Development and validation of neurological health score using machine learning algorithms

Researchers developed and validated a machine learning-based Neurological Health Score (NHS) using genetic, lifestyle, and biochemical data, demonstrating that higher scores significantly correlate with a reduced risk of various neurological disorders and improved social well-being.

The Gist

The "Brain Weather Forecast": Understanding the Neurological Health Score

Imagine you are trying to predict whether a massive storm is going to hit your city. You wouldn't just look at the clouds; you’d look at the humidity, the wind speed, the temperature, and even the ocean currents.

For a long time, doctors have been trying to do the same thing with our brains. They might look at your genetics (your "internal blueprint") or your lifestyle (how you "drive" your body), but they rarely put them all into one single "weather report" to tell you how healthy your nervous system is.

This research paper describes the creation of a new tool called the Neurological Health Score (NHS). Think of it as a "Brain Weather Forecast" that tells you how likely you are to face "storms" like dementia, stroke, or Parkinson’s disease.

---

1. The Ingredients: What goes into the recipe?

To make this score, the researchers didn't just look at one thing. They gathered a massive "recipe" of data from hundreds of thousands of people, including:

• The Blueprint (Genetics): Your inherited DNA that might make you more or less sensitive to certain conditions.
• The Fuel (Biochemistry): Things in your blood, like cholesterol, sugar (HbA1C), and vitamins.
• The Maintenance (Lifestyle): How much you sleep, how much you move, what you eat, and even how much stress you feel.
• The Social Climate (Well-being): Whether you feel lonely or if you have friends and family to talk to.

2. The Chef: Using "Smart Machines" (Machine Learning)

Instead of a human trying to crunch all these numbers by hand—which would be like trying to count every raindrop in a hurricane—the researchers used Machine Learning.

They used two types of "digital chefs":

1. Logistic Regression: This is like a traditional recipe book. It’s very clear and tells you exactly how much each ingredient (like "stress" or "diet") affects the final dish. It’s easy to read and understand.
2. XGBoost: This is like a high-tech, experimental kitchen. It’s much faster and better at finding "secret combinations"—for example, how a specific diet might interact with a specific gene in a way a human might never notice.

3. The Results: What did the forecast tell us?

The researchers tested this score on people in the UK and then "double-checked" it using data from people in India to make sure it worked for different ethnicities.

Here is what they found:

• The "High Score" Advantage: People in the top 25% (those with the highest "Brain Weather" scores) had a 1.3 times lower risk of developing neurological issues. They were essentially living in "sunny weather" zones.
• The Red Flags: The study found that certain things act like "dark clouds." High stress, loneliness, and certain blood markers (like high liver enzymes or low Vitamin D) were strong signals that a "storm" might be coming.
• The Silver Lining: Because many of these factors (like diet, exercise, and social connection) are things we can actually change, the score isn't just a warning—it’s a navigation map.

4. Why does this matter to you?

In the past, a doctor might tell you, "You have a genetic risk for dementia." That can feel like being told, "A storm is coming, and there's nothing you can do."

With this new Neurological Health Score, a doctor can say: "Your genetic blueprint shows some clouds, but your 'lifestyle weather' is currently very sunny. If you improve your sleep and reduce your stress, you can actually push those storm clouds away."

In short: This paper moves us from just "predicting the storm" to "learning how to build a better umbrella."

Technical Summary

Technical Summary: Development and Validation of a Neurological Health Score (NHS) Using Machine Learning

1. Problem Statement

Neurological disorders—including dementia, stroke, Parkinson’s disease, ALS, schizophrenia, bipolar disorder, multiple sclerosis, and migraine—are multifactorial, resulting from complex interactions between genetic susceptibility and environmental/lifestyle factors. While Genome-Wide Association Studies (GWAS) have identified many genetic risks, existing clinical tools often focus on a narrow set of modifiable factors (e.g., the 21-point Brain Care Score), frequently omitting critical genetic and biochemical data. There is a clinical need for a comprehensive, quantifiable framework that integrates genetic, lifestyle, and biochemical variables to identify high-risk individuals and recommend personalized preventive measures.

2. Methodology

The researchers developed a Neurological Health Score (NHS) using a multi-stage computational approach:

• Data Sources: The study utilized large-scale data from the UK Biobank. The development cohort consisted of Caucasian individuals, while a separate cohort of Indian ethnicity was used for external validation.
• Variable Selection:
  • Genetic: Polygenic Risk Scores (PRS) were calculated for all eight disorders using PLINK2.0, based on GWAS-identified SNPs (ranging from 6 to 193 SNPs per disorder) filtered for clinical portability (GSA-V3 chip).
  • Lifestyle & Socio-demographic: Variables included BMI, diet, alcohol intake, smoking, sleep, stress, social connections (loneliness/confiding), physical activity, walking pace, employment, and income.
  • Biochemical: Approximately 30 blood-related parameters (e.g., cholesterol, glucose, liver enzymes like ALT/AST, Vitamin D, and Cystatin C) were analyzed.
• Algorithmic Approach:
  • Feature Selection: Logistic regression and XGBoost (a gradient-boosted decision tree) were used to identify significant variables. XGBoost was specifically employed to capture non-linear relationships and complex interactions that standard regression might miss.
  • Model Construction: Variables were categorized and assigned scores (where higher scores indicated lower risk). The NHS was calculated as the sum of these scores.
  • Statistical Modeling: A logistic regression model was built using the NHS, adjusted for age and sex. The model's performance was evaluated using the Area Under the Curve (AUC) from Receiver Operating Characteristic (ROC) curves.

3. Key Contributions

• Multidimensional Integration: This is, to the authors' knowledge, the first study to integrate genetic (PRS), lifestyle, and biochemical data into a single unified score for neurological health.
• Hybrid ML Approach: By combining the interpretability of logistic regression with the predictive power of XGBoost, the study addresses the "black box" limitation of machine learning in clinical settings.
• Cross-Ethnic Validation: The model was not only trained on Caucasian data but also validated on an Indian cohort, testing its potential for broader applicability.

4. Results

• Significant Predictors: For the combined neurological risk model, the most significant lifestyle factors were stress, loneliness, and alcohol intake frequency. Key biochemical predictors included Cystatin C, Vitamin D, ALT, and GGT. Significant physical markers included walking pace and handgrip strength.
• Model Performance: The NHS proved to be a significant predictor of neurological disorder risk even after adjusting for age and sex.
  • AUC: The model achieved an AUC of 0.6 (improving from 0.58 when only age/sex were considered).
  • Odds Ratio (OR): The OR was 0.95, indicating that as the NHS increases, the risk of neurological disorders decreases.
• Risk Stratification: The top 25% of individuals (highest NHS) demonstrated a 1.3 times lower risk of neurological disorders compared to the rest of the population.

5. Significance

The study provides a foundational framework for quantifying neurological health in clinical settings. By providing a single, actionable score, healthcare providers can:

1. Identify high-risk individuals before the onset of clinical symptoms.
2. Personalize interventions by highlighting whether a patient's risk is driven more by genetic predisposition, biochemical imbalances, or modifiable lifestyle factors (e.g., stress or diet).
3. Promote preventive medicine through data-driven lifestyle recommendations.