Estimating chronological and brain age using risk-taking behavior under uncertainty

This study demonstrates that computational parameters derived from risk-taking tasks, such as the Iowa Gambling Task and Balloon Analogue Risk Task, serve as superior functional markers for predicting both chronological and brain age compared to traditional cognitive screening measures, revealing specific associations with distributed brain networks involved in aging.

The Gist

Imagine your brain is like a high-tech car. As the car gets older (chronological age), the engine might start to sputter, the paint might fade, and the sensors might get a little sluggish. Usually, mechanics (doctors) check the car's health by looking at the dashboard lights (standard cognitive tests like the MMSE or MoCA). But sometimes, the dashboard looks perfectly fine even though the engine is starting to wear out.

This paper suggests a new way to check the car's health: watching how the driver makes risky decisions.

Here is the story of the study, broken down simply:

1. The Experiment: Two Games of Chance

The researchers gathered two groups of people: 55 young adults (the "new cars") and 112 older adults (the "vintage cars"). They asked everyone to play two specific games:

• The Iowa Gambling Task (IGT): Imagine you are at a casino with four decks of cards. Some decks give you money but occasionally take a huge chunk away. Other decks give you smaller wins but rarely take big losses. You have to figure out which decks are "good" and which are "bad" just by playing. This tests long-term learning and memory.
• The Balloon Analogue Risk Task (BART): Imagine a balloon. Every time you pump it, you earn money. But if you pump it too much, it pops, and you lose everything. You have to decide: Do I pump one more time for a little extra cash, or stop now and bank my winnings? This tests moment-to-moment risk-taking.

2. The Problem: The Dashboard Lies

When the researchers looked at the final scores of these games (the "dashboard"), they found something surprising: The older adults played just as well as the young adults.

If you only looked at the final score, you'd think the older drivers were just as sharp as the young ones. But the researchers knew that "average" scores often hide the fact that the way people play has changed.

3. The Solution: The "Black Box" Decoder

Instead of just looking at the final score, the researchers used computational modeling. Think of this as opening the car's "black box" (the flight recorder) to see exactly how the driver was thinking during every single move.

They broke the decision-making process down into specific "ingredients":

• Learning Rate: How fast do you learn from a mistake?
• Memory Decay: How quickly do you forget what happened last time?
• Loss Aversion: How much does the fear of losing money stop you from taking a chance?
• Feedback Sensitivity: How much does a win or loss change your next move?

The Discovery:
When they looked at these "ingredients," they found clear differences. The older adults weren't playing differently in terms of results, but they were playing differently in process:

• They forgot past outcomes faster (like a leaky bucket).
• They were more afraid of losing money (more cautious).
• They were slower to update their strategy based on new information.

4. The Connection to the Brain

The researchers then took MRI scans of the participants' brains to measure "Brain Age." This is like checking the actual wear and tear on the engine, regardless of how many years the car has been on the road.

• The Standard Tests Failed: The traditional cognitive tests (the dashboard lights) couldn't predict the brain's actual age or condition.
• The Risk Games Succeeded: The "ingredients" from the games (like how sensitive someone was to feedback or how fast they learned) were highly accurate predictors of both the person's actual age and their brain's health.

The Map:

• The IGT (the casino game) was linked to the basal ganglia (the deep, habit-forming part of the brain, like the automatic transmission).
• The BART (the balloon game) was linked to a wider network including the prefrontal cortex (the CEO of the brain that handles planning and impulse control).

The Big Takeaway

This study is like realizing that to know if a car is truly aging, you shouldn't just ask the driver, "Do you feel okay?" (which is what standard tests do). Instead, you should watch how they merge onto the highway or react to a sudden brake light.

In simple terms:
Even if older adults can still pass a standard memory test, the way they take risks and learn from mistakes reveals subtle changes in their brain that we can't see otherwise. By analyzing these "risky decisions" with math, we can detect early signs of brain aging long before the person starts showing obvious signs of memory loss.

Why does this matter?
It gives us a new, sensitive tool to spot brain aging early. Just like a mechanic who can hear a tiny rattle in the engine before the car breaks down, these "risk-taking markers" might help doctors intervene earlier to keep our brains healthy as we age.

Technical Summary

1. Problem Statement

Risky decision-making under uncertainty relies on distributed brain networks (prefrontal, striatal, cingulate) that are vulnerable to aging. While age-related declines in decision-making often precede overt cognitive impairment, traditional behavioral metrics (e.g., net scores on tasks) and standard neuropsychological screening tools (e.g., MMSE, MoCA) often fail to detect these subtle, preclinical changes. There is a lack of understanding regarding whether latent decision-making mechanisms, when quantified via computational modeling, can serve as sensitive behavioral markers for both chronological aging and "brain age" (structural brain deterioration).

2. Methodology

Participants:

• Sample: 167 total participants (55 young adults, aged 19–31; 112 healthy older adults, aged 54–88).
• Exclusion Criteria: History of neurological/psychiatric disease, MRI contraindications, and left-handedness (to avoid lateralization confounds).
• Cognitive Screening: All participants underwent MoCA and MMSE assessments.

Experimental Tasks:

• Iowa Gambling Task (IGT): A modified version where participants choose between advantageous (C/D) and disadvantageous (A/B) decks to maximize rewards.
• Balloon Analogue Risk Task (BART): Participants pump a balloon (red/high-risk or blue/low-risk) to earn money, balancing potential gains against the risk of explosion.

Computational Modeling:

• IGT: Data were fitted to four models. The Value-Plus-Perseveration (VPP) model was selected as the best fit based on LOOIC and WAIC criteria. Key parameters included:
  • $\alpha$ (Feedback sensitivity), $\lambda$ (Loss aversion), $K$ (Perseverance), $w$ (Learning rate), $A$ (Memory decay).
• BART: Data were fitted to the Exponential-Weight Mean–Variance (EWMV) model. Parameters were analyzed separately for high-risk (red) and low-risk (blue) conditions. Key parameters included:
  • $\eta$ (Learning rate/Updating exponent), $\lambda$ (Loss aversion), $\phi$ (Prior belief of burst), $\tau$ (Inverse temperature).

Neuroimaging:

• Acquisition: 3.0T MRI with T1-weighted structural imaging.
• Processing: Voxel-Based Morphometry (VBM) using SPM12 and DARTEL for nonlinear registration.
• Brain Age Estimation: A Generalized Linear Model (GLM) predicted "brain age" based on gray matter volume (GMV) across 90 regions defined by the Automated Anatomical Labeling (AAL) atlas.

Statistical Analysis:

• Group comparisons (Young vs. Older) using t-tests and Mann-Whitney U tests.
• Partial correlations and linear regression models to assess the predictive power of computational parameters vs. traditional cognitive scores (MMSE/MoCA) for Chronological Age and Brain Age.
• Correlation analysis between model parameters and regional GMV.

3. Key Contributions

1. Superior Sensitivity of Computational Markers: Demonstrated that while raw behavioral scores (net gains/pumps) showed no significant difference between young and older adults, computational parameters revealed distinct, significant age-related alterations in latent cognitive processes.
2. Prediction of Brain Health: Established that computational parameters from risk-taking tasks significantly predict both chronological age and MRI-derived brain age, whereas standard cognitive screening tools (MMSE, MoCA) failed to show significant predictive power.
3. Task-Specific Neural Correlates: Mapped specific decision-making parameters to distinct neuroanatomical substrates, validating that IGT and BART engage different neural networks (Basal Ganglia vs. Prefrontal/Cingulate networks).
4. First Systematic Comparison: To the authors' knowledge, this is the first study to systematically compare the neural and cognitive mechanisms of IGT and BART in the context of aging using computational modeling and structural MRI.

4. Key Results

Behavioral & Modeling Differences:

• IGT (Older Adults): Showed higher memory decay ($A$), higher loss aversion ($\lambda$), and higher perseverance ($K$), but lower feedback sensitivity ($\alpha$) and learning rate ($w$).
• BART (Older Adults): Showed increased loss aversion ($\lambda$) and reduced learning rate ($\eta$) in both risk conditions. In high-risk conditions, they also showed altered prior beliefs ($\phi$).

Prediction of Age:

• Chronological Age:
  • Model 1 (MMSE/MoCA): No significant prediction ($R^2 = 0.037, p=0.123$).
  • Model 2/3 (IGT/BART parameters): Significant prediction ($R^2 = 0.117$ and $0.074$).
  • Combined Model: Strongest prediction ($R^2 = 0.171, p=0.002$). Key predictors were feedback sensitivity ($\alpha$) and updating exponent ($\eta$).
• Brain Age:
  • Model 1 (MMSE/MoCA): No significant prediction ($R^2 = 0.026, p=0.237$).
  • Combined Model (IGT + BART parameters): Significant prediction ($R^2 = 0.163, p=0.001$). Key predictors were feedback sensitivity ($\alpha$), perseverance decay ($K$), and updating exponent ($\eta$).

Neural Correlates (Gray Matter):

• IGT Parameters: Primarily associated with the basal ganglia (caudate, putamen, pallidum) and superior occipital gyrus.
• BART Parameters: Associated with a broader network including the prefrontal cortex, insula, and cingulate gyrus.
• Cognitive Scores: No significant correlations were found between GMV and MMSE/MoCA scores.

5. Significance and Implications

• Early Detection: Computational parameters of risk-taking behavior offer a more sensitive tool for detecting preclinical cognitive decline and brain aging than traditional neuropsychological tests, which often suffer from ceiling effects.
• Mechanistic Insight: By decomposing behavior into specific cognitive components (e.g., learning efficiency, loss aversion), the study identifies which specific processes are most vulnerable to aging, providing targets for future interventions.
• Biomarker Potential: Risk-taking parameters serve as reliable functional markers that correlate with structural brain deterioration, suggesting they could be used to monitor "brain health" and the rate of brain aging in clinical and research settings.
• Limitations: The study is cross-sectional (cannot infer causality), limited to two specific tasks (IGT/BART), and relies on a specific brain age prediction model. Future longitudinal studies are needed to validate these parameters as predictors of future cognitive decline.