FOR RESEARCH/OPERATIONAL PLANNING ONLY – NOT FOR CLINICAL DIAGNOSIS
This tool supports workforce capacity planning and operational decision-making. It is not a medical device and must not be used for individual clinical diagnosis, treatment decisions, or to determine fitness for duty. Clinical decisions require qualified healthcare professionals with access to full patient history.
Predict recovery trajectories for injured personnel.
Evidence-based recovery prediction tool with Cox proportional hazards modelling calibrated to peer-reviewed military and civilian literature.
Open SEKHMET Recovery Predictor - No installation required!
| Model | Description | Use Case |
|---|---|---|
| Cox PH (Evidence-based) | Cox proportional hazards with Weibull baseline, calibrated to peer-reviewed sources | Clinical decision support, research |
| Bayesian (Clinician-adjustable) | Adjustable parameters for local calibration | Site-specific tuning |
| XGBoost (ML/SHAP) | Machine learning with SHAP explainability | Research, model comparison |
- Input case details (injury type, body region, severity, risk factors)
- Traffic light summary (RTD likelihood at 3/6/12 months)
- Recovery timeline with survival curves and 90% CI
- Hazard ratio contributions
- Model agreement comparison (Cox vs XGBoost)
- Comparator benchmark (your case vs typical 30yo)
- Forecast recovery timelines across a team
- Gantt-style availability planning
- Band distribution analysis
- Full reference list with DOI links
- Parameter-to-source mapping table
- Evidence summary statistics and charts
- BibTeX and text export for citation managers
Click the Streamlit badge above or visit pj-sekhmet-ckltbh5thn6walldqpr2pt.streamlit.app
Click the Codespaces badge to open a fully-configured development environment. The Streamlit app will automatically launch on port 8501.
cd src/predictor
pip install -r ../../requirements.txt
streamlit run app.pyThe Cox model is calibrated to clinical literature including:
- 9 peer-reviewed sources from military and civilian populations
- 6 military-specific studies (UK and US)
- 2,235 total sample size across studies
- Documented risk factors with hazard ratios
| Source | Year | Focus | Military |
|---|---|---|---|
| Olivotto et al. | 2025 | MSKI prognostic factors | No |
| Marquina et al. | 2024 | ACL reconstruction meta-analysis | No |
| KCMHR Phase 4 | 2024 | UK military mental health | Yes |
| Anderson et al. | 2023 | Military academy epidemiology | Yes |
| Rhon et al. | 2022 | Spine rehabilitation | Yes |
| Antosh et al. | 2018 | ACL RTD in military | Yes |
| Wiggins et al. | 2016 | ACL reinjury rates | No |
| Hoge et al. | 2014 | Military PTSD | Yes |
| Shaw et al. | 2019 | Occupational LBP factors | No |
Full citations available in the References tab of the app.
| Factor | HR | Effect | Source |
|---|---|---|---|
| Age (per decade >25) | 1.15 | Delays recovery | Anderson 2023 |
| Prior same-region injury | 1.80 | Delays recovery | Wiggins 2016 |
| Smoking | 1.43 | Delays recovery | Anderson 2023 |
| BMI >= 30 | 1.20 | Delays recovery | Olivotto 2025 |
| OH Risk High | 1.30 | Delays recovery | Shaw 2019 |
| Supervised rehabilitation | 0.75 | Accelerates recovery | Olivotto 2025 |
| Type | Median Recovery | Evidence Grade |
|---|---|---|
| MSKI minor | 1-3 months | Moderate |
| MSKI moderate | 3-9 months | Moderate |
| MSKI major | 6-12 months | Moderate |
| MSKI severe | 12-24+ months | Low |
| Band | Range | Workforce Planning |
|---|---|---|
| Fast | 0-3 months | Short-term cover |
| Medium | 3-6 months | Medium-term adjustment |
| Slow | 6-12 months | Long-term planning |
| Complex | 12+ months | Permanent replacement |
Pj-SEKHMET/
├── src/
│ └── predictor/
│ ├── app.py # Streamlit UI (4 tabs)
│ ├── config.py # Enums, EvidenceBase loader
│ ├── cox_model.py # Cox PH survival model
│ ├── bayesian_model.py # Bayesian adjustable model
│ ├── xgb_model.py # XGBoost with SHAP
│ └── evidence_base.yaml # Clinical parameters
├── .devcontainer/ # GitHub Codespaces config
├── requirements.txt
└── README.md
Current Phase: Research & Demonstration
| Version | Feature | Status |
|---|---|---|
| V1 | Core Cox PH prediction model | Complete |
| V1 | Streamlit UI with configurable bands | Complete |
| V1 | Bayesian clinician-adjustable model | Complete |
| V2 | XGBoost model with SHAP explainability | Complete |
| V2 | Occupational Health risk factor | Complete |
| V3 | Traffic light RTD summary | Complete |
| V3 | Survival curve with 90% CI | Complete |
| V3 | Model agreement indicator | Complete |
| V3 | Comparator benchmark | Complete |
| V4 | References tab with full citations | Complete |
| V4 | BibTeX/text export | Complete |
| V5 | Regulatory disclaimers & governance | Complete |
| V5 | Riley framework compliance tracking | Complete |
| V5 | Calibration status & validation roadmap | Complete |
| - | GitHub Codespaces support | Complete |
| Feature | Status |
|---|---|
| CSV cohort upload | Planned |
| PDF report export | Planned |
| Real data validation | Pending data access |
Following the PROGRESS framework for clinical prediction model research (Riley et al., BMJ 2020):
| Criterion | Status | Notes |
|---|---|---|
| External Validation | Not done | No validation against real outcomes |
| Calibration Assessment | Pending | Requires outcome data |
| Discrimination (C-statistic) | Not calculated | Requires validation dataset |
| Clinical Utility | Not assessed | No decision curve analysis |
| Sample Size | Synthetic | XGBoost: n=5,000 simulated cases |
| Model Stability | Variable | Cox stable; XGBoost potentially unstable |
| Use Case | Appropriate? |
|---|---|
| Workforce capacity planning | Yes |
| Resource allocation scenarios | Yes |
| Research demonstrations | Yes |
| Individual clinical decisions | No |
| Fitness assessments | No |
- Obtain de-identified outcome data
- Perform calibration assessment (predicted vs observed)
- Calculate discrimination metrics (C-statistic)
- Conduct decision curve analysis for clinical utility
- External validation in independent cohort
- Clinical governance approval
- Riley RD et al. (2020). "Minimum sample size for developing a multivariable prediction model". BMJ. doi:10.1136/bmj.m441
- prognosisresearch.com - PROGRESS framework resources
pmsampsize- Sample size calculation tool
MIT
SEKHMET: Egyptian goddess of healing and war