Assessing the Completeness and Concordance of Disability Data in the All of Us Research Program

Presentation Time: 02:00 PM - 02:12 PM

Abstract Keywords: Disability, Accessibility, and Human Function, Data Mining, Knowledge Representation and Information Modeling
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Electronic Health Records (EHR) might have missingness for EHR codes needed to identify participants living with disabilities. The All of Us Research Program collected disability status from its participants. We compared EHR disability codes with self-reported disabilities collected via surveys. Our analysis showed that EHR disability documentation has some missingness. Moreover, we applied PheWAS on EHR-disability codes and self-reported disabilities to identify associated phecodes. Simialrties and difference were observed which highlight the importance of combining both sources.

Speaker:
Lina Sulieman, PhD
Vanderbilt University Medical Center

Authors:
Lina Sulieman, PhD - Vanderbilt University Medical Center; Paul Harris, PhD - Vanderbilt University; Brandy Mapes, MLIS - Vanderbilt University Medical Center;

---

Analyzing and Mitigating Model Drift in Acute Kidney Injury Prediction for Hospitalized Patients

Presentation Time: 02:12 PM - 02:24 PM

Abstract Keywords: Evaluation, Machine Learning, Data Mining, Fairness and elimination of bias
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Artificial intelligence and machine learning are transforming healthcare by improving clinical risk predictions and diagnostic precision. However, their performance can be compromised by data drifts due to changes in patient populations and evolving clinical practices. This study investigated performance drift in models predicting Acute Kidney Injury (AKI) using electronic health records from 249,749 inpatient encounters over ten years, analyzing performance across both the overall population and nine subgroups with unique health profiles. To mitigate the performance drift, we implemented two model updating strategies: an Overall Population Update (OPU) and a Specific Subgroup Update (SSU). Our results demonstrated significant reductions in drift, with OPU increasing the average area-under-the-precision-recall-curve (AUPRC) by 0.14 in the overall population and 0.11 across subgroups, and SSU improving the average AUPRC by 0.10 among subgroups. These findings highlight the importance of continuous model surveillance and adaptive updates to maintain reliable predictive performance in dynamic clinical environments.

Speaker:
Zijian Xu, MS
University of Florida

Author:
Alan Yu, MD - University of Kansas Medical Center;

---

Quantifying the Condition Completeness using Medications in the All of Us Research Program

Presentation Time: 02:24 PM - 02:36 PM

Abstract Keywords: Data Standards, Data Modernization, Data Mining, Healthcare Quality
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics

Electronic Health Records (EHRs) are valuable for advancing discovery and precision medicine, but data quality issues often undermine research credibility. A significant challenge is assessing data completeness, as most metrics are too general and lack phenotype specificity. We propose using medication-condition relationships from reputable datasets like DrugBank to measure gaps in condition documentation. Our analysis has revealed high levels of missing documentation for various phenotypes. Cancer codes had high missingness, probably due to healthcare fragmentation.

Speaker:
Lina Sulieman, PhD
Vanderbilt University Medical Center

Authors:
Lina Sulieman, PhD - Vanderbilt University Medical Center; Joshua Smith, PhD - Vanderbilt University Medical Center; Paul Harris, PhD - Vanderbilt University; Karthik Natarajan, PhD - Columbia University Dept of Biomedical Informatics;

---

Improved prediction of resource use through ensemble modeling of administrative claims data

Presentation Time: 02:36 PM - 02:48 PM

Abstract Keywords: Population Health, Healthcare Economics/Cost of Care, Quantitative Methods
Primary Track: Applications
Programmatic Theme: Clinical Informatics

The Johns Hopkins ACG system is a health analytic solution that leverages clinical information to predict risk of outcomes such as hospitalization and care expenditures. Despite its effective prediction in the general population, local calibration of ACG models is often recommended for subpopulations due to variance in context. In this analysis, we aim to evaluate the predictive value of multiple techniques for estimating clinical risk, using local calibrations of a claims database with identical features.

Speaker:
Hadi Kharrazi, MD, PhD, FAMIA, FACMI
Johns Hopkins University

Authors:
Christopher Kitchen, MS - Johns Hopkins University; Klaus Lemke, MS - Johns Hopkins University; Chintan Pandya, PhD, MPH, MBBS, FAMIA - Johns Hopkins Bloomberg School of Public Health; Hadi Kharrazi, MD, PhD, FAMIA, FACMI - Johns Hopkins University; Jonathan Weiner, DrPH - Johns Hopkins University;

---

Using National COVID cohort collaborative (N3C) data to explore impact of COVID-19 infection on kidney function in patients receiving lithium therapy

Presentation Time: 02:48 PM - 03:00 PM

Abstract Keywords: Real-World Evidence Generation, Quantitative Methods, Data Standards, Controlled Terminologies, Ontologies, and Vocabularies, Public Health, Patient Safety
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics

Lithium is a drug primarily used to treat psychiatric disorders and has shown significant efficacy in treating bipolar disorder (BD) and major depressive disorder (MDD). Although lithium can stabilize emotional fluctuations and prevent the onset of depressive and manic episodes, case reports of patients treated with lithium developing kidney dysfunction after being infected with SARS-CoV-2 emerged during the COVID-19 pandemic. We used the National COVID Cohort Collaborative (N3C) dataset to investigate the relationship between patients with lithium exposure and with SAR-Co-V-2 infection, and kidney dysfunction. Using data from March 1, 2021, to September 30, 2022, patients were classified as whether they were infected with SARS-CoV-2. We used the estimated glomerular filtration rate (eGFR) to observe whether there are significant differences between the two groups. Our findings suggest that the impact of lithium treatment and COVID-19 on kidney function may not be significant, consistent with most other studies' findings.

Speaker:
Yujia Tian, BS
University of Michigan

Authors:
Yujia Tian, BS - University of Michigan; Yongqun He, PhD - University of Michigan; Rachel Richesson, PhD, MPH, FACMI - University of Michigan Medical School; Melvin McInnis, M.D. - Department of Psychiatry, University of Michigan Medical School; Anastasia Yocum, Ph.D. - Department of Psychiatry, University of Michigan Medical School; Jinju Li, BS - Center for Computational Medicine and Bioinformatics, University of Michigan Medical School;

---

The Impact of the HITECH Act on Health IT Innovation

Presentation Time: 03:00 PM - 03:12 PM

Abstract Keywords: Policy, Causal Inference, Evaluation
Primary Track: Policy
Programmatic Theme: Clinical Informatics

The 2009 HITECH Act accelerated electronic health record adoption in the US, but less is known about the second-order effects this program has had on innovation in medical informatics. Drawing from the US Patent and Trademark Office PatentsView database, we analyze the universe of granted patents from 2000 to 2023 to estimate the effects of the HITECH Act and find large increases in innovation with potential crowd-out of new entrants and academic innovators.

Speaker:
Nate Apathy, PhD
University of Maryland

Authors:
Nate Apathy, PhD - University of Maryland; A J Holmgren, PhD - University of California, San Francisco;

---