ChatGPT-estimated occupational complexity predicts cognitive outcomes and cortical thickness above and beyond socioeconomic status among older adults

Abstract

Many aging cohort studies have collected data on participants’ job titles, yet these job titles were seldom analyzed within the cognitive aging context despite their relevance to neurocognition, due to difficulties in analyzing these job titles quantitatively. While it is possible to rate these jobs’ occupational complexity (OC) using job classification systems, this can be somewhat labor-intensive and prone to human errors. To this end, we demonstrate a novel and simple method to extract OC ratings from job titles using ChatGPT. Then, we showcased the utility of these ratings in predicting cognitive and structural brain outcomes, especially compared to other socioeconomic status (SES) indicators. Community-dwelling older adults (N = 238, age_mean = 70) completed cognitive assessments and underwent MRI scans. Regression models were fitted to predict 14 different cognitive outcomes, vertex-wise cortical thickness (CT), and subcortical gray matter volumes, using OC scores and/or SES predictors (e.g., education, housing type, and income levels), controlling for demographical covariates. OC scores outperformed SES indicators in predicting clusters of CT increases and most cognitive outcomes, including diagnoses of mild cognitive impairment. Furthermore, OC scores significantly predicted clusters of CT increases and various cognitive outcomes, even after controlling for SES. Meta-analytic decoding suggests these clusters of CT increases occurred in regions typically associated with sensorimotor and memory processing. These results highlight the significant and unique contribution of ChatGPT-derived OC scores in predicting cognitive and brain aging outcomes. These scores are easy to derive and can be helpful in fine-tuning predictions of cognitive and brain aging outcomes.