Understanding Prediabetes in a Medicare Advantage Population Using Data Adaptive Techniques.

Area of Science:

• Health Informatics
• Predictive Modeling
• Public Health

Background:

• Prediabetes is a significant public health concern, often undiagnosed.
• Administrative data offers a potential resource for identifying at-risk populations.
• Early identification of prediabetes can facilitate timely interventions.

Purpose of the Study:

• To evaluate adaptive predictive models for identifying undiagnosed prediabetes using administrative data.
• To compare the performance of logistic regression, decision trees, neural networks, and ensemble models.
• To assess the prevalence of prediabetes within a national Medicare Advantage Prescription Drug (MAPD) population.

Main Methods:

• Retrospective, cross-sectional study utilizing MAPD administrative data (N=279,903).
• Development and evaluation of data-adaptive logistic regression, decision tree, neural network, and ensemble models.
• Performance metrics included misclassification rate (MCR), average squared error (ASE), c-statistics, sensitivity (SN), and false positive (FP) rates.

Main Results:

• The ensemble model showed superior performance for metabolic syndrome prediction (c-statistic=0.83).
• Logistic regression demonstrated the best discrimination for prediabetes prediction (c-statistic=0.67, MCR=0.13, ASE=0.11).
• The logistic regression model predicted prediabetes in 44% of the MAPD population, aligning with national estimates.

Conclusions:

• Logistic regression is a robust and effective method for predicting undiagnosed prediabetes in MAPD individuals.
• Administrative data, when analyzed with appropriate models, can identify individuals with prediabetes.
• This approach supports targeted public health initiatives for prediabetes management.