Education and electronic medical records and genomics network, challenges, and lessons learned from a large-scale clinical trial using polygenic risk scores.

Polygenic risk scores (PRS) have potential to improve health care by identifying individuals that have elevated risk for common complex conditions. Use of PRS in clinical practice, however, requires careful assessment of the needs and capabilities of patients, providers, and health care systems. The electronic Medical Records and Genomics (eMERGE) network is conducting a collaborative study which will return PRS to 25,000 pediatric and adult participants. All participants will receive a risk report, potentially classifying them as high risk (∼2-10% per condition) for 1 or more of 10 conditions based on PRS. The study population is enriched by participants from racial and ethnic minority populations, underserved populations, and populations who experience poorer medical outcomes. All 10 eMERGE clinical sites conducted focus groups, interviews, and/or surveys to understand educational needs among key stakeholders-participants, providers, and/or study staff. Together, these studies highlighted the need for tools that address the perceived benefit/value of PRS, types of education/support needed, accessibility, and PRS-related knowledge and understanding. Based on findings from these preliminary studies, the network harmonized training initiatives and formal/informal educational resources. This paper summarizes eMERGE's collective approach to assessing educational needs and developing educational approaches for primary stakeholders. It discusses challenges encountered and solutions provided.

Engaging Adolescents and Young Adults in Decisions About Return of Genomic Research Results: a mixed-methods longitudinal clinical trial protocol.

Background: To protect minors' future autonomy, professional organizations have historically discouraged returning predictive adult-onset genetic test results and carrier status to children. Recent clinical guidance diverges from this norm, suggesting that when minors have genomic sequencing performed for clinical purposes, parents and children should have the opportunity to learn secondary findings, including for some adult-onset conditions. While parents can currently opt in or out of receiving their child's secondary findings, the American Society of Human Genetics Workgroup on Pediatric Genetic and Genomic Testing suggests including adolescents in the decision-making process. However, it is not clear what factors young people consider when given the opportunity to learn genetic findings for themselves. We are examining adolescents', young adults', and parents' (if applicable) decisions about learning genomic information for the adolescent. Methods: We are enrolling assenting (ages 13-17) adolescents and consenting (ages 18-21) young adults in a prospective genomic screening study to assess the choices they make about receiving individual genomic results. Participants use an online tool to indicate whether they want to learn their personal genetic risk for specific preventable, treatable, and adult-onset conditions, as well as carrier status for autosomal recessive conditions. We are examining 1) how choices differ between adolescent and young adult cohorts (as well as between adolescents/young adults and parents) and 2) decisional conflict and stability across study timepoints. Results are returned based on participants' choices. Qualitative interviews with a subset of participants explore decisional stability, adolescent/young adult engagement with parents in decision-making, and the impact of learning pathogenic/likely pathogenic and carrier results. Discussion: This study explores decision making and decision stability between adolescents and parents (where applicable), as well as the ethical implications and impact of return of clinical-grade genetic research results to adolescents and young adults. The results of this study will contribute empirical evidence to support best practices and guidance on engaging young people in genetic research studies and clinical care that offer return of results.

Artificial intelligence in bronchopulmonary dysplasia- current research and unexplored frontiers.

IMPACT: Provide an overview of bronchopulmonary dysplasia, its definitions, and their shortcomings. Explore the areas where machine learning may be used to further our understanding of bronchopulmonary dysplasia.

Developing video education materials for the return of genomic test results to parents and adolescents.

Objective: To describe the development, implementation, and revision of a video to provide information about genomic testing and the return of genomic research results to adolescents and parents. Methods: Formative, community-engaged research was conducted in three stages: development, implementation, and revision. Existing research participant advisory groups were used for focus groups and convenience sampling was used for interviews. Participants included parents, young adults without children, and adolescents. Transcripts of recorded sessions were used for formative analysis. Results: Video was the preferred format for delivering genomic testing information to adolescents during the development stage. During implementation, adolescents identified video length as an impediment to recall. During the revision stage, participants preferred the video in separate short segments, supported plan to require only one short video and leaving other short videos optional. Participants were divided on whether the required short video provided enough information, but all participants reported that watching additional videos would not have changed their decisions about receiving test results. Conclusion: Genomic education videos should be brief (<4 mins) to improve the odds that participants will view the entirety of any required video. Innovation: The development of participant materials should incorporate plans for monitoring implementation and plans for revising materials.

Usability of human-computer interaction in neonatal care.

While a goal for Electronic Health Record (EHR) technologies was to improve quality, efficiency, and safety, the usability of EHRs has remained poor. The relation to patient harm and user satisfaction cannot be ignored. Optimization of EHR usability is imperative to improving the outcomes for critically ill patients, especially neonates who are at the extremes of physiologic variability. Further development and integration of metadata with predictive modeling and clinical protocols can support provider decision making, increase efficiency and safety, and reduce clinician burnout. This paper reviews EHR usability and identifies opportunities to improve the EHR specific to neonatal care.

Machine Learning for Detection of Correct Peripherally Inserted Central Catheter Tip Position from Radiology Reports in Infants.

BACKGROUND: In critically ill infants, the position of a peripherally inserted central catheter (PICC) must be confirmed frequently, as the tip may move from its original position and run the risk of hyperosmolar vascular damage or extravasation into surrounding spaces. Automated detection of PICC tip position holds great promise for alerting bedside clinicians to noncentral PICCs. OBJECTIVES: This research seeks to use natural language processing (NLP) and supervised machine learning (ML) techniques to predict PICC tip position based primarily on text analysis of radiograph reports from infants with an upper extremity PICC. METHODS: Radiographs, containing a PICC line in infants under 6 months of age, were manually classified into 12 anatomical locations based on the radiologist's textual report of the PICC line's tip. After categorization, we performed a 70/30 train/test split and benchmarked the performance of seven different (neural network, support vector machine, the naïve Bayes, decision tree, random forest, AdaBoost, and K-nearest neighbors) supervised ML algorithms. After optimization, we calculated accuracy, precision, and recall of each algorithm's ability to correctly categorize the stated location of the PICC tip. RESULTS: A total of 17,337 radiographs met criteria for inclusion and were labeled manually. Interrater agreement was 99.1%. Support vector machines and neural networks yielded accuracies as high as 98% in identifying PICC tips in central versus noncentral position (binary outcome) and accuracies as high as 95% when attempting to categorize the individual anatomical location (12-category outcome). CONCLUSION: Our study shows that ML classifiers can automatically extract the anatomical location of PICC tips from radiology reports. Two ML classifiers, support vector machine (SVM) and a neural network, obtained top accuracies in both binary and multiple category predictions. Implementing these algorithms in a neonatal intensive care unit as a clinical decision support system may help clinicians address PICC line position.

Risk Calculator to Predict Cesarean Delivery Among Women Undergoing Induction of Labor.

OBJECTIVE: To develop and validate a predictive risk calculator for cesarean delivery among women undergoing induction of labor. METHODS: We performed a population-based cohort study of all women who had singleton live births after undergoing induction of labor from 32 0/7 to 42 6/7 weeks of gestation in the United States from 2012 to 2016. The primary objective was to build a predictive model estimating the probability of cesarean delivery after induction of labor using antenatal factors obtained from de-identified U.S. live-birth records. Multivariable logistic regression estimated the association of these factors on risk of cesarean delivery. K-fold cross validation was performed for internal validation of the model, followed by external validation using a separate live-birth cohort from 2017. A publicly available online calculator was developed after validation and calibration were performed for individual risk assessment. The seven variables selected for inclusion in the model by magnitude of influence were prior vaginal delivery, maternal weight at delivery, maternal height, maternal age, prior cesarean delivery, gestational age at induction, and maternal race. RESULTS: From 2012 to 2016, there were 19,844,580 live births in the United States, of which 4,177,644 women with singleton gestations underwent induction of labor. Among these women, 800,423 (19.2%) delivered by cesarean. The receiver operating characteristic curve for the seven-variable model achieved an area under the curve (AUC) of 0.787 (95% CI 0.786-0.788). External validation demonstrated a consistent measure of discrimination with an AUC of 0.783 (95% CI 0.764-0.802). CONCLUSION: This validated predictive model uses seven variables that were obtainable from the patient's medical record and discriminates between women at increased or decreased risk of cesarean delivery after induction of labor. This risk calculator, found at https://ob.tools/iol-calc, can be used in addition to the Bishop score by health care providers in counseling women who are undergoing an induction of labor and allocating appropriate resources for women at high risk for cesarean delivery.

Electronic Health Record-Embedded Decision Support Platform for Morphine Precision Dosing in Neonates.

Morphine is the opioid most commonly used for neonatal pain management. In intravenous form, it is administered as continuous infusions and intermittent injections, mostly based on empirically established protocols. Inadequate pain control in neonates can cause long-term adverse consequences; however, providing appropriate individualized morphine dosing is particularly challenging due to the interplay of rapid natural physiological changes and multiple life-sustaining procedures in patients who cannot describe their symptoms. At most institutions, morphine dosing in neonates is largely carried out as an iterative process using a wide range of starting doses and then titrating to effect based on clinical response and side effects using pain scores and levels of sedation. Our background data show that neonates exhibit large variability in morphine clearance resulting in a wide range of exposures, which are poorly predicted by dose alone. Here, we describe the development and implementation of an electronic health record-integrated, model-informed decision support platform for the precision dosing of morphine in the management of neonatal pain. The platform supports pharmacokinetic model-informed dosing guidance and has functionality to incorporate real-time drug concentration information. The feedback is inserted directly into prescribers' workflows so that they can make data-informed decisions. The expected outcomes are better clinical efficacy and safety with fewer side effects in the neonatal population.

Predictive Model for Failed Induction of Labor Among Obese Women.

OBJECTIVE: To evaluate antenatal risk factors associated with failed induction of labor among obese women to develop a predictive model for induction of labor outcome. METHODS: We conducted a population-based cohort study of all obese (body mass index higher than 30.0) women with singleton live births who underwent attempted induction of labor between 37 and 44 weeks of gestation in the United States from 2012 to 2016 using de-identified U.S live birth records. The primary objective was to build a predictive model for the probability of induction of labor failure using antenatal factors. Multivariable logistic regression estimated the association of these factors on risk of failed induction of labor. We performed k-fold cross-validation for internal validation and then externally validated the model using a separate live birth cohort from 2017 (n=197,982). An online calculator was developed after validation, and calibration was performed. The 10 variables selected for inclusion in the model in order of significance were prior vaginal delivery, prior cesarean delivery, maternal height, age, weight at delivery, parity, gestational weight gain, Medicaid insurance, pregestational diabetes, and chronic hypertension. RESULTS: Among 19,844,580 live births in the United States between 2012 and 2016, 1,098,981 obese women with singleton pregnancies underwent induction of labor, of which 273,184 (24.9%) were unsuccessful. The receiver operator characteristic curve for the 10 variable model achieved an area under the curve (AUC) of 0.79 (95% CI 0.78-0.79). External validation demonstrated a consistent measure of discrimination, with an AUC curve of 0.77 (95% CI 0.76-0.77). CONCLUSION: This model provides valuable estimation as to the cumulative effect of multiple factors on the risk of failed induction of labor among obese parturients. The predictive model identifies women at increased or decreased risk (ie, greater than 75% vs less than 20%) for cesarean delivery. This risk calculator may be a useful tool for practitioners in the counseling, triaging, risk stratifying, and delivery planning for obese women before attempted induction of labor.

Predictive Model of Factors Associated With Maternal Intensive Care Unit Admission.

OBJECTIVE: Severe maternal morbidity has increased in the United States over the past two decades by approximately 200%, to 144 cases per 10,000 delivery hospitalizations. There are limited data available to assist in identifying at-risk women before parturition. We sought to evaluate risk factors associated with maternal admission to an intensive care unit (ICU). METHODS: We conducted a population-based cohort study of all live births delivered between 20 and 44 weeks of gestation in the United States during 2012-2016. Our primary objective was to identify prenatal factors associated with increased risk of maternal ICU admission to build a multivariable predictive model to estimate the association of these factors with ICU admission risk. We performed k-fold cross-validation for internal validation and then externally validated the model on a separate live birth cohort (2006-2011, n=856,255). RESULTS: There were 18,745,615 live births in the United States between 2012 and 2016. Among the mothers of these live newborns, 27,602 (0.15%) were admitted to the ICU in the peripartum period. Fourteen variables were selected for inclusion in the predictive model for maternal ICU admission. The predicted minimal and maximal risk for ICU admission ranged 0-25%. The receiver operating characteristic curve for these 14 variables achieved an area under the curve (AUC) of 0.81 (95% CI 0.79-0.81). External validation with a separate live birth cohort demonstrated a consistent measure of discrimination with an AUC of 0.83 (95% CI 0.82-0.84). Using a relatively high cut point of 5.0% or more predicted risk for ICU admission, achieved a positive predictive value (PPV) of only 4.0%. CONCLUSION: This model provides insight as to the cumulative effect of multiple risk factors on maternal ICU admission risk. The predictive model achieves an AUC of 0.81, discriminating women with significantly increased risk (30-fold) for ICU admission. Nonetheless, because of the low frequency of maternal ICU admission, the PPV of the model was low and therefore whether models such as ours may be beneficial in future efforts to reduce the prevalence and burden of maternal morbidity is uncertain.

Assessing the Readability of Freely Available ICU Notes.

Unstructured data stored in an electronic health record (EHR) system can be very informative but require techniques such as natural language processing to extract the information. Developing such techniques requires shared data, but clinical data are often not easy to access. A freely available intensive care unit database, MIMIC-III, was released in 2016 to address this issue and benefit the informatics research community. While the database has been utilized by a few studies, the text characteristics of the notes have not been summarized. In this study, we present the summary of the basic text characteristics and the readability of the MIMIC-III ICU notes. We further compare the results with our previous study where proprietary EHR notes were used. The results show that the text characteristics of MIMIC-III notes were comparable with proprietary EHR notes, although the note readability index was slightly lower. The clinical notes in MIMIC-III can be a viable option for researchers who are interested in clinicians' language use but have no access to proprietary EHR systems.

Neonatal Lung Disease Associated with TBX4 Mutations.

Variable lung disease was documented in 2 infants with heterozygous TBX4 mutations; their clinical presentations, pathology, and outcomes were distinct. These findings demonstrate that TBX4 gene mutations are associated with neonatal respiratory failure and highlight the wide spectrum of clinicopathological outcomes that have implications for patient diagnosis and management.

A Randomized Trial Comparing Classical Participatory Design to VandAID, an Interactive CrowdSourcing Platform to Facilitate User-centered Design.

BACKGROUND: Early involvement of stakeholders in the design of medical software is particularly important due to the need to incorporate complex knowledge and actions associated with clinical work. Standard user-centered design methods include focus groups and participatory design sessions with individual stakeholders, which generally limit user involvement to a small number of individuals due to the significant time investments from designers and end users. OBJECTIVES: The goal of this project was to reduce the effort for end users to participate in co-design of a software user interface by developing an interactive web-based crowdsourcing platform. METHODS: In a randomized trial, we compared a new web-based crowdsourcing platform to standard participatory design sessions. We developed an interactive, modular platform that allows responsive remote customization and design feedback on a visual user interface based on user preferences. The responsive canvas is a dynamic HTML template that responds in real time to user preference selections. Upon completion, the design team can view the user's interface creations through an administrator portal and download the structured selections through a REDCap interface. RESULTS: We have created a software platform that allows users to customize a user interface and see the results of that customization in real time, receiving immediate feedback on the impact of their design choices. Neonatal clinicians used the new platform to successfully design and customize a neonatal handoff tool. They received no specific instruction and yet were able to use the software easily and reported high usability. CONCLUSIONS: VandAID, a new web-based crowdsourcing platform, can involve multiple users in user-centered design simultaneously and provides means of obtaining design feedback remotely. The software can provide design feedback at any stage in the design process, but it will be of greatest utility for specifying user requirements and evaluating iterative designs with multiple options.

Lethal neonatal meningoencephalitis caused by multi-drug resistant, highly virulent Escherichia coli.

Neonatal meningitis is a rare but devastating condition. Multi-drug resistant (MDR) bacteria represent a substantial global health risk. This study reports on an aggressive case of lethal neonatal meningitis due to a MDR Escherichia coli (serotype O75:H5:K1). Serotyping, MDR pattern and phylogenetic typing revealed that this strain is an emergent and highly virulent neonatal meningitis E. coli isolate. The isolate was resistant to both ampicillin and gentamicin; antibiotics currently used for empiric neonatal sepsis treatment. The strain was also positive for multiple virulence genes including K1 capsule, fimbrial adhesion fimH, siderophore receptors iroN, fyuA and iutA, secreted autotransporter toxin sat, membrane associated proteases ompA and ompT, type II polysaccharide synthesis genes (kpsMTII) and pathogenicity-associated island (PAI)-associated malX gene. The presence of highly-virulent MDR organisms isolated in neonates underscores the need to implement rapid drug resistance diagnostic methods and should prompt consideration of alternate empiric therapy in neonates with Gram negative meningitis.

The first reported case of compound heterozygous IL1RN mutations causing deficiency of the interleukin-1 receptor antagonist.

Interleukin-1 receptor antagonist (IL-1Ra) deficiency is a rare autoinflammatory disease involving neonatal onset of pustulosis, periostitis, and sterile osteomyelitis. We report the case of a 2-week-old male who presented with a swollen, erythematous left index finger and elevated serum markers of inflammation. He later developed cyclical fevers, diffuse pustular skin lesions, and thrombus formation. After not responding to broad-spectrum antimicrobial therapy and achieving only moderate success with systemic steroid therapy, he was ultimately treated with recombinant IL-1Ra, anakinra, and experienced significant clinical improvement. Sequencing of his IL1RN gene revealed that the patient was compound heterozygous for a known mutation (E77X) associated with IL-1Ra deficiency and a novel mutation in exon 2 of the gene (c.140delC; p.T47TfsX4). His case highlights IL-1Ra deficiency as an autoinflammatory disease that is distinct from neonatal-onset multisystem inflammatory disease but that also responds well to anakinra. Our patient is the first reported compound heterozygote for E77X and the novel mutation in exon 2 of the gene, the latter of which adds to what will surely be a growing database of pathologic mutations in IL1RN.

Metadata and annotations for multi-scale electrophysiological data.

The increasing use of high-frequency (kHz), long-duration (days) intracranial monitoring from multiple electrodes during pre-surgical evaluation for epilepsy produces large amounts of data that are challenging to store and maintain. Descriptive metadata and clinical annotations of these large data sets also pose challenges to simple, often manual, methods of data analysis. The problems of reliable communication of metadata and annotations between programs, the maintenance of the meanings within that information over long time periods, and the flexibility to re-sort data for analysis place differing demands on data structures and algorithms. Solutions to these individual problem domains (communication, storage and analysis) can be configured to provide easy translation and clarity across the domains. The Multi-scale Annotation Format (MAF) provides an integrated metadata and annotation environment that maximizes code reuse, minimizes error probability and encourages future changes by reducing the tendency to over-fit information technology solutions to current problems. An example of a graphical utility for generating and evaluating metadata and annotations for "big data" files is presented.