Optimizing Hip and Knee Arthroplasty Clinic Flow: A Prospective Evaluation of Artificial Intelligence Scribe Technology

Abstract

Background

The increasing burden of clinical documentation contributes to physician inefficiency and burnout, with electronic health record implementation significantly increasing documentation time from 16% to 28% of clinical time. While human medical scribes have shown benefits, artificial intelligence (AI) scribes represent a promising solution. This study evaluated the impact of AI scribe technology on clinical encounter duration and physician workload in a high-volume hip and knee arthroplasty clinic.

Methods

This prospective quality-improvement cohort study was conducted by a single surgeon at one institution from November 2024 to February 2025. Hip and knee arthroplasty patients were divided into 2 cohorts: a control group using traditional physician and medical assistant documentation within the electronic medical record and an intervention group using AI-powered clinical documentation. Clinical encounter duration was assessed by 3 independent reviewers using direct observation and stopwatch timing. The National Aeronautics and Space Administration Task Load Index (NASA-TLX) was completed after each clinic day to quantify subjective physician workload across 6 domains on a 20-point scale.

Results

A total of 112 patient encounters were analyzed: 55 with AI scribes and 57 with traditional documentation. Encounter time was significantly shorter in the AI group, with a mean of 12.03 minutes (range, 4.01-22.13) compared to 15.06 minutes (range, 6.20-32.27) for traditional documentation (P < .001). Overall, physician workload measured by NASA-TLX was lower with AI assistance, with a mean total score of 30.0 (range, 26-40) vs 40.2 (range, 11-73) for traditional documentation (P = .485). The AI group showed consistently lower mean scores across all NASA-TLX subdomains, including temporal demand (5.8 vs 10.4), effort (6.8 vs 10.0), and frustration (3.4 vs 5.4).

Conclusions

AI scribe implementation significantly reduced clinical encounter duration by approximately 20% and demonstrated promising improvements in physician workload metrics.

Background

The increasing burden of clinical documentation contributes to physician inefficiency and burnout, especially with the widespread adoption of electronic health records (EHRs). EHR implementation has significantly increased physicians' documentation time from approximately 16% to 28% of observed clinical time, adding strain to already busy schedules [1]. This trend is confirmed across studies, with evidence showing that after-hours documentation is perceived as one of the most burdensome aspects of physicians' work and is strongly associated with burnout symptoms [2]. The design and usability of current EHR systems exacerbate the problem, as physicians report that many documentation tasks are clerical in nature and do not require clinical training [3]. Furthermore, poorly designed interfaces and time-consuming navigation contribute to cognitive fatigue [3]. Ultimately, the longer work hours, heavy clerical demands, and interruptions to patient care that come with current documentation methods have created frustration among physicians [4]. This documentation burden has been associated with medical errors, risks to patient safety, poor documentation quality, job attrition, and burnout [4]. These findings highlight the need for targeted interventions and improvements to alleviate documentation burden and promote both efficiency and physician well-being in clinical practice.

To address this, many practices have implemented human medical scribes, which have been shown to reduce documentation time and improve physician satisfaction [5,6]. In high-volume specialties like orthopaedics, scribes can have a large impact: one orthopaedic sports medicine clinic found that scribes reduced half-day documentation time from 87 to 26 minutes [6]. More recently, “ambient” artificial intelligence (AI) scribes have demonstrated similar benefits. These systems use advanced speech recognition and natural language processing to automate clinical documentation, easing administrative burden [7]. Ambient AI scribe systems, such as Microsoft DAX Copilot, integrate directly into the clinical workflow by passively capturing patient-clinician conversations through a secure audio interface and automatically generating draft visit notes within the EHR. These tools use speech recognition and natural language processing to identify relevant clinical concepts, structure the encounter narrative, and populate the documentation in real time. The clinician then reviews, edits, and signs the AI-generated note, offloading the majority of clerical work to the system while preserving physician oversight and documentation quality. Ambient AI scribes in outpatient clinics can reduce time spent on notes by 20.4% and lower after-hours EHR work by 30.0% [8]. Additionally, Rotenstein et al. [9] showed that virtual scribe use reduced total EHR time per appointment by an average of 5.6 minutes. These findings suggest that AI-assisted documentation may improve efficiency and physician well-being. However, their impact in specialty practices, such as hip and knee arthroplasty, has not been well reported. This study presents a prospective evaluation of an AI scribe system in a high-volume hip and knee arthroplasty clinic, hypothesizing that integration of the technology would shorten encounter duration and lessen physician workload.

Material and methods

Study design and sample

Institutional review board exemption was obtained at our institution prior to data collection. This was a prospective quality-improvement cohort study at a single institution performed by a single surgeon. Hip and knee arthroplasty patients were divided into 2 cohorts based on the method of clinical documentation used during their arthroplasty clinic visits. The control cohort consisted of patients whose clinical encounters were documented using traditional methods by the physician and medical assistant (MA) within the electronic medical record (EMR) system (Cerner Oracle Health, Kansas City, MO). The AI intervention cohort included patients whose visits were documented with the use of an AI-powered clinical documentation assistant (Microsoft DAX Copilot, Redmond, WA), in collaboration with the orthopaedic surgeon. Patients were grouped according to clinic workflow, with the first series of encounters documented using the traditional physician-MA EMR workflow, followed by the next series of encounters documented using the AI-powered scribe system; no randomization or patient-level selection criteria were used, and assignment into each group was determined solely by the documentation method in use during that clinic session. All patients whose encounters were assigned to the AI intervention group were informed of the use of the AI scribe system prior to their visit and verbally consented to the recording of their encounter. No patients refused participation.

All consecutive adult hip and knee arthroplasty clinic encounters for a single surgeon during the study period were eligible for inclusion. We included all visit types, including new patient consultations, preoperative evaluations, routine postoperative follow-up, annual surveillance visits, and problem-focused return visits. Encounters were excluded only if complete timing data could not be obtained or if the AI scribe was not successfully activated for the duration of the visit. This approach allowed for a comprehensive assessment of AI scribe performance across all encounter types commonly seen in a high-volume arthroplasty clinic.

Clinical encounter duration and documentation time were assessed by 3 independent reviewers using direct observation. In our institution, MAs complete documentation in real time during the patient visit, entering data into the EMR within the patient examination room. Encounter timing was recorded using a stopwatch. Timing began with the MA's entry into the exam room and was paused when the MA exited the room. Timing resumed with the surgeon's entry and was again paused at the time of the surgeon's exit. To avoid inflating encounter times, the stopwatch was paused during patient transitions and restarted when the surgeon began chart documentation on his personal computer following the visit. This approach was designed to capture the total duration of clinical interaction as well as the subsequent documentation period for each patient encounter.

In the AI scribe group, at the beginning of each encounter the surgeon or MA activated the AI scribe, which passively captured the conversation between the patient and care team through a secure microphone. The system then generated a structured draft clinical note, including history, examination findings, assessment, and plan, which was imported into the EMR. The surgeon reviewed and edited this draft as needed for accuracy and completeness before signing the final note. Any additional time spent editing the note after being generated by the AI scribe was included in the total encounter time. No changes were made to rooming procedures, visit scheduling, or staffing between the traditional documentation and AI scribe workflows.

The surgeon also completed the National Aeronautics and Space Administration Task Load Index (NASA-TLX) after each clinic day to quantify subjective workload. The NASA-TLX is a validated instrument yielding scores for mental demand, physical demand, temporal demand, perceived performance, effort, and frustration assessed on a 20-point scale [10].

Data analyses

Continuous variables were presented as means with ranges. Primary outcomes included clinical encounter times and physician documentation workload. Comparisons between the control group and AI-intervention group were assessed using independent t-tests for the continuous variables. NASA-TLX scores were summed (0-120), with lower total scores indicating decreased perceived workload. The level of statistical significance was set at a P value of <.05. A priori power analysis was performed to detect a medium effect size difference in clinical encounter duration (Cohen's d = 0.6). This analysis indicated that 45 patients per group were needed to demonstrate a statistically significant difference (α = 0.05, power = 0.80, 2-tailed). Our final sample of 55 and 57 encounters exceeded this threshold. A total of 112 patient encounters were included in the final analysis. The control group comprised 57 patients, while the AI intervention group included 55 patients.

Results

A total of 112 patient encounters were analyzed, 55 with AI scribes and 57 with traditional documentation. Encounter time was significantly shorter in the AI group, with a mean of 12.03 minutes (range, 4.01-22.13) compared with 15.06 minutes (range, 6.20-32.27) for traditional documentation (P < .001). Overall physician workload, measured by the NASA-TLX, was lower with AI assistance, with a mean total score of 30.0 (range, 26-40) vs traditional documentation with a mean total score of 40.2 (range, 11-73) (P = .485). Across the 6 NASA-TLX subdomains, AI documentation yielded consistently lower mean scores. Temporal demand averaged 5.8 (range, 2-8) with AI vs 10.4 (range, 1-19) without (P = .282). Effort was 6.8 (range, 5-8) with AI vs 10.0 (range, 2-19) without (P = .433). Frustration averaged 3.4 (range, 1-6) with AI vs 5.4 (range, 2 to 13) without (P = .391). Mental demand averaged 6.2 (range, 2 to 10) with AI vs 8.0 (range, 1 to 14) without (P = .838). Physical demand averaged 4.5 (range, 2-5) with AI vs 6.0 (range, 1-7) without (P = .796). Performance averaged 3.3 (range, 4-7) with AI vs 5.1 (range, 1-7) without (P = .643).

All values are summarized in Table 1 and Figure 1.

Table 1.

Comparison of clinical encounter times and NASA-TLX scores between AI and traditional documentation methods.

Variables	AI	Traditional
Number of encounters	55	57
Encounter time (min) (P < .001)	12.03 ± 3.73 (4.01-22.13)	15.06 ± 5.40 (6.20-32.27)
NASA-TLX total (n = 5; P = .485)	30.0 ± 5.3 (26-40)	40.2 ± 6.1 (11-73)
Temporal demand (n = 5; P = .282)	5.8 ± 1.2 (2-8)	10.4 ± 1.6 (1-19)
Effort (n = 5; P = .433)	6.8 ± 1.3 (5-8)	10.0 ± 1.5 (2-19)
Frustration (n = 5; P = .391)	3.4 ± 1.0 (1-6)	5.4 ± 1.1 (2-13)
Performance (n = 5; P = .643)	3.3 ± 1.1 (4-7)	5.1 ± 1.3 (1-7)
Physical demand (n = 5; P = .796)	4.5 ± 1.2 (2-5)	6.0 ± 1.3 (1-7)
Mental demand (n = 5; P = .838)	6.2 ± 1.1 (2-10)	8.0 ± 1.4 (1-14)

Figure 1.

Comparison of NASA Task Load Index scores between AI and traditional documentation methods.

Discussion

In this high-volume arthroplasty clinic, implementing an AI scribe significantly accelerated clinic flow and showed promising reductions in physician workload. These findings align with the broader literature on documentation burden and scribes, especially in orthopaedics. Orthopaedic surgeons have been reported to experience the highest levels of EHR-related stress among medical specialties [11]. Given the clear link between heavy documentation loads and clinician burnout, solutions like medical scribes have gained traction [12]. Scribes can mitigate many unintended inefficiencies of EHRs and improve physician well-being, and emerging technologies (remote scribes, voice recognition, and AI) may further alleviate documentation burdens. Our results provide support for the conclusion that in the arthroplasty setting, such technology can enhance efficiency.

We found a ∼20% reduction in encounter time per patient with the AI scribe, equating to about 3 minutes saved on average. In our arthroplasty clinic, saving roughly 3 minutes per encounter could allow an extra 3–4 patients to be seen in a full day or simply give providers more time for direct patient care. This suggests streamlining documentation can translate into seeing more patients or finishing clinics earlier. Over weeks and months, these efficiency gains can meaningfully increase access and throughput in a busy surgical practice.

Our findings with an AI-driven ambient scribe mirror those from other specialties that have begun adopting such tools. Stults et al. [13] observed a significant decrease in documentation time and cognitive load measured using the NASA-TLX in ambulatory clinics with an AI documentation platform. In a systematic review of AI scribe systems, Sasseville et al. [14] found overall positive effects on provider efficiency and engagement, with clinicians reporting that AI scribes helped lighten their workload and allowed them to be more involved in patient care. Our study adds to current findings, demonstrating that these benefits extend to orthopaedic subspecialty care, specifically arthroplasty. Notably, we observed no downsides in clinic operations or staffing; support staff were freed from clerical tasks rather than made busier, which aligns with other reports that scribes (human or AI) can be integrated without disrupting team workflows. This suggests that implementing an AI scribe in orthopaedics can be a scalable way to improve efficiency without requiring additional personnel, especially important in high-volume practices.

Although the reduction in the surgeon's NASA-TLX workload score with AI assistance did not reach statistical significance in our sample, the trend favored the AI scribe across all measured domains. Subjectively, the surgeon reported less time pressure, effort, and frustration on clinic days when the AI scribe was used. This trend is noteworthy because even modest reductions in perceived workload can have a significant impact on burnout risk. For instance, prior research has shown that each 10% drop in task load is associated with a one-third lower odds of burnout symptoms [15]. In our study, the AI tool nearly halved the surgeon's frustration score (from 5.4 to 3.4), which could translate to meaningful improvements in long-term well-being if sustained. These findings echo recent qualitative data on ambient AI scribes. In a 2025 study of physicians piloting an AI scribe, 100% of participants reported lower cognitive demand, and the majority noted reduced temporal demand and improved work-life integration when using the system [16]. This aligns with the experience in our arthroplasty clinic, where the surgeon felt the AI assistant offloaded some of the documentation effort, which in turn may contribute to better work-life balance. Taken together, our data and these prior studies indicate that AI scribes have the potential to improve physician well-being by lightening the clerical load and cognitive burden inherent in clinic documentation.

The accuracy and frequency of corrections required for AI-generated notes were not formally tracked in this study, as the primary focus was clinical efficiency and physician workload. Future studies specifically designed to evaluate AI scribe accuracy, error typology, and correction frequency in orthopaedic subspecialty settings are warranted and represent an important direction for subsequent investigation. Although formal error tracking was not performed, the AI scribe subjectively demonstrated a high degree of accuracy in summarizing patient-physician conversations and required very few corrections to the generated note content. Where additions were necessary, they were largely limited to components the ambient scribe was not designed to capture passively, most notably physical examination findings, which were manually entered by the surgeon following note generation. Other reports caution that today's AI scribes often require human oversight to catch mistakes, which can erode some of the efficiency gains [16]. Going forward, continued advances in natural language processing and domain-specific training (in our case, orthopaedic terminology) will be critical to minimizing error rates and editing time. Future iterations that incorporate clinician feedback and adapt to individual practice styles are likely to streamline the process further.

Another consideration is cost and return on investment. Human scribes, whether in-person or virtual, incur significant personnel costs, but they can be cost-effective by allowing physicians to see more patients or by improving billing documentation. AI scribes, on the other hand, typically involve subscription or licensing fees for the software (and sometimes hardware). We did not perform a cost analysis in this study, but the potential for productivity gains (20% faster encounters in our data) suggests that an AI scribe could pay for itself if that time saved is redeployed into additional visits or improved throughput. For example, if a surgeon can consistently add even one more new patient appointment per clinic day thanks to time saved on documentation, the revenue from those visits may offset the AI service fee. Of course, this calculus will vary by practice and the specific payment model. The model used in this study, Microsoft DAX Copilot, has a flat-rate subscription cost of US $369 per provider per month (plus a one-time implementation fee of approximately US $700 per user) for unlimited patient encounters and full EMR integration. It will also be important to study which settings and providers benefit most from AI scribes. Not every physician experiences the same level of efficiency gain—factors such as baseline typing speed, visit complexity, and personal workflow preferences likely influence the impact of a scribe.

As with any third-party platform through which protected health information flows, the use of ambient AI scribe technology introduces additional data security considerations. Clinicians and institutions considering adoption of such tools should carefully vet vendor agreements regarding data transmission protocols, storage practices, and data ownership, including whether patient audio or transcription data may be used for model training or retained beyond the clinical encounter. We emphasize the importance of rigorous institutional review of any AI documentation tool prior to clinical implementation.

Limitations

Several potential limitations must be acknowledged when interpreting the results. First, our analysis was conducted by a single institution under the supervision of a single surgeon, with a relatively small sample size and short study duration. As such, there is an inherent risk of selection bias and generalizability concerns. While the single-surgeon design limits generalizability, this was an intentional methodological choice to isolate the effect of the AI scribe by eliminating intersurgeon variability in documentation style, note templates, and baseline clinical efficiency. Multicenter, multisurgeon studies are needed to establish external validity across diverse orthopaedic practice settings. The clinical workflow, documentation practices, and AI system implementation were specific to our setting, but these factors may differ elsewhere. Results might vary with different AI scribe platforms, EMR systems, or in centers using different clinic protocols. Thus, while our findings are internally valid for our setting, caution is warranted in extrapolating them to all orthopaedic practices.

Another limitation is the lack of long-term outcome data and patient-centered metrics. We focused on encounter duration and physician workload in the immediate clinic setting, but we did not formally assess patient satisfaction, documentation quality, or billing accuracy. It remains unclear whether the improved efficiency with AI scribes translates into better patient experiences or maintained clinical documentation standards.

The accuracy and reliability of AI-generated documentation also represent ongoing limitations. Current ambient AI systems require human oversight to identify and correct transcription errors, which can reduce some of the efficiency gains. Additionally, the NASA-TLX workload assessment, while validated, remains a subjective measure that may be influenced by individual perception and day-to-day variability. Finally, our study period was relatively brief, and a longer-term evaluation would be necessary to assess sustained benefits and potential adaptation.

Conclusions

This prospective cohort study demonstrated that the implementation of an AI scribe system significantly reduced clinical encounter duration and improved physician workload metrics in a hip and knee arthroplasty clinic. The 20% reduction in encounter time and consistently lower NASA-TLX scores across all domains suggest that AI-powered documentation can enhance clinic efficiency while reducing physician burden. These findings add to the growing body of literature supporting the integration of AI scribe technology as a valuable tool in healthcare documentation workflows. Further randomized controlled trials with larger sample sizes are warranted to validate these results and assess long-term outcomes, cost-effectiveness, and patient satisfaction measures across diverse orthopaedic practice settings.

Conflicts of interest

A. McDaid is a paid employee for The University of Auckland and Fios Health Ltd; and has stock or stock options in Fios Health Ltd. C.M. Lawrie receives royalties from Zimmer Biomet; is a paid consultant for Zimmer Biomet, Smith & Nephew, Mizuho OSI, Medtronic, Ignite Orthomotion, Ganymed Robotics, and Fios Health; holds stock or stock options in Fios Health; is an AAHKS Fellowship Oversight Committee member; and is the president of the Anterior Hip Foundation; all other authors declare no potential conflicts of interest.

For full disclosure statements refer to https://doi.org/10.1016/j.artd.2026.102095.

CRediT authorship contribution statement

Zachary Grand: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Resources, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Jonathan Brutti: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Resources, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Klaudia Greer: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Resources, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Jason Mirharooni: Writing – review & editing. Andrew McDaid: Writing – review & editing, Conceptualization. Charles M. Lawrie: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Resources, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization.