Using Deep Learning to Automate Eosinophil Counting in Pediatric Ulcerative Colitis Histopathological Images.

Background: Accurate identification of inflammatory cells from mucosal histopathology images is important in diagnosing ulcerative colitis. The identification of eosinophils in the colonic mucosa has been associated with disease course. Cell counting is not only time-consuming but can also be subjective to human biases. In this study we developed an automatic eosinophilic cell counting tool from mucosal histopathology images, using deep learning. Method: Four pediatric IBD pathologists from two North American pediatric hospitals annotated 530 crops from 143 standard-of-care hematoxylin and eosin (H & E) rectal mucosal biopsies. A 305/75 split was used for training/validation to develop and optimize a U-Net based deep learning model, and 150 crops were used as a test set. The U-Net model was then compared to SAU-Net, a state-of-the-art U-Net variant. We undertook post-processing steps, namely, (1) the pixel-level probability threshold, (2) the minimum number of clustered pixels to designate a cell, and (3) the connectivity. Experiments were run to optimize model parameters using AUROC and cross-entropy loss as the performance metrics. Results: The F1-score was 0.86 (95%CI:0.79-0.91) (Precision: 0.77 (95%CI:0.70-0.83), Recall: 0.96 (95%CI:0.93-0.99)) to identify eosinophils as compared to an F1-score of 0.2 (95%CI:0.13-0.26) for SAU-Net (Precision: 0.38 (95%CI:0.31-0.46), Recall: 0.13 (95%CI:0.08-0.19)). The inter-rater reliability was 0.96 (95%CI:0.93-0.97). The correlation between two pathologists and the algorithm was 0.89 (95%CI:0.82-0.94) and 0.88 (95%CI:0.80-0.94) respectively. Conclusion: Our results indicate that deep learning-based automated eosinophilic cell counting can obtain a robust level of accuracy with a high degree of concordance with manual expert annotations.

Therapeutic preference for Alzheimer's disease treatments: a discrete choice experiment with caregivers and neurologists.

BACKGROUND: Alzheimer's disease (AD) is a major global health crisis in need of more effective therapies. However, difficult choices to optimize value-based care will need to be made. While identifying preferred therapeutic attributes of new AD therapies is necessary, few studies have explored how preferences may vary between the stakeholders. In this study, the trade-offs among key attributes of amyloid plaque-lowering therapies for AD were assessed using a discrete choice experiment (DCE) and compared between caregivers and neurologists. METHODS: An initial pilot study was conducted to identify the potentially relevant features of a new therapy. The DCE evaluated seven drug attributes: clinical effects in terms of delay in AD progression over the standard of care (SOC), variation in clinical effects, biomarker response (achieving amyloid plaque clearance on PET scan), amyloid-related imaging abnormalities-edema (ARIA-E), duration of therapy, need for treatment titration as well as route, and frequency of drug administration. Respondents were then randomly presented with 12 choice sets of treatment options and asked to select their preferred option in each choice set. Hierarchical Bayesian regression modeling was used to estimate weighted preference attributes, which were presented as mean partial utility scores (pUS), with higher scores suggesting an increased preference. RESULTS: Both caregivers (n = 137) and neurologists (n = 161) considered clinical effects (mean pUS = 0.47 and 0.82) and a 5% incremental in ARIA-E (mean pUS = - 0.26 and - 0.52) to be highly impactful determinants of therapeutic choice. In contrast, variation in clinical effects (mean pUS = 0.12 and 0.14) and treatment duration (mean pUS = - 0.02 and - 0.13) were the least important characteristics of any new treatment. Neurologists' also indicated that subcutaneous drug delivery (mean pUS = 0.42 vs. 0.07) and administration every 4 weeks (mean pUS = 1.0 vs. 0.20) are highly desirable therapeutic features. Respondents were willing to accept up to a 9% increment in ARIA-E for one additional year of delayed progression. CONCLUSIONS: Caregivers and neurologists considered incremental clinical benefit over SOC and safety to be highly desirable qualities for a new drug that could clear amyloid plaques and delay clinical progression and indicated a willingness to accept incremental ARIA-E to achieve additional clinical benefits.

Predicting recurrence and recurrence-free survival in high-grade endometrial cancer using machine learning.

OBJECTIVE: To develop machine-learning models to predict recurrence and time-to-recurrence in high-grade endometrial cancer (HGEC) following surgery and tailored adjuvant treatment. METHODS: Data were retrospectively collected across eight Canadian centers including 1237 patients. Four models were trained to predict recurrence: random forests, boosted trees, and two neural networks. Receiver operating characteristic curves were used to select the best model based on the highest area under the curve (AUC). For time to recurrence, we compared random forests and Least Absolute Shrinkage and Selection Operator (LASSO) model to Cox proportional hazards. RESULTS: The random forest was the best model to predict recurrence in HGEC; the AUCs were 85.2%, 74.1%, and 71.8% in the training, validation, and test sets, respectively. The top five predictors were: stage, uterus height, specimen weight, adjuvant chemotherapy, and preoperative histology. Performance increased to 77% and 80% when stratified by Stage III and IV, respectively. For time to recurrence, there was no difference between the LASSO and Cox proportional hazards models (c-index 71%). The random forest had a c-index of 60.5%. CONCLUSIONS: A bootstrap random forest model may be a more accurate technique to predict recurrence in HGEC using multiple clinicopathologic factors. For time to recurrence, machine-learning methods performed similarly to the Cox proportional hazards model.

Assessment of Machine Learning-Based Medical Directives to Expedite Care in Pediatric Emergency Medicine.

Importance: Increased wait times and long lengths of stay in emergency departments (EDs) are associated with poor patient outcomes. Systems to improve ED efficiency would be useful. Specifically, minimizing the time to diagnosis by developing novel workflows that expedite test ordering can help accelerate clinical decision-making. Objective: To explore the use of machine learning-based medical directives (MLMDs) to automate diagnostic testing at triage for patients with common pediatric ED diagnoses. Design, Setting, and Participants: Machine learning models trained on retrospective electronic health record data were evaluated in a decision analytical model study conducted at the ED of the Hospital for Sick Children Toronto, Canada. Data were collected on all patients aged 0 to 18 years presenting to the ED from July 1, 2018, to June 30, 2019 (77 219 total patient visits). Exposure: Machine learning models were trained to predict the need for urinary dipstick testing, electrocardiogram, abdominal ultrasonography, testicular ultrasonography, bilirubin level testing, and forearm radiographs. Main Outcomes and Measures: Models were evaluated using area under the receiver operator curve, true-positive rate, false-positive rate, and positive predictive values. Model decision thresholds were determined to limit the total number of false-positive results and achieve high positive predictive values. The time difference between patient triage completion and test ordering was assessed for each use of MLMD. Error rates were analyzed to assess model bias. In addition, model explainability was determined using Shapley Additive Explanations values. Results: There was a total of 42 238 boys (54.7%) included in model development; mean (SD) age of the children was 5.4 (4.8) years. Models obtained high area under the receiver operator curve (0.89-0.99) and positive predictive values (0.77-0.94) across each of the use cases. The proposed implementation of MLMDs would streamline care for 22.3% of all patient visits and make test results available earlier by 165 minutes (weighted mean) per affected patient. Model explainability for each MLMD demonstrated clinically relevant features having the most influence on model predictions. Models also performed with minimal to no sex bias. Conclusions and Relevance: The findings of this study suggest the potential for clinical automation using MLMDs. When integrated into clinical workflows, MLMDs may have the potential to autonomously order common ED tests early in a patient's visit with explainability provided to patients and clinicians.

A Research Ethics Framework for the Clinical Translation of Healthcare Machine Learning.

The application of artificial intelligence and machine learning (ML) technologies in healthcare have immense potential to improve the care of patients. While there are some emerging practices surrounding responsible ML as well as regulatory frameworks, the traditional role of research ethics oversight has been relatively unexplored regarding its relevance for clinical ML. In this paper, we provide a comprehensive research ethics framework that can apply to the systematic inquiry of ML research across its development cycle. The pathway consists of three stages: (1) exploratory, hypothesis-generating data access; (2) silent period evaluation; (3) prospective clinical evaluation. We connect each stage to its literature and ethical justification and suggest adaptations to traditional paradigms to suit ML while maintaining ethical rigor and the protection of individuals. This pathway can accommodate a multitude of research designs from observational to controlled trials, and the stages can apply individually to a variety of ML applications.

Personalized application of machine learning algorithms to identify pediatric patients at risk for recurrent ureteropelvic junction obstruction after dismembered pyeloplasty.

PURPOSE: To develop a model that predicts whether a child will develop a recurrent obstruction after pyeloplasty, determine their survival risk score, and expected time to re-intervention using machine learning (ML). METHODS: We reviewed patients undergoing pyeloplasty from 2008 to 2020 at our institution, including all children and adolescents younger than 18 years. We developed a two-stage machine learning model from 34 clinical fields, which included patient characteristics, ultrasound findings, and anatomical variation. We fit and trained with a logistic lasso model for binary cure model and subsequent survival model. Feature importance on the model was determined with post-selection inference. Performance metrics included area under the receiver-operating-characteristic (AUROC), concordance, and leave-one-out cross validation. RESULTS: A total of 543 patients were identified, with a median preoperative and postoperative anteroposterior diameter of 23 and 10 mm, respectively. 39 of 232 patients included in the survival model required re-intervention. The cure and survival models performed well with a leave-one-out cross validation AUROC and concordance of 0.86 and 0.78, respectively. Post-selective inference showed that larger anteroposterior diameter at the second post-op follow-up, and anatomical variation in the form of concurrent anomalies were significant model features predicting negative outcomes. The model can be used at https://sickkidsurology.shinyapps.io/PyeloplastyReOpRisk/ . CONCLUSION: Our ML-based model performed well in predicting the risk of and time to re-intervention after pyeloplasty. The implementation of this ML-based approach is novel in pediatric urology and will likely help achieve personalized risk stratification for patients undergoing pyeloplasty. Further real-world validation is warranted.

The telomere length landscape of prostate cancer.

Replicative immortality is a hallmark of cancer, and can be achieved through telomere lengthening and maintenance. Although the role of telomere length in cancer has been well studied, its association to genomic features is less well known. Here, we report the telomere lengths of 392 localized prostate cancer tumours and characterize their relationship to genomic, transcriptomic and proteomic features. Shorter tumour telomere lengths are associated with elevated genomic instability, including single-nucleotide variants, indels and structural variants. Genes involved in cell proliferation and signaling are correlated with tumour telomere length at all levels of the central dogma. Telomere length is also associated with multiple clinical features of a tumour. Longer telomere lengths in non-tumour samples are associated with a lower rate of biochemical relapse. In summary, we describe the multi-level integration of telomere length, genomics, transcriptomics and proteomics in localized prostate cancer.

Automatically disambiguating medical acronyms with ontology-aware deep learning.

Modern machine learning (ML) technologies have great promise for automating diverse clinical and research workflows; however, training them requires extensive hand-labelled datasets. Disambiguating abbreviations is important for automated clinical note processing; however, broad deployment of ML for this task is restricted by the scarcity and imbalance of labeled training data. In this work we present a method that improves a model's ability to generalize through novel data augmentation techniques that utilizes information from biomedical ontologies in the form of related medical concepts, as well as global context information within the medical note. We train our model on a public dataset (MIMIC III) and test its performance on automatically generated and hand-labelled datasets from different sources (MIMIC III, CASI, i2b2). Together, these techniques boost the accuracy of abbreviation disambiguation by up to 17% on hand-labeled data, without sacrificing performance on a held-out test set from MIMIC III.

Accurate Classification of Pediatric Colonic Inflammatory Bowel Disease Subtype Using a Random Forest Machine Learning Classifier.

BACKGROUND: The pediatric inflammatory bowel disease (PIBD) classes algorithm was developed to bring consistency to labelling of colonic IBD, but labels are exclusively based on features atypical for ulcerative colitis (UC). AIM: The aim of the study was to develop an algorithm and identify features that discriminate between pediatric UC and colonic Crohn disease (CD). METHODS: Baseline clinical, endoscopic, radiologic, and histologic data, including the PIBD class features in 74 colonic IBD (56: UC, 18: colonic CD) patients were collected. The PIBD class features and additional features common to UC were used to perform initial clustering, using similarity network fusion (SNF). We trained a Random Forest (RF) classifier on the full dataset and used a leave-one-out approach to evaluate model accuracy. The top-features were used to build a new classifier, which we tested on 15 previously unused patients. We then performed clustering with SNF on the top RF features and assessed ability to discriminate between UC and colonic-CD independent of a supervised model. RESULTS: The initial SNF clustering with 58 patients demonstrated 2 groups: group 1 (n = 39, 90% UC) and group 2 (n = 19, 68% colonic-CD). Our RF classifier correctly labelled 97% of the 58 patients based on leave-one-out cross validation and identified the 7 most important features (3 histological and 4 endoscopic) to clinically distinguish these groups. We trained a new RF classifier with the top 7 features and found 100% accuracy in a set of 15 held-out patients. Finally, post hoc clustering with these 7 features revealed 2 groups of patients: group 1 (n = 55, 98% UC) and group 2 (n = 18, 94% colonic-CD). CONCLUSIONS: A combination of supervised and unsupervised analyses identified a short list of features, which consistently distinguish UC from colonic CD. Future directions include validation in other populations.

Trends and relevance in the bladder and bowel dysfunction literature: PlumX metrics contrasted with fragility indicators.

INTRODUCTION: The concepts of fragility index (FI) and fragility quotient (FQ) have been previously described. PlumX metrics encompass online "footprints" of research in addition to traditional citations. Herein we explore PlumX metrics against the quality of BBD literature. OBJECTIVE: To explore altmetrics against the quality of bladder and bowel dysfunction (BBD) literature. STUDY DESIGN: A literature search was conducted using Pubmed, Medline, Embase for BBD and related terms. A total of 54,045 abstracts were screened, followed by 693 full text reviews and data extraction from 126. Studies were included if they reported on 2 groups being compared, had dichotomous outcomes, and had significant results. RESULTS: The median FI score was 4 (0-500) and there were 20 studies which had a FI of 0. The FQ had a median of 0.04 (0-0.32). PlumX usage was 263 ± 540, captures were 45 ± 60 and social media attention was 2 ± 2. Overall, 42% of papers were clinical trials (RCTs). When compared to other study designs, we noted a significant difference in PlumX captures (57 ± 72 RCT vs. 35 ± 47 other; p = 0.03). RCTs had higher usage, social media engagement and citations however, the differences were not significant. H-Index had a significant correlation with FI (p = 0.036), however correlations for PlumX usage and captures, while modestly positive (0.04-0.10) for the FI and FQ, were not significant. A comparison of FI and FQ by topic can be reviewed in the Summary Table. DISCUSSION: When considering the FI and FQ robustness indicators of the BBD literature, we found similarities when compared to other studies. It was reported that overall, the hydronephrosis literature was fragile with many studies requiring only a few events to nullify significance, regardless of the study design. Similarly, in a review of pediatric vesicoureteral reflux (VUR) clinical trials, results were also fragile. When comparing fragility measures to altmetric variables we noted that despite the growing popularity of altmetrics, citation counts, and h-indices remain the traditional measures to monitor research consumption. There has been a reported correlation between manuscript citation counts, author h-index, altmetrics measures in several specialties and across many domains of research including medical sciences, arts, and the humanities, however in the present study only weak correlations were noted. CONCLUSION: The body of BBD comparative studies is fragile in keeping with other pediatric urology literature populations. Despite fragile results, RCTs generate slightly moreattention as measured by select PlumX metrics. These results suggest the need for including fragility measures in our literature, aiming to focus attention towards more robust articles.

From Clinic to Computer and Back Again: Practical Considerations When Designing and Implementing Machine Learning Solutions for Pediatrics.

Purpose of review: Machine learning (ML), a branch of artificial intelligence, is influencing all fields in medicine, with an abundance of work describing its application to adult practice. ML in pediatrics is distinctly unique with clinical, technical, and ethical nuances limiting the direct translation of ML tools developed for adults to pediatric populations. To our knowledge, no work has yet focused on outlining the unique considerations that need to be taken into account when designing and implementing ML in pediatrics. Recent findings: The nature of varying developmental stages and the prominence of family-centered care lead to vastly different data-generating processes in pediatrics. Data heterogeneity and a lack of high-quality pediatric databases further complicate ML research. In order to address some of these nuances, we provide a common pipeline for clinicians and computer scientists to use as a foundation for structuring ML projects, and a framework for the translation of a developed model into clinical practice in pediatrics. Throughout these pathways, we also highlight ethical and legal considerations that must be taken into account when working with pediatric populations and data. Summary: Here, we describe a comprehensive outline of special considerations required of ML in pediatrics from project ideation to implementation. We hope this review can serve as a high-level guideline for ML scientists and clinicians alike to identify applications in the pediatric setting, generate effective ML solutions, and subsequently deliver them to patients, families, and providers.

A population-based study of the treatment effect of first-line ipilimumab for metastatic or unresectable melanoma.

Ipilimumab is an anti-CTLA4 monoclonal antibody with demonstrated efficacy for metastatic melanoma in randomized controlled trials, including in the first-line setting. Population-based outcomes directly compared with historic chemotherapy treatment in metastatic or unresectable melanoma are lacking. Using population-based data from the province of Ontario, the benefit of first-line ipilimumab was estimated by comparing outcomes of patients treated with first-line dacarbazine over the period 2007-2009 with patients treated over the period 2010-2015 with first-line ipilimumab. Cutaneous and noncutaneous cases were included. The administrative data set utilized was high-dimensional; meaning, there was a large number of variables relative to the sample size. To adjust for important confounders among the many available variables, we utilized a double-selection method, a modified machine learning algorithm to extract the important variables that were related to both survival times and treatment usage. Time-dependent treatment modeling was utilized. Among the 2793 melanoma patients receiving palliative treatment (systemic therapy, surgery, or radiation) in Ontario (2007-2015), there were 289 patients treated with first-line ipilimumab (2010-2015) and 175 patients treated with first-line dacarbazine (2007-2009). For first-line ipilimumab, the adjusted hazard ratio compared with dacarbazine for overall survival (OS) was 0.63 (95% confidence interval: 0.47-0.84) with a 1-year survival of 46.5 versus 18.9% with dacarbazine. In subgroup analysis, ipilimumab was associated with improved OS across groups (age, sex, comorbidity, income quintile, previous interferon). First-line ipilimumab was found to have a significant OS benefit compared with historical controls in a population including those patients not routinely included in clinical trials. The treatment effect was similar to randomized controlled trials, suggesting a meaningful benefit when utilized in a population-based setting.

Mammary molecular portraits reveal lineage-specific features and progenitor cell vulnerabilities.

The mammary epithelium depends on specific lineages and their stem and progenitor function to accommodate hormone-triggered physiological demands in the adult female. Perturbations of these lineages underpin breast cancer risk, yet our understanding of normal mammary cell composition is incomplete. Here, we build a multimodal resource for the adult gland through comprehensive profiling of primary cell epigenomes, transcriptomes, and proteomes. We define systems-level relationships between chromatin-DNA-RNA-protein states, identify lineage-specific DNA methylation of transcription factor binding sites, and pinpoint proteins underlying progesterone responsiveness. Comparative proteomics of estrogen and progesterone receptor-positive and -negative cell populations, extensive target validation, and drug testing lead to discovery of stem and progenitor cell vulnerabilities. Top epigenetic drugs exert cytostatic effects; prevent adult mammary cell expansion, clonogenicity, and mammopoiesis; and deplete stem cell frequency. Select drugs also abrogate human breast progenitor cell activity in normal and high-risk patient samples. This integrative computational and functional study provides fundamental insight into mammary lineage and stem cell biology.