Current status of artificial intelligence methods for skin cancer survival analysis: a scoping review.

Skin cancer mortality rates continue to rise, and survival analysis is increasingly needed to understand who is at risk and what interventions improve outcomes. However, current statistical methods are limited by inability to synthesize multiple data types, such as patient genetics, clinical history, demographics, and pathology and reveal significant multimodal relationships through predictive algorithms. Advances in computing power and data science enabled the rise of artificial intelligence (AI), which synthesizes vast amounts of data and applies algorithms that enable personalized diagnostic approaches. Here, we analyze AI methods used in skin cancer survival analysis, focusing on supervised learning, unsupervised learning, deep learning, and natural language processing. We illustrate strengths and weaknesses of these approaches with examples. Our PubMed search yielded 14 publications meeting inclusion criteria for this scoping review. Most publications focused on melanoma, particularly histopathologic interpretation with deep learning. Such concentration on a single type of skin cancer amid increasing focus on deep learning highlight growing areas for innovation; however, it also demonstrates opportunity for additional analysis that addresses other types of cutaneous malignancies and expands the scope of prognostication to combine both genetic, histopathologic, and clinical data. Moreover, researchers may leverage multiple AI methods for enhanced benefit in analyses. Expanding AI to this arena may enable improved survival analysis, targeted treatments, and outcomes.

TCGA-Reports: A machine-readable pathology report resource for benchmarking text-based AI models.

In cancer research, pathology report text is a largely untapped data source. Pathology reports are routinely generated, more nuanced than structured data, and contain added insight from pathologists. However, there are no publicly available datasets for benchmarking report-based models. Two recent advances suggest the urgent need for a benchmark dataset. First, improved optical character recognition (OCR) techniques will make it possible to access older pathology reports in an automated way, increasing the data available for analysis. Second, recent improvements in natural language processing (NLP) techniques using artificial intelligence (AI) allow more accurate prediction of clinical targets from text. We apply state-of-the-art OCR and customized post-processing to report PDFs from The Cancer Genome Atlas, generating a machine-readable corpus of 9,523 reports. Finally, we perform a proof-of-principle cancer-type classification across 32 tissues, achieving 0.992 average AU-ROC. This dataset will be useful to researchers across specialties, including research clinicians, clinical trial investigators, and clinical NLP researchers.

Estimating the heritability of SARS-CoV-2 susceptibility and COVID-19 severity.

SARS-CoV-2 has infected over 340 million people, prompting therapeutic research. While genetic studies can highlight potential drug targets, understanding the heritability of SARS-CoV-2 susceptibility and COVID-19 severity can contextualize their results. To date, loci from meta-analyses explain 1.2% and 5.8% of variation in susceptibility and severity respectively. Here we estimate the importance of shared environment and additive genetic variation to SARS-CoV-2 susceptibility and COVID-19 severity using pedigree data, PCR results, and hospitalization information. The relative importance of genetics and shared environment for susceptibility shifted during the study, with heritability ranging from 33% (95% CI: 20%-46%) to 70% (95% CI: 63%-74%). Heritability was greater for days hospitalized with COVID-19 (41%, 95% CI: 33%-57%) compared to shared environment (33%, 95% CI: 24%-38%). While our estimates suggest these genetic architectures are not fully understood, the shift in susceptibility estimates highlights the challenge of estimation during a pandemic, given environmental fluctuations and vaccine introduction.

Inborn Errors of Immunity Contribute to the Burden of Skin Disease and Create Opportunities for Improving the Practice of Dermatology.

Opportunities to improve the clinical management of skin disease are being created by advances in genomic medicine. Large-scale sequencing increasingly challenges notions about single-gene disorders. It is now apparent that monogenic etiologies make appreciable contributions to the population burden of disease and that they are underrecognized in clinical practice. A genetic diagnosis informs on molecular pathology and may direct targeted treatments and tailored prevention strategies for patients and family members. It also generates knowledge about disease pathogenesis and management that is relevant to patients without rare pathogenic variants. Inborn errors of immunity are a large class of monogenic etiologies that have been well-studied and contribute to the population burden of inflammatory diseases. To further delineate the contributions of inborn errors of immunity to the pathogenesis of skin disease, we performed a set of analyses that identified 316 inborn errors of immunity associated with skin pathologies, including common skin diseases. These data suggest that clinical sequencing is underutilized in dermatology. We next use these data to derive a network that illuminates the molecular relationships of these disorders and suggests an underlying etiological organization to immune-mediated skin disease. Our results motivate the further development of a molecularly derived and data-driven reorganization of clinical diagnoses of skin disease.

SynTwin: A graph-based approach for predicting clinical outcomes using digital twins derived from synthetic patients.

The concept of a digital twin came from the engineering, industrial, and manufacturing domains to create virtual objects or machines that could inform the design and development of real objects. This idea is appealing for precision medicine where digital twins of patients could help inform healthcare decisions. We have developed a methodology for generating and using digital twins for clinical outcome prediction. We introduce a new approach that combines synthetic data and network science to create digital twins (i.e. SynTwin) for precision medicine. First, our approach starts by estimating the distance between all subjects based on their available features. Second, the distances are used to construct a network with subjects as nodes and edges defining distance less than the percolation threshold. Third, communities or cliques of subjects are defined. Fourth, a large population of synthetic patients are generated using a synthetic data generation algorithm that models the correlation structure of the data to generate new patients. Fifth, digital twins are selected from the synthetic patient population that are within a given distance defining a subject community in the network. Finally, we compare and contrast community-based prediction of clinical endpoints using real subjects, digital twins, or both within and outside of the community. Key to this approach are the digital twins defined using patient similarity that represent hypothetical unobserved patients with patterns similar to nearby real patients as defined by network distance and community structure. We apply our SynTwin approach to predicting mortality in a population-based cancer registry (n=87,674) from the Surveillance, Epidemiology, and End Results (SEER) program from the National Cancer Institute (USA). Our results demonstrate that nearest network neighbor prediction of mortality in this study is significantly improved with digital twins (AUROC=0.864, 95% CI=0.857-0.872) over just using real data alone (AUROC=0.791, 95% CI=0.781-0.800). These results suggest a network-based digital twin strategy using synthetic patients may add value to precision medicine efforts.

Using machine learning probabilities to identify effects of COVID-19.

Coronavirus disease 2019 (COVID-19), the disease caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) virus, has had extensive economic, social, and public health impacts in the United States and around the world. To date, there have been more than 600 million reported infections worldwide with more than 6 million reported deaths. Retrospective analysis, which identified comorbidities, risk factors, and treatments, has underpinned the response. As the situation transitions to an endemic, retrospective analyses using electronic health records will be important to identify the long-term effects of COVID-19. However, these analyses can be complicated by incomplete records, which makes it difficult to differentiate visits where the patient had COVID-19. To address this issue, we trained a random Forest classifier to assign a probability of a patient having been diagnosed with COVID-19 during each visit. Using these probabilities, we found that higher COVID-19 probabilities were associated with a future diagnosis of myocardial infarction, urinary tract infection, acute renal failure, and type 2 diabetes.

Benchmark Pathology Report Text Corpus with Cancer Type Classification.

In cancer research, pathology report text is a largely un-tapped data source. Pathology reports are routinely generated, more nuanced than structured data, and contain added insight from pathologists. However, there are no publicly-available datasets for benchmarking report-based models. Two recent advances suggest the urgent need for a benchmark dataset. First, improved optical character recognition (OCR) techniques will make it possible to access older pathology reports in an automated way, increasing data available for analysis. Second, recent improvements in natural language processing (NLP) techniques using AI allow more accurate prediction of clinical targets from text. We apply state-of-the-art OCR and customized post-processing to publicly available report PDFs from The Cancer Genome Atlas, generating a machine-readable corpus of 9,523 reports. We perform a proof-of-principle cancer-type classification across 32 tissues, achieving 0.992 average AU-ROC. This dataset will be useful to researchers across specialties, including research clinicians, clinical trial investigators, and clinical NLP researchers.

Natural Language Processing Can Automate Extraction of Barrett's Esophagus Endoscopy Quality Metrics.

Objectives: To develop an automated natural language processing (NLP) method for extracting high-fidelity Barrett's Esophagus (BE) endoscopic surveillance and treatment data from the electronic health record (EHR). Methods: Patients who underwent BE-related endoscopies between 2016 and 2020 at a single medical center were randomly assigned to a development or validation set. Those not aged 40 to 80 and those without confirmed BE were excluded. For each patient, free text pathology reports and structured procedure data were obtained. Gastroenterologists assigned ground truth labels. An NLP method leveraging MetaMap Lite generated endoscopy-level diagnosis and treatment data. Performance metrics were assessed for this data. The NLP methodology was then adapted to label key endoscopic eradication therapy (EET)-related endoscopy events and thereby facilitate calculation of patient-level pre-EET diagnosis, endotherapy time, and time to CE-IM. Results: 99 patients (377 endoscopies) and 115 patients (399 endoscopies) were included in the development and validation sets respectively. When assigning high-fidelity labels to the validation set, NLP achieved high performance (recall: 0.976, precision: 0.970, accuracy: 0.985, and F1-score: 0.972). 77 patients initiated EET and underwent 554 endoscopies. Key EET-related clinical event labels had high accuracy (EET start: 0.974, CE-D: 1.00, and CE-IM: 1.00), facilitating extraction of pre-treatment diagnosis, endotherapy time, and time to CE-IM. Conclusions: High-fidelity BE endoscopic surveillance and treatment data can be extracted from routine EHR data using our automated, transparent NLP method. This method produces high-level clinical datasets for clinical research and quality metric assessment.

ChatGPT and large language models in academia: opportunities and challenges.

The introduction of large language models (LLMs) that allow iterative "chat" in late 2022 is a paradigm shift that enables generation of text often indistinguishable from that written by humans. LLM-based chatbots have immense potential to improve academic work efficiency, but the ethical implications of their fair use and inherent bias must be considered. In this editorial, we discuss this technology from the academic's perspective with regard to its limitations and utility for academic writing, education, and programming. We end with our stance with regard to using LLMs and chatbots in academia, which is summarized as (1) we must find ways to effectively use them, (2) their use does not constitute plagiarism (although they may produce plagiarized text), (3) we must quantify their bias, (4) users must be cautious of their poor accuracy, and (5) the future is bright for their application to research and as an academic tool.

Generalizable and Automated Classification of TNM Stage from Pathology Reports with External Validation.

Cancer staging is an essential clinical attribute informing patient prognosis and clinical trial eligibility. However, it is not routinely recorded in structured electronic health records. Here, we present a generalizable method for the automated classification of TNM stage directly from pathology report text. We train a BERT-based model using publicly available pathology reports across approximately 7,000 patients and 23 cancer types. We explore the use of different model types, with differing input sizes, parameters, and model architectures. Our final model goes beyond term-extraction, inferring TNM stage from context when it is not included in the report text explicitly. As external validation, we test our model on almost 8,000 pathology reports from Columbia University Medical Center, finding that our trained model achieved an AU-ROC of 0.815-0.942. This suggests that our model can be applied broadly to other institutions without additional institution-specific fine-tuning.

Discovering Venom-Derived Drug Candidates Using Differential Gene Expression.

Venoms are a diverse and complex group of natural toxins that have been adapted to treat many types of human disease, but rigorous computational approaches for discovering new therapeutic activities are scarce. We have designed and validated a new platform-named VenomSeq-to systematically identify putative associations between venoms and drugs/diseases via high-throughput transcriptomics and perturbational differential gene expression analysis. In this study, we describe the architecture of VenomSeq and its evaluation using the crude venoms from 25 diverse animal species and 9 purified teretoxin peptides. By integrating comparisons to public repositories of differential expression, associations between regulatory networks and disease, and existing knowledge of venom activity, we provide a number of new therapeutic hypotheses linking venoms to human diseases supported by multiple layers of preliminary evidence.

Plasma Metabolomics Predicts Chemotherapy Response in Advanced Pancreatic Cancer.

Pancreatic cancer (PC) is one of the deadliest cancers. Developing biomarkers for chemotherapeutic response prediction is crucial for improving the dismal prognosis of advanced-PC patients (pts). To evaluate the potential of plasma metabolites as predictors of the response to chemotherapy for PC patients, we analyzed plasma metabolites using high-performance liquid chromatography-mass spectrometry from 31 cachectic, advanced-PC subjects enrolled into the PANCAX-1 (NCT02400398) prospective trial to receive a jejunal tube peptide-based diet for 12 weeks and who were planned for palliative chemotherapy. Overall, there were statistically significant differences in the levels of intermediates of multiple metabolic pathways in pts with a partial response (PR)/stable disease (SD) vs. progressive disease (PD) to chemotherapy. When stratified by the chemotherapy regimen, PD after 5-fluorouracil-based chemotherapy (e.g., FOLFIRINOX) was associated with decreased levels of amino acids (AAs). For gemcitabine-based chemotherapy (e.g., gemcitabine/nab-paclitaxel), PD was associated with increased levels of intermediates of glycolysis, the TCA cycle, nucleoside synthesis, and bile acid metabolism. These results demonstrate the feasibility of plasma metabolomics in a prospective cohort of advanced-PC patients for assessing the effect of enteral feeding as their primary source of nutrition. Metabolic signatures unique to FOLFIRINOX or gemcitabine/nab-paclitaxel may be predictive of a patient's response and warrant further study.

Structured deep embedding model to generate composite clinical indices from electronic health records for early detection of pancreatic cancer.

The high-dimensionality, complexity, and irregularity of electronic health records (EHR) data create significant challenges for both simplified and comprehensive health assessments, prohibiting an efficient extraction of actionable insights by clinicians. If we can provide human decision-makers with a simplified set of interpretable composite indices (i.e., combining information about groups of related measures into single representative values), it will facilitate effective clinical decision-making. In this study, we built a structured deep embedding model aimed at reducing the dimensionality of the input variables by grouping related measurements as determined by domain experts (e.g., clinicians). Our results suggest that composite indices representing liver function may consistently be the most important factor in the early detection of pancreatic cancer (PC). We propose our model as a basis for leveraging deep learning toward developing composite indices from EHR for predicting health outcomes, including but not limited to various cancers, with clinically meaningful interpretations.

Association between neighborhood socioeconomic status, built environment and SARS-CoV-2 infection among cancer patients treated at a Tertiary Cancer Center in New York City.

BACKGROUND: Racial and ethnic minority groups experience a disproportionate burden of SARS-CoV-2 illness and studies suggest that cancer patients are at a particular risk for severe SARS-CoV-2 infection. AIMS: The objective of this study was examine the association between neighborhood characteristics and SARS-CoV-2 infection among patients with cancer. METHODS AND RESULTS: We performed a cross-sectional study of New York City residents receiving treatment for cancer at a tertiary cancer center. Patients were linked by their address to data from the US Census Bureau's American Community Survey and to real estate tax data from New York's Department of City Planning. Models were used to both to estimate odds ratios (ORs) per unit increase and to predict probabilities (and 95% CI) of SARS-CoV2 infection. We identified 2350 New York City residents with cancer receiving treatment. Overall, 214 (9.1%) were infected with SARS-CoV-2. In adjusted models, the percentage of Hispanic/Latino population (aOR = 1.01; 95% CI, 1.005-1.02), unemployment rate (aOR = 1.10; 95% CI, 1.05-1.16), poverty rates (aOR = 1.02; 95% CI, 1.0002-1.03), rate of >1 person per room (aOR = 1.04; 95% CI, 1.01-1.07), average household size (aOR = 1.79; 95% CI, 1.23-2.59) and population density (aOR = 1.86; 95% CI, 1.27-2.72) were associated with SARS-CoV-2 infection. CONCLUSION: Among cancer patients in New York City receiving anti-cancer therapy, SARS-CoV-2 infection was associated with neighborhood- and building-level markers of larger household membership, household crowding, and low socioeconomic status. NOVELTY AND IMPACT: We performed a cross-sectional analysis of residents of New York City receiving treatment for cancer in which we linked subjects to census and real estate date. This linkage is a novel way to examine the neighborhood characteristics that influence SARS-COV-2 infection. We found that among patients receiving anti-cancer therapy, SARS-CoV-2 infection was associated with building and neighborhood-level markers of household crowding, larger household membership, and low socioeconomic status. With ongoing surges of SARS-CoV-2 infections, these data may help in the development of interventions to decrease the morbidity and mortality associated with SARS-CoV-2 among cancer patients.

Session Introduction: TOWARDS ETHICAL BIOMEDICAL INFORMATICS: LEARNING FROM OLELO NOEAU, HAWAIIAN PROVERBS.

Innovations in human-centered biomedical informatics are often developed with the eventual goal of real-world translation. While biomedical research questions are usually answered in terms of how a method performs in a particular context, we argue that it is equally important to consider and formally evaluate the ethical implications of informatics solutions. Several new research paradigms have arisen as a result of the consideration of ethical issues, including but not limited for privacy-preserving computation and fair machine learning. In the spirit of the Pacific Symposium on Biocomputing, we discuss broad and fundamental principles of ethical biomedical informatics in terms of Olelo Noeau, or Hawaiian proverbs and poetical sayings that capture Hawaiian values. While we emphasize issues related to privacy and fairness in particular, there are a multitude of facets to ethical biomedical informatics that can benefit from a critical analysis grounded in ethics.

Mural Wnt/ß-catenin signaling regulates Lama2 expression to promote neurovascular unit maturation.

Neurovascular unit and barrier maturation rely on vascular basement membrane (vBM) composition. Laminins, a major vBM component, are crucial for these processes, yet the signaling pathway(s) that regulate their expression remain unknown. Here, we show that mural cells have active Wnt/ß-catenin signaling during central nervous system development in mice. Bulk RNA sequencing and validation using postnatal day 10 and 14 wild-type versus adenomatosis polyposis coli downregulated 1 (Apcdd1-/-) mouse retinas revealed that Lama2 mRNA and protein levels are increased in mutant vasculature with higher Wnt/ß-catenin signaling. Mural cells are the main source of Lama2, and Wnt/ß-catenin activation induces Lama2 expression in mural cells in vitro. Markers of mature astrocytes, including aquaporin 4 (a water channel in astrocyte endfeet) and integrin-α6 (a laminin receptor), are upregulated in Apcdd1-/- retinas with higher Lama2 vBM deposition. Thus, the Wnt/ß-catenin pathway regulates Lama2 expression in mural cells to promote neurovascular unit and barrier maturation.

Alterations in the kallikrein-kinin system predict death after heart transplant.

Heart transplantation remains the definitive treatment for end stage heart failure. Because availability is limited, risk stratification of candidates is crucial for optimizing both organ allocations and transplant outcomes. Here we utilize proteomics prior to transplant to identify new biomarkers that predict post-transplant survival in a multi-institutional cohort. Microvesicles were isolated from serum samples and underwent proteomic analysis using mass spectrometry. Monte Carlo cross-validation (MCCV) was used to predict survival after transplant incorporating select recipient pre-transplant clinical characteristics and serum microvesicle proteomic data. We identified six protein markers with prediction performance above AUROC of 0.6, including Prothrombin (F2), anti-plasmin (SERPINF2), Factor IX, carboxypeptidase 2 (CPB2), HGF activator (HGFAC) and low molecular weight kininogen (LK). No clinical characteristics demonstrated an AUROC > 0.6. Putative biological functions and pathways were assessed using gene set enrichment analysis (GSEA). Differential expression analysis identified enriched pathways prior to transplant that were associated with post-transplant survival including activation of platelets and the coagulation pathway prior to transplant. Specifically, upregulation of coagulation cascade components of the kallikrein-kinin system (KKS) and downregulation of kininogen prior to transplant were associated with survival after transplant. Further prospective studies are warranted to determine if alterations in the KKS contributes to overall post-transplant survival.

A database of pediatric drug effects to evaluate ontogenic mechanisms from child growth and development.

BACKGROUND: Adverse drug effects (ADEs) in children are common and may result in disability and death, necessitating post-marketing monitoring of their use. Evaluating drug safety is especially challenging in children due to the processes of growth and maturation, which can alter how children respond to treatment. Current drug safety-signal-detection methods do not account for these dynamics. METHODS: We recently developed a method called disproportionality generalized additive models (dGAMs) to better identify safety signals for drugs across child-development stages. FINDINGS: We used dGAMs on a database of 264,453 pediatric adverse-event reports and found 19,438 ADEs signals associated with development and validated these signals against a small reference set of pediatric ADEs. Using our approach, we can hypothesize on the ontogenic dynamics of ADE signals, such as that montelukast-induced psychiatric disorders appear most significant in the second year of life. Additionally, we integrated pediatric enzyme expression data and found that pharmacogenes with dynamic childhood expression, such as CYP2C18 and CYP27B1, are associated with pediatric ADEs. CONCLUSIONS: We curated KidSIDES, a database of pediatric drug safety signals, for the research community and developed the Pediatric Drug Safety portal (PDSportal) to facilitate evaluation of drug safety signals across childhood growth and development. FUNDING: This study was supported by grants from the National Institutes of Health (NIH).

Deep learning on time series laboratory test results from electronic health records for early detection of pancreatic cancer.

The multi-modal and unstructured nature of observational data in Electronic Health Records (EHR) is currently a significant obstacle for the application of machine learning towards risk stratification. In this study, we develop a deep learning framework for incorporating longitudinal clinical data from EHR to infer risk for pancreatic cancer (PC). This framework includes a novel training protocol, which enforces an emphasis on early detection by applying an independent Poisson-random mask on proximal-time measurements for each variable. Data fusion for irregular multivariate time-series features is enabled by a "grouped" neural network (GrpNN) architecture, which uses representation learning to generate a dimensionally reduced vector for each measurement set before making a final prediction. These models were evaluated using EHR data from Columbia University Irving Medical Center-New York Presbyterian Hospital. Our framework demonstrated better performance on early detection (AUROC 0.671, CI 95% 0.667 - 0.675, p < 0.001) at 12 months prior to diagnosis compared to a logistic regression, xgboost, and a feedforward neural network baseline. We demonstrate that our masking strategy results greater improvements at distal times prior to diagnosis, and that our GrpNN model improves generalizability by reducing overfitting relative to the feedforward baseline. The results were consistent across reported race. Our proposed algorithm is potentially generalizable to other diseases including but not limited to cancer where early detection can improve survival.

System-wide transcriptome damage and tissue identity loss in COVID-19 patients.

The molecular mechanisms underlying the clinical manifestations of coronavirus disease 2019 (COVID-19), and what distinguishes them from common seasonal influenza virus and other lung injury states such as acute respiratory distress syndrome, remain poorly understood. To address these challenges, we combine transcriptional profiling of 646 clinical nasopharyngeal swabs and 39 patient autopsy tissues to define body-wide transcriptome changes in response to COVID-19. We then match these data with spatial protein and expression profiling across 357 tissue sections from 16 representative patient lung samples and identify tissue-compartment-specific damage wrought by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection, evident as a function of varying viral loads during the clinical course of infection and tissue-type-specific expression states. Overall, our findings reveal a systemic disruption of canonical cellular and transcriptional pathways across all tissues, which can inform subsequent studies to combat the mortality of COVID-19 and to better understand the molecular dynamics of lethal SARS-CoV-2 and other respiratory infections.