to sudden shifts. Early warning signals, such as critical fluctuations, often precede these transitions and can be detected in behavioral data. Dynamic complexity, which integrates volatility and distribution, is a promising indicator for identifying instability before significant behavioral changes. Previous studies have linked increased dynamic complexity to impending declines in physical activity, highlighting its potential as a predictor of relapse. However, its ability to predict increases in activity is less clear, suggesting different mechanisms for activity decline versus recovery. Current research is often limited by small, homogenous samples, emphasizing the need for validation in larger, more diverse real-world populations and robust examination across various definitions of sudden behavioral change.
The study analyzed real-world data from the Kiplin dataset, collected in France between January 2019 and April 2023, to investigate the connection between dynamic complexity and sudden alterations in walking behavior, focusing primarily on activity reduction and secondarily on activity increase. From 247,059 Kiplin app users, 21,425 participants were included after excluding those with insufficient data, extremely high daily step counts, or significant missing data. The average participant age was 43 years, with 60% being women. The mean daily step count was 7,817 steps, and data completeness was high, with an average time series of 256 days. Sudden changes, defined as a ±30% shift from a participant's median level lasting at least 7 days, were prevalent, with 96% of participants experiencing at least one sudden decrease and 94% experiencing at least one sudden increase. Participants typically had about three sudden gains and three sudden losses during the study. Sudden increases lasted a median of 20 days, while sudden decreases lasted longer, averaging 33 days. These patterns often followed a cyclical nature, with gains frequently followed by losses and vice versa. Dynamic complexity, normalized for each participant, consistently predicted sudden declines in activity, particularly for shorter and more pronounced decreases. This association remained statistically significant even for longer-duration losses. The Kiplin intervention was linked to a decreased probability of sudden losses and an increased probability of sudden gains, especially for shorter events. The predictive power of dynamic complexity was particularly strong among less active individuals, suggesting its relevance as an early warning indicator for this group. Other demographic factors like gender, age, BMI, and device type did not significantly moderate these associations. The study's findings indicate that monitoring fluctuations in daily step counts can offer valuable insights into impending changes in physical activity, paving the way for proactive interventions.
This study underscores the potential of critical fluctuations in daily step counts as early indicators of significant drops in physical activity. These findings provide a solid foundation for developing and testing timely interventions designed to avert sudden declines in physical activity at a population level. While the results are encouraging, certain limitations should be acknowledged, including the regression-tree method for detecting changes, the absence of direct wear-time data, potential biases from missing data, and the generalizability constraints imposed by the inclusion criteria. The models identified predictive associations at the group level but did not evaluate individualized risk prediction or its sensitivity and specificity. Furthermore, a formal assessment of selection bias was not possible due to unavailable data for excluded registrants. Despite established links between early warning signals and sudden activity losses, the precise underlying mechanisms warrant further investigation. Continued research is essential to enhance personalized risk prediction and determine if early warning signal-triggered interventions can effectively prevent abrupt declines in physical activity, fostering healthier and more active lives.