Abstract P1-08-01: Validation of automated Ki67 analysis to predict Oncotype DX recurrence score in early-stage breast cancer
Bibliographic record
Abstract
Abstract Background: Breast cancer prognosis and systemic treatment is mainly determined by its molecular subtype defined by the expression of driver genes including ER, PR, HER2 and Ki67. Several assays which include these key genes are currently available for refined prognostication and in some cases, prediction of chemotherapy benefit. In the most widely used assay, Oncotype DX®, Ki67 positivity is a heavily weighted contributor towards the calculation of the recurrence score (RS). However, reliable and reproducible scoring of Ki67 positive cells is challenged by the high inter- and intra-observer variability observed in the current manual scoring method. The objective of this study was to examine the correlation of an automated Ki67 analysis with Oncotype DX® RS and to develop a clinico-pathologic model for predicting Oncotype DX® RS category. Methods: A total of 606 hormone receptor-positive, HER2-negative and node negative breast cancers were analyzed (original cohort n=388; validation cohort n=218). Invasive breast cancers were annotated on hematoxylin and eosin slide by pathologist and the area was registered on Ki67 stained slides by HALO image analysis software (version 2.0.1145.14) where an algorithm was designed to identify Ki67-positive and negative nuclei. Tiles with the highest Ki67 percent scores were average to generate a Ki67 index. Pathologist manually scored subset of cases and the correlations of Ki67 index were tested by Pearson’s correlation coefficient. Correlations between Ki67, clinico-pathological variables and Oncotype DX® RS were assessed by nonparametric tests (Kruskal-Wallis; Wilcoxon rank and Fisher’s exact). A Random Forest machine learning model including clinico-pathological variables and Ki67 index was used to predict Oncotype DX® RS. All modelling was performed in R (version 3.6.0). The predicted RS obtained from the model were assigned to risk-of-recurrence groups (low <14, intermediate 14-28, high>28) and the performance was evaluated by a confusion matrix that matched Oncotype DX®-determined risk group assignment. The final model performance was based on the original cohort as the training set and validation cohort as the test set. Results: The Ki67 index calculated using our automated method showed high concordance with the manual scoring performed by a pathologist (original cohort r=0.9085; validation cohort r=0.912) and between operators (original cohort r=0.9835; validation cohort r=0.995). The Ki67 index correlated significantly with expected clinico-pathological variables, including tumor grade, nuclear grade, mitotic index, ER and PR. The Ki67 index strongly correlated with the Oncotype DX® RS and we were able to efficiently classify patients into the appropriate risk of recurrence categories. Furthermore, using the Ki67 index and thirteen clinical-pathological variables, we have developed a multivariate Random Forest machine learning model that accurately predicts RS with high accuracy (97%), sensitivity (98%) and specificity (80%). This model was further optimized to include only six variables (Ki67 index, ER total score, PR total score, tumor size, tumor grade, age at diagnosis) with an increased specificity and reduced false negatives while achieving a high accuracy (94.1%), sensitivity (91.3%) and specificity (100%). Conclusion: In summary, this automated Ki67 scoring method is reproducible, correlates strongly with the Oncotype DX® RS, and was incorporated into a clinic-pathologic model for prediction of Oncotype DX® RS with favorable performance characteristics. Citation Format: Angela MY Chan, Tasnima Abedin, Satbir Thakur, HaoCheng Li, Mie Konno, Emeka Enwere, Sasha Lupichuk, Nancy Nixon, Donald Morris, Hua Yang, Paola Neri. Validation of automated Ki67 analysis to predict Oncotype DX recurrence score in early-stage breast cancer [abstract]. In: Proceedings of the 2021 San Antonio Breast Cancer Symposium; 2021 Dec 7-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2022;82(4 Suppl):Abstract nr P1-08-01.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.003 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.000 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".