Clustering Runners' Response to Different Midsole Stack Heights: A Field Study

Jannik Koegel, Stacy Huerta, Markus Gambietz, Martin Ullrich, Christian Heyde, Eva Dorschky, Björn M. Eskofier2024 · DOI

TL;DR

This study investigates how runners' biomechanics change when wearing shoes with different midsole stack heights. By using wearable sensors and clustering techniques, they identified distinct groups of runners who respond differently to these shoe variations, suggesting personalized recommendations could be possible.

What Problem Does This Paper Address?

Advanced footwear technology (AFT), particularly shoes with high midsole stack heights, has shown promise in improving running economy. However, the physiological benefits are known to be highly individual, and the specific biomechanical responses that lead to these benefits remain unclear. Previous research has often been conducted in laboratory settings on treadmills, and the focus has primarily been on metabolic costs rather than detailed biomechanical adaptations in a real-world running environment. This paper aims to bridge this gap by exploring how individual runners' biomechanics adapt to varying midsole stack heights in a field setting, using wearable sensor data to identify distinct response patterns.

Methods

The study employed a field-based approach using wearable sensors to capture running biomechanics. Thirty-one runners participated, performing running trials on an outdoor course wearing three different experimental shoe conditions with varying midsole stack heights: 25 mm, 35 mm, and 45 mm. All shoes shared the same upper, outsole, and carbon plate, with only the midsole foam (TPEE) and stack height differing.

A STRYD sensor was attached to each runner's shoe to collect data on eight variables: cadence, speed, flight time, ground contact time, stride length, leg spring stiffness, vertical oscillation, and peak ground reaction force. The collected data was normalized to the 25 mm shoe condition for each participant.

To identify distinct response patterns, the researchers performed a Principal Component Analysis (PCA) to reduce the dimensionality of the data, retaining two principal components that explained 85.6% of the variance. These components were then used as input for a hierarchical clustering analysis using Ward's method, aiming to group runners with similar biomechanical responses. The optimal number of clusters was determined using Silhouette and Elbow methods, resulting in three distinct clusters.

Datasets

The study utilized data collected from 31 human participants during outdoor running trials. The dataset consists of biomechanical parameters captured by a STRYD sensor across three different shoe conditions. No pre-existing datasets were used for training or evaluation; instead, the paper introduces a novel dataset of field-based running biomechanics in response to varying footwear stack heights.

Key Insights and Ideas

  • Individualized Responses are Key: The core finding is that runners do not respond uniformly to changes in midsole stack height. This reinforces the idea that "one size fits all" doesn't apply to advanced footwear technology.
  • Clustering Reveals Distinct Patterns: By clustering biomechanical data, the study successfully identified three distinct groups of runners, suggesting that specific biomechanical adaptations are characteristic of subgroups.
  • Leg Spring Stiffness and Vertical Oscillation Drive Variance: These two parameters were identified as the most significant contributors to the observed differences in running patterns across the clusters.
  • Potential for "Non-Responders": One cluster appears to show minimal changes in their running mechanics across different stack heights, suggesting they might not benefit as much from AFT.
  • Economic vs. Less Economical Responders: The study proposes that clusters showing increased leg spring stiffness and decreased vertical oscillation with higher stack heights could be considered "positive" or more economical responders, while the opposite pattern might indicate "negative" responders.

Key Results

  • The study successfully clustered 31 runners into three distinct groups based on their biomechanical responses to different midsole stack heights.
  • No significant differences were found between the clusters in terms of anthropometric data (body height, body weight, leg length) or average running speed, indicating that these factors did not drive the observed biomechanical clustering.
  • Cluster 0: Showed a gradual increase in leg spring stiffness and a decrease in vertical oscillation and ground contact time with increasing stack heights. Flight times were longer, and peak ground reaction force was higher.
  • Cluster 1: Exhibited a decrease in leg spring stiffness and an increase in vertical oscillation and ground contact time with higher stack heights. Flight times and peak ground reaction forces were reduced.
  • Cluster 2: Displayed a more complex pattern, with leg spring stiffness decreasing initially then slightly increasing with the highest stack height. Vertical oscillation and ground contact time increased, with flight time reduced then increased. Peak ground reaction forces were slightly reduced.
  • The PCA explained 85.6% of the data variance with two principal components, justifying their use for clustering.

Limitations and Future Work

  • Weight-Matched Shoes: The study did not use weight-matched shoes, as the focus was on the realistic weight increase associated with higher stack heights. While the weight difference was small (44g), it's acknowledged as a potential factor.
  • No Oxygen Consumption Measurement: The study relied on biomechanical parameters and did not directly measure oxygen consumption or running economy. Therefore, the interpretation of "economical" vs. "less economical" responders is based on biomechanical proxies and requires verification.
  • Foam Properties: Using the same foam across all stack heights means that changes in cushioning and energy return are inherent to the stack height itself, making it difficult to isolate the effect of stack height alone. Future work could involve engineering foams specifically for different stack heights to better isolate parameters.
  • Sensor Validation: While the STRYD sensor is validated for several parameters, its accuracy for vertical oscillation and peak ground reaction force is still under investigation.
  • Speed Control: Runners were instructed to run at a self-selected speed reflecting their subjective effort. While this mimics real-world running, it introduces potential speed variations that could influence other biomechanical parameters. Future studies might explore speed-controlled runs.
  • Future Directions: The authors suggest increasing participant numbers to enable more sophisticated machine learning methods for predicting cluster classification. They also highlight the potential for using this approach to recommend the most economical shoe condition for individual runners based on wearable sensor data, which would require further validation with oxygen consumption measurements. Incorporating individual leg length for more personalized analysis is also recommended for future sensor development.

Summary updated

BibTeX
@misc{koegel2024clustering,
  author = {Jannik Koegel and Stacy Huerta and Markus Gambietz and Martin Ullrich and Christian Heyde and Eva Dorschky and Björn M. Eskofier},
  title = {{Clustering Runners' Response to Different Midsole Stack Heights: A Field Study}},
  year = {2024},
  doi = {10.3390/s24144694},
}
Figure