Skip to main content
Download PDF

Factors to improve odds of success following medial opening-wedge high tibial osteotomy: a machine learning analysis

BMC Musculoskeletal Disorders volume 25, Article number: 323 (2024) Cite this article

Abstract

Background

Although high tibial osteotomy (HTO) is an established treatment option for medial compartment osteoarthritis, predictive factors for HTO treatment success remain unclear. This study aimed to identify informative variables associated with HTO treatment success and to develop and internally validate machine learning algorithms to predict which patients will achieve HTO treatment success for medial compartmental osteoarthritis.

Methods

This study retrospectively reviewed patients who underwent medial opening-wedge HTO (MOWHTO) at our center between March 2010 and December 2015. The primary outcomes were a lack of conversion to total knee arthroplasty (TKA) and achievement of a minimal clinically important difference of improvement in the Knee Injury and Osteoarthritis Outcome Score (KOOS) at a minimum of five years postoperatively. Recursive feature selection was used to identify the combination of variables from an initial pool of 25 features that optimized model performance. Five machine learning algorithms (XGBoost, multilayer perception, support vector machine, elastic-net penalized logistic regression, and random forest) were trained using five-fold cross-validation three times and applied to an independent test set of patients. The performance of the model was evaluated by the area under the receiver operating characteristic curve (AUC).

Results

A total of 231 patients were included, and 200 patients (86.6%) achieved treatment success at the mean of 9 years of follow-up. A combination of seven variables optimized algorithm performance, and the following specific cutoffs increased the likelihood of MOWHTO treatment success: body mass index (BMI) ≤ 26.8 kg/m2, preoperative KOOS for pain ≤ 46.0, preoperative KOOS for quality of life ≤ 33.0, preoperative International Knee Documentation Committee score ≤ 42.0, preoperative Short-Form 36 questionnaire (SF-36) score > 42.25, three-month postoperative hip-knee-ankle angle > 1.0°, and three-month postoperative medial proximal tibial angle (MPTA) > 91.5° and ≤ 94.7°. The random forest model demonstrated the best performance (F1 score: 0.93; AUC: 0.81) and was transformed into an online application as an educational tool to demonstrate the capabilities of machine learning.

Conclusions

The random forest machine learning algorithm best predicted MOWHTO treatment success. Patients with a lower BMI, poor clinical status, slight valgus overcorrection, and postoperative MPTA < 94.7 more frequently achieved a greater likelihood of treatment success.

Level of evidence

Level III, retrospective cohort study.

Peer Review reports

Background

Medial opening-wedge high tibial osteotomy (MOWHTO) is a reliable treatment option for medial compartment osteoarthritis of the knee with varus alignment in young patients [1]. The benefits of MOWHTO have been demonstrated in multiple studies showing satisfactory clinical results, slowed progression of knee osteoarthritis, and delayed conversion to total knee arthroplasty (TKA) [2,3,4,5,6,7,8].

Previous studies have shown various survival rates ranging from 74 to 92% after 10 years of follow-up [9,10,11,12]. Several potential risk factors for failed MOWHTO have been identified: older age, higher body mass index (BMI), progressed stage of osteoarthritis, and undercorrected alignment of the lower extremity appear to be related to poor outcomes or survival after HTO [11, 13,14,15,16]. However, patients usually have multiple simultaneous risk factors, and due to the complex interactions and relationships between these factors, the ability to accurately predict and quantify the probability of failure of MOWTHO is challenging.

Machine learning is an artificial intelligence (AI) application that can analyze complex big data and generate algorithms to predict an outcome [17]. There has been a growing need to understand machine learning in medicine, and the influence of machine learning in orthopedic surgery has also recently attracted considerable interest [18,19,20,21,22,23]. If prognostic factors can be predicted preoperatively, appropriate preventive measures to delay the requirement for surgery can be performed to achieve successful postoperative outcomes in high-risk patients. Furthermore, factors that can allow clinicians to make an accurate prognosis and provide patients with customized risk predictions for outcomes are crucial.

To date, no study has developed a machine learning model to estimate the survival rates or success rates of MOWHTO based on predictive factors. Therefore, this study aimed to identify informative variables associated with the success of MOWHTO and to develop and internally validate machine learning algorithms to predict which patients will achieve treatment success following MOWHTO for medial compartmental osteoarthritis. We examined several variables that have previously been identified as factors influencing the outcome of MOWHTO and developed a machine learning algorithm based on them. Our hypothesis was that machine learning analysis could accurately predict patients who will achieve treatment success after MOWHTO for the follow-up duration.

Methods

Data and patients

This study was approved by the institutional review board of our institution, and all patients provided written informed consent. We retrospectively identified all patients who underwent MOWHTO at our institution for knee osteoarthritis between March 2010 and December 2015. Patients who underwent MOWHTO for isolated medial compartment osteoarthritis (Kellgren–Lawrence grad ≥ II) and who had varus malalignment with intact cruciate ligaments were included.

The exclusion criteria were as follows: (1) prior open knee procedures; (2) associated ligamentous insufficiency (anterior or posterior cruciate ligament) requiring reconstruction; (3) knee range of motion < 120° and flexion contracture > 15°; (4) either patellofemoral or lateral compartment osteoarthritis (Kellgren–Lawrence grade ≥ II); (5) inflammatory arthritis or traumatic osteoarthritis; and (6) a follow-up period < 5 years.

Ultimately, 231 knees (231 patients) with a mean follow-up period of 9 years (range, 5.0–11.5 years) were enrolled in the study.

Surgical techniques and postoperative management

The goal was to shift the weight-bearing line to the Fujisawa point and to create 3–5° of postoperative mechanical valgus [24]. All surgical procedures were performed by two experienced orthopedic surgeons. Arthroscopic examinations were performed regularly at the time of MOWHTO. Detected meniscal tears or articular cartilage injuries were treated with debridement or microfracture based on the surgeon’s judgment. Following arthroscopy, MOWHTO was performed in a biplanar fashion according to the technique developed by the Arbeitsgemeinschaft für Osteosynthesefragen International Knee Expert group using a locking plate [25].

Primary outcomes and candidate variables

The primary outcomes were defined as the achievement of a minimal clinically important difference (MCID) in the Knee Injury and Osteoarthritis Outcome Score (KOOS) for pain and clinical survival at least five years after HTO [26]. The endpoint of survival was conversion to TKA. Moreover, we selected 25 variables that have already been demonstrated to be predictive factors in the recent literature and significantly impact HTO outcomes and survival rates [11, 13,14,15,16, 27]. These included sex, age, BMI, meniscal status, cartilage status using the International Cartilage Repair Society (ICRS) grade, existence of kissing lesions, additional cartilage repair procedures, patient-reported outcome measures (PROMs) (KOOS, International Knee Documentation Committee [IKDC] score, Short-Form 36 [SF-36] questionnaire, Tegner activity scale score), and radiographic evaluations (mechanical hip-knee-ankle [HKA] angle), posterior tibial slope, medial proximal tibial angle [MPTA], and Kellgren–Lawrence grading). All the outcomes were assessed by two independent investigators blinded to the surgical procedures and study aim.

Traditional statistical analysis

We defined clinical success as a well-functioning MOWHTO with a KOOS for pain that exceeded the MCID after a minimum follow-up of five years without conversion to TKA. An independent t-test and Pearson’s Chi-square test were used for continuous and categorical variables, respectively. The odds ratio of each variable was calculated. Multivariate logistic regression analyses were performed to examine the effects of these variables on treatment success. Receiver operating characteristic (ROC) curves with area under the curve (AUC) analyses were performed to evaluate model sensitivity and specificity. All statistical analyses were performed using SPSS (version 25.0; SPSS, Chicago, IL, USA), and p < 0.05 was considered statistically significant.

Algorithm development

A binary classifier model that predicts the possibility of treatment success based on 25 variables was developed and constructed using Python. An 80:20 random sample split was used to partition the study population into training (n = 184) and independent test (n = 47) sets for algorithm development and internal validation, respectively. Five-fold cross-validation with five iterations of training using the training set was used to develop five unique machine learning algorithms: extreme gradient boosting (XGBoost), multi-layer perception (MLP), support vector machine (SVC), random forest (RF), and elastic-net penalized logistic regression (ENPLR).

Model performance assessment

We evaluated the performance of the models using independent test sets that were not used for model training and measured their performance using the Brier score and AUC. These metrics were used to identify the best-performing model. An ROC curve plots the probability of correctly identifying positive cases against the probability of falsely identifying positive cases at different threshold settings. The AUC score assigns a score of 1 to a perfect predictor and 0.5 to a predictor with random guesses. The Brier score is calculated by averaging the squared difference between the outcome and model prediction probabilities to provide an overall performance measure. Lower Brier scores indicate better model performance. We used the Brier score obtained from a traditional logistic regression as a null model, and models with better performance than this null model were considered adequate. Finally, the optimal model was selected based on the lowest Brier score.

Model fidelity and application development

It is important to understand individual predictions made by classifiers. The explanation of individual predictions allows informed decision-making about how much predictions can be trusted and provides insight to improve the model. Ribeir et al. [28] introduced LIME, which explains individual predictions using locally interpretable surrogate models. In this study, LIME provides quantitative data and visual representations of patient-specific predictions, enabling determination of what specific combinations of factors support or contradict the prediction that a specific patient will achieve treatment success with MOWHTO. We developed a web application that provides individualized prediction through the R packages ‘lime’ by Pedersen et al. [29] and ‘shiny’ by Chang et al. [30] However, in their current form, these predictions represent a proof-of-concept for machine learning in orthopedics and should not be used until additional validation studies are performed.

Results

A total of 231 patients were included, and 200 patients (86.6%) achieved treatment success at a mean of 9 years of follow-up. The demographic characteristics and clinical data are summarized in Table 1.

Table 1 Baseline Demographic Characteristics and Clinical Data*
Full size table

Traditional analysis − logistic regression analysis

Based on the multivariate regression analysis, the preoperative KOOS for pain, preoperative SF-36 Physical Component Summary (PCS) score, and preoperative MPTA influenced treatment success after MOWHTO (Table 2). The AUC of this regression model was 0.66 (95% confidence interval (CI), 0.420–0.890).

Table 2 Multiple Logistic Regression Analysis for Predictors of MOWHTO Treatment Success
Full size table

Assessment of algorithm performance

The relative algorithm performances of the five algorithms are described in Table 3. The best-performing algorithm was the random forest model with an AUC of 0.81 (95% CI, 0.771–0.849) (Fig. 1). The Brier score was 0.09 (95% CI, 0.089–0.099). The null-model Brier score was 0.107 (95% CI 0.101–0.114), indicating that this algorithm appropriately calibrated predictions.

Table 3 Performance of Each Machine Learning Algorithm in the Independent Test Set of Patients*
Full size table
Fig. 1
figure 1

Receiver operating characteristic (ROC) curve for the random forest machine learning model. AUC = area under the ROC curve

Full size image

Importance of the features

We evaluated the importance ranks that indicated the importance of the input features for the random forest algorithm. The performance of the algorithm was optimized by combining seven variables: BMI, preoperative IKDC score, preoperative KOOS for pain, preoperative KOOS for quality of life (QOL), preoperative SF-36 PCS score, postoperative HKA angle, and postoperative MPTA (Fig. 2). We created 56 unique cases of LIME with 5,000 permutations to determine the relative contribution of these features to the overall predictions. This allowed us to determine the levels of each feature and ranges of values that either supported or contradicted treatment success for categorical and continuous variables. A BMI ≤ 26.8 kg/m2, preoperative IKDC score ≤ 42.0, preoperative KOOS for pain ≤ 46.0, preoperative KOOS for QOL ≤ 33.0, preoperative SF-36 PCS score > 42.25, postoperative HKA angle > 1.0°, and a postoperative MPTA > 91.5° and ≤ 94.7° were associated with treatment success for patients undergoing MOWHTO for medial compartmental osteoarthritis.

Fig. 2
figure 2

Feature importance plot for the random forest model based on the independent test set. Each predictive weight of each variable is compared among the other seven variables chosen from recursive feature elimination with cross-validation. KOOS = Knee Injury and Osteoarthritis Outcome Score, SF-36 PCS = Short-Form 36 questionnaire Physical Component Summary, IKDC = International Knee Documentation Committee, HKA = hip–knee–ankle, QOL = quality of life, BMI = body mass index, and MPTA = medial proximal tibial angle

Full size image

Customized prediction application

We deployed the optimal algorithm as a web-based application (https://ailab.shinyapps.io/betterhto/). When the seven studied features are input into the algorithm, the probability of treatment success following MOWHTO is expressed as a percentage, and the importance of each feature used in the decision-making process is displayed in a graph (Fig. 3).

Fig. 3
figure 3

Demonstration of the possible clinical influence of the clinical decision-making tool derived from the random forest model. The probability of achieving treatment success following MOWHTO is 98.0% (left) and 42.0% (right). Factors marked in blue positively affected this patient’s ability to achieve treatment success. Factors marked in red had a negative impact. KOOS = Knee Injury and Osteoarthritis Outcome Score, SF-36 PCS = Short-Form 36 questionnaire Physical Component Summary, IKDC = International Knee Documentation Committee, HKA = hip–knee–ankle, QOL = quality of life, BMI = body mass index, MPTA = medial proximal tibial angle, and MOWHTO = medial opening-wedge high tibial osteotomy

Full size image

Discussion

This study’s principal finding was that the random forest machine learning-based model demonstrated the best performance for predicting treatment success of MOWHTO at a mean of nine years postoperatively. The algorithm required only seven factors to predict treatment success: BMI, preoperative IKDC score, preoperative KOOS for pain, preoperative KOOS for QOL, preoperative SF-36 PCS score, postoperative HKA angle, and postoperative MPTA. The AUC to predict treatment success was 0.81, which exceeds the threshold for good performance of ≥ 0.8 [20], and this study enabled individualized prediction of treatment success after MOWHTO using a web-based system. Our results demonstrate that machine learning algorithms are promising new approaches in clinical situations wherein several variables must be comprehensively assessed, such as in predicting treatment success of MOWHTO in patients with medial compartmental osteoarthritis.

This is the first study to predict the patient-specific treatment success of MOWHTO by applying a machine learning model. Predicting treatment success of MOWHTO is challenging due to the heterogeneity and diversity of associated variables; discriminating factors must be identified to guide treatment decisions, and accurately quantifying this risk is difficult [15]. Our seven identified variables are simple and intuitive and can guide the patient-specific discussion regarding surgical options and realistic outcome goals.

Predictors of MOWHTO treatment success must be identified to reduce the risk of failure requiring TKA, and numerous predictors have been described to identify ideal candidates for MOWHTO [5, 11, 13,14,15,16, 31]. Bonasia et al. [11] identified positive (Ahlback grade 0 arthritis of the medial compartment and a good preoperative Knee Society score) and negative prognostic factors (age > 56 years and postoperative knee flexion < 120°) associated with MOWHTO outcomes in a case series of 140 patients. Jin et al. [16] analyzed the risk factors for survival after MOWHTO, and the main failure criteria were conversion to TKA and inferior PROMs. They presented a regression analysis showing that age ≥ 65 years, grade 4 cartilage damage in the medial compartment, and grade ≥ 2 cartilage damage in the lateral compartment negatively influenced outcomes after MOWHTO. Bouguennec et al. [13] demonstrated that survival factors reducing the risk of MOWHTO failure included female sex, age < 54 years, BMI < 25 kg/m2, Ahlback grade 1 or 2, varus articular component < 0.9°, HKA angle correction > 180°, and absence of a hinge fracture. Patients usually have multiple simultaneous risk factors, and some studies did not address the confounding effects of other variables, which should be controlled for accurate analysis of the true effects on postoperative outcomes.

Machine learning involves techniques that model complex relationships between variables to predict an outcome. Applications of predictive machine learning have broadly impacted the medical field, especially in orthopedic surgery, and facilitate surgeon decision-making [20,21,22,23, 32,33,34,35,36,37,38]. Batailler et al. [39] recently determined the main predictive factors for long-term HTO survival and proposed a predictive score that includes age, BMI, and the presence or absence of a joint line and is particularly useful in borderline cases for decision-making regarding potential HTO surgery. However, this is not a true prediction, and a machine learning model may be a helpful decision aid in daily practice to determine HTO indications. Martin et al. [21] performed a machine learning analysis of the Norwegian Knee Ligament Register (NKLR), identified important risk factors related to subsequent revision of primary anterior cruciate ligament (ACL) reconstruction, and developed a clinically meaningful calculator for predicting revision of primary ACL reconstruction. Kunze et al. [23] developed machine learning algorithms capable of providing patient-specific predictions of which athletes will achieve clinically relevant improvement in sports-specific function after hip arthroscopy for femoroacetabular impingement syndromes. Their machine learning algorithms demonstrated excellent performance in predicting achievement of an MCID in clinical scores. Using this framework, orthopedic surgeons may consider various treatment options preoperatively according to the patients’ individual risk profiles.

Notably, the random forest machine learning model identified seven variables that differed from those highlighted in traditional regression analyses as crucial for predicting MOWHTO treatment success. This discrepancy may be due to the distinct methodologies applied. Traditional regression highlighted only three parameters as significant in a multivariate analysis, emphasizing the reliance on statistical significance (P values) for variable selection. However, the random forest model employs a feature importance mechanism and ranks variables based on their contribution to model accuracy rather than based on statistical significance alone. This approach led to the identification of variables such as BMI, preoperative IKDC score, preoperative KOOS for pain, preoperative KOOS for QOL, preoperative SF-36 PCS score, postoperative HKA angle, and postoperative MPTA as important predictors. We identified two types of predictive factors: preoperative variables and postoperative variables at three months after the initial surgery. A low BMI and poor clinical status except for the preoperative SF-36 PCS score were important preoperative variables associated with MOWHTO treatment success. A BMI ≤ 26.8 kg/m2 was associated with MOWHTO treatment success using LIME analysis, which is in agreement with the findings of previous studies (25.0–27.5 kg/m2) [39, 40]. Overweight patients put excessive stress on the knee joint, which may accelerate degenerative changes and surgical outcomes. Bouguennec et al. [13] reported that a BMI < 25 kg/m2 was associated with reduced HTO failure, and Howells et al. [41] showed inferior PROMs at 5 years after HTO in patients with a BMI > 30 kg/m2. Patients with more inferior PROMs preoperatively showed a better prognosis than those with less inferior PROMs. A more severe clinical status for osteoarthritic knees may create an opportunity for improvement with MOWHTO. Preoperative features are dynamic and may be optimized following a trial of nonsurgical management. These findings are quite similar to those of previous reports in that significantly more patients with more severe disease before TKA are satisfied with their procedure than those with less severe degenerative changes [42, 43].

The ideal degree of correction has been extensively evaluated, and correction from neutral up to extreme valgus corrections is recommended [10, 44,45,46]. Our results showed that an HKA angle ≥ 1° was a positive factor for MOWHTO treatment success, which is consistent with the literature [13, 39], and undercorrection is generally associated with worse results [1, 47]. Thus, achieving adequate operative correction to a relevant angle is necessary for good long-term outcomes after MOWHTO [48, 49]. Furthermore, our findings suggest that unloading effects of MOWHTO led to clinical success, with no association with meniscal/chondral status or additional cartilage repair procedures and postoperative clinical outcomes.

Although the association between excessive joint line obliquity (JLO) and inferior outcomes after HTO has not yet been demonstrated, considering JLO for HTO is crucial [50,51,52]. In their biomechanical study, Nakayama et al. reported that an MPTA > 95° was unacceptable [53]. Schuster et al. [52] retrospectively reviewed 79 patients with medial knee osteoarthritis and demonstrated that an overcorrected MPTA (> 95°) was related to inferior clinical outcomes during long-term follow-up. Kim et al. [50] also assessed the influence of the MPTA on HTO outcomes at a minimum four-year follow-up using a propensity score matching analysis and suggested that although an excessively increased MPTA after HTO had no significant effects on clinical outcomes and cartilage deterioration in the lateral compartment, lateral compartment pain was experienced significantly more frequently. Thus, our findings of a postoperative MPTA ≤ 94.70° by LIME analysis are clinically relevant, and concerns about the potential side effects of a certain extent of overcorrection of MPTA should be understood cautiously for the orthopedic surgeons.

The random forest machine learning algorithm demonstrated excellent performance for predicting MOWHTO treatment success in patients with medial compartmental osteoarthritis compared with the conventional logistic regression model based on the AUC. The conventional logistic regression model is prone to overfitting of training data when used as a prediction model [54], often resulting in poorer performance when presented with new data, which makes it difficult to use clinically. The random forest model achieved a 12% relative Brier score reduction over the traditional logistic regression analysis. However, random forest classifiers may require more training data to produce robust results and may also contain unnecessary predictors; thus, further study of model optimization through feature selection of input variables and data augmentation is necessary. LIME was used to explain the individual predictions of our model. It applies to any predictive model and has no assumptions about the model. This is advantageous, especially when the model is trained to be noninterpretable. However, the results may vary at each execution because the data points are sampled without considering the correlation between variables. Alvarez-Melis et al. [55] reported that the explanations of two close points are very different from each other, indicating that there is instability in the explanation; thus, it sometimes can be difficult to trust.

Our study has certain limitations. First, this was a retrospective nonrandomized study, and relatively few patients were recruited. Further large-sample studies may improve the machine learning model performance. Second, although we considered a variety of machine learning methods, a model that was not considered might have had superior performance. Third, there are other possible predictors associated with MOWHTO treatment success that inherently vary depending on institutional protocols and surgeon preference that we could not evaluate in the present study. Fourth, for each of the five unique machine learning algorithms, we evaluated their performance using a training dataset through five-fold cross-validation with five iterations. However, developing a meta-algorithm, which was not used in the present study, would be better to explore the integration and improve the performance of the model. Fifth, the machine learning algorithms developed in this study were internally validated using an independent test set. However, external validation using data from other centers would strengthen the robustness of the algorithms. As this study recruited patients from a single center, observer or selection biases cannot be eliminated. Sixth, there might have been selection bias among the included patients. In Asian populations, MOWHTO tends to be more frequently performed in females than in males. Our findings might have been biased by the disproportionate female sex predominance [56,57,58,59]. Seventh, this study was based on a retrospective analysis performed at a single institution, which may restrict the generalizability of the results. Finally, it is possible that there are other important variables, including pre- to postoperative (delta) posterior tibial slope or contralateral ICRS grade, that could have been evaluated and may have led to alternative results. Furthermore, it could be more practical to include these variables with only one PROM because time-consuming PROM assessments are not routinely conducted in daily clinical practice. The addition of a slimmed web-based algorithm would allow the clinicians to reduce the number of input factors and thus facilitate obtaining information more quickly; however, it will likely have limited statistical significance. Further studies with larger sample sizes and additional crucial variables as well as collaborations with multiple centers are necessary to validate our findings.

Conclusion

The correct indications are necessary to achieve MOWHTO treatment success. The random forest machine learning-based model used to evaluate patients who underwent MOWHTO showed demonstrated the best performance for predicting MOWHTO treatment success. According to our findings, patients with a lower BMI, poor clinical status, slight valgus overcorrection, and postoperative MPTA < 94.7° more frequently achieved a greater probability of treatment success. Our findings are clinically relevant and would allow patient and surgical information to guide shared clinical decision-making for patient-specific management.

Data availability

The datasets used and analyzed during the current study available are from the corresponding author upon reasonable request.

Abbreviations

MOWHTO:

Medial opening-wedge high tibial osteotomy

TKA:

Total knee arthroplasty

BMI:

Body mass index

AI:

Artificial intelligence

MCID:

Minimally clinically important difference

KOOS:

Knee Injury and Osteoarthritis Outcome Score

PROMs:

Patient-reported outcome measures

IKDC:

International Knee Documentation Committee

SF-36:

Short-Form 36

HKA:

Hip-knee-ankle angle

MPTA:

Medial proximal tibial angle

ROC:

Receiver operating characteristic

AUC:

Area under the curve

XGBoost:

Extreme gradient boosting

MLP:

Multi-layer perception

SVC:

Support vector machine

RF:

Random forest

ENPLR:

Elastic-net penalized logistic regression

NKLR:

Norwegian Knee Ligament Register

ACL:

Anterior cruciate ligament

References

  1. Amendola A, Bonasia DE. Results of high tibial osteotomy: review of the literature. Int Orthop. 2010;34(2):155–60.

    Article  PubMed  Google Scholar 

  2. Faschingbauer M, Nelitz M, Urlaub S, Reichel H, Dornacher D. Return to work and sporting activities after high tibial osteotomy. Int Orthop. 2015;39(8):1527–34.

    Article  PubMed  Google Scholar 

  3. Park JG, Han SB, Jang KM. Association of Preoperative Tibial Varus Deformity with Joint Line Orientation and Clinical Outcome after Open-Wedge High Tibial Osteotomy for Medial Compartment Osteoarthritis: a propensity score-matched analysis. Am J Sports Med. 2021;49(13):3551–60.

    Article  PubMed  Google Scholar 

  4. Yuwen P, Sun W, Yang Y, Feng C, Wu D, Chen W, Zhang Y. Medium to long term of self-report outcomes of high tibial osteotomy in advanced age patients. Int Orthop 2024.

  5. Yang HY, Kwak WK, Lee CH, Kang JK, Song EK, Seon JK. Extent of Preoperative Medial Meniscal Extrusion influences Intermediate-Term outcomes after medial opening-wedge high tibial osteotomy. J Bone Joint Surg Am. 2022;104(4):316–25.

    Article  PubMed  Google Scholar 

  6. Yang HY, Cheon JH, Lee CH, Song EK, Seon JK. Effect of prior knee arthroscopy on Midterm outcomes after medial opening-wedge high tibial osteotomy: a propensity score-matched analysis. Orthop J Sports Med. 2023;11(6):23259671231175457.

    Article  PubMed  PubMed Central  Google Scholar 

  7. Ollivier B, Berger P, Depuydt C, Vandenneucker H. Good long-term survival and patient-reported outcomes after high tibial osteotomy for medial compartment osteoarthritis. Knee Surg Sports Traumatol Arthrosc. 2021;29(11):3569–84.

    Article  PubMed  Google Scholar 

  8. Park JY, Kim JH, Cho JW, Kim MS, Choi W. Clinical and radiological results of high tibial of osteotomy over the age of 65 are comparable to that of under 55 at minimum 2-year follow-up: a propensity score matched analysis. Knee Surg Relat Res. 2024;36(1):10.

    Article  PubMed  PubMed Central  Google Scholar 

  9. Barton SB, McLauchlan GJ, Canty SJ. The incidence and impact of arthroscopy in the year prior to total knee arthroplasty. Knee. 2017;24(2):396–401.

    Article  CAS  PubMed  Google Scholar 

  10. Bode G, von Heyden J, Pestka J, Schmal H, Salzmann G, Südkamp N, Niemeyer P. Prospective 5-year survival rate data following open-wedge valgus high tibial osteotomy. Knee Surg Sports Traumatol Arthrosc. 2015;23(7):1949–55.

    Article  PubMed  Google Scholar 

  11. Bonasia DE, Dettoni F, Sito G, Blonna D, Marmotti A, Bruzzone M, Castoldi F, Rossi R. Medial opening wedge high tibial osteotomy for medial compartment overload/arthritis in the varus knee: prognostic factors. Am J Sports Med. 2014;42(3):690–8.

    Article  PubMed  Google Scholar 

  12. Duivenvoorden T, Brouwer RW, Baan A, Bos PK, Reijman M, Bierma-Zeinstra SM, Verhaar JA. Comparison of closing-wedge and opening-wedge high tibial osteotomy for medial compartment osteoarthritis of the knee: a randomized controlled trial with a six-year follow-up. J Bone Joint Surg Am. 2014;96(17):1425–32.

    Article  CAS  PubMed  Google Scholar 

  13. Bouguennec N, Mergenthaler G, Gicquel T, Briand C, Nadau E, Pailhé R, Hanouz JL, Fayard JM, Rochcongar G. Medium-term survival and clinical and radiological results in high tibial osteotomy: factors for failure and comparison with unicompartmental arthroplasty. Orthop Traumatol Surg Res. 2020;106(8s):S223–30.

    Article  PubMed  Google Scholar 

  14. Keenan OJF, Clement ND, Nutton R, Keating JF. Older age and female gender are independent predictors of early conversion to total knee arthroplasty after high tibial osteotomy. Knee. 2019;26(1):207–12.

    Article  CAS  PubMed  Google Scholar 

  15. Koh IJ, Kim MS, Sohn S, Song KY, Choi NY, Jung H, In Y. Predictive factors for satisfaction after contemporary unicompartmental knee arthroplasty and high tibial osteotomy in isolated medial femorotibial osteoarthritis. Orthop Traumatol Surg Res. 2019;105(1):77–83.

    Article  PubMed  Google Scholar 

  16. Jin C, Song EK, Santoso A, Ingale PS, Choi IS, Seon JK. Survival and risk factor analysis of Medial Open Wedge High Tibial Osteotomy for Unicompartment knee osteoarthritis. Arthroscopy. 2020;36(2):535–43.

    Article  PubMed  Google Scholar 

  17. Obermeyer Z, Emanuel EJ. Predicting the Future - Big Data, Machine Learning, and Clinical Medicine. N Engl J Med. 2016;375(13):1216–9.

    Article  PubMed  PubMed Central  Google Scholar 

  18. Fontana MA, Lyman S, Sarker GK, Padgett DE, MacLean CH. Can Machine Learning algorithms Predict which patients will achieve minimally clinically important differences from total joint arthroplasty? Clin Orthop Relat Res. 2019;477(6):1267–79.

    Article  PubMed  PubMed Central  Google Scholar 

  19. Shohat N, Goswami K, Tan TL, Yayac M, Soriano A, Sousa R, Wouthuyzen-Bakker M, Parvizi J. 2020 Frank Stinchfield Award: identifying who will fail following irrigation and debridement for prosthetic joint infection. Bone Joint J 2020, 102–b(7_Supple_B):11–19.

  20. Jo C, Ko S, Shin WC, Han HS, Lee MC, Ko T, Ro DH. Transfusion after total knee arthroplasty can be predicted using the machine learning algorithm. Knee Surg Sports Traumatol Arthrosc. 2020;28(6):1757–64.

    Article  PubMed  Google Scholar 

  21. Martin RK, Wastvedt S, Pareek A, Persson A, Visnes H, Fenstad AM, Moatshe G, Wolfson J, Engebretsen L. Predicting Anterior Cruciate Ligament Reconstruction Revision: a machine learning analysis utilizing the Norwegian knee ligament Register. J Bone Joint Surg Am. 2022;104(2):145–53.

    Article  PubMed  Google Scholar 

  22. Park YB, Kim H, Lee HJ, Baek SH, Kwak IY, Kim SH. The Clinical Application of Machine Learning Models for Risk Analysis of Ramp Lesions in Anterior Cruciate Ligament injuries. Am J Sports Med. 2023;51(1):107–18.

    Article  PubMed  Google Scholar 

  23. Kunze KN, Polce EM, Clapp I, Nwachukwu BU, Chahla J, Nho SJ. Machine learning algorithms predict functional improvement after hip arthroscopy for femoroacetabular impingement syndrome in athletes. J Bone Joint Surg Am. 2021;103(12):1055–62.

    Article  PubMed  Google Scholar 

  24. Fujisawa Y, Masuhara K, Shiomi S. The effect of high tibial osteotomy on osteoarthritis of the knee. An arthroscopic study of 54 knee joints. Orthop Clin North Am. 1979;10(3):585–608.

    Article  CAS  PubMed  Google Scholar 

  25. Lobenhoffer P, Agneskirchner JD. Improvements in surgical technique of valgus high tibial osteotomy. Knee Surg Sports Traumatol Arthrosc. 2003;11(3):132–8.

    Article  PubMed  Google Scholar 

  26. Jacquet C, Pioger C, Khakha R, Steltzlen C, Kley K, Pujol N, Ollivier M. Evaluation of the minimal clinically important difference (MCID) of the KOOS, KSS and SF-12 scores after open-wedge high tibial osteotomy. Knee Surg Sports Traumatol Arthrosc. 2021;29(3):820–6.

    Article  PubMed  Google Scholar 

  27. Batailler C, Gicquel T, Bouguennec N, Steltzlen C, Tardy N, Cartier JL, Mertl P, Pailhé R, Rochcongar G, Fayard JM. A predictive score of high tibial osteotomy survivorship to help in surgical decision-making: the SKOOP score. Arch Orthop Trauma Surg. 2023;143(8):4843–51.

    Article  PubMed  Google Scholar 

  28. Ribeiro MT, Singh S, Guestrin C. Why Should I Trust You? Explaining the Predictions of Any Classifier. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining San Francisco, California, USA: Association for Computing Machinery; 2016: 1135–1144.

  29. Pedersen TL, Benesty M. Package ‘lime’. In.; 2018.

  30. Chang W, Cheng J, Allaire J, Xie Y, McPherson J. Shiny: web application framework for R. R Package Version. 2017; 1 (5). In.; 2021.

  31. Kim HJ, Shin JY, Lee HJ, Kim DH, Jo Y, Kim JW, Oh CW, Kyung HS. Does preoperative bone marrow edema affect clinical outcomes after medial open-wedge high tibial osteotomy? Knee Surg Relat Res. 2024;36(1):9.

    Article  PubMed  PubMed Central  Google Scholar 

  32. Lee DW, Han HS, Lee MC, Ro DH. Prediction of postoperative gait speed change after bilateral primary total knee arthroplasty in female patients using a machine learning algorithm. Orthop Traumatol Surg Res 2024:103842.

  33. Chong YY, Chan PK, Chan VWK, Cheung A, Luk MH, Cheung MH, Fu H, Chiu KY. Application of machine learning in the prevention of periprosthetic joint infection following total knee arthroplasty: a systematic review. Arthroplasty. 2023;5(1):38.

    Article  PubMed  PubMed Central  Google Scholar 

  34. Kim MS, Cho RK, Yang SC, Hur JH, In Y. Machine learning for detecting total knee Arthroplasty Implant loosening on plain radiographs. Bioeng (Basel) 2023, 10(6).

  35. Wu G, Cheligeer C, Southern DA, Martin EA, Xu Y, Leal J, Ellison J, Bush K, Williamson T, Quan H, et al. Development of machine learning models for the detection of surgical site infections following total hip and knee arthroplasty: a multicenter cohort study. Antimicrob Resist Infect Control. 2023;12(1):88.

    Article  PubMed  PubMed Central  Google Scholar 

  36. Barahona M, Guzmán M, Cartes S, Arancibia A, Mora J, Barahona M, Palma D, Hinzpeter J, Infante C, Barrientos C. Development of a machine learning model for anterior knee Pain after total knee arthroplasty with patellar preservation using Radiological variables. J Arthroplasty 2024.

  37. Jeong HW, Kim M, Choi HG, Park SY, Lee YS. Development of a machine learning model to predict lateral hinge fractures by analyzing patient factors before open wedge high tibial osteotomy. Knee Surg Sports Traumatol Arthrosc. 2023;31(8):3070–8.

    Article  PubMed  Google Scholar 

  38. Nam HS, Pei Yuik Ho J, Park SY, Cho JH, Lee YS. Development of a machine learning model for identifying the optimal situation favoring double-level osteotomy over single-level high tibial osteotomy. Knee. 2024;47:196–207.

    Article  PubMed  Google Scholar 

  39. Batailler C, Gicquel T, Bouguennec N, Steltzlen C, Tardy N, Cartier JL, Mertl P, Pailhé R, Rochcongar G, Fayard JM. A predictive score of high tibial osteotomy survivorship to help in surgical decision-making: the SKOOP score. Arch Orthop Trauma Surg 2022.

  40. Li HB, Nie S, Lan M, Tang ZM, Liao XG. The risk factors for early conversion to total knee arthroplasty after high tibial osteotomy. Comput Methods Biomech Biomed Engin 2020:1–7.

  41. Howells NR, Salmon L, Waller A, Scanelli J, Pinczewski LA. The outcome at ten years of lateral closing-wedge high tibial osteotomy: determinants of survival and functional outcome. Bone Joint J 2014, 96–b(11):1491–1497.

  42. Keurentjes JC, Fiocco M, So-Osman C, Onstenk R, Koopman-Van Gemert AW, Pöll RG, Kroon HM, Vliet Vlieland TP, Nelissen RG. Patients with severe radiographic osteoarthritis have a better prognosis in physical functioning after hip and knee replacement: a cohort-study. PLoS ONE. 2013;8(4):e59500.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  43. Schnurr C, Jarrous M, Güdden I, Eysel P, König DP. Pre-operative arthritis severity as a predictor for total knee arthroplasty patients’ satisfaction. Int Orthop. 2013;37(7):1257–61.

    Article  PubMed  PubMed Central  Google Scholar 

  44. Babis GC, An KN, Chao EY, Rand JA, Sim FH. Double level osteotomy of the knee: a method to retain joint-line obliquity. Clinical results. J Bone Joint Surg Am. 2002;84(8):1380–8.

    Article  PubMed  Google Scholar 

  45. Lobenhoffer P, Agneskirchner JD. [Osteotomy around the knee vs. unicondylar knee replacement]. Orthopade. 2014;43(10):923–9.

    Article  CAS  PubMed  Google Scholar 

  46. Tsukada S, Wakui M. Is overcorrection preferable for repair of degenerated articular cartilage after open-wedge high tibial osteotomy? Knee Surg Sports Traumatol Arthrosc. 2017;25(3):785–92.

    Article  PubMed  Google Scholar 

  47. El-Azab H, Glabgly P, Paul J, Imhoff AB, Hinterwimmer S. Patellar height and posterior tibial slope after open- and closed-wedge high tibial osteotomy: a radiological study on 100 patients. Am J Sports Med. 2010;38(2):323–9.

    Article  PubMed  Google Scholar 

  48. Coventry MB, Ilstrup DM, Wallrichs SL. Proximal tibial osteotomy. A critical long-term study of eighty-seven cases. J Bone Joint Surg Am. 1993;75(2):196–201.

    Article  CAS  PubMed  Google Scholar 

  49. Sprenger TR, Doerzbacher JF. Tibial osteotomy for the treatment of varus gonarthrosis. Survival and failure analysis to twenty-two years. J Bone Joint Surg Am. 2003;85(3):469–74.

    Article  PubMed  Google Scholar 

  50. Kim GW, Kang JK, Song EK, Seon JK. Increased joint obliquity after open-wedge high tibial osteotomy induces pain in the lateral compartment: a comparative analysis of the minimum 4-year follow-up outcomes using propensity score matching. Knee Surg Sports Traumatol Arthrosc. 2021;29(10):3495–502.

    Article  PubMed  Google Scholar 

  51. Rosso F, Rossi R, Cantivalli A, Pilone C, Bonasia DE. Joint Line Obliquity does not affect the outcomes of opening Wedge High Tibial Osteotomy at an average 10-Year follow-up. Am J Sports Med. 2022;50(2):461–70.

    Article  PubMed  Google Scholar 

  52. Schuster P, Geßlein M, Schlumberger M, Mayer P, Mayr R, Oremek D, Frank S, Schulz-Jahrsdörfer M, Richter J. Ten-year results of Medial Open-Wedge High Tibial Osteotomy and Chondral Resurfacing in severe medial osteoarthritis and Varus Malalignment. Am J Sports Med. 2018;46(6):1362–70.

    Article  PubMed  Google Scholar 

  53. Nakayama H, Schröter S, Yamamoto C, Iseki T, Kanto R, Kurosaka K, Kambara S, Yoshiya S, Higa M. Large correction in opening wedge high tibial osteotomy with resultant joint-line obliquity induces excessive shear stress on the articular cartilage. Knee Surg Sports Traumatol Arthrosc. 2018;26(6):1873–8.

    Article  PubMed  Google Scholar 

  54. Tu JV. Advantages and disadvantages of using artificial neural networks versus logistic regression for predicting medical outcomes. J Clin Epidemiol. 1996;49(11):1225–31.

    Article  CAS  PubMed  Google Scholar 

  55. Alvarez-Melis D, Jaakkola TS. On the robustness of interpretability methods. arXiv Preprint arXiv:180608049 2018.

  56. Cho HJ, Chang CB, Kim KW, Park JH, Yoo JH, Koh IJ, Kim TK. Gender and prevalence of knee osteoarthritis types in elderly koreans. J Arthroplasty. 2011;26(7):994–9.

    Article  PubMed  Google Scholar 

  57. Goshima K, Sawaguchi T, Shigemoto K, Iwai S, Nakanishi A, Ueoka K. Patellofemoral Osteoarthritis Progression and Alignment changes after Open-Wedge High Tibial Osteotomy do not affect clinical outcomes at Mid-term follow-up. Arthroscopy. 2017;33(10):1832–9.

    Article  PubMed  Google Scholar 

  58. Koh IJ, Kim MW, Kim JH, Han SY, In Y. Trends in high tibial osteotomy and knee arthroplasty utilizations and demographics in Korea from 2009 to 2013. J Arthroplasty. 2015;30(6):939–44.

    Article  PubMed  Google Scholar 

  59. Moseley JB, O’Malley K, Petersen NJ, Menke TJ, Brody BA, Kuykendall DH, Hollingsworth JC, Ashton CM, Wray NP. A controlled trial of arthroscopic surgery for osteoarthritis of the knee. N Engl J Med. 2002;347(2):81–8.

    Article  PubMed  Google Scholar 

Download references

Acknowledgements

None.

Funding

This work was supported by the Korea Medical Device Development Fund grant funded by the Korea government (the Ministry of Science and ICT, the Ministry of Trade, Industry and Energy, the Ministry of Health & Welfare, the Ministry of Food and Drug Safety) (Project Number: 1711139068, KMDF_PR_20210525-KD000001).

Author information

Author notes

  1. Hong Yeol Yang and Yong Gwan Shin contributed equally to this work.

Authors and Affiliations

  1. Department of Orthopaedic Surgery, Chonnam National University Medical School and Hospital, 322, Seoyang-ro 322 Hwasun-gun, Chonnam, 58128, Republic of Korea

    Hong Yeol Yang, Hyun Ho Shin, Ji Hoon Choi & Jong Keun Seon

  2. R&D Center, XRAI inc, Gwangju, 61186, Republic of Korea

    Yong Gwan Shin

Authors
  1. Hong Yeol Yang
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Yong Gwan Shin
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Hyun Ho Shin
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Ji Hoon Choi
    View author publications

    You can also search for this author in PubMed Google Scholar

  5. Jong Keun Seon
    View author publications

    You can also search for this author in PubMed Google Scholar

Contributions

HYY collected the data, performed the measurement and analysis, participated in the study design, and drafted the manuscript. YGS, HHS participated in the study design, supervised the analysis, and helped to draft the manuscript. JHC collected the data, performed the measurement. JKS designed the study, supervised the whole study process, and helped to draft and review the manuscript. All authors read and approved the final manuscript.

Corresponding author

Correspondence to Jong Keun Seon.

Ethics declarations

Ethical approval

All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. This study has been approved by the Institutional Review Board (IRB) of Chonnam National University Hwasun Hospital, and informed written consent for participation in the study was obtained.

Consent for publication

Not Applicable.

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Yang, H.Y., Shin, Y.G., Shin, H.H. et al. Factors to improve odds of success following medial opening-wedge high tibial osteotomy: a machine learning analysis. BMC Musculoskelet Disord 25, 323 (2024). https://doi.org/10.1186/s12891-024-07441-x

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s12891-024-07441-x

Keywords

BMC Musculoskeletal Disorders

ISSN: 1471-2474

Contact us