Screening for Diabetes Mellitus in the U.S. Population Using Neural Network Models and Complex Survey Designs

Abstract: Complex survey designs are commonly employed in many medical cohorts. In such scenarios, developing case-specific predictive risk score models that reflect the unique characteristics of the study design is essential for minimizing selective biases in the statistical results. The objectives of this paper are to: (i) propose a general predictive framework for regression and classification using neural network (NN) modeling that incorporates survey weights into the estimation process; (ii) introduce an uncertainty quantification algorithm for model prediction tailored to data from complex survey designs; and (iii) apply this method to develop robust risk score models for assessing the risk of Diabetes Mellitus in the US population, utilizing data from the NHANES 2011-2014 cohort. The results indicate that models of varying complexity, each utilizing a different set of variables, demonstrate different discriminative power for predicting diabetes (with different economic cost), yet yield generalizable results at the population level. Although the focus is on diabetes, this NN predictive framework is adaptable for developing clinical models across a diverse range of diseases and medical cohorts. The software and data used in this paper are publicly available on GitHub.

• The paper introduces a neural network framework that integrates survey weights and applies conformal inference for uncertainty quantification.
• It demonstrates robust diabetes risk estimation by accurately modeling complex survey data from the NHANES 2011-2014 cohort.
• The work offers a versatile methodology for both regression and classification tasks, advancing precision public health research.

Deep Learning Framework with Uncertainty Quantification for Survey Data

Introduction

The assessment of risk for diseases such as Diabetes Mellitus within specific populations relies heavily on the accuracy and reliability of predictive models. These models, however, must account for the complex survey designs that are commonly employed to gather data, ensuring that the unique characteristics of the sample are considered to minimize potential biases. This paper presents a comprehensive framework that leverages neural network (NN) modeling in tandem with an uncertainty quantification algorithm tailored for complex survey designs, specifically applied to the United States NHANES 2011-2014 cohort for assessing diabetes mellitus risk.

Neural Network Models and Machine Learning for Survey Data

Historically, the integration of machine learning models, including neural networks (NN), with complex survey data has been limited, owing to the inherent challenges in adjusting for the survey's design. This paper bridges this gap by proposing a general NN framework for regression and classification tailored to survey data. The methodology encapsulates:

• The introduction of survey weights into the NN modeling process, effectively incorporating the survey's design into the estimation phase.
• The innovation of an uncertainty quantification mechanism via conformal inference techniques, aptly suited for complex survey designs. This approach crucially accounts for the survey design in the quantification of model prediction uncertainty.

Diabetes Study Case: NHANES Application

The urgency of the diabetes epidemic, particularly in the U.S., underscores the necessity of robust and accurate predictive models. By applying the proposed NN framework to the NHANES 2011-2014 cohort data, this research not only demonstrates the framework's applicability but also its potential in advancing diabetes research. The models developed showcase the ability to yield reliable predictions and offer novel insights, emphasizing the framework's adaptability to various diseases beyond diabetes.

Summary of Contributions

The paper's contributions are significant and multi-faceted, encompassing:

• The pioneering application of NN models as estimators for survey data, specifically for developing disease risk scores.
• The proposal of a novel method to quantify prediction uncertainty, a critical aspect when dealing with complex survey data.
• The provision of a versatile framework that is applicable across regression and classification settings, illustrating the framework's broad utility.
• Insights derived from the application analysis in the context of diabetes, highlighting the model's capability in navigating the trade-offs between prediction accuracy and economic cost.

Future Directions

Looking ahead, this research opens avenues for further advancements in AI and deep learning, particularly in their application to epidemiology and public health. The paper speculates on future work that could extend the proposed methods to time-to-event analysis, tackle covariate functions more comprehensively, or even refine diagnostic tests through conditional ROC curve analysis.

Conclusion

By presenting a novel NN framework tailored for complex survey designs, this research marks a significant step forward in the application of machine learning to epidemiological and biomedical studies. The methodology's robustness, evidenced through its application to the NHANES diabetes case study, underscores its potential to enhance precision public health initiatives and contribute meaningful advancements to the field of biomedical science.