Predicting Brain Aging (Brain Age Gap) from Biomedical and Lifestyle Variables
Poster No:
1144
Submission Type:
Abstract Submission
Authors:
Mostafa Mahdipour1,2, Somayeh Maleki Balajoo1,2, Federico Raimondo1,2, Eliana Nicolaisen-Sobesky1,2, Shammi More1,2, Felix Hoffstaedter1,2, Masoud Tahmasian1,2,3, Simon Eickhoff1,2, Sarah Genon1,2
Institutions:
1Institute of Systems Neuroscience, Heinrich Heine University Düsseldorf, Düsseldorf, Germany, 2Institute of Neuroscience and Medicine (INM-7: Brain and Behaviour), Research Centre Jülich, Jülich, Germany, 3Department of Nuclear Medicine, Faculty of Medicine and University Hospital Cologne, University of Cologne, Cologne, Germany
First Author:
Mostafa Mahdipour
Institute of Systems Neuroscience, Heinrich Heine University Düsseldorf|Institute of Neuroscience and Medicine (INM-7: Brain and Behaviour), Research Centre Jülich
Düsseldorf, Germany|Jülich, Germany
Institute of Systems Neuroscience, Heinrich Heine University Düsseldorf|Institute of Neuroscience and Medicine (INM-7: Brain and Behaviour), Research Centre Jülich
Düsseldorf, Germany|Jülich, Germany
Co-Author(s):
Somayeh Maleki Balajoo
Institute of Systems Neuroscience, Heinrich Heine University Düsseldorf|Institute of Neuroscience and Medicine (INM-7: Brain and Behaviour), Research Centre Jülich
Düsseldorf, Germany|Jülich, Germany
Institute of Systems Neuroscience, Heinrich Heine University Düsseldorf|Institute of Neuroscience and Medicine (INM-7: Brain and Behaviour), Research Centre Jülich
Düsseldorf, Germany|Jülich, Germany
Federico Raimondo
Institute of Systems Neuroscience, Heinrich Heine University Düsseldorf|Institute of Neuroscience and Medicine (INM-7: Brain and Behaviour), Research Centre Jülich
Düsseldorf, Germany|Jülich, Germany
Institute of Systems Neuroscience, Heinrich Heine University Düsseldorf|Institute of Neuroscience and Medicine (INM-7: Brain and Behaviour), Research Centre Jülich
Düsseldorf, Germany|Jülich, Germany
Eliana Nicolaisen-Sobesky
Institute of Systems Neuroscience, Heinrich Heine University Düsseldorf|Institute of Neuroscience and Medicine (INM-7: Brain and Behaviour), Research Centre Jülich
Düsseldorf, Germany|Jülich, Germany
Institute of Systems Neuroscience, Heinrich Heine University Düsseldorf|Institute of Neuroscience and Medicine (INM-7: Brain and Behaviour), Research Centre Jülich
Düsseldorf, Germany|Jülich, Germany
Shammi More
Institute of Systems Neuroscience, Heinrich Heine University Düsseldorf|Institute of Neuroscience and Medicine (INM-7: Brain and Behaviour), Research Centre Jülich
Düsseldorf, Germany|Jülich, Germany
Institute of Systems Neuroscience, Heinrich Heine University Düsseldorf|Institute of Neuroscience and Medicine (INM-7: Brain and Behaviour), Research Centre Jülich
Düsseldorf, Germany|Jülich, Germany
Felix Hoffstaedter
Institute of Systems Neuroscience, Heinrich Heine University Düsseldorf|Institute of Neuroscience and Medicine (INM-7: Brain and Behaviour), Research Centre Jülich
Düsseldorf, Germany|Jülich, Germany
Institute of Systems Neuroscience, Heinrich Heine University Düsseldorf|Institute of Neuroscience and Medicine (INM-7: Brain and Behaviour), Research Centre Jülich
Düsseldorf, Germany|Jülich, Germany
Masoud Tahmasian
Institute of Systems Neuroscience, Heinrich Heine University Düsseldorf|Institute of Neuroscience and Medicine (INM-7: Brain and Behaviour), Research Centre Jülich|Department of Nuclear Medicine, Faculty of Medicine and University Hospital Cologne, University of Cologne
Düsseldorf, Germany|Jülich, Germany|Cologne, Germany
Institute of Systems Neuroscience, Heinrich Heine University Düsseldorf|Institute of Neuroscience and Medicine (INM-7: Brain and Behaviour), Research Centre Jülich|Department of Nuclear Medicine, Faculty of Medicine and University Hospital Cologne, University of Cologne
Düsseldorf, Germany|Jülich, Germany|Cologne, Germany
Simon Eickhoff
Institute of Systems Neuroscience, Heinrich Heine University Düsseldorf|Institute of Neuroscience and Medicine (INM-7: Brain and Behaviour), Research Centre Jülich
Düsseldorf, Germany|Jülich, Germany
Institute of Systems Neuroscience, Heinrich Heine University Düsseldorf|Institute of Neuroscience and Medicine (INM-7: Brain and Behaviour), Research Centre Jülich
Düsseldorf, Germany|Jülich, Germany
Sarah Genon
Institute of Systems Neuroscience, Heinrich Heine University Düsseldorf|Institute of Neuroscience and Medicine (INM-7: Brain and Behaviour), Research Centre Jülich
Düsseldorf, Germany|Jülich, Germany
Institute of Systems Neuroscience, Heinrich Heine University Düsseldorf|Institute of Neuroscience and Medicine (INM-7: Brain and Behaviour), Research Centre Jülich
Düsseldorf, Germany|Jülich, Germany
Introduction:
The Brain Age Gap (BAG) can be considered as an indicator of the brain health [1, 2]. BAG is defined as the difference between an individual's chronological age and the age predicted by a machine learning (ML) algorithm based on individual brain features. While some studies demonstrated univariate associations between the BAG and several lifestyle and biomedical variables [2, 3], a substantial gap persists in understanding the multivariate association between these factors and BAG. Here, we addressed this question by using a wide range of biomedical, lifestyle, and sociodemographic variables conjointly to predict the BAG in a large population of the UK Biobank.
Methods:
We built a ML model to predict an individual's brain age using brain structural features (cortical and subcortical parcellated grey matter volume) in a subset of specifically healthy participants in the UK Biobank (n =5,025, age range 46-82 years, 2,579 females). In particular, we applied ridge regression implemented in the Julearn package [4]. Subsequently, the optimized predictive brain model was applied to predict brain age in the remaining UK Biobank participants (n =34,365, age range 44-82 years, 18,128 females) (Fig 1A). The BAG was then computed following the adjustment of predicted brain ages for proportional bias based on regression parameters from the training set [5].
For predicting an individual's BAG, we selected 157 variables covering biomedical (e.g., cardiovascular, respiratory, and body metabolism), lifestyle (e.g. smoking and diet) and sociodemographic (e.g.: socioeconomical status, family and social life) variables in a subset of 7,736 participants not included in the healthy subset used to build the age prediction model (age range 44-81 years, 4,272 females). In order to train the model, we used a random forest algorithm as implemented in Julearn [4] with a nested cross-validation approach including 10 inner and outer folds. Notably, we controlled for possible confounds (age, age2, sex, height, and volumetric scaling from the T1 image to standard space).
For predicting an individual's BAG, we selected 157 variables covering biomedical (e.g., cardiovascular, respiratory, and body metabolism), lifestyle (e.g. smoking and diet) and sociodemographic (e.g.: socioeconomical status, family and social life) variables in a subset of 7,736 participants not included in the healthy subset used to build the age prediction model (age range 44-81 years, 4,272 females). In order to train the model, we used a random forest algorithm as implemented in Julearn [4] with a nested cross-validation approach including 10 inner and outer folds. Notably, we controlled for possible confounds (age, age2, sex, height, and volumetric scaling from the T1 image to standard space).
Results:
We achieved a high level of accuracy in predicting chronological age using brain structural features. The average cross-validation Mean Absolute Error (MAE) of 3.75 years closely mirrored the MAE of the best model applied to the population data, which was 3.93 years and the correlation between chronological and predicted age was strong (r = 0.75 for the prediction model and 0.76 for the population data, Fig 1B). Using bias correction on the population's predicted brain age led to a relatively unbiased BAG with regards to age (Fig 1C).
We then found that individual BAG could be robustly predicted to some extent by phenotypical variables with an average (across test folds) correlation between the predicted BAG and calculated BAG of 0.239±0.02, MAE of 4.54±0.10 years, and a Root Mean Square Error of 5.72±0.12 years. Examining mean features importance (Fig 2) further reveals that most features are relevant for the prediction. Nevertheless, biomedical factors, closely followed by lifestyle factors, appear to play a relatively more important role than sociodemographic variables.
We then found that individual BAG could be robustly predicted to some extent by phenotypical variables with an average (across test folds) correlation between the predicted BAG and calculated BAG of 0.239±0.02, MAE of 4.54±0.10 years, and a Root Mean Square Error of 5.72±0.12 years. Examining mean features importance (Fig 2) further reveals that most features are relevant for the prediction. Nevertheless, biomedical factors, closely followed by lifestyle factors, appear to play a relatively more important role than sociodemographic variables.
Conclusions:
We successfully developed a brain age prediction model in a healthy sample that enables to compute brain age gap as a sensitive estimator of individual brain structural health in an aging population. Although the relationships between any individual factor and brain structural health can be seen as negligible based on previously reported effect sizes [6], here we showed that considering a range of variables jointly enables a decent prediction of BAG. Although no set of variables appear to play a crucial role here, biomedical factors related to body metabolism and cardiovascular systems, as well lifestyle factors directly influencing these later (such as smoking and alcohol consumption) appear relatively important for structural brain health.
Lifespan Development:
Aging 1
Normal Brain Development: Fetus to Adolescence
Lifespan Development Other
Modeling and Analysis Methods:
Classification and Predictive Modeling
Multivariate Approaches 2
Keywords:
Aging
Cortex
Data analysis
Data Registration
Machine Learning
Multivariate
Phenotype-Genotype
Sub-Cortical
Other - Uk Biobank
1|2Indicates the priority used for review
Provide references using author date format
2.Hamdan, S., et al. (2023). "Julearn: an easy-to-use library for leakage-free evaluation and inspection of ML models." arXiv preprint arXiv:2310.12568.
3.Miller, K. L., et al. (2016). "Multimodal population brain imaging in the UK Biobank prospective epidemiological study." Nature neuroscience 19(11): 1523-1536.
4.More, S., et al. (2023). "Brain-age prediction: A systematic comparison of machine learning workflows." NeuroImage 270: 119947.
5.Smith, S. M., et al. (2019). "Estimation of brain age delta from brain imaging." NeuroImage 200: 528-539.
6.Tian, Y. E., et al. (2023). "Heterogeneous aging across multiple organ systems and prediction of chronic disease and mortality." Nature Medicine 29(5): 1221-1231.