Machine learning model to predict 3-month nonunion(delayed union) of unstable distal clavicle fracture patients treated with open reduction and internal fixation

doi:10.21203/rs.3.rs-4280633/v1

Background Unstable distal clavicle fractures (UDCFs) exhibit a significant propensity for bone nonunion and are commonly managed through open reduction and internal fixation(ORIF). A notable drawback of this approach is the considerable incidence of fixation failure, including 3-month nonunion (also referred to as delayed union), and subsequent bone nonunion. Accurate prediction of union outcomes post-UDCF treatment could provide substantial evidence to inform and optimize intraoperative and postoperative strategies.

Methods Data were gathered from patients with UDCFs at Nanjing Luhe People's Hospital, China, from January 2015 to May 2023. The adverse outcome, characterized by the non-disappearance of the fracture line and uncontinuous callus formation, was defined as 3-month nonunion. To predict the risk of 3-month nonunion, five machine learning models were constructed, employing the area under the precision-recall curve (AUPRC) as the principal evaluation criterion. Subsequently, The Shapley additive explanations (SHAP) was further used to interpret the best model after comparing the models' performances.

Results In this investigation, a cohort of 248 patients was enrolled, among which 76 individuals (30.6%, median age: 55 years, males: 52) encountered adverse outcomes. The integration of variables such as the coracoclavicular distance（CCD）, anesthesia methodology, high-density lipoprotein (HDL) levels, and blood loss revealed that the CatBoost model was markedly more effective compared to four other predictive models, achieving an AUPRC of 0.801 (95% CI: 0.592-0.918). The SHAP values identified the CCD as the paramount predictor for assessing the risk of 3-month outcomes in patients with UDCFs within the Chinese demographic.

Conclusion The refined model incorporated four readily accessible variables, wherein the CCD, HDL levels, and blood loss were associated with an elevated risk of 3-month nonunion. Moreover, the CatBoost model has demonstrated a commendable capability in enhancing the prediction of 3-month outcomes for Chinese patients with UDCFs, effectively distinguishing individuals at high risk for delayed nonunion. This enhancement is pivotal for reinforcing intraoperative and postoperative management, thereby ameliorating patient prognoses.

Distal clavicle fracture

Machine learning

Prediction

Nonunion

Distal clavicle fractures constitute a prevalent type of upper limb fracture, representing 28% of all adult clavicle fractures [1]. Among these, 10% to 52% are classified as displaced and unstable, attributed to the multidirectional forces exerted by the robust trapezius, latissimus dorsi, and pectoral muscles[2]. Recent research indicates that the nonunion rate for unstable distal clavicle fractures (UDCFs) subjected to nonoperative treatment can reach up to 30% [3, 4]. Consequently, Open reduction and internal fixation (ORIF), incorporating anatomic locking plate and hook plate techniques, is widely recommended in China. Nevertheless, clinicians frequently encounters adverse outcomes, including internal fixation failure, re-fracture, 3-month nonunion (i.e., delayed union) and eventual bone nonunion [5-8]. Nonunion poses significant treatment challenges and incurs considerable financial burdens, with indirect costs like productivity losses being pivotal in escalating the total expenses for non-union patients. Hence, any strategy aimed at reducing healing duration, facilitating quicker return to work, and minimizing financial implications should be advocated for patients with fractures and non-unions. Accurately predicting the prognosis following UDCFs could yield compelling evidence to guide reasonable intraoperative and postoperative interventions, thereby abbreviating the disease course and alleviating discomfort.

Some studies reveal that various risk factors are associated with the adverse outcomes of delayed union and nonunion in fracture patients. These include physiological factors (e.g., advanced age, smoking), comorbidities (e.g., metabolic diseases, use of non-steroidal anti-inflammatory drugs, diabetes, nutritional deficiencies), and patient-specific factors (e.g., fracture pattern and location, displacement, infection, degree of bone loss, and severity of soft tissue injury) [9, 10] .Several scoring systems have been developed to predict fracture prognosis with the aim of improving risk stratification, such as the Constant-Murley Shoulder Function Scale, Neer Shoulder Function Scale, Rating Scale of the American Shoulder and Elbow Surgeons Score, Radiographic Union Score for Hip, and Radiographic Union Score for Tibia. However, certain scoring systems are not applicable to DCFs, and others primarily focus on shoulder function outcomes rather than bone union. Moreover, these scoring systems were developed based on data from White or Black populations, not Asians. In comparison to White patients, Chinese individuals generally have a smaller and slimmer physique. Consequently, these prognostic factors and models are not entirely accurate in predicting fracture union outcomes in the Chinese population, due to racial differences and various interventional factors that influence prognosis, including anatomical structure and surgical technique.

Recent research has demonstrated the growing application of ML models in the medical sector [11-16].ML models incorporate a variety of advanced algorithms to handle different data characteristics, including State-Action-Reward-State-Action , Random Forest Classifier, Deep Q Network, and Support Vector Machine[17].In comparison to traditional regression models, ML has been validated as an effective method for prognosis prediction in modeling due to its capacity to intricately analyze complex non-linear interrelations among variables [18, 19], and to enhance prediction accuracy through its superior algorithms, particularly when analyzing large datasets with numerous variables [20, 21]. Crucially, the widespread adoption of Electronic Patient Record systems and the extensive use of structured patient data have made sophisticated algorithmic modeling and bedside application feasible [22]. Aiming to enhance personalized fracture union strategies, we endeavored to devise and authenticate a prognosis model using ML methods. This model aims to forecast delayed union in Chinese UDCFs patients and subsequently compare the 3-month adverse outcomes with current clinical predictions.

Baseline characteristics

Figure 1 illustrates the inclusion of 248 patients treated for UDCFs from January 2015 to May 2023 in our analysis. In general, the overall median age was 55 years (IQR: 47-63 years), and with males constituting 60.1% of the cohort. Delayed union was observed in 76 patients (30.6%, median age: 55 years, number of male patients: 52), whereas 172 patients exhibited successful healing post-procedure (69.4%, median age: 54.5 years, number of male patients: 97). After data split, 173 patients were allocated to the training set and 75 to the test set. The detailed baseline information of the training set is shown in Table 1 and the training set and the testing set were similar in baseline characteristics.

Table 1 Baseline characteristics of patients in normal and delayed union groups in the training set

Variable	Normal	Delayed union	P-value
Variable	(n=120)	(n=53)	P-value
Age, years, median (IQR)	54.0 [44.8, 62.0]	56.0 [52.0, 62.0]	0.239
BMI, kg/m², median (IQR)	22.8 [21.0, 25.7]	23.2 [20.7, 25.7]	0.923
Height, cm, median (IQR)	168.0 [160.0, 172.0]	169.0 [163.0, 173.0]	0.462
Weight, kg, median (IQR)	64.0 [58.0, 72.3]	65.0 [56.5, 75.0]	0.475
Female, sex, n (%)	49 (40.8)	17 (32.1)	0.356
Smoking, n (%)	42 (35.0)	22 (41.5)	0.518
Drinking, n (%)	49 (40.8)	23 (43.4)	0.882
High education level, n (%)	16 (13.3)	2 (3.8)	0.103
Heavy physical labor, n (%)	50 (41.7)	25 (47.2)	0.612
Hypertension, n (%)	35 (29.2)	22 (41.5)	0.157
DM, n (%)	24 (20.0)	8 (15.1)	0.580
Other internal medicine diseases, n(%)	19 (15.8)	3 (5.7)	0.109
SBP, mmHg, median (IQR)	130.5 [122.0, 141.3]	138.0 [120.0, 151.0]	0.385
DBP, mmHg, median (IQR)	82.5 [76.0, 91.0]	85.0 [75.0, 93.0]	0.571
Emergency data, median (IQR)
Blood glu, mmol/L	5.7 [5.2, 6.4]	6.0 [5.6, 6.7]	0.089
WBC, ×10^9/L	7.1 [5.7, 8.6]	7.4 [5.7, 9.4]	0.623
Hemoglobin, g/L	126.0 [114.0, 140.3]	127.0 [117.0, 140.0]	0.666
Platelet, ×10^9/L	202.0 [166.0, 251.8]	203.0 [153.0, 256.0]	0.488
CRP, mg/L	7.6 [2.2, 17.7]	6.3 [1.6, 16.3]	0.500
D-dimer, μg/ml	0.9 [0.5, 2.0]	0.8 [0.5, 1.8]	0.394
Total protein, g/L	64.1 [60.4, 67.1]	63.6 [60.4, 67.1]	0.696
Albumin, g/L	39.0 [36.3, 41.4]	38.8 [36.8, 41.1]	0.877
LDH, U/L	191.0 [162.5, 220.5]	183.0 [162.0, 206.0]	0.212
Total cholesterol, mmol/L	4.1 [3.6, 4.7]	4.2 [3.6, 4.8]	0.895
Triglyceride, mmol/L	1.1 [0.8, 1.6]	1.1 [0.7, 1.3]	0.238
HDL, mmol/L	1.2 [1.0, 1.4]	1.3 [1.2, 1.5]	0.036*
LDL, mmol/L	2.5 [2.0, 3.0]	2.4 [1.9, 3.1]	0.746
Urea, mmol/L	4.9 [4.1, 6.0]	5.50 [4.40, 6.20]	0.239
Creatinine, μmol/L	65.5 [54.3, 75.6]	66.4 [53.7, 73.7]	0.865
Blood potassium, mmol/L	3.8 [3.5, 4.0]	3.8 [3.6, 4.0]	0.610
Blood calcium, mmol/L	2.2 [2.2, 2.3]	2.2 [2.2, 2.3]	0.596
Blood phosphorus, mmol/L	1.1 [0.9, 1.2]	1.0 [0.9, 1.1]	0.229
Blood magnesium, mmol/L	0.9 [0.8, 0.9]	0.9 [0.8, 0.9]	0.608
CCD, mm, median (IQR)	7.5 [6.2, 8.6]	10.2 [9.0, 11.7]	<0.001*
Comminuted fracture, n (%)	6 (5.0)	9 (17.0)	0.022*
Multiple fractures, n (%)	47 (39.2)	23 (43.4)	0.723
TBI, n (%)	10 (8.3)	2 (3.8)	0.445
Surgical duration, min, median (IQR)	70.0 [55.0, 85.0]	75.0 [60.0, 95.0]	0.080
Blood loss, ml, median (IQR)	50.0 [20.0, 100.0]	50.0 [50.0, 10.00]	0.047*
Cut length, cm, median (IQR)	10.0 [8.0, 10.0]	10.0 [8.0, 10.0]	0.359
Operation, n (%)			0.079
Outer panel	49 (40.8)	30 (56.6)
Hook plate	71 (59.2)	23 (43.4)
Anesthesia, n (%)			0.006*
Nerve block	76 (63.3)	21 (39.6)
General anesthesia	44 (36.7)	32 (60.4)
Postoperative ossotide dosage, mg, median (IQR)	0.0 [0.0, 420.0]	0.0 [0.0, 350.0]	0.883
Complication, n (%)	6 (5.0)	10 (18.9)	0.009*
Peripheral callus, n (%)	6 (5.0)	2 (3.8)	1.000

Abbreviations: BMI, body mass index; DM, diabetes mellitus; SBP, systolic blood pressure; DBP, diastolic blood pressure; WBC, white blood cell count; CRP, C-reactive protein; LDH, lactic dehydrogenase; HDL, high density lipoprotein; LDL, low density lipoprotein; CCD, coracoclavicular distance; TBI, traumatic brain injury; IQR, interquartile range

Note: *, P < 0.05

Feature selection

Seven of the 47 variables we gathered that may be connected to delayed union were statistically significant after univariate analysis. Furthermore, after exclusion of variables with zero coefficients through the LASSO process, we ultimately used four variables for model construction: CCD, anesthesia method, HDL, and blood loss. The detailed coefficients of these non-zero coefficient variables are detailed in Supplementary Table 1.

Model performance

All machine learning models that have been constructed have demonstrated similar performance, with no significant differences in AUROCs, as determined by the DeLong test (Supplementary Table 2). Among the five predictive models, the CatBoost model attained the greatest AUROC (0.863, 95% CI: 0.762-0.964) and AUPRC (0.801, 95% CI: 0.592-0.918), the highest accuracy (74.7%), F1 score (0.689), and PPV (55.3%) in the test set. AUROCs and AUPRCs for both the training and test sets are depicted in Figure 2. In addition, the calibration curve, presented in Figure 3, demonstrates a good alignment between the predictions of the CatBoost model and the actual outcomes. Evaluation metrics for all models are compiled in Table 2.

Table 2 The performance of the models in training set

		AUROC(95%CI)	AUPRC(95%CI)	Accuracy(%)	Sensitivity(%)	Specificity(%)	F1	PPV(%)	NPV(%)	Brier
LR	Training	0.862(0.795-0.929)	0.798(0.669-0.886)	83.8	71.7	89.2	0.731	74.5	87.7	0.126
LR	Test	0.844(0.739-0.948)	0.774(0.563-0.901)	72.0	60.9	76.9	0.571	53.8	81.6	0.138
RFC	Training	0.877(0.822-0.932)	0.728(0.594-0.831)	78.0	94.3	70.8	0.725	58.8	96.6	0.146
RFC	Test	0.847(0.744-0.949)	0.674(0.464-0.832)	70.7	95.7	59.6	0.667	51.2	96.9	0.151
XGB	Training	0.884(0.829-0.939)	0.771(0.640-0.865)	80.9	86.8	78.3	0.736	63.9	93.1	0.130
XGB	Test	0.852(0.753-0.952)	0.712(0.501-0.859)	73.3	91.3	65.4	0.677	53.8	94.4	0.137
MLP	Training	0.857(0.790-0.924)	0.785(0.655-0.876)	83.2	71.1	88.3	0.724	73.1	87.6	0.135
MLP	Test	0.838(0.731-0.945)	0.766(0.555-0.896)	70.7	60.9	75.0	0.560	51.9	81.2	0.139
Catboost	Training	0.920(0.879-0.960)	0.857(0.735-0.928)	82.7	90.6	79.2	0.762	65.8	95.0	0.122
Catboost	Test	0.863(0.762-0.964)	0.801(0.592-0.918)	74.7	91.3	67.3	0.689	55.3	94.6	0.143

Abbreviations: LR, logistic regression; RFC, random forest classifier; XGB, eXtreme gradient boosting; MLP, multi-layer perceptron; Catboost, category boosting; NPV, negative predictive value; PPV, positive predictive value; AUROC, the area under the receiver operating characteristic curve; AUPRC, the area under the precision-recall curve.

Model interpret

We offer Figure 4 as the global interpretation to the prediction behavior of the CatBoost model using SHAP. Our model identified CCD as the main driving factor based on the mean absolute SHAP value as Figure 4a shows. Figure 4b summarizes the direction of effects for each variable. Feature values are indicated by a spectrum with blue representing the lowest value and red color indicates higher risk of delayed union while blue color shows lower risk. As a result, the variables included in the CatBoost are all risk factors except for the anesthesia method of nerve block.

This study employed five distinct machine learning methodologies to develop a predictive model for the risk of 3-month nonunion in Chinese patients with UDCFs undergoing ORIF, based on four critical variables: CCD, anesthesia method (nerve block), HDL, and blood loss. Among these models, the CatBoost model emerged as the most effective, achieving an AUPRC of 0.801. Subsequently, risk stratification was executed using the optimal threshold defined by the CatBoost model, enabling the high-risk group identified by the model to benefit from preventive interventions targeting CCD、HDL and blood loss.

To begin with, a significant advantage of our model is its good performance. To our knowledge, accurately predicting the risk of nonunion in Chinese UDCFs patients following ORIF remains an unresolved challenge. In this study, we have effectively constructed a CatBoost model that achieves a notably high AUPRC of 0.801, a critical measure for assessing the predictive accuracy of models trained on imbalanced datasets. In addition, other evaluation metrics such as sensitivity and F1 score also achieve better performance. Consequently, our model presents a valuable asset for improving patient outcomes post-UDCFs, offering both medical and financial advantages by enabling the CatBoost to predict the likelihood of patients being at high risk of nonunion. In external and subsequent prospective studies, the clinical utility of this model will be evaluated in more detail and reported.

Despite the clinical efficacy of diverse surgical approaches in managing UDCFs, the issue of postoperative nonunion persists, influenced by numerous factors pre-, intra-, and post-operatively. Developing a prediction model for the risk of 3-month nonunion among UDCFs patients aids in identifying risk factors and facilitates early intervention to reduce the likelihood of delayed union. Qvist et al.[23]conducted a retrospective analysis of 150 clavicle fracture cases, assessing eight factors: age, sex, smoking status, comminuted fractures, fracture shortening exceeding 2 cm, Disabilities of the Arm, Shoulder and Hand (DASH) scores, Visual Analogue Scale (VAS) scores, and the VAS ratio (four-week VAS score/two-week VAS score). Their findings revealed that: (1) a rise in the absolute pain score 4 weeks post-fracture correlated with a heightened risk of nonunion at 6 months; (2) patients with a VAS ratio above 0.6 faced an 18-fold increased relative risk of nonunion compared to those with a VAS ratio below 0.6. Nicholson et al. [24] conducted a prospective follow-up of 200 patients with clavicle fractures, assessing them 6 weeks post-fracture to predict the likelihood of nonunion after 6 months. This study gathered data on 7 patient-related factors, 3 fracture characteristics, 5 outcomes from physical examinations, and 3 results from subjective functional evaluations. Analysis via a conditional stepwise regression model indicated that a QuickDASH score ≥ 40, absence of callus formation in X-ray images, and fracture displacement observed during physical examinations were predictors of nonunion. Similar to the aforementioned studies, our research serves as a risk alert for 3-month nonunion; however, we discovered that preoperative HDL levels, intraoperative blood loss, the usage of intraoperative general anesthesia, and postoperative CCD were linked to the risk of 3-month nonunion, diverge from those of previous research, likely due to its foundation on distinct risk factors and its focus on predicting nonunion risk in UDCFs patients undergoing ORIF. These predictors are readily ascertainable via laboratory tests, imaging examination, and surgical documentation. This significantly aids clinicians in making tailored and accurate treatment decisions for high-risk individuals, thereby mitigating the risk of delayed union and offering broad applicability in clinical settings.

Key Findings

Our findings indicate that an increased CCD is associated with a higher risk of 3-month nonunion. It is widely recognized that conservative treatment of UDCFs carries a higher risk of nonunion and may result in shoulder instability[25-27]. The coracoclavicular ligament is pivotal in stabilizing the acromioclavicular joint, consisting of the trapezoid and conoid ligaments. These components are instrumental in preventing forward and backward displacement, as well as upward rotation of the clavicle. Notably, the trapezoid ligament exhibits a significantly enhanced capacity to restrict backward displacement compared to the conoid ligament[28-30]. An increase in the CCD is predominantly attributed to injuries of the coracoclavicular ligament, which leads to stress concentration at the distal fracture end, thereby diminishing stability. Fleming et al. [31] highlighted that combined coracoclavicular ligament repair more effectively restores the CCD and stabilizes the acromioclavicular joint. Surgical strategies to restore the CCD typically encompass ORIF for anatomical realignment, Coracoclavicular Screw Fixation in cases where ligaments are intact, Ligament Reconstruction with grafts to replicate shoulder biomechanics, Hook Plate Fixation for provisional stability, and minimally invasive Arthroscopic Techniques utilizing various devices for less invasive interventions[32].

Second, our findings indicate that anesthesia impacts bone nonunion three months post-surgery, with nerve blocks potentially serving as a protective factor against nonunion outcomes. A variety of nerve blocks are employed in clavicular surgeries, such as interscalene, deep and superficial cervical plexus blocks, pectoralis nerve blocks, and selective supraclavicular nerve blocks [33-36]. An increasing body of literature advocates for the use of regional nerve blocks for anesthesia in patients with clavicle fractures . Nerve blocks alleviate pain by inhibiting nerve cells from transmitting electrical signals that the brain interprets as pain. This effect is accomplished through anesthetics that obstruct the nerve receptors tasked with detecting damage or injury. Additionally, nerve blocks diminish inflammation, which aids in nerve recovery and further lessens pain[1,37]. Certain studies have suggested that mitigating absolute pain may be advantageous for the union outcome following unstable distal clavicle fractures[37]. Ultrasound-guided interscalene brachial plexus block, when used in conjunction with superficial cervical nerve block, has been demonstrated to be a secure and efficacious method of anesthesia for clavicle procedures, as opposed to general anesthesia [36]. A separate retrospective cohort study revealed that employing regional anesthesia in ORIF for clavicle fractures correlated with a higher rate of same-day discharge [38]. Ryan et al.[39] discovered that the combined use of brachial plexus regional anesthesia with a modified superficial cervical plexus block constitutes a dependable and effective approach, potentially offering advantages in terms of anesthesia initiation and overall case duration. This technique appears to be a viable substitute for general anesthesia accompanied by an interscalene brachial plexus block. Our retrospective analysis of UDCFs patients who underwent ORIF contributes to the body of evidence suggesting that opting for nerve block anesthesia could diminish the risk of nonunion three months post-surgery.

Third, HDL has been the focus of extensive research for many years due to its crucial role in atherosclerosis prevention. Indeed, it may also have a significant impact on the pathology and management of conditions such as morbid obesity, nonalcoholic fatty liver disease, type 2 diabetes, and various central nervous system disorders [40]. Researchers have uncovered a novel function of HDL in the development of degenerative and metabolic bone diseases through experimental mouse studies[41, 42]. The findings indicate that reduced and dysfunctional HDL levels may contribute to a higher incidence of these diseases by influencing the molecular mechanisms involved in bone synthesis and degradation. Studies have reported that elevated HDL-C levels are linked to osteoporosis[43] and an increased risk of fractures [44, 45]. The relationship between high HDL-C levels, low bone mineral density, and osteoporosis [46], along with a genome-wide association study linking high HDL-C to low bone mineral density [47], may offer a pathophysiological explanation. A case-control study by Quan et al. [48] identified osteoporosis as an independent risk factor for bone nonunion. Furthermore, another study presented high-quality evidence of a negative correlation between HDL-C levels and bone mineral density [49]. Integrating these findings with our research suggests that an increased HDL-C levels may be associated with a heightened risk of bone nonunion. However, this view conflicts with other opinions that assert no association between HDL-C levels and osteoporosis [50]. Thus, whether high HDL levels exacerbates the risk of bone nonunion requires further investigation.

Fourth, our research identified a correlation between intraoperative blood loss and the risk of 3-month nonunion in patients with UDCFs. Research has demonstrated that inadequate blood supply at the fracture site contributes to delayed bone union or nonunion. The formation of callus beyond the fracture site predominantly relies on the vascular supply to the periosteum and adjacent soft tissues. Furthermore, factors like trauma, excessive displacement at the fracture site, comminuted fractures, and both open and closed soft tissue injuries can impair the local blood supply. Surgical interventions can further influence the vascular supply to the fracture site, necessitating procedures like extended wound exposure, periosteal stripping, and varying levels of damage to surrounding soft tissues. Following a fracture, the blood supply to the affected end is compromised, and full restoration of vascular supply to this area may take over 6 months [51]. Significant blood loss during surgery could result in an inadequate blood supply to the fracture site, hinder callus formation, and ultimately contribute to nonunion. To mitigate the risk of postoperative nonunion in UDCFs patients, a multifaceted approach is essential for managing intraoperative hemorrhage: (1) Minimizing surgical trauma and employing minimally invasive techniques are crucial for preserving the peripheral blood supply [32,52]; (2) The application of advanced hemostatic methods, including electrocoagulation and topical hemostatic agents, is recommended [52]; (3) It is imperative to secure stable fracture fixation to minimize displacement at the fracture site; (4) Postoperative care should encompass optimal nutrition and the initiation of controlled activities to enhance circulation; (5) The administration of anti-fibrinolytic agents, for instance, triamcinolone, is advocated to curtail bleeding [53]. Employing these comprehensive strategies is beneficial for sustaining an adequate blood supply at the fracture site, thereby diminishing the likelihood of nonunion.

Limitations

This study is subject to certain limitations that warrant consideration in subsequent research. Firstly, similar to numerous retrospective analyses, our model's efficacy requires prospective validation. Secondly, our evaluation of bone nonunion in UDCFs patients was confined to the initial 3 months post-ORIF, without subsequent follow-ups, despite the acknowledgment that delayed fracture union or nonunion could extend beyond this period. Thirdly, the dataset for this research was sourced from Nanjing Luhe People's Hospital, which may not provide as comprehensive a perspective as a multicenter study would. Although external validation data were utilized, these too originated from the same institution, lacking validation cohorts from diverse regions and nations. Consequently, the broader adoption and implementation of our model across various regions and countries may face challenges. Subsequently, additional datasets and prospective multicenter clinical trials are essential to confirm the reliability of our findings and model, further research should delve into the correlation between CCD and other variables concerning the ultimate outcome, adjusting for potential confounders such as the level of surgeon experience and patient ethnic diversity.

To our knowledge, this study introduces the inaugural machine learning model designed to forecast the risk of 3-month nonunion in Chinese patients with UDCFs undergoing ORIF, and shows the CCD, HDL levels, and blood loss are associated with an increased risk of 3-month nonunion. These insights advocate for surgeons to prioritize the assessment and adjustment of CCD to ensure mechanical stability. Compared with traditional evaluation systems encompassing clinical examination and serological markers, the CatBoost model enhances the prediction of 3-month outcomes in Chinese UDCFs patients. It demonstrates a commendable proficiency in identifying patients at elevated risk for delayed nonunion, underscoring its potential for clinical application.

Study Population

This retrospective study was performed in the Nanjing Luhe People's Hospital, China, where electronic health records from the orthopedic unit for 264 adult patients diagnosed with UDCFs were collected and analyzed, spanning from January 2015 to May 2023. The primary outcome was the union of UDCFs within the first three months post-operation, as determined by clinical signs of healing and corroborated by shoulder X-rays images. The exclusion criteria were patients with incomplete clinical or pre-treatment data, those receiving conservative treatment, individuals under 18, and patients with concurrent neurovascular injuries. We discarded all variables with more than 30% missing values from subsequent analyses. A total of 248 patients who underwent ORIF were included in the final analysis.

Predictors and outcomes

In this study, the candidate variables for our model included demographic factors such as age, sex, smoking status, drinking status, level of education, and labor intensity, medical histories like diabetes mellitus, hypertension, anemia, and hypoproteinemia, baseline characteristics including height, weight, diastolic and systolic blood pressures, presence of comminuted or multiple fractures, and results of blood biochemical tests, as well as surgical details such as the method of operation, the coracoclavicular distance（CCD）, type of anesthesia, duration of surgery, volume of blood loss, length of incision, and use of osteogenic stimuli, and postoperative outcomes, including 3-month fracture union status and complications.

We sought to predict a binary outcome, either the achievement of union within 3 months or a delayed union, and evaluated the status of fracture union during the initial 3 months postoperatively by analyzing shoulder X-rays images, focusing on the disappearance of the fracture line and the formation of continuous callus as indicators of union. CCD was defined as the measure between the coracoid process and the inferior aspect of the clavicle[30], determined from shoulder X-ray images taken within 7 days after surgery. In addition, Educational attainment was categorized into two levels: high, for those with college education, and low, for those without.

Data pre-processing

The entire dataset was divided into the training and test sets using a random allocation in a 7:3 ratio. Two methods were mainly used to deal with missing data in both training and test sets. Variables with a missing data rate exceeding 30% were removed outright from the dataset, and for the rest of remaining variables exhibiting missing values, multiple imputation employing predictive mean matching (PMM) was executed. The One-Hot Encoding technique was applied to transform multiple categorical variables into dummy variables. Whereas continuous variables underwent normalization by a z-score.

Model development and validation

In order to select the variables that would eventually develop models, we used the Least Absolute Shrinkage and Selection Operator (LASSO) algorithm to process variables that demonstrated statistical significance in univariate analysis. The variables retaining nonzero coefficients following the LASSO were selected as the final variables for model development.

Trials of five machine learning classifiers—logistic regression (LR), Random Forest Classifier (RFC), eXtreme Gradient Boosting (XGB), Multi-Layer Perceptron (MLP), and category boosting (CatBoost)—were employed to generate five models for the prediction of delayed union outcome in the training set. The optimal parameters for each algorithm were determined by the grid search algorithm coupled with 10-fold cross-validation. The test set in our study provided data for internal validation to demonstrate the generalization of the machine learning model, ensuring the model, developed from the training data, maintained its applicability.

Model evaluation

Due to the imbalance within our dataset, AUPRC was selected as the principal evaluation metric, which has been proven to be superior to AUROC, especially in the context of imbalanced data . In addition, accuracy, recall, precision, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and the F1 score were served as metrics to evaluate the models. A calibration curve analysis was also carried out to visually evaluate the degree of agreement between the model's predicted and actual probabilities.

Model interpret

SHAP method was used to evaluate variable importance and provide an explanation for the predictions generated by ML algorithms in order to gain a better understanding of the machine learning model's predictive process. SHAP is a useful tool that helps boost the model's credibility and acceptability in clinical environments, which illustrates how each variable contributes to and influences the risk of delayed union.

Statistical analysis

Baseline characteristics were compared between the control group and the delayed union cohort. The Shapiro-Wilk test was used firstly to determine whether continuous variables were normally distributed. Descriptive statistics were presented as means with standard deviations (for normally distributed variable), or medians with interquartile ranges (IQR) (for non-normally distributed variable), and frequency (percentages) for categorical variables, respectively. We used the Student's t-test or the Mann-Whitney U-test to assess the differences in continuous variables, while for categorical variables, the Chi-squared or Fisher's exact tests were performed. The above statistical analysis conducted with the statistical package R, version 4.1.3 (R Development Core Team, Auckland, New Zealand), and p-values less than 0.05 was considered statistically significant.

UDCFs Unstable distal clavicle fractures

(ORIF Open reduction and internal fixation)

AUPRC Area Under the Precision-Recall Curve

SHAP The SHapley Additive exPlanations

CCD Coracoclavicular distance

HDL High-density lipoprotein

LASSO Least Absolute Shrinkage and Selection Operator

LR Logistic regression

RFC Random Forest Classifier

XGB eXtreme Gradient Boosting

MLP Multi-Layer Perceptron

CatBoost Category boosting

PPV Positive predictive value

NPV Negative predictive value

IQR Interquartile ranges

ROC Receiver operating characteristic

PRC Precision-Recall Curve

DASH Disabilities of the Arm, Shoulder and Hand

VAS Visual Analogue Scale

Author contributions

CKM ,WL and LML collected data and wrote the manuscript. JJZ and CKM conceived and designed the study, and critically revised the manuscript. WL and KZH performed data collection and statistical analysis. All authors have revised the article critically for important intellectual content and approved the final version to be published.

Funding

This work was supported by Nanjing Medical Science and Technical Development Foundation (Grant Number YKK23237), National Natural Science Foundation of China (Grant Number 82173899), Jiangsu Pharmaceutical Association (Grant Number H202108, A2021024, Q202202, JY202207, A202309, Z04JKM2023E040), Open project of International Joint Laboratory of Recombinant Drug Protein Expression System in Henan Province (No.KFKTYB202208).

Data availability

The datasets analyzed during the current study are available from the corresponding authors on reasonable request.

Ethics approval and consent to participate

Ethical approval for this study was obtained from the Nanjing Luhe People's Hospital, Nanjing, China (number LHLL2022024). Because of the retrospective study, the requirement for informed consent was exempted.

Competing interest

The authors declare no competing interests.

Acknowledgments

We thank Zhen Wang and Qianming Du for the guidance in the data collection and sample size calculation.

Panagopoulos A, Tatani I, Kattou C, Kouzelis A, Kolios K, Athinodorou IP, et al. Clinical and Radiological Outcomes of Closed-Loop, Double Button, Coracoclavicular Fixation for Extralateral (Neer Type IIC) Fractures of the Distal Clavicle. Cureus. 2022;14(5):e25228. doi: 10.7759/cureus.25228.
Xiong J, Chen JH, Dang Y, Zhang DY, Fu ZG, Zhang PX. Treatment of unstable distal clavicle fractures (Neer type II): A comparison of three internal fixation methods. J Int Med Res. 2018 ;46(11):4678-4683. doi: 10.1177/0300060518788245.
Singh A, Schultzel M, Fleming JF, Navarro RA. Complications after surgical treatment of distal clavicle fractures. Orthop Traumatol Surg Res. 2019 ;105(5):853-859. doi: 10.1016/j.otsr.2019.03.012.
Uittenbogaard SJ, van Es LJM, den Haan C, van Deurzen DFP, van den Bekerom MPJ. Outcomes, Union Rate, and Complications After Operative and Nonoperative Treatments of Neer Type II Distal Clavicle Fractures: A Systematic Review and Meta-analysis of 2284 Patients. Am J Sports Med. 2023;51(2):534-544. doi: 10.1177/03635465211053336.
Li D, Qiao R, Yang N, Zhang K, Zhu Y, Song Z. Clinical analysis of the acromial height-measuring device combined with new-type clavicular hook plate and standard clavicular hook plate in the treatment of Neer type II distal clavicle fractures. J Orthop Surg Res. 2022;17(1):448. doi: 10.1186/s13018-022-03338-4.
Chen MJ, DeBaun MR, Salazar BP, Lai C, Bishop JA, Gardner MJ. Hook versus locking plate fixation for Neer type-II and type-V distal clavicle fractures: a retrospective cohort study. Eur J Orthop Surg Traumatol. 2020 ;30(6):1027-1031. doi: 10.1007/s00590-020-02658-7.
Stegeman SA, Nacak H, Huvenaars KH, Stijnen T, Krijnen P, Schipper IB. Surgical treatment of Neer type-II fractures of the distal clavicle: a meta-analysis. Acta Orthop. 2013;84(2):184-90. doi: 10.3109/17453674.2013.786637.
Lopiz Y, Checa P, García-Fernández C, Valle J, Vega ML, Marco F. Complications with the clavicle hook plate after fixation of Neer type II clavicle fractures. Int Orthop. 2019;43(7):1701-1708. doi: 10.1007/s00264-018-4108-3.
Hak DJ, Fitzpatrick D, Bishop JA, Marsh JL, Tilp S, Schnettler R, et al. Delayed union and nonunions: epidemiology, clinical issues, and financial aspects. Injury. 2014;45 Suppl 2:S3-7. doi: 10.1016/j.injury.2014.04.002.
Zhu Y, Hu J, Zhan T, Zhu K, Zhang C. Refracture after plate removal of midshaft clavicle fractures after bone union-incidence, risk factors, management and outcomes. BMC Musculoskelet Disord. 2023;24(1):308. doi: 10.1186/s12891-023-06391-0.
Kamal H, Lopez V, Sheth SA. Machine Learning in Acute Ischemic Stroke Neuroimaging. Front Neurol. 2018;9:945. doi: 10.3389/fneur.2018.00945.
Xiong W, Chen T, Li J, Xiang L, Zhang C, Xiang L, et al. Interpretable machine learning model to predict rupture of small intracranial aneurysms and facilitate clinical decision. Neurol Sci. 2022 ;43(11):6371-6379. doi: 10.1007/s10072-022-06351-x.
Hu Y, Yang T, Zhang J, Wang X, Cui X, Chen N, et al. Dynamic Prediction of Mechanical Thrombectomy Outcome for Acute Ischemic Stroke Patients Using Machine Learning. Brain Sci. 2022;12(7):938. doi: 10.3390/brainsci12070938.
Desautels T, Das R, Calvert J, Trivedi M, Summers C, Wales DJ, et al. Prediction of early unplanned intensive care unit readmission in a UK tertiary care hospital: a cross-sectional machine learning approach. BMJ Open. 2017;7(9):e017199. doi: 10.1136/bmjopen-2017-017199.
Lin X, Zheng X, Zhang J, Cui X, Zou D, Zhao Z,et al. Machine learning to predict futile recanalization of large vessel occlusion before and after endovascular thrombectomy. Front Neurol. 2022;13:909403. doi: 10.3389/fneur.2022.909403.
Quesada JA, Lopez-Pineda A, Gil-Guillén VF, Durazo-Arvizu R, Orozco-Beltrán D, López-Domenech A, et al. Machine learning to predict cardiovascular risk. Int J Clin Pract. 2019;73(10):e13389. doi: 10.1111/ijcp.13389.
Li X, Pan X, Jiang C, Wu M, Liu Y, Wang F, et al. Predicting 6-Month Unfavorable Outcome of Acute Ischemic Stroke Using Machine Learning. Front Neurol. 2020;11:539509. doi: 10.3389/fneur.2020.539509.
Deo RC. Machine Learning in Medicine. Circulation. 2015;132(20):1920-30. doi: 10.1161/CIRCULATIONAHA.115.001593.
Lin S, Zou Y, Hu J, Xiang L, Guo L, Lin X, et al. Development and assessment of machine learning models for predicting recurrence risk after endovascular treatment in patients with intracranial aneurysms. Neurosurg Rev. 2022 ;45(2):1521-1531. doi: 10.1007/s10143-021-01665-4.
Goldstein BA, Navar AM, Carter RE. Moving beyond regression techniques in cardiovascular risk prediction: applying machine learning to address analytic challenges. Eur Heart J. 2017 ;38(23):1805-1814. doi: 10.1093/eurheartj/ehw302.
Goto T, Jo T, Matsui H, Fushimi K, Hayashi H, Yasunaga H. Machine Learning-Based Prediction Models for 30-Day Readmission after Hospitalization for Chronic Obstructive Pulmonary Disease. COPD. 2019;16(5-6):338-343. doi: 10.1080/15412555.2019.1688278.
Li X, Wu M, Sun C, Zhao Z, Wang F, Zheng X, et al. Using machine learning to predict stroke-associated pneumonia in Chinese acute ischaemic stroke patients. Eur J Neurol. 2020;27(8):1656-1663. doi: 10.1111/ene.14295.
Qvist AH, Væsel MT, Jensen CM, Jakobsen T, Jensen SL. Minimal Pain Decrease Between 2 and 4 Weeks After Nonoperative Management of a Displaced Midshaft Clavicle Fracture Is Associated with a High Risk of Symptomatic Nonunion. Clin Orthop Relat Res. 2021;479(1):129-138. doi: 10.1097/CORR.0000000000001411.
Nicholson JA, Clement ND, Clelland AD, MacDonald D, Simpson AHRW, Robinson CM. Displaced Midshaft Clavicle Fracture Union Can Be Accurately Predicted with a Delayed Assessment at 6 Weeks Following Injury: A Prospective Cohort Study. J Bone Joint Surg Am. 2020;102(7):557-566. doi: 10.2106/JBJS.19.00955.
Lian J, Chan FJ, Levy BJ. Classification of Distal Clavicle Fractures and Indications for Conservative Treatment. Clin Sports Med. 2023;42(4):685-693. doi: 10.1016/j.csm.2023.05.007.
Oh JH, Kim SH, Lee JH, Shin SH, Gong HS. Treatment of distal clavicle fracture: a systematic review of treatment modalities in 425 fractures. Arch Orthop Trauma Surg. 2011;131(4):525-33. doi: 10.1007/s00402-010-1196-y.
Robinson CM, Cairns DA. Primary nonoperative treatment of displaced lateral fractures of the clavicle. J Bone Joint Surg Am. 2004;86(4):778-82. doi: 10.2106/00004623-200404000-00016.
Lee S, Bedi A. Shoulder acromioclavicular joint reconstruction options and outcomes. Curr Rev Musculoskelet Med. 2016 Dec;9(4):368-377. doi: 10.1007/s12178-016-9361-8.
Li Q, Hsueh PL, Chen YF. Coracoclavicular ligament reconstruction: a systematic review and a biomechanical study of a triple endobutton technique. Medicine (Baltimore). 2014;93(28):e193. doi: 10.1097/MD.0000000000000193.
Gstettner C, Tauber M, Hitzl W, Resch H. Rockwood type III acromioclavicular dislocation: surgical versus conservative treatment. J Shoulder Elbow Surg. 2008;17(2):220-5. doi: 10.1016/j.jse.2007.07.017.
Fleming MA, Dachs R, Maqungo S, du Plessis JP, Vrettos BC, Roche SJ. Angular stable fixation of displaced distal-third clavicle fractures with superior precontoured locking plates. J Shoulder Elbow Surg. 2015;24(5):700-4. doi: 10.1016/j.jse.2014.09.024.
Panagopoulos A, Solou K, Tatani I, Triantafyllopoulos IK, Lakoumentas J, Kouzelis A, et al. What is the optimal surgical treatment for Neer type IIB (IIC) distal clavicle fractures? A systematic review and meta-analysis. J Orthop Surg Res. 2022;17(1):215. doi: 10.1186/s13018-022-03108-2.
Balaban O, Dülgeroğlu TC, Aydın T. Ultrasound-Guided Combined Interscalene-Cervical Plexus Block for Surgical Anesthesia in Clavicular Fractures: A Retrospective Observational Study. Anesthesiol Res Pract. 2018 ;2018:7842128. doi: 10.1155/2018/7842128.
Vandepitte C, Latmore M, O'Murchu E, Hadzic A, Van de Velde M, Nijs S. Combined interscalene-superficial cervical plexus blocks for surgical repair of a clavicular fracture in a 15-week pregnant woman. Int J Obstet Anesth. 2014 ;23(2):194-5. doi: 10.1016/j.ijoa.2013.10.004.
Fugelli CG, Westlye ET, Ersdal H, Strand K, Bjørshol C. Combined Interscalene Brachial Plexus and Superficial Cervical Plexus Nerve Block for Midshaft Clavicle Surgery: A Case Series. AANA J. 2019 ;87(5):374-378.
Banerjee S, Acharya R, Sriramka B. Ultrasound-Guided Inter-scalene Brachial Plexus Block with Superficial Cervical Plexus Block Compared with General Anesthesia in Patients Undergoing Clavicular Surgery: A Comparative Analysis. Anesth Essays Res. 2019;13(1):149-154. doi: 10.4103/aer.AER_185_18.
Albi-Feldzer A, Dureau S, Ghimouz A, Raft J, Soubirou JL, Gayraud G, et al. Preoperative Paravertebral Block and Chronic Pain after Breast Cancer Surgery: A Double-blind Randomized Trial. Anesthesiology. 2021;135(6):1091-1103. doi: 10.1097/ALN.0000000000003989.
Beletsky A, Burton BN, Swisher MW, Finneran Iv JJ, Gabriel RA. Utilization of regional anesthesia for open reduction and internal fixation of clavicular fractures is associated with increased same-day discharge and not readmissions. Reg Anesth Pain Med. 2020;45(9):716-719. doi: 10.1136/rapm-2020-101567.
Ryan DJ, Iofin N, Furgiuele D, Johnson J, Egol K. Regional anesthesia for clavicle fracture surgery is safe and effective. J Shoulder Elbow Surg. 2021;30(7):e356-e360. doi: 10.1016/j.jse.2020.10.009.
Constantinou C, Karavia EA, Xepapadaki E, Petropoulou PI, Papakosta E, Karavyraki M, et al. Advances in high-density lipoprotein physiology: surprises, overturns, and promises. Am J Physiol Endocrinol Metab. 2016 ;310(1):E1-E14. doi: 10.1152/ajpendo.00429.2015.
Triantaphyllidou IE, Kalyvioti E, Karavia E, Lilis I, Kypreos KE, Papachristou DJ. Perturbations in the HDL metabolic pathway predispose to the development of osteoarthritis in mice following long-term exposure to western-type diet. Osteoarthritis Cartilage. 2013;21(2):322-30. doi: 10.1016/j.joca.2012.11.003.
Blair HC, Kalyvioti E, Papachristou NI, Tourkova IL, Syggelos SA, Deligianni D,et al. Apolipoprotein A-1 regulates osteoblast and lipoblast precursor cells in mice. Lab Invest. 2016;96(7):763-72. doi: 10.1038/labinvest.2016.51.
Li S, Guo H, Liu Y, Wu F, Zhang H, Zhang Z, et al. Relationships of serum lipid profiles and bone mineral density in postmenopausal Chinese women. Clin Endocrinol (Oxf). 2015;82(1):53-8. doi: 10.1111/cen.12616.
Hussain SM, Ebeling PR, Barker AL, Beilin LJ, Tonkin AM, McNeil JJ. Association of Plasma High-Density Lipoprotein Cholesterol Level With Risk of Fractures in Healthy Older Adults. JAMA Cardiol. 2023 ;8(3):268-272. doi: 10.1001/jamacardio.2022.5124.
Papachristou NI, Blair HC, Kypreos KE, Papachristou DJ. High-density lipoprotein (HDL) metabolism and bone mass. J Endocrinol. 2017 ;233(2):R95-R107. doi: 10.1530/JOE-16-0657.
Lisnyansky M, Kapelushnik N, Ben-Bassat A, Marom M, Loewenstein A, Khananshvili D, et al. Reduced Activity of Geranylgeranyl Diphosphate Synthase Mutant Is Involved in Bisphosphonate-Induced Atypical Fractures. Mol Pharmacol. 2018;94(6):1391-1400. doi: 10.1124/mol.118.113670.
Chen H, Shao Z, Gao Y, Yu X, Huang S, Zeng P. Are blood lipids risk factors for fracture? Integrative evidence from instrumental variable causal inference and mediation analysis using genetic data. Bone. 2020 ;131:115174. doi: 10.1016/j.bone.2019.115174.
Quan K, Xu Q, Zhu M, Liu X, Dai M. Analysis of Risk Factors for Non-union After Surgery for Limb Fractures: A Case-Control Study of 669 Subjects. Front Surg. 2021;8:754150. doi: 10.3389/fsurg.2021.754150.
Niu P, Li H, Liu D, Zhang YF, Liu Y, Liang C. Association Between HDL-C and Bone Mineral Density: An Cross-Sectional Analysis. Int J Gen Med. 2021;14:8863-8872. doi: 10.2147/IJGM.S334972.
Kan B, Zhao Q, Wang L, Xue S, Cai H, Yang S. Association between lipid biomarkers and osteoporosis: a cross-sectional study. BMC Musculoskelet Disord. 2021;22(1):759. doi: 10.1186/s12891-021-04643-5.
Reed AA, Joyner CJ, Isefuku S, Brownlow HC, Simpson AH. Vascularity in a new model of atrophic nonunion. J Bone Joint Surg Br. 2003;85(4):604-10. doi: 10.1302/0301-620x.85b4.12944.
Taghinia AH. Pediatric Replantation and Revascularization. Hand Clin. 2019 ;35(2):155-178. doi: 10.1016/j.hcl.2018.12.006.
Fourman MS. CORR Insights®: Minimal Pain Decrease Between 2 and 4 Weeks after Nonoperative Management of a Displaced Midshaft Clavicle Fracture Is Associated with a High Risk of Symptomatic Nonunion. Clin Orthop Relat Res. 2021;479(1):139-141. doi: 10.1097/CORR.0000000000001606.

No competing interests reported.

Machine learning model to predict 3-month nonunion(delayed union) of unstable distal clavicle fracture patients treated with open reduction and internal fixation

Status:

Version 1

Abstract

Figures

Introduction

Results

Discussion

Conclusion

Patients and Methods

Abbreviations

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1