In Silico Cognitive Decline Cohorts (Control, MCI, and Alzheimer's Disease) Generated with Large Language Models

This repository contains the code and synthetic data associated with a study investigating the use of large language models (LLMs) to generate in silico cohorts spanning healthy aging (Control), mild cognitive impairment (MCI), and Alzheimer’s disease (AD). The materials support experiments examining whether LLMs can reproduce clinically plausible cognitive decline profiles and whether simulated deficits reflect architectural constraints rather than superficial role-playing effects. The repository includes Python scripts used to generate and validate synthetic subjects through conversational prompting, administer a verbal neuropsychological assessment battery, and evaluate performance across multiple cognitive domains. It also contains the full set of synthetic subject outputs for two experiments: (1) graded cognitive profiling across clinical conditions, and (2) systematic manipulation of system prompt richness to dissociate structurally constrained memory deficits from manipulable generative language behaviors. All data are fully synthetic and generated in silico; no real patient data are included. The materials are intended to enable replication, methodological extension, and secondary analyses, and may serve as a testbed for digital biomarker development, clinical training, and research in computational neuroscience and neuropsychology.