Pros and cons of artificial intelligence implementation in diagnostic pathology.

The rapid introduction of digital pathology has greatly facilitated development of artificial intelligence (AI) models in pathology that have shown great promise in assisting morphological diagnostics and quantitation of therapeutic targets. We are now at a tipping point where companies have started to bring algorithms to the market, and questions arise whether the pathology community is ready to implement AI in routine workflow. However, concerns also arise about the use of AI in pathology. This article reviews the pros and cons of introducing AI in diagnostic pathology.

Validation of digital microscopy: Review of validation methods and sources of bias.

Digital microscopy (DM) is increasingly replacing traditional light microscopy (LM) for performing routine diagnostic and research work in human and veterinary pathology. The DM workflow encompasses specimen preparation, whole-slide image acquisition, slide retrieval, and the workstation, each of which has the potential (depending on the technical parameters) to introduce limitations and artifacts into microscopic examination by pathologists. Performing validation studies according to guidelines established in human pathology ensures that the best-practice approaches for patient care are not deteriorated by implementing DM. Whereas current publications on validation studies suggest an overall high reliability of DM, each laboratory is encouraged to perform an individual validation study to ensure that the DM workflow performs as expected in the respective clinical or research environment. With the exception of validation guidelines developed by the College of American Pathologists in 2013 and its update in 2021, there is no current review of the application of methods fundamental to validation. We highlight that there is high methodological variation between published validation studies, each having advantages and limitations. The diagnostic concordance rate between DM and LM is the most relevant outcome measure, which is influenced (regardless of the viewing modality used) by different sources of bias including complexity of the cases examined, diagnostic experience of the study pathologists, and case recall. Here, we review 3 general study designs used for previous publications on DM validation as well as different approaches for avoiding bias.

Deep learning-based grading of ductal carcinoma in situ in breast histopathology images.

Ductal carcinoma in situ (DCIS) is a non-invasive breast cancer that can progress into invasive ductal carcinoma (IDC). Studies suggest DCIS is often overtreated since a considerable part of DCIS lesions may never progress into IDC. Lower grade lesions have a lower progression speed and risk, possibly allowing treatment de-escalation. However, studies show significant inter-observer variation in DCIS grading. Automated image analysis may provide an objective solution to address high subjectivity of DCIS grading by pathologists. In this study, we developed and evaluated a deep learning-based DCIS grading system. The system was developed using the consensus DCIS grade of three expert observers on a dataset of 1186 DCIS lesions from 59 patients. The inter-observer agreement, measured by quadratic weighted Cohen's kappa, was used to evaluate the system and compare its performance to that of expert observers. We present an analysis of the lesion-level and patient-level inter-observer agreement on an independent test set of 1001 lesions from 50 patients. The deep learning system (dl) achieved on average slightly higher inter-observer agreement to the three observers (o1, o2 and o3) (κo1,dl = 0.81, κo2,dl = 0.53 and κo3,dl = 0.40) than the observers amongst each other (κo1,o2 = 0.58, κo1,o3 = 0.50 and κo2,o3 = 0.42) at the lesion-level. At the patient-level, the deep learning system achieved similar agreement to the observers (κo1,dl = 0.77, κo2,dl = 0.75 and κo3,dl = 0.70) as the observers amongst each other (κo1,o2 = 0.77, κo1,o3 = 0.75 and κo2,o3 = 0.72). The deep learning system better reflected the grading spectrum of DCIS than two of the observers. In conclusion, we developed a deep learning-based DCIS grading system that achieved a performance similar to expert observers. To the best of our knowledge, this is the first automated system for the grading of DCIS that could assist pathologists by providing robust and reproducible second opinions on DCIS grade.

Being fully digital: perspective of a Dutch academic pathology laboratory.

The introduction of fast and robust whole slide scanners has facilitated the implementation of 'digital pathology' with various uses, the final challenge being full digital diagnostics. In this article, we describe the implementation process of a fully digital workflow for primary diagnostics in 2015 at the University Medical Centre in Utrecht, The Netherlands, as one of the first laboratories going fully digital with a future-proof complete digital archive. Furthermore, we evaluated the experience of the first 2 years of working with the system by pathologists and residents. The system was successfully implemented in 6 months, including a European tender procedure. Most pathologists and residents had high confidence in working fully digitally, the expertise areas lagging behind being paediatrics, haematopathology, and neuropathology. Reported limitations concerned recognition of microorganisms and mitoses. Neither the age of respondents nor the number of years of pathology experience was correlated with the confidence level regarding digital diagnostics. The ergonomics of digital diagnostics were better than those of traditional microscopy. In this article, we describe our experiences in implementing our fully digital primary diagnostics workflow, describing in depth the implementation steps undertaken, the interlocking components that are required for a fully functional digital pathology system (laboratory management, hospital information systems, data storage, and whole slide scanners), and the changes required in workflow and slide production.

Validation of a whole-slide image-based teleconsultation network.

AIMS: Most validation studies on digital pathology diagnostics have been performed in single institutes. Because rapid consultation on cases with extramural experts is one of the most important uses for digital pathology laboratory networks, the aim of this study was to validate a whole-slide image-based teleconsultation network between three independent laboratories. METHODS AND RESULTS: Each laboratory contributed 30 biopsies and/or excisions, totalling 90 specimens (776 slides) of varying difficulty and covering a wide variety of organs and subspecialties. All slides were scanned centrally at ×40 scanning magnification and uploaded, and subsequently assessed digitally by 16 pathologists using the same image management system and viewer. Each laboratory was excluded from digital assessment of their own cases. Concordance rates between the two diagnostic modalities (light microscopic versus digital) were compared. Loading speed of the images, zooming latency and focus quality were scored. Leaving out eight minor discrepancies without any clinical significance, the concordance rate between remote digital and original microscopic diagnoses was 97.8%. The two cases with a major discordance (for which the light microscopic diagnoses were deemed to be the better ones) resulted from a different interpretation of diagnostic criteria in one case and an image quality issue in the other case. Average scores for loading speed of the images, zooming latency and focus quality were 2.37 (on a scale up to 3), 2.39 (scale up to 3) and 3.06 (scale up to 4), respectively. CONCLUSIONS: This validation study demonstrates the suitability of a teleconsultation network for remote digital consultation using whole-slide images. Such networks may contribute to faster revision and consultation in pathology while maintaining diagnostic standards.

Domain generalization across tumor types, laboratories, and species - Insights from the 2022 edition of the Mitosis Domain Generalization Challenge.

Recognition of mitotic figures in histologic tumor specimens is highly relevant to patient outcome assessment. This task is challenging for algorithms and human experts alike, with deterioration of algorithmic performance under shifts in image representations. Considerable covariate shifts occur when assessment is performed on different tumor types, images are acquired using different digitization devices, or specimens are produced in different laboratories. This observation motivated the inception of the 2022 challenge on MItosis Domain Generalization (MIDOG 2022). The challenge provided annotated histologic tumor images from six different domains and evaluated the algorithmic approaches for mitotic figure detection provided by nine challenge participants on ten independent domains. Ground truth for mitotic figure detection was established in two ways: a three-expert majority vote and an independent, immunohistochemistry-assisted set of labels. This work represents an overview of the challenge tasks, the algorithmic strategies employed by the participants, and potential factors contributing to their success. With an F1 score of 0.764 for the top-performing team, we summarize that domain generalization across various tumor domains is possible with today's deep learning-based recognition pipelines. However, we also found that domain characteristics not present in the training set (feline as new species, spindle cell shape as new morphology and a new scanner) led to small but significant decreases in performance. When assessed against the immunohistochemistry-assisted reference standard, all methods resulted in reduced recall scores, with only minor changes in the order of participants in the ranking.

CONFIDENT-trial protocol: a pragmatic template for clinical implementation of artificial intelligence assistance in pathology.

INTRODUCTION: Artificial intelligence (AI) has been on the rise in the field of pathology. Despite promising results in retrospective studies, and several CE-IVD certified algorithms on the market, prospective clinical implementation studies of AI have yet to be performed, to the best of our knowledge. In this trial, we will explore the benefits of an AI-assisted pathology workflow, while maintaining diagnostic safety standards. METHODS AND ANALYSIS: This is a Standard Protocol Items: Recommendations for Interventional Trials-Artificial Intelligence compliant single-centre, controlled clinical trial, in a fully digital academic pathology laboratory. We will prospectively include prostate cancer patients who undergo prostate needle biopsies (CONFIDENT-P) and breast cancer patients who undergo a sentinel node procedure (CONFIDENT-B) in the University Medical Centre Utrecht. For both the CONFIDENT-B and CONFIDENT-P trials, the specific pathology specimens will be pseudo-randomised to be assessed by a pathologist with or without AI assistance in a pragmatic (bi-)weekly sequential design. In the intervention group, pathologists will assess whole slide images (WSI) of the standard hematoxylin and eosin (H&E)-stained sections assisted by the output of the algorithm. In the control group, pathologists will assess H&E WSI according to the current clinical workflow. If no tumour cells are identified or when the pathologist is in doubt, immunohistochemistry (IHC) staining will be performed. At least 80 patients in the CONFIDENT-P and 180 patients in the CONFIDENT-B trial will need to be enrolled to detect superiority, allocated as 1:1. Primary endpoint for both trials is the number of saved resources of IHC staining procedures for detecting tumour cells, since this will clarify tangible cost savings that will support the business case for AI. ETHICS AND DISSEMINATION: The ethics committee (MREC NedMec) waived the need of official ethical approval, since participants are not subjected to procedures nor are they required to follow rules. Results of both trials (CONFIDENT-B and CONFIDENT-P) will be published in scientific peer-reviewed journals.

Deep learning supported mitoses counting on whole slide images: A pilot study for validating breast cancer grading in the clinical workflow.

Introduction: Breast cancer (BC) prognosis is largely influenced by histopathological grade, assessed according to the Nottingham modification of Bloom-Richardson (BR). Mitotic count (MC) is a component of histopathological grading but is prone to subjectivity. This study investigated whether mitoses counting in BC using digital whole slide images (WSI) compares better to light microscopy (LM) when assisted by artificial intelligence (AI), and to which extent differences in digital MC (AI assisted or not) result in BR grade variations. Methods: Fifty BC patients with paired core biopsies and resections were randomly selected. Component scores for BR grade were extracted from pathology reports. MC was assessed using LM, WSI, and AI. Different modalities (LM-MC, WSI-MC, and AI-MC) were analyzed for correlation with scatterplots and linear regression, and for agreement in final BR with Cohen's κ. Results: MC modalities strongly correlated in both biopsies and resections: LM-MC and WSI-MC (R2 0.85 and 0.83, respectively), LM-MC and AI-MC (R2 0.85 and 0.95), and WSI-MC and AI-MC (R2 0.77 and 0.83). Agreement in BR between modalities was high in both biopsies and resections: LM-MC and WSI-MC (κ 0.93 and 0.83, respectively), LM-MC and AI-MC (κ 0.89 and 0.83), and WSI-MC and AI-MC (κ 0.96 and 0.73). Conclusion: This first validation study shows that WSI-MC may compare better to LM-MC when using AI. Agreement between BR grade based on the different mitoses counting modalities was high. These results suggest that mitoses counting on WSI can well be done, and validate the presented AI algorithm for pathologist supervised use in daily practice. Further research is required to advance our knowledge of AI-MC, but it appears at least non-inferior to LM-MC.

A comprehensive multi-domain dataset for mitotic figure detection.

The prognostic value of mitotic figures in tumor tissue is well-established for many tumor types and automating this task is of high research interest. However, especially deep learning-based methods face performance deterioration in the presence of domain shifts, which may arise from different tumor types, slide preparation and digitization devices. We introduce the MIDOG++ dataset, an extension of the MIDOG 2021 and 2022 challenge datasets. We provide region of interest images from 503 histological specimens of seven different tumor types with variable morphology with in total labels for 11,937 mitotic figures: breast carcinoma, lung carcinoma, lymphosarcoma, neuroendocrine tumor, cutaneous mast cell tumor, cutaneous melanoma, and (sub)cutaneous soft tissue sarcoma. The specimens were processed in several laboratories utilizing diverse scanners. We evaluated the extent of the domain shift by using state-of-the-art approaches, observing notable differences in single-domain training. In a leave-one-domain-out setting, generalizability improved considerably. This mitotic figure dataset is the first that incorporates a wide domain shift based on different tumor types, laboratories, whole slide image scanners, and species.

External validation and clinical utility assessment of PREDICT breast cancer prognostic model in young, systemic treatment-naïve women with node-negative breast cancer.

BACKGROUND: The validity of the PREDICT breast cancer prognostic model is unclear for young patients without adjuvant systemic treatment. This study aimed to validate PREDICT and assess its clinical utility in young women with node-negative breast cancer who did not receive systemic treatment. METHODS: We selected all women from the Netherlands Cancer Registry who were diagnosed with node-negative breast cancer under age 40 between 1989 and 2000, a period when adjuvant systemic treatment was not standard practice for women with node-negative disease. We evaluated the calibration and discrimination of PREDICT using the observed/expected (O/E) mortality ratio, and the area under the receiver operating characteristic curve (AUC), respectively. Additionally, we compared the potential clinical utility of PREDICT for selectively administering chemotherapy to the chemotherapy-to-all strategy using decision curve analysis at predefined thresholds. RESULTS: A total of 2264 women with a median age at diagnosis of 36 years were included. Of them, 71.2% had estrogen receptor (ER)-positive tumors and 44.0% had grade 3 tumors. Median tumor size was 16 mm. PREDICT v2.2 underestimated 10-year all-cause mortality by 33% in all women (O/E ratio:1.33, 95%CI:1.22-1.43). Model discrimination was moderate overall (AUC10-year:0.65, 95%CI:0.62-0.68), and poor for women with ER-negative tumors (AUC10-year:0.56, 95%CI:0.51-0.62). Compared to the chemotherapy-to-all strategy, PREDICT only showed a slightly higher net benefit in women with ER-positive tumors, but not in women with ER-negative tumors. CONCLUSIONS: PREDICT yields unreliable predictions for young women with node-negative breast cancer. Further model updates are needed before PREDICT can be routinely used in this patient subset.

Mitosis domain generalization in histopathology images - The MIDOG challenge.

The density of mitotic figures (MF) within tumor tissue is known to be highly correlated with tumor proliferation and thus is an important marker in tumor grading. Recognition of MF by pathologists is subject to a strong inter-rater bias, limiting its prognostic value. State-of-the-art deep learning methods can support experts but have been observed to strongly deteriorate when applied in a different clinical environment. The variability caused by using different whole slide scanners has been identified as one decisive component in the underlying domain shift. The goal of the MICCAI MIDOG 2021 challenge was the creation of scanner-agnostic MF detection algorithms. The challenge used a training set of 200 cases, split across four scanning systems. As test set, an additional 100 cases split across four scanning systems, including two previously unseen scanners, were provided. In this paper, we evaluate and compare the approaches that were submitted to the challenge and identify methodological factors contributing to better performance. The winning algorithm yielded an F1 score of 0.748 (CI95: 0.704-0.781), exceeding the performance of six experts on the same task.

Implementation of Artificial Intelligence in Diagnostic Practice as a Next Step after Going Digital: The UMC Utrecht Perspective.

Building on a growing number of pathology labs having a full digital infrastructure for pathology diagnostics, there is a growing interest in implementing artificial intelligence (AI) algorithms for diagnostic purposes. This article provides an overview of the current status of the digital pathology infrastructure at the University Medical Center Utrecht and our roadmap for implementing AI algorithms in the next few years.

Deep learning-based breast cancer grading and survival analysis on whole-slide histopathology images.

Breast cancer tumor grade is strongly associated with patient survival. In current clinical practice, pathologists assign tumor grade after visual analysis of tissue specimens. However, different studies show significant inter-observer variation in breast cancer grading. Computer-based breast cancer grading methods have been proposed but only work on specifically selected tissue areas and/or require labor-intensive annotations to be applied to new datasets. In this study, we trained and evaluated a deep learning-based breast cancer grading model that works on whole-slide histopathology images. The model was developed using whole-slide images from 706 young (< 40 years) invasive breast cancer patients with corresponding tumor grade (low/intermediate vs. high), and its constituents nuclear grade, tubule formation and mitotic rate. The performance of the model was evaluated using Cohen's kappa on an independent test set of 686 patients using annotations by expert pathologists as ground truth. The predicted low/intermediate (n = 327) and high (n = 359) grade groups were used to perform survival analysis. The deep learning system distinguished low/intermediate versus high tumor grade with a Cohen's Kappa of 0.59 (80% accuracy) compared to expert pathologists. In subsequent survival analysis the two groups predicted by the system were found to have a significantly different overall survival (OS) and disease/recurrence-free survival (DRFS/RFS) (p < 0.05). Univariate Cox hazard regression analysis showed statistically significant hazard ratios (p < 0.05). After adjusting for clinicopathologic features and stratifying for molecular subtype the hazard ratios showed a trend but lost statistical significance for all endpoints. In conclusion, we developed a deep learning-based model for automated grading of breast cancer on whole-slide images. The model distinguishes between low/intermediate and high grade tumors and finds a trend in the survival of the two predicted groups.

Prognostic Value of Stromal Tumor-Infiltrating Lymphocytes in Young, Node-Negative, Triple-Negative Breast Cancer Patients Who Did Not Receive (neo)Adjuvant Systemic Therapy.

PURPOSE: Triple-negative breast cancer (TNBC) is considered aggressive, and therefore, virtually all young patients with TNBC receive (neo)adjuvant chemotherapy. Increased stromal tumor-infiltrating lymphocytes (sTILs) have been associated with a favorable prognosis in TNBC. However, whether this association holds for patients who are node-negative (N0), young (< 40 years), and chemotherapy-naïve, and thus can be used for chemotherapy de-escalation strategies, is unknown. METHODS: We selected all patients with N0 TNBC diagnosed between 1989 and 2000 from a Dutch population-based registry. Patients were age < 40 years at diagnosis and had not received (neo)adjuvant systemic therapy, as was standard practice at the time. Formalin-fixed paraffin-embedded blocks were retrieved (PALGA: Dutch Pathology Registry), and a pathology review including sTILs was performed. Patients were categorized according to sTILs (< 30%, 30%-75%, and ≥ 75%). Multivariable Cox regression was performed for overall survival, with or without sTILs as a covariate. Cumulative incidence of distant metastasis or death was analyzed in a competing risk model, with second primary tumors as competing risk. RESULTS: sTILs were scored for 441 patients. High sTILs (≥ 75%; 21%) translated into an excellent prognosis with a 15-year cumulative incidence of a distant metastasis or death of only 2.1% (95% CI, 0 to 5.0), whereas low sTILs (< 30%; 52%) had an unfavorable prognosis with a 15-year cumulative incidence of a distant metastasis or death of 38.4% (32.1 to 44.6). In addition, every 10% increment of sTILs decreased the risk of death by 19% (adjusted hazard ratio: 0.81; 95% CI, 0.76 to 0.87), which are an independent predictor adding prognostic information to standard clinicopathologic variables (χ2 = 46.7, P < .001). CONCLUSION: Chemotherapy-naïve, young patients with N0 TNBC with high sTILs (≥ 75%) have an excellent long-term prognosis. Therefore, sTILs should be considered for prospective clinical trials investigating (neo)adjuvant chemotherapy de-escalation strategies.

Digital pathology in the time of corona.

The 2020 COVID-19 crisis has had and will have many implications for healthcare, including pathology. Rising number of infections create staffing shortages and other hospital departments might require pathology employees to fill more urgent positions. Furthermore, lockdown measures and social distancing cause many people to work from home. During this crisis, it became clearer than ever what an asset digital diagnostics is to keep pathologists, residents, molecular biologists and pathology assistants engaged in the diagnostic process, allowing social distancing and a 'need to be there' on-the-premises policy, while working effectively from home. This paper provides an overview of our way of working during the 2020 COVID-19 crisis with emphasis on the virtues of digital pathology.

Predicting breast tumor proliferation from whole-slide images: The TUPAC16 challenge.

Tumor proliferation is an important biomarker indicative of the prognosis of breast cancer patients. Assessment of tumor proliferation in a clinical setting is a highly subjective and labor-intensive task. Previous efforts to automate tumor proliferation assessment by image analysis only focused on mitosis detection in predefined tumor regions. However, in a real-world scenario, automatic mitosis detection should be performed in whole-slide images (WSIs) and an automatic method should be able to produce a tumor proliferation score given a WSI as input. To address this, we organized the TUmor Proliferation Assessment Challenge 2016 (TUPAC16) on prediction of tumor proliferation scores from WSIs. The challenge dataset consisted of 500 training and 321 testing breast cancer histopathology WSIs. In order to ensure fair and independent evaluation, only the ground truth for the training dataset was provided to the challenge participants. The first task of the challenge was to predict mitotic scores, i.e., to reproduce the manual method of assessing tumor proliferation by a pathologist. The second task was to predict the gene expression based PAM50 proliferation scores from the WSI. The best performing automatic method for the first task achieved a quadratic-weighted Cohen's kappa score of κ = 0.567, 95% CI [0.464, 0.671] between the predicted scores and the ground truth. For the second task, the predictions of the top method had a Spearman's correlation coefficient of r = 0.617, 95% CI [0.581 0.651] with the ground truth. This was the first comparison study that investigated tumor proliferation assessment from WSIs. The achieved results are promising given the difficulty of the tasks and weakly-labeled nature of the ground truth. However, further research is needed to improve the practical utility of image analysis methods for this task.

1399 H&E-stained sentinel lymph node sections of breast cancer patients: the CAMELYON dataset.

Background: The presence of lymph node metastases is one of the most important factors in breast cancer prognosis. The most common way to assess regional lymph node status is the sentinel lymph node procedure. The sentinel lymph node is the most likely lymph node to contain metastasized cancer cells and is excised, histopathologically processed, and examined by a pathologist. This tedious examination process is time-consuming and can lead to small metastases being missed. However, recent advances in whole-slide imaging and machine learning have opened an avenue for analysis of digitized lymph node sections with computer algorithms. For example, convolutional neural networks, a type of machine-learning algorithm, can be used to automatically detect cancer metastases in lymph nodes with high accuracy. To train machine-learning models, large, well-curated datasets are needed. Results: We released a dataset of 1,399 annotated whole-slide images (WSIs) of lymph nodes, both with and without metastases, in 3 terabytes of data in the context of the CAMELYON16 and CAMELYON17 Grand Challenges. Slides were collected from five medical centers to cover a broad range of image appearance and staining variations. Each WSI has a slide-level label indicating whether it contains no metastases, macro-metastases, micro-metastases, or isolated tumor cells. Furthermore, for 209 WSIs, detailed hand-drawn contours for all metastases are provided. Last, open-source software tools to visualize and interact with the data have been made available. Conclusions: A unique dataset of annotated, whole-slide digital histopathology images has been provided with high potential for re-use.

Aberrant hepatic lipid storage and metabolism in canine portosystemic shunts.

Non-alcoholic fatty liver disease (NAFLD) is a poorly understood multifactorial pandemic disorder. One of the hallmarks of NAFLD, hepatic steatosis, is a common feature in canine congenital portosystemic shunts. The aim of this study was to gain detailed insight into the pathogenesis of steatosis in this large animal model. Hepatic lipid accumulation, gene-expression analysis and HPLC-MS of neutral lipids and phospholipids in extrahepatic (EHPSS) and intrahepatic portosystemic shunts (IHPSS) was compared to healthy control dogs. Liver organoids of diseased dogs and healthy control dogs were incubated with palmitic- and oleic-acid, and lipid accumulation was quantified using LD540. In histological slides of shunt livers, a 12-fold increase of lipid content was detected compared to the control dogs (EHPSS P<0.01; IHPSS P = 0.042). Involvement of lipid-related genes to steatosis in portosystemic shunting was corroborated using gene-expression profiling. Lipid analysis demonstrated different triglyceride composition and a shift towards short chain and omega-3 fatty acids in shunt versus healthy dogs, with no difference in lipid species composition between shunt types. All organoids showed a similar increase in triacylglycerols after free fatty acids enrichment. This study demonstrates that steatosis is probably secondary to canine portosystemic shunts. Unravelling the pathogenesis of this hepatic steatosis might contribute to a better understanding of steatosis in NAFLD.

Long-term prognosis of young breast cancer patients (≤40 years) who did not receive adjuvant systemic treatment: protocol for the PARADIGM initiative cohort study.

INTRODUCTION: Currently used tools for breast cancer prognostication and prediction may not adequately reflect a young patient's prognosis or likely treatment benefit because they were not adequately validated in young patients. Since breast cancers diagnosed at a young age are considered prognostically unfavourable, many treatment guidelines recommend adjuvant systemic treatment for all young patients. Patients cured by locoregional treatment alone are, therefore, overtreated. Lack of prognosticators for young breast cancer patients represents an unmet medical need and has led to the initiation of the PAtients with bReAst cancer DIaGnosed preMenopausally (PARADIGM) initiative. Our aim is to reduce overtreatment of women diagnosed with breast cancer aged ≤40 years. METHODS AND ANALYSIS: All young, adjuvant systemic treatment naive breast cancer patients, who had no prior malignancy and were diagnosed between 1989 and 2000, were identified using the population based Netherlands Cancer Registry (n=3525). Archival tumour tissues were retrieved through linkage with the Dutch nationwide pathology registry. Tissue slides will be digitalised and placed on an online image database platform for clinicopathological revision by an international team of breast pathologists. Immunohistochemical subtype will be assessed using tissue microarrays. Tumour RNA will be isolated and subjected to next-generation sequencing. Differences in gene expression found between patients with a favourable and those with a less favourable prognosis will be used to establish a prognostic classifier, using the triple negative patients as proof of principle. ETHICS AND DISSEMINATION: Observational data from the Netherlands Cancer Registry and left over archival patient material are used. Therefore, the Dutch law on Research Involving Human Subjects Act (WMO) is not applicable. The PARADIGM study received a 'non-WMO' declaration from the Medical Ethics Committee of the Netherlands Cancer Institute - Antoni van Leeuwenhoek hospital, waiving individual patient consent. All data and material used are stored in a coded way. Study results will be presented at international (breast cancer) conferences and published in peer-reviewed, open-access journals.