Profile-associated financial and access-related framing in LLM-generated pediatric asthma referral plans: a factorial audit of seven large language models - Scorecard - MDSpire
Advertisement
Profile-associated financial and access-related framing in LLM-generated pediatric asthma referral plans: a factorial audit of seven large language models
Clinical Scorecard: Evaluation of Financial and Access Framing in Pediatric Asthma Referral Plans Generated by Large Language Models: A Factorial Analysis of Seven Models
At a Glance
Category
Detail
Condition
Pediatric Asthma
Key Mechanisms
Large language models (LLMs) generate referral plans with varying financial-access and geographic-access framing.
Target Population
Children with moderate persistent asthma.
Care Setting
Clinical decision support and referral planning.
Key Highlights
LLM-generated referral plans varied significantly in financial-access and geographic-access language.
The name 'DeShawn' was associated with a higher financial-access term rate.
Geographic-access signals showed a strong association with referral plan content.
Human validation indicated moderate reliability in endpoint-specific assessments.
The study emphasizes the need for structural competence in LLM outputs.
Guideline-Based Recommendations
Diagnosis
Assess asthma control and environmental triggers.
Management
Implement appropriate anti-inflammatory therapy and specialist involvement as needed.
Monitoring & Follow-up
Evaluate the effectiveness of referral plans in addressing social determinants of health.
Risks
Potential for algorithmic bias in LLM-generated recommendations.
Patient & Prescribing Data
Children aged 5 years with moderate persistent asthma.
Referral plans should include actionable and contextualized support for treatment.
Clinical Best Practices
Ensure referral plans recognize real-world constraints without reducing patients to those constraints.
Evaluate LLM outputs for both biomedical accuracy and contextual framing.
