Social and Cardiovascular Risk Factors as Predictors of the Progression from Mild Cognitive Impairment to Dementia in a Large EHR Database

This study utilizing a large electronic health record database reveals that both cardiovascular factors (specifically higher systolic blood pressure and lower BMI) and social determinants (including non-English language preference, certain racial/ethnic backgrounds, and widowhood) independently predict a faster progression from mild cognitive impairment to dementia, highlighting the potential of integrating clinical and social data for early risk identification.

The Gist

Imagine your brain is a house. For years, it has been standing strong. Then, you notice a few creaky floorboards and a door that sticks a little. This is Mild Cognitive Impairment (MCI). It's not a collapsed house (Dementia) yet, but it's a warning sign that something is changing.

The big question for doctors and families is: Will this house stay standing for decades, or will it crumble into a full-blown collapse soon?

This study acted like a massive detective agency. Instead of looking at tiny, expensive blueprints (like brain scans or genetic tests that only rich hospitals have), the researchers looked at the "everyday maintenance logs" found in regular doctor's offices. They tracked over 6,500 people who had just been diagnosed with those "creaky floorboards" and watched to see who's house started falling apart faster.

Here is what they found, explained simply:

1. The "Heart-Brain" Connection (Blood Pressure)

Think of your blood vessels as the plumbing in your house. If the water pressure is too high, it strains the pipes.

• The Finding: High blood pressure (specifically the top number, systolic) was a major red flag.
• The Twist: It hit women much harder than men. For women, even a little extra pressure in the pipes seemed to speed up the house's decline. For men, the link was weaker.
• The Metaphor: Imagine a woman's house has slightly older, more brittle pipes. Turning up the water pressure a little bit causes leaks much faster than it would in a man's house with sturdier pipes.

2. The "Weighty" Mystery (Body Mass Index)

Usually, we think being overweight is bad for your health. But in this specific "creaky floorboard" phase, the opposite happened.

• The Finding: People with a slightly higher body weight actually had their houses stay standing longer.
• The Twist: This was especially true for men.
• The Metaphor: Think of body weight as a "safety cushion" or a reserve tank of fuel. When the brain starts to get sick, it often burns through fuel rapidly, causing people to lose weight. So, having a bit of extra "cushion" might mean the house has more resources to fight off the storm. It doesn't mean you should try to gain weight, but it explains why being very thin in old age can be a warning sign.

3. The "Loneliness" Factor (Marital Status)

Your social circle is like the neighborhood watch. If you have neighbors looking out for you, problems get caught early.

• The Finding: People who were widowed (lost their spouse) were much more likely to see their house crumble quickly.
• The Twist: Interestingly, people who were single (never married) didn't have this risk.
• The Metaphor: Being widowed is like having your main security guard suddenly leave the post. The sudden loss of daily support and routine is a shock to the system. Being single, however, is like having a different kind of security team (friends, community) that has been in place for years, so the house is still protected.

4. The "Language Barrier" and "Identity"

Imagine trying to fix a leaky roof, but the repair manual is in a language you don't speak, or the repair crew doesn't understand your specific needs.

• The Finding: People who spoke Spanish, Chinese, or other non-English languages, as well as Latinx and Asian patients, saw their houses decline faster than White, English-speaking patients.
• The Metaphor: This isn't about biology; it's about the "repair crew." If you can't easily talk to your doctor, or if the system doesn't speak your language, you might miss early warnings or get the wrong advice. It's like trying to navigate a maze with a map that has missing pages.

5. The "Neighborhood" Surprise

You might think living in a "bad neighborhood" (with less money or resources) would make the house fall faster.

• The Finding: Surprisingly, the specific "zip code" score didn't matter much in this study.
• The Metaphor: By the time the "creaky floorboards" (MCI) are noticed, the damage might already be done by the time you lived in that neighborhood. The immediate factors (like your blood pressure or who is taking care of you) mattered more than the street you live on at that specific moment.

The Big Takeaway

This study is like a new, simpler tool for doctors. Instead of needing a million-dollar MRI machine to guess who is at risk, they can look at the "easy stuff" in your medical file:

• Is your blood pressure high? (Especially if you are a woman).
• Have you lost a lot of weight?
• Did you recently lose your spouse?
• Do you face language barriers?

If the answer is "yes," the doctor knows to be extra vigilant. It's about catching the house before the roof falls in, using the tools we already have in our pockets, rather than waiting for expensive tests that not everyone can get.

Technical Summary

1. Problem Statement

The transition from Mild Cognitive Impairment (MCI) to Alzheimer's Disease and Related Dementias (AD/ADRD) is highly heterogeneous, with only 10–15% of MCI patients progressing annually. While biological and biomarker-based predictors (e.g., amyloid, tau, neuroimaging) are well-studied, they are often inaccessible in primary care settings and fail to capture the full spectrum of risk, particularly for underrepresented populations.

Existing literature suffers from several limitations:

• Selection Bias: Many studies rely on specialized cohorts (e.g., ADNI) that enroll prevalent (existing) rather than incident (newly diagnosed) MCI cases, obscuring the natural course of progression.
• Data Limitations: Prior studies often lack real-world clinical data, focusing on biomarkers unavailable in routine care.
• Underexamined Determinants: The roles of routine clinical measures (blood pressure, BMI) and social determinants of health (SDOH) in the specific progression phase (MCI to dementia) remain unclear, despite their known impact on general dementia incidence.

Objective: To evaluate the association of cardiovascular and social risk factors with the rate of progression from incident MCI to AD/ADRD using a large, diverse Electronic Health Record (EHR) database.

2. Methodology

Study Design and Population:

• Data Source: Retrospective cohort study using de-identified EHR data from UCSF Health (San Francisco Bay Area) spanning 2010–2024.
• Cohort: 6,529 patients aged ≥50 years with an incident diagnosis of MCI.
• Inclusion Criteria: MCI diagnosis recorded in structured EHR fields within dementia specialty clinics, geriatrics, internal medicine, or primary care.
• Outcome: Time to incident AD/ADRD diagnosis, death, or last clinical encounter (censoring).

Exposure Variables:

• Cardiovascular/Metabolic:
  • Median Systolic Blood Pressure (SBP) and Diastolic Blood Pressure (DBP).
  • Median Body Mass Index (BMI).
  • History of Type II Diabetes (binary).
  • Measurement: Calculated from all encounters in the 5 years prior to or including the MCI diagnosis date.
• Social Determinants:
  • Neighborhood Disadvantage: California Healthy Places Index (HPI) linked to the census tract of the patient's longest residence.
  • Marital Status: Categorized (Married/Partnered, Single, Divorced/Separated, Widowed, Unknown).
  • Language Preference: English, Spanish, Chinese, Other/Non-English.
  • Race/Ethnicity: White, Latinx, Black/African American, Asian, Other.

Statistical Analysis:

• Primary Model: Cox Proportional Hazards models to estimate Hazard Ratios (HR) for progression.
• Adjustments:
  • Social Models: Adjusted for age, sex, and clinic category.
  • Cardiovascular Models: Adjusted for age, sex, race/ethnicity, social factors, and clinic category.
• Non-linearity & Interactions:
  • Restricted cubic splines were used to test for non-linear relationships (SBP, DBP, BMI).
  • Interaction terms were tested for Sex. Significant interactions ($p<0.05$) led to sex-stratified models.
• Model Selection: Akaike Information Criterion (AIC) and Likelihood Ratio Tests (LRT) compared linear vs. spline models.

3. Key Results

Demographics:

• 21.5% (1,403/6,529) progressed to AD/ADRD.
• Progressors were older (mean 75.2 vs. 69.7), had higher SBP, lower DBP, lower BMI, and were more likely to be widowed, non-English speaking, and Asian/Latinx compared to censored patients.

Cardiovascular Risk Factors:

• Systolic Blood Pressure (SBP): Higher SBP was significantly associated with increased progression risk (HR per 10 mmHg: 1.09).
  • Sex Interaction: The association was significantly stronger in females (HR 1.13) than males (HR 1.05, non-significant).
  • Non-linearity: Spline models revealed a threshold effect; risk increased notably above ~120 mmHg, plateaued at 130–140 mmHg, and rose again in Stage 2 hypertension (>140 mmHg).
• Body Mass Index (BMI): Higher BMI was protective (inversely associated) with progression (HR per 1 unit: 0.97).
  • Sex Interaction: The protective effect was stronger in males (HR 0.94) than females (HR 0.98).
  • Linearity: The relationship was linear; spline models did not improve fit.
• Diastolic Blood Pressure (DBP): Showed a non-linear association (risk increased steeply >85 mmHg) but was less consistent and not statistically significant in linear models.
• Diabetes: History of Type II diabetes showed a non-significant trend toward faster progression (HR 1.08, $p=0.20$).

Social Risk Factors:

• Marital Status: Widowed status was associated with higher progression risk (HR 1.15) compared to married individuals. Conversely, being Single was associated with lower risk (HR 0.79).
• Language: Non-English speakers faced significantly higher risks:
  • Spanish-speaking (HR 1.38).
  • Chinese-speaking (HR 1.19).
  • Other Non-English (HR 1.24).
• Race/Ethnicity: Latinx (HR 1.22) and Asian (HR 1.14) patients had higher progression risks compared to White patients.
• Neighborhood Disadvantage: The California HPI score was not significantly associated with progression (HR 0.99), likely due to the relatively high socioeconomic status of the cohort.

4. Key Contributions

1. Real-World Generalizability: Unlike studies relying on specialized biomarker cohorts (e.g., ADNI), this study utilizes routine primary care EHR data, making findings directly applicable to standard clinical settings.
2. Incident Cohort Focus: By starting follow-up at the first MCI diagnosis, the study avoids the survival and selection biases inherent in studies of prevalent MCI cases.
3. Integration of Social and Clinical Data: The study successfully links geospatial SDOH data with longitudinal clinical metrics to predict progression, demonstrating a scalable framework for risk stratification.
4. Sex-Specific Nuance: The identification of distinct sex-specific patterns (SBP risk in women, BMI protection in men) challenges "one-size-fits-all" risk models.
5. Non-Linear Modeling: The use of spline models revealed that cardiovascular risk is not linear, identifying specific thresholds (e.g., SBP >120 mmHg) where risk accelerates, particularly in women.

5. Significance and Implications

• Clinical Practice: The findings suggest that routine vitals (BP, BMI) and social history (marital status, language) are powerful, accessible predictors of dementia progression. Clinicians can use these low-cost indicators to identify high-risk MCI patients for earlier intervention and monitoring without relying on expensive biomarkers.
• Equity: The study highlights significant disparities in progression rates for racial/ethnic minorities and non-English speakers. This suggests that structural barriers (access to care, communication gaps, chronic stress) may accelerate disease progression independent of biological factors.
• Sex-Specific Guidelines: The results imply that blood pressure management thresholds for dementia prevention might need to be lower or more aggressive for women with MCI, while weight loss in older men with MCI might warrant closer monitoring for frailty or underlying pathology.
• Future Research: The lack of association with neighborhood disadvantage (HPI) suggests that individual-level social factors (like widowhood) or specific structural barriers (language) may be more immediate drivers of progression than broad census-tract metrics in affluent urban settings.

Limitations: Observational design (causality cannot be inferred), reliance on ICD codes for diagnosis (potential misclassification), and lack of data on medication adherence or specific treatment of risk factors.