Using Natural Language Processing to Improve Discrete Data Capture From Interpretive Cervical Biopsy Diagnoses at a Large Health Care Organization.

CONTEXT.­: The terminology used by pathologists to describe and grade dysplasia and premalignant changes of the cervical epithelium has evolved over time. Unfortunately, coexistence of different classification systems combined with nonstandardized interpretive text has created multiple layers of interpretive ambiguity. OBJECTIVE.­: To use natural language processing (NLP) to automate and expedite translation of interpretive text to a single most severe, and thus actionable, cervical intraepithelial neoplasia (CIN) diagnosis. DESIGN.­: We developed and applied NLP algorithms to 35 847 unstructured cervical pathology reports and assessed NLP performance in identifying the most severe diagnosis, compared to expert manual review. NLP performance was determined by calculating precision, recall, and F score. RESULTS.­: The NLP algorithms yielded a precision of 0.957, a recall of 0.925, and an F score of 0.94. Additionally, we estimated that the time to evaluate each monthly biopsy file was significantly reduced, from 30 hours to 0.5 hours. CONCLUSIONS.­: A set of validated NLP algorithms applied to pathology reports can rapidly and efficiently assign a discrete, actionable diagnosis using CIN classification to assist with clinical management of cervical pathology and disease. Moreover, discrete diagnostic data encoded as CIN terminology can enhance the efficiency of clinical research.

Accuracy and Efficiency of Deep-Learning-Based Automation of Dual Stain Cytology in Cervical Cancer Screening.

BACKGROUND: With the advent of primary human papillomavirus testing followed by cytology for cervical cancer screening, visual interpretation of cytology slides remains the last subjective analysis step and suffers from low sensitivity and reproducibility. METHODS: We developed a cloud-based whole-slide imaging platform with a deep-learning classifier for p16/Ki-67 dual-stained (DS) slides trained on biopsy-based gold standards. We compared it with conventional Pap and manual DS in 3 epidemiological studies of cervical and anal precancers from Kaiser Permanente Northern California and the University of Oklahoma comprising 4253 patients. All statistical tests were 2-sided. RESULTS: In independent validation at Kaiser Permanente Northern California, artificial intelligence (AI)-based DS had lower positivity than cytology (P < .001) and manual DS (P < .001) with equal sensitivity and substantially higher specificity compared with both Pap (P < .001) and manual DS (P < .001), respectively. Compared with Pap, AI-based DS reduced referral to colposcopy by one-third (41.9% vs 60.1%, P < .001). At a higher cutoff, AI-based DS had similar performance to high-grade squamous intraepithelial lesions cytology, indicating a risk high enough to allow for immediate treatment. The classifier was robust, showing comparable performance in 2 cytology systems and in anal cytology. CONCLUSIONS: Automated DS evaluation removes the remaining subjective component from cervical cancer screening and delivers consistent quality for providers and patients. Moving from Pap to automated DS substantially reduces the number of colposcopies and also achieves excellent performance in a simulated fully vaccinated population. Through cloud-based implementation, this approach is globally accessible. Our results demonstrate that AI not only provides automation and objectivity but also delivers a substantial benefit for women by reduction of unnecessary colposcopies.

Fine-Needle Aspiration Biopsy of Liver Lesions Yields Higher Tumor Fraction for Molecular Studies: A Direct Comparison With Concurrent Core Needle Biopsy.

BACKGROUND: This retrospective study evaluated and compared the diagnostic accuracy and suitability of tissue specimens for advanced molecular diagnostic testing obtained via 2 different techniques for percutaneous biopsy of primary and metastatic liver tumors. PATIENTS AND METHODS: Samples from 137 patients with liver masses who underwent concurrent fine-needle aspiration biopsy with cell block (FNAB-CB) and core needle biopsy (CNB) at 2 hospitals were assessed for diagnostic accuracy, tumor fraction, and tumor cellularity. A subset of FNAB-CBs, that were deemed to have less or equal tumor cellularity compared with CNBs, had level sections performed and were reassessed for tumor cellularity. RESULTS: Diagnostic accuracy was 96% for FNAB and 93% for CNB (P=.267). In FNAB-CBs, tumor fraction was significantly higher than in CNB samples (67% vs 36%; P<.0001), whereas nontumor components were significantly lower (stromal component, 7% vs 29%; P<.0001; background benign hepatocytes, 25% vs 36%; P=.003). Additionally, in 44% of cases, FNAB-CB tumor cellularity was equal to or greater than that of the concurrent CNB. CONCLUSIONS: In the current age of personalized medicine, a minimally invasive, safe approach to obtaining adequate tissue for myriad molecular testing is paramount. We have shown that FNAB sampling is diagnostically accurate and produces higher tumor fractions than CNB. Thus, FNAB should be strongly considered as an initial sampling modality, especially for patients in whom molecular tests will determine management.

Clinical Evaluation of Human Papillomavirus Screening With p16/Ki-67 Dual Stain Triage in a Large Organized Cervical Cancer Screening Program.

Importance: As cervical cancer screening transitions from Papanicolaou cytologic screening to primary human papillomavirus (HPV) testing worldwide, effective triage tests are needed to decide who among the HPV-positive women should receive further diagnostic evaluation to avoid unnecessary colposcopies and biopsies. Objective: To evaluate the performance of the p16/Ki-67 dual stain (DS) and HPV16/18 genotyping for the triage of HPV-positive women. Design, Setting, and Participants: A prospective observational study was conducted within the cervical cancer screening program at Kaiser Permanente Northern California of 3225 HPV-positive women undergoing HPV and Papanicolaou cytologic testing with a valid DS result from September 16 to October 31, 2015, with follow-up through December 31, 2018. Exposures: Human papillomavirus screening with partial genotyping and cytologic triage compared with DS triage. Main Outcomes and Measures: Cervical intraepithelial neoplasia grade 3 or more severe (CIN3+) and grade 2 or more severe (CIN2+), diagnosed within 3 years after sample collection. Results: A total of 3225 women (mean [SD] age, 37.9 [11.3] years) participated in the study. For triage of HPV-positive women with partial genotyping, DS showed better risk stratification for CIN3+ than did Papanicolaou cytologic testing, with women with positive DS results having a higher risk than women with positive Papanicolaou test results for CIN3+ (218 of 1818 [12.0%; 95% CI, 10.5%-13.5%] vs 219 of 2128 [10.3%; 95% CI, 9.0%-11.6%]; P = .005). Similarly, DS showed better risk stratification for CIN3+ compared with Papanicolaou cytologic testing in HPV-positive women, irrespective of genotyping. The greatest reassurance against CIN3+ was observed in HPV16/18-negative women with negative DS results, with a risk low enough to extend retesting intervals. Dual stain triage strategies required substantially fewer colposcopies per detection of CIN3+ compared with Papanicolaou cytologic testing, with a 32.1% (859 of 2677) reduction of colposcopies compared with the currently recommended triage strategy of HPV screening with Papanicolaou cytologic testing. Results for CIN2+ were very similar. Conclusions and Relevance: Triage of HPV-positive women with DS was superior to Papanicolaou cytologic testing in this study, demonstrating equal immediate detection of precancerous lesions and substantially reduced referral to colposcopy. These findings suggest that DS can safely replace Papanicolaou cytologic testing as a triage strategy for primary HPV screening, and that retesting intervals in HPV16/18-negative women with negative DS results can be safely extended to 3 years.

Faculty and medical student attitudes about preclinical classroom attendance.

BACKGROUND: Technological advances have diminished reliance on classroom attendance for mastering preclinical medical school course content, but nonattendance may have unintended consequence on the learning environment. Perceptions among educators and students regarding the value of attendance and implications of nonattendance have not been systematically studied. PURPOSES: The purpose of this study was to investigate differences in medical student and faculty attitudes regarding preclinical classroom attendance and the impact of nonattendance on educators and the learning environment. METHODS: Using Internet-based surveys, we assessed attitudes about preclinical classroom attendance among medical students and teaching faculty at Washington University School of Medicine. Our primary hypothesis was that students would be less likely than faculty to place societal value on attendance and relate it to professionalism. RESULTS: A total of 382 (79%) of 484 eligible students and 248 (64%) of 387 eligible faculty completed the survey. Both groups recognized a negative impact of poor attendance on faculty enthusiasm for teaching (students 83%, faculty 75%), but faculty were significantly more likely to endorse a negative impact on effectiveness of lectures (75% vs. 42%, p<.0001) and small-groups (92% vs. 76%, p<.0001) and a relationship between attendance and professionalism (88% vs. 68%, p<.0001). Students were significantly more likely to support free choice among learning opportunities (90% vs. 41%, p<.0001) including regularly missing class for research and community service activities (70% vs. 14%, p<.0001) and to consider lecture videos an adequate substitute for attendance (70% vs. 15%, p<.0001). Free-text responses suggested that students tended to view class-going primarily as a tool for learning factual material, whereas many faculty viewed it as serving important functions in the professional socialization process. CONCLUSIONS: In this single-center cohort, medical student and teaching faculty attitudes differed regarding the importance of classroom attendance and its relationship to professionalism, findings that were at least partially explained by differing expectations of the purpose of the preclinical classroom experience.

Pulmonary interstitial glycogenosis: an unrecognized etiology of persistent pulmonary hypertension of the newborn in congenital heart disease?

BACKGROUND: Pulmonary interstitial glycogenosis (PIG) arises from a developmental disorder of the pulmonary mesenchyme and presents clinically with reversible neonatal respiratory distress and/or persistent pulmonary hypertension of the newborn (PPHN). OBJECTIVE: We report two cases of PIG in patients with congenital heart disease (CHD) and evidence of PPHN. RESULTS: Both cases demonstrated the hallmark PIG histologic finding of diffuse, uniform interstitial thickening due to the presence of immature interstitial cells containing abundant cytoplasmic glycogen. CONCLUSIONS: We report the second and third patients with PIG associated with CHD. Because histologic examination is required to establish the diagnosis, we speculate that PIG, although rare, may be underrecognized in neonates presenting with PPHN in the setting of CHD.

Pathologic characteristics of pediatric intracranial pilocytic astrocytomas and their impact on outcome in 3 countries: a multi-institutional study.

Pilocytic astrocytoma (PA) is one of the most common glial neoplasms in the pediatric population, and its gross total resection can be curative. Treatment of partially resected or recurrent tumors is challenging, and the factors associated with increased recurrence risk are not well defined. Identification of favorable and unfavorable factors can lead to a better understanding and management of patients with PA. We studied the pathologic characteristics of 116 intracranial PAs from 4 institutions representing 3 distinct geographic regions to identify factors that may be associated with biological behavior. The study included 65 boys and 51 girls with a median age of 6 years. Median follow-up for all patients was 101 months, during which time 38 patients had recurrence. Progression-free and overall survivals were better in patients who underwent gross total resection. We were not able to identify any differences in pathologic and molecular markers among the 4 institutions from 3 different countries. However, progression-free survival varied significantly among institutions. Sox-2 was the most prevalent stem cell marker in PA, and many tumors showed synaptophysin positivity. BRAF immunostaining was not useful in determining BRAF duplication. BRAF duplication was more typical of posterior fossa tumors. There was a strong correlation between BRAF duplication and pERK immunostaining, suggesting that the RAF/MEK/ERK pathway is active in these tumors. This finding has significant implications given its role in oncogen-induced senescence and possible influence on treatment decisions of subtotally resected tumors.

Clinical stratification of glioblastoma based on alterations in retinoblastoma tumor suppressor protein (RB1) and association with the proneural subtype.

A recent study of CDK4/6 inhibitors in glioblastoma (GBM) xenografts identified retinoblastoma tumor suppressor protein RB1 status as a determinant of tumor therapeutic efficacy. Because of the need for clinically applicable RB1 testing, we assessed the utility of 2 complementary methods for determining RB1 status in GBM. Using fluorescence in situ hybridization (FISH) and immunohistochemistry (IHC), we analyzed 34 GBMs that had also undergone molecular characterization as part of The Cancer Genome Atlas (TCGA). By IHC, 4 tumors (11.8%) had complete loss of RB protein expression, including 2 with homozygous deletion of RB1 by FISH and 1 with hemizygous deletion of RB1 by FISH combined with a novel nonsense mutation in RB1. Consistent with these results, in an independent set of 51 GBMs tested by IHC, we demonstrated loss of RB1 protein in 5 (9.8%). In GBM molecular subtype analysis of TCGA data, complete loss of RB1 transcript expression was seen in 18 (10.6%) of 170 tumors, and these were highly enriched for, but not exclusive to, the proneural subtype (p < 0.01). These data support the use of IHC for determining RB1 status in clinical GBM specimens and suggest that RB1 alterations may be more common in certain GBM subgroups.

Cyclic AMP suppression is sufficient to induce gliomagenesis in a mouse model of neurofibromatosis-1.

Current models of oncogenesis incorporate the contributions of chronic inflammation and aging to the patterns of tumor formation. These oncogenic pathways, involving leukocytes and fibroblasts, are not readily applicable to brain tumors (glioma), and other mechanisms must account for microenvironmental influences on central nervous system tumorigenesis. Previous studies from our laboratories have used neurofibromatosis-1 (NF1) genetically engineered mouse (GEM) models to understand the spatial restriction of glioma formation to the optic pathway of young children. Based on our initial findings, we hypothesize that brain region-specific differences in cAMP levels account for the pattern of NF1 gliomagenesis. To provide evidence that low levels of cAMP promote glioma formation in NF1, we generated foci of decreased cAMP in brain regions where gliomas rarely form in children with NF1. Focal cAMP reduction was achieved by forced expression of phosphodiesterase 4A1 (PDE4A1) in the cortex of Nf1 GEM strains. Ectopic PDE4A1 expression produced hypercellular lesions with features of human NF1-associated glioma. Conversely, pharmacologic elevation of cAMP with the PDE4 inhibitor rolipram dramatically inhibited optic glioma growth and tumor size in Nf1 GEM in vivo. Together, these results indicate that low levels of cAMP in a susceptible Nf1 mouse strain are sufficient to promote gliomagenesis, and justify the implementation of cAMP-based stroma-targeted therapies for glioma.

Regulation of glutamate transport in developing rat oligodendrocytes.

Glutamate released from synaptic vesicles mediates excitatory neurotransmission by stimulating glutamate receptors. Glutamate transporters maintain low synaptic glutamate levels critical for this process, a role primarily attributed to astrocytes. Recently, vesicular release of glutamate from unmyelinated axons in the rat corpus callosum has been shown to elicit AMPA receptor-mediated currents in glial progenitor cells. Glutamate transporters are the only mechanism of glutamate clearance, yet very little is known about the role of glutamate transporters in normal development of oligodendrocytes (OLs) or in excitotoxic injury to OLs. We found that OLs in culture are capable of sodium-dependent glutamate uptake with a K(m) of 10 +/- 2 microm and a V(max) of 2.6, 5.0, and 3.8 nmol x min(-1) x mg(-1) for preoligodendrocytes, immature, and mature OLs, respectively. Surprisingly, EAAC1, thought to be exclusively a neuronal transporter, contributes more to [(3)H]l-glutamate uptake in OLs than GLT1 or GLAST. These data suggest that glutamate transporters on oligodendrocytes may serve a critical role in maintaining glutamate homeostasis at a time when unmyelinated callosal axons are engaging in glutamatergic signaling with glial progenitors. Furthermore, GLT1 was significantly increased in cultured mature OLs contrary to in vivo data in which we have shown that, although GLT1 is present on developing OLs when unmyelinated axons are prevalent in the developing rat corpus callosum, after myelination, GLT1 is not expressed on mature OLs. The absence of GLT1 in mature OLs in the rat corpus callosum and its presence in mature rat cultured OLs may indicate that a signaling process in vivo is not activated in vitro.

Targeted inhibition of cyclic AMP phosphodiesterase-4 promotes brain tumor regression.

PURPOSE: As favorable outcomes from malignant brain tumors remain limited by poor survival and treatment-related toxicity, novel approaches to cure are essential. Previously, we identified the cyclic AMP phosphodiesterase-4 (PDE4) inhibitor Rolipram as a potent antitumor agent. Here, we investigate the role of PDE4 in brain tumors and examine the utility of PDE4 as a therapeutic target. EXPERIMENTAL DESIGN: Immunohistochemistry was used to evaluate the expression pattern of a subfamily of PDE4, PDE4A, in multiple brain tumor types. To evaluate the effect of PDE4A on growth, a brain-specific isoform, PDE4A1 was overexpressed in xenografts of Daoy medulloblastoma and U87 glioblastoma cells. To determine therapeutic potential of PDE4 inhibition, Rolipram, temozolomide, and radiation were tested alone and in combination on mice bearing intracranial U87 xenografts. RESULTS: We found that PDE4A is expressed in medulloblastoma, glioblastoma, oligodendroglioma, ependymoma, and meningioma. Moreover, when PDE4A1 was overexpressed in Daoy medulloblastoma and U87 glioblastoma cells, in vivo doubling times were significantly shorter for PDE4A1-overexpressing xenografts compared with controls. In long-term survival and bioluminescence studies, Rolipram in combination with first-line therapy for malignant gliomas (temozolomide and conformal radiation therapy) enhanced the survival of mice bearing intracranial xenografts of U87 glioblastoma cells. Bioluminescence imaging indicated that whereas temozolomide and radiation therapy arrested intracranial tumor growth, the addition of Rolipram to this regimen resulted in tumor regression. CONCLUSIONS: This study shows that PDE4 is widely expressed in brain tumors and promotes their growth and that inhibition with Rolipram overcomes tumor resistance and mediates tumor regression.

Evaluation of an Interactive Case-based Online Network (ICON) in a problem based learning environment.

PURPOSE: This study sought to assess the introduction of a web-based innovation in medical education that complements traditional problem-based learning curricula. Utilizing the case method as its fundamental educational approach, the Interactive Case-based Online Network (ICON) allows students to interact with each other, faculty and a virtual patient in difficult neurological cases. Given the paucity of available metrics to benchmark online systems, we complement user perceptions with data on system utilization. METHODS: We describe a case study of distinct, small group tutorials over 2 years as part of the Human Nervous System and Behavior (HNSB) course at the Harvard Medical School. Participating students and faculty were interviewed following completion of the course and their utilization of the system was recorded and examined. RESULTS: Students each spent 3.2+/-1.3 h (mean+/-SD) through 8.6+/-2.8 accessions per week using ICON outside of required tutorial time. Faculty each spent 4.8+/-3.4 h through 16.6+/-8.9 accessions per week on ICON. Students identified real-time engagement, stronger relationships with faculty, increased accountability to the tutorial group and self-selected pace as the most beneficial characteristics of the ICON-based tutorial in comparison to traditional problem based learning (PBL) tutorials. Faculty identified enhanced collaboration with students and more realistic student experiences as the most beneficial characteristics. Both students and faculty reported that limitations of ICON included increased time investment for faculty and increased reliance on good faculty mentorship. CONCLUSION: This is the first study of the ICON learning system in undergraduate medical education, a platform designed to facilitate collaboration outside of the classroom. Data on user perceptions and system utilization suggest that both faculty and students chose to adopt this online learning system as a means for collaboration. The study also outlines future avenues for research in assessing novel online technologies.