Predicting early return to sport after periacetabular osteotomy

Abstract

Aims

Periacetabular osteotomy (PAO) is the primary surgical treatment for developmental dysplasia of the hip (DDH), providing considerable pain relief and improved joint function. Return to sport (RTS) is a key outcome for young, active patients. This study aimed to identify preoperative predictors of RTS timing and develop a machine-learning-based prediction model to optimize patient counselling.

Methods

This retrospective analysis of prospectively collected data included 235 patients who underwent PAO between January 2019 and December 2023. Preoperative variables, including demographic, functional, and psychological assessments, were analyzed. RTS was assessed at three, six, and 12 months postoperatively. Logistic regression with recursive feature elimination and a conditional inference tree (ctree) model were used to identify predictors of RTS.

Results

At three months, 102 patients (43%) had returned to sports, increasing to 182 (77%) at six months and 223 (95%) at 12 months. Key predictors of early RTS included the minimally invasive surgical approach, higher preoperative physical activity (≥ two sessions/week), lower anxiety scores, and higher Hip disability and Osteoarthritis Outcome Score (HOOS) pain scores. Male sex and older age were associated with delayed RTS. The ctree model stratified patients based on their likelihood of early RTS, providing an individualized prognosis.

Conclusion

PAO enables early RTS in over 90% of patients within the first year. The use of a minimally invasive approach allowing immediate active hip flexion, higher preoperative activity levels, and lower anxiety scores significantly improves RTS timing. The machine-learning model provides precise, individualized RTS predictions, offering a valuable tool for patient counselling and rehabilitation planning.

Cite this article: Bone Jt Open 2025;6(6 Supple B):33–42.

Introduction

Developmental dysplasia of the hip (DDH) is a prevalent condition that can cause hip pain and functional limitations, particularly in young and active individuals.¹ In distinct cases of DDH, periacetabular osteotomy (PAO) is a well-established surgical intervention to prevent early secondary hip osteoarthritis.^2-4 This procedure aims to correct insufficient coverage of the femoral head restoring hip joint function, and has demonstrated favourable mid- and long-term outcomes.^5-7 Despite the documented success of PAO in preserving joint function, the immediate postoperative phase – characterized by the patient’s recovery speed, pain alleviation, and the ability to resume daily and sports activities – remains a pivotal concern for many patients, especially given the young demographic typically afflicted by DDH.^8,9 The return to sport (RTS) following PAO is therefore a key indicator of success for these patients, as evidenced by numerous studies published on this topic in recent years.^10-16

However, many of these studies present several limitations, including retrospective designs,^10,11,16 small cohort sizes,^10-14 and a focus on specific subgroups.^13,14 Furthermore, due to the retrospective nature of these studies, there is a paucity of information regarding preoperative factors influencing RTS. Consequently, the prognostic value of these studies is limited for a large average cohort undergoing this surgery.

This is of particular importance to patients, as they frequently need to plan their rehabilitation and set realistic expectations for their recovery process. A recent social media study has demonstrated that patients frequently seek information regarding rehabilitation and recovery timelines, underscoring the critical need for clear, evidence-based guidance.⁹ Furthermore, it is of the utmost importance to align patient expectations with clinical outcomes in order to ensure a successful rehabilitation process, as another recent study demonstrated.⁸

The present study aimed to explore preoperative factors that influence the likelihood and timing of RTS. The objective is to develop a prediction model which could serve as a valuable tool for clinicians, enabling them to provide more effective counsel to patients and set appropriate expectations regarding their recovery and RTS. It is hypothesized that machine-learning analysis of preoperative data can generate a predictive model, enabling a personalized prognosis for early or delayed RTS in future patients.

Methods

Study design and data collection

In this retrospective analysis of prospectively collected data, 235 patients undergoing PAO between January 2019 and December 2023 were included from our institutional registry for joint-preserving hip surgery. The study was conducted at a single high-volume centre, and all surgeries were performed by a single surgeon (GIW) with a total of 698 cases operated during the study period.

Institutional Review Board approval was obtained prior to the study’s initiation (BB099/20). Inclusion criteria for this analysis were a diagnosis of DDH (LCEA < 18°), complete written informed consent, and completed preoperative as well as follow-up assessments at three, six, and 12 months postoperatively. To ensure a highly homogeneous study group, numerous exclusion criteria were applied, as detailed in Figure 1. For example, cases were excluded if they involved contralateral PAOs during the study period, which could introduce confounding effects on RTS outcomes, incomplete follow-up data, or indications other than DDH such as acetabular retroversion. Additionally, patients who underwent additional surgeries that could impact recovery and RTS were not included in the final analysis.

Fig. 1.

STROBE flowchart of inclusion and exclusion criteria for the study.

Key demographic and clinical data, including age, sex, height, and weight, were extracted from standard clinical records. As part of our institutional registry for joint-preserving hip surgery, extensive preoperative data collection included assessment of the general activity level using the University of California, Los Angeles, activity scale (UCLA),¹⁷ hip joint functionality by the modified Harris Hip Score (mHHS),¹⁸ and the International Hip Outcome Tool-12 (iHOT-12).¹⁹ Additionally, the Hip disability and Osteoarthritis Outcome Score (HOOS), including its five subscales,²⁰ and the 36-Item Short-Form Health Survey questionnaire (SF-36)²¹ were used to evaluate QoL and daily functionality. To assess preoperative psychological distress, the Brief Symptom Inventory-18 (BSI-18) was used, including its three subscales.²²

Furthermore, the types of sports that patients engaged in prior to surgery were recorded. These were classified as low-impact, high-impact, or ‘no sport’ if the patient was inactive.²³

The frequency of sports participation per week prior to surgery was recorded, offering the following options: 0, 1, 2, 3, 4, or more than 4 sessions per week. The duration of each session was also documented, with options being 0 to 15 minutes, 16 to 30 minutes, 31 to 60 minutes, 61 to 120 minutes, or more than 120 minutes per session.

RTS was evaluated with a binary yes/no response. Patients were classified as having achieved RTS within three, six, or 12 months based on their responses at these intervals. Those who answered in the negative at the 12-month follow-up were classified as not having returned to sport. Postoperative follow-up parameters are included in Supplementary Table i. They were not part of the primary analysis but are provided for transparency and validation purposes.

Surgical details and rehabilitation guidelines

During the study period, a modified, minimally invasive Bernese PAO technique was performed in all patients. This approach involved the use of a bikini incision and one of two surgical techniques: the rectus sparing approach (RS) or the rectus and sartorius sparing approach (RASS). In the RS approach, which included bony detachment of the sartorius muscle from its origin at the anterior superior iliac spine, the muscle was reattached at the conclusion of the surgery. The RASS approach, as previously described, represents a completely minimally invasive technique that spared both the rectus femoris and sartorius muscles.²⁴ The decision between RS and RASS was made intraoperatively based on patient-specific anatomical conditions. In cases where adequate surgical access was available adjacent to the sartorius muscle, the RASS approach was chosen. Conversely, when space limitations made it difficult to safely perform all osteotomies, the RS approach was used. To evaluate whether the selection of surgical approach introduced bias, a comparison was made of baseline demographic characteristics between the RS and RASS. The analysis revealed no statistically significant differences between the two groups with respect to age (31.7 years (SD 9.0) compared with 32.1 years (SD 8.2); p = 0.522, Mann-Whitney U test), sex (98/121 (81.3%) compared with 98/114 (85.9%) female patients; p = 0.438, chi-squared test), or BMI (25.1 kg/m² (SD 4.4) compared with 23.6 kg/m² (SD 4.2); p = 0.181, Mann-Whitney U test), thereby suggesting that the selection of approach was predominantly influenced by intraoperative anatomical considerations rather than by patient-specific factors.

Postoperative care began with immediate physiotherapy and hip mobilization. The initial phase involved 15 kg partial weightbearing for six weeks. RS group patients were treated with a continuous passive motion (CPM) device that allowed controlled, passive mobilization of the hip with no active hip flexion allowed, while RASS group patients were permitted active flexion up to 90° immediately after surgery. At six weeks, following radiograph confirmation of appropriate healing of the osteotomies, patients transitioned to full weightbearing and unrestricted range of motion. Strength and mobility training intensified, with gradual resumption of sports as recovery progressed. By 12 weeks, all patients could fully engage in sports without restrictions, provided they were pain-free and functionally recovered.

Statistical analysis

Descriptive statistics were used to summarize the baseline characteristics of the study population. Means and SDs were reported for continuous variables, while categorical variables were presented as frequencies and percentages. For normally distributed data, the independent-samples t-test was used to assess significant differences between continuous variables. For non-normally distributed data, the Mann–Whitney U test was applied. The chi-squared test was used to analyze differences between categorical variables. For nonparametric paired data, the Wilcoxon signed-rank test was employed. A p-value < 0.05 was considered statistically significant for all analyses.

To identify the most relevant patient individual factors representing predictors of RTS after PAO, logistic regression analyses were performed for RTS within three months and within six months. The initial full models included all available preoperative variables, and recursive feature elimination (RFE) was used to optimize the models by systematically removing less predictive factors. The final statistical models retained only those predictors with significant contributions to the outcome.

Model performance and selection were evaluated using the Akaike Information Criterion (AIC), with lower AIC values indicating a better-fitting model.²⁵ Degrees of freedom for each model were considered to ensure appropriate complexity without overfitting. To identify key predictive factors for RTS, we employed a conditional inference tree (ctree) model, a machine-learning approach that enables unbiased variable selection while accounting for complex, non-linear interactions.²⁶ The algorithm iteratively partitions the dataset based on statistically significant differences in predictor variables, allowing for the identification of distinct patient subgroups with varying RTS probabilities. This hierarchical stratification offers a clinically interpretable framework that facilitates individualized prognosis estimation. This analysis was conducted using the ‘partykit’ package in R (version 4.1.2; R Foundation for Statistical Computing, Austria).²⁷

Results

Study population and timing of RTS

In total, the study included 235 hips in 235 patients with 195 female patients (83%), a mean age at the time of surgery of 31.9 years (SD 8.6; 14.4 to 55.5), and a mean BMI of 24.3 kg/m² (SD 4.3; 17.1 to 43.8). Among the 235 operations, 114 (48.5%) were performed using the RASS approach and 121 (51.5%) using the RS approach.

Prior to surgery, 106 patients (45%) participated in high-impact sports, 83 patients (35%) engaged in low-impact sports, and 46 patients (20%) reported no regular sports participation. Of the 189 patients (80%) who were active preoperatively, 47 (25%) participated in an average of one session per week, 43 (23%) in two sessions, 39 (21%) in three sessions, 33 (17%) in four sessions, and 27 (14%) in more than four sessions per week. In terms of exercise duration, 41 patients (21%) reported sessions lasting up to 15 minutes, 39 patients (21%) had sessions between 16 and 30 minutes, 42 patients (22%) between 31 and 60 minutes, 41 patients (22%) between 61 and 120 minutes, and 26 patients (14%) reported sessions lasting longer than 120 minutes.

Preoperative patient-reported outcome measures (PROMs) showed a mean UCLA activity score of 6 (SD 2.4), a mean mHHS of 53.1 (SD 18.1), and a mean iHOT-12 score of 46.9 (SD 19.7). The mean HOOS subscales were as follows: Sport/Recreation 44.8 (SD 27.1), Symptoms 56.2 (SD 22.8), Pain 57.6 (SD 27.2), Activities of Daily Living (ADL) 51.1 (SD 22.2), and Quality of Life (QoL) 30.4 (SD 19.7). Regarding preoperative psychological scores, the mean BSI-18 somatization score was 2.3 (SD 2.7), the mean BSI depression score was 2.6 (SD 3.6), the mean BSI anxiety score was 2.4 (SD 2.9), and the mean SF-36 general health perception score was 47.2 (SD 13.6).

At the three-month follow-up, 102 of the 235 patients (43%) had successfully resumed participation in sporting activities. By the six-month mark, the number of patients returning to sport had increased to 182 (77%). At the final follow-up 12 months after surgery, 223 patients (95%) had resumed participation in sports activities. In total, 12 patients (5%) had not yet resumed participation in sporting activities at the time of the final follow-up assessment (Figure 2).

Fig. 2.

Bar chart demonstrating the return to sports distributions. RTS, return to sport.

Regression analyses

In the logistic regression analysis predicting RTS within three months, the initial full model included all preoperative variables, with an AIC of 176.38 and 113° of freedom (Table I). Most variables were not significant predictors of early RTS, except for the RASS approach, which showed a trend towards significance (estimate: 1.102, p = 0.006). After RFE (Table II), the final model, with an AIC of 237.59 and 87° of freedom, identified the RASS surgical approach (estimate: 1.231, p < 0.001), preoperative exercise frequency (> two sessions/week; estimate: 0.718, p = 0.039), higher HOOS pain scores (estimate: 0.020, p = 0.011), lower BSI anxiety scores (estimate: -0.152, p = 0.011), and younger age as significant predictors. Male sex was negatively associated with early RTS (estimate: -1.281, p = 0.009).

Table I.

Multivariate logistic regression analysis of return to sport.

≤ 3 compared with > 3 mths			≤ 6 compared with > 6 mths
Variable	Estimate	p-value	Variable	Estimate	p-value
RASS	1.102 (0.404)	0.006	RASS	−0.205 (0.419)	0.063
Age (yrs)	0.039 (0.026)	0.131	Age	−0.033 (0.026)	0.208
BMI (kg/m²)	0.018 (0.059)	0.843	BMI	0.073 (0.057)	0.203
UCLA	0.045 (0.105)	0.669	UCLA	0.158 (0.106)	0.137
Low-impact	0.924 (0.810)	0.254	Low-impact	−0.029 (0.890)	0.974
High-impact	0.101 (0.869)	0.908	High-impact	0.729 (0.941)	0.439
Sess./wk. < 2	−0.583 (0.781)	0.456	Sess./wk. < 2	1.258 (0.857)	0.142
Sess./wk. > 2	0.239 (0.832)	0.774	Sess./wk. > 2	0.033 (0.907)	0.971
Sess. < 30 mins	0.249 (0.552)	0.653	Sess. < 30 mins	0.070 (0.582)	0.904
Sess. > 30 mins	0.790 (0.528)	0.125	Sess. > 30 mins	−0.414 (0.550)	0.452
mHHS	0.007 (0.018)	0.686	mHHS	0.012 (0.019)	0.529
iHOT-12	−0.008 (0.018)	0.614	iHOT-12	−0.014 (0.018)	0.416
HOOS Sport/Re.	0.008 (0.012)	0.497	HOOS Sport/Re.	−0.009 (0.013)	0.468
HOOS Symptoms	0.009 (0.013)	0.496	HOOS Symptoms	−0.009 (0.014)	0.506
HOOS QoL	−0.016 (0.013)	0.212	HOOS QoL	0.005 (0.013)	0.722
HOOS Pain	0.006 (0.018)	0.762	HOOS Pain	0.006 (0.019)	0.757
BSI Somatization	−0.011 (0.082)	0.896	BSI Somatization	0.011 (0.085)	0.900
BSI Depression	0.001 (0.073)	0.995	BSI Depression	−0.029 (0.072)	0.688
BSI Anxiety	−0.107 (0.091)	0.024	BSI Anxiety	0.113 (0.089)	0.205
Male	−0.570 (0.604)	0.345	Male	0.016 (0.563)	0.978
SF-36 GHP	0.003 (0.015)	0.856	SF-36 GHP	0.032 (0.016)	0.053

BSI, Brief Symptom Inventory; GHP, General Health Perception; HOOS, Hip disability and Osteoarthritis Outcome Score; iHOT-12, International Hip Outcome Tool-12; mHHS, modified Harris Hip Score; QoL, quality of life; RASS, rectus and sartorius sparing approach; Sess., session; SF-36, 36-Item Short-Form Health Survey questionnaire; Sport/Re., Sport/Recreation; UCLA, University of California, Los Angeles, activity scale.

Table II.

Multivariate logistic regression analysis of return to sport after recursive feature elimination.

≤ 3 compared with > 3 mths			≤ 6 compared with > 6 mths
Variable	Estimate	p-value	Variable	Estimate	p-value
RASS	1.231 (0.324)	< 0.001	BMI	0.074 (0.040)	0.066
Age (yrs)	0.035 (0.199)	0.081	Age	−0.040 (0.019)	0.029
Sess./wk > 2	0.718 (0.348)	0.039	Sess./wk > 2	1.248 (0.322)	< 0.001
Sess. > 30 mins	0.561 (0.330)	0.089	UCLA	0.190 (0.075)	0.011
HOOS Pain	0.020 (0.008)	0.011	HOOS Symptoms	−0.015 (0.007)	0.051
BSI Anxiety	−0.152 (0.061)	0.011
Male	−1.281 (0.489)	0.009

BSI, Brief Symptom Inventory; HOOS, Hip disability and Osteoarthritis Outcome Score; RASS, rectus and sartorius sparing approach; Sess., session.

For the six-month RTS model (Table I), with an initial AIC of 216.53 and 104° of freedom, age at surgery became a significant predictor (estimate: -0.040, p = 0.029) after RFE (Table II), with a final AIC of 239.71 and 92° of freedom. BMI showed a trend towards significance (estimate: 0.074, p = 0.066). The UCLA activity score remained a positive predictor (estimate: 0.190, p = 0.011), while sport frequency (> two sessions/week; estimate: 1.248, p < 0.001) and the HOOS Symptoms subscale (estimate: -0.015, p = 0.051) were also relevant factors.

Conditional interference tree models

The ctree model for predicting RTS within three months versus beyond three months had an AIC of 170.12, and identified several key variables that stratify patients based on their likelihood of achieving early RTS (Figure 3). The first significant split in the tree was determined by the RASS surgical approach (Node 1). Patients who underwent the RASS approach were generally more likely to RTS within three months. For those who did not undergo the RASS approach, the model further split based on the preoperative UCLA activity score (Node 2). Patients with a UCLA score of 4 or less were more likely to have a delayed RTS. These patients were further stratified by age (Node 4), with those aged older than 37.5 years showing a higher likelihood of returning to sport after three months (Node 6). Conversely, younger patients with a UCLA score above 4 (Node 5) were more likely to RTS within three months. In patients who underwent the RASS approach, the frequency of exercise sessions per week was the next significant factor (Node 7). Patients engaging in more than two sessions per week had a higher probability of RTS within three months. For these patients, the BSI Anxiety score provided further differentiation (Node 9). Patients with a BSI Anxiety score of 4 or less were particularly likely to RTS within three months (Node 12), while those with higher anxiety scores were more likely to have a delayed return (Node 13).

Fig. 3.

Conditional interference tree model for return to sport (RTS) < 3 months versus > 3 months. BSI, Brief Symptom Inventory; RASS, rectus and sartorius sparing approach; UCLA, University of California, Los Angeles, activity scale.

The full ctree model for predicting the timeline of RTS had an AIC of 203.45 and degrees of freedom consistent with the complexity of the model (Figure 4). The model first split by the preoperative UCLA activity score (Node 1). Patients with a UCLA score of 3 or lower were more likely to have a delayed RTS. Among these patients, the RASS approach further differentiated outcomes: patients undergoing the RASS approach were more likely to RTS within three months (Node 4), while those who did not undergo the RASS approach were more likely to have a delayed return (Node 3).

Fig. 4.

Conditional interference tree model for all return to sport (RTS) follow-ups. BSI, Brief Symptom Inventory; HOOS, Hip disability and Osteoarthritis Outcome Score; RASS, rectus and sartorius sparing approach; spw, sessions per week; UCLA, University of California, Los Angeles, activity scale.

For patients with a preoperative UCLA score greater than 3, the model further split by the RASS approach (Node 5) and BSI Anxiety scores (Node 6). Patients with lower anxiety scores (≤ 5) were more likely to RTS within six months (Node 7), whereas those with higher anxiety scores were predicted to return within 12 months or later (Node 8).

In another branch of the model, patients who did not undergo PAO by the RASS approach were stratified further based on the HOOS pain score. Female patients with a HOOS pain score of 42.5 or lower were more likely to RTS within three months (Node 12), while those reporting higher pain scores were more likely for delayed RTS (Node 13). Male patients in this group were generally predicted to RTS at around six months (Node 15).

Discussion

The main findings of this study were: 1) PAO enables early RTS within the first postoperative year in over 90% of patients; 2) key factors for earlier RTS are minimally invasive surgical techniques that allowed immediate active hip flexion and higher preoperative physical activity levels, with more than two sessions of sports per week, lower BSI anxiety scores, and higher preoperative HOOS pain scores; and 3) male sex and older age representing potential risk factors for a delayed RTS after PAO.

In comparison to existing literature, which often reports lower or less clearly defined RTS rates, our results are particularly noteworthy. For example, a study by Heyworth et al²⁸ reported that 80% of patients returned to play at a median of nine months postoperatively, but only 58% of competitive athletes returned to their preoperative level, while 89% of recreational athletes achieved this milestone. However, it is important to note that Heyworth et al’s study focused on athletes specifically and their return to play at a competitive level, whereas our study evaluated RTS in a representative non-athletic population. This difference in focus likely accounts for the earlier RTS observed in our study. Leopold et al¹⁶ conducted a similar study, using comparable surveys to assess RTS after PAO, but their study design was retrospective, with patients surveyed several years postoperatively. They reported a slightly higher overall RTS rate, with most patients resuming sports after six months (42%) or three to six months (38%), and only 3% not returning to sports. Overall, the timing of RTS in their study was generally later than in our findings, likely due to the retrospective design, which may introduce recall bias and affect the accuracy of the reported RTS timelines. In addition to the aforementioned factors, the clear postoperative guidelines provided to our patients played an important role in the early RTS observed in our study.

In contrast to previous studies on RTS, which habitually disregard the significance of mentioning a structured postoperative care regime, our methodology incorporated comprehensive recommendations that are pivotal for achieving early RTS. The Delphi consensus on PAO rehabilitation also highlights this lack of standardized protocols.²⁹ Our study not only is in accordance with, but also exceeds, these guidelines by making use of minimally invasive techniques that facilitate more rapid mobilization. Furthermore, our meticulous and periodic follow-ups throughout the initial postoperative year guaranteed precise and prompt RTS data, distinguishing our findings from those of existing literature.

Thus, this study is the first to prospectively examine factors influencing the timing of RTS after PAO, addressing a substantial gap in the existing research. For instance, Petrie et al¹⁵ reported on a large multicentre prospective cohort where 66% of patients saw an increase or no change in activity levels after PAO, with those in the high preoperative activity group experiencing a notable decrease due to a ceiling effect. Their research focused on identifying predictors of changes in activity levels following PAO, using multivariable regression analysis to assess factors influencing these changes as measured by the UCLA score. Their findings highlighted prior ipsilateral surgery as an independent predictor for improvement in postoperative UCLA scores. Nevertheless, the study did not investigate when patients resume participation in sports activities or identify the factors that influence this timeline.

In our analysis, we found that the surgical approach, particularly the use of minimally invasive techniques that allow immediate active flexion in the hip joint postoperatively, was a significant factor associated with earlier RTS. This was supported by both the logistic regression analysis and ctree model, where patients undergoing this minimally invasive technique had a higher likelihood of RTS within three months. It is important to note that this advantage may be partially influenced by institutional rehabilitation protocols rather than the surgical approach alone. Based on our findings, we believe that early active mobilization is likely beneficial for postoperative recovery, as it allows for faster functional adaptation and potentially accelerates RTS. This view is supported by the fact that many other surgeons allow early active hip flexion even in RS patients.

Given these considerations, we now believe that restricting active flexion following the RS approach may not be necessary. However, this has become less relevant in our current practice, as we have almost entirely transitioned to the RASS approach, where immediate active hip flexion is already permitted.

Additionally, preoperative physical activity levels as measured by the UCLA activity score, the frequency of engaging in more than two sessions per week of physical activity, lower BSI anxiety scores, and higher HOOS pain scores were all strongly predictive of earlier RTS. Specifically, the threshold of more than two sport sessions per week emerged as a critical predictor, indicating that patients who maintained this level of activity preoperatively had a markedly better chance of a quick recovery.

Conversely, male sex was identified as a negative predictor for early RTS within three months, suggesting that male patients might experience a slower recovery process compared to females. This finding highlights potential sex-specific differences in the recovery trajectory following PAO, a conclusion that is consistent with the results presented in previous studies.^7,15,16,30 Lower preoperative BSI Anxiety scores were significant, as they underscore the role of psychological wellbeing in promoting a faster return to physical activity.³¹ Higher preoperative HOOS pain scores were also positively associated with early RTS, indicating that patients experiencing less pain before surgery may recover more quickly and RTS sooner.

For RTS within six months, our findings highlighted the role of age, where older patients were less likely to RTS within this timeframe, contrasting with the younger age trend observed for earlier RTS. The consistency of the UCLA activity score as a positive predictor across both timeframes further underscores the value of maintaining a higher preoperative activity level.

This insight has important implications for the timing of surgery. If a patient is still able to engage in physical activity, even at a reduced level, they may benefit from undergoing PAO before their activity levels decline further. Our data suggest that waiting until a patient is no longer active might delay their postoperative recovery and prolong the time needed to return to activity. Therefore, timing the surgery when the patient still maintains a reasonable level of activity could enhance the likelihood of a faster and more successful recovery.

The implementation of a ctree model marks a novel approach in the field of hip-preserving surgery, enabling us to generate a prediction model for estimating the likelihood of RTS for future patients. This model not only facilitates a more personalized approach to patient care, but also sets a new standard for how RTS outcomes can be predicted and managed in clinical practice. A meta-analysis conducted by Curley et al³² highlighted the lack of tools for predicting RTS, with most analyses being descriptive rather than predictive. Our approach aligns with broader trends in medicine, where machine learning and artificial intelligence are being leveraged to better understand outcomes and inform clinical decision-making.³³ In orthopaedics, this technique has primarily been applied to predict outcomes following spinal surgery and to estimate the likelihood of hospital readmissions.^34-36 Moreover, our approach addresses the specific needs of PAO patients by focusing on the alignment of rehabilitation protocols with patient expectations, as highlighted in a qualitative interview study by Jacobsen et al,⁸ which explored the challenges and expectations of PAO patients in Denmark and Australia. This study underscores the importance of individualized rehabilitation strategies that align with patient hopes and realities, thereby enhancing the recovery process and overall patient satisfaction.

Therefore, using machine-learning techniques, the current study developed a novel prediction model that can be used to score future patients and provide personalized probabilities of their RTS outcomes. This predictive capability marks a major step forward in individualizing postoperative care and managing patient expectations more effectively.

However, the study has several limitations that must be acknowledged. The single-centre design and involvement of a single surgeon limit the generalizability of our findings. This is particularly relevant when considering the specific techniques used, including the minimally invasive RASS approach and the fast-track rehabilitation protocol, which are not yet widely implemented. Other studies have reported on different surgical techniques and rehabilitation protocols, making direct comparisons challenging. Furthermore, the selection of the RS and RASS approach was not randomized but rather based on patient-specific anatomical factors. Although this approach was intended to ensure an individualized approach, it introduces the possibility that unmeasured differences between the two groups could influence RTS outcomes.

While our study identifies key predictors of RTS using logistic regression and machine-learning approaches, it is important to acknowledge the inherent statistical limitations associated with multivariate analyses. Specifically, a risk of Type II error should be considered, particularly when multiple variables are analyzed simultaneously.

Additionally, our study used a ctree model to predict RTS following PAO. This machine-learning approach enables hierarchical stratification of predictive variables, offering an intuitive and clinically interpretable framework. However, ctree models also have several limitations that should be considered when interpreting the results: ctree models are sensitive to sample size and data distribution, meaning that uneven distributions of predictor variables could influence the structure of the tree. Unlike traditional regression models, ctree does not provide direct effect estimates, making it difficult to quantify the precise magnitude of influence each predictor has on RTS.

Finally, the external validity of our model has yet to be established, emphasizing the need for further validation in larger, independent cohorts. Future research should focus on external validation of our findings in multicentre studies to confirm the robustness and generalizability of our predictive model.

In conclusion, our study provides valuable insights into patient-individual factors influencing the timing of RTS following PAO. We demonstrated that minimally invasive surgical techniques, maintaining preoperative physical activity levels, and lower anxiety scores are significant predictors of an earlier RTS, particularly within the first three months postoperatively. Conversely, factors such as male sex and older age are associated with a delayed RTS, highlighting the importance of personalized patient management. Our findings suggest that timing the surgery while patients still maintain some level of physical activity could lead to a faster recovery and more favourable outcomes. Future studies should focus on validating these findings in larger, multicentre cohorts and further refining the prediction model to ensure its utility across diverse patient populations.

Take home message

- Periacetabular osteotomy (PAO) enables early return to sport (RTS) within the first postoperative year in over 90% of patients.

- Key factors for earlier RTS include immediate active hip flexion, higher preoperative physical activity levels, more than two sessions of sports per week prior to surgery, low Brief Symptom Inventory-18 (BSI-18) anxiety scores, and higher preoperative Hip disability and Osteoarthritis Outcome Score (HOOS) pain scores. Male sex and older age were associated with delayed RTS after PAO.

- The preoperative evaluation of these variables enables the prediction of the RTS timeline.

Data Availability

The datasets generated and analyzed in the current study are not publicly available due to data protection regulations. Access to data is limited to the researchers who have obtained permission for data processing. Further inquiries can be made to the corresponding author.