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.

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.

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.

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.

The effect of a statewide prescription drug monitoring program on opioid prescribing patterns.

PURPOSE: Determine whether opioid prescribing patterns have changed as a result of implementation of a prescription drug monitoring program (PDMP) in the state of Massachusetts. MATERIALS AND METHODS: A multicentered retrospective study was performed including patients who received tonsillectomy, parotidectomy, thyroidectomy or direct laryngoscopy and biopsy with or without rigid esophagoscopy and/or rigid bronchoscopy at Lahey Hospital and Medical Center (Burlington, MA) or Boston Medical Center (Boston, MA). Opioid prescribing patterns were compared for the 12 months prior to implementation of the Massachusetts Prescription Awareness Tool (MassPAT) to 36 months of prescribing patterns post implementation. Quantity of opioids prescribed was based on morphine milligram equivalents (MME). Continuous variables were compared using analysis of variance (ANOVA) while categorical variables were compared using chi-squared test or Fisher's exact test. Multivariate analysis was performed using linear regression. RESULTS: A total of 2281 patients were included in the study. There was a significant association in mean overall MME prescribed comparing pre-MassPAT and post-MassPAT data [tonsillectomy: 635.9 ± 175.6 vs 463.3 ± 177.7 (p < 0.0001), parotidectomy: 250.4 ± 71.33 vs 169.8 ± 79.26 (p < 0.0001), thyroidectomy: 186.2 ± 81.14 vs 118.3 ± 88.79 (p < 0.0001), direct laryngoscopy with biopsy: 308.3 ± 246.9 vs 308.3 ± 246.9 (p = 0.0201)]. There was also a significant association between length of opioid prescription (days) and implementation of MassPAT, but there was no significant difference in the percent of patients requiring refills pre- MassPAT and post-MassPAT. CONCLUSION: This study demonstrates that prescribers have been able to significantly decrease the amount of opioids prescribed for tonsillectomy, parotidectomy, thyroidectomy, and direct laryngoscopy and biopsy and patients have not required additional opioid refills.

Long-Term Opioid Use in Post-Surgical Management of Patients With Head and Neck Cancer.

OBJECTIVE: This study aims to identify clinical and socioeconomic factors associated with long-term, post-surgical opioid use in the head and neck cancer population. METHODS: A single center retrospective study was conducted including patients diagnosed with head and neck cancer between January 1, 2014 and July 1, 2019 who underwent primary surgical management. The primary outcome measure was continued opioid use 6 months after treatment completion. Both demographic and cancer-related variables were recorded to determine what factors were associated with prolonged opioid use. Univariate analysis was performed using chi-squared test for categorical variables and 2-sample t-test for continuous variables. Multivariate analysis was performed using logistic regression. RESULTS: A total of 359 patients received primary surgical management. Forty-five patients (12.53%) continued to take opioids 6 months after treatment completion. Using univariate analysis, patients less than 65 years of age (P = .0126), adjuvant chemoradiation (n = 25, P < .001), and overall length of hospital stay (8.60 ± 8.58 days, P = .0274) were significantly associated with long term opioid use. Multivariate logistic regression showed that adjuvant chemoradiation (OR = 3.446, 95% CI [1.742, 6.820], P = .0004) and overall length of hospital stay (OR = 0.949, 95% CI [0.903, 0.997], P = .0373) to be significantly associated with opioid use 6 months after head and neck cancer treatment. CONCLUSION: Long-term postoperative opioid use in head and neck cancer patients is significantly associated with adjuvant chemoradiation, and patients with longer length of hospital stay. Therefore, future research should focus on interventions to better manage opioid use during the acute treatment period to decrease long-term use.

Geographic Barriers Affect Follow-Up Care in Head and Neck Cancer.

OBJECTIVES/HYPOTHESIS: Follow-up care in head and neck cancers (HNC) is critical in managing patient health. However, social determinants of health (SDOH) can create difficulties in maintaining follow-up care. The study goal is to explore how SDOH impacts maintenance of HNC follow-up care appointments. METHODS: A systematic retrospective chart review of 877 HNC patients diagnosed in the past 10 years a safety-net tertiary care hospital with systems to help reduce care disparities. Cohort groups were identified and compared against protocols for follow-up. Data were analyzed using analysis of variance, chi-square tests, Fisher's exact tests, two-sample t-tests, and simple linear regression. RESULTS: The average length of follow-up time in months and average total number of follow-ups over 5 years were 32.96 (34.60) and 9.24 (7.87), respectively. There was no significant difference in follow-up care between United States (US) versus non-US born and English versus non-English speaking patients. Race/ethnicity, county median household income, insurance status, and county educational attainment were not associated with differences in follow-up. However, living a greater distance from the hospital was associated with lower follow-up length and less frequency in follow-up (P < .0001). CONCLUSION: While income, primary language, country of birth, race/ethnicity, insurance status, and markers of educational attainment do not appear to impact HNC follow-up at our safety-net, tertiary care institution, and distance from hospital remains an important contributor to disparities in care. This study shows that many barriers to care can be addressed in a model that addresses SDOH, but there are barriers that still require additional systems and resources. Laryngoscope, 132:1022-1028, 2022.

Social determinants of health and treatment decisions in head and neck cancer.

BACKGROUND: This study compares select social determinants of health (SDOH) with treatment modality selection and treatment completion in head and neck cancer (HNC) patients, to better understand disparities in health outcomes. METHODS: A retrospective cohort study of HNC (n = 1428) patients was conducted. Demographic and disease-specific variables were recorded, including treatment modality selection and completion. Data were analyzed using two-sample t tests, chi-square, and Fisher's exact tests. RESULTS: Primary language was significantly associated with treatment choice, where non-English speakers were less likely to choose treatment as recommended by the Tumor Board. Lower mean distance from the hospital (37.38 [48.31] vs. 16.92 [19.10], p < 0.0001) and a county-based higher mean percentage of bachelor degree or higher education (42.16 [8.82] vs. 44.95 [6.19], p < 0.0003) were associated with treatment selection. CONCLUSION: Language, distance from the hospital, and education affected treatment selection in this study and may be useful in understanding how to counsel patients on treatment selection for HNC.

Changes in Care Provision During COVID-19 Impact Patient Well-Being.

The fast onset and extensive impact of COVID-19 necessitated strict public health measures and temporary diversion of personnel and resources from other types of medical care. This study examined the prevalence of such disruptions and their impacts on patient-perceived well-being using an untargeted survey. The majority of surveyed patients experienced changes in their routine medical care. Of those whose appointments were postponed or canceled, most patients indicated an overall negative impact on their emotional and physical well-being. We highlighted the impact of disruptions in nonurgent medical care during a large-scale public health emergency.

Spontaneous Labor Onset and Outcomes in Obese Women at Term.

OBJECTIVE: The objective of this study was to compare the rates of spontaneous labor onset and its progression in obese and nonobese women after 37 weeks. STUDY DESIGN: We performed a secondary analysis of a retrospective cohort of all women who were admitted for delivery at ≥ 37 weeks of gestation at a university-based tertiary care center between 2004 and 2010. The cohort was stratified by weeks of gestation at which the patient presented for delivery. The rates of spontaneous labor, vaginal delivery, and augmentation with oxytocin were compared between obese (body mass index [BMI] ≥ 30) and nonobese (BMI < 30) women. RESULTS: Obese women had lower rates of spontaneous labor than nonobese women at every gestational week (37 weeks, 6.1 vs. 9.3%, p < 0.001; 38 weeks, 12.8 vs. 19.2%, p < 0.001; 39 weeks 26.0 vs. 37.0%, p < 0.001; 40 weeks, 39.6 vs. 50.2%, p < 0.001; 41 weeks, 30.8 vs. 38.0%, p < 0.012). Among women who presented in spontaneous labor, obesity was associated with higher rates of augmentation with oxytocin and lower rates of vaginal delivery. CONCLUSION: Obese women at or beyond 37 weeks are less likely to experience spontaneous labor compared with nonobese women. In addition, obese women presenting in spontaneous labor are less likely that nonobese women to have a vaginal delivery at 37 to 40 weeks, even after oxytocin augmentation.