BACKGROUND

Systemic inflammation is linked to Alzheimer's disease and related dementias (AD/ADRD)

Markers like neutrophil- and monocyte-to-lymphocyte ratios (NLR, MLR) are associated with AD/ADRD risk¹

In the emergency department (ED), these markers are often transiently elevated due to acute illness, biasing their predictive value

HYPOTHESIS

Incorporating pre-ED trajectories of NLR and MLR will improve AD/ADRD risk prediction

METHODS

Outcomes were assessed starting 3 months post-ED visit over 3–7-year windows.

ED-SIM model using NLR and MLR measurements obtained up to 2 years before the ED visit.

For each patient, μ₀ (mean NLR/MLR prior to ED), μ₁ (NLR/MLR at ED visit), and θ (time-trend in NLR/MLR up to baseline) were derived

Weighted pre-ED trends (θ) and mean systemic inflammation marker (SIM) levels, before and during ED, (μ) were combined via outer product and input into a logistic regression with age, sex, and race/ethnicity covariates to predict AD/ADRD incidence.

SAMPLE

Patients in Veteran's Health Administration (VA) with emergency department (ED) utilization from 2011 to 2020, age 60-85, no prior AD/ADRD history, and at least one NLR/MLR.

Split into 80/20% training/validation sets

RESULTS

Cohort: 52,816 patients; 3,017 (5.6%) developed AD/ADRD. Median NLR 3.48 [3.25–5.75] and MLR 0.46 [0.32–0.68] at initial ED visit, with slightly lower pre-ED values across 8 [6–13] samples.

Main Result: In the 20% validation set, the 5-year prediction model incorporating ≥8 pre-ED SIM values achieved an AUC of 0.725, outperforming ED-only SIMs (AUC 0.708) and demographics alone (AUC 0.684). Similar relative gains were observed in sub-cohorts with fewer SIM measurements.

Model Comparison (Figure 2): ROC curves stratified by prediction window (3, 5, 7 yrs) and minimum lab samples (3, 5, 7 pre-ED). The full ED-SIM trajectory model (Pre-ED + ED SIM + Demographics) outperformed the ED SIM + Demographics model and the demographics-only model. All models adjusted for age (linear/quadratic), sex, and race/ethnicity (Figure 2).

CONCLUSIONS AND FUTURE WORK

Incorporating markers of systemic inflammation (NLR, MLR) measured during and prior to an initial ED visit substantially improves AD/ADRD risk prediction. While SIMs from a single ED visit demonstrate strong predictive performance, inclusion of prior values provides additional improvement, substantiating the role of systemic inflammation in AD/ADRD.

Table 1 – Cohort Characteristics at Baseline

Figure 1 – Schematic diagram of predictive model

Means (μ) and trend (θ) of NLR, MLR up to baseline predicts AD/ADRD (Y):

Baseline: First Emergency Department (ED) visit

Outcome (Y): AD/ADRD

log Pr(Y) ∝ (μ, θ)(NLR) ⊗ (μ, θ)(MLR)

Variables shown: μ₀(NLR), μ₀(MLR), θ(NLR), θ(MLR), μ₁(NLR), μ₁(MLR)

Timeline: Up to 2 years prior to ED → 90-day exposure gap before ED → 90-day outcome gap after ED → Prediction window

Funding: This study was supported by the VA Cooperative Studies Program

Figure 2 – ROC curves comparison of ADRD prediction models from ED

Length of prediction window (8 or more lab samples prior to initial ED visit):

 Minimum lab samples prior to initial ED visit (5-year predictions):

REFERENCES