Observational Study on AI Accuracy in Diagnosing and Treating Failed or Painful Hip Arthroplasty (PAINGPT)

Istituto Ortopedico Rizzoli

Status

Enrolling

Conditions

Total Hip Arthroplasty (THA)

Treatments

Other: Specializing Resident (4th year) Assessment

Other: Arthroplasty Fellow Assessment

Other: Junior Resident (3rd year) Assessment

Other: GPT-4 Assessment

Study type

Observational

Funder types

Other

Identifiers

NCT07012577

203/2025/Oss/IOR

Details and patient eligibility

About

Primary Goal:

This study aims to evaluate the diagnostic and therapeutic accuracy of GPT-4 (an advanced AI language model) compared to three orthopedic surgeons with varying experience levels in cases of failed or painful total hip arthroplasty.

Key Research Questions:

Diagnostic Accuracy:

Does GPT-4 provide correct, partially correct, or incorrect diagnoses compared to human orthopaedic surgeons?

Diagnostic Completeness:

Are GPT-4's diagnostic suggestions complete, partially complete, or incomplete compared to those of orthopedic surgeons?

Treatment Accuracy:

Does GPT-4 recommend correct, partially correct, or incorrect treatments for failed hip arthroplasty?

Treatment Completeness:

Are GPT-4's treatment recommendations fully comprehensive, partially complete, or incomplete compared to those of orthopaedic surgeon?

Study Design:

Participants:

20 anonymized patient cases (ages 18-80) with failed or painful hip arthroplasties, treated at IRCCS Istituto Ortopedico Rizzoli (Bologna, Italy) between 2004-2024.

Cases were selected based on clear diagnostic and treatment records (no ambiguous or incomplete data).

Comparison Groups:

GPT-4 (via ChatGPT interface)

Three orthopedic doctors (with different experience levels: resident, specialist, senior surgeon)

Method:

Each case (clinical summary + X-ray image) is presented to GPT-4 and the three doctors.

They must provide a diagnosis and treatment recommendations.

Two independent evaluators (principal investigator + department head) blindly assess responses for correctness and completeness using a 3-point scale (0=wrong/incomplete, 2=correct/complete).

Statistical analysis compares GPT-4 vs. human performance.

Expected Outcomes:

Determine if AI can match or outperform doctors in diagnosing and treating hip arthroplasty failures.

Assess whether GPT-4 could serve as a supplementary tool in orthopedic decision-making.

Ethical & Privacy Considerations:

No real-time patient data is used-only anonymized past cases.

No personal/sensitive data is shared with OpenAI (GPT-4 is used via a standard web interface).

Study complies with GDPR, HIPAA, and ethical AI guidelines.

Timeline:

Study duration: ~8 months (from ethics approval to final analysis).

Results will be published regardless of outcome.

Why This Study Matters:

First study evaluating GPT-4's role in complex orthopedic diagnostics.

Could influence future AI-assisted clinical decision-making in joint replacement surgeries.

Enrollment

20 estimated patients

Sex

All

Ages

18 to 80 years old

Volunteers

No Healthy Volunteers

Inclusion criteria

Adults (≥18 and ≤80 years old).
Documented painful or failed total hip arthroplasty requiring clinical/radiological evaluation (2004-2024).
Complete pre-operative clinical history, imaging (X-ray/tomography), and surgical reports.
Clear diagnosis of failure mode (e.g., aseptic loosening, infection, fracture, wear).
Treatment and outcomes fully documented in the institutional database.
"Exemplary" cases with minimal diagnostic ambiguity (per Engh/MusculoSkleletal Infection Society criteria, etc.).

Exclusion criteria

total hip arthroplasty with no documented failure/pain (well-functioning implants).
Incomplete clinical/radiological records (e.g., missing pre-operative imaging or surgical notes).
Complex/multifactorial failures (e.g., concurrent infection + loosening + fracture).
Radiographs/images non-interpretable (poor quality, missing views).
Cases with conflicting diagnoses/treatments in original records.