Making sense of artificial intelligence and large language models-including ChatGPT-in pediatric hematology/oncology.

ChatGPT and other artificial intelligence (AI) systems have captivated the attention of healthcare providers and researchers for their potential to improve care processes and outcomes. While these technologies hold promise to automate processes, increase efficiency, and reduce cognitive burden, their use also carries risks. In this commentary, we review basic concepts of AI, outline some of the capabilities and limitations of currently available tools, discuss current and future applications in pediatric hematology/oncology, and provide an evaluation and implementation framework that can be used by pediatric hematologist/oncologists considering the use of AI in clinical practice.

Coronavirus Disease 2019 mRNA Vaccination Appears Safe in Pediatric Patients With Hypersensitivity to Polyethylene Glycolated Escherichia coli L-asparaginase.

Polyethylene glycol-asparaginase (PEGAsp) is an established component of acute leukemia therapy. Hypersensitivity reactions to PEGAsp occur in 10% to 15% of patients, with polyethylene glycol suggested as the antigenic culprit. As coronavirus disease 2019 (COVID-19) mRNA vaccines contain polyethylene glycol, the safety of administration of these vaccines to patients with prior PEGAsp hypersensitivity has been questioned. Between December 21, 2020 and March 3, 2022, 66 patients with acute leukemia and PEGAsp allergy received COVID-19 vaccination. No patients (0/66 0%, 95% CI: 0%-5.4%) experienced an allergic reaction to the vaccine. COVID-19 mRNA vaccination appears to be safe in pediatric and young adult patients with acute lymphoblastic leukemia with PEGAsp allergy.

Survey of international pediatric nutritional supportive care practices: a report from the Pediatric Study Group of the Multinational Association of Supportive Care in Cancer (MASCC).

PURPOSE: Malnutrition is common in children with cancer. While multiple validated malnutrition screens exist, there is no universal, standardized approach to screen or diagnose malnutrition. The Multinational Association of Supportive Care in Cancer (MASCC) Pediatric Study Group is focused on oncologic supportive care for children and young adults. This subgroup designed and administered a pilot study to characterize global malnutrition screening, diagnosis, and treatment practices for pediatric patients with cancer after identifying variations in malnutrition practice patterns within its members. METHODS: A novel, exploratory survey was iteratively developed and distributed in early 2020 to 45 MASCC Pediatric Study Group members. The survey included multiple questions with standard patient presentations and nutritional scenarios, and the respondents selected the answer that best reflected the care patients would receive at their institution. RESULTS: A validated screening tool to assess for malnutrition was routinely used by 15 of 26 respondents (58%). No single validated screen was used by more than 24% of responders, and 11 of 26 (42%) reported not having a standard malnutrition treatment screen. When the same patient was presented with the survey using different malnutrition indicators, patient care plans varied greatly. This was particularly true for z-scores compared to weight percentiles. CONCLUSIONS: Development of consensus recommendations for screening practices, preferred malnutrition indicators, and treatment guidelines could help reduce the underdiagnosis of malnutrition and subsequent variation in its management and ought to be a focus of the global pediatric cancer supportive care community.

Data standards in pediatric oncology: Past, present, and future.

In this commentary, we highlight the central role that data standards play in facilitating data-driven efforts to advance research in pediatric oncology. We discuss the current state of data standards for pediatric oncology and propose five steps to achieve an improved future state with benefits for clinicians, researchers, and patients.

Implementation science in pediatric oncology: A narrative review and future directions.

Implementation science (IS) has garnered attention within oncology, and most prior IS work has focused on adult, not pediatric, oncology. This narrative review broadly characterizes IS for pediatric oncology. It includes studies through 2020 using the following search terms in PubMed, Ovid Medline, and Cochrane: "implementation science," "pediatric," "childhood," "cancer," and "oncology." Systematic review was not performed due to the limited number of heterogeneous studies. Of 216 articles initially reviewed, nine were selected as specific to IS and pediatric oncology. All nine examined oncologic supportive care, cancer prevention, or cancer control. The supportive care focus is potentially due to the presence of cooperative study groups such as the Children's Oncology Group, which efficiently drive cancer-directed therapy changes through clinical trials. Future IS within pediatric oncology should embrace this ecosystem and focus on cancer control interventions that benefit patients across multiple cancer types and patients treated outside cooperative group studies.

Evaluation of an automated pediatric malnutrition screen using anthropometric measurements in the electronic health record: a quality improvement initiative.

PURPOSE: Malnutrition related to undernutrition in pediatric oncology patients is associated with worse outcomes including increased morbidity and mortality. At a tertiary pediatric center, traditional malnutrition screening practices were ineffective at identifying cancer patients at risk for undernutrition and needing nutrition consultation. METHODS: To efficiently identify undernourished patients, an automated malnutrition screen using anthropometric data in the electronic health record (EHR) was implemented. The screen utilized pediatric malnutrition (undernutrition) indicators from the 2014 Consensus Statement of the Academy of Nutrition and Dietetics/American Society for Parenteral and Enteral Nutrition with corresponding structured EHR elements. The time periods before (January 2016-August 2017) and after (September 2017-August 2018) screen implementation were compared. Process metrics including nutrition consults, timeliness of nutrition assessments, and malnutrition diagnoses documentation were assessed using statistical process control charts. Outcome metrics including change in nutritional status at least 3 months after positive malnutrition screen were assessed with the Cochran-Armitage trend test. RESULTS: After automated malnutrition screen implementation, all process metrics demonstrated center line shifts indicating special cause variation. For patient admissions with a positive screen for malnutrition of any severity level, no significant improvement in status of malnutrition was observed after 3 months (P = .13). Sub-analysis of patient admissions with screen-identified severe malnutrition noted improvement in degree of malnutrition after 3 months (P = .02). CONCLUSIONS: Select 2014 Consensus Statement indicators for pediatric malnutrition can be implemented as an automated screen using structured EHR data. The automated screen efficiently identifies oncology patients at risk of malnutrition and may improve clinical outcomes.

A robustness metric for biological data clustering algorithms.

BACKGROUND: Cluster analysis is a core task in modern data-centric computation. Algorithmic choice is driven by factors such as data size and heterogeneity, the similarity measures employed, and the type of clusters sought. Familiarity and mere preference often play a significant role as well. Comparisons between clustering algorithms tend to focus on cluster quality. Such comparisons are complicated by the fact that algorithms often have multiple settings that can affect the clusters produced. Such a setting may represent, for example, a preset variable, a parameter of interest, or various sorts of initial assignments. A question of interest then is this: to what degree do the clusters produced vary as setting values change? RESULTS: This work introduces a new metric, termed simply "robustness", designed to answer that question. Robustness is an easily-interpretable measure of the propensity of a clustering algorithm to maintain output coherence over a range of settings. The robustness of eleven popular clustering algorithms is evaluated over some two dozen publicly available mRNA expression microarray datasets. Given their straightforwardness and predictability, hierarchical methods generally exhibited the highest robustness on most datasets. Of the more complex strategies, the paraclique algorithm yielded consistently higher robustness than other algorithms tested, approaching and even surpassing hierarchical methods on several datasets. Other techniques exhibited mixed robustness, with no clear distinction between them. CONCLUSIONS: Robustness provides a simple and intuitive measure of the stability and predictability of a clustering algorithm. It can be a useful tool to aid both in algorithm selection and in deciding how much effort to devote to parameter tuning.

Health-related Google searches performed by parents of pediatric oncology patients.

BACKGROUND: Little is known about the specific information parents of children with cancer search for online. Understanding the content of parents' searches over time could offer insight into what matters most to parents and identify knowledge gaps that could inform more comprehensive approaches to family education and support. METHODS: We describe parents' health-related Google searches starting six months before cancer diagnosis and extending through the date of study enrollment, which was at least one month after initiating cancer treatment. Searches were obtained retrospectively and grouped into health-related and non-health-related categories. The median time to parent enrollment from date of cancer diagnosis was 264 days. RESULTS: Parents searched for health-related topics more frequently than the general population (13% vs 5%). Health-related searches increased in the months preceding the child's cancer diagnosis and most commonly pertained to symptoms and logistics, "directions to hospital." Health-related search volume peaked about a month after cancer diagnosis when general health-related searches were present in addition to cancer-specific searches. Eighteen percent of health-related searches were cancer specific, and of these cancer-specific searches, 54% pertained to support, for example "cancer quote for son." CONCLUSIONS: Google search content offers insight into what matters to parents of cancer patients. Understanding search content could inform more comprehensive approaches to family education and support initiatives.

Development and evaluation of a computable phenotype to identify pediatric patients with leukemia and lymphoma treated with chemotherapy using electronic health record data.

BACKGROUND: Widespread implementation of electronic health records (EHR) has created new opportunities for pediatric oncology observational research. Little attention has been given to using EHR data to identify patients with pediatric hematologic malignancies. METHODS: This study used EHR-derived data in a pediatric clinical data research network, PEDSnet, to develop and evaluate a computable phenotype algorithm to identify pediatric patients with leukemia and lymphoma who received treatment with chemotherapy. To guide early development, multiple computable phenotype-defined cohorts were compared to one institution's tumor registry. The most promising algorithm was chosen for formal evaluation and consisted of at least two leukemia/lymphoma diagnoses (Systematized Nomenclature of Medicine codes) within a 90-day period, two chemotherapy exposures, and three hematology-oncology provider encounters. During evaluation, the computable phenotype was executed against EHR data from 2011 to 2016 at three large institutions. Classification accuracy was assessed by masked medical record review with phenotype-identified patients compared to a control group with at least three hematology-oncology encounters. RESULTS: The computable phenotype had sensitivity of 100% (confidence interval [CI] 99%, 100%), specificity of 99% (CI 99%, 100%), positive predictive value (PPV) and negative predictive value (NPV) of 100%, and C-statistic of 1 at the development institution. The computable phenotype performance was similar at the two test institutions with sensitivity of 100% (CI 99%, 100%), specificity of 99% (CI 99%, 100%), PPV of 96%, NPV of 100%, and C-statistic of 0.99. CONCLUSION: The EHR-based computable phenotype is an accurate cohort identification tool for pediatric patients with leukemia and lymphoma who have been treated with chemotherapy and is ready for use in clinical studies.

Relationship Between State-Level Google Online Search Volume and Cancer Incidence in the United States: Retrospective Study.

BACKGROUND: In the United States, cancer is common, with high morbidity and mortality; cancer incidence varies between states. Online searches reflect public awareness, which could be driven by the underlying regional cancer epidemiology. OBJECTIVE: The objective of our study was to characterize the relationship between cancer incidence and online Google search volumes in the United States for 6 common cancers. A secondary objective was to evaluate the association of search activity with cancer-related public events and celebrity news coverage. METHODS: We performed a population-based, retrospective study of state-level cancer incidence from 2004 through 2013 reported by the Centers for Disease Control and Prevention for breast, prostate, colon, lung, and uterine cancers and leukemia compared to Google Trends (GT) relative search volume (RSV), a metric designed by Google to allow interest in search topics to be compared between regions. Participants included persons in the United States who searched for cancer terms on Google. The primary measures were the correlation between annual state-level cancer incidence and RSV as determined by Spearman correlation and linear regression with RSV and year as independent variables and cancer incidence as the dependent variable. Temporal associations between search activity and events raising public awareness such as cancer awareness months and cancer-related celebrity news were described. RESULTS: At the state level, RSV was significantly correlated to incidence for breast (r=.18, P=.001), prostate (r=-.27, P<.001), lung (r=.33, P<.001), and uterine cancers (r=.39, P<.001) and leukemia (r=.13, P=.003) but not colon cancer (r=-.02, P=.66). After adjusting for time, state-level RSV was positively correlated to cancer incidence for all cancers: breast (P<.001, 95% CI 0.06 to 0.19), prostate (P=.38, 95% CI -0.08 to 0.22), lung (P<.001, 95% CI 0.33 to 0.46), colon (P<.001, 95% CI 0.11 to 0.17), and uterine cancers (P<.001, 95% CI 0.07 to 0.12) and leukemia (P<.001, 95% CI 0.01 to 0.03). Temporal associations in GT were noted with breast cancer awareness month but not with other cancer awareness months and celebrity events. CONCLUSIONS: Cancer incidence is correlated with online search volume at the state level. Search patterns were temporally associated with cancer awareness months and celebrity announcements. Online searches reflect public awareness. Advancing understanding of online search patterns could augment traditional epidemiologic surveillance, provide opportunities for targeted patient engagement, and allow public information campaigns to be evaluated in ways previously unable to be measured.

GeneWeaver: finding consilience in heterogeneous cross-species functional genomics data.

A persistent challenge lies in the interpretation of consensus and discord from functional genomics experimentation. Harmonizing and analyzing this data will enable investigators to discover relations of many genes to many diseases, and from many phenotypes and experimental paradigms to many diseases through their genomic substrates. The GeneWeaver.org system provides a platform for cross-species integration and interrogation of heterogeneous curated and experimentally derived functional genomics data. GeneWeaver enables researchers to store, share, analyze, and compare results of their own genome-wide functional genomics experiments in an environment containing rich companion data obtained from major curated repositories, including the Mouse Genome Database and other model organism databases, along with derived data from highly specialized resources, publications, and user submissions. The data, largely consisting of gene sets and putative biological networks, are mapped onto one another through gene identifiers and homology across species. A versatile suite of interactive tools enables investigators to perform a variety of set analysis operations to find consilience among these often noisy experimental results. Fast algorithms enable real-time analysis of large queries. Specific applications include prioritizing candidate genes for quantitative trait loci, identifying biologically valid mouse models and phenotypic assays for human disease, finding the common biological substrates of related diseases, classifying experiments and the biological concepts they represent from empirical data, and applying patterns of genomic evidence to implicate novel genes in disease. These results illustrate an alternative to strict emphasis on replicability, whereby researchers classify experimental results to identify the conditions that lead to their similarity.

On finding bicliques in bipartite graphs: a novel algorithm and its application to the integration of diverse biological data types.

BACKGROUND: Integrating and analyzing heterogeneous genome-scale data is a huge algorithmic challenge for modern systems biology. Bipartite graphs can be useful for representing relationships across pairs of disparate data types, with the interpretation of these relationships accomplished through an enumeration of maximal bicliques. Most previously-known techniques are generally ill-suited to this foundational task, because they are relatively inefficient and without effective scaling. In this paper, a powerful new algorithm is described that produces all maximal bicliques in a bipartite graph. Unlike most previous approaches, the new method neither places undue restrictions on its input nor inflates the problem size. Efficiency is achieved through an innovative exploitation of bipartite graph structure, and through computational reductions that rapidly eliminate non-maximal candidates from the search space. An iterative selection of vertices for consideration based on non-decreasing common neighborhood sizes boosts efficiency and leads to more balanced recursion trees. RESULTS: The new technique is implemented and compared to previously published approaches from graph theory and data mining. Formal time and space bounds are derived. Experiments are performed on both random graphs and graphs constructed from functional genomics data. It is shown that the new method substantially outperforms the best previous alternatives. CONCLUSIONS: The new method is streamlined, efficient, and particularly well-suited to the study of huge and diverse biological data. A robust implementation has been incorporated into GeneWeaver, an online tool for integrating and analyzing functional genomics experiments, available at http://geneweaver.org. The enormous increase in scalability it provides empowers users to study complex and previously unassailable gene-set associations between genes and their biological functions in a hierarchical fashion and on a genome-wide scale. This practical computational resource is adaptable to almost any applications environment in which bipartite graphs can be used to model relationships between pairs of heterogeneous entities.

Efficient prediction of human protein-protein interactions at a global scale.

BACKGROUND: Our knowledge of global protein-protein interaction (PPI) networks in complex organisms such as humans is hindered by technical limitations of current methods. RESULTS: On the basis of short co-occurring polypeptide regions, we developed a tool called MP-PIPE capable of predicting a global human PPI network within 3 months. With a recall of 23% at a precision of 82.1%, we predicted 172,132 putative PPIs. We demonstrate the usefulness of these predictions through a range of experiments. CONCLUSIONS: The speed and accuracy associated with MP-PIPE can make this a potential tool to study individual human PPI networks (from genomic sequences alone) for personalized medicine.

Automation of Hematopoietic Cell Transplantation Outcomes Reporting Leads to Dramatic Reduction in Errors Reported to Real-World Data Registry.

Institutions that perform hematopoietic cell transplantation (HCT) are required by law to report standardized, structured data on transplantation outcomes. A key post-transplantation outcome is engraftment, the time between HCT infusion and reemergence of circulating neutrophils and platelets. At our center, we found that manual chart abstraction for engraftment data was highly error-prone. We developed a custom R/Shiny application that automatically calculates engraftment dates and displays them in an intuitive format to augment the manual chart review. Our hypothesis was that use of the application to assist with calculating and reporting engraftment dates would be associated with a decreased error rate. The study was conducted at a single tertiary care institution. The application was developed in a collaborative, multidisciplinary fashion by members of an embedded cellular therapy informatics team. Retrospective validation of the application's accuracy was conducted on all malignant HCTs from February 2016 to December 2020 (n = 198). Real-world use of the application was evaluated prospectively from April 2021 through April 2022 (n = 53). The Welch 2-sample t test was used to compare error rates preimplementation and postimplementation. Data were visualized using p charts, and standard special cause variation rules were applied. The accuracy of reported data postdeployment increased dramatically; the engraftment error rate decreased from 15% to 3.8% for neutrophils (P = .003) and from 28% to 1.9% for platelets (P < .001). This study demonstrates the effective deployment of a custom R/Shiny application that was associated with significantly reduced error rates in HCT engraftment reporting for operational, research, and regulatory purposes. Users reported subjective satisfaction with the application and that it addressed difficulties with the legacy manual process. Identifying and correcting erroneous data in engraftment reporting could lead to a more efficient and accurate nationwide assessment of transplantation success. Furthermore, we show that it is possible and practical for academic medical centers to create and support embedded informatics teams that can quickly build applications for clinical operations in a manner compliant with regulatory requirements.

Opportunities to improve diagnosis in emergency transfers to the pediatric intensive care unit.

BACKGROUND: Late recognition of in-hospital deterioration is a source of preventable harm. Emergency transfers (ET), when hospitalized patients require intensive care unit (ICU) interventions within 1 h of ICU transfer, are a proximal measure of late recognition associated with increased mortality and length of stay (LOS). OBJECTIVE: To apply diagnostic process improvement frameworks to identify missed opportunities for improvement in diagnosis (MOID) in ETs and evaluate their association with outcomes. DESIGN, SETTINGS, AND PARTICIPANTS: A single-center retrospective cohort study of ETs, January 2015 to June 2019. ET criteria include intubation, vasopressor initiation, or ≥ $\ge \phantom{\rule{}{0ex}}$ 60 mL/kg fluid resuscitation 1 h before to 1 h after ICU transfer. The primary exposure was the presence of MOID, determined using SaferDx. Cases were screened by an ICU and non-ICU physician. Final determinations were made by an interdisciplinary group. Diagnostic process improvement opportunities were identified. MAIN OUTCOME AND MEASURES: Primary outcomes were in-hospital mortality and posttransfer LOS, analyzed by multivariable regression adjusting for age, service, deterioration category, and pretransfer LOS. RESULTS: MOID was identified in 37 of 129 ETs (29%, 95% confidence interval [CI] 21%-37%). Cases with MOID differed in originating service, but not demographically. Recognizing the urgency of an identified condition was the most common diagnostic process opportunity. ET cases with MOID had higher odds of mortality (odds ratio 5.5; 95% CI 1.5-20.6; p = .01) and longer posttransfer LOS (rate ratio 1.7; 95% CI 1.1-2.6; p = .02). CONCLUSION: MOID are common in ETs and are associated with increased mortality risk and posttransfer LOS. Diagnostic improvement strategies should be leveraged to support earlier recognition of clinical deterioration.

Supportive Care in Pediatric Oncology: Opportunities and Future Directions.

The optimization of outcomes for pediatric cancer patients relies on the successful advancement of supportive care to ease the treatment burden and mitigate the long-term impacts of cancer therapy. Advancing pediatric supportive care requires research prioritization as well as the development and implementation of innovations. Like the prevailing theme throughout pediatric oncology, there is a clear need for personalized or precision approaches that are consistent, evidence-based, and guided by clinical practice guidelines. By incorporating technology and datasets, we can address questions which may not be feasible to explore in clinical trials. Now is the time to listen to patients' voices by using patient-reported outcomes (PROs) to ensure that their contributions and experiences inform clinical care plans. Furthermore, while the extrapolation of knowledge and approaches from adult populations may suffice in the absence of pediatric-specific evidence, there is a critical need to specifically understand and implement elements of general and developmental pediatrics like growth, nutrition, development, and physical activity into care. Increased research funding for pediatric supportive care is critical to address resource availability, equity, and disparities across the globe. Our patients deserve to enjoy healthy, productive lives with optimized and enriched supportive care that spans the spectrum from diagnosis to survivorship.

The maximum clique enumeration problem: algorithms, applications, and implementations.

BACKGROUND: The maximum clique enumeration (MCE) problem asks that we identify all maximum cliques in a finite, simple graph. MCE is closely related to two other well-known and widely-studied problems: the maximum clique optimization problem, which asks us to determine the size of a largest clique, and the maximal clique enumeration problem, which asks that we compile a listing of all maximal cliques. Naturally, these three problems are NP-hard, given that they subsume the classic version of the NP-complete clique decision problem. MCE can be solved in principle with standard enumeration methods due to Bron, Kerbosch, Kose and others. Unfortunately, these techniques are ill-suited to graphs encountered in our applications. We must solve MCE on instances deeply seeded in data mining and computational biology, where high-throughput data capture often creates graphs of extreme size and density. MCE can also be solved in principle using more modern algorithms based in part on vertex cover and the theory of fixed-parameter tractability (FPT). While FPT is an improvement, these algorithms too can fail to scale sufficiently well as the sizes and densities of our datasets grow. RESULTS: An extensive testbed of benchmark graphs are created using publicly available transcriptomic datasets from the Gene Expression Omnibus (GEO). Empirical testing reveals crucial but latent features of such high-throughput biological data. In turn, it is shown that these features distinguish real data from random data intended to reproduce salient topological features. In particular, with real data there tends to be an unusually high degree of maximum clique overlap. Armed with this knowledge, novel decomposition strategies are tuned to the data and coupled with the best FPT MCE implementations. CONCLUSIONS: Several algorithmic improvements to MCE are made which progressively decrease the run time on graphs in the testbed. Frequently the final runtime improvement is several orders of magnitude. As a result, instances which were once prohibitively time-consuming to solve are brought into the domain of realistic feasibility.

Development of an Informatics Algorithm to Link Seasonal Infectious Diseases to Birth-Dependent Diseases Across Species: A Case Study with Osteosarcoma.

Many diseases have been linked with birth seasonality, and these fall into four main categories: mental, cardiovascular, respiratory and women's reproductive health conditions. Informatics methods are needed to uncover seasonally varying infectious diseases that may be responsible for the increased birth month-dependent disease risk observed. We have developed a method to link seasonal infectious disease data from the USA to birth month dependent disease data from humans and canines. We also include seasonal air pollution and climate data to determine the seasonal factors most likely involved in the response. We test our method with osteosarcoma, a rare bone cancer. We found the Lyme disease incidence was the most strongly correlated significant factor in explaining the birth month-osteosarcoma disease pattern (R=0.418, p=2.80X10-23), and this was true across all populations observed: canines, pediatric, and adult populations.