A Multimodal Generative AI Copilot for Human Pathology.

The field of computational pathology[1,2] has witnessed remarkable progress in the development of both task-specific predictive models and task-agnostic self-supervised vision encoders[3,4]. However, despite the explosive growth of generative artificial intelligence (AI), there has been limited study on building general purpose, multimodal AI assistants and copilots[5] tailored to pathology. Here we present PathChat, a vision-language generalist AI assistant for human pathology. We build PathChat by adapting a foundational vision encoder for pathology, combining it with a pretrained large language model and finetuning the whole system on over 456,000 diverse visual language instructions consisting of 999,202 question-answer turns. We compare PathChat against several multimodal vision language AI assistants and GPT4V, which powers the commercially available multimodal general purpose AI assistant ChatGPT-4[7]. PathChat achieved state-of-the-art performance on multiple-choice diagnostic questions from cases of diverse tissue origins and disease models. Furthermore, using open-ended questions and human expert evaluation, we found that overall PathChat produced more accurate and pathologist-preferable responses to diverse queries related to pathology. As an interactive and general vision-language AI Copilot that can flexibly handle both visual and natural language inputs, PathChat can potentially find impactful applications in pathology education, research, and human-in-the-loop clinical decision making.

Demographic bias in misdiagnosis by computational pathology models.

Despite increasing numbers of regulatory approvals, deep learning-based computational pathology systems often overlook the impact of demographic factors on performance, potentially leading to biases. This concern is all the more important as computational pathology has leveraged large public datasets that underrepresent certain demographic groups. Using publicly available data from The Cancer Genome Atlas and the EBRAINS brain tumor atlas, as well as internal patient data, we show that whole-slide image classification models display marked performance disparities across different demographic groups when used to subtype breast and lung carcinomas and to predict IDH1 mutations in gliomas. For example, when using common modeling approaches, we observed performance gaps (in area under the receiver operating characteristic curve) between white and Black patients of 3.0% for breast cancer subtyping, 10.9% for lung cancer subtyping and 16.0% for IDH1 mutation prediction in gliomas. We found that richer feature representations obtained from self-supervised vision foundation models reduce performance variations between groups. These representations provide improvements upon weaker models even when those weaker models are combined with state-of-the-art bias mitigation strategies and modeling choices. Nevertheless, self-supervised vision foundation models do not fully eliminate these discrepancies, highlighting the continuing need for bias mitigation efforts in computational pathology. Finally, we demonstrate that our results extend to other demographic factors beyond patient race. Given these findings, we encourage regulatory and policy agencies to integrate demographic-stratified evaluation into their assessment guidelines.

A machine learning approach to predict mortality due to immune-mediated thrombotic thrombocytopenic purpura.

Background: Mortality due to immune-mediated thrombotic thrombocytopenic purpura (iTTP) remains significant. Predicting mortality risk may potentially help individualize treatment. The French Thrombotic Microangiopathy (TMA) Reference Score has not been externally validated in the United States. Recent advances in machine learning technology can help analyze large numbers of variables with complex interactions for the development of prediction models. Objectives: To validate the French TMA Reference Score in the United States Thrombotic Microangiopathy (USTMA) iTTP database and subsequently develop a novel mortality prediction tool, the USTMA TTP Mortality Index. Methods: We analyzed variables available at the time of initial presentation, including demographics, symptoms, and laboratory findings. We developed our model using gradient boosting machine, a machine learning ensemble method based on classification trees, implemented in the R package gbm. Results: In our cohort (n = 419), the French score predicted mortality with an area under the receiver operating characteristic curve of 0.63 (95% CI: 0.50-0.77), sensitivity of 0.35, and specificity of 0.84. Our gradient boosting machine model selected 8 variables to predict acute mortality with a cross-validated area under the receiver operating characteristic curve of 0.77 (95% CI: 0.71-0.82). The 2 cutoffs corresponded to sensitivities of 0.64 and 0.50 and specificities of 0.76 and 0.87, respectively. Conclusion: The USTMA Mortality Index was acceptable for predicting mortality due to acute iTTP in the USTMA registry, but not sensitive enough to rule out death. Identifying patients at high risk of iTTP-related mortality may help individualize care and ultimately improve iTTP survival outcomes. Further studies are needed to provide external validation. Our model is one of many recent examples where machine learning models may show promise in clinical prediction tools in healthcare.

A visual-language foundation model for computational pathology.

The accelerated adoption of digital pathology and advances in deep learning have enabled the development of robust models for various pathology tasks across a diverse array of diseases and patient cohorts. However, model training is often difficult due to label scarcity in the medical domain, and a model's usage is limited by the specific task and disease for which it is trained. Additionally, most models in histopathology leverage only image data, a stark contrast to how humans teach each other and reason about histopathologic entities. We introduce CONtrastive learning from Captions for Histopathology (CONCH), a visual-language foundation model developed using diverse sources of histopathology images, biomedical text and, notably, over 1.17 million image-caption pairs through task-agnostic pretraining. Evaluated on a suite of 14 diverse benchmarks, CONCH can be transferred to a wide range of downstream tasks involving histopathology images and/or text, achieving state-of-the-art performance on histology image classification, segmentation, captioning, and text-to-image and image-to-text retrieval. CONCH represents a substantial leap over concurrent visual-language pretrained systems for histopathology, with the potential to directly facilitate a wide array of machine learning-based workflows requiring minimal or no further supervised fine-tuning.

Towards a general-purpose foundation model for computational pathology.

Quantitative evaluation of tissue images is crucial for computational pathology (CPath) tasks, requiring the objective characterization of histopathological entities from whole-slide images (WSIs). The high resolution of WSIs and the variability of morphological features present significant challenges, complicating the large-scale annotation of data for high-performance applications. To address this challenge, current efforts have proposed the use of pretrained image encoders through transfer learning from natural image datasets or self-supervised learning on publicly available histopathology datasets, but have not been extensively developed and evaluated across diverse tissue types at scale. We introduce UNI, a general-purpose self-supervised model for pathology, pretrained using more than 100 million images from over 100,000 diagnostic H&E-stained WSIs (>77 TB of data) across 20 major tissue types. The model was evaluated on 34 representative CPath tasks of varying diagnostic difficulty. In addition to outperforming previous state-of-the-art models, we demonstrate new modeling capabilities in CPath such as resolution-agnostic tissue classification, slide classification using few-shot class prototypes, and disease subtyping generalization in classifying up to 108 cancer types in the OncoTree classification system. UNI advances unsupervised representation learning at scale in CPath in terms of both pretraining data and downstream evaluation, enabling data-efficient artificial intelligence models that can generalize and transfer to a wide range of diagnostically challenging tasks and clinical workflows in anatomic pathology.

Targeting a higher plasma VWF level at time of delivery in pregnant individuals with von Willebrand disease: Outcomes at a single-institution cohort study.

INTRODUCTION: Guidelines on the management of pregnant individuals with von Willebrand disease (VWD) at the time of delivery recommend that von Willebrand factor (VWF) and factor VIII:C (FVIII:C) levels be ≥50% to prevent postpartum haemorrhage (PPH). Yet, high PPH rates persist despite these levels or with prophylactic factor replacement therapy to achieve these levels. AIMS: The current practice at our centre has been to target peak plasma VWF and FVIII:C levels of ≥100 IU/dL at time of delivery. The objective of this study was to describe obstetric outcomes in pregnant individuals with VWD who were managed at our centre. METHODS: Demographics and outcomes on pregnant individuals with VWD who delivered between January 2015 and April 2023 were collected. RESULTS: Forty-seven singleton deliveries (among 41 individuals) resulting in 46 live births and one foetal death were included. Twenty-one individuals had at least one prior birth by the start date of this study, of which 11 (52.4%) self-reported a history of PPH. Early PPH occurred in 12.8% (6/47) of deliveries. Two individuals required blood transfusion, of which one also had an unplanned hysterectomy and transfer to ICU. There were no thrombotic events reported. CONCLUSION: The strategy of targeting higher peak plasma VWF and FVIII:C levels (≥100 IU/dL) at the time of delivery may be effective in reducing the risk of delivery-associated bleeding complications in VWD patients. Yet, the rate of early PPH remains unsatisfactory compared to the non-VWD population.

Observational cohort study of long-term outcomes of liver transplantation in haemophilia.

INTRODUCTION: Gene therapy is now a reality for individuals with haemophilia, yet little is known regarding the quality-of-life impact of factor correction. As few data exist, and recognizing the analogy to liver transplantation (OLTX), we identified OLTX+ and OLTX- men in the ATHNdataset to compare post-OLTX factor VIII and IX on quality of life (QoL) by Haem-A-QoL and PROMIS-29. METHODS: OLTX- were matched to OLTX+ by age, race, and haemophilia type and severity. Deidentified demographic data, including post-transplant factor levels, genotype and target joint disease were analysed by descriptive statistics. Haem-A-Qol and PROMIS-29 were compared in OLTX+ and OLTX- by student's t-test and univariate regression models. RESULTS: Of 86 people with haemophilia A (HA) or haemophilia B (HB) cared for at 10 haemophilia treatment centers (HTCs), 21 (24.4%) OLTX+ and 65 (75.6%) OLTX- were identified. OLTX+ and OLTX- had a similar frequency of target joint disease (p = .806), HA genotypes, null versus non-null (p = .696), and HIV infection (p = .316). At a median 9.2 years post-OLTX, median FVIII, .63 IU/mL [IQR 0.52-0.97] and FIX, .91 IU/mL [IQR .63-1.32], Haem-A-QoL, PROMIS-29, and HOT scores were comparable. Severe HA/HB had lower post-OLTX 'dealing with haemophilia' scores (p = .022) and higher 'sports and leisure' (p = .010) and 'view of yourself' scores (p = .024) than OLTX+ non-severe participants. Non-caucasian OLTX+ had significantly lower scores in sports and leisure (p = .042), future expectations (p = .021) and total score (p = .010). CONCLUSION: Nine years after OLTX, QoL is comparable to OLTX-, but significantly better in OLTX+ with severe than non-severe disease and in caucasians than non-caucasians.

How I diagnose and treat antiphospholipid syndrome in pregnancy.

ABSTRACT: Antiphospholipid syndrome (APS) is a rare autoimmune disease characterized by arterial, venous, or microvascular thrombosis, pregnancy morbidities, or nonthrombotic manifestations in patients with persistently positive antiphospholipid antibodies. These antibodies bind cellular phospholipids and phospholipid-protein complexes resulting in cellular activation and inflammation that lead to the clinical features of APS. Our evolving understanding of APS has resulted in more specific classification criteria. Patients meeting these criteria should be treated during pregnancy according to current guidelines. Yet, despite treatment, those positive for lupus anticoagulant have at least a 30% likelihood of adverse pregnancy outcomes. Patients with recurrent early miscarriage or fetal death in the absence of preeclampsia or placental insufficiency may not meet current classification criteria for APS. Patients with only low titer anticardiolipin or anti-ß(2)-glycoprotein I antibodies or immunoglobulin M isotype antibodies will not meet current classification criteria. In such cases, clinicians should implement management plans that balance potential risks and benefits, some of which involve emotional concerns surrounding the patient's reproductive future. Finally, APS may present in pregnancy or postpartum as a thrombotic microangiopathy, a life-threatening condition that may initially mimic preeclampsia with severe features but requires a very different treatment approach.

Quality improvement approaches to heparin-induced thrombocytopenia: a scoping review.

Background: Heparin-induced thrombocytopenia (HIT) is a relatively uncommon condition characterized by 2 exceedingly common phenomena in hospitalized patients: thrombocytopenia and heparin exposure. Consequently, HIT is frequently overdiagnosed and inappropriately treated. These issues are the focus of many quality improvement (QI) initiatives. Objectives: In this scoping review, we identified and characterized all published QI studies on improving the diagnosis and management of HIT. Methods: We conducted a systematic literature search through April 2022 for studies reporting on QI interventions regarding the diagnosis, treatment, and/or prevention of HIT. Results: Thirty studies were included in the final review. Studies were separated into 5 groups based on the focus of the interventions: increasing HIT recognition, reducing HIT incidence, reducing HIT overdiagnosis, promoting safer HIT management, and creating HIT task forces. Nine studies focused on the implementation of 4Ts score calculator into electronic medical record orders for HIT testing, while only 1 evaluated the impact of reducing unfractionated heparin use in favor of low-molecular-weight heparin. Six studies focused on the implementation of direct thrombin inhibitor management protocols, while none evaluated the use of alternative anticoagulants in HIT management. Conclusion: The bulk of published HIT QI research focused on reducing overdiagnosis and promoting safer direct thrombin inhibitor therapy, while minimal attention has been devoted to HIT prevention and the use of evidence-based alternative HIT therapies.

Trends in dedicated care for females with bleeding disorders within U.S. hemophilia treatment centers.

Graphical representation of increasing percentage of female patients seen at HTCs, percentage of females by diagnosis, number of clinics in existence, and absolute number of female patients seen over a 10-year period (top left then clockwise).

A General-Purpose Self-Supervised Model for Computational Pathology.

Tissue phenotyping is a fundamental computational pathology (CPath) task in learning objective characterizations of histopathologic biomarkers in anatomic pathology. However, whole-slide imaging (WSI) poses a complex computer vision problem in which the large-scale image resolutions of WSIs and the enormous diversity of morphological phenotypes preclude large-scale data annotation. Current efforts have proposed using pretrained image encoders with either transfer learning from natural image datasets or self-supervised pretraining on publicly-available histopathology datasets, but have not been extensively developed and evaluated across diverse tissue types at scale. We introduce UNI, a general-purpose self-supervised model for pathology, pretrained using over 100 million tissue patches from over 100,000 diagnostic haematoxylin and eosin-stained WSIs across 20 major tissue types, and evaluated on 33 representative CPath clinical tasks in CPath of varying diagnostic difficulties. In addition to outperforming previous state-of-the-art models, we demonstrate new modeling capabilities in CPath such as resolution-agnostic tissue classification, slide classification using few-shot class prototypes, and disease subtyping generalization in classifying up to 108 cancer types in the OncoTree code classification system. UNI advances unsupervised representation learning at scale in CPath in terms of both pretraining data and downstream evaluation, enabling data-efficient AI models that can generalize and transfer to a gamut of diagnostically-challenging tasks and clinical workflows in anatomic pathology.

Incidence of inguinofemoral lymph node metastases at the first local recurrence of vulvar cancer: a Dutch nationwide study.

BACKGROUND: Up to 40% of vulvar cancer patients present with local recurrence within 10 years of follow-up. An inguinofemoral lymphadenectomy (IFL) is indicated if not performed at primary treatment. The incidence and risk factors for lymph node metastases (LNM) at first local recurrence, however, are unclear. Our aim was to determine the incidence of LNM at first local recurrence, in relation to previous groin treatment and clinicopathological factors. METHODS: A multicenter cohort study including vulvar cancer patients with a first macroinvasive local recurrence after primary surgical treatment between 2000 and 2015 was conducted in the Netherlands. Groin status at local recurrence was defined as positive (N+), negative (N-) or unknown (N?) and based on histology, imaging and follow-up. Patient-, tumour- and treatment characteristics of primary and recurrent disease were analysed. RESULTS: Overall, 16.3% (66/404) had a N+ groin status at first local recurrence, 66.4% (268/404) N- and 17.3% (70/404) N? groin status. The incidence of a N+ groin status was comparable after previous SLN and IFL, 11.5% and 13.8%, respectively. A N+ groin status was related to tumour size (25 vs.12 mm; P < 0.001), depth of invasion (5 vs. 3 mm; P < 0.001) and poorly differentiated tumours (22.9 vs. 11.9%; P = 0.050) at local recurrence. CONCLUSIONS: The incidence of LNM at first local recurrence in vulvar cancer patients was 16.3%, and independent of previous type of groin surgery. In accordance with primary diagnosis, tumour size, depth of invasion, and tumour grade were significantly associated with a positive groin status.

Algorithmic fairness in artificial intelligence for medicine and healthcare.

In healthcare, the development and deployment of insufficiently fair systems of artificial intelligence (AI) can undermine the delivery of equitable care. Assessments of AI models stratified across subpopulations have revealed inequalities in how patients are diagnosed, treated and billed. In this Perspective, we outline fairness in machine learning through the lens of healthcare, and discuss how algorithmic biases (in data acquisition, genetic variation and intra-observer labelling variability, in particular) arise in clinical workflows and the resulting healthcare disparities. We also review emerging technology for mitigating biases via disentanglement, federated learning and model explainability, and their role in the development of AI-based software as a medical device.

Differentiating and Managing Rare Thrombotic Microangiopathies During Pregnancy and Postpartum.

The most common thrombotic microangiopathy (TMA) of pregnancy is the well-recognized syndrome of preeclampsia with hemolysis, elevated liver enzymes, and low platelet count (HELLP) syndrome. However, rare TMAs, including thrombotic thrombocytopenic purpura, complement-mediated hemolytic-uremic syndrome, and catastrophic antiphospholipid syndrome, may occur during pregnancy or postpartum and present with features similar to those of preeclampsia with severe features. Early recognition and treatment of these infrequently encountered conditions are key for avoiding serious maternal morbidities with long-term sequelae and possible maternal or fetal death. Differentiating between preeclampsia with severe features and these rare TMAs is diagnostically challenging as there is significant overlap in their clinical and laboratory presentation. Given the rarity of these TMAs, high-quality evidence-based recommendations on diagnosis and management during pregnancy are lacking. Using current objective information and recommendations from working groups, this report provides practical clinical approaches to diagnose and manage these rare TMAs. This report also discusses how to manage individuals with a history of these rare TMAs who are planning to conceive. To optimize favorable outcomes, a multidisciplinary approach including obstetricians, maternal-fetal medicine specialists, hematologists, and nephrologists alongside close clinical and laboratory monitoring is vital.

Pregnancy and childbirth in women with bleeding disorders: A retrospective cohort study.

INTRODUCTION: Reproductive-age women with bleeding disorders (BDs) are underdiagnosed and understudied, despite their increased risk for adverse health outcomes and pregnancy complications. AIM: This study examines pregnancy outcomes and obstetric complications of Utah women with BDs. METHODS: This retrospective cohort study utilized linked birth records and clinical billing data from two large Utah healthcare systems. Utah residents who had their first birth at > 20 weeks gestation (2008-2015) and who received non-emergent care within either system before delivery were included (n = 61 226). Multivariable logistic regression models were used to examine relationships between BDs and neonatal and obstetric outcomes. RESULTS: A total of 295 women (.48%) were included in the BD study population. Women with BDs had significantly increased odds of preterm birth (aOR 1.85, 95% CI 1.32-2.60), Caesarean delivery (aOR 1.38, 95% CI 1.06-1.79), postpartum blood transfusion (aOR 2.55, 95% CI 1.05-6.22), unplanned postpartum hysterectomy (aOR 33.96, 95% CI 7.30-157.89) and transfer to an intensive care unit (aOR 18.18, 95% CI 7.17-46.08). All of the women with BDs who experienced these serious complications were not diagnosed with a BD until the year of their first birth. Additionally, those with BDs were more likely to experience maternal and infant mortality. CONCLUSION: Women with BDs had an increased risk for preterm birth, Caesarean delivery, blood transfusion, unplanned hysterectomy, intensive care unit admission, maternal and infant mortality. Those who were not diagnosed with a BD before the year of their first birth were at an increased risk for serious pregnancy complications.

An Unusual Case of Hyperhemolysis Syndrome and Delayed Hemolytic Transfusion Reaction due to Anti-Jk(a) and Anti-P1 Antibodies.

Background: Hyperhemolysis syndrome (HS) is a severe hemolytic transfusion reaction that can cause hemoglobin and hematocrit levels to drop below pretransfusion levels, leading to severe anemia. HS most commonly occurs in patients with a pre-existing hemoglobinopathy such as sickle cell disease (SCD) or beta-thalassemia. Methods: We report a case of HS, occurring in the absence of hemoglobinopathy, making the diagnosis challenging. The patient reported was also affected by a CIC-rearranged sarcoma. As part of the workup, the patient received a bone marrow biopsy for suspected hemophagocytic lymphohistiocytosis. Results: This provided a rare biopsy specimen to correlate reticulocytopenia with marked erythroid hyperplasia in the marrow, supporting the hypothesis of reticulocyte destruction as a contributing cause of anemia in these patients. This patient had demonstrable alloantibodies to the Jk(a) and P1 antigens as potential triggers for HS. Conclusions: It is vital that a diagnosis of HS be correctly made in these patients with severe anemia, as blood transfusions generally lead to worsening of their conditions.

A deep-learning model for transforming the style of tissue images from cryosectioned to formalin-fixed and paraffin-embedded.

Histological artefacts in cryosectioned tissue can hinder rapid diagnostic assessments during surgery. Formalin-fixed and paraffin-embedded (FFPE) tissue provides higher quality slides, but the process for obtaining them is laborious (typically lasting 12-48 h) and hence unsuitable for intra-operative use. Here we report the development and performance of a deep-learning model that improves the quality of cryosectioned whole-slide images by transforming them into the style of whole-slide FFPE tissue within minutes. The model consists of a generative adversarial network incorporating an attention mechanism that rectifies cryosection artefacts and a self-regularization constraint between the cryosectioned and FFPE images for the preservation of clinically relevant features. Transformed FFPE-style images of gliomas and of non-small-cell lung cancers from a dataset independent from that used to train the model improved the rates of accurate tumour subtyping by pathologists.