guest ::
| Home > Publications database > Handgrip strength prediction using behavioural and anthropometric features in 179K individuals from the UK Biobank |
| Home > Publications database > Handgrip strength prediction using behavioural and anthropometric features in 179K individuals from the UK Biobank |
| Poster (After Call) | FZJ-2024-01225 |
; ; ; ; ; ;
2023
This record in other databases:
Please use a persistent id in citations: doi:10.34734/FZJ-2024-01225
Abstract: Handgrip strength (HGS) is an inexpensive, and non-invasive biomarker for assessing motor performance and identifying individuals at risk of motor impairment. Normative models can be used to identify abnormalities and brain changes that affect HGS and other motor functions. The current work examined machine learning-based approaches to predict HGS using a wide range of behavioural phenotypes and anthropometric measures. The data were obtained from healthy controls in the UK Biobank. Participants with dominant handgrip strength < 4 kg were excluded. We included 30 behavioural phenotypes, such as cognitive functions, anxiety, depression, and neuroticism. Anthropometric measures included BMI, height, and waist-to-hip circumference ratio. These features were analysed with or without the inclusion of gender as an additional feature, as males and females are known to show HGS differences. Linear SVM and Random Forest were used as predictive models. The performance was evaluated using the R2 score. The cross-validation scheme consisted of 10 repetitions and 5 folds. We also examined sample size effects. The study included 2,143 subjects with completed assessments of all behavioural phenotypes. The behavioural features had the lowest R2 score (median = 0.09), while anthropometric features showed a relatively higher R2 score (median = 0.49). Adding gender as a feature significantly increased the prediction scores but building the model separately for males and females decreased accuracy. In males, predictions were better than in females with anthropometric features. Furthermore, (non-)dominant HGS could be predicted, but the Left + Right HGS predictions were more accurate. Also, the sample size effects were analysed for anthropometrics from 10% (N=17,957) to 100% (N=179,542). The results demonstrate that the lower sample size of anthropometric features shows a high variance and the performance saturates around N=71k. Gender is a strong confound when modelling HGS. Anthropometric features are better predictors of HGS than behavioural features. Future work involves applying the pipeline to stroke patients to identify brain correlates, such as white matter intensity, and assessing its feasibility in clinical settings. In conclusion, HGS prediction using anthropometric and behavioural features can inform future research in the field of motor performance and have implications for early disease diagnosis and treatment.
|
The record appears in these collections: |
PDF