Transparent deep learning to identify autism spectrum disorders (ASD) in EHR using clinical notes.

OBJECTIVE: Machine learning (ML) is increasingly employed to diagnose medical conditions, with algorithms trained to assign a single label using a black-box approach. We created an ML approach using deep learning that generates outcomes that are transparent and in line with clinical, diagnostic rules. We demonstrate our approach for autism spectrum disorders (ASD), a neurodevelopmental condition with increasing prevalence. METHODS: We use unstructured data from the Centers for Disease Control and Prevention (CDC) surveillance records labeled by a CDC-trained clinician with ASD A1-3 and B1-4 criterion labels per sentence and with ASD cases labels per record using Diagnostic and Statistical Manual of Mental Disorders (DSM5) rules. One rule-based and three deep ML algorithms and six ensembles were compared and evaluated using a test set with 6773 sentences (N = 35 cases) set aside in advance. Criterion and case labeling were evaluated for each ML algorithm and ensemble. Case labeling outcomes were compared also with seven traditional tests. RESULTS: Performance for criterion labeling was highest for the hybrid BiLSTM ML model. The best case labeling was achieved by an ensemble of two BiLSTM ML models using a majority vote. It achieved 100% precision (or PPV), 83% recall (or sensitivity), 100% specificity, 91% accuracy, and 0.91 F-measure. A comparison with existing diagnostic tests shows that our best ensemble was more accurate overall. CONCLUSIONS: Transparent ML is achievable even with small datasets. By focusing on intermediate steps, deep ML can provide transparent decisions. By leveraging data redundancies, ML errors at the intermediate level have a low impact on final outcomes.

Multi-center implementation of rapid whole genome sequencing provides additional evidence of its utility in the pediatric inpatient setting.

Objective: Multi-center implementation of rapid whole genome sequencing with assessment of the clinical utility of rapid whole genome sequencing (rWGS), including positive, negative and uncertain results, in admitted infants with a suspected genetic disease. Study design: rWGS tests were ordered at eight hospitals between November 2017 and April 2020. Investigators completed a survey of demographic data, Human Phenotype Ontology (HPO) terms, test results and impacts of results on clinical care. Results: A total of 188 patients, on general hospital floors and intensive care unit (ICU) settings, underwent rWGS testing. Racial and ethnic characteristics of the tested infants were broadly representative of births in the country at large. 35% of infants received a diagnostic result in a median of 6 days. The most common HPO terms for tested infants indicated an abnormality of the nervous system, followed by the cardiovascular system, the digestive system, the respiratory system and the head and neck. Providers indicated a major change in clinical management because of rWGS for 32% of infants tested overall and 70% of those with a diagnostic result. Also, 7% of infants with a negative rWGS result and 23% with a variant of unknown significance (VUS) had a major change in management due to testing. Conclusions: Our study demonstrates that the implementation of rWGS is feasible across diverse institutions, and provides additional evidence to support the clinical utility of rWGS in a demographically representative sample of admitted infants and includes assessment of the clinical impact of uncertain rWGS results in addition to both positive and negative results.

The International SCN8A Patient Registry: A Scientific Resource to Advance the Understanding and Treatment of a Rare Pediatric Neurodevelopmental Syndrome.

Genetic variants in the SCN8A gene underlie a wide spectrum of neurodevelopmental phenotypes that range from severe epileptic encephalopathy to benign familial infantile epilepsy to neurodevelopmental delays with or without seizures. A host of additional comorbidities also contribute to the phenotypic spectrum. As a result of the recent identification of the genetic etiology and the length of time it often takes to diagnose patients, little data are available on the natural history of these conditions. The International SCN8A Patient Registry was developed in 2015 to fill gaps in understanding the spectrum of the disease and its natural history, as well as the lived experiences of individuals with SCN8A syndrome. Another goal of the registry is to collect longitudinal data from participants on a regular basis. In this article, we describe the construction and structure of the International SCN8A Patient Registry, present the type of information available, and highlight particular analyses that demonstrate how registry data can provide insights into the clinical management of SCN8A syndrome.

DMD Gene and Dystrophinopathy Phenotypes Associated With Mutations: A Systematic Review for Clinicians.

ABSTRACT: The diagnosis of Duchenne and Becker muscular dystrophy (DBMD) is made by genetic testing in approximately 95% of cases. Although specific mutations can be associated with skeletal muscle phenotype, pulmonary and cardiac comorbidities (leading causes of death in Duchenne) have not been associated with Duchenne muscular dystrophy mutation type or location and vary within families. Therefore, identifying predictors for phenotype severity beyond frameshift prediction is important clinically. We performed a systematic review assessing research related to genotype-phenotype correlations in DBMD. While there are severity differences across the spectrum and within mild and severe forms of DBMD, few protective or exacerbating mutations within the dystrophin gene were reported. Except for intellectual disability, clinical test results reporting genotypic information are insufficient for clinical prediction of severity and comorbidities and the predictive validity is too low to be useful when advising families. Including expanded information coupled with proposed severity predictions in clinical genetic reports for DBMD is critical for improving anticipatory guidance.

Postfolate spina bifida lesion level change.

BACKGROUND: Spina bifida accounts for a large proportion of birth defects in the United States. Studies have evaluated the decrease in prevalence at birth after folate fortification of food grains, but little is known about neurologic functional changes related to fortification. This study assesses the functional level of lesions in the prefortification and postfortification eras. METHODS: Data were collected through retrospective review of medical records from a regional multispecialty clinic in Arizona. This study included individuals born between 1981-1995 (prefortification) and 1999-2013 (postfortification). Patients were included if they had a primary diagnosis of spina bifida with or without hydrocephalus. RESULTS: There was a significant difference in functional lesion level with an 85% reduction in thoracic level lesions in the postfortification era (p < .005). There were no differences in gender or ethnicity across eras; however, Hispanic ethnicity had a higher number of cases overall (51.7%). The most common lesion level in both eras was mid-lumbar, accounting for 35.7 and 34.4% of cases in the prefolate and postfolate eras, respectively. CONCLUSIONS: This study demonstrates a significant difference in the distribution of lesion level of spina bifida patients born in the postfortification era, based on neurologic function. Further research with a larger sample size is needed to determine if this observation holds true nationally.

Optimizing Corpus Creation for Training Word Embedding in Low Resource Domains: A Case Study in Autism Spectrum Disorder (ASD).

Automating the extraction of behavioral criteria indicative of Autism Spectrum Disorder (ASD) in electronic health records (EHRs) can contribute significantly to the effort to monitor the condition. Word embedding algorithms such as Word2Vec can encode semantic meanings of words in vectors and assist in automated vocabulary discovery from EHRs. However, text available for training word embeddings for ASD is miniscule compared to the billions of tokens typically used. We evaluate the importance of corpus specificity versus size and hypothesize that for specific domains small corpora can generate excellent word embeddings. We custom-built 6 ASD-themed corpora (N=4482), using ASD EHRs and abstracts from PubMed (N=39K) and PsychInfo (N=69K) and evaluated them. We were able to generate the most useful 200-dimension embeddings based on the small ASD EHR data. Due to diversity in its vocabulary, the abstract-based embeddings generated fewer related terms and saw minimal improvement when the size of the corpus increased.