Conference Coverage

Autonomous AI Replies Rated Acceptable in 97% of Messages to Knee or Hip Arthroplasty Patients

Edited by:
Brian McCurdy

Key Highlights

  • Ten arthroplasty surgeons evaluated AI responses delivered to their own patients without human involvement.
  • Responses received a mean rating of 4.54 out of 5, with surgeon-specific acceptability rates ranging from 91% to 100%.
  • Missing or outdated surgeon preferences accounted for 25 of 31 unacceptable responses; 2 involved medication safety concerns.

In a prospective cohort study of total joint arthroplasty practices, surgeons rated autonomous, surgeon-specific artificial intelligence (AI) responses acceptable in 96.9% of patient messaging exchanges. The abstract was presented at the American Association of Hip and Knee Surgeons Annual Meeting 2026 in Dallas.

The study examined an approach to address growing demands for patient portal messaging. Although generative AI can automate replies, the investigators noted uncertainty about response safety and appropriateness. Their evaluation asked surgeons to assess messages sent to their own patients during the perioperative period.

Ten fellowship-trained total joint arthroplasty surgeons, all existing users of an AI patient-messaging program, each graded 110 inquiry-response pairs. For each surgeon, we randomly selected 100 pairs from the 500 most recent exchanges, with a maximum of 10 per patient. Another 10 identical standardized scenarios—5 involving total knee arthroplasty (TKA) and 5 involving total hip arthroplasty (THA)—represented ideal responses and were graded by every surgeon to assess grading stringency.

Responses were scored on a 5-point Likert scale, ranging from unacceptable to exceeding expectations. Scores of ≥3 were considered acceptable, while scores of ≤2 were categorized as failures.

Study Findings

Among 1,000 own-patient responses, 619 involved TKA and 381 involved THA. Overall acceptability was 96.9% (95% CI, 95.2%–98.3%), with a mean score of 4.54. Individual surgeons’ acceptability rates ranged from 91% to 100%. All standardized scenario ratings were acceptable. Mean standardized and own-practice ratings correlated strongly across surgeons (r=0.96), which the investigators interpreted as consistent with variation in grading stringency rather than AI performance.

Of 31 unacceptable responses, 25 (81%) involved missing or outdated stored treatment preferences. Two involved medication safety concerns. No failure reflected misapplication of a correct, current preference.

Clinical Implications

According to the authors, the high response rates and autonomous delivery support surgeon-specific generative AI as a way to reduce messaging burden in arthroplasty practice.

Expert Commentary

“Because responses were delivered autonomously, these inquiries bypassed the clinical inbox, supporting surgeon-specific generative AI to reduce messaging burden in arthroplasty practice,” the researchers concluded.

Reference

Ehmke A, King CA, Raley J, et al. Reducing clinician messaging burden with autonomous surgeon-specific AI in total joint arthroplasty. Paper #10. Abstract presented at the American Association of Hip and Knee Surgeons Annual Meeting, November 5–8, 2026, Dallas, TX.