Slide 1 of 15

Fetal Growth Restriction

What It Is and Why It Matters

Chukwuma Onyeije, MD, FACOG
Medical Director, Atlanta Perinatal Associates
Founder, DoctorsWhoCode.blog · OpenMFM.org
Intern Orientation · June 2026

Welcome, Interns!

Summer 2026 Clinical Research Sprint at Atlanta Perinatal Associates.

Our Central Question: Can AI-assisted review of the medical record identify patients at risk for FGR earlier and more consistently than traditional clinical workflows — while keeping physicians responsible for all clinical decisions?

Your Team

Matthew Nwazue — Kennesaw State
Teyah Walker — Tuskegee University
Dayyid Dixon — Albany State
India Williams — Morris Brown College

Collaboration Partners

APA Physicians
Serelora Software Developers
OpenMFM.org / DoctorsWhoCode.blog

What Is Fetal Growth Restriction?

FGR is a high-risk pregnancy condition where a fetus fails to reach its growth potential.

Key Insight: FGR is not simply "a small baby." It describes a pathological process, not just a size measurement.

Concept	Definition	Clinical Significance
SGA (Small for Gestational Age)	Fetal size < 10th percentile for gestational age	Statistical descriptor; may be constitutionally normal
FGR (Fetal Growth Restriction)	Fetus failing to reach its genetic growth potential	Pathological; associated with placental insufficiency
Constitutionally Small	Small but otherwise healthy fetus with normal Dopplers	Requires surveillance to distinguish from true FGR
Placental Insufficiency	Impaired placental blood flow limiting fetal nutrients	The primary mechanism driving FGR in most cases

How Is FGR Diagnosed?

Diagnosis relies primarily on ultrasound biometry and Doppler velocimetry.

Primary Biometric Criteria

Estimated Fetal Weight (EFW) < 10th percentile
Abdominal Circumference (AC) < 10th percentile
Declining growth velocity across serial scans (crossing percentile lines)

Early vs. Late Onset FGR

Early-onset (< 32 weeks): More severe; strongly associated with placental insufficiency and abnormal Dopplers; higher risk of adverse outcomes
Late-onset (≥ 32 weeks): More common; often subtle; may have normal Dopplers; still carries elevated risk

Key Frameworks: SMFM Consult Series #52 (primary guidelines) · FIGO Initiative · Delphi Consensus (Gordijn et al., 2016)

Doppler Ultrasound in FGR

Umbilical artery (UA) Doppler measures placental resistance and is the cornerstone of FGR surveillance.

✅

Normal
Diastolic flow present
Routine surveillance

⚠️

Elevated S/D
Increased resistance
Increased surveillance

🔶

Absent EDV
Severe insufficiency
Hospitalization / delivery

🔴

Reversed EDV
Critical finding
Urgent delivery planning

Surveillance Note: MCA Doppler, Ductus Venosus, and Uterine Artery Dopplers are used in specific clinical scenarios but are not routine management drivers under SMFM #52.

Why FGR Matters: The Stakes

FGR is associated with a spectrum of adverse short-term and long-term outcomes.

Leading cause of stillbirth

Preterm birth risk

NICU admission

Long-term neurodevelopment

Short-Term Risks

Stillbirth (severe Doppler abnormalities)
Preterm birth (iatrogenic/spontaneous)
Neonatal hypoglycemia, polycythemia
Respiratory distress syndrome
NICU admission

Long-Term Risks

Neurodevelopmental delay (TRUFFLE study)
Cardiovascular disease in adulthood
Metabolic syndrome
Impaired cognitive function
Increased risk of chronic disease

Maternal and Pregnancy Risk Factors

Identifying patients at elevated risk for FGR relies on a thorough review of maternal and pregnancy history.

Category	Risk Factors
Maternal Medical	Chronic hypertension, preeclampsia, pregestational diabetes, autoimmune disease (lupus, APS), renal disease, thrombophilia
Obstetric History	Prior FGR, prior stillbirth, prior preeclampsia, prior preterm birth
Current Pregnancy	Multiple gestation, placental abnormalities (previa, abruption), cord abnormalities (single umbilical artery), IVF conception
Lifestyle / Demo	Smoking, substance use, low pre-pregnancy BMI, advanced maternal age, socioeconomic factors
Prevention Window	Low-dose aspirin (started ≤ 16 weeks) reduces preeclampsia and associated FGR risk in high-risk patients (ASPRE trial)

The Detection Gap: Our Core Problem

The interval between when FGR was first identifiable in the record and when it was formally documented.

Why this matters: Every week of delay in recognizing FGR is a week of missed surveillance, missed aspirin initiation, and potentially missed opportunity to prevent stillbirth.

Earliest-Identifiable Point

The gestational age at which the available record first contained enough information to meet FGR criteria — based on data available at that time.

Delayed Detection

FGR detection that occurred later than a pre-specified acceptable interval, or was not recognized antenatally at all before delivery.

Key Evidence: The TRUFFLE Study

LANDMARK TRIAL (Trial of Randomized Umbilical and Fetal Flow in Europe) in early-onset FGR.

TRUFFLE Question: In very preterm FGR (26–32 weeks), should delivery timing be guided by ductus venosus (DV) Doppler changes or by CTG short-term variation (STV)?

Key Finding

Delivery triggered by DV changes (absent/reversed a-wave) was associated with the best 2-year neurodevelopmental outcomes compared to CTG-based triggers alone.

Why It Matters to Us

Demonstrates that timing of delivery in FGR profoundly affects long-term outcomes
Supports the value of structured, guideline-based surveillance
Reinforces why missed/delayed FGR detection has lasting consequences

The Two-Engine AI Model

An architecture designed to be transparent, auditable, and physician-supervised.

🔍 Probabilistic Engine

What it does: Reads unstructured clinical text — notes, ultrasound reports, consultation letters — and extracts possible FGR-relevant signals.

Technology: Large Language Model (LLM)

→

⚙️ Deterministic Engine

What it does: Applies fixed, rule-based guideline logic (SMFM #52 via FGRManager) to the extracted data to recommend surveillance and timing.

Technology: FGRManager rule engine

The Physician Checkpoint: Both engines surface information for physician review. A physician — not the AI — makes all clinical decisions.

Our Study Design: Two Linked Phases

Structured steps to validate and test the AI-assisted review logic.

Phase A: Retrospective Study

→

Phase B: Prospective Silent Pilot

Phase A: Diagnostic Accuracy

Review prior APA pregnancies with growth ultrasound data
Determine when FGR criteria were met and measure the detection gap
Test whether AI-assisted review would have flagged cases earlier
Reference standard: Physician adjudication panel

Phase B: Silent Pilot

AI runs silently in parallel with usual care
No change to patient management during the pilot
Measures: alert rate, positive predictive value, safety
Prerequisite for any clinician-facing implementation

Your Role This Summer

You are research team members, not clinicians. Your contributions are essential.

✅ You WILL Do

Review and summarize assigned literature
Abstract approved data fields from records (after training)
Help maintain the data dictionary and perform QA checks
Map clinical workflows and identify detection gaps
Prepare figures, tables, posters, and manuscript sections

🚫 You Will NOT Do

Make or suggest clinical diagnoses
Change any clinical documentation in the records
Use patient information in personal AI tools (OpenAI, etc.)
Contact patients about the study
Share PHI outside approved APA secure systems

Golden Rule: When in doubt, document the uncertainty and escalate to your supervisor. The correct response to ambiguity is never guessing.

Week 1: Building Your Foundation

Learning the ropes. No production chart abstraction begins until you understand the clinical terms.

Day	Focus	Deliverable
Day 1	Orientation · Why FGR matters · Privacy onboarding	One-page reflection: "Why missed FGR detection matters"
Day 2	SGA vs. FGR · SMFM, FIGO, Delphi frameworks	Comparison table: SGA / FGR / Constitutional / Placental insufficiency
Day 3	Placenta · Doppler categories · Surveillance logic	Doppler category summary with student-level definitions
Day 4	Data dictionary · Workflow mapping · Documented vs. inferred data	Data dictionary contributions for assigned variables
Day 5	Literature synthesis · Abstraction examples · Competency check	Literature review outline · Gap analysis · Readiness sign-off

What You Will Produce This Summer

By the end of the sprint, you will have contributed to a suite of academic, data, and technical products.

📄 Academic Products

Annotated bibliography
2–3 page literature review
One-page gap analysis
Poster or slide contribution
Manuscript paragraph, table, or figure draft

📊 Data & Technical Products

Completed chart abstraction forms
Data dictionary entries
Double-abstraction quality-assurance notes
Clinical workflow map
AI logic documentation in plain language
Validation notes and missingness summary

Bigger Picture: This work supports abstract submission, a regional conference presentation, and a peer-reviewed manuscript — all with your name on it.

Let's Make a Difference

This project is about disciplined clinical research, not technology enthusiasm. The goal is to learn whether structured, AI-assisted review can help APA surface FGR risk earlier, more consistently, and more equitably — for mothers and babies throughout metropolitan Atlanta.

Welcome to the team. Let's get to work.

Chukwuma Onyeije, MD, FACOG · Medical Director, Atlanta Perinatal Associates
DoctorsWhoCode.blog · OpenMFM.org · CodeCraftMD