Prenatal and childhood exposure to smoking and retinal nerve fibre layer thickness: A meta-analysis of three independent birth cohorts.

Measurement of patient functional outcomes during hospitalization is crucial to understanding the impact of illnesses and healthcare interventions. Outcome scoring systems provide a quantitative mechanism for evaluating functional status during and after hospitalization. Within pediatrics, the functional status scale (FSS)() has emerged as a validated, reliable metric for quantitative assessment of functioning among hospitalized children. FSS is comprised of six domains of functioning: mental status, sensory functioning, communication, motor functioning, feeding, and respiratory status. Each domain is scored from normal (a score of 1) to severe dysfunction (5) with total scores ranging from 6-30. The FSS has been successfully applied to a variety of pediatric populations including those experiencing trauma(-) and acute respiratory failure() and has been translated and tested in multiple languages and countries(, ). Though the tool has many advantages over prior outcome scoring systems, score determination requires manual data collection (performed by a nurse, respiratory therapist, or physician in the original study)() and often involves interviewing the child’s parent/caregiver(). If automation of FSS scores were possible, such a tool could vastly increase the number of patients for whom a quantitative outcome score is available, accelerating outcomes research among hospitalized children. Recent studies have suggested that FSS scores can be calculated through chart review alone(, ). However, given the complex information being analyzed, manual chart review and FSS score determination are still required. Large language models (LLMs) can rapidly summarize and analyze large volumes of free text. However, because of appropriate HIPAA protections, clinical note text is not publicly available and thus not present within the training data of most LLMs. A fine-tuned model inclusive of clinical note text and outcome labels, however, could overcome this barrier. As such, we hypothesized a fine-tuned LLM, designed specifically to estimate FSS scores from patient notes, could accurately determine FSS scores as compared to manually determined scores assigned through manual data collection and caregiver interview.