Question it answers
How far can a lean staged model go before loop-heavy healthcare behavior needs richer state, routing, and resource detail?
Sepsis Cases
TLDR: The public Sepsis Cases log is a much harder test than BPI Offer. It still maps into Simply Solver as a staged DES abstraction, but the heavy diagnostic loops and repeated tests make it a weaker one-to-one mirror of a simple Simio or FlexSim flowline. It is valuable precisely because it shows where the abstraction starts to strain.
Start Here
How to read this case study
This is an advanced benchmark, not a toy healthcare example. Read it as a stress test of the abstraction: what still holds, what starts to compress, and why the diagnostic loop is the main story.
- Begin with the loop structure. The diagnostic cluster matters more than the intake spine, so start there before reading the rest of the board.
- Use the validation table to set expectations. The mean stays directionally credible, but the median and tail tell you the abstraction is compressing a richer real process.
- Open the template as an advanced benchmark. Test whether faster diagnostics beat downstream capacity changes before you simplify the routing.
Open The Template
Launch the benchmark directly
One click creates a runnable copy so you can inspect the dense routing, run the baseline, and try the recommended experiments without a library detour.
- Best for: a loop-heavy healthcare benchmark with visible abstraction limits.
- First experiment: shorten Leucocytes, CRP, and LacticAcid together.
- Watch: mean cycle time, throughput/day, and pressure in the diagnostic cluster.
What to focus on
The diagnostic loops, the admission branches that materially change path shape, and the gap between directional fit and trace fidelity.
Why it matters
It shows where Simply Solver is still useful for structured scenario thinking and where a richer Simio or FlexSim-style model would earn its complexity.
Cases
1,050
Real hospital pathways represented as anonymized sepsis cases.
Activities
16
Diagnostics, treatment, admission, release, and return events.
Repeated Activity Cases
70.9%
Most traces revisit at least one activity, which drives complexity.
Mean Cycle
683.26 h
The mean is manageable, but the tail is extremely long.
Where activity concentrates
The workload clusters around repeated lab and diagnostic stages, not around a clean single-pass care pathway.
Why this is harder than a clean flowline
The strongest signals are loops between diagnostics and triage. That is exactly the kind of behavior that makes a simplistic staged model less faithful.
These are top observed transitions from the public preprocessing artifact, not hand-picked hypothetical paths.
Why This Problem
The Sepsis Cases log is a better second case study than a second clean flow model because it stresses a different weakness. The public dataset describes about 1,000 hospital cases, roughly 15,000 events, 16 different activities, and 39 data attributes. In local preprocessing, 70.86% of cases show repeated activities and the top transitions repeatedly bounce between diagnostics. That makes it much closer to a loop-heavy clinical pathway than to a one-directional line.
| Signal | Value |
|---|---|
| Public dataset | Sepsis Cases - Event Log |
| DOI | 10.4121/uuid:915d2bfb-7e84-49ad-a286-dc35f063a460 |
| Cases parsed | 1,050 |
| Mean cycle time | 683.259 hours |
| Median cycle time | 128.241 hours |
| P90 cycle time | 2,225.617 hours |
| Repeated activity traces | 744 cases (70.86%) |
| Same-activity reentry events | 1,034 |
How The Model Was Built
The public abstraction keeps explicit entry, triage, diagnostics, treatment, and release-style outcomes, but it does not try to model every clinical nuance. The main processors stay visible so the diagnostic loop is not hidden. Release and return events are treated as terminal abstractions for the case-study view even though the real traces are more nuanced. That choice keeps the model explainable while still preserving the high-signal routing behavior.
Now available as an advanced template
The loadable template keeps the calibrated loop-heavy structure from the public validation artifact, including a source anchored to the first observed event in the benchmark rather than to a hand-authored intake chain. It is intentionally denser than the BPI offer model, so treat it as an advanced benchmark board rather than a first-time starter flow.
| Observed signal | Case-study abstraction |
|---|---|
| ER intake and triage | Explicit processors for registration, triage, and sepsis triage |
| Repeated diagnostics | Preserved as routed loops among Leucocytes, CRP, and LacticAcid |
| Treatment steps | IV Liquid and IV Antibiotics remain explicit processors |
| Admission branching | Admission NC and Admission IC stay visible where they materially change the path |
| Discharge and return outcomes | Curated release and return abstractions keep the case study readable |
Baseline Validation
The Sepsis abstraction is good enough to keep mean cycle time and throughput in the neighborhood, but it is materially less trace-faithful than the BPI offer case. That is not a failure. It is evidence that loop-heavy healthcare pathways demand richer state, timing, and resource logic if the goal is to reproduce the shape of the full distribution.
Empirical vs simulated
The model stays directionally credible on the mean, but the median and tail tell you that this abstraction is compressing a much more complex real process.
Mean cycle time-3.50%
Throughput / day-5.19%
Median cycle time+258.25%
P90 cycle time-33.02%
| Metric | Empirical | Simulated | Difference |
|---|---|---|---|
| Mean cycle time | 683.259 h | 659.312 +/- 17.516 h | -3.50% |
| Arrivals / throughput per day | 2.494 | 2.365 +/- 0.101 | -5.19% |
| Median cycle time | 128.241 h | 459.418 +/- 22.907 h | +258.25% |
| P90 cycle time | 2,225.617 h | 1,490.762 +/- 58.819 h | -33.02% |
| Mean wait | Not directly observed | 3.555 min | Loop interactions remain simplified |
How Close Is This To Simio Or FlexSim?
The answer is more qualified here than it is for BPI Offer. The basic object language still maps well: a source feeds staged processors, routing sends cases through diagnostics and treatment, and release-style sinks close the abstraction. But the moment repeated diagnostics, self-loops, and clinical escalation become the story, richer routing logic and more explicit state become much more important. In other words, Simply Solver still looks like the same DES family, but this case exposes the limits of a lean staged model faster.
| Dimension | Simply Solver case study | Simio / FlexSim comparison |
|---|---|---|
| Core object pattern | Still recognizable as source -> processors -> sinks | Close at the topological level |
| Loop handling | Possible, but quickly becomes dense and assumption-heavy | Richer state and routing tooling helps more in Simio/FlexSim on cases like this |
| Diagnostic repetition | Represented through repeated routed returns | Conceptually similar, but harder to keep clean in a lean staged abstraction |
| Distribution fidelity | Mean can be calibrated; shape remains hard | Same challenge exists everywhere, but richer models make targeted structure easier to add |
| Best use | Decision framing and abstraction stress-test | Less like a textbook flowline, more like a model-selection boundary case |
Bottom line
Sepsis is useful because it is not clean. It shows that Simply Solver can still produce a serious public-data case study on a loop-heavy healthcare log, but it also makes clear where a lean staged model stops being a close mirror of a richer DES implementation.
What This Model Does Not Claim
- It does not model clinical decision quality or patient safety.
- It does not preserve every repeated diagnostic path with full trace fidelity.
- It does not model staffing, calendars, or treatment resources at hospital-operational depth.
- It should be read as a complex public benchmark case, not a hospital digital twin.