Medical and social aspects of vocational guidance for persons with disabilities: assessing types of work activity using large language models

BACKGROUND: Vocational guidance for persons with disabilities requires comparable and reproducible criteria that allow correlating persistent impairments of body functions with the requirements of specific types of work activity and factors of the working environment.

AIM: To form and validate a database of assessment indicators for types of work activity as a tool for their ranking within the framework of vocational guidance.

METHODS: A total of 435 types of work activity from the initial groups of the All-Russian Classifier of Occupations were analyzed. For each of them, a profile was formed based on 21 medico-social indicators, grouped into blocks of safety, ability to perform work functions, and working conditions. Primary assessments were obtained using several large language models on a unified ten-point scale; post-processing was performed for unified interpretation (“the higher, the more suitable”) and calculation of integral values by averaging inter-model assessments. External validation was conducted on independent expert survey data (87 questionnaires): in each question, five professions were compared on a 0–100 scale; a total of 9,135 assessments were processed. Comparisons were made after intra-question min-max normalization; the consistency of ranking (Spearman coefficient, Kendall coefficient), the proportion of concordant pairs, and the normalized rank difference were assessed.

RESULTS: The proportion of informative comparison blocks was 92.3%, and constant expert assessments were identified in 7.6% of cases. The consistency of the model ranking with experts was moderate: average Spearman coefficient ρ = 0.416, average Kendall coefficient τ-b = 0.363, proportion of concordant pairs 0.653, average normalized rank difference 0.278. The highest validity was demonstrated by indicators related to impairments of neuromuscular, skeletal, and movement-related (statodynamic) functions (necessity of using a wheelchair), while the indicator of emotional tension showed minimal consistency.

CONCLUSION: The formed assessment database is applicable for profiling and preliminary selection of types of work activity in the vocational guidance of persons with disabilities. For contextual factors of working conditions, clarification of operational definitions and expert calibration are required.