Stability Mechanisms
The Clarus Clinical Stability Benchmark organizes datasets according to two dimensions:
- Instability mechanism
- System domain
Each dataset represents a specific combination of these two dimensions.
This structure ensures the benchmark grows in a coherent way rather than as a collection of unrelated datasets.
Instability Mechanisms
These mechanisms describe the underlying dynamics that cause systems to become unstable.
Pressure Overload
Demand exceeds system capacity.
Examples:
- circulatory overload
- hospital workload spikes
- metabolic demand surges
Buffer Exhaustion
Protective reserves are depleted.
Examples:
- renal buffering failure
- endocrine metabolic exhaustion
- coagulation reserve collapse
Coupling Cascade
Multiple subsystems begin amplifying each other’s failures.
Examples:
- organ coupling cascades
- inflammatory cascades
- metabolic feedback loops
Delay Instability
The system response occurs too late to restore stability.
Examples:
- delayed antibiotic treatment
- delayed hemorrhage control
- delayed ventilation support
Control Loop Failure
Regulatory feedback mechanisms stop stabilizing the system.
Examples:
- respiratory drive failure
- autonomic instability
- thermoregulation breakdown
Recovery Window Closure
A system may be recoverable early but becomes unrecoverable once a stability threshold is crossed.
Examples:
- hemorrhage compensation collapse
- sepsis recovery window loss
- cardiac arrest progression
System Domains
Instability mechanisms can occur in different physiological or operational systems.
Examples of domains included in the benchmark:
- circulation
- microcirculation
- cellular energy metabolism
- respiratory regulation
- endocrine regulation
- thermoregulation
- coagulation
- electrolyte balance
- hospital operational systems
Dataset Design Rule
Each dataset in the Clarus benchmark represents: instability mechanism + system domain
Example:
| Mechanism | Domain | Dataset |
|---|---|---|
| Pressure overload | circulation | clinical-hemodynamic-collapse |
| Buffer exhaustion | metabolism | clinical-endocrine-instability |
| Coupling cascade | organs | clinical-organ-coupling-cascade |
| Delay instability | treatment | clinical-intervention-delay-failure |
| Control loop failure | respiration | clinical-respiratory-drive-instability |
| Recovery window closure | trauma | clinical-hemorrhage-compensation-collapse |
Benchmark Expansion
New datasets should follow this rule.
When expanding the benchmark:
- Choose a stability mechanism
- Choose a system domain
- Design variables that represent that regime
This ensures the benchmark remains structured while allowing the dataset suite to grow over time.