Skip to content

Simulation Details

The simulation generates synthetic interrelated FHIR R5 resources, including:

Implementation Details

The discrete-event simulation is implemented in SimPy. An SQLite database is used to log events and manage practitioner and patient data. The data export script (data_export.py), located in the code repository, allows exporting the data saved in the SQLite database as serialized FHIR resources (JSON) or in a custom simplified format (JSON).

Simulation Objective

The implemented discrete-event simulation aims to produce (as good as possible) realistic event sequences that can be classified as either normal or anomalous patterns of healthcare system usage. It aligns with the confidentiality scenario described by McGlade and Scott-Hayward (2019)1, where various event sequences are generated to mimic normal and anomalous usage patterns.

In this context, 'access' events by clinician users against patients can be defined, within a 24-hour period, as anomalous when a user accesses a patient record without a legitimate relationship to the patient and without a 'break-glass' event. A relationship is defined by the presence or absence of appointments, observations, or encounters linking the patient and the clinician user. A 'break-glass' event is when a clinician legitimately accesses a patient record in an emergency situation.

Classification Table

When event sequences are aggregated on a 24-hour basis, they can be classified according to the following classification table, as specified by McGlade and Scott-Hayward (2019)1:

ID Appointments Observations Encounters Break-Glass Class Description
1 ❌ ❌ ❌ ❌ Anomaly No relationship between clinician and patient, no emergency access.
2 ❌ ❌ ❌ ✅ Normal Emergency 'break-glass' event occurred.
3 ❌ ❌ ✅ ❌ Anomaly Clinician met patient, no future appointment, no emergency access.
4 ❌ ❌ ✅ ✅ Normal Emergency access to patient records during an encounter.
5 ❌ ✅ ❌ ❌ Anomaly Clinician observed patient, no future appointment, no emergency access.
6 ❌ ✅ ❌ ✅ Normal Emergency 'break-glass' event occurred during an observation.
7 ❌ ✅ ✅ ❌ Anomaly Clinician met and observed patient, no future appointment, no emergency access.
8 ❌ ✅ ✅ ✅ Normal Emergency access to patient records during an encounter and observation.
9 ✅ ❌ ❌ ❌ Normal Future appointment exists, no emergency access.
10 ✅ ❌ ❌ ✅ Anomaly Future appointment exists, but emergency 'break-glass' event occurred.
11 ✅ ❌ ✅ ❌ Normal Clinician met patient, future appointments exist, no emergency access.
12 ✅ ❌ ✅ ✅ Normal Emergency access to patient records during an encounter, future appointments exist.
13 ✅ ✅ ❌ ❌ Normal Clinician observed patient, future appointments exist, no emergency access.
14 ✅ ✅ ❌ ✅ Normal Emergency access to patient records during an observation, future appointments exist.
15 ✅ ✅ ✅ ❌ Normal Normal hospital visit, no emergency access.
16 ✅ ✅ ✅ ✅ Normal Emergency access to patient records during an encounter and observation, future appointments exist.

Simulation Assumptions

The discrete-event simulation is based on several key assumptions to ensure it produces realistic event sequences. The table below summarizes these assumptions:

Assumption Description
Discrete Time Units The simulation operates in discrete time units (minutes).
Practitioner Availability Practitioners have predefined work schedules (full-time, evening-shift, split-shift, part-time, weekend-only, and rotating 8-hour shifts) to ensure the simulation mimics real-world constraints and availability of healthcare practitioners.
Event Probabilities Events occur with predefined probabilities, and the time between events occurs based on predefined probability distributions.
Patient Cooldown Patients have a cooldown period between appointments to simulate realistic scheduling and prevent unrealistic frequent appointments.
Dynamic Patient Population The simulation dynamically adjusts the patient population based on discharge and new patient addition probabilities.
Sequence Generation of Events The simulation can generate singleton events or sequences of appointments, encounters, and observations:
- Appointments: Can be scheduled, cancelled, or result in no-shows.
- Encounters: Occur as actual visits and can follow appointments.
- Observations: Made by practitioners and can occur during encounters.
Random BTG/Normal Access Events The simulation can generate random Break The Glass (BTG) or normal access events at any time:
- BTG Access Events: Occur with a predefined probability and represent emergency access to patient records.
- Normal Access Events: Occur as part of regular scheduled interactions.

Event Transition Diagram

The following sequence diagram illustrates the (high-level) event generation process in the discrete-event simulation. It shows how different types of events (appointments, encounters, observations, and Break The Glass (BTG) events) are generated, along with their respective probabilities. The diagram also highlights the concurrent generation of BTG and normal access events, which can occur independently of the main event sequences.

sequenceDiagram
    participant MainProcess
    participant Appointment
    participant Encounter
    participant Observation
    participant BTGAccessEvent
    participant NormalAccessEvent

    MainProcess->>MainProcess: Sample Patient-Practitioner Pair

    loop Event Sequence Generation
        alt Sequence Type 1
            MainProcess->>Appointment: Generate Appointment (33.33% chance)
            alt Sequence Type 1.1
                Appointment->>Observation: Generate Observation (33.33% chance)
            end

            alt Sequence Type 1.2
                Appointment->>Encounter: Generate Encounter (66.66% chance)
                Encounter->>Observation: Generate Observation (33.33% chance)
            end
        end

        alt Sequence Type 2
            MainProcess->>Encounter: Generate Encounter (33.33% chance)
            Encounter->>Observation: Generate Observation (33.33% chance)
        end

        alt Sequence Type 3
            MainProcess->>Observation: Generate Observation (33.33% chance)
        end

        Observation->>BTGAccessEvent: Generate BTG Event (5% chance)

    end

    loop Concurrent BTG/Normal Access Event Generation
        alt BTG Access Event
            MainProcess->>BTGAccessEvent: Generate BTG Access Event (5% chance)
        end

        alt Normal Access Event
            MainProcess->>NormalAccessEvent: Generate Normal Access Event (5% chance)
        end
    end

  1. McGlade, D., & Scott-Hayward, S. (2019). ML-based cyber incident detection for Electronic Medical Record (EMR) systems. Smart Health, 12, 3–23. doi:10.1016/j.smhl.2018.05.001 ↩↩