ChatGPT-4 vs Physician Hypertension Treatment Recommendations from a Single-Centre Study

A retrospective comparison of 100 treatment-naïve hypertensive patients assesses the concordance between ChatGPT-4's antihypertensive recommendations and real-world physician prescriptions. The study, authored by Kadri Murat Gürses and posted on figshare in 2026, found substantial agreement (κ = 0.67) and an association between concordant treatment and higher rates of short-term blood pressure control.

Use Cases

• Benchmarking AI-generated treatment recommendations against real-world physician decisions based on clinical case data.
• Studying the association between guideline-adherent therapy and short-term blood pressure control outcomes.
• Analyzing the distribution of therapy classes (e.g., ACEi + thiazide) recommended by AI versus human prescribers.

Strengths

• Dataset is derived from a structured clinical study with 100 patient cases.
• Primary outcome (concordance) and secondary outcomes (BP control, therapy distribution) are clearly defined with statistical results (κ = 0.67, OR = 3.8).
• Data generation followed a standardized prompt based on ESC/ESH and ACC/AHA guidelines.

Limitations

• The dataset is very small at 29.7 KB, indicating limited scope and sample size.
• Column-level documentation is absent; field semantics must be inferred after download.
• Data is from a single-centre, retrospective study, which may limit generalizability.

Provenance

Source

figshare

Collection Method

Retrospective entry of anonymized clinical data from 100 newly diagnosed hypertensive patients into ChatGPT-4 using a standardized prompt.

Freshness

Last updated 2026-05-12 13:09:29; freshness should be verified.

Geography

Single-centre study; specific location unknown.