There is a widespread belief in the Bayesian network (BN) community that while the overall accuracy of the results of BN inference is not sensitive to the precision of parameters, it is sensitive to the structure. We report on the results of a study focusing on the parameters in a companion paper, while this paper focuses on the BN graphical structure. We present the results of several experiments in which we test the impact of errors in the BN structure on its accuracy in the context of medical diagnostic models. We study the deterioration in model accuracy under structural changes that systematically modify the original gold standard model, notably the node and edge removal and edge reversal. Our results confirm the popular belief that the BN structure is important, and we show that structural errors may lead to a serious deterioration in the diagnostic accuracy. At the same time, most BN models are forgiving to single errors. In light of these results and the results of the companion paper, we recommend that knowledge engineers focus their efforts on obtaining a correct model structure and worry less about the overall precision of parameters.

There is a widely spread belief in the Bayesian network (BN) community that the overall accuracy of results of BN inference is not too sensitive to the precision of their parameters. We present the results of several experiments in which we put this belief to a test in the context of medical diagnostic models. We study the deterioration of accuracy under random symmetric noise but also biased noise that represents overconfidence and underconfidence of human experts.Our results demonstrate consistently, across all models studied, that while noise leads to deterioration of accuracy, small amounts of noise have minimal effect on the diagnostic accuracy of BN models. Overconfidence, common among human experts, appears to be safer than symmetric noise and much safer than underconfidence in terms of the resulting accuracy. Noise in medical laboratory results and disease nodes as well as in nodes forming the Markov blanket of the disease nodes has the largest effect on accuracy. In light of these results, knowledge engineers should moderately worry about the overall quality of the numerical parameters of BNs and direct their effort where it is most needed, as indicated by sensitivity analysis.

Objectives Since the publication of our study demonstrating high negative predictive values (>99% for women in their 40s) of benign-appearing endometrial cells (nEMCs), we have begun to include an educational comment in Papanicolaou (Pap) test reports with nEMCs that recommends routine periodic screening for asymptomatic premenopausal women (APW). The current study evaluated how the inclusion of this comment has affected clinical practice patterns at our institution. Methods The 2017 to 2019 database identified 175 reports containing the educational comment in women aged 45 to 54 years with a follow-up time of 11 to 37 months. Data, including age, menopause status, symptoms, imaging, and outcome, were collected. The procedure rate and the impact of clinical modifiers were assessed. Results Thirty-seven (20.6%) patients had biopsies within 6 months, which decreased from 48.1% as we previously reported. All nine (5%) APW with biopsies triggered only by nEMCs had benign histopathology. The remaining 28 biopsied patients had abnormal bleeding or a thickened endometrium, or they were postmenopausal, including a 53-year-old patient with complex atypical hyperplasia. None of the 138 patients with conservative follow-up developed atypical/malignant lesions. Conclusions A qualifying educational note included in Pap reports significantly reduced follow-up biopsies in APW. Optimal follow-up of nEMCs should be based on relevant clinical modifiers.

Background: Cervical screening could potentially be improved by better stratifying individual risk for the development of cervical cancer or precancer, possibly even allowing follow-up of individual patients differently than proposed under current guidelines that focus primarily on recent screening test results. We explore the use of a Bayesian decision science model to quantitatively stratify individual risk for the development of cervical squamous neoplasia. Materials and Methods: We previously developed a dynamic multivariate Bayesian network model that uses cervical screening and histopathologic data collected over 13 years in our system to quantitatively estimate the risk of individuals for the development of cervical precancer or invasive cervical cancer. The database includes 1,126,048 liquid-based cytology test results belonging to 389,929 women. From-the-vial, high risk human papilloma virus (HPV) test results and follow-up gynecological surgical procedures were available on 33.6% and 12% of these results (378,896 and 134,727), respectively. Results: Historical data impacted 5-year cumulative risk for both histopathologic cervical intraepithelial neoplasia 3 (CIN3) and squamous cell carcinoma (SCC) diagnoses. The risk was highest in patients with prior high grade squamous intraepithelial lesion cytology results. Persistent abnormal cervical screening test results, either cytologic or HPV results, were associated with variable increasing risk for squamous neoplasia. Risk also increased with prior histopathologic diagnoses of precancer, including CIN2, CIN3, and adenocarcinoma in situ. Conclusions: Bayesian modeling allows for individualized quantitative risk assessments of system patients for histopathologic diagnoses of significant cervical squamous neoplasia, including very rare outcomes such as SCC.

•Discuss ASCCP guideline.•CIN3 reliable surrogates for cervical cancer?•The Pittsburgh Cervical Cancer Screening Model.

© American Society for Clinical Pathology, 2020. This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0/), which permits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited. For commercial re-use, please contact journals.permissions@oup.com c o r r e s p o n d e n c e

Background As antibiotic resistance is becoming a major problem nowadays in a treatment of infections, bacteriophages (also known as phages) seem to be an alternative. However, to be used in a therapy, their life cycle should be strictly lytic. With the growing popularity of Next Generation Sequencing (NGS) technology, it is possible to gain such information from the genome sequence. A number of tools are available which help to define phage life cycle. However, there is still no unanimous way to deal with this problem, especially in the absence of well-defined open reading frames. To overcome this limitation, a new tool is definitely needed. Results We developed a novel tool, called PhageAI, that allows to access more than 10 000 publicly available bacteriophages and differentiate between their major types of life cycles: lytic and lysogenic. The tool included life cycle classifier which achieved 98.90% accuracy on a validation set and 97.18% average accuracy on a test set. We adopted nucleotide sequences embedding based on the Word2Vec with Ship-gram model and linear Support Vector Machine with 10-fold cross-validation for supervised classification. PhageAI is free of charge and it is available at . PhageAI is a REST web service and available as Python package. Conclusions Machine learning and Natural Language Processing allows to extract information from bacteriophages nucleotide sequences for lifecycle prediction tasks. The PhageAI tool classifies phages into either virulent or temperate with a higher accuracy than any existing methods and shares interactive 3D visualization to help interpreting model classification results. ### Competing Interest Statement In accordance with PhageAI - Bacteriophage Life Cycle Recognition with Machine Learning and Natural Language Processing policy, the authors are reporting that the PhageAI platform was developed by the authors and Proteon Pharmaceuticals is the owner of the platform. * AUC : Area Under Curve ML : Machine Learning MLP : Multi-layer Perceptron classifier NGS : Next Generation Sequencing NLP : Natural Language Processing RFECV : Feature ranking with recursive feature elimination and cross-validated selection of the best number of features ROC AUC : area under an ROC curve SVM : Support Vector Machine UMAP : Uniform Manifold Approximation and Projection

Objectives Given the recent debate challenging the contribution of cytology in cervical screening, we evaluated results of liquid-based cytology (LBC) and human papillomavirus (HPV) testing in cotesting preceding cervical cancer (CxCa) and precancer diagnoses in a national, heterogeneous population. Methods We assessed the results of cotesting, performed by Quest Diagnostics, in 13,633,071 women 30 years and older, tested 2010 to 2018. Cotest results preceding CxCa or precancer diagnoses were analyzed and stratified by histopathology. Results Among all screening results, 1,615 cotests preceded 1,259 CxCa diagnoses, and 11,164 cotests preceded 8,048 cervical precancer diagnoses. More women who were subsequently diagnosed with CxCa within 1 year were identified by the LBC result than by the HPV result (85.1%, 1,015/1,193 vs 77.5%, 925/1,193). Among all women with CxCa, the overall rate of nondetection was 13.1% (212/1,615) for cotesting results (LBC negative/HPV negative) and this rate increased substantially when testing exceeded 12 months compared to within 1 year prediagnosis of either CxCa or precancer. Conclusions Analysis of 9-year cotest results from a national reference laboratory confirms the value of LBC element in cotesting. This supports that LBC/HPV cotesting enhances screening for the identification of CxCa in women 30 years and older, more so than LBC or HPV alone within cotesting.

Introduction: Cervical screening has decreased the incidence of cervical carcinoma around the world primarily by preventing cervical squamous carcinoma, with significantly less measurable protective benefits in prevention of cervical adenocarcinoma. In this study, we apply Bayesian modeling of cervical clinical, screening, and biopsy data from a large integrated health system to explore the feasibility of calculating personalized risk assessments on screened system patients for subsequent histopathologic diagnoses of invasive cervical adenocarcinoma (AdCa) or cervical adenocarcinoma in situ (AIS). Materials and Methods: Diagnoses of cervical AIS or AdCa rendered between 2005 and 2018 were identified in our large health system database with 1,053,713 cytology results, 354,843 high-risk (hr) human papillomavirus (HPV) test results, and 99,012 cervical histopathologic results. Using our continuously updated Bayesian cervical cancer screening model which includes clinical data, cervical screening results, and cervical biopsy results, we projected quantitative estimates of patients’ 5-year cumulative risk for cervical AIS or AdCa. Results: 161 patients were identified with AIS (ages 17–75, mean 37 years), and 99 patients had diagnoses of cervical AdCa (ages 26–91, mean 48 years). Quantitative Bayesian 5-year cumulative risk projections for diagnoses of cervical AdCa or AIS in patients with different cervical screening test and biopsy histories were determined. The highest patient risk projections for subsequent cervical AdCa and/or AIS histopathologic diagnoses were associated with prior cervical screening test results of HPV-positive atypical glandular cells. Prior squamous cytologic abnormalities were associated with lower risk estimates. Prior histopathologic diagnoses of squamous abnormalities also influenced quantitative risk. A prior histopathologic diagnosis of AIS was associated with a very low risk of subsequent AdCa, consistent with effective excisional treatment. AdCa risk was greatest in women aged 30–65 years with prior CIN3 biopsy results, whereas AIS risk was greatest in women <30. Conclusion: Prevention of cervical AdCa in screened patients remains a major challenge for cervical screening. Individualized risk projections for cervical glandular neoplasia reflecting patient age, prior cervical screening test results, and prior cervical biopsy history are feasible using Bayesian modeling of health system data.

Uterine carcinosarcomas (UCS) are rare and highly aggressive tumors. Although it is currently accepted that the majority of UCS are metaplastic carcinomas, their aggressive behavior is unparalleled to that of any other high-grade endometrial neoplasms. Therefore, the search for the distinct immunohistochemical and molecular features that could help in the development of new treatment strategies continues. We evaluated the expression of PDL-1, growth hormone releasing hormone receptor, p53, WT1, PAX-8, estrogen receptor, HNF-1, and mismatch repair proteins in 43 UCS. Tumors were selected from the archives of the Magee-Womens Hospital University of Pittsburgh Medical Center Department of Pathology. Seventeen were stage I, 4 were stage II, 15 were stage III, and 7 were stage IV. The median age was 67yr and median overall survival was 3.2yr. Immunostaining for PAX8, HNF-1, and estrogen receptor showed statistically significant difference between epithelial and stromal components. Expression of p53 was significantly associated with clinical high stage, but other markers did not correlate with stage or survival. Immunostaining for programmed death ligand-1 was strongly positive in 30 UCS (70%), including 24 cases with tumor cell positivity, 12 cases with tumor cell and tumor-infiltrating immune cell positivity, and 6 cases with tumor-infiltrating immune cell positivity only. Of 27 tumors tested for mismatch repair expression, 12 (44%) showed loss of expression, 7 of which were PDL-1 positive. Growth hormone releasing hormone receptor was positive in 38 tumors (88%) and predominantly expressed in the epithelial component. The range of positivity for programmed death ligand-1 and growth hormone releasing hormone receptor suggests a possible potential adjuvant treatment that may be considered for UCS.

Purpose - The aim of this study was to use probabilistic graphical models to determine dental caries risk factors in three-year-old children. The analysis was conducted on the basis of the questionnaire data and resulted in building probabilistic graphical models to investigate dependencies among the features gathered in the surveys on dental caries. Materials and Methods - The data available in this analysis came from dental examinations conducted in children and from a questionnaire survey of their parents or guardians. The data represented 255 children aged between 36 and 48 months. Self-administered questionnaires contained 34 questions of socioeconomic and medical nature such as nutritional habits, wealth, or the level of education. The data included also the results of oral examination by a dentist. We applied the Bayesian network modeling to construct a model by learning it from the collected data. The process of Bayesian network model building was assisted by a dental expert. Results - The model allows to identify probabilistic relationships among the variables and to indicate the most significant risk factors of dental caries in three-year-old children. The Bayesian network model analysis illustrates that cleaning teeth and falling asleep with a bottle are the most significant risk factors of dental caries development in three-year-old children, whereas socioeconomic factors have no significant impact on the condition of teeth. Conclusions - Our analysis results suggest that dietary and oral hygiene habits have the most significant impact on the occurrence of dental caries in three-year-olds.

Biomarker analysis of metastatic breast carcinoma (MBC) is routinely recommended by ASCO/CAP guidelines, and establishing a diagnosis of MBC often requires immunohistochemistry (IHC). The reliability of breast tumor biomarkers and breast-specific markers on decalcified tissues has not been extensively studied. We performed IHC studies on breast tumors exposed to hydrochloric acid (HCl) and formic acid (FA) decalcification solutions, and HER2 fluorescence in situ hybridization on a subset of these tumors to establish a protocol for handling bone specimens with suspicion for MBC. Fifteen fresh cases of primary breast carcinoma and 8 HER2+ paraffin-embedded core biopsy cases were studied. Fresh tissue was divided into 5 fragments to approximate a bone core biopsy. One fragment (control) was fixed in 10% neutral buffered formalin. The remaining fragments were also exposed to FA or HCl decalcification for 1 or 5 hours. All fragments were embedded in 1 block and tested with an IHC panel. The known HER2+ cases were exposed to either 1 or 5 hours of FA, and HER2 fluorescence in situ hybridization was also performed. Results were interpreted as follows: H-scores for estrogen receptor, progesterone receptor, and GATA-3 were assigned from 0 to 300; HER2, cytokeratin 7, gross cystic disease fluid protein-15, Pax-8, TTF-1, cytokeratin 20, and mammaglobin were scored from 0 to 3+; and Ki67 from 0% to 100%. Mean scores were compared using the t test or Wilcoxon test for paired samples. No significant differences in mean score were seen between NF and 1 hour FA for any IHC immunoreactivity. After 5 hours of FA, only Ki67 average score was significantly less than NF. Mean scores for estrogen receptor, progesterone receptor, HER2, Ki67, and GATA-3 were significantly lower than NF in the tissue after either 1 or 5 hours of HCl. Mean scores for gross cystic disease fluid protein-15, mammaglobin, and cytokeratin 7 staining were not significantly lower than NF after 1 or 5 hours of HCl.

BACKGROUND:In the era of extensive data collection, there is a growing need for a large scale data analysis with tools that can handle many variables in one modeling framework. In this article, we present our recent applications of Bayesian network modeling to pathology informatics.METHODS:Bayesian networks (BNs) are probabilistic graphical models that represent domain knowledge and allow investigators to process this knowledge following sound rules of probability theory. BNs can be built based on expert opinion as well as learned from accumulating data sets. BN modeling is now recognized as a suitable approach for knowledge representation and reasoning under uncertainty. Over the last two decades BN have been successfully applied to many studies on medical prognosis and diagnosis.RESULTS:Based on data and expert knowledge, we have constructed several BN models to assess patient risk for subsequent specific histopathologic diagnoses and their related prognosis in gynecological cytopathology and breast pathology. These models include the Pittsburgh Cervical Cancer Screening Model assessing risk for histopathologic diagnoses of cervical precancer and cervical cancer, modeling of the significance of benign-appearing endometrial cells in Pap tests, diagnostic modeling to determine whether adenocarcinoma in tissue specimens is of endometrial or endocervical origin, and models to assess risk for recurrence of invasive breast carcinoma and ductal carcinoma in situ.CONCLUSIONS:Bayesian network models can be used as powerful and flexible risk assessment tools on large clinical datasets and can quantitatively identify variables that are of greatest significance in predicting specific histopathologic diagnoses and their related prognosis. Resulting BN models are able to provide individualized quantitative risk assessments and prognostication for specific abnormal findings commonly reported in gynecological cytopathology and breast pathology.

Biomarker analysis of invasive breast carcinoma is useful for prognosis, as surrogate for molecular subtypes of breast cancer, and prediction of response to adjuvant and neoadjuvant systemic therapies. Breast cancer intratumoral heterogeneity is incompletely studied. Comprehensive biomarker analysis of estrogen receptor (ER), progesterone receptor (PR), HER2, and Ki67 labeling index was performed on each tissue block of 100 entirely submitted breast tumors in 99 patients. Invasive carcinoma and in situ carcinoma was scored using semiquantitative histologic score (H-score) for ER and PR, HER2 expression from 0 to 3+, and percentage positive cells for Ki67. Core biopsy results were compared with surgical excision results, invasive carcinoma was compared with in situ carcinoma, and interblock tumoral heterogeneity was assessed using measures of dispersion (coefficient of variation and quartile coefficient of dispersion). Overall concordance between core biopsy and surgical excision was 99% for ER and 95% for PR. Mean histologic score of ER was significantly lower in invasive carcinoma between core biopsy and surgical excision (p = 0.000796). Intratumoral heterogeneity was higher for PR than for ER (mean coefficient of variation for ER 0.08 stdv 0.13 vs. PR 0.26 stdv 0.41). Ki67 labeling index was significantly higher in invasive carcinoma as compared with associated ductal carcinoma in situ on surgical resection specimen (p ≤ 0.0001). Ki67 hotspots were identified in 47% of cases. Of 52 HER2 negative cases on core biopsy, 10 were scored as equivocal on surgical resection. None (0/10) were amplified by Her-2/neu fluorescence in situ hybridization. Overall, biomarkers on core biopsy showed concordance with the surgical excision specimen in the vast majority of cases. Biomarker expression of in situ closely approximates associated invasive carcinoma. Intratumoral heterogeneity of PR is greater than ER. Biomarker expression on diagnostic core biopsy or single tumor block is representative of breast carcinoma as a whole in most cases and is appropriate for clinical decision-making.

Aims: Pathologists provide expert tissue assessment of breast cancer, yet their value to guide the appropriate use of breast cancer gene expression profile tests (GEPT) is underutilised. The specific aims of this study are to report morpho-immunohistological characteristics of breast tumours with Oncotype DX (R) (ODx) recurrence scores (RS) of 10 or fewer (ultra-low risk) and 25 or fewer (low risk) in order to determine if pathologists can identify prospectively patient tumours that do not require ODx testing. Methods and results: Oncotype DX (R) cases with RS < 10 from 2005 to 2010 comprised 441 of 2594 (17%) of clinical cases; this cohort had 5 years' follow-up and was treated with endocrine therapy alone. Tumours were analysed for tumour type, Nottingham grade, mitosis score (MS) semi-quantitative (H-score) hormone receptor content and Magee equation 3. Knowledge derived from this data set was used to develop algorithms in order to identify prospectively tumours with RS of 10 or fewer or 25 or fewer. Thirty-four per cent of tumours were low-grade special types, while the remainder were enriched with high hormone receptor content with MS of 1. These algorithmic selection criteria identified correctly all patient cases below the chemotherapy cut-point, i.e. RS < 25, indicating that these oncotype test orders were an unnecessary cost. Conclusions: This unique study demonstrates that (i) pathologists add great value to triage breast cancer for GEPT; and (ii) can identify prospectively low-grade tumour biology with high sensitivity and high specificity for those cases which do not require chemotherapy (RS < 25) using MS and hormone receptor content.

ObjectivesCervical screening strives to prevent cervical cancer (CxCa), minimizing morbidity and mortality. Most large US reports on cytology and human papillomavirus (HPV) cotesting of women aged 30 years and older are from one laboratory, which used conventional Papanicolaou (Pap) smears from 2003 to 2009.MethodsWe quantified detection of CxCa and precancer (cervical intraepithelial neoplasia 3/adenocarcinoma in situ [CIN3/AIS]) in 300,800 cotests at Magee Womens Hospital since 2005. Screening histories preceding CxCa and CIN3/AIS diagnoses were examined to assess the contribution of cytology and HPV testing. Cotesting utilized Food and Drug Administration-approved imaged liquid-based cytology (LBC) and from-the-vial HPV tests.ResultsLBC identified more women subsequently diagnosed with CxCa and CIN3/AIS than HPV testing. HPV-negative/cytology-positive results preceded 13.1% of CxCa and 7.2% of CIN3/AIS diagnoses.ConclusionsLBC enhanced cotesting detection of CxCa and CIN3/AIS to a greater extent than previously reported with conventional Pap smear and HPV cotesting.

Abstract Introduction The majority of publications regarding breast cancer GEPTs rarely supply detailed breast tumor histopathology in their outcome studies. As a result, the cost effective role of clinical risk assessment with histopathology of breast carcinomas tends to be minimized. The aims of this study are to characterize the details of breast tumor histopathology of patients with Oncotype Dx Recurrence Scores (RS) of 10 or less, and determine if Oncotype Dx offers value and clinical utility for patients with these low grade tumors Methods A total of 459 patients (18%) with Oncotype Dx RS of 10 or less were retrieved from a registry of 2558 patients with Oncotype Dx results. Patients had five years of follow-up with tumor registry and were treated with endocrine therapy alone. Tissue slides were available to review on 441/459 patients. Recorded details included (1) histopathologic type of carcinoma (2) mitotic score (MS), tubule formation, nuclear pleomorphism and Notttingham histologc (NG) grade. (3) Estrogen (ER) and progesterone (PgR) semiquantitated by Allred Score and Histologic Score (H Score: strong 200-300, moderate 100-199, weak <100). (4) Lymph node status. (5) overall survival and breast cancer specific survival. Results Patient ages were 33-92, with mean/median age of 60, and all had endocrine therapy alone. 148 of 441(34%) patients had carcinomas of “special types”, notable for low grade/good prognosis including tubular 22(15%), cribriform 15 (10.1%), papillary 17 (11.5%), and mucinous 28 (21%), along with 63 (42.5%) low grade classic lobular carcinomas and 3 (2%) low grade mixed ductal and lobular carcinomas. All 148 tumors had a MS of 1, were NG1 and had high ER HScores (280 median/263 mean) (Allred Scores 7-8) and high PR HScores (210 median/201 mean) (Allred Scores 6-8). The remaining 293 tumors were ductal carcinomas of no special type (NST), and 261/293 (89%) of these had a MS of 1/NG2. Of the remaining cases, 10 (3%) had a MS of 2/NG2, 18 (6%) had MS of 2/NG3 and four (1%) were MS3/NG3. Estrogen receptor H Score/Allred Score was strong (Allred Score 7-8) in 395/441 (89.6%), moderate in 45 (10.2%) and weak in 1 patient (0.2%). Progesterone HScores were strong in (Allred Score 6-8) 269/441 (76%) and moderate in the remainder. Strong and moderate ER comprised 99.8% of tumors. Thus, tumors with MS1, and NG1, all with ER HScore >200 (Allred Score of 7-8) were enriched in the RS <10, and these features distinguished this group from other tumors with a MS1. At 5 years, 433 patients (98%) were alive, 8 were dead, 1 from breast cancer due to distant recurrence. The 5-year breast cancer specific survival for this group was 99.7%. [95%CI 98.5-99.9.] 87 cases were accrued in the ongoing prospective study to date. There were 15/87 (17%) cases, 95% of which were correctly identified by pathologists as having an RS <10 using the criteria defined, with sensitivity 95%, specificity 86%, PPV 63% 95% CI(49.76-75.08), NPV 99% 95% CI(90.7-99.78). No patient had a recurrence score >22. Conclusions Pathologists can identify these low risk tumors with high accuracy. Oncotype Dx lacks clinical value and utility in this setting. Citation Format: Dabbs DJ, Serdy K, Onisko A, Clark BZ, Bhargava R, Smalley S, Perkins S, Brufsky AM. The clinical utility of oncotype Dx for patients with recurrence scores of 10 or less: A value based pathology study of tumor histopathology and outcomes analysis in an integrated delivery and finance health system [abstract]. In: Proceedings of the 2017 San Antonio Breast Cancer Symposium; 2017 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2018;78(4 Suppl):Abstract nr P4-08-04.