Table 2_Uncovering non-linear dietary predictors of cardiovascular disease risk in older a

Data from three NHANES cycles (2009–2014) and the MyPyramid Equivalents Database (MPED) was used to investigate non-linear relationships between diet and cardiovascular disease risk in older adults with periodontitis. The dataset includes results from an external validation study of 183 older adults and features selected using the BORUTA algorithm. It was authored by Jiefan Liu and last updated on March 18, 2026.

Use Cases

• Training interpretable ML models like XGBoost to predict CVD risk based on dietary trace elements and food intake.
• Applying SHAP and LIME algorithms to identify key protective dietary factors, such as theobromine, lycopene, and red meat consumption.
• Validating non-linear dietary predictors of disease in a specific inflammatory population using an external hospital-based cohort.
• Comparing the predictive performance of different ML models for clinical nutrition guidance.

Strengths

• Includes an external validation cohort of 183 older adults with periodontitis, supporting generalizability.
• Model performance is quantified, with XGBoost achieving a Validation AUC-ROC of 0.889 for the MPED model.
• Key predictors are identified through SHAP analysis, including specific nutrients like theobromine, lycopene, and food groups like meat and whole grains.
• Results are supported by multivariable logistic regression showing significant odds ratios (e.g., OR=0.46 for red meat).

Limitations

• Row count and column-level documentation are absent; field semantics must be inferred after download.
• The dataset is very small at 13.8 KB, indicating limited raw data scope, likely containing summary results or a single table.
• Data may reflect temporal and geographic bias inherent to the NHANES survey cycles used.

Provenance

Source

Integrated data from NHANES (2009–2014) and the MyPyramid Equivalents Database (MPED), with an external hospital-based cohort.

Collection Method

Cross-sectional analysis with feature selection using correlation coefficients and the BORUTA algorithm; six ML models were constructed and interpreted with SHAP/LIME.

Time Range

2009–2014 (NHANES cycles)

Freshness

Last updated 2026-03-18 06:42:30

Geography

null