Luan FF, Chen JY, Zheng YZ, Hu MH, Huang F, Zheng CX. Efficacy of a machine learning model integrating clinical-psychosocial factors in predicting posttraumatic bone nonunion and bone defects. World J Psychiatry 2026; 16(9): 119455 [DOI: 10.5498/wjp.119455]
Corresponding Author of This Article
Chen-Xiao Zheng, PhD, Zhongshan Hospital of Traditional Chinese Medicine Affiliated to Guangzhou University of Traditional Chinese Medicine, The Tenth Clinical Medical College of Guangzhou University of Traditional Chinese Medicine, No. 3 Kangxin Road, West District, Zhongshan 528400, Guangdong Province, China. cuokoo1973@163.com
Research Domain of This Article
Psychology
Article-Type of This Article
research-article
Open-Access Policy of This Article
This article is an open-access article which was selected by an in-house editor and fully peer-reviewed by external reviewers. It is distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/
Baishideng Publishing Group Inc, 7041 Koll Center Parkway, Suite 160, Pleasanton, CA 94566, USA
Share the Article
Luan FF, Chen JY, Zheng YZ, Hu MH, Huang F, Zheng CX. Efficacy of a machine learning model integrating clinical-psychosocial factors in predicting posttraumatic bone nonunion and bone defects. World J Psychiatry 2026; 16(9): 119455 [DOI: 10.5498/wjp.119455]
Fei-Fan Luan, Jia-Yi Chen, Yu-Zhong Zheng, Min-Hua Hu, Chen-Xiao Zheng, Zhongshan Hospital of Traditional Chinese Medicine Affiliated to Guangzhou University of Traditional Chinese Medicine, The Tenth Clinical Medical College of Guangzhou University of Traditional Chinese Medicine, Zhongshan 528400, Guangdong Province, China
Fei-Fan Luan, Feng Huang, The First Clinical Medical School, Guangzhou University of Chinese Medicine, Guangzhou 510405, Guangdong Province, China
Co-corresponding authors: Feng Huang and Chen-Xiao Zheng.
Author contributions: Luan FF designed the research and wrote the first manuscript; Luan FF, Chen JY, and Zheng YZ contributed to conceiving the research and analyzing data; Luan FF and Hu MH conducted the analysis; Huang F and Zheng CX provided guidance for the research; they contributed equally to this manuscript and are co-corresponding authors; all authors reviewed and approved the final manuscript.
AI contribution statement: The authors declare that no AI tools were used in the development or writing of this manuscript and take full responsibility for its integrity, accuracy, and originality.
Institutional review board statement: This study was approved by the Ethics Committee of Zhongshan Hospital of Traditional Chinese Medicine Affiliated to Guangzhou University of Traditional Chinese Medicine, No. 2026ZSZY-LL-KY-033.
Informed consent statement: The requirement for patients’ informed consent for this study was waived due to its retrospective nature.
Conflict-of-interest statement: All the authors report no relevant conflicts of interest for this article.
Data sharing statement: No additional data are available.
Corresponding author: Chen-Xiao Zheng, PhD, Zhongshan Hospital of Traditional Chinese Medicine Affiliated to Guangzhou University of Traditional Chinese Medicine, The Tenth Clinical Medical College of Guangzhou University of Traditional Chinese Medicine, No. 3 Kangxin Road, West District, Zhongshan 528400, Guangdong Province, China. cuokoo1973@163.com
Received: April 17, 2026 Revised: May 22, 2026 Accepted: June 15, 2026 Published online: September 19, 2026 Processing time: 128 Days and 20.8 Hours
Abstract
BACKGROUND
Traumatic long bone fractures with segmental bone defects represent a major challenge in orthopedics. Nonunion after bone grafting can result in persistent pain, dysfunction, and repeated operations. Currently, traditional clinical index-based predictive models show limited discriminative power and do not systematically integrate psychosocial factors into risk assessment.
AIM
To develop and validate a machine learning model integrating clinical and psychosocial factors to predict the risk of nonunion in patients with traumatic bone defects after bone transport.
METHODS
In this study, the development cohort included patients (n = 145) treated between January 2016 and November 2021, while an independent external validation cohort included patients (n = 53) treated between February 2022 and December 2024. Demographic data, injury- and treatment-related variables, and psychosocial scale scores (9-item Patient Health Questionnaire, 7-item Generalized Anxiety Disorder Scale, and Multidimensional Scale of Perceived Social Support) were collected. The Extreme Gradient Boosting algorithm was used to construct a traditional clinical indicator-based model and an integrated model integrating clinical and psychosocial factors. Model performance was evaluated through internal cross-validation and independent external validation using the area under the receiver operating characteristic curve, F1 score, and decision curve analysis.
RESULTS
During external validation, the integrated model demonstrated superior discriminative performance compared with the traditional clinical model (area under the receiver operating characteristic curve: 0.888 vs 0.823; P < 0.05). SHapley Additive exPlanations analysis showed that anxiety and depression contributed substantially to integrated model predictions. Decision curve analysis further demonstrated that the integrated model provided greater clinical net benefit across a broader decision-threshold range (20%-80%).
CONCLUSION
The machine learning model integrating clinical and psychosocial factors demonstrated superior discriminative performance and potential clinical applicability compared with models based solely on clinical indicators in predicting post-bone transport nonunion risk. Psychosocial factors constitute an important component of nonunion risk prediction.
Core Tip: In summary, the integrated model developed in this study demonstrates favorable discriminative ability (operating characteristic curve = 0.888 in external validation) and potential clinical utility for nonunion risk prediction, although further validation is needed. Importantly, our findings highlight the significant contribution of psychosocial factors - particularly anxiety, depression, and social support - to bone healing. These results provide strong evidence for promoting the deep integration of the “bio‑psycho‑social” medical model into routine orthopedic clinical practice, thereby facilitating more individualized and holistic patient management.
Citation: Luan FF, Chen JY, Zheng YZ, Hu MH, Huang F, Zheng CX. Efficacy of a machine learning model integrating clinical-psychosocial factors in predicting posttraumatic bone nonunion and bone defects. World J Psychiatry 2026; 16(9): 119455
Traumatic long bone fractures with segmental bone defects pose a severe challenge in orthopedics. Bone transport, an effective treatment for such injuries, carries a risk of nonunion, a major postoperative complication affecting long-term outcomes[1,2]. Nonunion-induced persistent pain, dysfunction, and repeated surgical procedures seriously compromise well-being while imposing a substantial socioeconomic burden[3-5]. Therefore, establishing a method capable of accurately identifying high-risk patients at an early stage to enable targeted monitoring, intervention, and prognostic improvement is crucial.
Current clinical risk assessment for nonunion mainly depends on traditional biomedical factors, including patient age, injury mechanism (open/closed), bone defect length, soft-tissue condition, and comorbidities such as diabetes[6-10]; additional studies have further confirmed the role of these factors[11-13]. Prediction models based on these factors, often constructed using conventional methods such as logistic regression, often have limited discriminative power. Moreover, they cannot adequately capture the complex nonlinear interactions among the factors, thereby limiting predictive accuracy.
Fracture healing is a complex biological process regulated by biomechanics, cellular and molecular biology, and the patients’ overall condition[14-16]. Bone nonunion itself also imposes profound psychosocial burdens on patients. Compared with patients whose fractures heal successfully, individuals with long-bone nonunion demonstrate significantly poorer physical and psychological health outcomes and markedly reduced quality of life compared with the general population[17,18]. The prolonged treatment course, repeated surgical interventions, severe pain, and long-term functional limitation of the affected limb contribute substantially to psychological distress. Previous studies reported a moderate-to-severe anxiety incidence of 70.6% among patients with bone nonunion[4]. Bone transport, which commonly requires long-term external fixation lasting several months or even more than a year, may further aggravate this psychological burden. The visible external fixator, demands of pin-tract care, and restriction of daily activities associated with bone transport may induce or aggravate anxiety and depressive symptoms. Previous studies confirmed that the anxiety scores of patients treated with Ilizarov external fixation increased significantly during treatment and persisted even after external fixator removal[19].
In recent years, preclinical and clinical studies have clarified several biological pathways through which psychological distress adversely affects bone repair. First, chronic psychological stress leads to the overactivation of the hypothalamus-pituitary-adrenal (HPA) axis and continuously elevated cortisol levels, thereby inhibiting osteoblast activity and promoting osteoclast formation, ultimately reducing bone mineral density[20]. Second, the sympathetic nervous system is an important mediator of stress-induced bone damage. The β2-adrenoceptor signaling pathway in chondrocytes plays a key role, and targeting this receptor facilitates fracture healing[21]. Furthermore, recent evidence indicates that psychological stress induces neutrophil expression of tyrosine hydroxylase within fracture hematomas, triggering local catecholamine production and release that directly affects chondrocytes in osteogenic regions and interferes with endochondral osteogenesis[22]. These findings indicate that psychological distress impairs bone healing through multiple interconnected pathways. Consequently, the psychological burden associated with nonunion and bone transport treatment may create a vicious cycle that eventually leads to adverse outcomes.
However, currently available clinical risk assessments for post-bone transport nonunion are based almost exclusively on traditional biomedical parameters, including patient age, bone defect length, and complications. Psychological dimensions, including anxiety, depression, social support, and coping strategies, have not been fully quantified or systematically integrated into existing forecasting frameworks. This limitation may significantly reduce the predictive accuracy of current models. As aforementioned, bone transport involves a prolonged treatment cycle and visible external fixators, imposing a psychological burden far exceeding that of conventional internal fixation; therefore, integrating psychosocial factors into predictive assessment is important and urgently needed.
With advances in artificial intelligence, machine learning (ML) algorithms have emerged as powerful for constructing high-dimensional and nonlinear predictive models because of their strong feature-recognition capability and ability to process complex patterns[23,24]. These developments provide an opportunity to address the aforementioned limitations. Accordingly, we hypothesized that integrating standardized psychosocial assessment indicators with traditional clinical parameters could establish an ML model with significantly improved discriminative performance, thereby enabling earlier, more individualized, and more accurate prediction of nonunion risk. To test whether the inclusion of psychosocial factors improves the prediction of post-bone transport nonunion, we combined clinical indicators with depression, anxiety, and social support scores to construct a predictive model and compared its performance with that of a model based solely on clinical indicators. The model was trained using a cohort from our institution and subsequently validated in an independent external cohort, with the aim of providing a more comprehensive framework for clinical evaluation.
MATERIALS AND METHODS
Study design and patient cohort
This retrospective study aimed to develop and validate an ML model for predicting the risk of nonunion following bone transport. Two independent cohorts were included: A development cohort and an external validation cohort. The development cohort comprised patients (n = 145) who underwent bone transport from January 2016 to November 2021 and was used for model construction and internal validation. To rigorously assess model generalizability, a temporal external validation design was adopted, using patients (n = 53) admitted between February 2022 and December 2024 as an independent external validation cohort.
Sample size estimation
The sample size was primarily determined by the number of consecutive cases meeting the selection criteria throughout the study. For prediction model studies, the sample size should meet the basic requirements of model development by ensuring an adequate number of events per variable (EPV) for each predictive variable, including candidate features. An EPV of at least 10-20 is generally recommended. In the development cohort, there were 17 outcome events (bone nonunion). After feature screening, the final number of predictive variables included in the model was < 17, based on a minimum EPV requirement of > 1, providing a basis for internal validation using cross-validation. The sample size of the external validation cohort (n = 53, including 10 events) was derived from consecutively available cases during the later study stage and was primarily used for the initial assessment of model generalizability. These results may support the calculation of confidence intervals for performance indicators and provide a reference for future larger-scale validation studies.
Patient selection criteria
Inclusion criteria: Age ≥ 18 years; segmental bone defects (defect length ≥ 2 cm) caused by traumatic long bone fractures; definitive treatment using the Ilizarov bone transport technique for unilateral lower-extremity defects; complete preoperative baseline clinical data and regular postoperative follow-up data (at least 6 months after the removal of the external fixator or until the clinical outcome was clear).
Exclusion criteria: Insufficient follow-up duration or missing key data, including psychosocial scale assessments and critical imaging data; bone defects caused by non-traumatic conditions such as tumors and infections; systemic diseases severely affecting bone metabolism (e.g., uncontrolled hyperthyroidism, hyperparathyroidism, or renal failure); major trauma involving other body regions or concomitant traumatic brain injury preventing the completion of psychological and social assessments.
According to these criteria, all patients undergoing bone transplant during the study period were screened, and eligible patients were assigned to the development or external validation cohort.
Data collection and definition
Data were collected through the electronic medical record system. The primary endpoint is the occurrence of post-bone transport nonunion. Nonunion was defined as the absence of imaging evidence of progressive healing for at least nine postoperative months, with no further healing progression during the subsequent three months. Imaging criteria included monthly X-ray films for at least three consecutive months showing clearly visible fracture lines, no bridging callus formation, and no cortical bone continuity. Clinical criteria included persistent pain at the fracture site, inability to fully bear weight without support, and abnormal activity at the bone defect site. Two orthopedists, blinded to the patients’ psychosocial status and model prediction outcomes, independently established the diagnosis of nonunion by integrating clinical and imaging findings. Disagreements were resolved through discussion, and a third senior orthopedist made the final decision when consensus could not be reached. Outcome assessments were performed at each follow-up visit. Follow-up evaluations were conducted at postoperative months 1, 2, 3, 6, 9, and 12, and quarterly thereafter until the outcome was clear. The collected variables were divided into three groups: (1) Demographic and clinical variables: Age, sex, and body mass index (BMI); (2) Injury- and treatment-related variables: Bone defect length, open fracture status (yes/no), distraction start time, distraction period, and bone mineralization time; and (3) Psychosocial variables: Depressive symptoms were assessed using the Patient Health Questionnaire-9 (PHQ-9), a 9-item tool with each item scored from 0 to 3. Total scores were 0-27 and were classified as no (0-4), mild (5-9), moderate (10-14), moderate-to-severe (15-19), or severe (20-27). Anxiety symptoms were evaluated using the Generalized Anxiety Disorder 7-item scale (GAD-7; score range: 0-21), categorized as no anxiety (0-4), mild (5-9), moderate (10-14), or severe (15-21). Perceived social support was assessed using the 12-item Multidimensional Scale of Perceived Social Support (MSPSS), which evaluates social support from family, friends, and significant others across three dimensions (four items per domain). Total scores ranged from 12 to 84, with higher scores indicating greater perceived social support. Smoking history and income status were also recorded.
Development and comparison of prediction models
Two prediction models were constructed and compared: (1) Traditional clinical model: Classification was performed using the XGBClassifier ML algorithm, with bone nonunion as the outcome variable. Included variables were age, sex, BMI, hypertension, diabetes, atherosclerosis, open fracture, bone defect length, distraction start time, distraction period, and bone mineralization time; and (2) Integrated model: Classification was also performed using the XGBClassifier ML algorithm, with bone nonunion serving as the outcome variable. Variables included age, sex, BMI, hypertension, diabetes, atherosclerosis, open fracture, bone defect length, distraction start time, distraction period, bone mineralization time-days, PHQ-9 score (depression), GAD-7 score (anxiety), MSPSS score (social support), current smoking status, and low-income status. Both models were constructed using Extreme Gradient Boosting (XGBoost). In the development cohort, hyperparameters were optimized through five-fold cross-validation with grid search, and the optimal parameters were selected accordingly. Model performance was internally evaluated within the cross-validation framework.
Model validation and performance evaluation
Model performance was evaluated from three dimensions:
Discriminating ability: The area under the receiver operating characteristic curve (AUC) was calculated, and differences in AUC between the two models were compared using DeLong’s test.
Comprehensive classification performance: Under the optimal threshold determined using the Youden index, accuracy, sensitivity, specificity, precision, and F1 score were calculated.
Clinical practicability: Decision curve analysis (DCA) was performed to evaluate standardized net benefits when using the prediction models to guide clinical decision-making, compared with “treat-all” and “treat-none” strategies across different threshold probabilities.
Performance was first evaluated during the internal validation of the development cohort. Subsequently, the optimized model was applied to the independent external validation cohort to assess generalizability.
Model interpretability analysis
To understand the prediction logic and key contributing factors, SHapley Additive exPlanations (SHAP) analysis was performed as a post hoc interpretability assessment of the best-performing integrated model. By calculating SHAP values for each feature across individual predictions, a summary plot was generated to visually illustrate the direction and relative importance of feature contributions to the model output.
Statistical analysis
Continuous variables are presented as mean ± SD and were compared using independent sample t-tests. Categorical variables are presented as frequencies (percentages) and were compared using χ2 tests. All statistical analyses were performed using R version 4.2.3 and Python version 3.11.4 with the partial receiver operating characteristic, XGBoost, and SHAP packages. Statistical significance was defined as P < 0.05.
RESULTS
Study cohort and baseline characteristics
Between January 2016 and November 2021, 145 patients were included in the development cohort. Subsequently, 53 patients were independently enrolled between February 2022 and December 2024 to form the external validation cohort. In the development cohort, patients with nonunion showed significant differences from the bone-healing group in age, BMI, bone defect length, distraction start time, bone mineralization time, psychosocial scores (depression, anxiety, and social support), and the proportions of patients with diabetes, open fracture, smoking history, and low-income status (P < 0.05; Table 1). In the external validation, anxiety scores, open fracture, and social support showed significant differences between the nonunion group and the bone-healing group (P < 0.05); however, there were no statistically significant differences in the remaining variables between the two groups (P > 0.05; Table 2). Except for BMI, depression score (PHQ-9), and anxiety score (GAD-7), which showed significant differences (P < 0.05), no statistically significant differences were observed in the remaining 14 indicators between the external validation cohort and the development cohort(P > 0.05; Table 3).
Table 1 Baseline data comparison (nonunion vs healing), n (%).
Figure 1, Tables 4 and 5 present the performance comparison between the traditional clinical model and the integrated model in internal and independent external validation sets. During internal validation, the integrated model developed using the XGBoost algorithm, integrating clinical and psychosocial factors, demonstrated strong overall predictive performance (AUC = 0.923, range: 0.856-0.990). Conversely, the traditional XGBoost model based solely on clinical indicators achieved an AUC of 0.854 (range: 0.788-0.919). During external validation, the integrated model achieved an AUC of 0.888 (range: 0.764-1.000), compared with 0.823 (range: 0.692-0.954) for the traditional clinical model. DeLong’s test confirmed that the integrated model had a significantly higher AUC than the traditional clinical model.
Feature importance and model interpretability analysis
To clarify the decision-making mechanism of the integrated model, the SHAP analysis was performed. Figure 2A presents the SHAP summary plot of the five most influential features contributing to predictions generated by the traditional clinical model: Bone defect length, BMI, bone mineralization time, open fracture, and age (ranked by significance). Figure 2B shows the five most influential features contributing to the predictions generated by the integrated model, namely PHQ-9 score, GAD-7 score, bone mineralization time, bone defect length, and BMI. Notably, GAD-7 and PHQ-9 scores contributed more significantly to model predictions than traditional clinical indicators such as bone defect length and BMI.
Figure 2 Bar chart showing SHapley Additive exPlanations global feature importance based on the external validation cohort.
A: The traditional clinical model; B: The integrated model. BMI: Body mass index; SHAP: SHapley Additive exPlanations; PHQ-9: Patient Health Questionnaire-9; GAD-7: Generalized Anxiety Disorder-7; MSPSS: Multidimensional Scale of Perceived Social Support.
Clinical DCA
As shown by DCA (Figure 3), the integrated model demonstrated significantly greater clinical applicability than the traditional clinical model. The traditional model provided a net clinical benefit only within the 20%-60% threshold probability range, with a peak value of approximately 0.10. In contrast, the integrated model was effective across a wider threshold range (20%-80%) and achieved a peak net benefit of 0.175, approximately 75% higher than that of the traditional model. These findings suggest that the integrated model may provide a higher expected clinical net benefit for preventing bone nonunion across varying clinical decision-making thresholds.
Figure 3 Clinical decision curve analysis based on the external validation cohort.
A: The traditional clinical model; B: The integrated model. XGBoost: EXtreme gradient boosting.
DISCUSSION
In this study, a model integrating clinical and psychosocial factors was constructed to predict the risk of nonunion following bone transport. In the independent external validation cohort, the model maintained excellent discriminant performance (AUC = 0.897), outperforming the model based solely on clinical indicators (AUC = 0.823), with statistical significance confirmed via DeLong’s test (P < 0.05). Recent studies have generally reported AUC values ranging from 0.82 to 0.86 for ML models[25,26], which are lower than those achieved by our model. Furthermore, our model outperformed traditional logistic regression-based nomograms for nonunion prediction (AUC approximately 0.8)[27]. This remarkable improvement suggests that the inclusion of the psychosocial dimensions may address previously overlooked determinants of bone healing variability.
SHAP analysis demonstrated that anxiety (GAD-7) and depression (PHQ-9) scores contributed strongly to model predictions, supporting recent evidence regarding the role of psychological factors in orthopedic rehabilitation. For example, Weinerman and colleagues highlighted the importance of perioperative mental health interventions in improving orthopedic trauma surgery outcomes and patient quality of life[28]. Our findings further support the pathophysiological concept of “psychological-skeletal” interaction. Previous studies have shown that chronic psychological stress can overactivate the HPA axis, resulting in persistently elevated cortisol levels that inhibit osteoblast activity and impair the bone healing process[29]. Beyond HPA-axis dysregulation, emerging evidence also implicates the sympathetic nervous system and inflammatory pathways. Stress-induced catecholamine release has been shown to promote osteoclast activity while inhibiting osteoblast differentiation. In patients with depression or anxiety, elevated levels of proinflammatory cytokines, including interleukin-6 and tumor necrosis factor-alpha, may form a catabolic microenvironment at the fracture site[22,30-32]. Another clinical study demonstrated positive correlations between tyrosine hydroxylase expression in fracture hematoma and patient-reported stress, depression, pain scores, and post-fracture healing disorders[22]. Collectively, these findings indicate that severe psychological stress may adversely affect bone growth and fracture healing. From a clinical prediction perspective, our model confirmed that these psychological stress mechanisms contribute independent risk weights beyond traditional clinical variables. Therefore, psychosocial dimensions may represent an indispensable component of future bone nonunion risk assessment systems.
The predictive model reported in this study addresses an important gap in the quantitative and psychosocial assessment of nonunion risk. Using routinely available clinical data and validated psychosocial scales (PHQ-9, GAD-7, and MSPSS), the model may provide individualized risk assessment before surgery or during the early postoperative period. Potential clinical applications include: (1) Preoperative risk stratification and joint decision-making, helping clinicians and patients establish realistic expectations while incorporating psychological support into operation planning; (2) Dynamic monitoring of psychosocial status throughout bone transportation to facilitate adaptive management; and (3) Staged interventions according to risk thresholds, including routine care for low-risk patients, combined psychological and physical therapy for moderate-risk patients, and multidisciplinary management involving orthopedics, psychology, and social work for high-risk patients. Thus, the model may support clinical decision-making and promote the implementation of the “bio-psycho-social” medical model in bone transport management.
Notably, the model showed good generalizability in the independent external validation cohort. Compared with reports describing performance degradation during external validation[33,34], these findings warrant attention. The model’s generalizability may be explained by the following factors: (1) The model appears to preferentially identify biologically stable core features. SHAP analysis demonstrated that predictions relied highly on anxiety (GAD-7), depression (PHQ-9), and bone defect length. This suggests that the model preferentially weights psychological status and anatomical injury severity as core factors. Such findings reflect the crucial pathophysiological role of the “psychological-skeletal” axis during bone healing. The relatively stable contribution of these factors across populations may underlie the model’s cross-population generalizability. Recent evidence has confirmed that perioperative psychological stress is an independent predictor of complications following Ilizarov bone transport[19]. By capturing the effects of the “psychological-skeletal” axis, the model may therefore achieve biologically grounded and population-generalizable risk prediction; and (2) The model showed certain noise robustness and reduced the risk of over-fitting. Baseline analyses showed that although variables such as diabetes and smoking history differed significantly between the nonunion and bone-healing groups within the development cohort, their contributions to model predictions were minimized through feature weighting. This indicates that the model can identify and suppress spurious correlations resulting from sampling variability in small samples, consistent with the latest general principles for constructing robust ML models[35]. When robust cross-environment features are captured while dataset-specific noise is minimized, external validation performance may remain stable or even improve[36-38]. Also, there were significant differences in BMI, depression score (PHQ-9), and anxiety score (GAD-7) between the development cohort and the external validation cohort, but the integrated model still demonstrated excellent discriminative performance in external validation, significantly outperforming the traditional clinical model. These findings further confirm the robustness of the model.
DCA further provided quantitative evidence supporting the model’s clinical practicability. Within the 20%-80% threshold probability range, the integrated model demonstrated greater clinical net benefit than the traditional clinical model. For example, at a threshold probability of 40%, the integrated model achieved a net benefit of approximately 0.175, nearly 75% higher than that of the traditional clinical model (approximately 0.10). These findings suggest that the model may better balance the identification of high-risk patients against unnecessary intervention during clinical decision-making. Evaluating the clinical net benefit is an important method for evaluating the value of predictive models[11]. Our findings provide preliminary support for future clinical implementation studies.
This study also has several limitations. The relatively small external validation cohort resulted in wide confidence intervals for performance indicators. Although the AUC values compare favorably with recent large-sample orthopedic studies (AUC: 0.82-0.86)[26,39], further validation of model calibration and generalizability in multi-center, large-sample prospective cohorts remains necessary. Future research should focus on translating prediction into intervention by determining whether targeted psychological interventions or enhanced physical therapy for high-risk patients can effectively improve clinical outcomes.
CONCLUSION
In summary, the integrated model developed in this study showed promising discriminative ability, acceptable robustness, and potential clinical applicability for predicting nonunion risk following bone transport. The findings also suggest that psychosocial factors may play a notable role in bone healing and provide supportive evidence for considering a broader integration of the “bio-psycho-social” medical model into orthopedic clinical practice. Further studies addressing the current limitations are needed before clinical translation.
Liu Z, Gu Y, Jin X, Tian W, Qi H, Sun Y, Li G, Wang H, Xiao X, Li P, Hu Y, Jia J. Comparison of Outcomes Following TiRobot-Assisted Sacroiliac Screw Fixation with Bone Grafting and Traditional Screw Fixation without Bone Grafting for Unstable Osteoporotic Sacral Fracture: A Single-Center Retrospective Study of 33 Patients.Med Sci Monit. 2021;27:e932724.
[RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)][Cited by in Crossref: 1][Cited by in RCA: 7][Article Influence: 1.4][Reference Citation Analysis (0)]
Jin M, Gu M, Kong K, Fan W, Ng S, Hu Y, Wang Z, Su J, Yuan WE, Li H, Zhai Z. Chemical Gambit in Bone Microenvironment: pH-Responsive Cu-Mn Nanoplatform Breaks the Inflammation-Osteoclast Vicious Cycle in Prosthesis-Associated Osteolysis.ACS Nano. 2026;20:1826-1843.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 4][Cited by in RCA: 5][Article Influence: 5.0][Reference Citation Analysis (0)]
Wang Y, Zwanenburg A, Zhang J, Teng X, Lam SK, Cao J, Ma Z, Zhou T, Zhang Y, Ge H, Cai J. Establishment of a CT-based radiomic feature robustness databank for OPC patients via image perturbation in a multi-institutional study: a practical method to safeguard model generalizability.Front Oncol. 2025;15:1464884.
[RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)][Cited by in RCA: 2][Reference Citation Analysis (0)]
