Sex-specific differences in nonlinear associations between glycaemia and brain health in UK Biobank

In a large UK Biobank study, glycaemia exhibits non-linear, inverted J-shaped associations with brain structure across the entire glucose spectrum, with higher glucose levels linked to reduced brain volumes and steeper declines observed in females compared to males.

The Gist

The Big Picture: Sugar, Sex, and the Brain

Imagine your brain is a high-end, complex city. It needs a steady supply of fuel (sugar/glucose) to keep the lights on and the traffic moving. But just like a city, it has a "Goldilocks Zone"—a perfect amount of fuel where everything runs smoothly. Too little fuel, and the city freezes; too much, and the streets get clogged, the pipes rust, and the buildings start to crumble.

This study looked at 36,000 people in the UK to see how their blood sugar levels (both the "instant" sugar and the "long-term average" sugar) affected the size and health of their brain's "buildings" (grey matter, white matter, and total volume).

The biggest surprise? The relationship isn't a straight line. It's more like a rollercoaster, and the ride feels different for men and women.

---

1. The "Inverted J" Rollercoaster

Most people think of health risks as a straight line: More sugar = More damage.
But this study found that the relationship is shaped like an inverted "J" (or a hill that drops off a cliff).

• The Flat Top (The Sweet Spot): When blood sugar is in a healthy, normal range (roughly between 4.4 and 4.7 mmol/L for glucose), the brain volume is at its peak. The city is thriving.
• The Cliff (High Sugar): As soon as blood sugar climbs above that sweet spot, the brain volume starts to drop. The higher the sugar goes, the steeper the drop.
• The Dip (Very Low Sugar): There was also a hint that very low sugar might be slightly less ideal than the perfect middle ground, though the danger of high sugar is much more obvious.

The Analogy: Think of your brain like a balloon.

• Perfect Inflation: The balloon is round, strong, and full (Healthy brain).
• Over-inflation: If you keep blowing air in (high sugar), the rubber stretches too thin and eventually starts to shrink or pop (brain atrophy).
• Under-inflation: If you let too much air out (very low sugar), it's also not quite right, but the damage from over-inflation is much more dramatic.

2. The Gender Gap: Who Gets Hit Harder?

This is where the study gets really interesting. The researchers looked at men and women separately, and they found a distinct difference in how their brains react to high sugar.

• The Men's Curve: When men's blood sugar gets high, their brain volume does start to shrink, but it's a gentle, gradual slope. It's like a slow slide down a playground slide.
• The Women's Curve: When women's blood sugar gets high, the drop is much steeper. It's like stepping off a diving board. For the same amount of extra sugar, women's brains appear to lose volume faster than men's.

The Analogy: Imagine two cars driving off a cliff.

• The Male Car: It has a parachute that deploys slowly. It still falls, but it descends gently.
• The Female Car: The parachute deploys late or not at all. It plummets much faster.
• The Lesson: Women might be more sensitive to the "toxic" effects of high blood sugar on brain structure than men are.

3. Two Different Types of Sugar

The study looked at two ways of measuring sugar:

1. Random Glucose: A snapshot of your sugar right now (like a photo).
2. HbA1c: Your average sugar over the last 3 months (like a video recording).

Both showed the same "Inverted J" pattern. However, the "video" (HbA1c) showed a very clear pattern for brain health, while the "photo" (glucose) showed the strongest difference between men and women. This suggests that both your current sugar spike and your long-term sugar habits matter for your brain.

4. Why Does This Matter? (The "Aha!" Moment)

For a long time, doctors have treated sugar like a light switch: Off (Normal) or On (Diabetes). If you aren't diabetic, you're usually considered "safe."

This study says: No, it's a dimmer switch, not a light switch.

• The Old Way: "You don't have diabetes, so your brain is fine."
• The New Way: "Even if you don't have diabetes, if your sugar is on the higher end of 'normal,' your brain might already be starting to shrink, especially if you are a woman."

Summary in Plain English

1. Sugar isn't just bad for diabetics: Even people without diabetes can have brain shrinkage if their blood sugar is consistently on the high side of "normal."
2. It's not a straight line: The damage happens in a specific curve. There is a perfect middle ground, and going above it is bad news.
3. Women are more vulnerable: When blood sugar gets high, women's brains seem to suffer structural damage (shrinking) more quickly and severely than men's brains.
4. The Takeaway: We shouldn't just wait until someone gets a diabetes diagnosis to worry about their brain health. Keeping blood sugar in that "Goldilocks Zone" is crucial for everyone, but especially for women, to keep their brains big, healthy, and functioning well as they age.

The Bottom Line: Think of your blood sugar as the thermostat for your brain. You don't need to freeze it or burn it down; you just need to keep it in that perfect, comfortable middle range to keep the city of your brain running smoothly. And if you're a woman, you might need to be extra careful to keep that thermostat from creeping up too high.

Technical Summary

1. Problem Statement

Diabetes and hyperglycaemia are established risk factors for adverse brain outcomes, including reduced brain volume, grey matter atrophy, and increased white matter hyperintensity (WMH) burden. However, existing neuroimaging literature suffers from three primary limitations:

1. Categorical Focus: Most studies rely on diagnosed diabetes status rather than treating glycaemia as a continuous variable, potentially missing associations within the non-diabetic range.
2. Linearity Assumption: Studies typically assume linear relationships between glycaemic markers and brain health, ignoring potential non-monotonic (e.g., U-shaped or inverted J-shaped) associations.
3. Lack of Sex Stratification: Analyses are often pooled, despite evidence suggesting females with diabetes face higher risks of adverse outcomes than males, even with lower incidence rates.

The study aims to investigate whether glycaemic markers (random glucose and HbA1c) exhibit non-linear associations with brain structural biomarkers across the full glycaemic spectrum and whether these associations differ by sex.

2. Methodology

Study Population:

• Data Source: UK Biobank.
• Sample Size: 36,321 participants (17,105 males, 19,216 females) aged 40–69 at baseline.
• Inclusion Criteria: Availability of brain MRI, at least one glycaemic measure, no baseline dementia, and self-reported White European ethnicity.
• Exclusion: Participants were not excluded based on diabetes status initially, but sensitivity analyses excluded those with diagnosed diabetes or on hypoglycaemic medication.

Variables:

• Exposures:
  • Random Glucose: Measured via serum hexokinase analysis (acute fluctuation).
  • HbA1c: Measured via whole blood HPLC (chronic average exposure).
• Outcomes (MRI-derived):
  • Whole Brain Volume (WBV), Grey Matter (GM), White Matter (WM).
  • Hippocampal Volume (HV).
  • White Matter Hyperintensity Volume (WMHV).
  • Note: All volumes were adjusted for Total Intracranial Volume (TIV); WMHV was log-transformed.
• Covariates: Age, education, socioeconomic deprivation (Townsend index), BMI, and smoking status.

Statistical Analysis:

• Primary Approach: Fractional Polynomial (FP) regression to flexibly model non-linear associations (up to two powers selected from $\{-2, -0.5, 0, 0.5, 1, 2, 3\}$).
• Stratification: Analyses were conducted sex-stratified (males vs. females) to detect effect modification.
• Robustness Checks:
  • Restricted Cubic Splines (RCS): Used with five knots (5th, 27.5th, 50th, 72.5th, 95th percentiles) to confirm non-linearity ($p_{non-linearity}$) and test for sex interactions ($p_{interaction}$).
  • Range-Restricted Linear Models: Fitted within ascending and descending portions of the curves to estimate sex-specific slopes.
  • Peak Derivative Metrics: Identified turning points (peaks) and maximum gradient points to characterize exposure-response profiles.
• Software: Stata 18.0.

3. Key Contributions

• Continuous Exposure Modeling: Demonstrates that the relationship between glycaemia and brain structure is not limited to diabetic thresholds but operates across the entire population distribution.
• Non-Linearity Characterization: Identifies specific inverted J-shaped (for brain volume) and J-shaped (for WMH) associations, challenging the assumption of simple linear dose-response relationships.
• Sex-Specific Nuance: Provides granular evidence that the shape and slope of these associations differ by sex, particularly for random glucose, where females exhibit steeper declines in brain volume at higher glycaemic levels compared to males.
• Methodological Rigor: Utilizes a multi-model approach (FP, RCS, linear segments) to validate findings and quantify specific turning points in the glycaemic spectrum.

4. Key Results

Non-Linear Associations:

• Brain Volume (WBV, GM, WM): Both glucose and HbA1c showed inverted J-shaped associations with whole brain and grey matter volumes.
  • Peak Values: Predicted brain volumes peaked at glucose levels of 4.4–4.7 mmol/L and HbA1c levels of 33–37 mmol/mol.
  • Decline: Beyond these peaks, higher glycaemia was associated with significantly lower brain volumes.
  • Low End: Evidence for lower volumes at the very low end of the glycaemic spectrum was less consistent.
• White Matter Hyperintensities (WMHV): Showed a J-shaped association, particularly with HbA1c, indicating increased WMH burden at both very low and very high levels, though the rise at the high end was more pronounced.

Sex Differences:

• Glucose: Significant sex interactions were found for Whole Brain Volume ($p=0.02$) and Grey Matter ($p=0.03$).
  • Slope: In the descending range (high glucose >10 mmol/L), the decline in WBV and GM was steeper in females than males.
  • Example: For WBV, $\beta_{females} = -7.22$ cm³/mmol/L vs. $\beta_{males} = -1.7$ cm³/mmol/L.
• HbA1c: Non-linearity was confirmed for all volume outcomes, but sex interactions were generally non-significant ($p \geq 0.4$) for most outcomes, suggesting the curve shape is similar between sexes for chronic glycaemia, though peak values varied slightly.
• Hippocampal Volume: No consistent non-linearity was found for glucose or HbA1c, though a strong sex interaction was observed for HbA1c ($p < 0.001$).

Sensitivity Analyses:

• Excluding participants with diagnosed diabetes or those on glucose-lowering medication did not materially alter the non-linear patterns, confirming that these associations exist across the non-diabetic spectrum.

5. Significance and Implications

Clinical and Public Health:

• Beyond Diagnostic Thresholds: The findings suggest that "normal" or "high-normal" glycaemic levels (below the diabetes diagnostic threshold) are already associated with suboptimal brain structural profiles. This implies that risk operates on a continuum rather than emerging only at the point of clinical diagnosis.
• Sex-Specific Risk: The steeper decline in brain volume for females at higher glucose levels suggests that women may be more vulnerable to the neurotoxic effects of acute hyperglycaemia. This supports the need for sex-stratified risk assessment and potentially tighter glycaemic control targets for women.
• Continuous Monitoring: Clinicians should consider glycaemia as a continuous exposure when evaluating neurological risk, rather than relying solely on categorical definitions (e.g., normal vs. diabetic).

Research Implications:

• Future Studies: Longitudinal studies are required to establish causality and determine if these structural changes translate to cognitive decline or dementia.
• Mechanistic Understanding: The non-linear nature suggests complex underlying mechanisms (e.g., oxidative stress, endothelial dysfunction, insulin resistance) that may vary in dominance across different glycaemic ranges.

Limitations:

• Cross-Sectional Design: Causality cannot be inferred; reverse causation (e.g., brain atrophy affecting glucose metabolism) is possible, particularly at the lower end of the glycaemic spectrum.
• Generalizability: The cohort is predominantly White European, limiting applicability to diverse ethnic groups with different glycaemic physiologies.
• Healthy Volunteer Bias: UK Biobank participants are generally healthier than the general population, which may attenuate the magnitude of observed associations.

In conclusion, this study provides robust evidence that the relationship between glycaemia and brain structure is non-linear and sex-specific, with females showing heightened vulnerability to high glucose levels. These findings advocate for a paradigm shift in how glycaemic risk is conceptualized in brain health research and clinical practice.