Proposed Quality Indicators for Aspects of Pediatric Acute Otitis Media Management.

BACKGROUND: The high incidence of pediatric acute otitis media (AOM) makes the implications of overdiagnosis and overtreatment far-reaching. Quality indicators (QIs) for AOM are limited, drawing from generalized upper respiratory infection QIs, or locally developed benchmarks. Recognizing this, we sought to develop pediatric AOM QIs to build a foundation for future quality improvement efforts. METHODS: Candidate indicators (CIs) were extracted from existing guidelines and position statements. The modified RAND Corporation/University of California, Los Angeles (RAND/UCLA) appropriateness methodology was used to select the final QIs by an 11-member expert panel consisting of otolaryngology-head and neck surgeons, a pediatrician and family physician. RESULTS: Twenty-seven CIs were identified after literature review, with an additional CI developed by the expert panel. After the first round of evaluations, the panel agreed on 4 CIs as appropriate QIs. After an expert panel meeting and subsequent second round of evaluations, the panel agreed on 8 final QIs as appropriate measures of high-quality care. The 8 final QIs focus on topics of antimicrobial management, specialty referral, and tympanostomy tube counseling. CONCLUSIONS: Evidence of variable and substandard care persists in the diagnosis and management of pediatric AOM despite the existence of high-quality guidelines. This study proposes 8 QIs which compliment guideline recommendations and are meant to facilitate future quality improvement initiatives that can improve patient outcomes.

Comparing environmental impacts of deep-seabed and land-based mining: A defensible framework.

The crises of climate change and biodiversity loss are interlinked and must be addressed jointly. A proposed solution for reducing reliance on fossil fuels, and thus mitigating climate change, is the transition from conventional combustion-engine to electric vehicles. This transition currently requires additional mineral resources, such as nickel and cobalt used in car batteries, presently obtained from land-based mines. Most options to meet this demand are associated with some biodiversity loss. One proposal is to mine the deep seabed, a vast, relatively pristine and mostly unexplored region of our planet. Few comparisons of environmental impacts of solely expanding land-based mining versus extending mining to the deep seabed for the additional resources exist and for biodiversity only qualitative. Here, we present a framework that facilitates a holistic comparison of relative ecosystem impacts by mining, using empirical data from relevant environmental metrics. This framework (Environmental Impact Wheel) includes a suite of physicochemical and biological components, rather than a few selected metrics, surrogates, or proxies. It is modified from the "recovery wheel" presented in the International Standards for the Practice of Ecological Restoration to address impacts rather than recovery. The wheel includes six attributes (physical condition, community composition, structural diversity, ecosystem function, external exchanges and absence of threats). Each has 3-5 sub attributes, in turn measured with several indicators. The framework includes five steps: (1) identifying geographic scope; (2) identifying relevant spatiotemporal scales; (3) selecting relevant indicators for each sub-attribute; (4) aggregating changes in indicators to scores; and (5) generating Environmental Impact Wheels for targeted comparisons. To move forward comparisons of land-based with deep seabed mining, thresholds of the indicators that reflect the range in severity of environmental impacts are needed. Indicators should be based on clearly articulated environmental goals, with objectives and targets that are specific, measurable, achievable, relevant, and time bound.

American College of Rheumatology White Paper: The Effects of Climate Change on Rheumatic Conditions-An Evolving Landscape and a Path Forward.

OBJECTIVE: Increases in global temperatures and extreme weather events associated with climate change have complex yet poorly understood detrimental impacts on human health. We reviewed the current published literature on climate change-related effects and rheumatic conditions. METHODS: To summarize our current understanding of the likely effects of climate change, including increased air pollution, on rheumatic disease, we searched the published, peer-reviewed English-language literature from January 2000 to December 2022. Articles were reviewed by a team of rheumatologists and clinical and translational science researchers. Systematic review articles were not included but informed additional literature searches. RESULTS: After extensive examination and adjudication, 88 articles met inclusion criteria and were selected for review. Much of the epidemiologic investigations assessed associations between air pollution and increased risk of development of rheumatoid arthritis, anti-citrullinated protein antibodies, flares of gout, and hospitalizations for systemic lupus erythematosus. Increased heat vulnerability was associated with higher odds of recurrent hospitalizations across rheumatic conditions. Mechanisms for observed associations are poorly understood but could include the effects of epigenetic changes, oxidative stress, and inflammatory cytokines. Studies had limitations, including restricted geography and populations studied without focus on historically marginalized communities at highest risk for adverse effects from pollution and climate change, the relative lack of mechanistic evaluations, and most with only indirect links to climate change. CONCLUSION: To date, the published literature lacks studies that directly examine effects of climate change on rheumatic diseases. Collaborative translational and epidemiologic research is needed to enhance our understanding and awareness in this area.

Health equity assessment of machine learning performance (HEAL): a framework and dermatology AI model case study.

Background: Artificial intelligence (AI) has repeatedly been shown to encode historical inequities in healthcare. We aimed to develop a framework to quantitatively assess the performance equity of health AI technologies and to illustrate its utility via a case study. Methods: Here, we propose a methodology to assess whether health AI technologies prioritise performance for patient populations experiencing worse outcomes, that is complementary to existing fairness metrics. We developed the Health Equity Assessment of machine Learning performance (HEAL) framework designed to quantitatively assess the performance equity of health AI technologies via a four-step interdisciplinary process to understand and quantify domain-specific criteria, and the resulting HEAL metric. As an illustrative case study (analysis conducted between October 2022 and January 2023), we applied the HEAL framework to a dermatology AI model. A set of 5420 teledermatology cases (store-and-forward cases from patients of 20 years or older, submitted from primary care providers in the USA and skin cancer clinics in Australia), enriched for diversity in age, sex and race/ethnicity, was used to retrospectively evaluate the AI model's HEAL metric, defined as the likelihood that the AI model performs better for subpopulations with worse average health outcomes as compared to others. The likelihood that AI performance was anticorrelated to pre-existing health outcomes was estimated using bootstrap methods as the probability that the negated Spearman's rank correlation coefficient (i.e., "R") was greater than zero. Positive values of R suggest that subpopulations with poorer health outcomes have better AI model performance. Thus, the HEAL metric, defined as p (R >0), measures how likely the AI technology is to prioritise performance for subpopulations with worse average health outcomes as compared to others (presented as a percentage below). Health outcomes were quantified as disability-adjusted life years (DALYs) when grouping by sex and age, and years of life lost (YLLs) when grouping by race/ethnicity. AI performance was measured as top-3 agreement with the reference diagnosis from a panel of 3 dermatologists per case. Findings: Across all dermatologic conditions, the HEAL metric was 80.5% for prioritizing AI performance of racial/ethnic subpopulations based on YLLs, and 92.1% and 0.0% respectively for prioritizing AI performance of sex and age subpopulations based on DALYs. Certain dermatologic conditions were significantly associated with greater AI model performance compared to a reference category of less common conditions. For skin cancer conditions, the HEAL metric was 73.8% for prioritizing AI performance of age subpopulations based on DALYs. Interpretation: Analysis using the proposed HEAL framework showed that the dermatology AI model prioritised performance for race/ethnicity, sex (all conditions) and age (cancer conditions) subpopulations with respect to pre-existing health disparities. More work is needed to investigate ways of promoting equitable AI performance across age for non-cancer conditions and to better understand how AI models can contribute towards improving equity in health outcomes. Funding: Google LLC.

A Direct Comparison of Three Screening Methods for Autism Spectrum Disorder in a High-Likelihood Sibling Population.

Targeted screening of children at increased likelihood of autism is recommended. However, autism screening tools are usually validated for use mainly in low-likelihood populations. This study compared the accuracy of the Modified Checklist for Autism in Toddlers, Revised with Follow-up (M-CHAT-R/F), the ASDetect app, and the Social Attention and Communication Surveillance, Revised (SACS-R). Siblings of autistic children underwent autism screening at 12, 18 and 30 months old. At each visit, caregivers completed the M-CHAT-R/F and ASDetect while trained nurses tested the siblings using the SACS-R. At 36 to 48 months, the siblings underwent an Autism Diagnostic Observation Schedule-Second Edition (ADOS-2) assessment. 189 siblings were screened, 141 completed the study, and 32 were confirmed to have autism. Although not validated for use at 12 months, the M-CHAT-R/F had the best sensitivity among the three tools for this age group, suggesting that early signs are already apparent to caregivers. The M-CHAT-R/F had overall better sensitivity (0.72-0.83) across all age groups, but with overall lower specificity (0.55-0.77). The SACS-R and ASDetect had better positive predictive values at 18 and 30 months (0.60-0.68), while the M-CHAT-R/F was 0.43-0.48. Negative predictive values were generally high across all three tools across all age groups (0.78-0.93). Targeted screening of children at high likelihood of autism yielded a detection rate of 22.7% and should therefore be implemented routinely to facilitate early detection and intervention. The performance of autism screening tools should be examined in higher-likelihood populations for targeted screening of these children.

An adaptive text message intervention to promote psychological well-being and reduce cardiac risk: The Text4Health controlled clinical pilot trial.

OBJECTIVE: In a two-arm pilot trial, we examined the feasibility, acceptability, and preliminary efficacy of a 12-week, adaptive text message intervention (TMI) to promote health behaviors and psychological well-being in 60 individuals with multiple cardiac risk conditions (i.e., hypertension, hyperlipidemia, and/or type 2 diabetes) and suboptimal adherence to exercise or dietary guidance. METHODS: Participants were allocated to receive the TMI or enhanced usual care (eUC). The TMI included daily adaptive text messages promoting health behaviors, twice-weekly messages to set goals and monitor progress, and monthly phone check-ins. Feasibility (primary outcome) and acceptability were measured by rates of successful text message delivery and daily participant ratings of message utility (0-10 Likert scale). We also assessed impact on health behavior adherence, psychological health, and functional outcomes. RESULTS: The TMI was feasible (99.3% of messages successfully sent) and well-accepted (mean utility = 7.4/10 [SD 2.6]). At 12 weeks, the TMI led to small-sized greater improvements in moderate to vigorous physical activity (d = 0.37), overall physical activity (d = 0.23), optimism (d = 0.20), anxiety (d = -0.36), self-efficacy (d = 0.22), and physical function (d = 0.20), compared to eUC. It did not impact other outcomes substantially at this time point. CONCLUSION: This 12-week, adaptive TMI was feasible, well-accepted, and associated with small-sized greater improvements in health behavior and psychological outcomes. Though larger studies are needed, it has the potential to be a scalable, low-intensity program that could be used in clinical practice. CLINICALTRIALS: govregistration:NCT04382521.

Characteristics and engagement among English-language online forums for addiction recovery available in the US.

In developing public resources for the Networks Enhancing Addiction Recovery - Forum Activity Roadmap (NEAR-FAR), we completed a systematic observational study of English-language online forums related to recovery from alcohol or other drug addiction in late 2021. Among 207 identified forums, the majority were classified as "general addiction" or alcohol-focused, though classifications related to other substances were common on websites hosting multiple forums. Commonly used social media platforms such as Reddit, Facebook, or Quora offered easily accessible venues for individuals seeking online support related to a variety of substances. Forums were related to established recovery programs such as 12-step and SMART Recovery as well as other nonprofit and for-profit recovery programs, and to community forums without formal recovery programming. Among 148 forums with any observed user activity, the median time between unique user engagements was 27 days (inter-quartile range: 2-74). Among 98 forums with past-month posting activity, we found a median of <10 posts per week (inter-quartile range: 1-78). This study compares three metrics of observed forum activity (posts per week, responses per post, time between unique user engagements) and operationalizes forum characteristics that may potentiate opportunities for enhanced engagement and social support in addiction recovery.

Macrophage RAGE activation is proinflammatory in NASH.

Intrahepatic macrophages in nonalcoholic steatohepatitis (NASH) are heterogenous and include proinflammatory recruited monocyte-derived macrophages. The receptor for advanced glycation endproducts (RAGE) is expressed on macrophages and can be activated by damage associated molecular patterns (DAMPs) upregulated in NASH, yet the role of macrophage-specific RAGE signaling in NASH is unclear. Therefore, we hypothesized that RAGE-expressing macrophages are proinflammatory and mediate liver inflammation in NASH. Compared with healthy controls, RAGE expression was increased in liver biopsies from patients with NASH. In a high-fat, -fructose, and -cholesterol-induced (FFC)-induced murine model of NASH, RAGE expression was increased, specifically on recruited macrophages. FFC mice that received a pharmacological inhibitor of RAGE (TTP488), and myeloid-specific RAGE KO mice (RAGE-MKO) had attenuated liver injury associated with a reduced accumulation of RAGE+ recruited macrophages. Transcriptomics analysis suggested that pathways of macrophage and T cell activation were upregulated by FFC diet, inhibited by TTP488 treatment, and reduced in RAGE-MKO mice. Correspondingly, the secretome of ligand-stimulated BM-derived macrophages from RAGE-MKO mice had an attenuated capacity to activate CD8+ T cells. Our data implicate RAGE as what we propose to be a novel and potentially targetable mediator of the proinflammatory signaling of recruited macrophages in NASH.

Stanford University School of Medicine: Our Neurosurgical Heritage.

The legacy of Stanford University's Department of Neurosurgery began in 1858, with the establishment of a new medical school on the West Coast. Stanford Neurosurgery instilled an atmosphere of dedication to neurosurgical care, scientific research, education, and innovation. We highlight key historical events leading to the formation of the medical school and neurosurgical department, the individuals who shaped the department's vision and expansion, as well as pioneering advances in research and clinical care. The residency program was started in 1961, establishing the basis of the current education model with a strong emphasis on training future leaders, and the Moyamoya Center, founded in 1991, became the largest Moyamoya referral center in the United States. The opening of Stanford Stroke Center (1992) and seminal clinical trials resulted in a significant impact on cerebrovascular disease by expanding the treatment window of IV thrombolysis and intra-arterial thrombectomy. The invention and implementation of CyberKnife® (1994) marks another important event that revolutionized the field of radiosurgery, and the development of Stanford's innovative Brain Computer Interface program is pushing the boundaries of this specialty. The more recent launch of the Neurosurgery Virtual Reality and Simulation Center (2017) exemplifies how Stanford is continuing to evolve in this ever-changing field. The department also became a model for diversity within the school as well as nationwide. The growth of Stanford Neurosurgery from one of the youngest neurosurgery departments in the country to a prominent comprehensive neurosurgery center mirrors the history of neurosurgery itself: young, innovative, and willing to overcome challenges.

Erratum to "A case of de novo seronegative inflammatory oligoarthritis associated with COVID-19 infection" Clinical Immunology Communications 2 (2022) 159-161/.

[This corrects the article DOI: 10.1016/j.clicom.2022.10.001.].

Matched-pair analysis of patients with ischemic stroke undergoing thrombectomy using next-generation balloon guide catheters.

BACKGROUND: Balloon guide catheters (BGCs) have not been widely adopted, possibly due to the incompatibility of past-generation BGCs with large-bore intermediate catheters. The next-generation BGC is compatible with large-bore catheters. We compared outcomes of thrombectomy cases using BGCs versus conventional guide catheters. METHODS: We conducted a retrospective study of 110 thrombectomy cases using BGCs (n=55) and non-BGCs (n=55). Sixty consecutive thrombectomy cases in whom the BOBBY BGC was used at a single institution between February 2021 and March 2022 were identified. Of these, 55 BGC cases were 1:1 matched with non-BGC cases by proceduralists, age, gender, stent retriever + aspiration device versus aspiration-only, and site of occlusion. First-pass effect was defined as Thrombolysis In Cerebral Infarction 2b or higher with a single pass. RESULTS: The BGC and non-BGC cohorts had similar mean age (67.2 vs 68.9 years), gender distribution (43.6% vs 47.3% women), median initial National Institutes of Health Stroke Scale score (14 vs 15), and median pretreatment ischemic core volumes (12 mL vs 11.5 mL). BGC and non-BGC cases had similar rates of single pass (60.0% vs 54.6%), first-pass effect (58.2% vs 49.1%), and complications (1.8% vs 9.1%). In aspiration-only cases, the BGC cohort had a significantly higher rate of first-pass effect (100% vs 50.0%, p=0.01). BGC was associated with a higher likelihood of achieving a modified Rankin Scale score of 2 at discharge (OR 7.76, p=0.02). No additional procedural time was required for BGC cases (46.7 vs 48.2 min). CONCLUSION: BGCs may be safely adopted with comparable procedural efficacy, benefits to aspiration-only techniques, and earlier functional improvement compared with conventional guide catheters.

ZFP92, a KRAB domain zinc finger protein enriched in pancreatic islets, binds to B1/Alu SINE transposable elements and regulates retroelements and genes.

Repressive KRAB domain-containing zinc-finger proteins (KRAB-ZFPs) are abundant in mammalian genomes and contribute both to the silencing of transposable elements (TEs) and to the regulation of developmental stage- and cell type-specific gene expression. Here we describe studies of zinc finger protein 92 (Zfp92), an X-linked KRAB-ZFP that is highly expressed in pancreatic islets of adult mice, by analyzing global Zfp92 knockout (KO) mice. Physiological, transcriptomic and genome-wide chromatin binding studies indicate that the principal function of ZFP92 in mice is to bind to and suppress the activity of B1/Alu type of SINE elements and modulate the activity of surrounding genomic entities. Deletion of Zfp92 leads to changes in expression of select LINE and LTR retroelements and genes located in the vicinity of ZFP92-bound chromatin. The absence of Zfp92 leads to altered expression of specific genes in islets, adipose and muscle that result in modest sex-specific alterations in blood glucose homeostasis, body mass and fat accumulation. In islets, Zfp92 influences blood glucose concentration in postnatal mice via transcriptional effects on Mafb, whereas in adipose and muscle, it regulates Acacb, a rate-limiting enzyme in fatty acid metabolism. In the absence of Zfp92, a novel TE-Capn11 fusion transcript is overexpressed in islets and several other tissues due to de-repression of an IAPez TE adjacent to ZFP92-bound SINE elements in intron 3 of the Capn11 gene. Together, these studies show that ZFP92 functions both to repress specific TEs and to regulate the transcription of specific genes in discrete tissues.

Predicting lymph node metastasis from primary tumor histology and clinicopathologic factors in colorectal cancer using deep learning.

BACKGROUND: Presence of lymph node metastasis (LNM) influences prognosis and clinical decision-making in colorectal cancer. However, detection of LNM is variable and depends on a number of external factors. Deep learning has shown success in computational pathology, but has struggled to boost performance when combined with known predictors. METHODS: Machine-learned features are created by clustering deep learning embeddings of small patches of tumor in colorectal cancer via k-means, and then selecting the top clusters that add predictive value to a logistic regression model when combined with known baseline clinicopathological variables. We then analyze performance of logistic regression models trained with and without these machine-learned features in combination with the baseline variables. RESULTS: The machine-learned extracted features provide independent signal for the presence of LNM (AUROC: 0.638, 95% CI: [0.590, 0.683]). Furthermore, the machine-learned features add predictive value to the set of 6 clinicopathologic variables in an external validation set (likelihood ratio test, p < 0.00032; AUROC: 0.740, 95% CI: [0.701, 0.780]). A model incorporating these features can also further risk-stratify patients with and without identified metastasis (p < 0.001 for both stage II and stage III). CONCLUSION: This work demonstrates an effective approach to combine deep learning with established clinicopathologic factors in order to identify independently informative features associated with LNM. Further work building on these specific results may have important impact in prognostication and therapeutic decision making for LNM. Additionally, this general computational approach may prove useful in other contexts.

When colorectal cancers spread to the lymph nodes, it can indicate a poorer prognosis. However, detecting lymph node metastasis (spread) can be difficult and depends on a number of factors such as how samples are taken and processed. Here, we show that machine learning, which involves computer software learning from patterns in data, can predict lymph node metastasis in patients with colorectal cancer from the microscopic appearance of their primary tumor and the clinical characteristics of the patients. We also show that the same approach can predict patient survival. With further work, our approach may help clinicians to inform patients about their prognosis and decide on appropriate treatments.

Pathologist Validation of a Machine Learning-Derived Feature for Colon Cancer Risk Stratification.

Importance: Identifying new prognostic features in colon cancer has the potential to refine histopathologic review and inform patient care. Although prognostic artificial intelligence systems have recently demonstrated significant risk stratification for several cancer types, studies have not yet shown that the machine learning-derived features associated with these prognostic artificial intelligence systems are both interpretable and usable by pathologists. Objective: To evaluate whether pathologist scoring of a histopathologic feature previously identified by machine learning is associated with survival among patients with colon cancer. Design, Setting, and Participants: This prognostic study used deidentified, archived colorectal cancer cases from January 2013 to December 2015 from the University of Milano-Bicocca. All available histologic slides from 258 consecutive colon adenocarcinoma cases were reviewed from December 2021 to February 2022 by 2 pathologists, who conducted semiquantitative scoring for tumor adipose feature (TAF), which was previously identified via a prognostic deep learning model developed with an independent colorectal cancer cohort. Main Outcomes and Measures: Prognostic value of TAF for overall survival and disease-specific survival as measured by univariable and multivariable regression analyses. Interpathologist agreement in TAF scoring was also evaluated. Results: A total of 258 colon adenocarcinoma histopathologic cases from 258 patients (138 men [53%]; median age, 67 years [IQR, 65-81 years]) with stage II (n = 119) or stage III (n = 139) cancer were included. Tumor adipose feature was identified in 120 cases (widespread in 63 cases, multifocal in 31, and unifocal in 26). For overall survival analysis after adjustment for tumor stage, TAF was independently prognostic in 2 ways: TAF as a binary feature (presence vs absence: hazard ratio [HR] for presence of TAF, 1.55 [95% CI, 1.07-2.25]; P = .02) and TAF as a semiquantitative categorical feature (HR for widespread TAF, 1.87 [95% CI, 1.23-2.85]; P = .004). Interpathologist agreement for widespread TAF vs lower categories (absent, unifocal, or multifocal) was 90%, corresponding to a κ metric at this threshold of 0.69 (95% CI, 0.58-0.80). Conclusions and Relevance: In this prognostic study, pathologists were able to learn and reproducibly score for TAF, providing significant risk stratification on this independent data set. Although additional work is warranted to understand the biological significance of this feature and to establish broadly reproducible TAF scoring, this work represents the first validation to date of human expert learning from machine learning in pathology. Specifically, this validation demonstrates that a computationally identified histologic feature can represent a human-identifiable, prognostic feature with the potential for integration into pathology practice.

Cholesteatoma in Congenital Aural Atresia and External Auditory Canal Stenosis: A Systematic Review.

OBJECTIVE: Patients with congenital external auditory canal (EAC) abnormalities are at risk of developing cholesteatoma and often undergo surveillance imaging to detect it. The aims of this systematic review are to determine the incidence of cholesteatoma in patients with congenital aural atresia (CAA) and patients with congenital EAC stenosis and to investigate the most common age of cholesteatoma diagnosis. This information will help clinicians decide which patients require surveillance scanning, as well as the timing of imaging. DATA SOURCES: Ovid MEDLINE, Embase, CENTRAL, and Web of Science databases. REVIEW METHODS: A systematic literature review following the PRISMA guidelines was performed. The data sources were searched by 2 independent reviewers, and articles were included that reported on CAA or congenital EAC stenosis with a confirmed diagnosis of cholesteatoma. The selected articles were screened separately by 3 reviewers before reaching a consensus on the final articles to include. Data collection on the number of patients with cholesteatoma and the age of diagnosis was performed for these articles. RESULTS: Eight articles met the inclusion criteria. The incidence of cholesteatoma was 1.7% (4/238) in CAA and 43.0% (203/473) in congenital EAC stenosis. The majority of patients with congenital EAC stenosis that developed cholesteatoma were diagnosed at age <12 years. CONCLUSION: CAA is associated with a low risk of cholesteatoma formation, and surveillance imaging is unnecessary in asymptomatic patients. EAC stenosis is strongly associated with cholesteatoma, and a surveillance scan for these patients is recommended prior to 12 years of age with close follow-up into adulthood.

Robustness to Spurious Correlations Improves Semantic Out-of-Distribution Detection.

Methods which utilize the outputs or feature representations of predictive models have emerged as promising approaches for out-of-distribution (ood) detection of image inputs. However, these methods struggle to detect ood inputs that share nuisance values (e.g. background) with in-distribution inputs. The detection of shared-nuisance out-of-distribution (sn-ood) inputs is particularly relevant in real-world applications, as anomalies and in-distribution inputs tend to be captured in the same settings during deployment. In this work, we provide a possible explanation for sn-ood detection failures and propose nuisance-aware ood detection to address them. Nuisance-aware ood detection substitutes a classifier trained via Empirical Risk Minimization (erm) and cross-entropy loss with one that 1. is trained under a distribution where the nuisance-label relationship is broken and 2. yields representations that are independent of the nuisance under this distribution, both marginally and conditioned on the label. We can train a classifier to achieve these objectives using Nuisance-Randomized Distillation (NURD), an algorithm developed for ood generalization under spurious correlations. Output- and feature-based nuisance-aware ood detection perform substantially better than their original counterparts, succeeding even when detection based on domain generalization algorithms fails to improve performance.

Deep learning models for histologic grading of breast cancer and association with disease prognosis.

Histologic grading of breast cancer involves review and scoring of three well-established morphologic features: mitotic count, nuclear pleomorphism, and tubule formation. Taken together, these features form the basis of the Nottingham Grading System which is used to inform breast cancer characterization and prognosis. In this study, we develop deep learning models to perform histologic scoring of all three components using digitized hematoxylin and eosin-stained slides containing invasive breast carcinoma. We first evaluate model performance using pathologist-based reference standards for each component. To complement this typical approach to evaluation, we further evaluate the deep learning models via prognostic analyses. The individual component models perform at or above published benchmarks for algorithm-based grading approaches, achieving high concordance rates with pathologist grading. Further, prognostic performance using deep learning-based grading is on par with that of pathologists performing review of matched slides. By providing scores for each component feature, the deep-learning based approach also provides the potential to identify the grading components contributing most to prognostic value. This may enable optimized prognostic models, opportunities to improve access to consistent grading, and approaches to better understand the links between histologic features and clinical outcomes in breast cancer.