ER continuous observations: 2368 Cases: 296 Pathologists: 4 Merged data rows: 1184 25 ER Low-Level Interpretation Analysis
26 ER Low-Level Interpretation Analysis
26.1 Background & Clinical Context
The estrogen receptor (ER) is one of the most critical biomarkers in breast cancer, directly determining endocrine therapy eligibility. While most breast cancers show either clearly negative (0%) or clearly positive (>10%) ER expression, the ER-low positive (1-9%) range represents a clinically controversial category.
26.1.1 ASCO/CAP 2020 Guidelines
The 2020 ASCO/CAP update reclassified ER expression into three categories:
| Category | ER% Range | Clinical Designation | Endocrine Therapy |
|---|---|---|---|
| Negative | <1% | ER-negative | Not indicated |
| Low Positive | 1-9% | ER-low positive | Consider (controversial) |
| Positive | >=10% | ER-positive | Standard of care |
Key clinical issue: The ER 1-9% (low positive) category:
- Shows limited benefit from endocrine therapy in most studies
- Behaves more like triple-negative breast cancer biologically
- ASCO/CAP recommends reporting as a separate category since 2020
- The 1% threshold (Negative vs any positivity) determines whether a patient is eligible for endocrine therapy
- The 10% threshold separates cases with clear benefit from those with uncertain benefit
Note for Pathologist: At low ER levels (0-10%), visual estimation is most subjective and interobserver variability highest. AI may systematically shift scores in this range. This analysis examines whether AI assistance improves or worsens agreement at these clinically critical thresholds.
26.2 Load and Prepare Data
26.3 ER Score Distribution (Pre-AI vs Post-AI)
26.3.1 Distribution Across Three Categories
| ER Category Distribution | ||||
| Pre-AI vs Post-AI Assessment (Negative <1%, Low 1-9%, Positive >=10%) | ||||
| Phase | ER Category1 | N | Total | Percentage |
|---|---|---|---|---|
| Pre-AI | Negative | 194 | 1175 | 16.5 |
| Pre-AI | Low | 23 | 1175 | 2.0 |
| Pre-AI | Positive | 958 | 1175 | 81.5 |
| Post-AI | Negative | 185 | 1175 | 15.7 |
| Post-AI | Low | 28 | 1175 | 2.4 |
| Post-AI | Positive | 962 | 1175 | 81.9 |
| 1 Low positive (1-9%) highlighted — controversial category per ASCO/CAP 2020 | ||||
26.3.2 Continuous Distribution with Threshold Lines
26.3.3 Zoomed View: ER 0-15% Range
26.3.4 Per-Pathologist Distribution
26.4 ER Category Transition Matrix
26.4.1 Overall 3x3 Transition Matrix (Pre-AI to Post-AI)
| ER Category Transition Matrix1 | |||
| Pre-AI (rows) to Post-AI (columns)1 | |||
| Pre-AI Category |
Post-AI Category
|
||
|---|---|---|---|
| Negative | Low | Positive | |
| Negative | 185 | 9 | 0 |
| Low | 0 | 15 | 8 |
| Positive | 0 | 4 | 954 |
| 1 Diagonal = consistent classification; off-diagonal = reclassification | |||
26.4.2 Transition Percentages
| ER Category Transition Matrix (Row Percentages) | |||
| Pre-AI (rows) to Post-AI (columns) — row % shown | |||
| Pre-AI Category |
Post-AI Category
|
||
|---|---|---|---|
| Negative | Low | Positive | |
| Negative | 185 (95.4%) | 9 (4.6%) | 0 (0.0%) |
| Low | 0 (0.0%) | 15 (65.2%) | 8 (34.8%) |
| Positive | 0 (0.0%) | 4 (0.4%) | 954 (99.6%) |
26.4.3 Clinically Relevant Transitions
| ER Category Transitions by Clinical Impact | ||
| Treatment-relevant reclassifications highlighted | ||
| Transition Type | N Cases | Percentage |
|---|---|---|
| No change | 1154 | 98.2 |
| Negative -> Low (Gained endocrine eligibility) | 9 | 0.8 |
| Low -> Positive (Clear endocrine benefit) | 8 | 0.7 |
| Positive -> Low (Uncertain benefit) | 4 | 0.3 |
Clinical Interpretation:
KEY FINDINGS - ER Category Transitions:
1% THRESHOLD CROSSINGS (Negative <-> Low):
- 9 cases (0.8%) Negative -> Low: GAINED endocrine therapy eligibility
- 0 cases (0%) Low -> Negative: LOST endocrine therapy eligibility
- Net: 9 additional cases eligible for endocrine therapy
10% THRESHOLD CROSSINGS (Low <-> Positive):
- 8 cases (0.7%) Low -> Positive: Moved to clear endocrine benefit
- 4 cases (0.3%) Positive -> Low: Moved to uncertain benefit range
MAJOR CHANGES (Negative <-> Positive):
- 0 cases (0%) Negative -> Positive
- 0 cases (0%) Positive -> Negative
Total reclassified: 21 out of 1175 (1.8%)26.5 Focus on Low-Level Cases (ER <10%)
26.5.1 Scatter Plot: Pre vs Post for Cases with ER <10%
26.5.2 Cases Crossing the 1% Threshold
| Cases Crossing the 1% Threshold | |||||
| ER-Negative <-> ER-Low Positive transitions | |||||
| Case ID | Pathologist | ER Pre-AI (%) | ER Post-AI (%) | Direction | Biopsy Type |
|---|---|---|---|---|---|
| 23175-25 | Pathologist 2 | 0.0 | 2.0 | Gained positivity (Neg->Low) | Excision |
| 21524-25 | Pathologist 2 | 0.0 | 1.0 | Gained positivity (Neg->Low) | Excision |
| 18689-25 | Pathologist 2 | 0.0 | 2.0 | Gained positivity (Neg->Low) | Tru-cut |
| 19713-25 | Pathologist 1 | 0.0 | 1.0 | Gained positivity (Neg->Low) | Tru-cut |
| 18689-25 | Pathologist 1 | 0.0 | 1.0 | Gained positivity (Neg->Low) | Tru-cut |
| 11985-25 | Pathologist 1 | 0.0 | 2.0 | Gained positivity (Neg->Low) | Excision |
| 19266-25 | Pathologist 3 | 0.0 | 8.0 | Gained positivity (Neg->Low) | Excision |
| 18961-25 | Pathologist 3 | 0.0 | 1.0 | Gained positivity (Neg->Low) | Excision |
| 14324-25 | Pathologist 3 | 0.0 | 2.0 | Gained positivity (Neg->Low) | Tru-cut |
26.5.3 Cases Crossing the 10% Threshold
| Cases Crossing the 10% Threshold | |||||
| ER-Low <-> ER-Positive transitions | |||||
| Case ID | Pathologist | ER Pre-AI (%) | ER Post-AI (%) | Direction | Biopsy Type |
|---|---|---|---|---|---|
| 14058-25 | Pathologist 2 | 5.0 | 10.0 | Low -> Positive | Tru-cut |
| 32963-25 | Pathologist 1 | 5.0 | 10.0 | Low -> Positive | Tru-cut |
| 14982-25 | Pathologist 1 | 5.0 | 16.0 | Low -> Positive | Excision |
| 14058-25 | Pathologist 1 | 5.0 | 10.0 | Low -> Positive | Tru-cut |
| 29841-25 | Pathologist 4 | 5.0 | 10.0 | Low -> Positive | Excision |
| 20256-25 | Pathologist 4 | 5.0 | 10.0 | Low -> Positive | Tru-cut |
| 29841-25 | Pathologist 3 | 5.0 | 10.0 | Low -> Positive | Excision |
| 14058-25 | Pathologist 3 | 5.0 | 10.0 | Low -> Positive | Tru-cut |
| 32973-25 | Pathologist 2 | 10.0 | 6.0 | Positive -> Low | Excision |
| 18961-25 | Pathologist 2 | 10.0 | 4.0 | Positive -> Low | Excision |
| 20256-25 | Pathologist 3 | 20.0 | 8.0 | Positive -> Low | Tru-cut |
| 20256-25 | Pathologist 2 | 40.0 | 8.0 | Positive -> Low | Tru-cut |
26.6 Interobserver Agreement for ER Categories
26.6.1 Fleiss’ Kappa: 3-Category Classification
| ER Category Agreement (3-category: Negative / Low / Positive) | |||||
| Fleiss' Kappa for all 4 pathologists | |||||
| Phase | Fleiss' Kappa | z-statistic | p-value | N Cases | Delta Kappa |
|---|---|---|---|---|---|
| Pre-AI | 0.938 | 43.201 | 0.00 | 295 | NA |
| Post-AI | 0.941 | 43.735 | 0.00 | 287 | 0.003 |
26.6.2 Binary Agreement: Negative vs Non-Negative (>=1%)
| Binary Agreement at Key ER Thresholds | |||||
| Fleiss' Kappa for binary classifications | |||||
| Threshold | Phase | Fleiss' Kappa | z | p-value | N Cases |
|---|---|---|---|---|---|
| 1% (Neg vs Non-Neg) | Pre-AI | 0.967 | 40.699 | 0.00 | 295 |
| 1% (Neg vs Non-Neg) | Post-AI | 0.945 | 39.208 | 0.00 | 287 |
| 10% (Low vs Positive) | Pre-AI | 0.660 | 25.308 | 0.00 | 245 |
| 10% (Low vs Positive) | Post-AI | 0.873 | 33.054 | 0.00 | 239 |
26.6.3 Pairwise Cohen’s Kappa per Pathologist Pair
| Pairwise Cohen's Kappa for ER Categories | |||
| 3-category (Negative/Low/Positive) per pathologist pair | |||
| Pathologist Pair | Pre-AI Kappa | Post-AI Kappa | Delta |
|---|---|---|---|
| Pathologist 2 vs Pathologist 1 | NA | NA | NA |
| Pathologist 2 vs Pathologist 4 | NA | NA | NA |
| Pathologist 2 vs Pathologist 3 | NA | NA | NA |
| Pathologist 1 vs Pathologist 4 | NA | NA | NA |
| Pathologist 1 vs Pathologist 3 | NA | NA | NA |
| Pathologist 4 vs Pathologist 3 | NA | NA | NA |
26.7 Confusion Matrix with Precision/Recall/F1
26.7.1 Aggregate Confusion Matrix (Pre-AI as Reference)
| ER Category Confusion Matrix: Pre-AI vs Post-AI | |||
| Pre-AI as reference, Post-AI as prediction | |||
| Reference (Pre-AI) |
Post-AI (Prediction)
|
||
|---|---|---|---|
| Negative | Low | Positive | |
| Negative | 185 | 9 | 0 |
| Low | 0 | 15 | 8 |
| Positive | 0 | 4 | 954 |
26.7.2 Precision, Recall, and F1 Scores
| Precision, Recall, and F1: ER 3-Category Classification | |||||||
| Pre-AI as reference standard | |||||||
| Category | TP | FP | FN | Precision | Recall | F1 | Accuracy |
|---|---|---|---|---|---|---|---|
| Negative | 185 | 0 | 9 | 1.000 | 0.954 | 0.976 | 0.982 |
| Low1 | 15 | 13 | 8 | 0.536 | 0.652 | 0.588 | 0.982 |
| Positive | 954 | 8 | 4 | 0.992 | 0.996 | 0.994 | 0.982 |
| 1 Low category (1-9%) highlighted — typically hardest to classify consistently | |||||||
26.7.3 Per-Pathologist Precision/Recall/F1
| Per-Pathologist Precision/Recall/F1 for ER Categories | |||||
| Pre-AI as reference, Post-AI as prediction | |||||
| Pathologist | Category | Precision | Recall | F1 | Accuracy |
|---|---|---|---|---|---|
| Pathologist 1 | Negative | 1.000 | 0.936 | 0.967 | 0.979 |
| Pathologist 1 | Low | 0.625 | 0.625 | 0.625 | 0.979 |
| Pathologist 1 | Positive | 0.988 | 1.000 | 0.994 | 0.979 |
| Pathologist 2 | Negative | 1.000 | 0.939 | 0.968 | 0.976 |
| Pathologist 2 | Low | 0.333 | 0.750 | 0.462 | 0.976 |
| Pathologist 2 | Positive | 0.996 | 0.988 | 0.992 | 0.976 |
| Pathologist 3 | Negative | 1.000 | 0.939 | 0.968 | 0.980 |
| Pathologist 3 | Low | 0.429 | 0.600 | 0.500 | 0.980 |
| Pathologist 3 | Positive | 0.992 | 0.996 | 0.994 | 0.980 |
| Pathologist 4 | Negative | 1.000 | 1.000 | 1.000 | 0.993 |
| Pathologist 4 | Low | 1.000 | 0.667 | 0.800 | 0.993 |
| Pathologist 4 | Positive | 0.992 | 1.000 | 0.996 | 0.993 |
26.7.4 Confusion Matrix Heatmap
26.8 Per-Pathologist Behavior at Low ER Levels
26.8.1 Score Changes at Low ER Levels
| Per-Pathologist ER Changes at Low Levels (<10%) | ||||||||||
| Direction and magnitude of AI-associated changes | ||||||||||
| Pathologist | N Low Cases | Mean Change | Median Change | N Upward | N Downward | N No Change | % Upward | % Downward | Crossed 1% | Crossed 10% |
|---|---|---|---|---|---|---|---|---|---|---|
| Pathologist 1 | 56 | 0.6 | 0.0 | 8 | 1 | 47 | 14.3 | 1.8 | NA | NA |
| Pathologist 2 | 56 | −0.6 | 0.0 | 5 | 4 | 47 | 8.9 | 7.1 | 3 | 4 |
| Pathologist 3 | 56 | 0.1 | 0.0 | 5 | 2 | 49 | 8.9 | 3.6 | NA | NA |
| Pathologist 4 | 55 | 0.2 | 0.0 | 2 | 0 | 53 | 3.6 | 0.0 | 0 | 2 |
26.8.2 Visualization of Per-Pathologist Shifts
26.8.3 Concordance Rates per Pathologist
| ER Category Concordance Rate per Pathologist | |||
| Pre-AI vs Post-AI agreement on 3-category classification | |||
| Pathologist | N Cases | N Concordant | Concordance (%) |
|---|---|---|---|
| Pathologist 1 | 292 | 286 | 97.9 |
| Pathologist 2 | 294 | 287 | 97.6 |
| Pathologist 3 | 293 | 287 | 98.0 |
| Pathologist 4 | 296 | 294 | 99.3 |
26.9 Biopsy Type Stratification
26.9.1 ER Category Transitions by Specimen Type
| ER Category Transitions by Biopsy Type | |||
| Does specimen type affect reclassification at low ER levels? | |||
| Biopsy Type | Low -> Pos | Neg -> Low | Pos -> Low |
|---|---|---|---|
| Excision | 3 | 5 | 2 |
| Tru-cut | 5 | 4 | 2 |
26.9.2 Reclassification Rate by Biopsy Type
| ER Reclassification Rate by Biopsy Type | |||||
| Overall and 1% threshold crossings | |||||
| Biopsy Type | N Total | N Reclassified | Reclass Rate (%) | Crossed 1% | % Crossed 1% |
|---|---|---|---|---|---|
| Excision | 692 | 10 | 1.4 | 5 | 0.7 |
| Tru-cut | 483 | 11 | 2.3 | 4 | 0.8 |
26.10 Impact on Molecular Subtype
26.10.1 ER-Driven Subtype Reclassifications
| ER Reclassification Impact on Molecular Subtype | |||
| How often does an ER category change cascade into subtype reclassification? | |||
| ER Transition | N ER Reclassified | N Subtype Changed | % Subtype Changed |
|---|---|---|---|
| Negative -> Low | 9 | 2 | 22.2 |
| Low -> Positive | 8 | 6 | 75.0 |
| Positive -> Low | 4 | 0 | 0.0 |
26.10.2 Triple Negative <-> Hormone Receptor Positive Transitions
| Triple-Negative <-> HR-Positive Transitions Driven by ER Change | |||||
| Cases where ER reclassification caused triple-negative status change | |||||
| Case ID | Pathologist | ER Pre-AI | ER Post-AI | Subtype Pre-AI | Subtype Post-AI |
|---|---|---|---|---|---|
| 21524-25 | Pathologist 2 | Negative | Low | Triple Negative | Hormone Weak Positive |
| 11985-25 | Pathologist 1 | Negative | Low | Triple Negative | Hormone Weak Positive |
26.10.3 Subtype Transition Summary Visualization
26.11 Clinical Implications & Summary
26.11.1 Key Findings
SUMMARY - ER Low-Level Interpretation Analysis
================================================
SAMPLE:
- Total paired assessments: 1175
- Reclassified: 21 (1.8%)
- Unchanged: 1154 (98.2%)
1% THRESHOLD (Negative <-> Low):
- Negative -> Low: 9 (gained endocrine eligibility)
- Low -> Negative: 0 (lost endocrine eligibility)
- Net change at 1%: 9 cases
10% THRESHOLD (Low <-> Positive):
- Low -> Positive: 8 (moved to clear benefit)
- Positive -> Low: 4 (moved to uncertain benefit)
AGREEMENT:
- 3-category Fleiss' Kappa: 0.938 (Pre-AI) -> 0.941 (Post-AI), Delta = 0.00326.11.2 Clinical Recommendations
1. ER 1% Threshold Monitoring
- Any AI-suggested change crossing the 1% threshold should trigger manual pathologist review
- The 1% threshold determines endocrine therapy eligibility
- Cases at 0-2% ER are inherently subjective and benefit from consensus scoring
2. ER Low Positive (1-9%) Reporting
- Report ER 1-9% as a distinct category per ASCO/CAP 2020 guidelines
- These cases have uncertain endocrine therapy benefit
- AI assistance may help standardize but does not eliminate subjectivity in this range
3. Quality Assurance
- Monitor reclassification rates at the 1% and 10% thresholds
- Consider double-reading for cases near clinical thresholds
- Track correlation between ER category changes and molecular subtype reclassification
4. Future Directions
- AI algorithms specifically calibrated for the 0-10% ER range may improve accuracy
- Validation against clinical outcomes (endocrine therapy response) needed
- Multi-institutional studies to assess generalizability of these findings
Analysis completed: 2026-02-10
ER low-level interpretation is clinically critical for endocrine therapy decisions
AI provides assistance but expert judgment remains essential at low ER levels