Synapse
⌘+K
Synapse
PulseExploreClubsResearchersJournals
Instagram
HomeClubsExplore
June 7, 2026Diabetes

The LightGBM model using discrete SDOH features achieved a testing-sample adjusted R2 of 0.948 and RMSE 0.671, outperforming OLS using ADI (adjusted R2 0.381, RMSE 2.51).

View Full Paper
Ask AI
Bookmark
Share

Population

Geolocated Maryland land parcels with 1317 features from calculating routing, distance, and count density of…

Comparison

Gradient boosted model using 1317 discrete… vs Ordinary Least Squares regression using area…

Design

Cross-sectional

Authors

SHSHUO J. HUANGRLRoland LaboulayeMBMatthew Bandos

Discussion

Loading...

Member takes

Overview

Machine learning models utilizing discrete address-level SDOH features significantly outperform traditional area deprivation indices in predicting tract-level diabetes prevalence.

Key Points

  • The study aims to identify the impact of discrete address-level social determinants of health on diabetes prevalence in Maryland.
  • Used a geolocated dataset with 1,317 features related to discrete social determinants of health (n=2,369,365)
  • Employed a gradient boosted machine learning model (LightGBM) on tract-level diabetes prevalence data
  • Compared model performance to ordinary least squares (OLS) analysis of area deprivation index (ADI)
  • The machine learning model achieved an adjusted R2 of 0.949 on training data and 0.948 on testing data.
  • Of the top 30 features, 8 were linked to local economic value and 8 to proximity to hospitals.
  • OLS analysis of ADI for diabetes prevalence produced a much lower adjusted R2 of 0.381.

Structured PICO

P
Population
Geolocated Maryland land parcels (n=2,369,365) with 1317 features from calculating routing, distance, and count density of geocoded discrete SDOH resources and hazards, plus ancillary data.
I
Intervention
Gradient boosted model (LightGBM) using 1317 discrete address-level SDOH features
C
Comparator
Ordinary Least Squares (OLS) regression using area deprivation index (ADI)
O
Outcome
Tract-level diabetes prevalence in CDC PLACES

Machine learning models utilizing discrete address-level SDOH features significantly outperform traditional area deprivation indices in predicting tract-level diabetes prevalence.

Limitations

  • cross-sectional analysis

Cite This Study

HUANG et al. (2026) studied this question.

synapsesocial.com/papers/6a250c7d7def13d035e1cb22https://doi.org/10.2337/db26-2994-lb
View Full Paper
Ask AI
Bookmark
Share