{"slug": "machine-learning-assessment-of-the-predictive-value-of-inflammatory-biomarkers", "title": "Machine-Learning Assessment of the Predictive Value of Inflammatory Biomarkers for Cognitive Impairment in an Older Hispanic Adult Cohort", "summary": "A leakage-safe Bernoulli/Categorical Naive Bayes classifier identified I-309 (CCL1) as the dominant incremental predictor of cognitive impairment in the Panama Aging Research Initiative--Health Disparities (PARI-HD) cohort of 165 older Hispanic adults, raising ROC-AUC by 0.110 over a demographic baseline of 0.630 +/- 0.017, according to an arXiv preprint (2609.19374v1). In the pre-specified primary analysis, I-309 yielded a fixed-partition DeLong p=0.0018, with a median p-value of 0.0011 across 200 random partitions, and a Benjamini-Hochberg-adjusted q=0.032 on the frozen partition within an exploratory family of 18 candidate markers. No other marker showed reliable incremental predictive value, and the authors present I-309/CCL1 as an interpretable candidate feature pending external validation.", "body_md": "arXiv:2609.19374v1 Announce Type: new \nAbstract: Small clinical tabular datasets require interpretable machine learning because deep learning is often impractical and ensemble models can be difficult to inspect. A key pitfall is that statistical significance does not necessarily imply predictive utility. Using data from the Panama Aging Research Initiative--Health Disparities (PARI-HD) cohort (n=165), we implemented a leakage-safe threshold-likelihood Bernoulli/Categorical Naive Bayes (BNB/CNB) classifier. Within every training fold, each continuous predictor was reduced to a supervised chi-square-derived state, while income entered the model through a categorical likelihood. All data-dependent steps were performed within repeated stratified 10-fold cross-validation with 30 repeats. The demographic baseline achieved a ROC-AUC of 0.630 +/- 0.017. I-309 (CCL1) was the dominant incremental feature, increasing AUC by 0.110, with paired DeLong tests yielding p<0.05 in 100% of repeats. In the pre-specified primary analysis, I-309 produced a fixed-partition DeLong p=0.0018, with robustness assessed across 200 random partitions, where the median p-value was 0.0011. Within the exploratory family of 18 candidate markers, I-309 achieved a Benjamini-Hochberg-adjusted q=0.032 on the frozen partition and satisfied q<0.05 in 85% of random partitions, whereas no other marker demonstrated reliable incremental predictive value. Because the fitted model is an inspectable table of thresholds and class-conditional probabilities, these results identify I-309/CCL1 as an interpretable candidate feature for tabular prediction of cognitive impairment, pending external validation.", "url": "https://wpnews.pro/news/machine-learning-assessment-of-the-predictive-value-of-inflammatory-biomarkers", "canonical_source": "https://arxiv.org/abs/2609.19374", "published_at": "2026-09-18 04:00:00+00:00", "updated_at": "2026-09-18 04:23:37.638231+00:00", "lang": "en", "topics": ["machine-learning", "ai-research"], "entities": ["Panama Aging Research Initiative--Health Disparities (PARI-HD)", "I-309 (CCL1)", "arXiv", "Bernoulli/Categorical Naive Bayes"], "alternates": {"html": "https://wpnews.pro/news/machine-learning-assessment-of-the-predictive-value-of-inflammatory-biomarkers", "markdown": "https://wpnews.pro/news/machine-learning-assessment-of-the-predictive-value-of-inflammatory-biomarkers.md", "text": "https://wpnews.pro/news/machine-learning-assessment-of-the-predictive-value-of-inflammatory-biomarkers.txt", "jsonld": "https://wpnews.pro/news/machine-learning-assessment-of-the-predictive-value-of-inflammatory-biomarkers.jsonld"}}