Truace
Subscribe

identifying current MCI status among older adults with chronic conditions using prespecified XGBoost model

Source article: Explainable machine learning for identifying mild cognitive impairment in older adults with chronic diseases: Model development and temporal validation

Objective This study aimed to develop and validate an interpretable machine-learning classification model for identifying MCI at the time of assessment among older adults with chronic conditions. We also evaluated model performance and the stability of TreeSHAP-based feature explanations across survey waves. Methods Data were obtained from the CHARLS. A total of 8222 participants from the 2018 wave were stratified by MCI status and randomly divided into a development set (n = 5754) and an internal validation set…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-1283· cite & verify
Explainable machine learning for identifying mild cognitive impairment in older adults with chronic diseases: Model development and temporal validation
"Healthy-Aging-in-Colorado-Infographic-Health-of-Older-Adult-Population" by cdphe.ched is licensed under CC BY 2.0. To view a copy of this license, visit https://creativecommons.org/licenses/by/2.0/.
Trace impact readingContested
P 69The P score combines the specificity and measured human impact of the grounded problem claim with the strength of this Trace’s cited sources.

Both sides are scored from claims and sources, not community votes.

G 69The G score combines the specificity and measured human impact of the grounded gain claim with the strength of this Trace’s cited sources.

In brief

Using CHARLS data, researchers built an interpretable machine-learning framework to identify MCI at time of assessment in older adults with chronic conditions. After LASSO selection of 27 predictors, a prespecified XGBoost model achieved AUROC 0.692 internally in 2018 and 0.702 in 2020 temporal validation, with TreeSHAP highlighting number of living children, sleep duration, education, ADL and age.

The work matters because it demonstrates relatively stable discrimination and explanation stability across survey waves for a low-cost screening aid, but clinical value remains constrained by low sensitivity, limited calibration and lack of external validation. Authors position it only as adjunctive risk-stratification before formal cognitive assessment, not as a diagnostic or prognostic tool.

Main points

  1. Study used CHARLS data: 8222 participants from 2018 wave split into development (n=5754) and internal validation (n=2468), plus 5558 from 2020 wave for temporal validation.
  2. LASSO selected 27 predictors; seven models tested with encoding, standardization, SMOTENC and tuning inside 10-fold CV to minimize leakage.
  3. Internal validation AUROC 0.692, AUPRC 0.282, sensitivity 0.511, specificity 0.746, Brier 0.109; temporal AUROC 0.702, AUPRC 0.305, sensitivity 0.390, specificity 0.843, Brier 0.137.
  4. TreeSHAP identified number of living children, sleep duration, educational attainment, activities of daily living (ADL), and age as strongest contributors.

The gain

XGBoost model achieved moderate and relatively stable discrimination for identifying current MCI status among older adults with chronic conditions across 2018 internal and 2020 temporal validation, with highly stable TreeSHAP explanations.

The problem

Model had limited sensitivity and calibration, with sensitivity dropping to 0.390 in temporal validation and Brier score worsening to 0.137, limiting clinical utility.

The rundown

Researchers developed seven classifiers using 27 LASSO-selected predictors from CHARLS 2018, applying encoding, standardization, SMOTENC resampling and hyperparameter tuning within training folds of 10-fold cross-validation while retaining original class distributions in validation sets.

Performance was evaluated with AUROC, AUPRC, sensitivity, specificity, Brier score, calibration analysis and decision curve analysis, which showed positive net benefit mainly across low-to-moderate threshold probabilities.

TreeSHAP was applied to the prespecified XGBoost model to characterize contributions and stability, finding consistent global importance rankings across waves despite modest discrimination.

What this doesn’t fix

Model showed limited sensitivity and calibration and requires further recalibration and external validation before clinical implementation, and should not replace formal diagnosis or predict future decline.

Sources

  1. Peer-reviewedActa Psychologica2026-10-03

The debate