Novel Machine Learning Identifies 5 Asthma Phenotypes Using Cluster Analysis of Real-World Data.

BACKGROUND: Asthma classification into different subphenotypes is important to guide personalized therapy and improve outcomes. OBJECTIVES: To further explore asthma heterogeneity through determination of multiple patient groups by using novel machine learning (ML) approaches and large-scale real-world data. METHODS: We used electronic health records of patients with asthma followed at the Cleveland Clinic between 2010 and 2021. We used k-prototype unsupervised ML to develop a clustering model where predictors were age, sex, race, body mass index, prebronchodilator and postbronchodilator spirometry measurements, and the usage of inhaled/systemic steroids. We applied elbow and silhouette plots to select the optimal number of clusters. These clusters were then evaluated through LightGBM's supervised ML approach on their cross-validated F1 score to support their distinctiveness. RESULTS: Data from 13,498 patients with asthma with available postbronchodilator spirometry measurements were extracted to identify 5 stable clusters. Cluster 1 included a young nonsevere asthma population with normal lung function and higher frequency of acute exacerbation (0.8 /patient-year). Cluster 2 had the highest body mass index (mean ± SD, 44.44 ± 7.83 kg/m2), and the highest proportion of females (77.5%) and Blacks (28.9%). Cluster 3 comprised patients with normal lung function. Cluster 4 included patients with lower percent of predicted FEV1 of 77.03 (12.79) and poor response to bronchodilators. Cluster 5 had the lowest percent of predicted FEV1 of 68.08 (15.02), the highest postbronchodilator reversibility, and the highest proportion of severe asthma (44.9%) and blood eosinophilia (>300 cells/µL) (34.8%). CONCLUSIONS: Using real-world data and unsupervised ML, we classified asthma into 5 clinically important subphenotypes where group-specific asthma treatment and management strategies can be designed and deployed.

The ChatGPT conundrum: Human-generated scientific manuscripts misidentified as AI creations by AI text detection tool.

AI Chat Bots such as ChatGPT are revolutionizing our AI capabilities, especially in text generation, to help expedite many tasks, but they introduce new dilemmas. The detection of AI-generated text has become a subject of great debate considering the AI text detector's known and unexpected limitations. Thus far, much research in this area has focused on the detection of AI-generated text; however, the goal of this study was to evaluate the opposite scenario, an AI-text detection tool's ability to discriminate human-generated text. Thousands of abstracts from several of the most well-known scientific journals were used to test the predictive capabilities of these detection tools, assessing abstracts from 1980 to 2023. We found that the AI text detector erroneously identified up to 8% of the known real abstracts as AI-generated text. This further highlights the current limitations of such detection tools and argues for novel detectors or combined approaches that can address this shortcoming and minimize its unanticipated consequences as we navigate this new AI landscape.

Machine learning and artificial intelligence for the diagnosis of infectious diseases in immunocompromised patients.

PURPOSE OF REVIEW: Immunocompromised patients are at high risk for infection. During the coronavirus disease (COVID-19) pandemic, immunocompromised patients exhibited increased odds of intensive care unit admission and death. Early pathogen identification is essential to mitigating infection related risk in immunocompromised patients. Artificial intelligence (AI) and machine learning (ML) have tremendous appeal to address unmet diagnostic needs. These AI/ML tools often rely on the wealth of data found in healthcare to enhance our ability to identify clinically significant patterns of disease. To this end, our review provides an overview of the current AI/ML landscape as it applies to infectious disease testing with emphasis on immunocompromised patients. RECENT FINDINGS: Examples include AI/ML for predicting sepsis in high risk burn patients. Likewise, ML is utilized to analyze complex host-response proteomic data to predict respiratory infections including COVID-19. These same approaches have also been applied for pathogen identification of bacteria, viruses, and hard to detect fungal microbes. Future uses of AI/ML may include integration of predictive analytics in point-of-care (POC) testing and data fusion applications. SUMMARY: Immunocompromised patients are at high risk for infections. AI/ML is transforming infectious disease testing and has great potential to address challenges encountered in the immune compromised population.

Preface: Artificial intelligence (AI), machine learning ML) and digital pathology integration are the next major chapter in our diagnostic pathology and laboratory medicine arena.

This timely captivating topic is organized and presented in this special issue of the journal of Seminar in diagnostic pathology. This special issue will be dedicated to the utilization of machine learning within the digital pathology and laboratory medicine fields. Special thanks to all the authors whose contributions to this review series has not only enhanced our overall understanding of this exciting new field but will also enrich the reader's understanding of this important discipline.

Artificial intelligence and machine learning overview in pathology & laboratory medicine: A general review of data preprocessing and basic supervised concepts.

Machine learning (ML) is becoming an integral aspect of several domains in medicine. Yet, most pathologists and laboratory professionals remain unfamiliar with such tools and are unprepared for their inevitable integration. To bridge this knowledge gap, we present an overview of key elements within this emerging data science discipline. First, we will cover general, well-established concepts within ML, such as data type concepts, data preprocessing methods, and ML study design. We will describe common supervised and unsupervised learning algorithms and their associated common machine learning terms (provided within a comprehensive glossary of terms that are discussed within this review). Overall, this review will offer a broad overview of the key concepts and algorithms in machine learning, with a focus on pathology and laboratory medicine. The objective is to provide an updated useful reference for those new to this field or those who require a refresher.

Common statistical concepts in the supervised Machine Learning arena.

One of the core elements of Machine Learning (ML) is statistics and its embedded foundational rules and without its appropriate integration, ML as we know would not exist. Various aspects of ML platforms are based on statistical rules and most notably the end results of the ML model performance cannot be objectively assessed without appropriate statistical measurements. The scope of statistics within the ML realm is rather broad and cannot be adequately covered in a single review article. Therefore, here we will mainly focus on the common statistical concepts that pertain to supervised ML (i.e. classification and regression) along with their interdependencies and certain limitations.

Machine learning in the coagulation and hemostasis arena: an overview and evaluation of methods, review of literature, and future directions.

Artificial Intelligence and machine-learning (ML) studies are increasingly populating the life science space and some have also started to integrate certain clinical decision support tasks. However, most of the activities within this space understandably remain within the investigational domain and are not yet ready for broad use in healthcare. In short, artificial intelligence/ML is still in an infancy stage within the healthcare arena, and we are nowhere near reaching its full potential. Various factors have contributed to this slow adoption rate within healthcare, which include but are not limited to data accessibility and integrity issues, paucity of specialized data science personnel, certain regulatory measures, and various voids within the ML operational platform domain. However, these obstacles and voids have also introduced us to certain opportunities to better understand this arena as we fully embark on this new journey, which undoubtedly will become a major part of our future patient care activities. Considering the aforementioned needs, this review will be concentrating on various ML studies within the coagulation and hemostasis space to better understand their shared study needs, findings, and limitations. However, the ML needs within this subspecialty of medicine are not unique and most of these needs, voids, and limitations also apply to the other medical disciplines. Therefore, this review will not only concentrate on introducing the audience to ML concepts and ML study design elements but also on where the future within this arena in medicine is leading us.

Comparative performance of two automated machine learning platforms for COVID-19 detection by MALDI-TOF-MS.

The 2019 novel coronavirus infectious disease (COVID-19) pandemic has resulted in an unsustainable need for diagnostic tests. Currently, molecular tests are the accepted standard for the detection of SARS-CoV-2. Mass spectrometry (MS) enhanced by machine learning (ML) has recently been postulated to serve as a rapid, high-throughput, and low-cost alternative to molecular methods. Automated ML is a novel approach that could move mass spectrometry techniques beyond the confines of traditional laboratory settings. However, it remains unknown how different automated ML platforms perform for COVID-19 MS analysis. To this end, the goal of our study is to compare algorithms produced by two commercial automated ML platforms (Platforms A and B). Our study consisted of MS data derived from 361 subjects with molecular confirmation of COVID-19 status including SARS-CoV-2 variants. The top optimized ML model with respect to positive percent agreement (PPA) within Platforms A and B exhibited an accuracy of 94.9%, PPA of 100%, negative percent agreement (NPA) of 93%, and an accuracy of 91.8%, PPA of 100%, and NPA of 89%, respectively. These results illustrate the MS method's robustness against SARS-CoV-2 variants and highlight similarities and differences in automated ML platforms in producing optimal predictive algorithms for a given dataset.

Identification of Endogenous Peptides in Nasal Swab Transport Media used in MALDI-TOF-MS Based COVID-19 Screening.

Mass spectrometry (MS) based diagnostic detection of 2019 novel coronavirus infectious disease (COVID-19) has been postulated to be a useful alternative to classical PCR based diagnostics. These MS based approaches have the potential to be both rapid and sensitive and can be done on-site without requiring a dedicated laboratory or depending on constrained supply chains (i.e., reagents and consumables). Matrix-assisted laser desorption ionization (MALDI)-time-of-flight (TOF) MS has a long and established history of microorganism detection and systemic disease assessment. Previously, we have shown that automated machine learning (ML) enhanced MALDI-TOF-MS screening of nasal swabs can be both sensitive and specific for COVID-19 detection. The underlying molecules responsible for this detection are generally unknown nor are they required for this automated ML platform to detect COVID-19. However, the identification of these molecules is important for understanding both the mechanism of detection and potentially the biology of the underlying infection. Here, we used nanoscale liquid chromatography tandem MS to identify endogenous peptides found in nasal swab saline transport media to identify peptides in the same the mass over charge (m/z) values observed by the MALDI-TOF-MS method. With our peptidomics workflow, we demonstrate that we can identify endogenous peptides and endogenous protease cut sites. Further, we show that SARS-CoV-2 viral peptides were not readily detected and are highly unlikely to be responsible for the accuracy of MALDI based SARS-CoV-2 diagnostics. Further analysis with more samples will be needed to validate our findings, but the methodology proves to be promising.

Prediction of Tuberculosis Using an Automated Machine Learning Platform for Models Trained on Synthetic Data.

High-quality medical data is critical to the development and implementation of machine learning (ML) algorithms in healthcare; however, security, and privacy concerns continue to limit access. We sought to determine the utility of "synthetic data" in training ML algorithms for the detection of tuberculosis (TB) from inflammatory biomarker profiles. A retrospective dataset (A) comprised of 278 patients was used to generate synthetic datasets (B, C, and D) for training models prior to secondary validation on a generalization dataset. ML models trained and validated on the Dataset A (real) demonstrated an accuracy of 90%, a sensitivity of 89% (95% CI, 83-94%), and a specificity of 100% (95% CI, 81-100%). Models trained using the optimal synthetic dataset B showed an accuracy of 91%, a sensitivity of 93% (95% CI, 87-96%), and a specificity of 77% (95% CI, 50-93%). Synthetic datasets C and D displayed diminished performance measures (respective accuracies of 71% and 54%). This pilot study highlights the promise of synthetic data as an expedited means for ML algorithm development.

Analyzing Anatomic Pathology On-Call Summaries to Improve Resident Education.

CONTEXT.­: Pathology on-call experiences help prepare trainees for successful transition from residency to independent practice, and as such are an integral component of training. However, few data exist on anatomic pathology resident on-call workload and experience. OBJECTIVE.­: To obtain an overall picture of the anatomic pathology on-call experience to inform and improve resident education. DESIGN.­: Retrospective and prospective review of daily anatomic pathology on-call summaries from July 2016 to June 2020. RESULTS.­: During the first 2 years of the study (ie, retrospective portion), only 19% of on-call summaries (138 of 730) were available for review. After interventions, the on-call summary submission rate jumped to 98% (716 of 731). After-hours calls were most frequent on weekdays from 5 to 8 pm. The most frequent requests were for frozen sections (55%; 619 of 1125 calls), inquiries regarding disposition of fresh placentas (13%; 148 of 1125 calls), and inquiries regarding disposition of various other specimens (6%; 68 of 1125 calls). After-hours frozen section requests were most frequent for gynecologic and head and neck specimens. Notably, a significant number of after-hours calls were recurring preanalytic issues amenable to system-level improvements. We were able to eliminate the most common of these recurring preanalytic calls with stepwise interventions. CONCLUSIONS.­: To our knowledge, this is the first study analyzing the anatomic pathology resident on-call experience. In addition to obtaining a broad overview of the residents' clinical exposure on this service, we identified and resolved issues critical to optimal patient care (eg, inconsistent "patient hand-off") and improved the resident on-call experience (eg, fewer preanalytic calls increased resident time for other clinical, educational, or wellness activities).

Automated machine learning for endemic active tuberculosis prediction from multiplex serological data.

Serological diagnosis of active tuberculosis (TB) is enhanced by detection of multiple antibodies due to variable immune responses among patients. Clinical interpretation of these complex datasets requires development of suitable algorithms, a time consuming and tedious undertaking addressed by the automated machine learning platform MILO (Machine Intelligence Learning Optimizer). MILO seamlessly integrates data processing, feature selection, model training, and model validation to simultaneously generate and evaluate thousands of models. These models were then further tested for generalizability on out-of-sample secondary and tertiary datasets. Out of 31 antigens evaluated, a 23-antigen model was the most robust on both the secondary dataset (TB vs healthy) and the tertiary dataset (TB vs COPD) with sensitivity of 90.5% and respective specificities of 100.0% and 74.6%. MILO represents a user-friendly, end-to-end solution for automated generation and deployment of optimized models, ideal for applications where rapid clinical implementation is critical such as emerging infectious diseases.

Immune response dynamics in COVID-19 patients to SARS-CoV-2 and other human coronaviruses.

COVID-19 serological test must have high sensitivity as well as specificity to rule out cross-reactivity with common coronaviruses (HCoVs). We have developed a quantitative multiplex test, measuring antibodies against spike (S) proteins of SARS-CoV-2, SARS-CoV, MERS-CoV, and common human coronavirus strains (229E, NL63, OC43, HKU1), and nucleocapsid (N) protein of SARS-CoV viruses. Receptor binding domain of S protein of SARS-CoV-2 (S-RBD), and N protein, demonstrated sensitivity (94% and 92.5%, respectively) in COVID-19 patients (n = 53), with 98% specificity in non-COVID-19 respiratory-disease (n = 98), and healthy-controls (n = 129). Anti S-RBD and N antibodies appeared five to ten days post-onset of symptoms, peaking at approximately four weeks. The appearance of IgG and IgM coincided while IgG subtypes, IgG1 and IgG3 appeared soon after the total IgG; IgG2 and IgG4 remained undetectable. Several inflammatory cytokines/chemokines were found to be elevated in many COVID-19 patients (e.g., Eotaxin, Gro-α, CXCL-10 (IP-10), RANTES (CCL5), IL-2Rα, MCP-1, and SCGF-b); CXCL-10 was elevated in all. In contrast to antibody titers, levels of CXCL-10 decreased with the improvement in patient health suggesting it as a candidate for disease resolution. Importantly, anti-N antibodies appear before S-RBD and differentiate between vaccinated and infected people-current vaccines (and several in the pipeline) are S protein-based.

Machine learning in health care and laboratory medicine: General overview of supervised learning and Auto-ML.

Artificial Intelligence (AI) and machine learning (ML) have now spawned a new field within health care and health science research. These new predictive analytics tools are starting to change various facets of our clinical care domains including the practice of laboratory medicine. Many of these ML tools and studies are also starting to populate our literature landscape as we know it but unfamiliarity of the average reader to the basic knowledge and critical concepts within AI/ML is now demanding a need to better prepare our audience to such relatively unfamiliar concepts. A fundamental knowledge of such platforms will inevitably enhance cross-disciplinary literacy and ultimately lead to enhanced integration and understanding of such tools within our discipline. In this review, we provide a general outline of AI/ML along with an overview of the fundamental concepts of ML categories, specifically supervised, unsupervised, and reinforcement learning. Additionally, since the vast majority of our current approaches within ML in laboratory medicine and health care involve supervised algorithms, we will predominantly concentrate on such platforms. Finally, the need for making such tools more accessible to the average investigator is becoming a major driving force for the need of automation within these ML platforms. This has now given rise to the automated ML (Auto-ML) world which will undoubtedly help shape the future of ML within health care. Hence, an overview of Auto-ML is also covered within this manuscript which will hopefully enrich the reader's understanding, appreciation, and the need for embracing such tools.

Effects of Image Quantity and Image Source Variation on Machine Learning Histology Differential Diagnosis Models.

AIMS: Histology, the microscopic study of normal tissues, is a crucial element of most medical curricula. Learning tools focused on histology are very important to learners who seek diagnostic competency within this important diagnostic arena. Recent developments in machine learning (ML) suggest that certain ML tools may be able to benefit this histology learning platform. Here, we aim to explore how one such tool based on a convolutional neural network, can be used to build a generalizable multi-classification model capable of classifying microscopic images of human tissue samples with the ultimate goal of providing a differential diagnosis (a list of look-alikes) for each entity. METHODS: We obtained three institutional training datasets and one generalizability test dataset, each containing images of histologic tissues in 38 categories. Models were trained on data from single institutions, low quantity combinations of multiple institutions, and high quantity combinations of multiple institutions. Models were tested against withheld validation data, external institutional data, and generalizability test images obtained from Google image search. Performance was measured with macro and micro accuracy, sensitivity, specificity, and f1-score. RESULTS: In this study, we were able to show that such a model's generalizability is dependent on both the training data source variety and the total number of training images used. Models which were trained on 760 images from only a single institution performed well on withheld internal data but poorly on external data (lower generalizability). Increasing data source diversity improved generalizability, even when decreasing data quantity: models trained on 684 images, but from three sources improved generalization accuracy between 4.05% and 18.59%. Maintaining this diversity and increasing the quantity of training images to 2280 further improved generalization accuracy between 16.51% and 32.79%. CONCLUSIONS: This pilot study highlights the significance of data diversity within such studies. As expected, optimal models are those that incorporate both diversity and quantity into their platforms.s.

Novel application of automated machine learning with MALDI-TOF-MS for rapid high-throughput screening of COVID-19: a proof of concept.

The 2019 novel coronavirus infectious disease (COVID-19) pandemic caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has created an unsustainable need for molecular diagnostic testing. Molecular approaches such as reverse transcription (RT) polymerase chain reaction (PCR) offers highly sensitive and specific means to detect SARS-CoV-2 RNA, however, despite it being the accepted "gold standard", molecular platforms often require a tradeoff between speed versus throughput. Matrix assisted laser desorption ionization (MALDI)-time of flight (TOF)-mass spectrometry (MS) has been proposed as a potential solution for COVID-19 testing and finding a balance between analytical performance, speed, and throughput, without relying on impacted supply chains. Combined with machine learning (ML), this MALDI-TOF-MS approach could overcome logistical barriers encountered by current testing paradigms. We evaluated the analytical performance of an ML-enhanced MALDI-TOF-MS method for screening COVID-19. Residual nasal swab samples from adult volunteers were used for testing and compared against RT-PCR. Two optimized ML models were identified, exhibiting accuracy of 98.3%, positive percent agreement (PPA) of 100%, negative percent agreement (NPA) of 96%, and accuracy of 96.6%, PPA of 98.5%, and NPA of 94% respectively. Machine learning enhanced MALDI-TOF-MS for COVID-19 testing exhibited performance comparable to existing commercial SARS-CoV-2 tests.

Enhancing Military Burn- and Trauma-Related Acute Kidney Injury Prediction Through an Automated Machine Learning Platform and Point-of-Care Testing.

CONTEXT.­: Delayed recognition of acute kidney injury (AKI) results in poor outcomes in military and civilian burn-trauma care. Poor predictive ability of urine output (UOP) and creatinine contribute to the delayed recognition of AKI. OBJECTIVE.­: To determine the impact of point-of-care (POC) AKI biomarker enhanced by machine learning (ML) algorithms in burn-injured and trauma patients. DESIGN.­: We conducted a 2-phased study to develop and validate a novel POC device for measuring neutrophil gelatinase-associated lipocalin (NGAL) and creatinine from blood samples. In phase I, 40 remnant plasma samples were used to evaluate the analytic performance of the POC device. Next, phase II enrolled 125 adults with either burns that were 20% or greater of total body surface area or nonburn trauma with suspicion of AKI for clinical validation. We applied an automated ML approach to develop models predicting AKI, using a combination of NGAL, creatinine, and/or UOP as features. RESULTS.­: Point-of-care NGAL (mean [SD] bias: 9.8 [38.5] ng/mL, P = .10) and creatinine results (mean [SD] bias: 0.28 [0.30] mg/dL, P = .18) were comparable to the reference method. NGAL was an independent predictor of AKI (odds ratio, 1.6; 95% CI, 0.08-5.20; P = .01). The optimal ML model achieved an accuracy, sensitivity, and specificity of 96%, 92.3%, and 97.7%, respectively, with NGAL, creatinine, and UOP as features. Area under the receiver operator curve was 0.96. CONCLUSIONS.­: Point-of-care NGAL testing is feasible and produces results comparable to reference methods. Machine learning enhanced the predictive performance of AKI biomarkers including NGAL and was superior to the current techniques.

Automated En Masse Machine Learning Model Generation Shows Comparable Performance as Classic Regression Models for Predicting Delayed Graft Function in Renal Allografts.

BACKGROUND: Several groups have previously developed logistic regression models for predicting delayed graft function (DGF). In this study, we used an automated machine learning (ML) modeling pipeline to generate and optimize DGF prediction models en masse. METHODS: Deceased donor renal transplants at our institution from 2010 to 2018 were included. Input data consisted of 21 donor features from United Network for Organ Sharing. A training set composed of ~50%/50% split in DGF-positive and DGF-negative cases was used to generate 400 869 models. Each model was based on 1 of 7 ML algorithms (gradient boosting machine, k-nearest neighbor, logistic regression, neural network, naive Bayes, random forest, support vector machine) with various combinations of feature sets and hyperparameter values. Performance of each model was based on a separate secondary test dataset and assessed by common statistical metrics. RESULTS: The best performing models were based on neural network algorithms, with the highest area under the receiver operating characteristic curve of 0.7595. This model used 10 out of the original 21 donor features, including age, height, weight, ethnicity, serum creatinine, blood urea nitrogen, hypertension history, donation after cardiac death status, cause of death, and cold ischemia time. With the same donor data, the highest area under the receiver operating characteristic curve for logistic regression models was 0.7484, using all donor features. CONCLUSIONS: Our automated en masse ML modeling approach was able to rapidly generate ML models for DGF prediction. The performance of the ML models was comparable with classic logistic regression models.

Novel application of an automated-machine learning development tool for predicting burn sepsis: proof of concept.

Sepsis is the primary cause of burn-related mortality and morbidity. Traditional indicators of sepsis exhibit poor performance when used in this unique population due to their underlying hypermetabolic and inflammatory response following burn injury. To address this challenge, we developed the Machine Intelligence Learning Optimizer (MILO), an automated machine learning (ML) platform, to automatically produce ML models for predicting burn sepsis. We conducted a retrospective analysis of 211 adult patients (age ≥ 18 years) with severe burn injury (≥ 20% total body surface area) to generate training and test datasets for ML applications. The MILO approach was compared against an exhaustive "non-automated" ML approach as well as standard statistical methods. For this study, traditional multivariate logistic regression (LR) identified seven predictors of burn sepsis when controlled for age and burn size (OR 2.8, 95% CI 1.99-4.04, P = 0.032). The area under the ROC (ROC-AUC) when using these seven predictors was 0.88. Next, the non-automated ML approach produced an optimal model based on LR using 16 out of the 23 features from the study dataset. Model accuracy was 86% with ROC-AUC of 0.96. In contrast, MILO identified a k-nearest neighbor-based model using only five features to be the best performer with an accuracy of 90% and a ROC-AUC of 0.96. Machine learning augments burn sepsis prediction. MILO identified models more quickly, with less required features, and found to be analytically superior to traditional ML approaches. Future studies are needed to clinically validate the performance of MILO-derived ML models for sepsis prediction.