A Machine Learning Classification Model for Gastrointestinal Health in Cancer Survivors: Roles of Telomere Length and Social Determinants of Health

Claire J. Han, Xia Ning, Christin E. Burd, Fode Tounkara, Matthew F. Kalady, Anne M. Noonan, Diane Von AhInternational Journal of Environmental Research and Public Health 2024 · DOI

TL;DR

This study developed machine learning models to predict gastrointestinal (GI) health in cancer survivors, revealing that biological aging (measured by telomere length) and social determinants of health (SDOH) are crucial predictors. The findings highlight the potential for personalized interventions to improve GI health outcomes in this population.

What Problem Does This Paper Address?

Cancer survivors often experience persistent gastrointestinal (GI) distress, impacting their quality of life and overall well-being. While GI health screening is important, accurate methods for classifying GI health status in survivors are lacking. This paper aims to bridge this gap by developing machine learning (ML) models that can classify GI health as "better" or "worse" by incorporating biological aging markers, specifically telomere length (TL), and social determinants of health (SDOH) alongside demographic and clinical data. Understanding these factors is crucial for developing targeted interventions to manage and prevent GI distress in this growing population.

Methods

The researchers employed a machine learning approach to classify GI health. They utilized six supervised ML models: logistic regression (LR), support vector machine (SVM), decision tree (DT), random forest (RF), gradient boosting machine (GBM), and extreme gradient boosting (XGBoost). The models were trained on 75% of the data and validated on the remaining 25%. Feature importance was assessed using permutation feature importance (PFI) to identify the most influential factors for predicting GI health.

Datasets

The study used data from the National Health and Nutrition Examination Survey (NHANES) from 1999–2002. This dataset was chosen because it included telomere length (TL) data from leukocyte DNA samples. The analysis included 645 adult cancer survivors who self-reported a cancer diagnosis and did not have chronic non-cancer GI disorders.

Key Insights and Ideas

  • Biological Age Matters: Telomere length, a marker of biological aging, emerged as a highly influential factor in predicting GI health, suggesting that how "old" a person's cells are is more relevant than their chronological age for GI well-being.
  • SDOH is Crucial: Social determinants of health, particularly poverty level and food security, were found to be significant predictors of GI health. This underscores the impact of socioeconomic factors on the health outcomes of cancer survivors.
  • Holistic Approach Needed: The study demonstrates that a combination of biological markers (TL) and social factors (SDOH) provides a more comprehensive picture of GI health risk than clinical or demographic data alone.
  • ML for Personalized Care: The developed ML models show promise in identifying cancer survivors at higher risk for GI distress, paving the way for more personalized and proactive interventions.

Key Results

  • Model Performance: In the training dataset, the Random Forest (RF) model achieved the highest performance with an AUC of 0.9842. In the test dataset, the Gradient Boosting Machine (GBM) model performed best with an AUC of 0.8035.
  • Top Predictors: For better GI health, the most influential features identified were longer telomere length and an annual household income above the poverty level. Other significant factors included routine physical activity, lower white blood cell counts, and food security.
  • Feature Importance: Telomere length was consistently ranked as the most important feature for predicting better GI health across both training and testing datasets. Lower poverty level was the second most significant factor.

Limitations and Future Work

The authors acknowledge several limitations:

  • Cross-Sectional Data: The use of cross-sectional NHANES data limits the ability to establish causality and predict future GI health trajectories. Longitudinal studies are needed to confirm these findings.
  • Inflammatory Markers: The utility of WBC and CRP as predictors was unclear, possibly due to their levels falling within the clinically normal range in the study population.
  • Test Dataset Size: The relatively small size of the test dataset may affect the generalizability of the findings.
  • GI Health Measurement: A single question was used to assess GI health, which might not capture the full complexity of GI distress.
  • Unknown SDOH-TL Link: The specific mechanisms linking SDOH to telomere length in the context of GI health remain to be fully elucidated.

Future work should focus on longitudinal studies, external validation of the models, and incorporating a wider range of clinical characteristics (e.g., cancer stage, treatment types) to enhance model predictability. The researchers also suggest exploring the integration of ML models into patient portals for personalized insights and interventions.

Summary updated

Abstract

Our findings provide valuable insights into classifying and identifying risk factors related to GI health, including biological aging and SDOH indicators. To enhance model predictability, further longitudinal studies and external clinical validations are necessary.
BibTeX
@article{han2024machine,
  author = {Claire J. Han and Xia Ning and Christin E. Burd and Fode Tounkara and Matthew F. Kalady and Anne M. Noonan and Diane Von Ah},
  title = {{A Machine Learning Classification Model for Gastrointestinal Health in Cancer Survivors: Roles of Telomere Length and Social Determinants of Health}},
  journal = {International Journal of Environmental Research and Public Health},
  year = {2024},
  doi = {10.3390/ijerph21121694},
}
Figure