ChatGPT artificial intelligence in clinical data analysis: an example comparing standard vs fusion prostate biopsy...

To compare statistical outputs from ChatGPT 4.0 and human experts in both comparative and correlation analyses in the evaluation of multiparametric MRI/ultrasound fusion-targeted biopsy plus random biopsy versus standard random biopsy alone, in terms of upstaging.

Authors performed a retrospective evaluation on 101 patients undergoing robot-assisted radical prostatectomy (RaRP) between 2021 and 2023. Patients were divided in two groups, according to the type of prostatic biopsy received: combined fusion (MRI/US) targeted and random biopsy versus standard random biopsy. Clinical and histological data were anonymized and analyzed using logistic regression models, ANOVA, and Chi-square tests. Analysis generated by ChatGPT and by an experienced human statistician were compared. The Q-EVAL and Q-EVA tools were used to assess the quality of user-formulated questions and AI-generated answers, respectively.

Results revealed high concordance between statistical outputs generated by AI and expert human statistician with perfect concordance using Cohen's kappa coefficient (κ = 1.0). Logistic regression analysis demonstrated that fusion biopsy was associated with a reduced likelihood of upstaging, a consistent finding across statistical evaluations. Additionally, user interaction assessments indicated high-quality in question formulation.

ChatGPT (version 4.0) proved reliable for statistical analysis, showing strong concordance with human statisticians (κ = 1.0) in performing logistic regression, chi-square, and ANOVA tests. The Q-EVAL tool could reduce query errors, though ChatGPT's lack of automatic citations remains a limitation. Fusion biopsy significantly lowered upstaging risk after RaRP. In conclusion, ChatGPT is a valuable assistive tool but further research is required to optimize human-AI collaboration in clinical research.