The authors designed a modular event-based architecture to enable dynamic, context-aware evaluation of best-practice clinical care using routine electronic medical record (eMR) data. The architecture separates data preparation from pathway rules and evaluates whether recommended care steps were delivered in the correct sequence and within relevant timeframes as encounter events accumulate.
Data were drawn from adult emergency department encounters across four hospitals between July 2022 and June 2024. The complete pipeline processed 451,124 encounters. The study reports that all data were transformed into encounter-level event streams to support longitudinal, timeline-based analysis of care processes.
The architecture operates through three sequential layers:
Event-stream transformation: heterogeneous eMR records were standardized and organized chronologically for each encounter, producing a stream of time-ordered clinical events.
Pathway-relevance filtering: events were filtered to retain only those relevant to the evaluated pathway. After filtering, 43,951,146 events remained, representing 76.3% of the original 57,631,084 extracted events.
State-based pathway evaluation: the framework updated pathway state as new clinical information arrived and assessed adherence to recommended actions based on state, timing, and sequence.
This layered approach isolates data normalization from clinical rule application, supporting reuse and clarity in both data handling and pathway logic.
For each encounter, the standardized event stream formed a chronological timeline. The evaluation layer used a state model derived from the New South Wales Adult Sepsis Pathway to track progression through pathway states as new events occurred. This allowed the system to assess whether care elements had been delivered within state-specific time windows and in the expected order.
The architecture was capable of prioritizing evaluations toward higher-acuity states; that is, it could recognize and evaluate progression to more severe states (for example, shock) without forcing passage through intermediate states if higher-acuity indicators appeared.
The architecture extracted and transformed 57,631,084 clinical events into encounter-level event streams. After applying pathway-relevance filtering, 43,951,146 events were retained (76.3% of extracted events). In total, 451,124 encounters were processed through the three-layer pipeline to support pathway evaluation.
Of the processed encounters, 129,907 (34.6%) entered the sepsis pathway. Within that cohort:
50,118 encounters (38.6% of those who entered the pathway) entered YELLOW state, reflecting progression to suspected sepsis.
3,349 encounters (2.6% of encounters progressing from GREEN) moved to RED state, indicating an initial shock state.
Among the 50,118 encounters that entered YELLOW state:
2,426 (4.8%) subsequently progressed to shock.
8,700 (17.4%) were recognized as having sepsis through a blood culture order.
4,287 (8.6%) were recognized through an antibiotic order.
These figures illustrate the architecture's capacity to detect sepsis-recognition events (blood culture or antibiotic orders) and to prioritize higher-acuity identification pathways when relevant events occur.
The study emphasizes that separating data preparation from pathway rules provides several practical benefits:
Transparency: clearly defined transformation and filtering steps disclose how raw eMR records are converted into the timeline used for evaluation.
Maintainability: modular components can be updated independently; data standardization steps and pathway rules are not tightly coupled.
Adaptability: the architecture can be reused for different pathways or modified to support alternative evaluation criteria, and it provides a foundation for interoperable, near-real-time pathway monitoring.
Together, these features support reproducible retrospective analyses and point toward possibilities for operational monitoring of care adherence.
The study reports ethics approval from the Western Sydney Local Health District Human Research Ethics Committee (2019/ETH09607). The authors declare no competing interests. All data produced in the study are available on reasonable request to the authors.
The authors conclude that the architecture demonstrates feasibility for transforming fragmented routine eMR data into a reusable longitudinal representation that supports context-dependent pathway evaluation. They note that the separation of data preparation and pathway rules enhances transparency and maintainability and serves as a foundation for future interoperable, near-real-time pathway monitoring.