Using machine learning to parse breast pathology reports.

PURPOSE: Extracting information from electronic medical record is a time-consuming and expensive process when done manually. Rule-based and machine learning techniques are two approaches to solving this problem. In this study, we trained a machine learning model on pathology reports to extract pertinent tumor characteristics, which enabled us to create a large database of attribute searchable pathology reports. This database can be used to identify cohorts of patients with characteristics of interest. METHODS: We collected a total of 91,505 breast pathology reports from three Partners hospitals: Massachusetts General Hospital, Brigham and Women's Hospital, and Newton-Wellesley Hospital, covering the period from 1978 to 2016. We trained our system with annotations from two datasets, consisting of 6295 and 10,841 manually annotated reports. The system extracts 20 separate categories of information, including atypia types and various tumor characteristics such as receptors. We also report a learning curve analysis to show how much annotation our model needs to perform reasonably. RESULTS: The model accuracy was tested on 500 reports that did not overlap with the training set. The model achieved accuracy of 90% for correctly parsing all carcinoma and atypia categories for a given patient. The average accuracy for individual categories was 97%. Using this classifier, we created a database of 91,505 parsed pathology reports. CONCLUSIONS: Our learning curve analysis shows that the model can achieve reasonable results even when trained on a few annotations. We developed a user-friendly interface to the database that allows physicians to easily identify patients with target characteristics and export the matching cohort. This model has the potential to reduce the effort required for analyzing large amounts of data from medical records, and to minimize the cost and time required to glean scientific insight from these data.

Perceptions of pathology informatics by non-informaticist pathologists and trainees.

BACKGROUND: Although pathology informatics (PI) is essential to modern pathology practice, the field is often poorly understood. Pathologists who have received little to no exposure to informatics, either in training or in practice, may not recognize the roles that informatics serves in pathology. The purpose of this study was to characterize perceptions of PI by noninformatics-oriented pathologists and to do so at two large centers with differing informatics environments. METHODS: Pathology trainees and staff at Cleveland Clinic (CC) and Massachusetts General Hospital (MGH) were surveyed. At MGH, pathology department leadership has promoted a pervasive informatics presence through practice, training, and research. At CC, PI efforts focus on production systems that serve a multi-site integrated health system and a reference laboratory, and on the development of applications oriented to department operations. The survey assessed perceived definition of PI, interest in PI, and perceived utility of PI. RESULTS: The survey was completed by 107 noninformatics-oriented pathologists and trainees. A majority viewed informatics positively. Except among MGH trainees, confusion of PI with information technology (IT) and help desk services was prominent, even in those who indicated they understood informatics. Attendings and trainees indicated desire to learn more about PI. While most acknowledged that having some level of PI knowledge would be professionally useful and advantageous, only a minority plan to utilize it. CONCLUSIONS: Informatics is viewed positively by the majority of noninformatics pathologists at two large centers with differing informatics orientations. Differences in departmental informatics culture can be attributed to the varying perceptions of PI by different individuals. Incorrect perceptions exist, such as conflating PI with IT and help desk services, even among those who claim to understand PI. Further efforts by the PI community could address such misperceptions, which could help enable a better understanding of what PI is and is not, and potentially lead to increased acceptance by non-informaticist pathologists.

Spinal cord glioblastoma: 25years of experience from a single institution.

Accounting for less than 0.2% of all glioblastomas, high grade gliomas of the spinal cord are very rare. Here, we discuss our approach to managing patients with high grade spinal cord glioma and review the literature on the subject. Six patients with high grade spinal cord gliomas who presented to our institution between 1990 and 2015 were reviewed. Each patient underwent subtotal surgical resection, with a subset receiving adjuvant chemotherapy and radiation. Our primary outcomes of interest were pre-operative and post-operative functional status. One year survival rate was 100%. All patients had stable or improved American Spine Injury Association score immediately after surgery, which was maintained at 3months in 83.3% of patients. Karnofsky Performance Status (KPS) was stable at 3month follow up in 50% of patients, but all had decreased KPS 1year after surgery. A subset of patients received post-operative radiation and chemotherapy with 0% tumor recurrence rate at 3months. We assessed the molecular profiles of tumors from two patients in our series and found that each had mutations in TP53, but had wildtype BRAF, IDH-1, and MGMT. Taken together, our data show that patients with high grade spinal cord gliomas have an excellent survival at 1year, but with some decline in functional status within this period. Further studies are needed to elucidate the natural history of the disease and to explore the role of adjuvant targeted molecular therapies.

Surgical excision of radial scars diagnosed by core biopsy may help predict future risk of breast cancer.

Radial scars (RS's) are benign breast lesions known to be associated with carcinomas and other high-risk lesions (HRL's). The upgrade rate to carcinoma after core biopsy revealing RS is 0-40 %. We sought to determine the outcomes of RS with and without HRL diagnosed by core biopsy. Patients who underwent core biopsy revealing RS without carcinoma at our institution between 1/1996 and 11/2012 were identified from a surgical pathology database. Retrospective chart review was utilized to classify patients as RS-no HRL or RS-HRL. HRL was defined as ADH, LCIS, and/or ALH. We determined upgrade rate to carcinoma at surgical excision, and upgrade to HRL for RS-no HRL patients. Univariate analysis was performed to identify risk factors for upgrade in RS-no HRL patients. 156 patients underwent core biopsy revealing RS, 131 RS-no HRL (84 %), and 25 RS-HRL (16 %). The overall rate of upgrade to invasive carcinoma was 0.8 % (1/124). 1.0 % (1/102) of RS-no HRL and 13.6 % (3/22) of RS-HRL patients were upgraded to DCIS (P = 0.0023). The upgrade of RS-no HRL to HRL at excision was 21.6 % (22/102). By univariate analysis, RS-no HRL with radiologic appearance of a mass/architectural distortion had a significantly higher rate of upgrade to HRL or carcinoma compared with calcifications (P = 0.03). Excision of RS to rule out associated invasive carcinoma is not warranted, given a <1 % rate of upgrade at excision. However, excision to evaluate for non-invasive cancer or HRL may be considered to help guide clinical decision-making about use of chemoprevention.

Cutaneous and mammary apocrine carcinomas have different immunoprofiles.

Often the distinction of cutaneous apocrine carcinoma from metastatic mammary apocrine carcinoma to the skin can be a diagnostic dilemma because both tumors share similar histologic features and have overlapping immunohistochemical profile. We compared the expression of adipophilin, cytokeratin 5/6, p63, GATA3, mammaglobin, androgen receptor, estrogen receptor, progesterone receptor, and HER2 by immunohistochemistry in 14 cutaneous apocrine carcinomas (11 primary tumors, 3 metastases) and 26 primary apocrine carcinomas of the breast. Whereas focal adipophilin staining was seen in 36% (5/14) of cutaneous apocrine carcinoma, strong and diffuse adipophilin staining was seen in 88% (22/25) of mammary apocrine carcinoma (P = .0013). Differences in estrogen receptor and progesterone receptor expression were also statistically significant (P = .018 and .043). Androgen receptor was strongly positive in all cutaneous and mammary cases. Although there was no significant difference in the frequency of expression of cytokeratin 5/6, p63, HER2, GATA3, and mammaglobin in cutaneous apocrine carcinoma versus mammary apocrine carcinoma, strong and diffuse cytokeratin 5/6 and/or mammaglobin expression were seen only in cutaneous apocrine carcinoma. In conclusion, cutaneous apocrine carcinoma is likely adipophilin- ER+ PR+/- HER2- and can exhibit strong and diffuse cytokeratin 5/6 and/or mammaglobin expression. On the contrary, a mammary apocrine carcinoma is likely adipophilin+ ER- PR- and often exhibit 3+ HER2 with corresponding HER2 gene amplification. A panel of adipophilin, ER, PR, HER2, cytokeratin 5/6, and mammaglobin may be helpful in distinguishing cutaneous apocrine carcinoma from mammary apocrine carcinoma.

The feasibility of using natural language processing to extract clinical information from breast pathology reports.

OBJECTIVE: The opportunity to integrate clinical decision support systems into clinical practice is limited due to the lack of structured, machine readable data in the current format of the electronic health record. Natural language processing has been designed to convert free text into machine readable data. The aim of the current study was to ascertain the feasibility of using natural language processing to extract clinical information from >76,000 breast pathology reports. APPROACH AND PROCEDURE: Breast pathology reports from three institutions were analyzed using natural language processing software (Clearforest, Waltham, MA) to extract information on a variety of pathologic diagnoses of interest. Data tables were created from the extracted information according to date of surgery, side of surgery, and medical record number. The variety of ways in which each diagnosis could be represented was recorded, as a means of demonstrating the complexity of machine interpretation of free text. RESULTS: There was widespread variation in how pathologists reported common pathologic diagnoses. We report, for example, 124 ways of saying invasive ductal carcinoma and 95 ways of saying invasive lobular carcinoma. There were >4000 ways of saying invasive ductal carcinoma was not present. Natural language processor sensitivity and specificity were 99.1% and 96.5% when compared to expert human coders. CONCLUSION: We have demonstrated how a large body of free text medical information such as seen in breast pathology reports, can be converted to a machine readable format using natural language processing, and described the inherent complexities of the task.

The pathology informatics curriculum wiki: Harnessing the power of user-generated content.

BACKGROUND: The need for informatics training as part of pathology training has never been so critical, but pathology informatics is a wide and complex field and very few programs currently have the resources to provide comprehensive educational pathology informatics experiences to their residents. In this article, we present the "pathology informatics curriculum wiki", an open, on-line wiki that indexes the pathology informatics content in a larger public wiki, Wikipedia, (and other online content) and organizes it into educational modules based on the 2003 standard curriculum approved by the Association for Pathology Informatics (API). METHODS AND RESULTS: In addition to implementing the curriculum wiki at http://pathinformatics.wikispaces.com, we have evaluated pathology informatics content in Wikipedia. Of the 199 non-duplicate terms in the API curriculum, 90% have at least one associated Wikipedia article. Furthermore, evaluation of articles on a five-point Likert scale showed high scores for comprehensiveness (4.05), quality (4.08), currency (4.18), and utility for the beginner (3.85) and advanced (3.93) learners. These results are compelling and support the thesis that Wikipedia articles can be used as the foundation for a basic curriculum in pathology informatics. CONCLUSIONS: The pathology informatics community now has the infrastructure needed to collaboratively and openly create, maintain and distribute the pathology informatics content worldwide (Wikipedia) and also the environment (the curriculum wiki) to draw upon its own resources to index and organize this content as a sustainable basic pathology informatics educational resource. The remaining challenges are numerous, but largest by far will be to convince the pathologists to take the time and effort required to build pathology informatics content in Wikipedia and to index and organize this content for education in the curriculum wiki.

Health of U.S. Navy submarine crew during periods of isolation.

BACKGROUND: An essential element in planning for long-term space missions is prediction of the medical support required. Medical data for analogous populations serving in isolated and/or contained environments are useful in predicting health risks for astronauts. METHODS: This study evaluated the rates of health events that occurred among a highly screened, healthy military population during periods of isolation using a centralized database of medical encounter records from U.S. Navy submarines. The study population was composed of U.S. Navy officers and enlisted men deployed on 240 submarine patrols between 1 January 1997 and 30 September 2000. RESULTS: A total of 1389 officers and 11,952 enlisted crew members served aboard participating submarines for 215,086 and 1,955,521 person-days at sea, respectively, during the study period. Officers had 214 initial visits to medical staff with 79 re-visits for the same condition during these patrols, while enlisted men had 3345 initial visits and 1549 re-visits. Among officers, the most common category of medical events was respiratory illnesses (primarily upper respiratory infections), followed by injury, musculoskeletal conditions, infectious diseases, symptoms and ill-defined conditions, and skin problems. Among enlisted men, the most common category of medical events was injury, followed by respiratory illnesses (upper respiratory infections), skin problems, symptoms and ill-defined conditions, digestive disorders, infectious conditions, sensory organ problems (ear infections and eye problems), and musculoskeletal conditions. CONCLUSIONS: Potential mission-impacting medical events reported were rare, i.e., among a crew of seven officers, only one medical event would be expected to occur during a 6-mo mission and result in 3/4 d or less of limited or no duty. Among a crew of seven enlisted men, about two medical events would be expected during a 6-mo mission and result in about 1 d of limited or no duty per medical event.