Total knee arthroplasty (TKA) is a well-established and effective treatment for end-stage osteoarthritis [1]. The number of TKA procedures performed is increasing annually, both locally and globally. In both Australia [2] and the United States [3], it is projected that the number of TKAs undertaken will rise by 276% and 181%, respectively, by 2030, placing an increased burden on health care services. Surgical success rates are high and are currently based on a range of objective medical outcomes including survival, prosthesis performance, and revision rates [4]. However, despite this, there are consistent reports that up to 20% of patients report dissatisfaction with the outcomes of the procedure, reflected by enduring pain, limited function, and diminished quality of life [5-7].

Rehabilitation is an important factor that improves recovery from this surgery. However, no clinical guidelines are currently available to inform the recovery pathways. This is because there is no evidence to suggest who may benefit from rehabilitation, when and where this service should be delivered, and what services should be provided. Consequently, some patients receive inpatient (7-10 days) rehabilitation or at home rehabilitation through exercise monitoring (up to 10 weeks), whereas others are not referred to any form of rehabilitation, depending on the surgeon and the clinic. This inconsistent approach to rehabilitation may be because of the general lack of data demonstrating the effectiveness of any specific pathway. For example, a recent study demonstrated no difference in patient mobility at 26 weeks after the surgery when comparing inpatient rehabilitation with home-based monitoring [8]. Furthermore, a systematic review concluded that the setting did not significantly influence the outcomes [9].

During rehabilitation programs, the primary emphasis is on assisting patients in regaining muscle strength, enhancing joint stability, improving range of motion, restoring neuromuscular function, and recovering proprioception [10]. The approaches taken and outcomes vary widely among therapists, and to date, no clear guidelines exist for the most effective therapeutic strategies to be applied [9]. In addition, several psychological factors have also been reported to be linked to recovery following knee replacement or have the potential to impact recovery [11,12]. This includes psychological “vulnerability factors,” such as depression, anxiety [13-15], and pain catastrophizing [16,17], and “protective factors,” including resilience and committed action. However, despite their potential impact on recovery, quality of life, and satisfaction [12], these factors are not routinely reviewed in patients before surgery and remain overlooked in rehabilitation referral and ongoing clinical decision-making.

Determining whether rehabilitation programs are effective is also complicated by the use of inconsistent measures of “success.” Many studies to date have used a single traditional outcome measure, such as a physician’s or patient’s report of functional mobility, pain, or quality of life [18]. Although surgical success has been characterized by a number of objective medical outcomes, including survival, prosthesis performance, and revision rates [4], there has been increasing interest in considering and combining longer-term outcomes of success, such as patient satisfaction and social integration, with these other traditional measures. Over the last decade, the integration of patient-reported outcome measures (PROMs) in the evaluation of the success of TKA has become more widespread [19]. However, the impact of PROMs on clinical practices remains limited. Although there are clear benefits of assessing the patient’s perspective of recovery, including their symptoms, functional status, and quality of life [20,21], there remains a lack of consensus regarding the appropriate measures to be used, together with concerns of the cost-effectiveness involved in their application [22].

Regardless of the adopted outcome definition, no consistent set of factors that predict a successful surgery outcome have been identified. Individual studies have identified some factors associated with long-term outcomes. These include BMI, radiological arthritis severity (Kellgren-Lawrence score), mood status (eg, anxiety and depression), quality of life, and functional mobility [18,23,24]. Nevertheless, it is important to note that the collective predictive value of these metrics has not been externally validated, nor has their generalizability been demonstrated across various settings and populations [25].

Mounting evidence suggests that it may be possible to predict, via the use of a model, which patients could benefit from a more multifaceted approach to rehabilitation, which could encompass a broad range of biopsychosocial parameters, and how this can be precisely tailored to address each patient’s individual needs [26]. Importantly, if such a prediction model can be developed and validated, it would be extremely useful to guide the future provision of rehabilitation services to support superior outcomes for each patient. In a recent systematic review, we demonstrated that the number of clinically meaningful, well-validated prediction tools for recovery following total knee replacement is limited [26]. Several clinical and lifestyle factors have been linked to poor outcomes and dissatisfaction following a variety of surgical procedures, including TKA, such as female sex, obesity [27,28], lower social support level [29], and comorbidity status [15]. Indeed, predictive models have demonstrated that age and sex can predict improvement in knee function [30] and satisfaction [31]. However, few studies have reported the inclusion of psychological parameters in their model development. Moreover, many of the models developed have limited predictive success and lack external validation, which limits their implementation in routine clinical use. There are also constraints when predictive models repurpose existing data sets owing to the limitations of data quality and completeness [32].

To date, there is an unmet need for the development of a clinical tool based on a comprehensive range of parameters that can assist in determining individual recovery trajectories following TKA, which can be easily adopted in the clinical setting. Through the identification of factors influencing patient recovery trajectories across a range of success outcomes, rehabilitation pathways will be better informed regarding the tailoring of services to meet patients’ individual needs. This has the potential to improve long-term outcomes and satisfaction by addressing a broader range of pre- and postoperative parameters.

The major aim of this study is to develop a clinical tool that can be used in the prediction of recovery trajectories, which can inform rehabilitative services regarding the nature of potential interventions and the timely delivery of the right services to each individual patient. Our approach will be to look beyond the impact of biomedical features alone on patient outcomes by adopting a biopsychosocial approach. The biopsychosocial model of health systematically considers biological, psychological, and socioenvironmental factors and their complex interactions in understanding illness, health, and health care delivery. It encompasses a more holistic approach to understanding illness and wellness than a purely biological vision. The tool will be constructed by assessing the predictive capacity of a comprehensive battery of biopsychosocial factors on a number of patient outcomes following TKA, including improvements in knee symptomology, quality of life, and satisfaction. This approach has the potential to facilitate the allocation of rehabilitation resources toward a more holistic approach to recovery following surgery and could improve the effectiveness of TKA across a range of success outcomes, including patient satisfaction. By supporting the delivery of rehabilitative interventions that are precisely targeted to patient needs, this clinical decision-making tool could also improve the efficiency of health care resource delivery.

This study was undertaken with patients who underwent TKA surgery between November 2019 and June 2022 across 4 participating Ramsay Health Care facilities in New South Wales (NSW), Australia, including Lake Macquarie Private Hospital, Gateshead; Kareena Hospital, Caringbah; Baringa Hospital, Coffs Harbour; and Wollongong Private Hospital, Wollongong. Surgical procedures were performed by 11 collaborating orthopedic surgeons.

This study had a prospective observational design. Participants were involved in the study for a duration of 4 months, with data collection taking place at 2 time points during this period: within the month preceding their scheduled TKA and 3 months (12 weeks) following the TKA procedure. Figure 1 outlines the participants’ timeline for this study and the number of participants recruited and excluded at each stage.

All patients identified by the participating orthopedic surgeons as requiring a total knee replacement and who fulfilled the eligibility criteria were invited to participate in the study. Patients were included as eligible if they were aged ≥18 years, could complete the PROMs questionnaires, and confirmed to require a primary total knee replacement (bilateral or unilateral procedure). Patients were excluded from participating if they were receiving a primary total knee replacement owing to trauma, had undergone previous knee surgery within the past 6 months, were planning to undergo additional knee surgery within the next 12 months, or did not have the capacity to provide informed consent to the research.

For machine learning modeling, the standard required sample size used should ensure at least 10 events for each included predictor parameter [33]. In this study, we have a total of 39 input variables and 76 input events when applying the conversion of categorical variables to numerical data (one-hot-encoding). Applying the standard approach, we would require 390 participants for models without one-hot encoding and 760 participants when using one-hot encoding.

Recently, Riley et al [34] have developed a more refined calculation of the required sample size, which depends not only on the number of candidate predictor parameters but also on the total number of participants, the outcome proportion (incidence) in the study population, and the expected predictive performance of the model. Applying the method proposed by Riley et al [34] to our classification modeling with an aim of 90% accuracy and an outcome proportion of 20% (ie, 20% of participants were not expected to meet the minimal clinically important difference [MCID] improvement threshold), a predicted R² of 0.5, and shrinkage of 0.85, returns a required sample size of 864.

In this study, we aimed to recruit at least 1000 patients undergoing total knee replacement. This will ensure sufficient data for inclusion in our machine learning predictive modeling analyses and allow for 13.6% (136/1000) of data to be nonevaluable owing to missing data fields.

Eligible participants were contacted by the study team through a follow-up phone call made after they received the study Patient Information Sheet from their surgeon. Eligible participants who chose not to participate when they received the Patient Information Sheet from their surgeon were not contacted. If the participant verbally agreed to participate in the study, they were provided access to study materials either on the web or by post, and if they consented to undertaking physical assessments of mobility via a video call, a time for this appointment was made at a time convenient to the participant.

If patients opted to have their study questionnaires accessed via the study-specific web-based platform, they were provided with a copy of the Patient Information Sheet when they accessed this system via a participant-specific log-in. Following a review of the Patient Information Sheet, they were asked to acknowledge their consent to participate in the study and provide electronic consent on the Electronic Letter of Consent Form. Questionnaire templates were not accessible to the participants until the consent field on this form had been completed. Participants also had the option to consent to all study assessments being undertaken or to only complete the study questionnaires.

If patients opted to have their study questionnaires posted to them, a copy of the Patient Information Sheet and Letter of Consent were included in the posted package. The participants were asked to sign the consent form before completing the questionnaires. The completed consent form and questionnaires were subsequently posted back to the study team.

Participation in this study was voluntary. Patients had the option to decline participation at any time point during the study and were informed that this would not impact their ongoing medical care. If patients decided to withdraw from the study, they could specify whether their data were to be removed from the study or if they were willing for it to be used as part of the analysis.

This research methodology was peer-reviewed and approved by the School of Medicine and Public Health at the University of Newcastle, NSW, Australia, in accordance with the Australian Code for the Responsible Conduct of Research. This study was conducted in accordance with the National Statement on Ethical Conduction in Human Research (2007). Ethics approval was granted by the University of Newcastle Human Research Ethics Committee (approval H-2019-0109) on June 21, 2019. Approval for each study site’s participation was reviewed and approved by the Ramsay Research and Governance Office in NSW, Australia (RHC RG-2019.008).

This study responds to an unmet clinical need to address factors that can improve outcomes in patients undergoing total knee replacement. With a substantial increase in the number of TKA procedures expected to take place over the next decade, there is the potential for a significant number of patients to be dissatisfied with their surgical outcomes, unless the issue can be effectively addressed.

We anticipate the development of a clinical predictive tool that has the potential to improve patient-centered TKA outcomes. The uniqueness of the model being used is the inclusion of a comprehensive assessment of modifiable psychological, resilience, and clinical factors that have been linked to TKA recovery outcomes.

The approach adopted in this protocol builds on existing research demonstrating the involvement of factors influencing recovery of patients following total knee replacement and the inclusion of these factors in predictive modeling of postsurgery outcomes. We are unaware of any published predictive models of patient outcomes following TKA that include such a comprehensive battery of biopsychosocial parameters as those being evaluated in our study protocol. Our systematic review of predictive models for TKA outcomes [26] and those undertaken by others [32] revealed that the most consistently used presurgical parameters included clinical and medical factors such as age, sex, BMI, and concomitant health conditions, whereas psychological factors were incorporated to a lesser extent and, when applied, were restricted to measures of mood status (depression and anxiety) or mental health quality of life. There have been several specific modifiable psychological factors, including pain catastrophizing [18,62] and resilience [63], which have been associated with recovery following TKA and could inform the optimization of patient outcomes. However, their predictive capacity for patient outcomes after TKA has not been widely explored.

Previous limitations of predictive models are associated with using data derived from existing registries or retrospectively collected data sources [26]. Such approaches can be problematic because of the need for data imputation caused by missing values or limitations in predicting specific outcomes owing to insufficient presurgery variables collected. The prospective collection of specific variables of interest, as we propose in this study, enables the collection of specific input data that have been associated with specific outcomes. This approach has the potential to inform appropriate rehabilitation interventions to improve patient outcomes.

If successful, this project will provide a tool that can help guide health professionals with respect to comprehensive rehabilitation decisions individualized to each patient. The tool will underpin and support the identification of biopsychosocial risk factors for poor postoperative holistic outcomes and facilitate the development of biopsychosocial prehabilitative and rehabilitative interventions, which could lead to improved overall outcomes. In addition, the tool will enable efficient patient stratification according to identified individual needs, which allows a vastly more targeted service provision, maximizing the potential for impact. Currently, patients are infrequently evaluated using formal, standardized assessments for the type of rehabilitative service they would most likely need and benefit from. This project will provide a means of stratification of patients in a clinical setting, identification of “at risk” individuals early in their recovery journey, and support appropriate follow-up services to be implemented.

The scope of the factors influencing decision-making will be evolved. Typically, when patient information is collected, it generally focuses on selected medical or targeted joint parameters [64] and does not take consider parameters such as patient anxiety, stress resilience, quality of pain, social support, and sense of purpose, all of which have recently been shown to be associated with patient outcomes [12]. The model used in this study will explore the influence of these additional parameters on recovery and well-being following TKA, thereby providing evidence for a more comprehensive therapeutic strategy.

In this study, we are collecting data prospectively, which enables the scope of features that may impact patient outcomes to be extended beyond that routinely collected in existing data sets and provide a focus on the biopsychosocial parameters of interest. The development of a clinical decision tool that is focused on a broader scope of patient features affords the opportunity for a more holistic approach to be taken in rehabilitation programs that extend beyond physical recovery strategies alone. The design of the study has attempted to reduce surgeon and hospital bias by including patients who underwent TKA by 11 different orthopedic surgeons across 4 different hospitals. These facilities were located across a large geographical range and enabled the inclusion of participants residing in metropolitan and rural communities to take part in the study. This was further enhanced by the capacity to collect data on the web and via video calls, which enabled the study to be managed at a central location and include participants across a large geographical area.

Although the planned study will evaluate the predictive capacity of a comprehensive battery of biopsychosocial parameters on TKA patient outcomes and offer significant contributions to the existing literature, there are potential limitations. In this study, we are assessing the predictive capacity of presurgical parameters on recovery outcomes assessed at 3 months after TKA. This may be considered early on in a patient’s recovery journey. It has been demonstrated that the maximal changes in patient-reported knee symptomology occur rapidly within the first 3 months, with subsequent improvements continuing, albeit at a slower rate, up to 12 months after surgery [65,66]. It is possible that the impact of psychological factors on patient outcomes may differ at later time points, as patients’ circumstances and needs change, and as such, predictive modeling developed in this study may be limited to early recovery outcomes.

We recognize that our study protocol does not encompass all factors that could potentially influence outcomes after TKA. Our analyses do not include surgical factors known to impact outcomes, including factors such as the surgeon or hospital, prosthesis type, and surgical procedures [67], and radiological factors, such as the Kellgren-Lawrence score [23]. Therefore, it is important to note that our predictive models are unlikely to account for 100% of the variability in patient outcomes that we observe. Instead, they will provide insight into the extent to which the presurgical biopsychosocial factors assessed in our model contribute to the observed variance.

As per the data analysis plan for this study, a rigorous internal validation of the predictive models generated will be performed. Ongoing external validation of the models will also enable the suitability and specificity of the findings to be applied to a wider clinical setting than that included in this study. In addition, through future engagement with key stakeholders, including clinical care teams (surgical and rehabilitation specialists), health care providers, and patient advocates, we will develop the predictive models to optimize their clinical implementation and usefulness.

The classification models proposed will facilitate the evaluation of intervention studies, whereby biopsychosocial risk factors are identified, targeted interventions are provided, and the benefits to postoperative outcomes are evaluated. This will enable the determination of whether targeting these factors in an individualized manner leads to improved satisfaction and PROMs.

In conclusion, the findings of this study will contribute to the development of a comprehensive biopsychosocial clinical decision–predictive tool. Such a tool will have the potential to inform clinical care teams regarding individual patient outcomes, enable the stratification of potentially “at risk” individuals, and inform the evolution of rehabilitation services through the identification of potential interventions, all of which could lead to an improvement in patient outcomes and satisfaction following TKA.