CoChat
Sign Up
Cited by 9Artificial Intelligence in Healthcare and EducationTotal Knee Arthroplasty OutcomesArtificial Intelligence in Healthcare

Comparative Efficacy of ChatGPT and DeepSeek in Addressing Patient Queries on Gonarthrosis and Total Knee Arthroplasty

Open access
Gurbuz, Serhat, Bahar, Hakan, Yavuz, Ulas, Keskin, Ahmet, Karslioglu, Bulent, Solak, Yener·2025·Arthroplasty Today
DOI 10.1016/j.artd.2025.101730PMC PMC12167088
Open this paper in CoChat — free accountOpen PDF

Abstract

Background: The advent of artificial intelligence (AI) in healthcare has opened new avenues for patient education and anxiety reduction. This study aims to compare the efficacy of 2 prominent AI platforms, ChatGPT and DeepSeek, in providing accurate and satisfactory responses to patients with gonarthrosis contemplating total knee arthroplasty (TKA). Methods: A prospective, comparative trial was conducted involving 100 patients diagnosed with gonarthrosis and indicated for TKA. Each patient posed 5 questions regarding the surgery and postoperative rehabilitation to both ChatGPT and DeepSeek. Responses were evaluated by 2 blinded orthopaedic specialists on a 10-point scale for accuracy and patient satisfaction. Patients also rated their satisfaction with each response on a 10-point scale. The primary outcome measures were the mean accuracy scores from specialists and mean satisfaction scores from patients. Results: < .001). ChatGPT demonstrated superior performance with a mean accuracy score of 8.7 ± 0.9 compared to DeepSeek's 7.4 ± 1.2. Patient satisfaction scores aligned with expert evaluations, with ChatGPT achieving a mean satisfaction score of 8.9 ± 0.8 vs DeepSeek's 7.6 ± 1.1. Notably, ChatGPT excelled in providing comprehensive explanations of surgical procedures (mean score 9.2 ± 0.7) and postoperative care (9.1 ± 0.8), while DeepSeek performed better in offering concise summaries of recovery timelines (8.3 ± 0.9). Conclusions: This study demonstrates that ChatGPT offers more accurate and satisfactory responses to patient queries regarding gonarthrosis and TKA compared to DeepSeek. The findings suggest that AI platforms, particularly ChatGPT, can serve as valuable tools in augmenting patient education and potentially reducing preoperative anxiety. Future research should explore the integration of AI-assisted information delivery in clinical practice and its long-term impact on patient outcomes.

AI snapshot

Written from the paper’s full text.

TL;DR

In a prospective comparative trial of 100 patients with gonarthrosis indicated for total knee arthroplasty, ChatGPT produced more accurate and more satisfactory responses to standardized patient questions than DeepSeek. ChatGPT scored higher on both expert-rated accuracy and patient satisfaction across all five question categories.

Key findings

  • ChatGPT had a mean expert-rated accuracy score of 8.7 ± 0.9 vs DeepSeek’s 7.4 ± 1.2, and a mean patient satisfaction score of 8.9 ± 0.8 vs DeepSeek’s 7.6 ± 1.1; differences were significant at P < .001.
  • Inter-rater reliability between the two blinded orthopaedic surgeons was excellent, with an intraclass correlation coefficient of 0.92 (95% CI: 0.88-0.95).
  • ChatGPT scored higher in surgical procedure explanation (accuracy 9.2 ± 0.7; satisfaction 9.3 ± 0.6), postoperative pain management (accuracy 9.1 ± 0.8; satisfaction 9.2 ± 0.7), and long-term outcomes (accuracy 8.7 ± 1.0; satisfaction 8.8 ± 0.9). DeepSeek performed better in offering concise summaries of recovery timelines (accuracy 8.3 ± 0.9; satisfaction 8.4 ± 0.8), but ChatGPT still scored higher in that category (accuracy 8.6 ± 0.9; satisfaction 8.7 ± 0.8).
  • Response times were comparable: ChatGPT 37.2 ± 5.3 seconds vs DeepSeek 35.8 ± 4.9 seconds. ChatGPT responses averaged 157 ± 23 words vs DeepSeek’s 112 ± 18 words.

Methods

  • Prospective, comparative trial conducted between January 2023 and January 2024 at Metin Sabancı Baltalimanı Bone Diseases Training and Research Hospital; 100 consecutive patients diagnosed with gonarthrosis and indicated for TKA completed the study (58 females, 42 males; mean age 67.3 ± 8.5 years, range 52-84).
  • Each patient posed five standardized core questions about surgery and postoperative rehabilitation to both ChatGPT and DeepSeek, phrased in their own words; responses were saved and evaluated by two blinded board-certified orthopaedic surgeons on a 10-point scale for accuracy, while patients rated satisfaction on a 10-point Likert-type scale.
  • Statistical analysis used SPSS version 28.0; paired t-tests compared mean scores between platforms, with P < .05 considered significant, and inter-rater reliability was assessed using the intraclass correlation coefficient.

Limitations

  • The authors state that allowing patients to phrase predefined questions in their own words introduced heterogeneity in question phrasing, and no subgroup or regression analyses were performed to investigate whether this variation affected accuracy or satisfaction ratings.
  • The authors state that no formal analysis was performed to identify factors preventing perfect satisfaction scores, and that most ratings clustered between 7 and 9, suggesting a possible ceiling effect; they also note that the patient satisfaction scale was not previously validated, though it was designed for this study and pilot-tested on 10 participants.
  • The authors state that AI-generated responses occasionally lack the nuance and personalization of human healthcare providers, that even ChatGPT did not consistently receive perfect expert scores, and that longitudinal studies are needed to assess long-term impact on clinical outcomes, patient satisfaction, and healthcare resource utilization.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1016/j.artd.2025.101730

Cites

18 of 18 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 6 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    For risks and complications, ChatGPT had a satisfaction score of 9.0 ± 0.7 compared with DeepSeek's 8.2 ± 0.8.

    “Risks and complications: Both platforms performed well, with ChatGPT (accuracy: 8.9 ± 0.8; satisfaction: 9.0 ± 0.7) slightly outperforming DeepSeek (accuracy: 8.1 ± 0.9; satisfaction: 8.2 ± 0.8).”
    Quote in the paperNot a substantial claim
  • Findingown data

    For long-term outcomes, ChatGPT had an accuracy score of 8.7 ± 1.0 compared with DeepSeek's 7.2 ± 1.3.

    “Long-term outcomes: ChatGPT provided more detailed and nuanced responses (accuracy: 8.7 ± 1.0; satisfaction: 8.8 ± 0.9) compared to DeepSeek (accuracy: 7.2 ± 1.3; satisfaction: 7.4 ± 1.2).”
    Quote in the paperSubstantial claim
  • Findingown data

    For risks and complications, ChatGPT had an accuracy score of 8.9 ± 0.8 compared with DeepSeek's 8.1 ± 0.9.

    “Risks and complications: Both platforms performed well, with ChatGPT (accuracy: 8.9 ± 0.8; satisfaction: 9.0 ± 0.7) slightly outperforming DeepSeek (accuracy: 8.1 ± 0.9; satisfaction: 8.2 ± 0.8).”
    Quote in the paperSubstantial claim
  • Findingown data

    For postoperative pain management, ChatGPT had an accuracy score of 9.1 ± 0.8 compared with DeepSeek's 7.6 ± 1.2.

    “Postoperative pain management: ChatGPT provided more comprehensive information (accuracy: 9.1 ± 0.8; satisfaction: 9.2 ± 0.7) compared to DeepSeek (accuracy: 7.6 ± 1.2; satisfaction: 7.8 ± 1.1).”
    Quote in the paperSubstantial claim
  • Findingown data

    For long-term outcomes, ChatGPT had a satisfaction score of 8.8 ± 0.9 compared with DeepSeek's 7.4 ± 1.2.

    “Long-term outcomes: ChatGPT provided more detailed and nuanced responses (accuracy: 8.7 ± 1.0; satisfaction: 8.8 ± 0.9) compared to DeepSeek (accuracy: 7.2 ± 1.3; satisfaction: 7.4 ± 1.2).”
    Quote in the paperSubstantial claim
  • May combine two claims. 9
  • Classified as probably not a substantial claim. 12

  • May combine two claims.

    Findingown data

    Both AI platforms showed consistent performance across genders, with no statistically significant differences (P > .05).

    “Notably, both AI platforms showed consistent performance across different age groups and genders, with no statistically significant differences observed (P > .05).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    ChatGPT consistently outperformed DeepSeek across all question categories, with higher accuracy and satisfaction scores.

    “Our findings reveal that ChatGPT consistently outperformed DeepSeek across all question categories, with significantly higher accuracy scores as evaluated by orthopaedic specialists and higher satisfaction scores reported by patients.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Both AI platforms showed consistent performance across different age groups, with no statistically significant differences (P > .05).

    “Notably, both AI platforms showed consistent performance across different age groups and genders, with no statistically significant differences observed (P > .05).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    No subgroup or regression analyses were performed to investigate whether variation in question formulation contributed to differences in accuracy or satisfaction ratings.

    “No subgroup or regression analyses were performed to investigate whether variation in question formulation contributed to differences in accuracy or satisfaction ratings.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Study design and conductreasoning

    Patients documented their interactions with both AI platforms by saving screenshots or copying text into Word documents, and this approach ensured the authenticity of the AI-generated responses.

    “Patients were required to document their interactions with both AI platforms by saving screenshots or copying the text into Word documents. They were instructed to bring these records to their follow-up appointments for evaluation. This approach ensured the authenticity of the AI-generated responses and allowed for comprehensive analysis by the research team.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    Two board-certified orthopaedic surgeons, blinded to the source of the responses, independently evaluated the accuracy and appropriateness of each AI-generated answer on a 10-point Likert scale.

    “Two board-certified orthopaedic surgeons, blinded to the source of the responses, independently evaluated the accuracy and appropriateness of each AI-generated answer on a 10-point Likert scale.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductcitation

    The patient satisfaction scale was not previously validated but was specifically designed for this study based on similar AI-patient interaction studies and pilot-tested on 10 participants to ensure usability.

    “While this was not a previously validated scale, it was specifically designed for this study based on similar AI–patient interaction studies and pilot-tested on 10 participants to ensure usability [11,12].”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [11] Busch, 2025, Evaluation of a retrieval-augmented generation-powered chatbot for pre-CT informed consent: a prospective comparative study, J Imaging Inform Med source

      Not checked in the free run.

    • cites [12] Yong, 2024, Performance of large language models in patient complaint resolution: web-based cross-sectional survey, J Med Internet Res source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    Allowing patients to phrase the predefined questions in their own words introduced heterogeneity in question phrasing that could have influenced how the AI platforms interpreted and responded to the queries.

    “another limitation of this study concerns the variability introduced by allowing patients to phrase the predefined questions in their own words. While this design choice aimed to simulate real-world interactions and enhance ecological validity, it also introduced heterogeneity in question phrasing that could have influenced how the AI platforms interpreted and responded to the queries.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    The finding of no significant differences in AI performance across age groups contradicts the common assumption that older patients might struggle with or be less satisfied with technology-based interventions.

    “Interestingly, our study found no significant differences in AI performance across different age groups, contradicting the common assumption that older patients might struggle with or be less satisfied with technology-based interventions.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

    Interpretationreasoning

    The ability to provide consistent, patient-friendly explanations of complex medical procedures and recovery processes may help alleviate fears and misconceptions, potentially leading to improved patient preparedness and postoperative outcomes.

    “The ability to provide consistent, patient-friendly explanations of complex medical procedures and recovery processes may help alleviate fears and misconceptions, potentially leading to improved patient preparedness and postoperative outcomes.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning

    The ability of AI platforms to provide instant, round-the-clock access to information may contribute significantly to patient empowerment and informed decision-making.

    “The ability of AI platforms to provide instant, round-the-clock access to information may contribute significantly to patient empowerment and informed decision-making.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    AI platforms are inherently limited in their ability to personalize responses to reflect the full clinical context of individual patients.

    “While AI platforms can provide standardized information that may reduce general uncertainty and anxiety, they are inherently limited in their ability to personalize responses to reflect the full clinical context of individual patients.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Wide interindividual variability in preoperative knee function, comorbidities, and health literacy substantially influences the risk profile, surgical outcomes, and rehabilitation expectations for each patient undergoing TKA.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    By offering instant, round-the-clock access to accurate and comprehensive information about TKA, AI platforms can serve as valuable tools in addressing patient concerns and uncertainties.

    “By offering instant, round-the-clock access to accurate and comprehensive information about TKA, these AI platforms can serve as valuable tools in addressing patient concerns and uncertainties.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    No formal analysis was performed to identify specific trends or factors preventing perfect scores.

    “No formal analysis was performed to identify specific trends or factors preventing perfect scores.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Previous study reported increased patient engagement and reduced anxiety levels when AI-assisted education was incorporated into preoperative counseling for orthopaedic procedures.

    “These findings are consistent with previous study which reported increased patient engagement and reduced anxiety levels when AI-assisted education was incorporated into preoperative counseling for orthopaedic procedures [16].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [16] Morya, 2024, Application of ChatGPT for orthopedic surgeries and patient care, Clin Orthop Surg source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Ethical concerns regarding data privacy, the potential for misinformation, and the need for regular content updates remain significant challenges for AI applications in orthopaedics.

    “Ethical concerns regarding data privacy, the potential for misinformation, and the need for regular content updates remain significant challenges for AI applications in orthopaedics [18].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [18] Lisacek-Kiosoglous, 2023, Artificial intelligence in orthopaedic surgery, Bone Joint Res source

      Not checked in the free run.

16 of 16

  • The paper relies on this without support.

    Wide interindividual variability in preoperative knee function, comorbidities, and health literacy substantially influences the risk profile, surgical outcomes, and rehabilitation expectations for each patient undergoing TKA.

    “Another important consideration in the context of patient education is the wide interindividual variability in preoperative knee function, comorbidities, and health literacy. These factors substantially influence the risk profile, surgical outcomes, and rehabilitation expectations for each patient undergoing TKA.”
1 of 1

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

  • DeepSeek overall satisfaction does not match its category scores

    Major

    DeepSeek's five category satisfaction means average 7.94, which rounds to 7.9 rather than the reported overall mean of 7.6.

    Reported 7.6 · recomputed 7.94

1 of 1

Cites

Flagged first, then by citations

1 of the 19 references couldn’t be matched to a known paper.

Cited by

Sorted by citations

Related

Authors
Gurbuz, S., Bahar, H., Yavuz, U., Keskin, A., Karslioglu, B., & Solak, Y.
ORCID
Venue
Arthroplasty Today
Publisher
Elsevier BV
Published
2025-06
Volume
33
Pages
101730
ISSN
2352-3441 (ISSN-L)
DOI
10.1016/j.artd.2025.101730
PMC
PMC12167088
OpenAlex
W4410953768
Cited by
9
References
19
Citations by year
2026: 6 · 2025: 3
FWCI
2.17
Type
journal article
Language
en
Open access
Licence
Locations
  • Arthroplasty Today (publishedVersion, cc-by-nc-nd)
  • PubMed Central (publishedVersion, cc-by-nc-nd)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Topics
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Health Sciences › Medicine › Surgery › Total Knee Arthroplasty Outcomes
  • Health Sciences › Health Professions › Health Information Management › Artificial Intelligence in Healthcare
Keywords
ChatGPT, DeepSeek, large language models, knee osteoarthritis, total knee arthroplasty, patient education, preoperative anxiety, postoperative rehabilitation, response accuracy, patient satisfaction

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-07.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-07.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review