Home › Research › APCD Maternal Health
2024 UNLV Faculty Opportunity Award

Pilot Testing an All-Payer Claims Database for Improving
Maternal Risk-Appropriate Care

A $35,000 interdisciplinary pilot study harnessing 4 terabytes of Virginia health data to combat severe maternal morbidity — using machine learning to unlock answers traditional data sources cannot provide.

View Impact Data Meet the Investigators
$35K
Project Budget
4 TB
Claims Data
500K
Patient Records
4
Pilot Studies
2
Grant Proposals
The Problem We're Solving

Severe Maternal Morbidity: A National Crisis

Severe maternal morbidity is one of the most urgent and under-addressed challenges in U.S. healthcare. The numbers demand action.

200%
Increase in severe maternal morbidity since 1993 — now a national emergency
50K+
Women and their newborns affected by severe maternal morbidity every year
40%
Of severe maternal morbidity cases are preventable with timely identification & treatment
$23K
Additional healthcare cost per mother-infant dyad affected by severe maternal morbidity

Disparities Make This Worse

Women of color face a disproportionately higher risk of severe maternal morbidity — including increased mortality — compared to white women. Systemic inequities in access to risk-appropriate care are a root cause our research directly targets.

Project Overview

A New Kind of Data for a Stubborn Problem

Traditional hospital discharge data captures only a snapshot of the delivery hospitalization — missing the full prenatal history that shapes maternal risk. All-Payer Claims Databases (APCDs) change everything.

APCDs aggregate all medical claims, pharmacy claims, dental claims, and eligibility records across all insurers into a single database — providing the most complete picture of healthcare delivery ever assembled at a population scale.

"Up to 40% of cases of severe maternal morbidity could be prevented with timely identification and treatment."

This project pilots the use of Virginia's APCD — five years of data covering approximately 500,000 pregnancies — to establish the team's competency with this novel data source and lay the groundwork for two major federal funding proposals.

What Is an APCD?

All-Payer Claims Database

Introduced in 2011, APCDs are statewide databases mandated by 18 states (including Nevada) to consolidate all insurance claims. They provide a longitudinal, population-level record from prenatal care through delivery and beyond.

Why AI?

4TB of Complexity Requires Deep Learning

Virginia's APCD slice is 4 terabytes of records. Machine learning and deep learning algorithms running on UNLV's $500K NSF-funded GPU cluster (housed at Switch Cloud) are the only viable path to extracting actionable patterns at this scale.

Computing Infrastructure

The UNLV Data Analytics Lab hosts 18 machines each with NVIDIA RTX™ A4000 GPUs, plus a high-end cluster at Switch — providing the backbone for all deep learning analyses in this project.

Scientific Aims

Four Pilot Research Questions

Each question builds our team's competency with APCD data and generates publishable findings to support our R01 federal grant proposals.

01

Prenatal vs. Hospital Claims Agreement

What proportion of risk-factors identified on claims during the prenatal period can also be identified in the delivery hospitalization claim record? This demonstrates our ability to link prenatal claims to delivery records across the APCD.

02

Risk Identification & Cost of Care

Do women with high-risk delivery complications have lower costs of care when their high-risk conditions are identified in the delivery hospitalization claim record? This tests our ability to work with the APCD cost data.

03

Risk-Appropriate Care & Outcomes

Do women with high-risk complications have lower odds of severe maternal morbidity when they receive risk-appropriate care? This links APCD data to the American Hospital Association Database to measure hospital level of care.

04

AI-Driven Risk Discovery

Can machine learning and data mining identify risk factors for severe maternal morbidity not yet recognized by clinical literature? This pilot leverages UNLV's GPU cluster for deep learning-based multinomial categorization on the full 4TB dataset.

Scientific Methods

Rigorous Methodology

Key Measures & Variables

1

Obstetric Comorbidity Index

Uses ICD codes to identify 20 conditions that elevate risk for maternal end-organ damage. Our team developed a validated binary classifier (high-risk yes/no) applied to the full sample.

2

CDC Standardized Severe Maternal Morbidity Protocol

Uses 21 ICD diagnostic and procedure codes to identify conditions where maternal deterioration requires intervention to prevent death — the gold standard outcome measure.

3

Risk-Appropriate Care Measure

A validated dichotomous variable indicating whether a high-risk woman delivered at a facility with obstetric critical care services, using our team's validated hospital-level-of-care method.

4

Cost-to-Charge Ratio (AHRQ)

Costs of care derived from insurance claim charges using the Agency for Healthcare Research and Quality's standardized Cost-to-Charge Ratio for Inpatient Files.

Analysis Framework

Generalized Linear Models (GLM)

The primary statistical framework, guided by research question, underlying theory, data availability, and nature of the data. Substantial attention to measurement error, estimation bias, and model assumption failures.

Deep Learning for Multinomial Classification

The 4TB Virginia APCD slice requires advanced deep learning for multinomial categorization. The CS team will clean and prepare data for GPU-cluster-based algorithms to surface previously unidentified risk patterns.

Sample & Scope

Women receiving healthcare in Virginia during pregnancy, 2019–2023. ~95,000 births/year → ~500,000 pregnancies over 5 years. Identified by any ICD or DRG code indicating pregnancy.

Project Investigators

The Research Team

A purposeful pairing of clinical nursing expertise and computational power.

Dr. Jennifer Vanderlaan

Dr. Jennifer Vanderlaan

Principal Investigator · Associate Professor, School of Nursing

A certified nurse-midwife with a Master's in Public Health and PhD in Nursing (Emory University), Dr. Vanderlaan sits at the intersection of clinical care, health policy, and data science. She holds an AHRQ R36 grant for risk-appropriate care research, has developed and validated novel measures for hospital maternal level of care and sample selection methodology for high-risk women, and leads the American College of Nurse-Midwives Midwifery Workforce Study (Johnson & Johnson Foundation).

Maternal Health Risk-Appropriate Care Nurse-Midwifery Health Policy

ROLE IN THIS PROJECT

Conceptualization · Funding Acquisition · Project Administration · Formal Analysis · Writing

Dr. Kazem Taghva

Dr. Kazem Taghva

Chair & Professor, Department of Computer Science

A pioneer in large-scale information retrieval and database systems, Dr. Taghva led development of the U.S. Department of Energy's Licensing Support Network database from 1989 to 2009 — one of the largest legal-discovery databases ever built. Known internationally for model-based retrieval from noisy text (featured in Croft, Metzler & Strohman's Search Engines), he brings foundational expertise in data engineering, OCR, and NLP to the team's AI pipeline.

Data Engineering Machine Learning Information Retrieval NSF

ROLE IN THIS PROJECT

Data Curation · Resources & Infrastructure · Software Development · Machine Learning Lead · Writing

Project Plan

Four Steps to Data Mastery

1

Acquire Data

Purchase 5 years of Virginia APCD data (~$35,000 for 25M records over 2 years). Data stored on UNLV's Computer Science servers. Timeline and cost learnings will directly inform our R01 budget proposals.

2

Prepare Data for Use

Assess data completeness; apply the Obstetric Comorbidity Index to every record; add hospital level of care using our validated method; identify data slices for each research question.

3

Conduct Pilot Analyses

GLM-based analyses for questions 1–3; GPU-accelerated deep learning for question 4. Prepare manuscripts and conference abstracts demonstrating competency with APCD data.

4

Submit Grant Proposals

Submit two federal R01 proposals — one to AHRQ (February 2026) and one to NIH IMPROVE Initiative (June 2026) — using pilot data as proof of competency and feasibility.

Estimated Project Timeline

Acquire Data
Data Preparation
Pilot Analyses
ML Pilot
Submit Manuscripts
Prepare AHRQ R01
Prepare NIH R01

24-month project timeline. Each bar represents approximate duration and start point.

Federal Grant Targets

AHRQ R01 (February 2026): Missed opportunities to diagnose and treat high-risk maternal conditions.

NIH R01 (June 2026): System-level factors in disparities in risk-appropriate care (PAR-24-059).

Expected Outputs

Project Deliverables

The project will produce at least two peer-reviewed publications demonstrating APCD competency, plus two federal grant proposals targeting $2M+ in follow-on funding.

Type Topic Lead Status
Manuscript Risk for severe maternal morbidity: Agreement between prenatal and hospital claims data Dr. Vanderlaan In Progress
Manuscript Machine learning to identify previously unrecognized risks for severe maternal morbidity Dr. Taghva In Progress
Grant Missed opportunity to diagnose and treat maternal high-risk (AHRQ R01) Dr. Vanderlaan Due Feb 2026
Grant System-level factors in disparities in risk-appropriate care (NIH R01, PAR-24-059) Dr. Vanderlaan Due Jun 2026
Get Involved

Interested in This Research?

We welcome collaborations with clinicians, health systems, state health departments, and researchers who share our commitment to reducing maternal morbidity disparities.

Contact the Team All Research