Decomposition of the systemic immune-inflammation index reveals neutrophil- and lymphocyte-driven, non-linear risk of progression from mild cognitive impairment to dementia

The progression from mild cognitive impairment (MCI) to dementia represents a critical intervention window. The systemic immune-inflammation index (SII) integrates neutrophil, platelet, and lymphocyte counts and has been proposed as an accessible marker of this progression. Because SII is a mathematical ratio of its three components, we additionally examined whether its predictive value is attributable to specific constituent cell populations. This study investigated whether SII, and its decomposed components, predict MCI-to-dementia conversion and how each relates to established Alzheimer’s disease (AD) biomarkers. We analyzed 517 MCI patients from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database with complete baseline blood counts, cerebrospinal fluid (CSF) biomarkers, and longitudinal follow-up. Participants were stratified by median SII into high and low groups. Correlations between SII (and its components) and clinical variables were assessed using Spearman’s correlation. Multivariable Cox proportional hazards regression, with covariates pre-specified on biological plausibility, identified independent predictors of dementia conversion, both for SII as a composite variable (Model B) and for its log2-transformed neutrophil, platelet, and lymphocyte components entered separately (Model C). Restricted cubic spline (RCS) analysis examined non-linear dose-response relationships. During a mean follow-up of 31 months, 190 of 517 patients (37%) progressed to dementia. High SII was independently associated with increased conversion risk (adjusted HR 1.5, 95% CI: 1.1-2.0, P = 0.011). When SII was decomposed into its three components, log2-transformed neutrophil count (HR 1.7, 95% CI: 1.2–2.3, P = 0.002) and log2-transformed lymphocyte count (HR 0.7, 95% CI: 0.5-1.0, P = 0.035) were independently associated with dementia conversion in opposite directions, whereas log2-transformed platelet count was not (HR 1.1, 95% CI: 0.7–1.8, P = 0.572). The decomposed model (Model C) showed improved discrimination over the SII-based model (Model B) (Harrell’s C-index 0.76 vs. 0.75) and provided statistically significant incremental information (likelihood ratio test, χ²=14.36, df = 3, P = 0.003). SII, and each of its components, correlated only weakly with CSF AD pathology markers.RCS analysis showed a non-linear association between SII and conversion risk (non-linearity P = 0.006), driven specifically by a non-linear effect of log2-neutrophil count (non-linearity P = 0.008). SII is independently, but modestly, associated with progression from MCI to dementia, and this association is driven predominantly by neutrophil and lymphocyte counts rather than platelets. Given the weak correlation between SII and core CSF markers of AD pathology, elevated SII more plausibly reflects a non-specific state of systemic inflammatory or frailty-related vulnerability than a biomarker specific to AD pathophysiology. Decomposing SII into its constituent cell counts offers a more biologically interpretable and statistically appropriate alternative to the composite ratio for risk stratification; validation in independent cohorts is warranted before any clinical application.