Keyword: Data Accuracy
1 result found.
Original Article
Epidemiology and Health Data Insights, 2(5), 2026, ehdi059, https://doi.org/10.63946/ehdi/19672
ABSTRACT:
Background: Family-reported complaints at admission shape rehabilitation planning, but they are usually stored as free text and are rarely analyzed.
Methods: We analyzed the admission complaint text of 318 children admitted to a national pediatric rehabilitation center in one admission course (22 August–8 September 2025), written in Russian in discharge summaries. We coded the text into ten complaint categories with a rule-based dictionary, compared the coding with hand coding of 209 children by a single clinician (including a test set of 40 children not used to build the dictionary), compared the categories by diagnosis group with permutation tests, and explored complaint profiles with non-negative matrix factorization of lemmatized text.
Results: On the test set the dictionary agreed with the clinician’s hand coding on 98.3% of category decisions (95% CI 97.0 to 99.5%). Speech and language complaints were most frequent (73%), followed by motor (59%) and social and behavioral (36%) complaints. Motor complaints were recorded in 98% of children with cerebral palsy, and 61% of these children also had speech or language complaints. Categories differed by diagnosis (all q<0.05 except learning and memory), but motor complaints were also frequent in Down syndrome and other diagnoses. Three complaint profiles were found.
Conclusions: Family-reported complaints in routine records can be coded reproducibly, and speech and language concerns were frequent in children admitted for motor disorders. Agreement was measured against a single clinician and may overestimate agreement with other clinicians.
Methods: We analyzed the admission complaint text of 318 children admitted to a national pediatric rehabilitation center in one admission course (22 August–8 September 2025), written in Russian in discharge summaries. We coded the text into ten complaint categories with a rule-based dictionary, compared the coding with hand coding of 209 children by a single clinician (including a test set of 40 children not used to build the dictionary), compared the categories by diagnosis group with permutation tests, and explored complaint profiles with non-negative matrix factorization of lemmatized text.
Results: On the test set the dictionary agreed with the clinician’s hand coding on 98.3% of category decisions (95% CI 97.0 to 99.5%). Speech and language complaints were most frequent (73%), followed by motor (59%) and social and behavioral (36%) complaints. Motor complaints were recorded in 98% of children with cerebral palsy, and 61% of these children also had speech or language complaints. Categories differed by diagnosis (all q<0.05 except learning and memory), but motor complaints were also frequent in Down syndrome and other diagnoses. Three complaint profiles were found.
Conclusions: Family-reported complaints in routine records can be coded reproducibly, and speech and language concerns were frequent in children admitted for motor disorders. Agreement was measured against a single clinician and may overestimate agreement with other clinicians.
