Table 5.
Post-test probabilities for the Random Forest ATTR-CM and cardiac amyloid Random Forest models based on model performance in the Northwestern Medicine Enterprise Data Warehouse Heart Failure Cohort.
| Model | Pre-test probability of ATTR-CMa | Random Forest model output cutoff for the diagnosis of ATTR-CM | LR+ | LR− | Post-test probability, LR+ | Post-test probability, LR− |
|---|---|---|---|---|---|---|
| Random Forest ATTR-CM model | 4% | >0.50 | 2.86 | 0.40 | 10.7% | 1.7% |
| 4% | >0.55 | 4.12 | 0.43 | 14.8% | 1.8% | |
| 4% | >0.60 | 5.85 | 0.52 | 19.7% | 2.1% | |
| 4% | >0.65 | 8.24 | 0.66 | 25.7% | 2.7% | |
| 4% | >0.70 | 11.07 | 0.79 | 31.7% | 3.2% | |
| 4% | >0.75 | 15.97 | 0.90 | 40.1% | 3.6% | |
| Random Forest cardiac amyloid model | 4% | >0.50 | 4.38 | 0.43 | 15.5% | 1.8% |
| 4% | >0.55 | 7.13 | 0.53 | 23.0% | 2.2% | |
| 4% | >0.60 | 12.37 | 0.66 | 34.2% | 2.7% | |
| 4% | >0.65 | 21.78 | 0.79 | 47.8% | 3.2% | |
| 4% | >0.70 | 39.37 | 0.89 | 62.3% | 3.6% | |
| 4% | >0.75 | 72.18 | 0.96 | 75.2% | 3.9% |
The random forest ATTR-CM model was derived using diagnosis codes specifically for wild-type ATTR-CM. The random forest cardiac amyloid model was derived using the more nonspecific umbrella diagnosis code for cardiac amyloidosis.
ATTR-CM amyloidogenic transthyretin cardiomyopathy, LR+ positive likelihood ratio, LR− negative likelihood ratio.
aPre-test probability was estimated to be 4% based on a prior publication (Kazi et al.20) that modeled the estimated prevalence of ATTR-CM in heart failure patients.