Introduction to Artificial Intelligence and Machine Learning for Pathology.

CONTEXT.­: Recent developments in machine learning have stimulated intense interest in software that may augment or replace human experts. Machine learning may impact pathology practice by offering new capabilities in analysis, interpretation, and outcomes prediction using images and other data. The principles of operation and management of machine learning systems are unfamiliar to pathologists, who anticipate a need for additional education to be effective as expert users and managers of the new tools. OBJECTIVE.­: To provide a background on machine learning for practicing pathologists, including an overview of algorithms, model development, and performance evaluation; to examine the current status of machine learning in pathology and consider possible roles and requirements for pathologists in local deployment and management of machine learning systems; and to highlight existing challenges and gaps in deployment methodology and regulation. DATA SOURCES.­: Sources include the biomedical and engineering literature, white papers from professional organizations, government reports, electronic resources, and authors' experience in machine learning. References were chosen when possible for accessibility to practicing pathologists without specialized training in mathematics, statistics, or software development. CONCLUSIONS.­: Machine learning offers an array of techniques that in recent published results show substantial promise. Data suggest that human experts working with machine learning tools outperform humans or machines separately, but the optimal form for this combination in pathology has not been established. Significant questions related to the generalizability of machine learning systems, local site verification, and performance monitoring remain to be resolved before a consensus on best practices and a regulatory environment can be established.

Streamlining Quality Review of Mass Spectrometry Data in the Clinical Laboratory by Use of Machine Learning.

CONTEXT.­: Turnaround time and productivity of clinical mass spectrometric (MS) testing are hampered by time-consuming manual review of the analytical quality of MS data before release of patient results. OBJECTIVE.­: To determine whether a classification model created by using standard machine learning algorithms can verify analytically acceptable MS results and thereby reduce manual review requirements. DESIGN.­: We obtained retrospective data from gas chromatography-MS analyses of 11-nor-9-carboxy-delta-9-tetrahydrocannabinol (THC-COOH) in 1267 urine samples. The data for each sample had been labeled previously as either analytically unacceptable or acceptable by manual review. The dataset was randomly split into training and test sets (848 and 419 samples, respectively), maintaining equal proportions of acceptable (90%) and unacceptable (10%) results in each set. We used stratified 10-fold cross-validation in assessing the abilities of 6 supervised machine learning algorithms to distinguish unacceptable from acceptable assay results in the training dataset. The classifier with the highest recall was used to build a final model, and its performance was evaluated against the test dataset. RESULTS.­: In comparison testing of the 6 classifiers, a model based on the Support Vector Machines algorithm yielded the highest recall and acceptable precision. After optimization, this model correctly identified all unacceptable results in the test dataset (100% recall) with a precision of 81%. CONCLUSIONS.­: Automated data review identified all analytically unacceptable assays in the test dataset, while reducing the manual review requirement by about 87%. This automation strategy can focus manual review only on assays likely to be problematic, allowing improved throughput and turnaround time without reducing quality.

The College of American Pathologists Biorepository Accreditation Program: Results from the First 5 Years.

The College of American Pathologists (CAP) developed the Biorepository Accreditation Program (BAP) in 2012. This program integrates best practices from the International Society for Biological and Environmental Biorepositories, the National Cancer Institute, the Organisation for Economic Cooperation and Development, the Center for Medicare and Medicaid Services, and the CAP Laboratory Accreditation Program. The goal of this elective program is to provide requirements for standardization in biorepository processes that will result in high-quality specimens that can be used to support research, drug discovery, and personalized medicine. CAP uses a peer inspection model to ensure the inspectors have proper expertise and to promote educational efforts through information sharing. Lead inspectors are comprised of pathologists, PhDs, and managers of biorepositories and they are often supported by CAP staff inspectors. Accreditation is a 3-year continuous cycle of quality with a peer inspection occurring at the start of year 1 and a self-inspection and CAP desk assessment at the start of year 2 and 3. At this time 53 biorepositories are fully CAP BAP accredited and 13 are in the process of obtaining accreditation. There are currently 273 established standards with requirement lists customized based on the scope of activities performed by a biorepository. A total of 90 inspections were completed between May 2012 and December 2016. Sixty-one were initial inspections and 29 were reinspections. A total of 527 deficiencies were identified in the areas of Equipment/Instrumentation (22%), Information Technology (18%), Specimen Handling and QC (15%), Quality Management (16%), Personnel (11%), Safety (10%), Facilities (6%), and Regulatory (2%). Assessment of common deficiencies identifies areas of focus for continuous improvement and educational opportunities. Overall success of the program is high based on the current enrollment of 66 biorepositories, anecdotal participant feedback and increasing national recognition of the BAP in federal documents.

Early Palliative Care Reduces End-of-Life Intensive Care Unit (ICU) Use but Not ICU Course in Patients with Advanced Cancer.

BACKGROUND: Early palliative care for advanced cancer patients improves quality of life and survival, but less is known about its effect on intensive care unit (ICU) use at the end of life. This analysis assessed the effect of a comprehensive early palliative care program on ICU use and other outcomes among patients with advanced cancer. PATIENTS AND METHODS: A retrospective cohort of patients with advanced cancer enrolled in an early palliative care program (n = 275) was compared with a concurrent control group of patients receiving standard care (n = 195) during the same time period by using multivariable logistic regression analysis. The multidisciplinary outpatient palliative care program used early end-of-life care planning, weekly interdisciplinary meetings to discuss patient status, and patient-reported outcomes assessment integrated within the electronic health record. RESULTS: Patients in the control group had statistically significantly higher likelihood of ICU admission at the end of life (odds ratios [ORs]: last 6 months, 3.07; last month, 3.59; terminal admission, 4.69), higher likelihood of death in the hospital (OR, 4.14) or ICU (OR, 5.57), and lower likelihood of hospice enrollment (OR, 0.13). Use of chemotherapy or radiation did not significantly differ between groups, nor did length of ICU stay, code status, ICU procedures (other than cardiopulmonary resuscitation), disposition location, and outcomes after ICU admission. CONCLUSION: Early palliative care significantly reduced ICU use at the end of life but did not change ICU events. This study supports early initiation of palliative care for advanced cancer patients before hospitalizations and intensive care. The Oncologist 2017;22:318-323 IMPLICATIONS FOR PRACTICE: Palliative care has shown clear benefit in quality of life and survival in advanced cancer patients, but less is known about its effect on intensive care. This retrospective cohort study at a university hospital showed that in the last 6 months of life, palliative care significantly reduced intensive care unit (ICU) and hospital admissions, reduced deaths in the hospital, and increased hospice enrollment. It did not, however, change patients' experiences within the ICU, such as number of procedures, code status, length of stay, or disposition. The findings further support that palliative care exerts its benefit before, rather than during, the ICU setting.

Pathology Informatics Essentials for Residents: A Flexible Informatics Curriculum Linked to Accreditation Council for Graduate Medical Education Milestones.

CONTEXT: -Recognition of the importance of informatics to the practice of pathology has surged. Training residents in pathology informatics has been a daunting task for most residency programs in the United States because faculty often lacks experience and training resources. Nevertheless, developing resident competence in informatics is essential for the future of pathology as a specialty. OBJECTIVE: -To develop and deliver a pathology informatics curriculum and instructional framework that guides pathology residency programs in training residents in critical pathology informatics knowledge and skills, and meets Accreditation Council for Graduate Medical Education Informatics Milestones. DESIGN: -The College of American Pathologists, Association of Pathology Chairs, and Association for Pathology Informatics formed a partnership and expert work group to identify critical pathology informatics training outcomes and to create a highly adaptable curriculum and instructional approach, supported by a multiyear change management strategy. RESULTS: -Pathology Informatics Essentials for Residents (PIER) is a rigorous approach for educating all pathology residents in important pathology informatics knowledge and skills. PIER includes an instructional resource guide and toolkit for incorporating informatics training into residency programs that vary in needs, size, settings, and resources. PIER is available at http://www.apcprods.org/PIER (accessed April 6, 2016). CONCLUSIONS: -PIER is an important contribution to informatics training in pathology residency programs. PIER introduces pathology trainees to broadly useful informatics concepts and tools that are relevant to practice. PIER provides residency program directors with a means to implement a standardized informatics training curriculum, to adapt the approach to local program needs, and to evaluate resident performance and progress over time.

Pathology Informatics Essentials for Residents: A flexible informatics curriculum linked to Accreditation Council for Graduate Medical Education milestones.

CONTEXT: Recognition of the importance of informatics to the practice of pathology has surged. Training residents in pathology informatics have been a daunting task for most residency programs in the United States because faculty often lacks experience and training resources. Nevertheless, developing resident competence in informatics is essential for the future of pathology as a specialty. OBJECTIVE: The objective of the study is to develop and deliver a pathology informatics curriculum and instructional framework that guides pathology residency programs in training residents in critical pathology informatics knowledge and skills and meets Accreditation Council for Graduate Medical Education Informatics Milestones. DESIGN: The College of American Pathologists, Association of Pathology Chairs, and Association for Pathology Informatics formed a partnership and expert work group to identify critical pathology informatics training outcomes and to create a highly adaptable curriculum and instructional approach, supported by a multiyear change management strategy. RESULTS: Pathology Informatics Essentials for Residents (PIER) is a rigorous approach for educating all pathology residents in important pathology informatics knowledge and skills. PIER includes an instructional resource guide and toolkit for incorporating informatics training into residency programs that vary in needs, size, settings, and resources. PIER is available at http://www.apcprods.org/PIER (accessed April 6, 2016). CONCLUSIONS: PIER is an important contribution to informatics training in pathology residency programs. PIER introduces pathology trainees to broadly useful informatics concepts and tools that are relevant to practice. PIER provides residency program directors with a means to implement a standardized informatics training curriculum, to adapt the approach to local program needs, and to evaluate resident performance and progress over time.

Smartphone monitoring of pneumatic tube system-induced sample hemolysis.

BACKGROUND: Pneumatic tube systems (PTSs) are convenient methods of patient sample transport in medical centers, but excessive acceleration force and time/distance traveled in the PTS have been correlated with increased blood-sample hemolysis. We investigated the utility of smartphones for monitoring of PTS-related variables. METHODS: Smartphones were sent through the PTS from several hospital locations. Each smartphone used 2 apps as data-loggers to record force of acceleration vs time. To relate the smartphone data to sample integrity, blood samples were collected from 5 volunteers, and hemolysis of the samples was analyzed after they were transported by hand or via 1 of 2 PTS routes. Increased sample hemolysis as measured by plasma lactate dehydrogenase (LD) was also related to the amount of transport in the PTS. RESULTS: The smartphones showed higher duration of forceful acceleration during transport through 1 of the 2 PTS routes, and the increased duration correlated with significant increases in hemolysis (H)-index and plasma LD. In addition, plasma LD showed a positive linear relationship with number of shock forces experienced during transport through the PTS. CONCLUSIONS: Smartphones can monitor PTS variables that cause sample hemolysis. This provides an accessible method for investigating specific PTS routes in medical centers.

Longitudinal patterns of cancer patient reported outcomes in end of life care predict survival.

PURPOSE: Patients with advanced cancer typically demonstrate sharp deterioration in physical function and psychological status during the last months of life. This study evaluates the relationship between survival in patients with advanced cancer and longitudinal assessment of anxiety, depression, fatigue, pain interference, and/or physical function using the US National Institute of Health Patient Reported Outcomes Information System. METHODS: Mixed-effects models were used to evaluate patient-reported outcome trajectories over time among patients with advanced loco-regional or metastatic cancer receiving care in a hospital-based palliative care clinic. Cox regression analysis was used to assess the statistical significance of differences in the probability of survival associated with patient-reported outcome scores. RESULTS: A total of 472 patients completed 1992 assessments during the 18-month study period. Longitudinal scores for fatigue, pain interference, and physical function demonstrated statistically significant non-linear trajectories. Scores for depression, fatigue, pain interference, and physical function were highly statistically significant predictors of survival (p < 0.01). Clinically meaningful differences in the probability of survival were demonstrated between patients with scores at the 25th vs. 75th percentiles, with absolute differences in survival at 6 and 12 months after assessment from 10 to 18 percentage points. CONCLUSIONS: Patient-reported outcomes can be used to reliably estimate where patients are along the trajectory of deteriorating health status leading toward the end of life, and for identifying patients with declining symptoms in need of referral to palliative care or more aggressive symptom management.

Pathology Informatics Essentials for Residents: A Flexible Informatics Curriculum Linked to Accreditation Council for Graduate Medical Education Milestones (a secondary publication).

CONTEXT: Recognition of the importance of informatics to the practice of pathology has surged. Training residents in pathology informatics has been a daunting task for most residency programs in the United States because faculty often lacks experience and training resources. Nevertheless, developing resident competence in informatics is essential for the future of pathology as a specialty. OBJECTIVE: To develop and deliver a pathology informatics curriculum and instructional framework that guides pathology residency programs in training residents in critical pathology informatics knowledge and skills, and meets Accreditation Council for Graduate Medical Education Informatics Milestones. DESIGN: The College of American Pathologists, Association of Pathology Chairs, and Association for Pathology Informatics formed a partnership and expert work group to identify critical pathology informatics training outcomes and to create a highly adaptable curriculum and instructional approach, supported by a multiyear change management strategy. RESULTS: Pathology Informatics Essentials for Residents (PIER) is a rigorous approach for educating all pathology residents in important pathology informatics knowledge and skills. PIER includes an instructional resource guide and toolkit for incorporating informatics training into residency programs that vary in needs, size, settings, and resources. PIER is available at http://www.apcprods.org/PIER (accessed April 6, 2016). CONCLUSIONS: PIER is an important contribution to informatics training in pathology residency programs. PIER introduces pathology trainees to broadly useful informatics concepts and tools that are relevant to practice. PIER provides residency program directors with a means to implement a standardized informatics training curriculum, to adapt the approach to local program needs, and to evaluate resident performance and progress over time.

Concept similarity in publications precedes cross-disciplinary collaboration.

Innovative science frequently occurs as a result of cross-disciplinary collaboration, the importance of which is reflected by recent NIH funding initiatives that promote communication and collaboration. If shared research interests between collaborators are important for the formation of collaborations,methods for identifying these shared interests across scientific domains could potentially reveal new and useful collaboration opportunities. MEDLINE represents a comprehensive database of collaborations and research interests, as reflected by article co-authors and concept content. We analyzed six years of citations using information retrieval based methods to compute articles conceptual similarity, and found that articles by basic and clinical scientists who later collaborated had significantly higher average similarity than articles by similar scientists who did not collaborate.Refinement of these methods and characterization of found conceptual overlaps could allow automated discovery of collaboration opportunities that are currently missed.

Introduction to the mining of clinical data.

The increasing volume of medical data online, including laboratory data, represents a substantial resource that can provide a foundation for improved understanding of disease presentation, response to therapy, and health care delivery processes. Data mining supports these goals by providing a set of techniques designed to discover similarities and relationships between data elements in large data sets. Currently, medical data have several characteristics that increase the difficulty of applying these techniques, although there have been notable medical data mining successes. Future developments in integrated medical data repositories, standardized data representation, and guidelines for the appropriate research use of medical data will decrease the barriers to mining projects.

The development of health care data warehouses to support data mining.

Clinical data warehouses offer tremendous benefits as a foundation for data mining. By serving as a source for comprehensive clinical and demographic information on large patient populations, they streamline knowledge discovery efforts by providing standard and efficient mechanisms to replace time-consuming and expensive original data collection, organization, and processing. Building effective data warehouses requires knowledge of and attention to key issues in database design, data acquisition and processing, and data access and security. In this article, the authors provide an operational and technical definition of data warehouses, present examples of data mining projects enabled by existing data warehouses, and describe key issues and challenges related to warehouse development and implementation.

Multi-database mining.

Biomedical data useful for data mining are often distributed across multiple databases. These databases may be aggregated using several techniques to create single data sets that may be mined using standard approaches; however, separate databases may, in their design or data representation, capture information that is analytically useful and that is lost on integration. Recent techniques for mining multiple databases simultaneously but separately may preserve and leverage the unique perspectives within each database. This article presents an example, "dual mining," in which concurrent analysis of a target database with a related knowledge base can improve the identification of association patterns in the target most likely to be of interest for further analysis.

Temporal data mining.

Large-scale clinical databases provide a detailed perspective on patient phenotype in disease and the characteristics of health care processes. Important information is often contained in the relationships between the values and timestamps of sequences of clinical data. The analysis of clinical time sequence data across entire patient populations may reveal data patterns that enable a more precise understanding of disease presentation, progression, and response to therapy, and thus could be of great value for clinical and translational research. Recent work suggests that the combination of temporal data mining methods with techniques from artificial intelligence research on knowledge-based temporal abstraction may enable the mining of clinically relevant temporal features from these previously problematic general clinical data.

Regional and national health care data repositories.

Efforts are underway to define a national framework for secondary analysis of health-related data. In the meantime, regional health databases have been constructed using insurance claims data, clinical data from single large health care providers, clinical data from multiple collaborating health care providers, and public health data. Large-scale survey data also are available in government databases. Clinical laboratory results are an important component of all these databases because they can provide validation for manually assigned diagnostic and procedure codes and can support inference of key information not provided by coding, such as severity of disease and prevalence of risk factors.

PROTEMPA: a method for specifying and identifying temporal sequences in retrospective data for patient selection.

OBJECTIVE: To specify and identify disease and patient care processes represented by temporal patterns in clinical events and observations, and retrieve patient populations containing those patterns from clinical data repositories, in support of clinical research, outcomes studies, and quality assurance. DESIGN: A data processing method called PROTEMPA (Process-oriented Temporal Analysis) was developed for defining and detecting clinically relevant temporal and mathematical patterns in retrospective data. PROTEMPA provides for portability across data sources, "pluggable" data processing environments, and the creation of libraries of pattern definitions and data processing algorithms. MEASUREMENTS: A proof-of-concept implementation of PROTEMPA in Java was evaluated against standard SQL queries for its ability to identify patients from a large clinical data repository who show the features of HELLP syndrome, and categorize those patients by disease severity and progression based on time sequence characteristics in their clinical laboratory test results. RESULTS were verified by manual case review. RESULTS: The proof-of-concept implementation was more accurate than SQL in identifying patients with HELLP and correctly assigned severity and disease progression categories, which was not possible using SQL only. CONCLUSIONS: PROTEMPA supports the identification and categorization of patients with complex disease based on the characteristics of and relationships between time sequences in multiple data types. Identifying patient populations who share these types of patterns may be useful when patient features of interest do not have standard codes, are poorly-expressed in coding schemes, may be inaccurately or incompletely coded, or are not represented explicitly as data values.

Abstraction-based temporal data retrieval for a Clinical Data Repository.

Disease and patient care processes often create characteristic states, trends, and temporal patterns in clinical events and observations, called temporal abstractions. Identifying patient populations who share similar abstractions may be useful for clinical research, outcomes studies, and quality assurance. In these settings, abstractions may be specific to a query, and thus allowing the specification of abstractions directly in the query would be desirable. We propose a query language for specifying and retrieving clinical data sets that allows specifying abstractions directly, and automatically selects data for retrieval based on the presence of abstractions inferred from the data. We describe the language and a prototype implementation, demonstrate its features with two queries constructed in response to clinical researcher-initiated data requests submitted to our institution's Clinical Data Repository, and report preliminary results from an evaluation of the implementation's performance.

Definition of an XML markup language for clinical laboratory procedures and comparison with generic XML markup.

BACKGROUND: Clinical laboratory procedure manuals are typically maintained as word processor files and are inefficient to store and search, require substantial effort for review and updating, and integrate poorly with other laboratory information. Electronic document management systems could improve procedure management and utility. As a first step toward building such systems, we have developed a prototype electronic format for laboratory procedures using Extensible Markup Language (XML). METHODS: Representative laboratory procedures were analyzed to identify document structure and data elements. This information was used to create a markup vocabulary, CLP-ML, expressed as an XML Document Type Definition (DTD). To determine whether this markup provided advantages over generic markup, we compared procedures structured with CLP-ML or with the vocabulary of the Health Level Seven, Inc. (HL7) Clinical Document Architecture (CDA) narrative block. RESULTS: CLP-ML includes 124 XML tags and supports a variety of procedure types across different laboratory sections. When compared with a general-purpose markup vocabulary (CDA narrative block), CLP-ML documents were easier to edit and read, less complex structurally, and simpler to traverse for searching and retrieval. CONCLUSION: In combination with appropriate software, CLP-ML is designed to support electronic authoring, reviewing, distributing, and searching of clinical laboratory procedures from a central repository, decreasing procedure maintenance effort and increasing the utility of procedure information. A standard electronic procedure format could also allow laboratories and vendors to share procedures and procedure layouts, minimizing duplicative word processor editing. Our results suggest that laboratory-specific markup such as CLP-ML will provide greater benefit for such systems than generic markup.