Data quality and discharge-time readmission risk in diabetes hospitalisations

BACKGROUND: Electronic health record (EHR) and administrative data are widely reused for readmission modelling, but models may learn from documentation completeness, coding pathways and measurement availability as well as from clinical status. OBJECTIVE: This study assessed how documentation completeness, administrative fields and interpretable clinical predictors shaped discharge-time 30-day readmission risk stratification among hospitalised patients with diabetes. METHOD: A retrospective secondary analysis of the public Diabetes 130-US hospitals for years 1999-2008 dataset was conducted. Encounters with death or hospice-related discharge dispositions were excluded. The outcome was 30-day readmission. Four calibrated logistic regression models compared core clinical variables, administrative fields and missingness indicators under a discharge-time prediction frame. Uncertainty was estimated using patient-level clustered bootstrap resampling. RESULTS: The cohort included 99,343 encounters from 69,990 patient identifiers; 11,314 encounters (11.4%) had 30-day readmission. Weight, maximum glucose serum, A1C result, medical specialty and payer code were missing or not recorded in 96.9%, 94.8%, 83.1%, 48.9% and 39.7% of encounters, respectively. Adding administrative fields increased the area under the receiver operating characteristic curve (AUROC) from 0.640 to 0.667 (ΔAUROC 0.0266, 95% clustered bootstrap confidence interval (CI) 0.0178 to 0.0347). Missingness indicators alone produced little change (ΔAUROC 0.0019, 95% CI -0.0007 to 0.0046). The full model containing administrative fields and missingness indicators showed similar discrimination (AUROC 0.666). CONCLUSION: The modest performance gain was driven mainly by administrative coding and discharge-pathway variables. EHR readmission studies should define prediction timing, report measurement availability and interpret model performance as a product of clinical data generation as well as clinical risk.Implications for health information management practice:Secondary-use EHR modelling should profile variable availability, establish predictor timing and provenance, test administrative-field contributions and report calibrated performance before local use.

Authors

Institutions

Publication Details

Journal
Health Information Management Journal
Published
2026-11-12
DOI
https://doi.org/10.1177/18333583261502956
Primary Topic
Medical Coding and Health Information
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Data quality and discharge-time readmission risk in diabetes hospitalisations

Qidong Yang
Health Information Management Journal
Medical Coding and Health Information
article

Data quality and discharge-time readmission risk in diabetes hospitalisations

Qidong Yang
article en

Abstract

BACKGROUND: Electronic health record (EHR) and administrative data are widely reused for readmission modelling, but models may learn from documentation completeness, coding pathways and measurement availability as well as from clinical status. OBJECTIVE: This study assessed how documentation completeness, administrative fields and interpretable clinical predictors shaped discharge-time 30-day readmission risk stratification among hospitalised patients with diabetes. METHOD: A retrospective secondary analysis of the public Diabetes 130-US hospitals for years 1999-2008 dataset was conducted. Encounters with death or hospice-related discharge dispositions were excluded. The outcome was 30-day readmission. Four calibrated logistic regression models compared core clinical variables, administrative fields and missingness indicators under a discharge-time prediction frame. Uncertainty was estimated using patient-level clustered bootstrap resampling. RESULTS: The cohort included 99,343 encounters from 69,990 patient identifiers; 11,314 encounters (11.4%) had 30-day readmission. Weight, maximum glucose serum, A1C result, medical specialty and payer code were missing or not recorded in 96.9%, 94.8%, 83.1%, 48.9% and 39.7% of encounters, respectively. Adding administrative fields increased the area under the receiver operating characteristic curve (AUROC) from 0.640 to 0.667 (ΔAUROC 0.0266, 95% clustered bootstrap confidence interval (CI) 0.0178 to 0.0347). Missingness indicators alone produced little change (ΔAUROC 0.0019, 95% CI -0.0007 to 0.0046). The full model containing administrative fields and missingness indicators showed similar discrimination (AUROC 0.666). CONCLUSION: The modest performance gain was driven mainly by administrative coding and discharge-pathway variables. EHR readmission studies should define prediction timing, report measurement availability and interpret model performance as a product of clinical data generation as well as clinical risk.Implications for health information management practice:Secondary-use EHR modelling should profile variable availability, establish predictor timing and provenance, test administrative-field contributions and report calibrated performance before local use.

Health Information Management Journal
Nanyang Technological University (SG)
Good health and well-being
Medical Coding and Health Information
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon