| Net Reclassification Across Clinical Thresholds | ||
| marker | direction | count |
|---|---|---|
| er | up | 8 |
| er | down | 13 |
| pr | up | 28 |
| pr | down | 43 |
| ki67 | up | 5 |
| ki67 | down | 152 |
| her2 | up | 47 |
| her2 | down | 18 |
16 Clinical Impact and Reclassification
16.1 Aim
Quantify how AI-driven changes cross treatment-relevant thresholds and how subtypes migrate.
Note for Pathologist: This document focuses on the consequences of score changes. It matters less if Ki67 changes from 50% to 55%, but it matters a lot if it changes from 15% to 25% (crossing the Luminal B threshold). We count exactly how many patients would see a “Class Change” based on the AI’s input.
16.2 Clinical Cutoffs
- ER/PR: 0% = Negative; 1–9% = Low Positive; ≥10% = Positive.
- Ki67: <30% = Low; ≥30% = High (also consider 20% in sensitivity runs).
- HER2: 0/1 = Negative; 2 = Equivocal; 3 = Positive.
16.3 Net Reclassification Counts
Note for Pathologist: The reclassification table shows how many cases moved up or down across clinical cutoffs after AI. “Up” means the category increased (e.g., ER Negative to Low Positive), and “down” means it decreased. The most critical reclassifications are those that cross treatment-relevant boundaries, such as ER going from Negative to Positive or Ki67 moving from Low to High.
16.4 HER2 0/1 vs 3 Changes
| HER2 Negative/Positive Crossovers | ||||
| case_id | pathologist | her2_pre | her2_post | comment |
|---|---|---|---|---|
16.5 Subtype Migration Matrix
| Molecular Subtype Transition Counts (Pre → Post) | ||
| molecular_subtype_pre | molecular_subtype_post | n |
|---|---|---|
| HER2 Positive | HER2 Positive | 275 |
| HER2 Positive | Hormone Weak Positive | 7 |
| HER2 Positive | Luminal A | 11 |
| HER2 Positive | Luminal B | 26 |
| HER2 Positive | Triple Negative | 5 |
| Hormone Weak Positive | HER2 Positive | 5 |
| Hormone Weak Positive | Hormone Weak Positive | 150 |
| Hormone Weak Positive | Luminal A | 11 |
| Hormone Weak Positive | Luminal B | 23 |
| Luminal A | HER2 Positive | 7 |
| Luminal A | Hormone Weak Positive | 13 |
| Luminal A | Luminal A | 285 |
| Luminal A | Luminal B | 82 |
| Luminal A | Triple Negative | 3 |
| Luminal B | HER2 Positive | 5 |
| Luminal B | Hormone Weak Positive | 1 |
| Luminal B | Luminal A | 3 |
| Luminal B | Luminal B | 137 |
| Triple Negative | Hormone Weak Positive | 2 |
| Triple Negative | Triple Negative | 133 |
Note for Pathologist: The subtype transition matrix shows the flow of cases from one molecular subtype to another after AI assistance. Diagonal counts represent stable classifications. Off-diagonal counts represent reclassifications, with the most concerning being transitions that change the treatment paradigm entirely (e.g., Luminal A to Triple Negative).
16.6 Variance Reduction Test
# A tibble: 3 × 3
marker median_delta p_value
<chr> <dbl> <dbl>
1 er -0.674 2.40e-16
2 ki67 0.331 6.87e- 2
3 pr -0.231 2.65e-1916.7 Reporting Plan
- Highlight counts of clinically meaningful upgrades/downgrades per marker.
- Call out HER2 negative/positive crossovers explicitly.
- Show subtype migration matrix and net directionality.
- Report Wilcoxon results on variance deltas to support claims of improved consensus.