(BSPOT39) Building Patient Journeys for Prader-Willi Syndrome Patients: Insights from Electronic Health Records Through Natural Language Processing

Background: Prader-Willi Syndrome (PWS) manifests heterogeneously, complicating the understanding of the disease’s trajectory using electronic health record (EHR) data, particularly when critical information is embedded within unstructured clinical notes. Extracting such data could greatly enhance our understanding of disease progression.

Objectives: To comprehensively extract information on phenotypes and comorbidities associated with PWS through natural language processing (NLP), assess their prevalence, and analyze their longitudinal patterns.

Methods: We utilized EHRs from PWS patients seeking care at the University of Florida Health, spanning 2000 to 2022. Individuals with three or more clinical notes confirming PWS diagnoses were included. Clinical variables were extracted from both structured data and unstructured clinical notes using NLP, covering physiological and behavioral outcomes and comorbidities. We analyzed the percentage of patients exhibiting specific symptoms or behavioral outcomes across various age groups and tracked longitudinal changes in their status. Additionally, we examined sex disparities in each comorbidity.

Results: We identified 433 pediatric patients with PWS, including 224 males (51.7%), all of whom had varied initiation times of growth hormone treatment. We calculated the percentage of patients of various ages exhibiting one of 15 different phenotypes as recorded in their clinical notes. For instance, excoriation (all skin-picking behaviors) was documented in 143 patients (33.3%) between ages 2-14 years, with respective frequencies of 3.1%, 5.9%, 11.7%, 15.6%, 22.1%, 22.8%, 18.4%, 25.3%, 28.7%, 21.8%, 28.1%, 19.5%, and 20.0%. In terms of comorbidities, scoliosis (n=164, 37.9%), diagnosed across various age groups, emerged as the most frequently reported comorbidity, followed by obstructive sleep apnea (n=144, 33.3%) and hypopituitarism (n=117, 27.0%). While sex disparities were generally absent in most top-ranked comorbidities, central sleep apnea (n=42, 9.7%) demonstrated a higher prevalence in females (29) than in males (13, p=0.0011); unlike obstructive sleep apnea, which exhibited a comparable distribution between males (78) and females (66, p=0.1949). Precocious puberty (n=38, 8.8%) and congenital deformities of the hip (n=36, 8.3%) were more frequently reported in females (n=29, p < 0.0001 and n=26, p=0.0004, respectively), while convulsions (n=29, 6.7%) were more frequently reported in males (n=20, p=0.0086).

Conclusions: Our real-world study unveils the intricate longitudinal trajectory of diverse behavioral phenotypes and the timing of comorbidity onset with pediatric PWS, with noticeable sex disparities in certain comorbidities.