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.

Payment Reform in the Era of Advanced Diagnostics, Artificial Intelligence, and Machine Learning.

Health care is undergoing a profound transformation driven by an increase in new types of diagnostic data, increased data sharing enabled by interoperability, and improvements in our ability to interpret data through the application of artificial intelligence and machine learning. Paradoxically, we are also discovering that our current paradigms for implementing electronic health-care records and our ability to create new models for reforming the health-care system have fallen short of expectations. This article traces these shortcomings to two basic issues. The first is a reliance on highly centralized quality improvement and measurement strategies that fail to account for the high level of variation and complexity found in human disease. The second is a reliance on legacy payment systems that fail to reward the sharing of data and knowledge across the health-care system. To address these issues, and to better harness the advances in health care noted above, the health-care system must undertake a phased set of reforms. First, efforts must focus on improving both the diagnostic process and data sharing at the local level. These efforts should include the formation of diagnostic management teams and increased collaboration between pathologists and radiologists. Next, building off current efforts to develop national federated research databases, providers must be able to query national databases when information is needed to inform the care of a specific complex patient. In addition, providers, when treating a specific complex patient, should be enabled to consult nationally with other providers who have experience with similar patient issues. The goal of these efforts is to build a health-care system that is funded in part by a novel fee-for-knowledge-sharing paradigm that fosters a collaborative decentralized approach to patient care and financially incentivizes large-scale data and knowledge sharing.

Quantifying the competitiveness of the electronic health record market and its implications for interoperability.

OBJECTIVE: The objective of this study was to quantify both the competitiveness of the EHR vendor market in the United States of America (US) and the degree of fragmentation of individual Medicare beneficiaries' medical records across the differing EHR vendors found in the US healthcare system. METHODS AND MATERIALS: We determined the Part A and Part B Medicare-expenditure weighted market shares of EHR vendors and estimated the rate of attestation of meaningful use (MU) for EHRs among Medicare Part A & B providers from 2011 to 2016. Based on these data we calculated the annual Herfindahl-Hirschman Index to quantify the competitiveness of the EHR market as well as the number of vendors individual Medicare beneficiaries' medical records were stored in for the period 2014-2016. RESULTS: We find that as of 2016 the EHR vendor environment was competitive but trending towards becoming highly concentrated soon. We also found that patient medical records were highly fragmented as only 4.5 % of expenditure-weighted individual Medicare beneficiaries had their MU medical records associated with a single vendor, while 19.8 % of expenditure-weighted beneficiaries had their MU medical records stored in 8 or more vendors. DISCUSSION: These results indicate that there are tradeoffs between EHR market competition, and the challenges associated with achieving interoperability across numerous competing vendors. CONCLUSION: Uncertainty of interoperability among different EHR vendors may make transmission of medical records among different providers challenging, mitigating the benefit of vendor competition. This highlights the critical importance of current interoperability efforts moving forward.

Identifying and Coordinating Care for Complex Patients: Findings from the Leading Edge of Analytics and Health Information Technology.

In the United States, a relatively small proportion of complex patients---defined as having multiple comorbidities, high risk for poor outcomes, and high cost---incur most of the nation's health care costs. Improved care coordination and management of complex patients could reduce costs while increasing quality of care. However, care coordination efforts face multiple challenges, such as segmenting populations of complex patients to better match their needs with the design of specific interventions, understanding how to reduce spending, and integrating care coordination programs into providers' care delivery processes. Innovative uses of analytics and health information technology (HIT) may address these challenges. Rudin and colleagues at RAND completed a literature review and held discussions with subject matter experts, reaching the conclusion that analytics and HIT are being used in innovative ways to coordinate care for complex patients but that the capabilities are limited, evidence of their effectiveness is lacking, and challenges are substantial, and important foundational work is still needed.

A Comparison of Disease Burden Between Twins and Control Pairs in Medicare: Quantification of Heredity's Role in Human Health.

To quantify heredity's effects on the burden of illness in the Medicare population, this study linked information between participants in a research twin registry to a comprehensive set of Medicare claims. To calculate disease categories, the authors used the Centers for Medicare & Medicaid Services Hierarchical Conditions Categories (HCC) model that was developed to risk adjust Medicare's capitation payments to private health care plans based on the health expenditure risk of their enrollees. Using the Medicare database, 2 sets of unrelated but demographically matched control pairs (MCPs) were generated, one specific for the monozygotic twin population and the second specific for the dizygotic twin population. The concordance and correlation rates of the 70 HCC categories for the 2 twin populations, in comparison to their corresponding MCP, was then calculated using Medicare claims data from 1991 through 2011. When indicated, HCCs for which there was a statistically significant difference between the twin and corresponding MCP control group were analyzed by calculating concordance and correlation rates of the International Classification of Diseases, Ninth Revision codes that compose the HCC. Findings reveal that monozygotic twins share 6.5% more HCC disease categories than their MCP while dizygotic twins share 3.8% more HCC disease categories than their MCP. Atrial fibrillation is a highly heritable disease category, a finding consistent with prior literature describing the heritability of the cardiac arrhythmias. These findings are consistent with qualitative assessments of heredity's role found in previous models of population health, and provide both novel methods and quantitative evidence to support future model development.

Temporal variation in patterns of comorbidities in the medicare population.

It is widely accepted that Medicare beneficiaries with multiple comorbidities (ie, patients with combinations of more than 1 disease) account for a disproportionate amount of mortality and expenditures. The authors previously studied this phenomenon by analyzing Medicare claims data from 2008 to determine the pattern of disease combinations (DCs) for 32,220,634 beneficiaries. Their findings indicated that 22% of these individuals mapped to a long-tailed distribution of approximately 1 million DCs. The presence of so many DCs, each populated by a small number of individuals, raises the possibility that the DC distribution varies over time. Measuring this variability is important because it indicates the rate at which the health care system must adapt to the needs of new patients. This article analyzes Medicare claims data for 3 consecutive calendar years, using 2 algorithms based on the Centers for Medicare & Medicaid Services (CMS)-Hierarchical Conditions Categories (HCC) claims model. These algorithms make different assumptions regarding the degree to which the CMS-HCC model could be disaggregated into its underlying International Classification of Diseases, Ninth Revision, Clinical Modification codes. The authors find that, although a large number of beneficiaries belong to a set of DCs that are nationally stable across the 3 study years, the number of DCs in this set is large (in the range of several hundred thousand). Furthermore, the small number of beneficiaries associated with the larger number of variable DCs (ie, DCs that were not constantly populated in all 3 study years) represents a disproportionally high level of expenditures and death.

Integrating pathology and radiology disciplines: an emerging opportunity?

Pathology and radiology form the core of cancer diagnosis, yet the workflows of both specialties remain ad hoc and occur in separate "silos," with no direct linkage between their case accessioning and/or reporting systems, even when both departments belong to the same host institution. Because both radiologists' and pathologists' data are essential to making correct diagnoses and appropriate patient management and treatment decisions, this isolation of radiology and pathology workflows can be detrimental to the quality and outcomes of patient care. These detrimental effects underscore the need for pathology and radiology workflow integration and for systems that facilitate the synthesis of all data produced by both specialties. With the enormous technological advances currently occurring in both fields, the opportunity has emerged to develop an integrated diagnostic reporting system that supports both specialties and, therefore, improves the overall quality of patient care.

The complexity of disease combinations in the Medicare population.

Developing systems of care that address the mortality, morbidity, and expenditures associated with Medicare beneficiaries with multiple diseases would benefit from a greater understanding of the complexity of disease combinations (DCs) found in the Medicare population. To develop estimates of the number of DCs, we performed an observational analysis on 32,220,634 beneficiaries in the Medicare Fee-for-Service claims database based on a set of records containing each beneficiary's Part A and B International Classification of Diseases, 9(th) Revision, Clinical Modification (ICD-9-CM) claims data for the year of 2008. We made 2 simplifying adjustments. First, we mapped the individual ICD-9-CM codes to the Centers for Medicare and Medicaid Services-Hierarchical Conditions Categories (HCC) model that was developed in 2004 to risk adjust capitation payments to private health care plans based on the health expenditure risk of their enrollees. Second, we aggregated beneficiaries with identical HCCs regardless of the temporal order of these findings within the 2008 claims year; thus the DC to which they are assigned represents the summation of their 2008 claims data. We defined 3 distinct populations at the HCC level. The first consisted of 35% of the beneficiaries who did not fall into any HCC category and accounted for 6% of expenditures. The second was represented by the 100 next most prevalent DCs that accounted for 33% of the beneficiaries and 15% of expenditures. The final population, accounting for 32% of the beneficiaries and 79% of expenses, was complex and consisted of over 2 million DCs. Our results indicate that the majority of expenditures are associated with a complex set of beneficiaries.

A data review and re-assessment of ovarian cancer serum proteomic profiling.

BACKGROUND: The early detection of ovarian cancer has the potential to dramatically reduce mortality. Recently, the use of mass spectrometry to develop profiles of patient serum proteins, combined with advanced data mining algorithms has been reported as a promising method to achieve this goal. In this report, we analyze the Ovarian Dataset 8-7-02 downloaded from the Clinical Proteomics Program Databank website, using nonparametric statistics and stepwise discriminant analysis to develop rules to diagnose patients, as well as to understand general patterns in the data that may guide future research. RESULTS: The mass spectrometry serum profiles derived from cancer and controls exhibited numerous statistical differences. For example, use of the Wilcoxon test in comparing the intensity at each of the 15,154 mass to charge (M/Z) values between the cancer and controls, resulted in the detection of 3,591 M/Z values whose intensities differed by a p-value of 10-6 or less. The region containing the M/Z values of greatest statistical difference between cancer and controls occurred at M/Z values less than 500. For example the M/Z values of 2.7921478 and 245.53704 could be used to significantly separate the cancer from control groups. Three other sets of M/Z values were developed using a training set that could distinguish between cancer and control subjects in a test set with 100% sensitivity and specificity. CONCLUSION: The ability to discriminate between cancer and control subjects based on the M/Z values of 2.7921478 and 245.53704 reveals the existence of a significant non-biologic experimental bias between these two groups. This bias may invalidate attempts to use this dataset to find patterns of reproducible diagnostic value. To minimize false discovery, results using mass spectrometry and data mining algorithms should be carefully reviewed and benchmarked with routine statistical methods.