Implementation determinants of a planned machine learning-enabled surgical scheduling system in a high-volume orthopaedic centre in Canada: qualitative findings

Abstract

Objectives

Elective non-emergent surgical wait times have increased across countries such as Canada, straining operating room (OR) resources and affecting patient outcomes and healthcare spending. Manual scheduling systems in Ontario orthopaedic centres create wide variations in wait times, with recent declines in meeting benchmark targets despite increased procedure volumes. Challenges stem from fragmented referral processes, outdated scheduling methods and resource constraints. Artificial intelligence and machine learning (ML) offer potential solutions for optimising scheduling; however, their implementation remains inconsistent. This study aims to identify determinants affecting the rollout of a new ML-driven automated scheduling system at a high-volume elective orthopaedic surgery centre.

Design

A qualitative description approach supported by implementation science frameworks.

Setting

A high-volume elective orthopaedic surgery unit at a Canadian tertiary care centre.

Participants

17 individuals from clinical, administrative and leadership roles who were directly involved in surgical scheduling.

Interventions

A new ML-driven automated surgical scheduling system.

Outcomes

Perceptions of the proposed new surgical scheduling system (barriers and enablers of implementation, recommendations for improvement).

Results

Three main themes were identified, capturing challenges and enablers in the existing scheduling system: system functionality, process-related factors and resource constraints.

Participants described substantial inefficiencies in the existing manual scheduling system, including outdated software, fragmented information systems, inconsistent communication and resource constraints. Across interest-holder groups, there was broad but variable perceived support for a planned ML-enabled scheduling system, particularly for improving duration prediction, access to scheduling data and reporting, alongside concerns about system complexity, workflow fit, training and resource implications. Interest-holders emphasised the importance of user-friendly design, interoperability, responsive training, phased implementation and ongoing feedback.

Conclusions

This pre-implementation qualitative study identified significant process and resource limitations in manual orthopaedic surgical scheduling, but interest-holder support for a well-designed ML-driven system is strong. While participants anticipated potential benefits for scheduling accuracy, throughput and resource allocation, these perceived advantages will require meaningful user engagement, robust training, phased rollout and evaluation in subsequent implementation and outcome studies.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• Our interest-holder engagement approach captured perspectives across scheduling, clinical and leadership roles.

• We used robust implementation science frameworks (Consolidated Framework for Implementation Research, Knowledge-to-Action, Theoretical Domains Framework), providing theoretical grounding and structure for data collection and analysis.

• By using a qualitative description methodology, the research captures nuanced perspectives essential to real-world implementation.

• The single-site focus may limit generalisability to other hospital contexts; however, our context-specific focus at a high-volume orthopaedic site provides practical relevance to other surgical centres exploring similar innovations.

• Upcoming changes, including the implementation of new hospital information systems and expanded weekend scheduling with surgeons from other hospitals, will require additional evaluation to ensure successful implementation and scale-up.

Introduction

Across Canada and other Organisation for Economic Co-operation and Development (OECD) countries, wait times for elective (non-emergent) surgeries have steadily increased, a trend that predates the COVID-19 pandemic.¹ Because operating rooms (OR) are among the most resource-intensive areas of a hospital, optimising their use is essential for reducing surgical wait times and managing healthcare spending.² Adults waiting for non-urgent surgeries in Ontario also experience wide variation in wait times. For example, median wait times range from 43 weeks to 298 weeks for hip replacement and from 24 weeks to 368 weeks for knee replacement.³ This wide range is attributed to fragmented, non-centralised referral systems, limited or inefficient use of ORs and human resources and the potential for continued impact of COVID-19.³ In 2023, only 66% of hip replacement patients and 59% of knee replacement patients (excluding emergency cases) received treatment within the recommended national benchmark wait times.⁴ This represents a decline from pre-COVID levels in 2019, when 75% of hip and 70% of knee replacement patients received surgery within benchmarks,⁵ despite more surgeries being completed in 2023.⁴ This decline can be attributed to a backlog of delayed procedures and an increasing population-level demand for joint replacements.⁴

Long surgical wait times have measurable effects on patient health and quality of life. Delays are associated with poorer preoperative health and outcomes, including longer hospital stays and reduced functional recovery, which can result in increased healthcare costs.^{6 7} Pain, limited mobility and mental/emotional distress are also common effects of long waits.^{8 9} In 2023, surgical delays led to an estimated average annual loss of $C2871 in wages and productivity among Canadian patients.¹⁰ Other drivers of increased demand for surgical services include the ageing and growing population.^{5 11}

Efforts to reduce surgical wait times through traditional supply-side interventions (eg, expanding infrastructure/staff) have yielded limited success across OECD nations, including Canada.^{12 13} While additional resource allocation, such as extended operating days and weekend ORs, can certainly reduce wait times, this comes at a high financial cost. Artificial intelligence (AI) and machine learning (ML) can offer new opportunities to enhance OR efficiency and optimise surgical waitlist management. While these approaches have shown theoretical promise,^{14 15} they remain fragmented in their implementation, require further validation and face challenges related to data access, privacy concerns and the risk of perpetuating healthcare inequities.^{16 17} In response, our team at Sunnybrook Research Institute and the University of Toronto proposed a ‘smart joint prediction-scheduling system’ to enhance OR resource allocation at the Holland Orthopaedic and Arthritic Centre (a part of Sunnybrook Health Sciences Centre (SHSC)) in Toronto, Ontario, Canada. This data-driven approach integrates predictive analytics with automated scheduling to increase surgical throughput, defined as the number of surgical cases completed within a given time period,¹⁸ while also addressing systemic inequities that AI can reinforce if not explicitly mitigated.^{19 20} More specifically, a team of researchers and clinicians developed a new surgical scheduling system, a data-driven, automated system designed to leverage hospital datasets to increase throughput, minimise after-hours care, generate cost savings, decrease surgical wait times and enhance OR time utilisation. Its goal is to optimise the surgical scheduling process and patient throughput using ML-predicted operative times and schedule-optimisation formulas. We hypothesise that an end-to-end ML-based approach that intelligently leverages readily available hospital datasets and patient preoperative factors will increase throughput, minimise after-hours care, generate cost savings, decrease wait times and improve OR time utilisation compared with current, manual practices. Beyond informing implementation at our own high-volume elective orthopaedic centre, this study could support other surgical programmes considering AI and ML-enabled scheduling tools.

In addition to considering an ML-based approach to surgical scheduling, it is also important to ensure the optimised implementation of such innovations. However, this requires individual and collective behaviour change across the healthcare system, in which innovation will be implemented.²¹ Adoption requires a nuanced understanding of the contextual and behavioural factors that influence change, as well as the input, collaboration and leadership of healthcare professionals and support staff throughout the innovation’s development, implementation and ongoing management.²² By using implementation science frameworks to identify multilevel determinants, we provide a structured, transferable account of barriers, enablers and context conditions that other hospitals can use to assess their own readiness, adapt similar systems and plan implementation strategies.

Accordingly, the objective of our study was to conduct a pre-implementation qualitative study to understand existing surgical scheduling processes and to identify implementation determinants relevant to the design and planned rollout of an ML-enabled scheduling system at a high-volume orthopaedic centre. Our aim was not to evaluate the performance of a deployed system, but to generate context-specific, implementation-focused evidence to inform its development, piloting and future evaluation.

Methods

Setting

At the time of the study, elective orthopaedic surgical scheduling at the Holland Orthopaedic Centre was supported by a hybrid system involving legacy software, hospital-specific electronic tools and manual processes. Core intraoperative scheduling and documentation were managed through PICIS OR Manager, whereas preoperative information and waitlist data were accessed through SunnyCare, a hospital-specific clinical information system. In addition, administrative assistants and schedulers relied heavily on paper forms, locally maintained spreadsheets, email and telephone communication to coordinate referrals, bookings and day-of-surgery adjustments. Master surgical time blocks were constructed in Microsoft Word and maintained by a designated knowledge user; availability information was collated using personal scheduling software and transferred into Excel; and clinic offices relied on a hybrid of paper and electronic lists to manage waitlists and communicate case information. A fully integrated commercial electronic health record-embedded scheduling module (eg, Epic) was not yet in place during the study period, although implementation of a new hospital information system was planned. This hybrid environment shaped both the limitations of the existing process and interest-holder expectations for the proposed ML-enabled scheduling system. Figure 1 summarises the resulting six-step scheduling and workflow continuum, illustrating the flow of information between systems.

Figure 1. Existing surgical scheduling process at Sunnybrook's Holland Centre. OR, operating room.

Study design

We conducted a qualitative description study, which employs low-inference interpretation (ie, description), allowing researchers to remain close to the data.23,25 This methodology was appropriate for addressing our objective of identifying the perceptions of individuals who are directly involved in using the current surgical scheduling system or who will implement and use the proposed new system.^{24 26} Our work was grounded in implementation science to uncover the determinants of implementing the new surgical scheduling system and to inform the system’s future scale-up to other surgical sites. These included consideration of the ‘assess barriers to knowledge use’ action step of the Knowledge-to-Action (KTA) model, which outlines processes to ensure the optimised uptake of knowledge from research into practice.²⁷ The interview guide was informed by the Theoretical Domains Framework (TDF)²¹ to ensure coverage of key individual-level determinants, including knowledge, skills, beliefs about capabilities and consequences, environmental context and social influences across interest-holder groups. We then used the Consolidated Framework for Implementation Research (CFIR)²⁸ as the primary analytical framework to organise emergent determinants related to the innovation, inner setting, implementation processes and individuals involved in implementation. This complementary use enabled us to structure data collection around behavioural determinants while interpreting findings within a widely used multilevel implementation framework for complex healthcare innovations. To guide the reporting of our study, we used the Standards for Reporting Qualitative Research (SRQR).²⁹

Patient and public involvement

This research did not involve patients or the public.

Participant recruitment

Participants were purposively recruited from the Holland Orthopaedic Centre at SHSC, one of Canada’s largest centres for orthopaedic surgery, which performs over 5000 elective orthopaedic surgical cases (including 4000 hip and knee replacement procedures) annually using criterion sampling.³⁰ We recruited individuals across the surgical scheduling pathway who are using the current surgical scheduling system and would be involved in implementing or using the new system: administrative staff (orthopaedic surgery administrative assistants, surgical schedulers, booking clerks, charge nurses, surgical administrative staff), leadership and decision-makers (decision support staff, unit managers, programme directors) and surgeons, who were identified with input from study collaborators and the orthopaedic surgical division chief. To ensure thematic saturation and the inclusion of relevant knowledge users,³¹ we invited an additional four individuals beyond those initially identified. The invitations included a link to an online consent form and a demographic survey (Qualtrics V.2.14.0), which the research team pilot-tested for clarity of the questions. Reminders were sent 1 and 2 weeks after the study started, with one phone follow-up. On consent, each participant received a unique ID to maintain data anonymity.

Data collection

Given the wide range of knowledge user roles in implementing the new surgical scheduling system, data were collected through individual interviews, thereby enhancing the breadth of perspectives. This also allowed us to gain an in-depth understanding of their current role and perspective on the future system.³² We iteratively developed a semistructured and open-ended interview guide, which was refined and pilot-tested by our integrated knowledge translation (IKT) team, comprising health services researchers, physicians and decision-makers at SHSC (online supplemental table 1). IKT is a collaborative research approach that engages all relevant knowledge users across most aspects of the research process to enhance the relevance of findings.³³ The development of the interview guide was informed by the TDF, which outlines 14 behaviour change constructs (across cognitive, affective, social and environmental factors) that have been shown to influence implementation.²¹ Eight of these behaviour change constructs were relevant to the development of our interview guide, which addresses our objectives. Questions prompted participants (1) to describe the processes they currently use to schedule surgeries and their experiences with the existing system for elective surgeries, including any challenges and enablers and how they overcome them (recommendations), and (2) to share their perceptions of the proposed new scheduling system and its optimised implementation. Before asking for their perceptions of the new system, our team developed a brief presentation of its features and functions to ensure they were sufficiently informed about how it might be used and implemented, enabling them to answer our questions.

Interview procedures

Interviews were conducted via videoconference (Zoom, Microsoft Teams or phone) between June and August 2023. Each session lasted 45–60 min. Sessions were moderated by two researchers (JM, IH), who had no prior relationships with the participants, and were conducted in a conversational tone to align with the semistructured approach. Sessions began with introductions and a review of the study objectives, followed by a reconfirmation of their consent. After each interview session, the research team conducted a debriefing to identify preliminary themes, assess whether any process modifications were warranted and discuss how their roles and relationships might influence data collection and interpretation.^{34 35} Data saturation, defined as the point at which no new codes are identified in consecutive interviews, was determined through team debrief discussions and the moderator’s field notes.³⁶

Data analysis

Audio recordings of the interviews were transcribed verbatim. All transcripts were deidentified using unique participant identification numbers. Data analysis was conducted using directed content analysis, which involved both inductive and deductive coding approaches.³⁷ We did not consider the TDF during the analysis because we anticipated that the identified determinants extended to organisational factors and innovation design in addition to individual behaviours.²¹ For data from the existing surgical scheduling system, inductive coding was used to identify emerging themes on enablers, challenges and recommendations. Challenges and enablers were defined as any factors, characteristics or beliefs that either impeded or facilitated the implementation of existing or proposed new scheduling systems and their processes. A two-stage coding approach was applied to analyse data on the new surgical scheduling system. First, we conducted deductive coding using a codebook based on the CFIR.²⁸ Second, inductive coding was used to identify additional themes across the CFIR domains.³⁸ Codes were initially categorised into four segments—challenges, enablers, recommendations and descriptive segments—using a Microsoft Excel spreadsheet. Data on the existing surgical scheduling system at the Sunnybrook Holland Centre were mapped onto a figure depicting this scheduling continuum (figure 1). The continuum highlights six main steps consisting of the prescheduling consultation (steps 1 and 2), postscheduling (steps 3–5), day of surgery (step 6) and an evaluation process (steps 1–6) to track metrics to assess OR efficiency, inform guidelines and drive quality improvement initiatives.

To ensure analytical rigour, the first three transcripts were independently analysed by three reviewers (IH, DM, JM) to develop the codebook. For subsequent transcripts, one reviewer coded the transcripts, while other reviewers audited all of the primary reviewer’s codes. Text fragments were coded multiple times as necessary to ensure comprehensive coding. All codes were reviewed for consistency, and any disagreements were resolved by a third researcher (MK) and through team discussions. To maintain researcher reflexivity, interview moderators kept notes to document personal reflections and potential biases. An audit trail of conversations and analytical decisions was also maintained. Patients or community members were not involved in the analysis of this study.

Data synthesis

To synthesise the data, team members (DM, IH, MK) organised qualitative data on challenges, related recommendations and enablers into tables. To synthesise the qualitative data on the existing scheduling system, we mapped challenges, enablers and recommendations onto an end-to-end surgical scheduling continuum. This continuum spans prescheduling consultation and referral, waitlist entry and triage, booking and schedule construction, day-of-surgery activities and post hoc review and evaluation. We depicted this continuum in figure 1 and indicated where specific issues occurred at each step to illustrate how local determinants cluster along the pathway. After organising coded data into tables, team members collaboratively reviewed themes for coherence, overlap and alignment with the study objective. Representative quotes were selected to illustrate key themes. Descriptive data were organised into a continuum to reflect the existing surgical scheduling processes. Quantitative data from demographic survey responses were analysed using descriptive statistics (frequencies, means, ranges) in Microsoft Excel.

Results

Table 1 presents the demographic characteristics of the 17 participants, who were primarily female (53%), identified as white (71%) and women (53%) and aged 51–60 years (47%). 47% of participants held their positions at the Holland Centre for five or more years with the following primary roles in surgical scheduling: administrative staff (41.2%), decision-makers/leadership (35.3%) and orthopaedic surgeons (23.5%). Our findings were organised according to identified challenges and enablers of (1) the existing surgical scheduling system, including its related processes, and (2) the proposed new system, including its design and implementation.

Table 1. Participant characteristics^*.

Characteristic		n (%) n=17
Age range (years)
	30–40	3 (17.7)
	41–50	4 (23.5)
	51–60	8 (47.1)
	>60	2 (11.8)
Biological sex
	Female	9 (52.9)
	Male	8 (47.1)
Gender identity
	Woman	9 (52.9)
	Man	8 (47.1)
Ethnic or racial background†
	White	12 (70.6)
	Chinese	2 (11.8)
	Black	1 (5.9)
	Latin American	1 (5.9)
	South Asian	1 (5.9)
	European	1 (5.9)
Primary role in surgical scheduling
	Administrative staff Includes surgical schedulers, OR booking clerks and charge nurses. They coordinate elective surgical case bookings or monitor the schedule, ensuring efficient resource utilisation and successful execution.	7 (41.2)
	Leadership/decision-makers Division leads, managers and decision-makers who drive volume targets and quality improvement. They oversee the master surgical schedule, allocate OR time, establish policies and manage performance reporting.	6 (35.3)
	Orthopaedic surgeons Provide patient care and collaborate with surgical schedulers to advise on case prioritisation, timing and complexity.	4 (23.5)
Years in current role
	0–5	9 (52.9)
	6–10	3 (17.6)
	11–15	2 (11.8)
	>20	3 (17.6)

^*Percentages may not add to 100% due to rounding.

^†One participant self-identified as white and European.

OR, operating room.

Existing surgical scheduling system and processes

We developed three themes related to the existing Holland Centre surgical scheduling system: (1) system functionality, (2) process-related factors and (3) resource-related factors. We describe the challenges and enablers across these themes in the context of the scheduling process steps that we identified in the existing surgical scheduling continuum (table 2; see more details in online supplemental online supplemental tables 2-4).

Table 2. Summary of challenges and enablers across the process steps of the existing surgical schedule system.

Context (surgical scheduling process steps)	Challenge themes: subthemes	Main challenge themes	Main enabler themes
Master grid set-up and access (steps 1–2)	Functionality related: system usability Process related: schedule coordination; users’ scheduling skills	Reliance on Word and manual processes; inconsistent communication of closures; block allocation constrained by limited OR time, staffing and historical patterns rather than optimal use.	Centralised control of master schedule edits supports version control and organisation; use of personal scheduling tools assists allocation.
Prescheduling and waitlist management (steps 1–2)	Functionality related: priority and surgical codes Process related: process adherence; schedule coordination	P1–P4 urgency codes lack granularity; inconsistent waitlist updating; absence of central intake leads to variable wait times across surgeons.	None reported.
Booking on grid (step 3): prediction and information	Functionality related: predicted surgery duration; access to scheduling information	Prediction algorithm uses outdated, averaged historical data; does not incorporate team efficiency, turnover, teaching, anaesthesia input or detailed patient risk; limited access to timely data on beds, case durations, teaching status, cancellations and equipment.	System provides core information (wait times, preferences, history, equipment, pre-op status, Bayview schedule, case notes) and allows documenting case-specific needs.
Booking on grid (step 3): codes, usability and data quality	Functionality related: surgical codes and priority; system usability Process related: data quality and process adherence	Limited, outdated procedure codes; key factors (risk profile, BMI, insurance, revision type, history) not reflected; perceived miscoding to increase booked cases; PICIS is slow, text heavy and cumbersome; schedule sharing relies on printing; data may be inaccurate, contributing to delays, overtime and cancellations.	System flags some equipment conflicts; experienced AAs and schedulers work around system limitations and troubleshoot issues.
Booking on grid (step 3): coordination and resources	Process related: schedule coordination; human and physical resources	Late scheduling reduces flexibility; increased workload with electronic and 7-day scheduling; complex case mix; manual assessment of bed needs; limited relationships and equipment (eg, imaging, power tools) constrain scheduling; limited overtime flexibility.	Longer operating days can support more efficient OR use.
Schedule review and finalisation (steps 4–5)	Process related: schedule coordination	Specialised equipment needs may be reported late; difficult to refill last-minute cancellations.	Strong culture of high OR utilisation; unbooked blocks reallocated ≥10 days in advance; multiple pre-op reviews and experienced staff support daily schedule adjustments.
Day-of-surgery execution (step 6)	Functionality related: surgical codes Process related: process adherence; schedule coordination; human resources	Inability to correct codes retrospectively; lateness and poor practices cause delays; wrong codes lead to wrong equipment; anaesthesia delays, overtime, recovery bottlenecks and late anaesthesia changes drive cancellations; staffing shortages (nurses, cleaning staff) slow turnover.	None reported.
Evaluation and cross-cutting workflow (steps 1–6)	Evaluation related: metrics and reporting Process related: schedule coordination and information flow	Limited, fragmented metrics; manual data extraction; poor interoperability and reporting; key drivers of OR time not captured; heavy reliance on manual tracking and multiple data sources across the pathway.1	Some OR utilisation, costing and item-usage reports are available; existing processes and central contacts provide a practical gateway to perioperative planning and troubleshooting.

AAs, administrative assistants; BMI, body mass index; OR, operating room.

System functionality challenges included reliance on outdated or fragmented tools (eg, a Word-based master grid, a slow DOS-based OR Manager and multiple disconnected platforms), limited and non-granular priority and procedure codes and a prediction algorithm that does not adequately reflect recent data, case complexity, teaching or patient risk. At the same time, the current system provides useful features, such as documenting patient preferences and case-specific needs, as well as basic reporting on OR utilisation.

Process-related factors reflected how existing workflows and practices shaped scheduling. Participants described inconsistent communication of OR closures, variable waitlist management and process adherence across offices, late bookings and cancellations and coordination difficulties across campuses and services, all of which reduced flexibility to optimise OR time. Enablers in this area included centralised control of master schedule edits, use of personal scheduling tools by key knowledge users, proactive reallocation of unused OR blocks and multiple presurgical reviews that helped identify issues before the day of surgery.

Resource-related themes cut across steps and highlighted the impact of constrained physical and human resources on scheduling performance. Limited availability of specialised beds and equipment, constrained overtime and shortages in nursing, cleaning and anaesthesia staff contributed to delays, slower room turnover and cancellations, particularly for complex or short-day schedules. Participants emphasised that any new system must be sensitive to these constraints, noting that longer OR days, better alignment of staffing with case types and more deliberate planning around bed and equipment availability could enable more efficient use of existing capacity.

Proposed new scheduling system design and implementation

The following four CFIR domains represented the themes that we developed from our data: (1) innovation, (2) inner setting of the implementation, (3) implementation and (4) individual involved in implementation (table 3; for more details, see online supplemental tables 5-8.

Table 3. New surgical scheduling system: summary of challenges and enablers mapped to the CFIR domains.

CFIR domain	Construct definition	Main challenge themes	Main enabler themes
Innovation	Innovation design (how the system is designed and presented)	Limited granularity in urgency coding; inconsistent waitlist practices; inaccurate or incomplete case-time prediction inputs; manual, fragmented scheduling and equipment workflows.	Opportunity to redesign scheduling around more granular data, clearer workflows and automation of routine tasks (eg, equipment selection, bed needs).
	Innovation complexity (perceived complexity and number of steps)	Concern about notification overload, overflagging of complex cases and additional steps that may increase workload.	None explicitly reported; implied need to streamline notifications and preserve familiar features.
	Relative advantage (advantages over current system)	Current system perceived as ‘good enough’; uncertainty that a new system can improve throughput under resource constraints; risk that change effort may not yield meaningful gains.	Anticipated improvements in efficiency, throughput, workload, reporting, resource allocation and revenue if key features (eg, automatic tracking, better data) are realised.
	Evidence base (perceived strength of evidence)	Scepticism about automation and ML based on mixed experiences in other hospitals; desire to see local proof of concept.	Confidence drawn from literature and other sectors’ use of ML; perception that greater automation is necessary and timely.
	Adaptability (fit with local needs and constraints)	Uncertainty about how the system will handle site-specific protocols, resource constraints, teaching demands and patient-specific factors; dependence on variable human data entry.	Adaptability viewed as both essential and feasible if the system supports local tailoring and override capability.
Inner setting	Access to knowledge and information (training/support)	Risk of insufficient or poorly timed training; variable digital skills; potential early system glitches; slow IT support.	Strong appetite for role-specific, multimodal training and responsive support.
	Structural characteristics—work infrastructure	Unclear future roles and responsibilities for schedulers; concern that workload may increase without added staffing.	None explicitly reported.
	Structural characteristics—IT infrastructure	Hybrid paper/electronic systems, inconsistent data quality and uncertainty related to upcoming hip replacement.	None explicitly reported; implied that better integration could improve access and performance.
	Available resources—funding	Uncertain funding for implementation and ongoing oversight; possible need for additional position(s).	None explicitly reported.
	Tension for change (need for change)	Some feel the current system works well; concern that change could worsen performance.	Others see clear room for improvement and would welcome enhancements.
	Incentive systems	Surgeons may resist if assistants’ workload increases, especially when they fund these roles.	None explicitly reported.
	Mission alignment	Past efficiency initiatives seen as undermining work–life balance and teaching; fear this may recur.	Holland Centre seen as open to innovation and strongly focused on scheduling and efficiency.
	Compatibility and relative priority	Concerns about missing features and misfit with existing workflows; some see low priority if current system already optimises OR time.	Holland Centre viewed as an ideal starting site due to case homogeneity, high utilisation and perceived potential for even modest efficiency gains.
Implementation	Doing (approach to rollout)	Risk of top-down implementation without early user input, leading to late discovery of problems.	Preference for phased rollout, pilots and iterative testing with early and ongoing KU engagement.
	Reflecting and evaluating	Unclear plans for monitoring, sustaining use and validating predicted times.	Recognition of the need for ongoing measurement and feedback to refine the system.
Individuals	Motivation	Anticipated resistance and scepticism due to past consultant-led ‘efficiency’ projects and perceived loss of control; uncertain buy-in from surgeons and anaesthetists.	Many KUs are pro-change and supportive if engaged, able to retain some flexibility and see clear rationale and benefits.
	Capability	Potential variability in technology comfort.	Overall confidence that KUs have, or can easily acquire, the skills needed to use the new system.

CFIR, Consolidated Framework for Implementation Research; IT, information technology; KU, knowledge user; ML, machine learning; OR, operating room.

Across CFIR domains, participants highlighted both opportunities and challenges for implementing an automated surgical scheduling system. Within the innovation domain, priorities centred on usability, interoperability and accurate case-time estimation. While participants identified clear potential benefits such as improved efficiency and reduced workload, concerns about complexity, notification burden and the strength of the evidence base contributed to mixed perceptions of relative advantage. Flexibility and the ability to override automated decisions were viewed as essential for real-world applicability. In the inner setting CFIR domain, successful implementation was seen as dependent on robust training, reliable information technology (IT) support and alignment with existing workflows and organisational priorities. Participants noted structural and resource-related constraints, including data quality issues, hybrid systems and uncertainty around funding and staffing. Readiness for change varied, though the organisation was generally perceived as innovation oriented. For the implementation domain, participants strongly recommended a phased, participatory approach, including pilot testing, parallel system use and early engagement with key interest-holders. Ongoing evaluation, feedback loops and clear governance structures were considered critical to ensure iterative improvement and accountability. Finally, within the individuals domain, motivation to adopt the system was mixed, shaped by prior experiences and trust in the innovation. However, most participants expressed confidence in their ability to use the system, suggesting that capability is unlikely to be a major barrier with appropriate support.

Discussion

This study aimed to understand existing (current) surgical scheduling processes and identify the challenges, enablers and knowledge users to support implementation of a new, ML-enabled system at a high-volume elective orthopaedic surgery centre. Our findings reveal both challenges in the existing scheduling system and broad, though varied, support for the new. Participants noted significant inefficiencies in the manual scheduling process, citing outdated software, inconsistent communication and resource limitations. The existing system’s shortcomings included inaccurate predictions of surgery times and limited access to scheduling information, resulting in cancelled procedures and underused ORs. Participants generally supported the ML-based innovation, anticipating improvements in throughput, accuracy and resource allocation; however, concerns about system flexibility, complexity and user training persisted. Interest-holders emphasised the importance of responsive training, a phased rollout, ongoing user feedback and strong leadership engagement for a successful implementation. These insights will directly inform the design and phased implementation of this automated scheduling system, which we believe will become the standard for surgical scheduling.

Consistent with emerging literature, participants perceived that an ML-based system could improve scheduling accuracy and hospital efficiency, reduce overtime and optimise resource utilisation, particularly for elective surgeries such as total knee and hip arthroplasties. However, our study did not evaluate implementation outcomes or patient-level effects. Our conclusions, therefore, focus on perceived value and the implementation conditions that interest-holders deem necessary for such a system to deliver on its potential. ML models have demonstrated greater accuracy in predicting surgical case durations and optimising overall schedules compared with traditional methods. For example, Lex et al and Abbas et al showed that ML-driven approaches reduced postoperative wait times and increased throughput for orthopaedic cases.1439,42 Evidence also indicates that ML models provide substantially improved predictions of case duration compared with manual systems.⁴³ However, other studies emphasise the importance of integrating AI recommendations (ie, algorithmic decision support) with human oversight and allowing for adaptability to local needs, as hybrid systems appear to be most effective in enhancing staff satisfaction and transparency. AI systems need workflow customisation and override options to maximise trust in these systems. Implementation science research further highlights the critical need for iterative, context-responsive planning, customisation (eg, tailored training) and multilevel organisational engagement to ensure successful rollout and uptake. For example, Dryden-Palmer et al described the importance of piloting innovations in cycles, gathering feedback and adjusting implementation strategies to local needs.⁴⁴ Barriers such as user resistance, workflow disruption and insufficient organisational buy-in can limit the successful adoption of these digital solutions.^{44 45} For example, user reluctance, cost-related resistance, workflow incompatibility and limited uptake of automation in healthcare organisations can be major barriers to digital scheduling interventions. Dryden-Palmer et al emphasised the critical role of local leadership engagement and organisational culture, noting that insufficient buy-in and resistance within teams can undermine the implementation of complex hospital innovations.⁴⁴ In fact, the lack of organisational buy-in is frequently cited in reviews of the adoption of automated scheduling in hospitals.⁴⁶ Overall, our findings are consistent with other studies, suggesting that engaging relevant knowledge users, leadership facilitation, phased implementation and transparent evaluation of system performance are essential to implementing best practices for translating innovations, such as our surgical schedule system, within complex hospital environments.⁴⁵

Targets for the ML-enabled system, organisational change and structural constraints

Our findings suggest that not all problems in the current scheduling process are equally amenable to technological solutions. Several challenges appear directly addressable through the design and configuration of the ML-enabled system itself. These include fragmented access to scheduling information, a lack of granular procedure codes, manual and duplicative data entry and the absence of automated support for equipment selection and reporting. Designing the system to integrate with existing clinical and perioperative systems, streamline data entry and provide real-time visibility into schedules and waitlists may directly mitigate these issues.

Other problems primarily require organisational and workflow changes. For example, inconsistencies in how patients are entered onto the waitlist, variation in booking practices across surgeon offices, unclear governance over flow and prioritisation decisions and communication gaps between clinics, surgeon offices and the OR will not be resolved by the technology alone. Addressing these issues will require building shared protocols, clarifying roles and responsibilities and establishing mechanisms for oversight and feedback. Finally, some constraints are structural and outside the scope of a scheduling tool, such as chronic shortages of nursing and anaesthesia staff, restrictions on overtime, limited availability of specialised beds and equipment and broader funding policies. Participants were realistic about these limitations and emphasised the need for flexibility in the system (eg, override options, scenario planning) to work within these constraints. Recognising which determinants are technical, organisational or structural can help decision-makers align expectations and pair the ML-enabled system with appropriate non-technical interventions.

Implications for implementation and transferability

Our findings highlight several implementation conditions that are likely to be relevant across surgical centres, even though the study was conducted at a single elective orthopaedic site. Many of the challenges identified—such as reliance on manual, fragmented tools; limited interoperability between clinical and scheduling systems; inconsistent waitlist entry and review; and constrained physical and human resources—are not unique to our institution and have been described in other surgical contexts.47,49 These issues represent common targets for ML-enabled scheduling systems and should be explicitly assessed when similar tools are considered elsewhere. The use of the KTA model (assess barriers to knowledge use step) and CFIR offers a practical structure that other centres can apply. Locally, teams can map their current scheduling pathway, identify determinants across CFIR domains (eg, innovation characteristics, inner setting, individuals, implementation process) and use these findings to define context-appropriate implementation strategies. For example, our results suggest that role-specific training, early and visible leadership engagement, phased pilot implementation, mechanisms for rapid feedback and system modification and clear governance over waitlist processes are likely to be important in many settings. At the same time, we recognise that certain aspects of our context are specific to a high-volume elective orthopaedic centre in a publicly funded system. We therefore encourage readers to adapt our framework and determinants to their own organisational structures, information systems and resource environments rather than mechanically replicating our approach.

Our findings provide a practical framework that other surgical programmes can apply when planning ML-enabled scheduling systems. First, institutions can map their local scheduling processes onto the six-step continuum we describe, using this as a structured starting point to surface where information handoffs, manual workarounds and decision points occur. Second, our CFIR-informed determinants highlight a set of challenges that are likely common and transferable to other surgical settings as well as enablers that can be deliberately leveraged, including centralised scheduling expertise, phased implementation and responsive training and IT support. In contrast, some barriers (such as specific staffing models, local bed availability and cross-campus arrangements) are context dependent and may differ in other organisations. Third, the determinant tables can be used as a checklist to guide site-specific implementation planning, for example, by specifying which stakeholder groups to engage, what data infrastructure is needed and how to structure pilot cycles and feedback loops prior to full rollout. We therefore view our process map and determinant tables as adaptable tools to support local mapping and planning, rather than as a prescriptive model.

Strengths and limitations

Major strengths of this study include its interest-holder engagement, which captures perspectives across scheduling, clinical and leadership roles. Our qualitative description design was supported by implementation science frameworks (TDF, CFIR and the KTA model), which enabled an in-depth, context-specific examination of scheduling processes and implementation determinants and provided structure for data collection and analysis. This methodology enabled us to remain close to participants’ accounts while systematically examining these determinants at multiple levels. By employing qualitative description grounded in implementation science theory, the research captures nuanced perspectives essential to real-world implementation. Our study also had some limitations. Our work was conducted at a single site, which may limit generalisability to other hospital contexts. However, the underlying implementation determinants and strategies are relevant to other high-volume surgical centres considering similar AI-enabled scheduling innovations. While we captured a range of professional roles, we did not include patients or community representatives. Additionally, some system challenges identified by our participants lacked corresponding recommendations, highlighting areas for further exploration and research. Upcoming changes, including the implementation of new hospital information systems and expanded weekend scheduling with surgeons from other hospitals, will require additional evaluation to ensure successful implementation and scale-up.

Future directions

Future research should focus on implementing and piloting the proposed scheduling platform, evaluating its impact (via shadow and pilot studies including technical evaluations) and adaptation in dynamic hospital environments and incorporating diverse user and patient perspectives to achieve a comprehensive impact assessment. This should be done in a staged manner, ideally through an initial shadow study of the new system, followed by a pilot implementation. Additional technical infrastructure will be required to support this process.

Conclusions

This study demonstrates that in the context of significant process and resource limitations, knowledge at a high-volume orthopaedic centre expressed substantial support (though not universal) for the introduction of an automated ML-driven scheduling system. Successful implementation will require robust engagement with schedulers, clinicians and leadership; careful attention to training and workflow integration; and iterative evaluation of system performance and impact. The anticipated benefits for OR efficiency and wait times will need to be confirmed in future quantitative and mixed-methods evaluations.

Data availability statement

All data relevant to the study are included in the article or uploaded as supplementary information.