Leader: Massimo Montalto (UNICATT); Other collaborator(s): RICCARDO CALVANI (UCSC)
Based on the available databases from longitudinal cohorts (identified in task 1.1), multivariate/multivariable models will be built to unveil the bidirectional relationships between clinical, functional and biological parameters. Statistical approaches will be based on variable selection methods coupled with classification algorithms
Brief description of the activities and of the intermediate results: The identification of variables of interest to be used as input variables in the predictive models has been completed from pilot analyses. Multimorbidity, defined as the presence of two or more diseases in a same individual has been identified as the condition ‘at risk’ of adverse health outcomes. Outcomes include functional and cognitive decline, disability, burden of multimorbidity, other health adverse outcomes such as hospitalization and death. Frailty, polypharmacy, clinical and biological parameters related to muscle mass, hormone and metabolic parameters, inflammatory markers, nutritional and environmental factors have been selected to be explored in predictive models. Cohorts available include Ospedale San Raffaele (Milano) and UCSC (Policlinico A. Gemelli- Roma) memory clinic datasets, CHROnOS, SPRINT-T, FRASNET, IlSirente, CASSIOPEA.
Main policy, industrial and scientific implications: The integration of multiple predictors using a multidimensional approach is expected to produce tools that may adequately capture the complexity of older adults in clinical practice.
A secondary analysis from the SPRINTT dataset has been completed to identify predictors of incident mobility disability in older adults with physical frailty and sarcopenia. Classification models based on partial least squares discriminant analysis (PLS-DA) allowed correct classification of participants with approximately 70% accuracy.
Findings from the analysis on the identification of predictors of incident disability have been shared with Spoke 3 researchers and reported during the July internal Spoke meeting. Manuscript is under preparation.
Analysis of plasmatic biomarkers of cognitive disability has been planned and preliminary associations have been identified using a cohort of 100 individuals with mild cognitive impairment. Additional cohorts, such as the SPRINTT cohort, will serve to further develop the identification of predictors of incident cognitive disability and to validate the observed associations.
In the past quarter, longitudinal analyses were conducted to identify predictors of incident cognitive disability. Specifically, the predictive power of circulating biomarkers on the risk of dementia was studied in a population with MCI. The analysis on the effectiveness of the circulating neutrophil-to-lymphocyte ratio in predicting the risk of conversion to dementia has been completed. Several classification models have been constructed using multiple chemometric and machine learning strategies (i.e., PLS-DA, Random Forest, K-Nearest Neighbours) to identify predictors of incident motor disability in subjects enrolled in the SPRINTT study. All models were subjected to stringent double repeated cross-validation and permutation testing, which are regarded as the most effective means of ensuring robust model validation and mitigating the risks of overfitting and spurious feature selection. In general, the best performing models in classification were those built using only continuous variables, regardless of the analytical method used.
Analytical files have been prepared for the development of predictive models of cognitive health outcomes using the SPRINT-T dataset. In particular, preliminary analyses have been conducted on the available cognitive measures, namely the Mini-Mental State Examination and the Trail Making Test (TMT) - A and B. Furthermore, through a series of meetings and preliminary analytical evaluations, the possibility of using other datasets available within the consortium for the study of cognitive outcomes has been explored. Predictive models of cognitive status variation are currently under development. Regarding the completed analysis on the role of the neutrophil to lymphocyte ratio as a predictor of conversion to dementia, the drafting phase of the manuscript has begun and is currently underway.
The analysis of the predictive role of the NLR ratio has been further refined based on feedback received during an internal review process. The manuscript has been finalized and it has at the submission stage in the journal Alzheimer’s & Dementia. In addition, based on the results of the analysis on predictors of motor disability conducted in the past months, an individual risk prediction algorithm has been developed and presented during the most recent WP4 general meeting. The investigation of key predictors of cognitive decline, based on a reanalysis of data from the SPRINT-T cohort, is currently under revision.
In the third quarter of 2025 our work aimed at exploring predictors of cognitive decline in older adults from the SPRINTT trial, focusing on early markers of vulnerability in cognitively intact participants.
This secondary analysis started from 759 participants, excluding those with baseline MMSE ≤26, yielding a final sample of 475 older adults.
The study aimed to identify predictors of clinically significant cognitive decline, defined as ≥2 MMSE points lost at 12 months or ≥3 points at 24 months.
At 24 months, 154 participants declined, while 321 maintained stable cognition.
Machine learning identified baseline MMSE, years of education, depressive symptoms, and gait speed <0.8 m/s as main predictors.
Logistic regression (χ²=24.97, p<0.001, AUC=0.65) showed older age and slow gait increased risk, whereas higher education was protective.
Kaplan–Meier curves confirmed faster decline among slow walkers (24-month survival: 65% vs 76%).
Cox models showed consistent effects: age HR 1.03, slow gait HR ≈1.47, education HR 0.93 (all p<0.05), while comorbidity burden (CIRS) was not significant.
Overall model discrimination was moderate but stable across methods.
Mobility and cognitive reserve emerged as key predictors of cognitive decline.
A parallel analysis is ongoing to explore similar predictors of functional outcomes, aiming to integrate cognitive and mobility trajectories in frailty research.
During the reference quarter, research activities focused on completing and disseminating the analyses aimed at identifying predictors of incident mobility disability in older adults with physical frailty and sarcopenia (PF&S) enrolled in the SPRINTT trial.
A secondary analysis was conducted on participants allocated to the control arm of the SPRINTT study. The analytical sample included community-dwelling older adults with preserved mobility at baseline. Incident mobility disability was defined as the inability to complete the 400-meter walk test within 15 minutes. A comprehensive set of 88 baseline variables, covering demographic, clinical, functional, social, and laboratory domains, was evaluated as candidate predictors. Multivariate classification and machine learning approaches were applied, including Partial Least-Squares Discriminant Analysis (PLS-DA), Random Forest, and Gradient Boosting, and model performance was subsequently assessed using Cox proportional hazards regression models. Among 759 participants (median age 78 years; 71.3% women), 354 individuals (47%) developed mobility disability over a median follow-up of 26 months. Predictive performance was comparable across modeling approaches, with the best accuracy achieved by the PLS-DA model (67.6%). Across all methods, baseline 400-meter walk time emerged as the strongest and most consistent predictor of incident mobility disability. In multivariable Cox regression analyses, longer baseline walking time, older age, lower functional performance (SPPB), and higher levels of depressive symptoms (CES-D) and sarcopenia severity (SARC-F) were independently associated with an increased risk of disability. A significant interaction was observed between gait speed and depressive symptoms (p < 0.001), indicating a synergistic effect: participants characterized by both slow gait (≥480 seconds) and clinically relevant depressive symptoms (CES-D ≥9) showed the highest risk of developing mobility disability (HR = 2.69; 95% CI: 1.89–3.84). Kaplan–Meier analyses further confirmed the additive effect of functional impairment and depression on disability-free survival. Overall, this task provided robust evidence that slower gait speed and depressive symptoms are key and synergistic predictors of incident mobility disability in older adults with PF&S. The identification of simple and clinically applicable thresholds (400 m walk time ≥480 seconds and CES-D ≥9) supports their potential use for early risk stratification and targeted preventive strategies within the PNRR Age-It framework. The results of this work were presented at the Annual Scientific Meeting of the Gerontological Society of America (GSA), held in Boston in November, contributing to the international dissemination of the project findings.
The manuscript reporting evidence on the longitudinal association between the neutrophil-to-lymphocyte ratio has been completed. Title: Neutrophil to lymphocyte ratio as predictive biomarker of dementia: findings from a prospective cohort study in individuals with mild cognitive impairment. Submitted to Alzheimer’s & Dementia. The manuscript describing the entire model building process, the different classification performance of the models developed, together with the identification of predictors of incident motor disability, is currently being drafted and will be submitted in a high impact journal in the coming months.
Presentation at EUGMS meeting, Valencia 2024 : “Predictors of incident mobility disability in older adults with physical frailty and sarcopenia: a secondary analysis from the SPRINTT clinical trial”