ASA Physical Status Classification, Preoperative Risk Assessment
Conditions
Keywords
Artificial Intelligence, Large Language Models, Preoperative Evaluation, ASA Classification
Brief summary
This prospective observational study aims to evaluate the performance of multiple artificial intelligence-based large language models in assigning American Society of Anesthesiologists Physical Status (ASA-PS) classifications in adult preoperative patients. AI-generated ASA scores obtained using both prompted and unprompted clinical scenario inputs will be compared with assessments performed by experienced anesthesiologists. The agreement, accuracy, readability, and overall quality of AI outputs will be analyzed to determine the potential role of artificial intelligence in supporting preoperative risk stratification.
Detailed description
The American Society of Anesthesiologists Physical Status (ASA-PS) classification is widely used for perioperative risk stratification but is subject to interobserver variability. Recent advances in artificial intelligence and large language models have introduced new opportunities for clinical decision support. This prospective observational study includes adult patients undergoing routine preoperative anesthesia evaluation at Bursa City Hospital. Demographic data, medical history, comorbidities, functional capacity, laboratory findings, electrocardiography, chest imaging results, and planned surgical procedures are recorded to construct standardized clinical scenarios. Multiple artificial intelligence models, including large language model-based systems, are provided with patient scenarios using both structured prompts and unstructured inputs. Each model assigns an ASA-PS classification and provides explanatory text. AI-generated classifications are compared with assessments performed independently by experienced anesthesiologists. Primary outcomes include agreement and accuracy between AI-generated and clinician-assigned ASA classifications using Cohen's Kappa statistics. Secondary outcomes include readability assessment using the Ateşman Turkish Readability Index and response quality evaluation using the Global Quality Scale. The study aims to explore whether artificial intelligence can improve standardization, objectivity, and efficiency in preoperative risk assessment while highlighting the strengths and limitations of current AI technologies in clinical anesthesia practice.
Interventions
None listed
Sponsors
Bursa City Hospital
Study design
Observational model
COHORT
Time perspective
PROSPECTIVE
Eligibility
Sex/Gender
ALL
Age
18 Years to No maximum
Healthy volunteers
Yes
Inclusion criteria
* Adult patients aged 18 years or older * Undergoing routine preoperative anesthesia evaluation * Classified as ASA Physical Status I-IV * Availability of complete clinical data required for AI assessment
Exclusion criteria
* Patients younger than 18 years * Refusal to participate * Incomplete or missing clinical information
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Agreement Between AI-Generated and Clinician-Assigned ASA Physical Status Classification | Preprocedural/Perioperative | Level of agreement between artificial intelligence models and anesthesiologists in assigning ASA Physical Status classification measured using Cohen's Kappa coefficient |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Accuracy of AI Models in ASA Classification | Preprocedural/Perioperative | Proportion of correct ASA Physical Status classifications generated by artificial intelligence models compared with anesthesiologist assessments |
| Readability of AI-Generated Clinical Responses | Preprocedural/Perioperative | Readability scores of artificial intelligence-generated clinical responses assessed using the Ateşman Turkish Readability Index (range: 0-100), where higher scores indicate better readability. |
Countries
Turkey (Türkiye)
Outcome results
None listed