Enriching representation learning using 53 million patient notes through human phenotype ontology embedding.

The Human Phenotype Ontology (HPO) is a dictionary of >15,000 clinical phenotypic terms with defined semantic relationships, developed to standardize phenotypic analysis. Over the last decade, the HPO has been used to accelerate the implementation of precision medicine into clinical practice. In addition, recent research in representation learning, specifically in graph embedding, has led to notable progress in automated prediction via learned features. Here, we present a novel approach to phenotype representation by incorporating phenotypic frequencies based on 53 million full-text health care notes from >1.5 million individuals. We demonstrate the efficacy of our proposed phenotype embedding technique by comparing our work to existing phenotypic similarity-measuring methods. Using phenotype frequencies in our embedding technique, we are able to identify phenotypic similarities that surpass current computational models. Furthermore, our embedding technique exhibits a high degree of agreement with domain experts' judgment. By transforming complex and multidimensional phenotypes from the HPO format into vectors, our proposed method enables efficient representation of these phenotypes for downstream tasks that require deep phenotyping. This is demonstrated in a patient similarity analysis and can further be applied to disease trajectory and risk prediction.

Child neurology telemedicine: Analyzing 14 820 patient encounters during the first year of the COVID-19 pandemic.

AIM: To determine the long-term impact of telemedicine in child neurology care during the COVID-19 pandemic and with the reopening of outpatient clinics. METHOD: We performed an observational cohort study of 34 837 in-person visits and 14 820 telemedicine outpatient visits across 26 399 individuals. We assessed differences in care across visit types, time-period observed, time between follow-ups, patient portal activation rates, and demographic factors. RESULTS: We observed a higher proportion of telemedicine for epilepsy (International Classification of Diseases, 10th Revision G40: odds ratio [OR] 1.4, 95% confidence interval [CI] 1.3-1.5) and a lower proportion for movement disorders (G25: OR 0.7, 95% CI 0.6-0.8; R25: OR 0.7, 95% CI 0.6-0.9) relative to in-person visits. Infants were more likely to be seen in-person after reopening clinics than by telemedicine (OR 1.6, 95% CI 1.5-1.8) as were individuals with neuromuscular disorders (OR 1.6, 95% CI 1.5-1.7). Self-reported racial and ethnic minority populations and those with highest social vulnerability had lower telemedicine participation rates (OR 0.8, 95% CI 0.8-0.8; OR 0.7, 95% CI 0.7-0.8). INTERPRETATION: Telemedicine continued to be utilized even once in-person clinics were available. Pediatric epilepsy care can often be performed using telemedicine while young patients with neuromuscular disorders often require in-person assessment. Prominent barriers for socially vulnerable families and racial and ethnic minorities persist.

Visits of concern in child neurology telemedicine.

AIM: To characterize child neurology telemedicine visits flagged as requiring in-person evaluation during the COVID-19 pandemic. METHOD: We analyzed 7130 audio-video telemedicine visits between March and November 2020. Visits of concern (VOCs) were defined as telemedicine visits where the clinical scenario necessitated in-person follow-up evaluation sooner than if the visit had been conducted in-person. RESULTS: VOCs occurred in 5% (333/7130) of visits for 292 individuals (148 females, 144 males). Providers noted technical challenges more often in VOCs (40%; 133/333) than visits without concern (non-VOCs) (28%; 1922/6797) (p < 0.05). The median age was younger in VOCs (9 years 3 months, interquartile range [IQR] 2 years 0 months-14 years 3 months) than non-VOCs (11 years 3 months, IQR 5 years 10 months-15 years 10 months) (p < 0.05). Median household income was lower for patients with VOCs ($74 K, IQR $55 K-$97 K) compared to non-VOCs ($80 K, IQR $61 K-$100 K) (p < 0.05). Compared with all other race categories, families who self-identified as Black were more likely to have a VOC (odds ratio 1.53, 95% confidence interval 1.21-2.06). Epilepsy and headache represented the highest percentages of VOCs, while neuromuscular disorders and developmental delay had a higher proportion of VOCs than other neurological disorders. INTERPRETATION: These findings suggest that telemedicine is an effective platform for most child neurology visits. Younger children and those with neuromuscular disorders or developmental delays are more likely to require in-person evaluation. WHAT THIS PAPER ADDS: It is possible to successfully flag patients who need in-person assessment. Providers can manage issues arising during telemedicine in 95% of visits. Visits flagged as concerning were likely unrelated to modality of patient care. Provider concern was independent of technical difficulties for most telehealth visits. Younger age may be correlated with need for in-person assessment.

Computational analysis of neurodevelopmental phenotypes: Harmonization empowers clinical discovery.

Making a specific diagnosis in neurodevelopmental disorders is traditionally based on recognizing clinical features of a distinct syndrome, which guides testing of its possible genetic etiologies. Scalable frameworks for genomic diagnostics, however, have struggled to integrate meaningful measurements of clinical phenotypic features. While standardization has enabled generation and interpretation of genomic data for clinical diagnostics at unprecedented scale, making the equivalent breakthrough for clinical data has proven challenging. However, increasingly clinical features are being recorded using controlled dictionaries with machine readable formats such as the Human Phenotype Ontology (HPO), which greatly facilitates their use in the diagnostic space. Improving the tractability of large-scale clinical information will present new opportunities to inform genomic research and diagnostics from a clinical perspective. Here, we describe novel approaches for computational phenotyping to harmonize clinical features, improve data translation through revising domain-specific dictionaries, quantify phenotypic features, and determine clinical relatedness. We demonstrate how these concepts can be applied to longitudinal phenotypic information, which represents a critical element of developmental disorders and pediatric conditions. Finally, we expand our discussion to clinical data derived from electronic medical records, a largely untapped resource of deep clinical information with distinct strengths and weaknesses.

Extracting seizure frequency from epilepsy clinic notes: a machine reading approach to natural language processing.

OBJECTIVE: Seizure frequency and seizure freedom are among the most important outcome measures for patients with epilepsy. In this study, we aimed to automatically extract this clinical information from unstructured text in clinical notes. If successful, this could improve clinical decision-making in epilepsy patients and allow for rapid, large-scale retrospective research. MATERIALS AND METHODS: We developed a finetuning pipeline for pretrained neural models to classify patients as being seizure-free and to extract text containing their seizure frequency and date of last seizure from clinical notes. We annotated 1000 notes for use as training and testing data and determined how well 3 pretrained neural models, BERT, RoBERTa, and Bio_ClinicalBERT, could identify and extract the desired information after finetuning. RESULTS: The finetuned models (BERTFT, Bio_ClinicalBERTFT, and RoBERTaFT) achieved near-human performance when classifying patients as seizure free, with BERTFT and Bio_ClinicalBERTFT achieving accuracy scores over 80%. All 3 models also achieved human performance when extracting seizure frequency and date of last seizure, with overall F1 scores over 0.80. The best combination of models was Bio_ClinicalBERTFT for classification, and RoBERTaFT for text extraction. Most of the gains in performance due to finetuning required roughly 70 annotated notes. DISCUSSION AND CONCLUSION: Our novel machine reading approach to extracting important clinical outcomes performed at or near human performance on several tasks. This approach opens new possibilities to support clinical practice and conduct large-scale retrospective clinical research. Future studies can use our finetuning pipeline with minimal training annotations to answer new clinical questions.

Phenotypic homogeneity in childhood epilepsies evolves in gene-specific patterns across 3251 patient-years of clinical data.

While genetic studies of epilepsies can be performed in thousands of individuals, phenotyping remains a manual, non-scalable task. A particular challenge is capturing the evolution of complex phenotypes with age. Here, we present a novel approach, applying phenotypic similarity analysis to a total of 3251 patient-years of longitudinal electronic medical record data from a previously reported cohort of 658 individuals with genetic epilepsies. After mapping clinical data to the Human Phenotype Ontology, we determined the phenotypic similarity of individuals sharing each genetic etiology within each 3-month age interval from birth up to a maximum age of 25 years. 140 of 600 (23%) of all 27 genes and 3-month age intervals with sufficient data for calculation of phenotypic similarity were significantly higher than expect by chance. 11 of 27 genetic etiologies had significant overall phenotypic similarity trajectories. These do not simply reflect strong statistical associations with single phenotypic features but appear to emerge from complex clinical constellations of features that may not be strongly associated individually. As an attempt to reconstruct the cognitive framework of syndrome recognition in clinical practice, longitudinal phenotypic similarity analysis extends the traditional phenotyping approach by utilizing data from electronic medical records at a scale that is far beyond the capabilities of manual phenotyping. Delineation of how the phenotypic homogeneity of genetic epilepsies varies with age could improve the phenotypic classification of these disorders, the accuracy of prognostic counseling, and by providing historical control data, the design and interpretation of precision clinical trials in rare diseases.

Modeling seizures in the Human Phenotype Ontology according to contemporary ILAE concepts makes big phenotypic data tractable.

OBJECTIVE: The clinical features of epilepsy determine how it is defined, which in turn guides management. Therefore, consideration of the fundamental clinical entities that comprise an epilepsy is essential in the study of causes, trajectories, and treatment responses. The Human Phenotype Ontology (HPO) is used widely in clinical and research genetics for concise communication and modeling of clinical features, allowing extracted data to be harmonized using logical inference. We sought to redesign the HPO seizure subontology to improve its consistency with current epileptological concepts, supporting the use of large clinical data sets in high-throughput clinical and research genomics. METHODS: We created a new HPO seizure subontology based on the 2017 International League Against Epilepsy (ILAE) Operational Classification of Seizure Types, and integrated concepts of status epilepticus, febrile, reflex, and neonatal seizures at different levels of detail. We compared the HPO seizure subontology prior to, and following, our revision, according to the information that could be inferred about the seizures of 791 individuals from three independent cohorts: 2 previously published and 150 newly recruited individuals. Each cohort's data were provided in a different format and harmonized using the two versions of the HPO. RESULTS: The new seizure subontology increased the number of descriptive concepts for seizures 5-fold. The number of seizure descriptors that could be annotated to the cohort increased by 40% and the total amount of information about individuals' seizures increased by 38%. The most important qualitative difference was the relationship of focal to bilateral tonic-clonic seizure to generalized-onset and focal-onset seizures. SIGNIFICANCE: We have generated a detailed contemporary conceptual map for harmonization of clinical seizure data, implemented in the official 2020-12-07 HPO release and freely available at hpo.jax.org. This will help to overcome the phenotypic bottleneck in genomics, facilitate reuse of valuable data, and ultimately improve diagnostics and precision treatment of the epilepsies.

Computational analysis of 10,860 phenotypic annotations in individuals with SCN2A-related disorders.

PURPOSE: Pathogenic variants in SCN2A cause a wide range of neurodevelopmental phenotypes. Reports of genotype-phenotype correlations are often anecdotal, and the available phenotypic data have not been systematically analyzed. METHODS: We extracted phenotypic information from primary descriptions of SCN2A-related disorders in the literature between 2001 and 2019, which we coded in Human Phenotype Ontology (HPO) terms. With higher-level phenotype terms inferred by the HPO structure, we assessed the frequencies of clinical features and investigated the association of these features with variant classes and locations within the NaV1.2 protein. RESULTS: We identified 413 unrelated individuals and derived a total of 10,860 HPO terms with 562 unique terms. Protein-truncating variants were associated with autism and behavioral abnormalities. Missense variants were associated with neonatal onset, epileptic spasms, and seizures, regardless of type. Phenotypic similarity was identified in 8/62 recurrent SCN2A variants. Three independent principal components accounted for 33% of the phenotypic variance, allowing for separation of gain-of-function versus loss-of-function variants with good performance. CONCLUSION: Our work shows that translating clinical features into a computable format using a standardized language allows for quantitative phenotype analysis, mapping the phenotypic landscape of SCN2A-related disorders in unprecedented detail and revealing genotype-phenotype correlations along a multidimensional spectrum.

The Human Phenotype Ontology in 2021.

The Human Phenotype Ontology (HPO, https://hpo.jax.org) was launched in 2008 to provide a comprehensive logical standard to describe and computationally analyze phenotypic abnormalities found in human disease. The HPO is now a worldwide standard for phenotype exchange. The HPO has grown steadily since its inception due to considerable contributions from clinical experts and researchers from a diverse range of disciplines. Here, we present recent major extensions of the HPO for neurology, nephrology, immunology, pulmonology, newborn screening, and other areas. For example, the seizure subontology now reflects the International League Against Epilepsy (ILAE) guidelines and these enhancements have already shown clinical validity. We present new efforts to harmonize computational definitions of phenotypic abnormalities across the HPO and multiple phenotype ontologies used for animal models of disease. These efforts will benefit software such as Exomiser by improving the accuracy and scope of cross-species phenotype matching. The computational modeling strategy used by the HPO to define disease entities and phenotypic features and distinguish between them is explained in detail.We also report on recent efforts to translate the HPO into indigenous languages. Finally, we summarize recent advances in the use of HPO in electronic health record systems.

Correction: A longitudinal footprint of genetic epilepsies using automated electronic medical record interpretation.

An amendment to this paper has been published and can be accessed via a link at the top of the paper.

Semantic Similarity Analysis Reveals Robust Gene-Disease Relationships in Developmental and Epileptic Encephalopathies.

More than 100 genetic etiologies have been identified in developmental and epileptic encephalopathies (DEEs), but correlating genetic findings with clinical features at scale has remained a hurdle because of a lack of frameworks for analyzing heterogenous clinical data. Here, we analyzed 31,742 Human Phenotype Ontology (HPO) terms in 846 individuals with existing whole-exome trio data and assessed associated clinical features and phenotypic relatedness by using HPO-based semantic similarity analysis for individuals with de novo variants in the same gene. Gene-specific phenotypic signatures included associations of SCN1A with "complex febrile seizures" (HP: 0011172; p = 2.1 × 10-5) and "focal clonic seizures" (HP: 0002266; p = 8.9 × 10-6), STXBP1 with "absent speech" (HP: 0001344; p = 1.3 × 10-11), and SLC6A1 with "EEG with generalized slow activity" (HP: 0010845; p = 0.018). Of 41 genes with de novo variants in two or more individuals, 11 genes showed significant phenotypic similarity, including SCN1A (n = 16, p < 0.0001), STXBP1 (n = 14, p = 0.0021), and KCNB1 (n = 6, p = 0.011). Including genetic and phenotypic data of control subjects increased phenotypic similarity for all genetic etiologies, whereas the probability of observing de novo variants decreased, emphasizing the conceptual differences between semantic similarity analysis and approaches based on the expected number of de novo events. We demonstrate that HPO-based phenotype analysis captures unique profiles for distinct genetic etiologies, reflecting the breadth of the phenotypic spectrum in genetic epilepsies. Semantic similarity can be used to generate statistical evidence for disease causation analogous to the traditional approach of primarily defining disease entities through similar clinical features.

A longitudinal footprint of genetic epilepsies using automated electronic medical record interpretation.

PURPOSE: Childhood epilepsies have a strong genetic contribution, but the disease trajectory for many genetic etiologies remains unknown. Electronic medical record (EMR) data potentially allow for the analysis of longitudinal clinical information but this has not yet been explored. METHODS: We analyzed provider-entered neurological diagnoses made at 62,104 patient encounters from 658 individuals with known or presumed genetic epilepsies. To harmonize clinical terminology, we mapped clinical descriptors to Human Phenotype Ontology (HPO) terms and inferred higher-level phenotypic concepts. We then binned the resulting 286,085 HPO terms to 100 3-month time intervals and assessed gene-phenotype associations at each interval. RESULTS: We analyzed a median follow-up of 6.9 years per patient and a cumulative 3251 patient years. Correcting for multiple testing, we identified significant associations between "Status epilepticus" with SCN1A at 1.0 years, "Severe intellectual disability" with PURA at 9.75 years, and "Infantile spasms" and "Epileptic spasms" with STXBP1 at 0.5 years. The identified associations reflect known clinical features of these conditions, and manual chart review excluded provider bias. CONCLUSION: Some aspects of the longitudinal disease histories can be reconstructed through EMR data and reveal significant gene-phenotype associations, even within closely related conditions. Gene-specific EMR footprints may enable outcome studies and clinical decision support.

Whole-exome and HLA sequencing in Febrile infection-related epilepsy syndrome.

Febrile infection-related epilepsy syndrome (FIRES) is a devastating epilepsy characterized by new-onset refractory status epilepticus with a prior febrile infection. We performed exome sequencing in 50 individuals with FIRES, including 27 patient-parent trios and 23 single probands, none of whom had pathogenic variants in established genes for epilepsies or neurodevelopmental disorders. We also performed HLA sequencing in 29 individuals with FIRES and 529 controls, which failed to identify prominent HLA alleles. The genetic architecture of FIRES is substantially different from other developmental and epileptic encephalopathies, and the underlying etiology remains elusive, requiring novel approaches to identify the underlying causative factors.

Analyzing 2,589 child neurology telehealth encounters necessitated by the COVID-19 pandemic.

OBJECTIVE: To assess the rapid implementation of child neurology telehealth outpatient care with the onset of the coronavirus disease 2019 (COVID-19) pandemic in March 2020. METHODS: This was a cohort study with retrospective comparison of 14,780 in-person encounters and 2,589 telehealth encounters, including 2,093 audio-video telemedicine and 496 scheduled telephone encounters, between October 1, 2019 and April 24, 2020. We compared in-person and telehealth encounters for patient demographics and diagnoses. For audio-video telemedicine encounters, we analyzed questionnaire responses addressing provider experience, follow-up plans, technical quality, need for in-person assessment, and parent/caregiver satisfaction. We performed manual reviews of encounters flagged as concerning by providers. RESULTS: There were no differences in patient age and major ICD-10 codes before and after transition. Clinicians considered telemedicine satisfactory in 93% (1,200 of 1,286) of encounters and suggested telemedicine as a component for follow-up care in 89% (1,144 of 1,286) of encounters. Technical challenges were reported in 40% (519 of 1,314) of encounters. In-person assessment was considered warranted after 5% (65 of 1,285) of encounters. Patients/caregivers indicated interest in telemedicine for future care in 86% (187 of 217) of encounters. Participation in telemedicine encounters compared to telephone encounters was less frequent among patients in racial or ethnic minority groups. CONCLUSIONS: We effectively converted most of our outpatient care to telehealth encounters, including mostly audio-video telemedicine encounters. Providers rated the vast majority of telemedicine encounters to be satisfactory, and only a small proportion of encounters required short-term in-person follow-up. These findings suggest that telemedicine is feasible and effective for a large proportion of child neurology care. Additional strategies are needed to ensure equitable telemedicine use.

The Associations Between Pain-related Beliefs, Pain Intensity, and Patient Functioning: Hypnotizability as a Moderator.

OBJECTIVES: On the basis of the idea that thoughts held about pain may represent "self-suggestions" and evidence indicating that people with higher levels of trait hypnotizability are more responsive to suggestions, the current study evaluated hypothesized moderating effects of hypnotizability on the associations between pain-related thoughts and both pain intensity and pain interference. METHODS: Eighty-five individuals with chronic pain were given measures of hypnotizability, pain intensity, pain interference, and pain-related thoughts (control beliefs, catastrophizing). RESULTS: Analyses supported a moderating role of hypnotizability on the association between control beliefs and pain interference. Specifically, the negative association between pain control beliefs and pain interference were stronger among those with higher trait hypnotizability than between those with lower trait hypnotizability. DISCUSSION: The study findings, if replicated in additional samples of individuals with chronic pain, have important clinical and theoretical implications. For example, if trait hypnotizability is found to predict an individual's response to a particular technique of cognitive therapy-such as focusing on and repeating pain control belief self-statements-measures of hypnotizability could be used to identify individuals who might be most responsive to this technique. The current findings indicate that research to further examine this possibility is warranted.