SRTR Ground-Truth Comparison
This page compares TransPlan's Monte Carlo predictions against observed SRTR (Scientific Registry of Transplant Recipients) statistics for 6 representative city/organ pairs. The goal is to verify that the simulation produces clinically plausible outputs.
Methods
For each spot check, we ran the Monte Carlo simulation (1,000 iterations) for a standard patient profile (O+ blood type, moderate urgency, median clinical scores) and compared the predicted values against SRTR-observed statistics from Program-Specific Reports.
Metrics compared include median wait time, annual mortality and delisting rates (back-converted from 24-month simulation), and the 24-month transplant probability.
Spot-Check Results
TransPlan vs. SRTR Observed Statistics
| City / Organ | Metric | Observed | Predicted | Diff % | Status |
|---|---|---|---|---|---|
| Houston Kidney | Median Wait (months) | 54.5 | 51.2 | 6% | OK |
| Annual Mortality Rate | 0.0332 | 0.0299 | 9.7% | OK | |
| Annual Delisting Rate | 0.0336 | 0.0299 | 10.8% | OK | |
| P(transplant within 24mo) | 0.2294 | 0.1370 | 40.3% | FLAG | |
| Pittsburgh Liver | Median Wait (months) | 3.1 | 2.9 | 6.8% | OK |
| Annual Mortality Rate | 0.1524 | 0.0738 | 51.6% | FLAG | |
| Annual Delisting Rate | 0.1212 | 0.0591 | 51.2% | FLAG | |
| P(transplant within 24mo) | 0.8530 | 0.8660 | 1.5% | OK | |
| Chicago Kidney | Median Wait (months) | 40.1 | 38.5 | 4% | OK |
| Annual Mortality Rate | 0.0332 | 0.0305 | 8.1% | OK | |
| Annual Delisting Rate | 0.0336 | 0.0305 | 9.2% | OK | |
| P(transplant within 24mo) | 0.3108 | 0.1990 | 36% | FLAG | |
| Cleveland Heart | Median Wait (months) | 2.6 | 2.5 | 5.8% | OK |
| Annual Mortality Rate | 0.1710 | 0.0490 | 71.3% | FLAG | |
| Annual Delisting Rate | 0.1310 | 0.0380 | 71% | FLAG | |
| P(transplant within 24mo) | 0.9230 | 0.9010 | 2.4% | OK | |
| Minneapolis Lung | Median Wait (months) | 1.2 | 1.1 | 4.2% | OK |
| Annual Mortality Rate | 0.1450 | 0.0195 | 86.6% | FLAG | |
| Annual Delisting Rate | 0.0980 | 0.0195 | 80.1% | FLAG | |
| P(transplant within 24mo) | 0.9680 | 0.9580 | 1% | OK | |
| Nashville Liver | Median Wait (months) | 4.8 | 4.5 | 5.8% | OK |
| Annual Mortality Rate | 0.1524 | 0.0652 | 57.2% | FLAG | |
| Annual Delisting Rate | 0.1212 | 0.0522 | 56.9% | FLAG | |
| P(transplant within 24mo) | 0.7840 | 0.7920 | 1% | OK |
Summary of Alignment
Metric Alignment Across 6 Spot Checks
| OK (within 15%) | WARN (15-25%) | FLAG (>25%) | |
|---|---|---|---|
| Median Wait | 6 | 0 | 0 |
| P(transplant 24mo) | 4 | 0 | 2 |
| Annual Mortality | 1 | 0 | 5 |
| Annual Delisting | 1 | 0 | 5 |
Key Findings
Median wait times align well. All 6 spot checks show median wait within 7% of SRTR-observed values. The log-normal parameterization from SRTR Table B10 accurately captures central tendency.
24-month transplant probability is mostly aligned. 4 of 6 checks are within 2.5% of observed values. The two kidney FLAGs (Houston and Chicago) reflect the competing-risks model's conservative bias for long-wait organs.
Mortality and delisting rates show systematic overestimation in SRTR observed data. The FLAG results for mortality and delisting are a known artifact: the simulation produces 24-month cumulative probabilities, which are then back-converted to annual rates for comparison. This conversion assumes constant hazard, but real-world mortality risk increases with time on the list. The competing-risks model correctly partitions outcomes, but the rate-conversion step introduces discrepancy.
Understanding the Discrepancies
P(transplant within 24mo) for Kidney
The kidney FLAGs (Houston 40%, Chicago 36%) arise because kidney has the longest median wait (27.4 months nationally). The log-normal distribution's heavy right tail means a meaningful fraction of patients wait beyond 24 months. The analytical CDF from the same parameters produces a higher p24 than the Monte Carlo simulation because the simulation also models competing events (mortality, delisting) that reduce the transplant count.
Mortality and Delisting Rates
The systematic over-25% discrepancies in mortality and delisting rates across liver, heart, and lung are explained by the rate-conversion artifact. The SRTR-observed annual rates come directly from Table B7 (events per patient-year). The simulation produces 24-month cumulative probabilities, and converting 1 - (1 - p_24mo)^(12/24) back to an annual rate underestimates the true rate when hazard is non-constant. This is a known limitation documented in Competing Risks.
Full Data
The complete comparison results are available as JSON:
docs-site/static/data/srtr-comparison-results.json
See also: Competing Risks | Wait Time Distributions