Search

Abstract Foundation models are transforming artificial intelligence (AI) in healthcare by providing modular components adaptable for various downstream tasks, making AI development more scalable and cost-effective. Foundation models for structured electronic health records (EHR), trained on coded medical records from millions of patients, demonstrated benefits including increased performance with fewer training labels, and improved robustness to distribution shifts. However, questions remain on the feasibility of sharing these models across hospitals and their performance in local tasks. This multi-center study examined the adaptability of a publicly accessible structured EHR foundation model (FMSM), trained on 2.57 M patient records from Stanford Medicine. Experiments used EHR data from The Hospital for Sick Children (SickKids) and Medical Information Mart for Intensive Care (MIMIC-IV). We assessed both adaptability via continued pretraining on local data, and task adaptability compared to baselines of locally training models from scratch, including a local foundation model. Evaluations on 8 clinical prediction tasks showed that adapting the off-the-shelf FMSM matched the performance of gradient boosting machines (GBM) locally trained on all data while providing a 13% improvement in settings with few task-specific training labels. Continued pretraining on local data showed FMSM required fewer than 1% of training examples to match the fully trained GBM’s performance, and was 60 to 90% more sample-efficient than training local foundation models from scratch. Our findings demonstrate that adapting EHR foundation models across hospitals provides improved prediction performance at less cost, underscoring the utility of base foundation models as modular components to streamline the development of healthcare AI.

Standard testing for disease evaluation in B-cell acute lymphoblastic leukemia (B-ALL) includes examination of the bone marrow and cerebrospinal fluid. Radiographic or functional imaging are indicated when clinical signs of non-CNS extramedullary disease are present but are not standard in the relapsed/refractory setting. We describe two cases of patients with relapsed/refractory B-ALL with prior exposure to blinatumomab and/or inotuzumab ozogamicin presenting for CAR-T cell treatment. Both patients were thought to only have minimal residual disease (MRD) at the pre-CAR disease assessment, with MRD of 6,648 (0.66%) and 100 (0.01%) cells per million cells, respectively, as measured by next-generation sequencing (NGS) in their bone marrows. Both patients for distinct reasons unrelated to non-CNS extra-medullary (EM) symptoms had PET-MRIs prior to lymphodepletion and CAR T cell infusion. In both cases patients were found to have significant bulky subclinical EM disease that required changes in clinical management. In the newly-emergent era of antigen-targeted immunotherapy, it is foundational that incidence and relapse patterns following targeted therapy are well-understood. Herein we contribute to a growing body of literature addressing this fundamental clinical gap and highlight a future role for formal prospective imaging studies to better establish response, toxicity and relapse patterns following CAR-T cell therapy in EM B-ALL.

Abstract Background Diagnostic codes are commonly used as inputs for clinical prediction models, to create labels for prediction tasks, and to identify cohorts for multicenter network studies. However, the coverage rates of diagnostic codes and their variability across institutions are underexplored. The primary objective was to describe lab- and diagnosis-based labels for 7 selected outcomes at three institutions. Secondary objectives were to describe agreement, sensitivity, and specificity of diagnosis-based labels against lab-based labels. Methods This study included three cohorts: SickKids from The Hospital for Sick Children, and StanfordPeds and StanfordAdults from Stanford Medicine. We included seven clinical outcomes with lab-based definitions: acute kidney injury, hyperkalemia, hypoglycemia, hyponatremia, anemia, neutropenia and thrombocytopenia. For each outcome, we created four lab-based labels (abnormal, mild, moderate and severe) based on test result and one diagnosis-based label. Proportion of admissions with a positive label were presented for each outcome stratified by cohort. Using lab-based labels as the gold standard, agreement using Cohen’s Kappa, sensitivity and specificity were calculated for each lab-based severity level. Results The number of admissions included were: SickKids (n = 59,298), StanfordPeds (n = 24,639) and StanfordAdults (n = 159,985). The proportion of admissions with a positive diagnosis-based label was significantly higher for StanfordPeds compared to SickKids across all outcomes, with odds ratio (99.9% confidence interval) for abnormal diagnosis-based label ranging from 2.2 (1.7–2.7) for neutropenia to 18.4 (10.1–33.4) for hyperkalemia. Lab-based labels were more similar by institution. When using lab-based labels as the gold standard, Cohen’s Kappa and sensitivity were lower at SickKids for all severity levels compared to StanfordPeds. Conclusions Across multiple outcomes, diagnosis codes were consistently different between the two pediatric institutions. This difference was not explained by differences in test results. These results may have implications for machine learning model development and deployment.

Abstract Objective Examine the influence of household income on health‐related quality of life (HRQOL) among children with newly diagnosed acute myeloid leukemia (AML). Design Secondary analysis of data prospectively collected from pediatric patients receiving treatment for AML at 14 hospitals across the United States. Exposure Household income was self‐reported on a demographic survey. The examined mediators included the acuity of presentation and treatment toxicity. Outcome Caregiver proxy reported assessment of patient HRQOL from the Peds QL 4.0 survey. Result Children with AML (n = 131) and caregivers were prospectively enrolled to complete PedsQL assessments. HRQOL scores were better for patients in the lowest versus highest income category (mean ± SD: 76.0 ± 14 household income

Abstract Background There is substantial heterogeneity in symptom management provided to pediatric patients with cancer. The primary objective was to describe the adaptation process and specific adaptation decisions related to symptom management care pathways based on clinical practice guidelines. The secondary objective evaluated if institutional factors were associated with adaptation decisions. Methods Fourteen previously developed symptom management care pathway templates were reviewed by an institutional adaptation team composed of two clinicians at each of 10 institutions. They worked through each statement for all care pathway templates sequentially. The institutional adaptation team made the decision to adopt, adapt or reject each statement, resulting in institution-specific symptom management care pathway drafts. Institutional adaption teams distributed the 14 care pathway drafts to their respective teams; their feedback led to care pathway modifications. Results Initial care pathway adaptation decision making was completed over a median of 4.2 (interquartile range 2.0-5.3) weeks per institution. Across all institutions and among 1350 statements, 551 (40.8%) were adopted, 657 (48.7%) were adapted, 86 (6.4%) were rejected and 56 (4.1%) were no longer applicable because of a previous decision. Most commonly, the reason for rejection was not agreeing with the statement (70/86, 81.4%). Institutional-level factors were not significantly associated with statement rejection. Conclusions Acceptability of the 14 care pathways was evident by most statements being adopted or adapted. The adaptation process was accomplished over a relatively short timeframe. Future work should focus on evaluation of care pathway compliance and determination of the impact of care pathway-consistent care on patient outcomes. Trial registration clinicaltrials.gov, NCT04614662. Registered 04/11/2020, https://clinicaltrials.gov/ct2/show/NCT04614662?term=NCT04614662&draw=2&rank=1 .

Background: Chimeric antigen receptor (CAR) T cell therapy is an effective salvage therapy for pediatric relapsed B-cell acute lymphoblastic leukemia (B-ALL), yet is challenged by high rates of post-CAR relapse. Literature describing specific relapse patterns and extramedullary (EM) sites of involvement in the post-CAR setting remains limited, and a clinical standard for post-CAR disease surveillance has yet to be established. We highlight the importance of integrating peripheral blood minimal residual disease (MRD) testing and radiologic imaging into surveillance strategies, to effectively characterize and capture post-CAR relapse. Main body: Here, we describe the case of a child with multiply relapsed B-ALL who relapsed in the post-CAR setting with gross non-contiguous medullary and EM disease. Interestingly, her relapse was identified first from peripheral blood flow cytometry MRD surveillance, in context of a negative bone marrow aspirate (MRD

Clinical pathways are evidence-based tools that have been integrated into many aspects of pediatric hospital medicine and have proven effective at reducing in-hospital complications from a variety of diseases. Adaptation of similar tools for specific, high-risk patient populations in pediatric oncology has been slower, in part due to patient complexities and variations in management strategies. There are few published studies of clinical pathways for pediatric oncology patients. Pediatric patients with a new diagnosis of leukemia or lymphoma often present with one or more “oncologic emergencies” that require urgent intervention and deliberate multidisciplinary care to prevent significant consequences. Here, we present two clinical pathways that have recently been developed using a multidisciplinary approach at a single institution, intended for the care of patients who present with hyperleukocytosis or an anterior mediastinal mass. These clinical care pathways have provided a critical framework for the immediate care of these patients who are often admitted to the pediatric intensive care unit for initial management. The goal of the pathways is to facilitate multidisciplinary collaborations, expedite diagnosis, and streamline timely treatment initiation. Standardizing the care of high-risk pediatric oncology patients will ultimately decrease morbidity and mortality associated with these diseases to increase the potential for excellent outcomes.

BackgroundGiven the costs of machine learning implementation, a systematic approach to prioritizing which models to implement into clinical practice may be valuable. ObjectiveThe primary objective was to determine the health care attributes respondents at 2 pediatric institutions rate as important when prioritizing machine learning model implementation. The secondary objective was to describe their perspectives on implementation using a qualitative approach. MethodsIn this mixed methods study, we distributed a survey to health system leaders, physicians, and data scientists at 2 pediatric institutions. We asked respondents to rank the following 5 attributes in terms of implementation usefulness: the clinical problem was common, the clinical problem caused substantial morbidity and mortality, risk stratification led to different actions that could reasonably improve patient outcomes, reducing physician workload, and saving money. Important attributes were those ranked as first or second most important. Individual qualitative interviews were conducted with a subsample of respondents. ResultsAmong 613 eligible respondents, 275 (44.9%) responded. Qualitative interviews were conducted with 17 respondents. The most common important attributes were risk stratification leading to different actions (205/275, 74.5%) and clinical problem causing substantial morbidity or mortality (177/275, 64.4%). The attributes considered least important were reducing physician workload and saving money. Qualitative interviews consistently prioritized implementations that improved patient outcomes. ConclusionsRespondents prioritized machine learning model implementation where risk stratification would lead to different actions and clinical problems that caused substantial morbidity and mortality. Implementations that improved patient outcomes were prioritized. These results can help provide a framework for machine learning model implementation.