JAMIA open · 2026
Using clinical notes to identify children with speech-language delay and understand differences in diagnostic timing.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Shu Jiang, Fannin Danai Kasambira, Dawson Geraldine, Maslow Gary, Engelhard Matthew M, Franz Lauren, Goldstein Benjamin A
Аннотация
OBJECTIVE: Speech-language delay (SLD) is a developmental condition often identified by pediatricians. While early intervention is recommended, it is common for pediatricians to take a "watch and wait" approach. We sought to assess whether there is discordance between documented concerns and diagnostic coding of SLD and whether discordances differ based on sociodemographic, clinical, and service utilization characteristics.
MATERIALS AND METHODS: We generated an age, sex, payer and race/ethnicity matched cohort of children with and without coded SLD to train a BioClinicalBERT natural learning processing (NLP) model to identify SLD in clinical notes. We applied the model to a population-based test set of well-child visits with no prior SLD diagnosis. We analyzed factors for encounters where SLD was mentioned, but there was no associated ICD-10 code present.
RESULTS: The model attained AUCs of 0.98 (internal validation) and 0.99 (population-based test set). Among encounters where notes documented concerns, only 52% had an associated ICD-10 code. Among the remainder, 39% subsequently received a code (follow-up period up to 650 days), indicating at least 26% of SLD encounters have delayed diagnostic coding. Younger and privately-insured children were less likely to have an ICD-10 code when SLD was documented in notes. Younger children with more outpatient visits were more likely to receive a future SLD diagnosis.
DISCUSSION: NLP can effectively recognize SLD concern in clinical notes, helping to capture those with early documented concerns. Gaps between documentation and coding could suggest "watch and wait" approaches.
CONCLUSION: Real world SLD research should consider potential discordance between documented concerns and diagnostic codes.
MATERIALS AND METHODS: We generated an age, sex, payer and race/ethnicity matched cohort of children with and without coded SLD to train a BioClinicalBERT natural learning processing (NLP) model to identify SLD in clinical notes. We applied the model to a population-based test set of well-child visits with no prior SLD diagnosis. We analyzed factors for encounters where SLD was mentioned, but there was no associated ICD-10 code present.
RESULTS: The model attained AUCs of 0.98 (internal validation) and 0.99 (population-based test set). Among encounters where notes documented concerns, only 52% had an associated ICD-10 code. Among the remainder, 39% subsequently received a code (follow-up period up to 650 days), indicating at least 26% of SLD encounters have delayed diagnostic coding. Younger and privately-insured children were less likely to have an ICD-10 code when SLD was documented in notes. Younger children with more outpatient visits were more likely to receive a future SLD diagnosis.
DISCUSSION: NLP can effectively recognize SLD concern in clinical notes, helping to capture those with early documented concerns. Gaps between documentation and coding could suggest "watch and wait" approaches.
CONCLUSION: Real world SLD research should consider potential discordance between documented concerns and diagnostic codes.