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.

Significant inter- and intra-laboratory variation in grading of invasive breast cancer: A nationwide study of 33,043 patients in the Netherlands.

Accurate, consistent and reproducible grading by pathologists is of key-importance for identification of individual patients with invasive breast cancer (IBC) that will or will not benefit from adjuvant systemic treatment. We studied the laboratory-specific grading variation using nationwide real-life data to create insight and awareness in grading variation. Synoptic pathology reports of all IBC resection-specimens, obtained between 2013 and 2016, were retrieved from the nationwide Dutch Pathology Registry (PALGA). Absolute differences in laboratory-proportions of Grades I-III were compared to the national reference. Multivariable logistic regression provided laboratory-specific odds ratios (ORs) for high- vs. low-grade IBC. 33,792 IBC pathology reports of 33,043 patients from 39 laboratories were included, of which 28.1% were reported as Grade I (range between laboratories 16.3-43.3%), 47.6% as Grade II (38.4-57.8%), and 24.3% as Grade III (15.5-34.3%). Based on national guidelines, the indication for adjuvant chemotherapy was dependent on histologic grade in 29.9% of patients. After case-mix correction, 20 laboratories (51.3%) showed a significantly deviant OR. Significant grading differences were also observed among pathologists within laboratories. In this cohort of 33,043 breast cancer patients, we observed substantial inter- and intra-laboratory variation in histologic grading. It can be anticipated that this has influenced outcome including exposure to unnecessary toxicity, since choice of adjuvant chemotherapy was dependent on grade in nearly a third of patients. Better standardization and training seems warranted.

The effect of an e-learning module on grading variation of (pre)malignant breast lesions.

Histologic grade is a biomarker that is widely used to guide treatment of invasive breast cancer (IBC) and ductal carcinoma in situ of the breast (DCIS). Yet, currently, substantial grading variation between laboratories and pathologists exists in daily pathology practice. This study was conducted to evaluate whether an e-learning may be a feasible tool to decrease grading variation of (pre)malignant breast lesions. An e-learning module, representing the key-concepts of grading (pre)malignant breast lesions through gold standard digital images, was designed. Pathologists and residents could take part in either or both the separate modules on DCIS and IBC. Variation in grading of a digital set of lesions before and after the e-learning was compared in a fully-crossed study-design. Multiple outcome measures were assessed: inter-rater reliability (IRR) by Light's kappa, the number of images graded unanimously, the number of images with both extreme scores (i.e., grade I and grade III), and the average number of discrepancies from expert-consensus. Participants were included as they completed both the pre- and post-e-learning set (DCIS-module: n = 36, IBC-module: n = 21). For DCIS, all outcome measures improved after e-learning, with the IRR improving from fair (kappa: 0.532) to good (kappa: 0.657). For IBC, all outcome measures for the subcategories tubular differentiation and mitosis improved, with >90% of participants agreeing on almost 90% of the images after the e-learning. In contrast, the IRR for the subcategory of nuclear pleomorphism remained fair (kappa: 0.523 vs. kappa: 0.571). This study shows that an e-learning module, in which pathologists and residents are trained in histologic grading of DCIS and IBC, is a feasible and promising tool to decrease grading variation of (pre)malignant breast lesions. This is highly relevant given the important role of histologic grading in clinical decision making of (pre)malignant breast lesions.

Hormone- and HER2-receptor assessment in 33,046 breast cancer patients: a nationwide comparison of positivity rates between pathology laboratories in the Netherlands.

PURPOSE: Patient management of invasive breast cancer (IBC) is to a large extent based on hormone- and HER2-receptor assessment. High-quality, reliable receptor assessment is of key importance as false results may lead to under- or overtreatment of patients. Surveillance of case-mix adjusted positivity rates has been suggested as a tool to identify laboratories with insufficient testing assays, as this covers the whole process of receptor assessment and enables laboratories to benchmark their positivity rates against other laboratories. We studied laboratory-specific variation in hormone- and HER2 positivity rates of 33,046 breast cancer patients using real-life nationwide data. METHODS: All synoptic pathology reports of IBC resection-specimens, obtained between 2013 and 2016, were retrieved from the nationwide Dutch pathology registry (PALGA). Absolute and case-mix adjusted receptor positivity rates were compared to the mean national proportion and presented in funnel plots in separate analyses for estrogen (ER), progesterone (PR) and HER2. Case-mix adjustment was performed by multivariable logistic regression. RESULTS: 33,794 IBC lesions from 33,046 patients of 39 pathology laboratories were included. After case-mix adjustment, mean positivity rates were 87.2% for ER (range 80.4-94.3), 71.3% for PR (62.5-77.5%), and 9.9% for HER2 (5.5-12.7%). Overall, 14 (35.9%), 17 (43.6%) and 11 (28.2%) laboratories showed positivity rates outside the 95% confidence interval for ER, PR and HER2, respectively. CONCLUSION: This nationwide study shows that absolute variation in hormone- and HER2-receptor positivity rates between Dutch pathology laboratories is limited. Yet, the considerable number of outlying laboratories shows that there is still need for improvement. Continuous monitoring and benchmarking of positivity rates may help to realize this.

Significant inter- and intra-laboratory variation in grading of ductal carcinoma in situ of the breast: a nationwide study of 4901 patients in the Netherlands.

PURPOSE: A considerable part of ductal carcinoma in situ (DCIS) lesions may never progress into invasive breast cancer. However, standard treatment consists of surgical excision. Trials aim to identify a subgroup of low-risk DCIS patients that can safely forgo surgical treatment based on histologic grade, which highlights the importance of accurate grading. Using real-life nationwide data, we aimed to create insight and awareness in grading variation of DCIS in daily clinical practice. METHODS: All synoptic pathology reports of pure DCIS resection specimens between 2013 and 2016 were retrieved from PALGA, the nationwide Dutch Pathology Registry. Absolute differences in proportions of grade I-III were visualized using funnel plots. Multivariable analysis was performed by logistic regression to correct for case-mix, providing odds ratios and 95% confidence intervals for high-grade (III) versus low-grade (I-II) DCIS. RESULTS: 4952 DCIS reports from 36 laboratories were included, of which 12.5% were reported as grade I (range 6.1-24.4%), 39.5% as grade II (18.2-57.6%), and 48.0% as grade III (30.2-72.7%). After correction for case-mix, 14 laboratories (38.9%) reported a significantly lower (n = 4) or higher (n = 10) proportion of high-grade DCIS than the reference laboratory. Adjusted ORs (95%CI) ranged from 0.52 (0.31-0.87) to 3.83 (1.42-10.39). Significant grading differences were also observed among pathologists within laboratories. CONCLUSION: In this cohort of 4901 patients, we observed substantial inter- and intra-laboratory variation in DCIS grading, not explained by differences in case-mix. Therefore, there is an urgent need for nationwide standardization of grading practices, especially since the future management of DCIS may alter significantly depending on histologic grade.

[Breast cancer during pregnancy and the postpartum period: challenging diagnosis of an aggressive disease]. / Borstkanker tijdens en na de zwangerschap.

Breast cancer is the most common malignancy diagnosed during pregnancy. Breast cancer during pregnancy or within the first postpartum year is commonly described as Pregnancy-Associated Breast Cancer (PABC). PABC often exhibits poorer histopathologic features and has a worse prognosis when compared to young breast cancer patients without a current or recent pregnancy. Here, we describe two cases of PABC in which the presenting symptoms of the patients were interpreted as pregnancy-related changes, causing a diagnostic delay. Therefore, we argue that every pregnant or postpartum woman with changes in the breast must be thoroughly evaluated to exclude the possibility of a malignancy. In case of any suspicion, patients must be referred to a breast cancer center for further evaluation.

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.

Interlaboratory Gleason grading variation affects treatment: a Dutch historic cohort study in 30 509 patients with prostate cancer.

AIM: Substantial variation in Gleason grading (GG) of prostate cancer (PCa) exists between Dutch pathology laboratories. This study investigates its impact on treatment strategies. METHODS: Pathology reports of prostate needle biopsies and clinical data of patients with PCa diagnosed between 2017 and 2019 were retrieved from the Dutch nationwide network and registry of histopathology and cytopathology and The Netherlands Cancer Registry. We investigated the impact of grading variation on treatment strategy for patients whose grade was decisive in treatment choice. First, we evaluated the effect of grading practice (low, average or high grading) on active treatment (AT) versus active surveillance in patients with prostate-specific antigen (PSA) <10 ng/mL and cT1c/cT2a disease. Second, we assessed the association of grading practice with performance of pelvic lymph node dissection (PLND) in patients with PSA 10-20 ng/mL or cT2b disease. We used multivariable logistic regression to analyse the relation between laboratories' grading practices and AT or PLND. RESULTS: We included 30 509 patients. GG was decisive in treatment strategy for 11 925 patients (39%). AT was performed significantly less often in patients diagnosed by laboratories that graded lower than average (OR=0.77, 95% CI 0.68 to 0.88). Conversely, patients received AT significantly more often when diagnosed in high-grading laboratories versus average-grading laboratories (OR=1.21, 95% CI 1.03 to1.43). PLND was performed significantly less often in patients diagnosed by low-grading versus average-grading laboratories (OR=0.66, 95% CI 0.48 to 0.90). CONCLUSION: Our study shows that the odds that a patient undergoes AT or PLND, depends on laboratories' grading practices in a substantial number of patients. This likely influences patient prognosis and outcome, necessitating standardisation of GG to prevent suboptimal patient outcome.

Nationwide differences in cytology fixation and processing methods and their impact on interlaboratory variation in PD-L1 positivity.

Programmed death ligand-1 (PD-L1) immunostaining, which aids clinicians in decision-making on immunotherapy for non-small cell lung cancer (NSCLC) patients, is sometimes performed on cytological specimens. In this study, differences in cytology fixation and cell block (CB) processing between pathology laboratories were assessed, and the influence of these differences on interlaboratory variation in PD-L1 positivity was investigated. Questionnaires on cytology processing were sent to all Dutch laboratories. Information gathered from the responses was added to data on all Dutch NSCLC patients with a mention of PD-L1 testing in their cytopathology report from July 2017 to December 2018, retrieved from PALGA (the nationwide network and registry of histo- and cytopathology in the Netherlands). Case mix-adjusted PD-L1 positivity rates were determined for laboratories with known fixation and CB method. The influence of differences in cytology processing on interlaboratory variation in PD-L1 positivity was assessed by comparing positivity rates adjusted for differences in the variables fixative and CB method with positivity rates not adjusted for differences in these variables. Twenty-eight laboratories responded to the survey and reported 19 different combinations of fixation and CB method. Interlaboratory variation in PD-L1 positivity was assessed in 19 laboratories. Correcting for differences in the fixative and CB method resulted in a reduction (from eight (42.1%) to five (26.3%)) in the number of laboratories that differed significantly from the mean in PD-L1 positivity. Substantial variation in cytology fixation and CB processing methods was observed between Dutch pathology laboratories, which partially explains the existing considerable interlaboratory variation in PD-L1 positivity.

Pregnancy-associated breast cancer: the influence of gestational age.

Whether pregnancy-associated breast cancer (PABC) arise before or during pregnancy and whether this histopathology is affected by gestational age are currently unclear. The present study assesses the influence of gestational age and lactation on the histopathologic profile of PABC. We identified 744 patients with PABC (defined as breast cancer during pregnancy or 6 months following delivery). Histopathologic features were compared between pregnant and postpartum patients. We found that age at diagnosis was 34.2 years, and a majority of cancers were diagnosed during pregnancy (71.3%). Within pregnant patients, tumors were significantly more often estrogen receptor (ER)-negative in second and third trimesters (57.4%), as compared to first trimesters (41.9%) (P = 0.036). Similarly, a progesterone receptor (PR)-negative status was reported significantly less often within first trimesters (38.0%) compared to second and third trimesters (57.1%) (P = 0.032). For human EGF receptor 2 status, no significant differences were observed between gestational trimesters or lactating vs non-lactating patients. In postpartum patients, grade III tumors were found in over 80%, with high percentages of ER-negative tumors reaching 63% in those lactating vs 49% in non-lactating patients. This study demonstrates the varying histopathologic profile of PABC by gestational age and lactation status. Second- and third-trimester cancers display most typically the common ER/PR-negative phenotype, which is commonly reported in the literature. The increased ER-negative status and percentage grade III tumors in lactating vs non-lactating patients also suggest the presence of additional factors further diversifying histology. This indicates the need for clear definitions of PABC and the role of potential subgroups, which may provide a stepping stone for further in-depth research into PABC-carcinogenesis.

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.

Receptor status of breast cancer diagnosed during pregnancy: A literature review.

The definition of PABC is inconsistently given as either breast cancer diagnosed exclusively during pregnancy, or combined with breast cancer diagnosed within six months to five years after delivery, and sometimes even longer. The longer away from the delivery date breast cancer is diagnosed, the less clear this association with pregnancy may become. Therefore, breast cancer diagnosed during pregnancy (BCdP) may not necessarily be the same disease entity as PABC. This review aims to provide an overview of BCdP receptor status, as this has not been assessed before. BCdP tumors were predominantly ER negative (56.6 %), PR negative (57.2 %) or both ER and PR negative (47.9 %). Moreover, HER2-overexpression was seen in 33.2 % of BCdP patients and 27.6 % had triple negative disease. This predominantly ER and PR negative profile with more often HER2 overexpression is aggressive and distinct from non-pregnant similar-aged patients, warranting future comparative research.

Considerable interlaboratory variation in PD-L1 positivity in a nationwide cohort of non-small cell lung cancer patients.

OBJECTIVES: Immunohistochemical expression of programmed death-ligand 1 (PD-L1) is used as a predictive biomarker for prescription of immunotherapy to non-small cell lung cancer (NSCLC) patients. Accurate assessment of PD-L1 expression is therefore crucial. In this study, the extent of interlaboratory variation in PD-L1 positivity in the Netherlands was assessed, using real-world clinical pathology data. MATERIALS AND METHODS: Data on all NSCLC patients in the Netherlands with a mention of PD-L1 testing in their pathology report from July 2017 to December 2018 were extracted from PALGA, the nationwide network and registry of histo- and cytopathology in the Netherlands. PD-L1 positivity rates were determined for each laboratory that performed PD-L1 testing, with separate analyses for histological and cytological material. Two cutoffs (1% and 50%) were used to determine PD-L1 positivity. Differences between laboratories were assessed using funnel plots with 95% confidence limits around the overall mean. RESULTS: 6,354 patients from 30 laboratories were included in the analysis of histology data. At the 1% cutoff, maximum interlaboratory variation was 39.1% (32.7%-71.8%) and ten laboratories (33.3%) differed significantly from the mean. Using the 50% cutoff, four laboratories (13.3%) differed significantly from the mean and maximum variation was 23.1% (17.2%-40.3%). In the analysis of cytology data, 1,868 patients from 23 laboratories were included. Eight laboratories (34.8%) differed significantly from the mean in the analyses of both cutoffs. Maximum variation was 41.2% (32.2%-73.4%) and 29.2% (14.7%-43.9%) using the 1% and 50% cutoffs, respectively. CONCLUSION: Considerable interlaboratory variation in PD-L1 positivity was observed. Variation was largest using the 1% cutoff. At the 50% cutoff, analysis of cytology data demonstrated a higher degree of variation than the analysis of histology data.

Significant Inter- and Intralaboratory Variation in Gleason Grading of Prostate Cancer: A Nationwide Study of 35,258 Patients in The Netherlands.

PURPOSE: Our aim was to analyze grading variation between pathology laboratories and between pathologists within individual laboratories using nationwide real-life data. METHODS: We retrieved synoptic (n = 13,397) and narrative (n = 29,377) needle biopsy reports from the Dutch Pathology Registry and prostate-specific antigen values from The Netherlands Cancer Registration for prostate cancer patients diagnosed between January 2017 and December 2019. We determined laboratory-specific proportions per histologic grade and unadjusted odds ratios (ORs) for International Society of Urological Pathologists Grades 1 vs. 2-5 for 40 laboratories due to treatment implications for higher grades. Pathologist-specific proportions were determined for 21 laboratories that consented to this part of analysis. The synoptic reports of 21 laboratories were used for analysis of case-mix correction for PSA, age, year of diagnosis, number of biopsies and positive cores. RESULTS: A total of 38,321 reports of 35,258 patients were included. Grade 1 ranged between 19.7% and 44.3% per laboratory (national mean = 34.1%). Out of 40 laboratories, 22 (55%) reported a significantly deviant OR, ranging from 0.48 (95% confidence interval (CI) 0.39-0.59) to 1.54 (CI 1.22-1.93). Case-mix correction was performed for 10,294 reports, altering the status of 3/21 (14%) laboratories, but increasing the observed variation (20.8% vs. 17.7%). Within 15/21 (71%) of laboratories, significant inter-pathologist variation existed. CONCLUSION: Substantial variation in prostate cancer grading was observed between and within Dutch pathology laboratories. Case-mix correction did not explain the variation. Better standardization of prostate cancer grading is warranted to optimize and harmonize treatment.

Grading variation in 2,934 patients with ductal carcinoma in situ of the breast: the effect of laboratory- and pathologist-specific feedback reports.

BACKGROUND: Histologic grade of ductal carcinoma in situ of the breast (DCIS) may become the single biomarker that decides whether patients will be treated. Yet, evidence shows that grading variation in daily practice is substantial. To facilitate quality improvement, feedback reports, in which laboratory-specific case-mix adjusted proportions per grade were benchmarked against other laboratories, were sent to the individual laboratories by March 1, 2018. One year later, the effect of these feedback reports on inter-laboratory variation was studied. METHODS: Synoptic pathology reports of all pure DCIS resection specimens between March 1, 2017 and March 1, 2019 were retrieved from PALGA (the nationwide Dutch pathology registry). Laboratory-specific proportions per grade were compared to the overall proportion in the year before and after feedback. The absolute deviation for all three grades at once, represented by the overall deviation score (ODS), was calculated as the sum of deviations from the grade-specific overall proportions. Case-mix adjusted, laboratory-specific odds ratios (ORs) for high- (grade III) versus low-grade (grade I-II) DCIS were obtained by multivariable logistic regression. RESULTS: Overall, 2954 DCIS reports from 31 laboratories were included. After feedback, the range between laboratories decreased by 22 and 6.5% for grades II and III, while an increase of 6.2% was observed for grade I. Both the mean ODS (27.2 to 24.1%) and maximum ODS (87.7 to 59.6%) decreased considerably. However, the range of case-mix adjusted ORs remained fairly stable and substantial (0.39 (95% CI: 0.18-0.86) to 3.69 (95% CI: 1.30-10.51)). CONCLUSION: A promising decrease in grading variation was observed after laboratory-specific feedback for DCIS grades II-III, while this was not observed for DCIS grade I. Overall, grading variation remained substantial which needs to be addressed considering its clinical implications. Nationwide consensus on a classification, and training of (expert breast) pathologists, for example by e-learning, may help to further improve grading standardization.

Variation in breast cancer grading: the effect of creating awareness through laboratory-specific and pathologist-specific feedback reports in 16 734 patients with breast cancer.

AIMS: Histological grade is widely used to guide the management of invasive breast cancer (IBC). Yet, substantial interlaboratory and intralaboratory grading variations exist in daily pathology practice. To create awareness and to facilitate quality improvement, feedback reports, containing case-mix-adjusted laboratory-specific grades benchmarked against other laboratories, were sent to the individual laboratories by 1 March 2018. We studied the effect of these feedback reports on interlaboratory grading variation up till 1 year later. METHODS: Overall, 17 102 synoptic pathology reports of IBC resection specimens from 33 laboratories, obtained between 1 March 2017 and 1 March 2019 were retrieved from the Dutch Pathology Registry (PALGA). An overall deviation score (ODS), representing the sum of deviations from the grade-specific overall proportions, was calculated to compare the absolute deviation for all grades at once. Case-mix correction was performed by two multivariable logistic regression analyses, providing laboratory-specific ORs for high-grade versus low-grade IBC. RESULTS: After feedback, the overall range between laboratories decreased by 3.8%, 6.4% and 6.6% for grades I, II and III, respectively. Though the mean ODS remained similar (13.8% vs 13.7%), the maximum ODS decreased from 34.1% to 29.4%. The range of laboratory-specific ORs decreased by 21.9% for grade III versus grades I-II. CONCLUSIONS: An encouraging decrease in grading variation of IBC was observed after laboratory-specific feedback. Nevertheless, the overall grading variation remains substantial. In view of the important role of grading in patient management, it is adamant that not only feedback should be provided on a regular basis but also other interventions, such as additional training, are required.