Analyzing Healthcare Interoperability Vulnerabilities: Formal Modeling and Graph-Theoretic Approach

Jawad Mohammed; Gahangir Hossain

cs.CRApr 3, 2026

Analyzing Healthcare Interoperability Vulnerabilities: Formal Modeling and Graph-Theoretic Approach

Authors: Jawad Mohammed, Gahangir Hossain

Organizations: University of North Texas, Denton, TX, USA

Abstract

In a healthcare environment, the healthcare interoperability platforms based on HL7 FHIR allow concurrent, asynchronous access to a set of shared patient resources, which are independent systems, i.e., EHR systems, pharmacy systems, lab systems, and devices. The FHIR specification lacks a protocol for concurrency control, and the research on detecting a race condition only targets the OS kernel. The research on FHIR security only targets authentication and injection attacks, considering concurrent access to patient resources to be sequential. The gap in the research in this area is addressed through the introduction of FHIR Resource Access Graph (FRAG), a formally defined graph G = (P,R,E, λ, τ, S), in which the nodes are the concurrent processes, the typed edges represent the resource access events, and the race conditions are represented as detectable structural properties. Three clinically relevant race condition classes are formally specified: Simultaneous Write Conflict (SWC), TOCTOU Authorization Violation (TAV), and Cascading Update Race (CUR). The FRAG model is implemented as a three-pass graph traversal detection algorithm and tested against a time window-based baseline on 1,500 synthetic FHIR R4 transaction logs. Under full concurrent access (C2), FRAG attains a 90.0% F1 score vs. 25.5% for the baseline, a 64.5 pp improvement.

Figures & tables

Figure 1: Cyber threat landscape in FHIR-based healthcare interoperability. Multiple independent systems access shared patient resources concurrently with no synchronization protocol (

S=\emptyset

), giving rise to race condition classes SWC, TAV, and CUR.

Approach	SWC	TAV	CUR	Primary barrier
DB row locking	Partial	No	No	Single-system only
ETag / opt. lock	If adopted	Partial	No	Inconsistent adoption
Pessimistic locking	Yes	Partial	Partial	HTTP statelessness
Event sourcing / CRDT	Yes	No	Better	Full system redesign
FHIR Subscriptions	N/A	N/A	Partial	Client adoption
FRAG (this paper)	Detect	Detect	Detect	—

Table 1: Coverage of synchronization approaches across race classes

EHR ( $p_{1}$ )		Lab ( $p_{2}$ )
$\downarrow$ read		$\downarrow$ write
AllergyIntolerance ( $r_{1}$ )
$\uparrow$ read
CDS ( $p_{3}$ )

Figure 2: FRAG showing

RC(p_{1},p_{2},r_{1})

between EHR READ and Lab WRITE on AllergyIntolerance, and a

\text{CUR}(p_{1},p_{2},p_{3})

chain where CDS acts on EHR’s stale read.

S=\emptyset

throughout.

Class	FRAG pattern	Clinical risk	Prop.
SWC	WRITE $\\|$ WRITE, $S{=}\emptyset$	Lost prescription update	1
TAV	READ $\to$ act, WRITE intervenes	Auth bypass	2
CUR	Stale READ, path $\geq 2$	Contraindicated drug	3

Table 2: Healthcare race condition classification in FRAG

Figure 3: The three healthcare race condition classes formally defined in FRAG: (a) Simultaneous Write Conflict (SWC) — two concurrent WRITE operations with no synchronization; (b) TOCTOU Authorization Violation (TAV) — permission revoked inside the vulnerability window; (c) Cascading Update Race (CUR) — a stale-read dependency chain of depth

\geq 2

Class	P	R	F1	F1	$\Delta$ F1
	(FRAG)	(FRAG)	(FRAG)	(Base)	(pp)
SWC	97.4%	98.7%	98.0%	99.2%	$-$ 1.2
TAV	99.7%	100.0%	99.9%	98.8%	$+$ 1.1
CUR	33.9%	53.5%	41.5%	98.4%	$-$ 56.9
Overall	72.5%	85.5%	78.5%	98.8%	$-$ 20.3

Table 3: Detection performance: FRAG vs. Baseline (C2 — 500 logs)

Figure 4: FRAG vs. Baseline detection performance under C2 (concurrent, unsynchronized access). Lines show Precision, Recall, and F1 per race class. The shaded region highlights the detection gap between FRAG and the Baseline F1.

Figure 5: F1 score comparison between conditions C2 and C3. Arrows and labels show the recall-driven F1 drop under partial ETag synchronization. Baseline performance remains flat across both conditions.

Figure 6: Per-class Precision, Recall, and F1 for FRAG under C2 (left) and C3 (right). Red

\times

markers indicate corresponding Baseline F1 scores. Precision remains high under C3 while recall decreases due to ETag-protected write filtering.

Class	P	R	F1	F1	$\Delta$ F1
	(FRAG)	(FRAG)	(FRAG)	(Base)	(pp)
SWC	96.6%	8.6%	15.7%	99.1%	$-$ 83.4
TAV	99.0%	28.6%	44.3%	98.5%	$-$ 54.2
CUR	31.9%	16.7%	21.9%	98.8%	$-$ 76.9
Overall	61.0%	16.9%	26.4%	98.8%	$-$ 72.4

Table 4: Detection performance: FRAG vs. Baseline (C3 — 500 logs)

Component	Tool	Purpose
Log Generation	Python + NumPy	Synthetic FHIR transaction logs
FRAG Construction	NetworkX DiGraph	Build directed access graph
Race Detection	Custom Python (Alg. 1)	3-pass SWC/TAV/CUR detection
Baseline Scanner	Python sliding window	Time-window comparison
Evaluation	Custom Python (eval_v2)	Race-instance P/R/F1
Visualization	Matplotlib + NetworkX	Graph plots, result charts

Table 5: Simulation implementation stack

Healthcare

Healthcare AI develops and evaluates models and systems for clinical care, medical research, public health, and healthcare operations.

See all 616 papers

Reinforcement Learning for Tool-Calling Agents in Fast Healthcare Interoperability Resources (FHIR)
May 13, 2026Marius S. Knorr, Robert Müller, Jan P. Bremer +1Healthcare RL for Language Model Reasoning
IDM gGmbH, University Medical Center Hamburg-Eppendorf, Hamburg, Germany · Aganthos
Logit-Boundary Geometric Belief Interfaces and Sparse Sheaf-Enclave Protocols: A Self-Contained Substrate for Secure Network Electronic Health Record (EHR) Interoperability
Aug 10, 2026Alvin Spivey, Yu HuangLLM Reliability Electronic Health Records
Light Imaging Technologies, Inc.
FHIRPath-QA: Executable Question Answering over FHIR Electronic Health Records
Feb 26, 2026Michael Frew, Nishit Bheda, Bryan TrippEHR QA Clinical QA
University of British Columbia · University of Waterloo

May 13, 2026J/K move · Enter open · S save

Reinforcement Learning for Tool-Calling Agents in Fast Healthcare Interoperability Resources (FHIR)

Marius S. Knorr, Robert Müller, Jan P. Bremer, Nils Schweingruber

IDM gGmbH, University Medical Center Hamburg-Eppendorf, Hamburg, Germany · Aganthos

Healthcare RL for Language Model Reasoning AI Agent Benchmarks LLM Agent Training RL for Tool Use

Open paper AskPDF arXiv

Fast Healthcare Interoperability Resources (FHIR) is the dominant standard for interoperable exchange of healthcare data. In FHIR, electronic health records form a directed graph of resources. Answering clinically meaningful questions over FHIR requires agents to perform multi-step reasoning, filtering, and aggregation across multiple resource types. Prior work shows that even tool-augmented LLM agents (retrieval, code execution, multi-turn planning) often select the wrong resources or violate traversal constraints. We study this problem in the context of FHIR-AgentBench, a benchmark for realistic question answering over real-world hospital data, and frame reasoning on FHIR as a sequential decision-making problem over a queryable structured graph. We implement a multi-turn CodeAct agent and post-train it with reinforcement learning using a custom harness and tools. A LLM Judge provides execution-grounded rewards. Compared to prompt-based, closed-model baselines, RL post-training improves performance while enforcing data-integrity constraints. Empirically, our approach improves answer correctness from 50% (o4-mini) to 77% on FHIR-AgentBench using a smaller and cheaper Qwen3-8B model. We present an end-to-end post-training pipeline (environment building, harness construction, model training and custom evaluation) that reliably improves multi-turn reasoning over structured clinical graphs.

Abstract

Healthcare

Explore similar work

Reinforcement Learning for Tool-Calling Agents in Fast Healthcare Interoperability Resources (FHIR)

Logit-Boundary Geometric Belief Interfaces and Sparse Sheaf-Enclave Protocols: A Self-Contained Substrate for Secure Network Electronic Health Record (EHR) Interoperability

FHIRPath-QA: Executable Question Answering over FHIR Electronic Health Records