Abstract S4-6: An international Ki67 reproducibility study
Bibliographic record
Abstract
Abstract Introduction: In breast cancer, immunohistochemical assessment of the cell proliferation marker Ki67 is of interest for potential use in clinical management (e.g. prognosis, prediction of treatment response, monitoring the effect of neoadjuvant therapy). However, lack of consistency across labs has limited the value of Ki67. A working group was assembled to devise a strategy to harmonize Ki67 analysis and scoring and to identify procedures to improve concordance (Dowsett et al. JNCI,'11). As a result, we have conducted an international Ki67 reproducibility study. Methods: Step 1: 100 breast cancer cases arranged into 1 mm core tissue microarrays (TMAs) were centrally stained using MIB-1 antibody. Eight labs scored Ki67 as the percentage of positively stained invasive tumor cells using their own local method. Six labs repeated scoring of 50 cases on the same TMA 3 times. Ki67 data were log2-transformed to approximate a normal distribution. Pairwise intra-lab and inter-lab concordance was displayed using Bland-Altman plots. Sources of variation (e.g. patient and lab) were analyzed using two-way crossed random effects models with quantification of reproducibility via intraclass correlation coefficient (ICC; range of 0–1, 1 = highest agreement). Step 2: A web-based scoring calibration interface was created to mitigate systematic differences in Ki67 interpretation observed in Step 1. Digital images of TMA cores from 9 “training” and 9 “test” cases representing a range of Ki67 values were assembled onto the web-based interface. An initial set of observers yielding consistent scores using a standardized scoring method served as reference labs. Detailed instructions for the standardized method were provided to 8 other labs, who were asked to score the cases and learn from discrepancies observed via the web tool. Statistical criteria for passing were pre-specified. Results: In Step 1, intra-lab reproducibility was high (ICC = 0.94). Inter-lab reproducibility was only moderate (ICC = 0.71). Absolute mean Ki67 values across the series ranged from 7% to 24%. Contributing to inter-lab discordance were tumor region selection, formal counting of nuclei versus visual estimation, and subjective assessment of staining positivity. Labs using formal counting methods gave more consistent results than those using visual estimation. The calibration exercise (Step 2) helped to improve consistency: all 8 labs passed the training exercise, 4 at first attempt and 4 after 1–2 rounds of retraining. When the same 8 labs scored the 9 test calibration cases, 3 passed on first attempt. Although overall concordance improved, discrepancies for cases with low Ki67 were largely responsible for the failure of the remaining 5 labs to pass the testing. Conclusions: Absolute values and cutpoints for Ki67, and associated clinical decisions, cannot be transferred between laboratories without careful standardization of scoring methodology. Our calibration training tool offers an initial process to improve inter-lab concordance, but further studies are required to establish what concordance can be achieved in practice. A large subsequent study will assess whether a pre-specified target of success (ICC = 0.9) can be achieved on glass slides after training with the web-based calibration tools developed as part of this study Citation Information: Cancer Res 2012;72(24 Suppl):Abstract nr S4-6.
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 | 0.049 | 0.047 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.003 | 0.002 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.002 | 0.001 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.006 | 0.002 |
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".