Development and incremental value assessment of a risk prediction model for post-traumatic osteoarthritis following ankle fracture surgery: a prospective study Authors: Xuehua Hu, Hong Cao, Min He, Yunying Wang, Ming Fang, Yu Li, Ying Li Published: 2026 ABSTRACT BackgroundPost-traumatic osteoarthritis (PTOA) is a common and clinically important complication after ankle fracture surgery, but reliable individualized risk stratification remains limited. This study aimed to develop and externally validate a Cox-based risk prediction model for postoperative PTOA after ankle fracture surgery and to assess its incremental value over traditional early postoperative clinical evaluation indicators.MethodsThis prospective two-center cohort study consecutively enrolled patients with surgically treated ankle fractures from Center I (March 2020 to March 2024) and Center II (March 2021 to March 2024). After screening 964 patients, 812 were included, with 605 assigned to the training cohort and 207 to the external validation cohort. Structural PTOA, the primary endpoint, was defined as the first occurrence of Kellgren–Lawrence grade ≥ 2 on postoperative weight-bearing ankle radiographs. Five incremental Cox models were developed using clinical, biochemical, nursing/perioperative, radiographic, and surgical variables. LASSO was used for feature selection, followed by multivariable Cox regression. Internal validation was performed using 1,000 bootstrap resamples, and external validation was conducted in an independent geographic cohort.ResultsThe 2-year incidence of structural PTOA was 24.0% (145/605) in the training cohort and 29.0% (60/207) in the validation cohort. Fourteen variables entered the full multivariable Cox model, and 11 independent predictors were retained for the final parsimonious model. The final comprehensive model showed the best performance, with an apparent C-index of 0.892, a bootstrap-corrected C-index of 0.887, and an external validation C-index of 0.865. Relative to the postoperative 3-month AOFAS score as a clinically relevant reference comparator, the comprehensive model showed higher predictive performance, with a continuous net reclassification improvement of 0.785 and an integrated discrimination improvement of 0.285. Time-dependent ROC analysis demonstrated favorable discrimination at 6, 12, and 24 months in both cohorts. Calibration and decision curve analyses also supported good agreement and clinical utility.ConclusionsA prospective, multi-source Cox prediction model for structural PTOA after ankle fracture surgery was successfully developed and externally validated. The final 11-variable model demonstrated good discrimination, robust validity, and clear incremental value over traditional early postoperative clinical evaluation indicators, supporting its potential use for individualized risk stratification after ankle fracture surgery.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12891-026-10081-y. Introduction Post-traumatic osteoarthritis (PTOA) is the dominant form of ankle osteoarthritis and represents a major source of long-term pain, disability, and impaired quality of life after ankle injury [1, 2]. Unlike hip or knee osteoarthritis, primary ankle osteoarthritis is relatively uncommon, and approximately 70%–80% of ankle osteoarthritis is post-traumatic, most commonly following severe sprain or intra-articular fracture [3, 4]. Among these injuries, ankle fractures are particularly important because they are common in orthopaedic trauma practice and may result in persistent articular incongruity, altered load transmission, and progressive cartilage degeneration despite surgical fixation [5]. A recent systematic review reported that the incidence of radiographic osteoarthritis after ankle fractures was 25% (95% CI, 18%–32%) overall and increased to 34% (95% CI, 23%–45%) in more severe fractures involving the posterior malleolus, underscoring the substantial long-term burden of this complication [6]. Although several studies have identified important prognostic factors for PTOA after ankle fracture, such as fracture-dislocation, posterior malleolar involvement, higher BMI, older age, and postoperative incongruity, the available evidence has largely remained at the level of association analysis rather than clinically applicable individualized risk prediction [7, 8]. In a long-term follow-up study of surgically treated malleolar fractures, Lübbeke et al. found that advanced radiographic ankle osteoarthritis developed in 36.3% of patients after a mean follow-up of 17.9 years; Weber C fracture, associated medial malleolar fracture, fracture-dislocation, higher body mass index, age ≥ 30 years, and longer time since surgery were significant risk factors, and the probability of developing PTOA reached 60%–70% in patients with three or more risk factors [9]. More recently, Wang et al. analyzed 477 patients with a minimum follow-up of 3 years and reported a radiographic PTOA incidence of 34.2%; age ≥ 60 years, physical labor intensity > grade II, BMI ≥ 28 kg/m², posterior malleolar involvement, larger posterior malleolar articular surface involvement, and postoperative articular incongruity were independently associated with PTOA [10]. In a cohort of ankle fracture-dislocations, Fonkoue et al. further showed that early PTOA occurred in 43.2% of patients after a mean follow-up of 27.2 months, and non-anatomical reduction was strongly associated with PTOA (OR 11.07, 95% CI 2.10–58.77) [11]. Despite these important observations, clinically usable prediction tools for PTOA after ankle fracture surgery remain scarce. Existing studies are predominantly retrospective, often single-center, and frequently focus on isolated risk factors or selected fracture patterns rather than constructing and validating comprehensive multivariable prediction models. In addition, the available literature is characterized by substantial heterogeneity in fracture classification, outcome definition, and follow-up duration, which limits direct translation into patient-level risk stratification. This limitation is particularly important because ankle osteoarthritis research is increasingly expected to move beyond radiographic description alone. In an international consensus study involving 100 participants from 18 countries, five core domains were identified for ankle osteoarthritis research: pain severity, health-related quality of life, function, disability, and ankle range of motion [12]. Therefore, a clinically meaningful model should ideally integrate structural risk with functional relevance and provide individualized estimates that can support early surveillance and targeted intervention. In this prospective two-center cohort study, we aimed to develop and externally validate a Cox-based prediction model for structural post-traumatic osteoarthritis (PTOA) after ankle fracture surgery. By integrating multi-source clinical data, including baseline clinical characteristics, biochemical indices, nursing/perioperative variables, fracture and radiographic features, and surgical/reduction-quality parameters, we sought to establish a more comprehensive risk stratification tool. We further evaluated whether this model could provide incremental predictive value over traditional early postoperative clinical evaluation indicators. In addition, an individualized nomogram was constructed, and its discrimination, calibration, clinical utility, and time-dependent predictive performance were systematically assessed. Materials and methods Study design and study population This study was designed as a prospective, two-center cohort study of adults with surgically treated isolated malleolar ankle fractures. In this study, “ankle fractures” referred to fractures involving one or more malleoli (lateral, medial, and/or posterior malleolus), with or without associated syndesmotic injury or ankle dislocation/subluxation. Consecutive patients were enrolled from two institutions: Center I (Xiangyang Central Hospital) between March 2020 and March 2024, and Center II (Xiangyang Sixth People’s Hospital) between March 2021 and March 2024. All enrolled patients were followed according to a predefined protocol, with a minimum follow-up duration of 2 years and a follow-up cutoff date of March 2026. A total of 964 patients were initially screened, of whom 152 were excluded according to the predefined eligibility criteria. The final study population consisted of 812 patients, including 605 patients from Center I and 207 patients from Center II. Center I was used as the training cohort for model development and bootstrap-based internal validation, whereas Center II served as an independent geographic external validation cohort. This study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of Xiangyang Central Hospital, the Affiliated Hospital of Hubei University of Arts and Science and Xiangyang Sixth People’s Hospital, Affiliated Hospital of Hubei University of Arts and Science (Approval No.: 2020-01-142 and 2021KYXZ12). Written informed consent was obtained from all participants (Fig. 1). Fig. 1Study flowchart of patient screening, cohort allocation, outcome ascertainment, model development, and external validation Study flowchart of patient screening, cohort allocation, outcome ascertainment, model development, and external validation Inclusion and exclusion criteria and ethical approval Patients were eligible for inclusion if they met all of the following criteria: (1) age ≥ 18 years; (2) radiographically confirmed ankle fracture; (3) surgical treatment performed for the index fracture; (4) availability of baseline clinical, laboratory, imaging, and perioperative data; and (5) completion of at least 2 years of follow-up or occurrence of the primary endpoint during follow-up. Only patients with surgically treated isolated malleolar ankle fractures as the primary injury of interest were included to ensure cohort homogeneity and reliable endpoint assessment. Patients were excluded if they had non-target fracture types, including pilon fractures, talus fractures, or other peri-ankle fracture patterns beyond the scope of isolated malleolar ankle fractures. Patients were excluded if they met any of the following criteria: (1) age < 18 years or no surgical treatment; (2) non-target fracture types, including pilon fractures, talus fractures, or other injuries clearly beyond the scope of isolated ankle fractures; (3) pre-existing ipsilateral ankle conditions that could affect endpoint assessment, including established ankle osteoarthritis/post-traumatic osteoarthritis, bone infection, bone tumor, or severe ankle-foot deformity. Pre-existing ankle osteoarthritis was excluded by combined review of pre-injury medical history and baseline imaging; patients with a documented history of ankle osteoarthritis/post-traumatic osteoarthritis or baseline radiographic/CT evidence of established degenerative changes not attributable to the acute index fracture were excluded; (4) severe prior ipsilateral ankle history or major concomitant interfering injuries, including previous ipsilateral ankle fracture or surgery, old fracture/malunion requiring reoperation, bilateral ankle fractures, or severe ipsilateral lower-extremity injuries affecting symptom and functional assessment; (5) missing key imaging data or poor image quality precluding radiographic evaluation; or (6) follow-up < 2 years or incomplete key clinical/follow-up data. The study protocol was approved by the ethics committees of the participating centers, and written informed consent was obtained from all participants before enrollment. All patient data were anonymized prior to analysis to ensure confidentiality and data security. Follow-up and outcome definitions All patients underwent regular postoperative follow-up through outpatient visits and standardized radiographic and clinical assessments. As part of the routine postoperative follow-up protocol, the AOFAS ankle-hindfoot score, Olerud-Molander Ankle Score (OMAS), and visual analogue scale (VAS) pain score were assessed at the predefined 3-month postoperative follow-up visit. These assessments were performed using the same standardized framework across both centers by trained clinical assessors. These 3-month indicators were collected as traditional early postoperative clinical evaluation measures for comparative analyses The primary endpoint was structural post-traumatic osteoarthritis (structural PTOA), defined as the first occurrence of Kellgren-Lawrence (K-L) grade ≥ 2 on postoperative weight-bearing ankle radiographs, including anteroposterior, lateral, and mortise views. Event time was defined as the follow-up time point at which this radiographic criterion was first met [13]. Data collection and candidate predictors Multi-source heterogeneous clinical data were collected for all eligible patients. Candidate predictors included: (1) baseline clinical characteristics, such as age, sex, body mass index (BMI), smoking history, alcohol use, diabetes, hypertension, osteoporosis, injury mechanism, injured side, previous ankle injury history, and occupational or daily activity intensity; (2) preoperative biochemical indices, including white blood cell count, neutrophil percentage, C-reactive protein (CRP), erythrocyte sedimentation rate (ESR), blood glucose or HbA1c, albumin, and lipid parameters; (3) objective nursing/perioperative variables, including time from injury to admission, time from injury to surgery, preoperative pain score, soft-tissue condition, nutritional risk screening score, and frailty score in elderly patients; (4) fracture and radiographic variables, including AO/OTA or Danis-Weber classification, fracture-dislocation, medial/posterior malleolar involvement, posterior malleolar fragment size, syndesmotic injury, preoperative articular incongruity, talar shift, medial clear space widening, and articular comminution; and (5) surgical and reduction-quality parameters, including fixation strategy, posterior malleolar fixation, syndesmotic fixation, operative time, intraoperative blood loss, reduction quality, and early postoperative weight-bearing protocol. For descriptive Kaplan–Meier stratification, reduction quality was treated as a composite summary of postoperative articular congruity and syndesmotic alignment. Poor reduction quality was defined as the presence of postoperative residual articular step-off/gap > 2 mm and/or postoperative incomplete syndesmotic reduction. To improve reproducibility, several key predictor variables were defined a priori. Poor preoperative soft-tissue condition was defined as severe swelling with loss of skin wrinkling and/or fracture blisters documented before definitive surgery. Severe articular comminution was defined as a multifragmentary intra-articular fracture pattern with substantial articular surface fragmentation and/or impaction on preoperative radiographs or CT. Early postoperative weight-bearing was defined as surgeon-permitted initiation of protected partial or progressive weight-bearing within 6 weeks after surgery; later weight-bearing was defined as initiation after 6 weeks. Postoperative residual articular step-off/gap was assessed on postoperative imaging, and a residual step-off/gap > 2 mm was considered abnormal. Postoperative incomplete syndesmotic reduction was defined as persistent syndesmotic malalignment on postoperative imaging according to our predefined radiographic/CT criteria. Key imaging variables were independently assessed by two experienced reviewers, and disagreements were resolved by senior adjudication. Clinical, laboratory, and radiographic data were extracted using standardized case report forms by trained investigators. Key imaging variables and radiographic endpoints were independently assessed by two experienced reviewers. In cases of disagreement, adjudication was performed by a senior reviewer to ensure consistency of imaging interpretation. To quantify interobserver reliability, weighted kappa was used for ordinal radiographic grading, Cohen’s kappa was used for binary and categorical radiographic variables, and intraclass correlation coefficients (ICCs) were calculated for continuous radiographic measurements when applicable. Interobserver agreement was evaluated for the primary endpoint assessment and the main radiographic variables, including Danis–Weber classification, fracture-dislocation / subluxation, syndesmotic injury, posterior malleolar fragment involvement and size, talar shift, widening of medial clear space, severe articular comminution, postoperative residual articular step-off/gap, and postoperative syndesmotic reduction. Model development strategy The prediction model was developed within a Cox proportional hazards framework. In the training cohort, candidate variables were first screened using univariable Cox regression analysis. To reduce dimensionality and minimize overfitting, least absolute shrinkage and selection operator (LASSO) regression was then applied for feature selection. Variables retained after LASSO were entered into a full multivariable Cox regression model. For selected predictors, categorical thresholds were used in the main model to improve clinical interpretability. BMI was dichotomized at 24 kg/m² as a prespecified clinically meaningful threshold for overweight in the local population context, and time to surgery was dichotomized at > 7 days as a prespecified clinically meaningful delay threshold based on perioperative practice considerations and prior literature [14, 15]. These cutoffs were defined for the main model before formal multivariable modeling and were not selected by outcome-driven optimization from the present dataset. To evaluate the incremental value of multi-source data integration, five stepwise Cox models were constructed: Model 1 included baseline clinical variables only; Model 2 added preoperative biochemical indices; Model 3 further incorporated objective nursing/perioperative factors; Model 4 additionally included fracture and radiographic characteristics; and Model 5 further incorporated surgical and reduction-quality parameters, representing the full multimodal model. Predictors that remained statistically significant in the full multivariable Cox model were retained to construct the final parsimonious prediction model, which was subsequently used for the correlation heatmap, nomogram construction, time-dependent ROC analysis, calibration analysis, and decision curve analysis. Internal and external validation Internal validation of the model developed in Center I was performed using 1,000 bootstrap resamples. Model optimism was estimated by comparing apparent and bootstrap-corrected performance, and the bootstrap-corrected C-index, calibration slope, calibration intercept, and Brier score were calculated. This procedure was used to assess model stability and to reduce the risk of overfitting. External validation was conducted in the independent geographic cohort from Center II. Model discrimination, calibration, and clinical utility were evaluated in the validation cohort using the same analytic framework as in the training cohort. This design allowed assessment of model reproducibility and transportability across institutions. Assessment of model performance, incremental value, and clinical utility Model discrimination was primarily assessed using the C-index in the training cohort, bootstrap-corrected training cohort, and external validation cohort. In addition, time-dependent receiver operating characteristic (ROC) analysis was used to evaluate predictive discrimination of the final model at 6, 12, and 24 months, and the corresponding time-specific areas under the curve (AUCs) were calculated. Model calibration was assessed using calibration plots, with bootstrap-corrected calibration performed in the training cohort. To quantify incremental value over traditional early postoperative clinical evaluation indicators, the integrated performance of the five stepwise models was compared with that of traditional 3-month postoperative indicators. The postoperative 3-month AOFAS, OMAS, and VAS scores were included as clinically relevant reference comparators to contextualize model performance. These single postoperative clinical indicators were not considered fully equivalent to a multivariable structural prediction model and were used for comparative benchmarking rather than one-to-one replacement. Continuous net reclassification improvement (NRI) and integrated discrimination improvement (IDI) were calculated to assess improvement in risk stratification. Clinical usefulness was further evaluated using decision curve analysis (DCA). Based on the final parsimonious model, an individualized nomogram was constructed to estimate 6-, 12-, and 24-month risks of structural PTOA after ankle fracture surgery. Statistical analysis Continuous variables were tested for distributional characteristics before analysis. Normally distributed variables are presented as mean ± standard deviation (SD) and were compared using the independent-samples t test. Non-normally distributed variables are presented as median and interquartile range (IQR) and were compared using the Mann-Whitney U test. Categorical variables are expressed as counts and percentages and were compared using the chi-square test or Fisher’s exact test, as appropriate. PTOA-free survival was estimated using the Kaplan-Meier method and compared using the log-rank test. Cox proportional hazards regression was used for univariable and multivariable analyses, with results reported as hazard ratios (HRs) and 95% confidence intervals (CIs). Restricted cubic spline (RCS) analysis was used to explore potential nonlinear relationships between continuous predictors and structural PTOA risk. All statistical analyses were two-sided, and a P value < 0.05 was considered statistically significant. Results Study cohort and baseline characteristics A total of 964 patients with surgically treated ankle fractures were initially screened, of whom 152 were excluded according to the predefined eligibility criteria. The final study population comprised 812 patients, including 605 patients in Center I and 207 patients in Center II. Center I served as the training cohort and Center II as the external validation cohort. Structural PTOA occurred in 145 patients (24.0%) in the training cohort and 60 patients (29.0%) in the external validation cohort (Fig. 1). In the training cohort, patients who developed structural PTOA were generally older and had a higher BMI than those without PTOA. They also more frequently had diabetes, osteoporosis, previous ankle injury, heavy physical labor or high daily activity, and high-energy injury mechanisms. In addition, the PTOA group showed significantly higher inflammatory and metabolic burden, as reflected by increased WBC, neutrophil ratio, CRP, ESR, fasting glucose, HbA1c, D-dimer, and total cholesterol levels, together with lower albumin levels. These patients also had longer time from injury to surgery, worse preoperative soft-tissue condition, higher rates of open fracture, higher nutritional risk and frailty scores, and more severe pain before surgery. Radiographically and surgically, the PTOA group was more likely to present with Danis-Weber type C fractures, pronation-type injuries, fracture-dislocation, posterior malleolar involvement with > 25% articular surface participation, syndesmotic injury, talar shift, widening of the medial clear space, severe articular comminution, longer operative time, greater intraoperative blood loss, poorer reduction quality, and less frequent early weight-bearing. Early postoperative clinical scores were also significantly worse in patients who developed PTOA (Table 1). Table 1Distribution of candidate predictor variables and early postoperative clinical evaluation indicators in the training cohort, stratified by structural PTOA statusVariable Stratification and ParametersTotal (N = 605)Non-PTOA Group(N = 460)PTOA Group (N = 145)P ValueBasic Clinical Characteristics Age (years)46.1 ± 12.244.5 ± 12.351.2 ± 10.5< 0.001 Male (n, %)335 (55.4%)250 (54.3%)85 (58.6%)0.362 BMI (kg/m²)24.9 ± 3.624.1 ± 3.227.5 ± 3.8< 0.001 Smoking history (n, %)188 (31.1%)133 (28.9%)55 (37.9%)0.041 Alcohol consumption (n, %)142 (23.5%)105 (22.8%)37 (25.5%)0.491 History of diabetes (n, %)92 (15.2%)55 (12.0%)37 (25.5%)< 0.001 History of hypertension (n, %)135 (22.3%)95 (20.7%)40 (27.6%)0.082 History of osteoporosis (n, %)85 (14.0%)52 (11.3%)33 (22.8%)0.001 High-energy injury mechanism (n, %)275 (45.5%)180 (39.1%)95 (65.5%)< 0.001 Injury side (Right, n, %)315 (52.1%)240 (52.2%)75 (51.7%)0.923 History of previous ankle injury (n, %)68 (11.2%)41 (8.9%)27 (18.6%)0.001 Heavy physical labor/high daily activity level (n, %)208 (34.4%)145 (31.5%)63 (43.4%)0.01 ASA classification (≥ III, n, %)72 (11.9%)48 (10.4%)24 (16.6%)0.045Preoperative Biochemical Indices WBC (×10⁹/L)8.2 ± 2.57.9 ± 2.29.1 ± 3.1< 0.001 Neutrophil ratio (NEU%)68.5 ± 8.467.2 ± 8.172.6 ± 7.9< 0.001 CRP (mg/L)16.3 [8.5–25.4]12.5 [6.8–19.2]28.6 [15.4–42.1]< 0.001 ESR (mm/h)22.4 ± 9.520.8 ± 8.227.5 ± 11.4< 0.001 Fasting blood glucose (mmol/L)5.8 ± 1.45.6 ± 1.16.4 ± 1.9< 0.001 Glycosylated hemoglobin (HbA1c, %)5.9 ± 0.85.8 ± 0.76.3 ± 1.0< 0.001 Albumin (g/L)41.2 ± 4.541.8 ± 4.239.3 ± 4.8< 0.001 D-dimer (mg/L)1.8 [0.9–3.2]1.5 [0.8–2.6]2.9 [1.4–4.5]< 0.001 Total cholesterol (TC, mmol/L)4.6 ± 1.14.5 ± 1.04.8 ± 1.20.006Nursing and Perioperative Status Time from injury to admission (days)1 [0–2]1 [0–2]1 [0–3]0.214 Time from injury to surgery (days)6 [4–9]5 [3–8]8 [5–12]< 0.001 Poor preoperative soft tissue condition (severe swelling/blisters, n, %)215 (35.5%)140 (30.4%)75 (51.7%)< 0.001 Open fracture (n, %)45 (7.4%)20 (4.3%)25 (17.2%)< 0.001 NRS-2002 nutritional risk screening ≥ 3 points (n, %)122 (20.2%)75 (16.3%)47 (32.4%)< 0.001 CFS clinical frailty scale ≥ 4 points (n, %)88 (14.5%)55 (12.0%)33 (22.8%)0.001 Preoperative pain score (VAS)6.5 ± 1.26.3 ± 1.17.1 ± 1.3< 0.001Fracture and Radiographic CharacteristicsDanis-Weber classification (n, %)< 0.001 - Type A110 (18.2%)100 (21.7%)10 (6.9%) - Type B345 (57.0%)280 (60.9%)65 (44.8%) - Type C150 (24.8%)80 (17.4%)70 (48.3%) Lauge-Hansen pronation-type injury (PER/PAD, n, %)165 (27.3%)105 (22.8%)60 (41.4%)< 0.001 Combined with ankle dislocation/subluxation (n, %)170 (28.1%)90 (19.6%)80 (55.2%)< 0.001 Fracture involving medial malleolus (n, %)380 (62.8%)285 (62.0%)95 (65.5%)0.443 Fracture involving posterior malleolus (n, %)295 (48.8%)205 (44.6%)90 (62.1%)< 0.001 Posterior malleolar fragment involving > 25% of articular surface (n, %)125 (20.7%)65 (14.1%)60 (41.4%)< 0.001 Syndesmotic injury (n, %)185 (30.6%)105 (22.8%)80 (55.2%)< 0.001 Talar shift > 2 mm (n, %)148 (24.5%)85 (18.5%)63 (43.4%)< 0.001 Widening of medial clear space (n, %)162 (26.8%)102 (22.2%)60 (41.4%)< 0.001 Severe articular comminution (n, %)115 (19.0%)60 (13.0%)55 (37.9%)< 0.001Surgical Details and Reduction Quality Surgical approach (Minimally invasive/percutaneous, n, %)85 (14.0%)72 (15.7%)13 (9.0%)0.042 Specific fixation of posterior malleolus (n, %)198 (32.7%)145 (31.5%)53 (36.6%)0.254 Syndesmotic fixation (n, %)172 (28.4%)100 (21.7%)72 (49.7%)< 0.001 Operative time (minutes)85.4 ± 25.680.2 ± 22.4101.8 ± 28.5< 0.001 Intraoperative blood loss (mL)100 [50–150]80 [50–120]150 [100–200]< 0.001 Postoperative residual articular step-off/gap > 2 mm (n, %)55 (9.1%)20 (4.3%)35 (24.1%)< 0.001 Postoperative incomplete syndesmotic reduction (n, %)42 (6.9%)18 (3.9%)24 (16.6%)< 0.001 Early weight-bearing protocol (< 6 weeks, n, %)195 (32.2%)165 (35.9%)30 (20.7%)0.001Traditional Early Postoperative Clinical Evaluation Models (for comparative validation) Postoperative 3-month AOFAS ankle-hindfoot score82.5 ± 8.484.6 ± 7.275.8 ± 9.1< 0.001 Postoperative 3-month OMAS score78.2 ± 9.580.5 ± 8.170.9 ± 10.2< 0.001 Postoperative 3-month VAS pain score2.1 ± 1.01.8 ± 0.83.1 ± 1.2< 0.001Continuous variables are presented as mean ± standard deviation (SD) or median [interquartile range (IQR)], as appropriate; categorical variables are presented as n (%). P values were calculated using the independent-samples t test, Mann–Whitney U test, chi-square test, or Fisher’s exact test, as appropriate Table 1Distribution of candidate predictor variables and early postoperative clinical evaluation indicators in the training cohort, stratified by structural PTOA statusVariable Stratification and ParametersTotal (N = 605)Non-PTOA Group(N = 460)PTOA Group (N = 145)P ValueBasic Clinical Characteristics Age (years)46.1 ± 12.244.5 ± 12.351.2 ± 10.5< 0.001 Male (n, %)335 (55.4%)250 (54.3%)85 (58.6%)0.362 BMI (kg/m²)24.9 ± 3.624.1 ± 3.227.5 ± 3.8< 0.001 Smoking history (n, %)188 (31.1%)133 (28.9%)55 (37.9%)0.041 Alcohol consumption (n, %)142 (23.5%)105 (22.8%)37 (25.5%)0.491 History of diabetes (n, %)92 (15.2%)55 (12.0%)37 (25.5%)< 0.001 History of hypertension (n, %)135 (22.3%)95 (20.7%)40 (27.6%)0.082 History of osteoporosis (n, %)85 (14.0%)52 (11.3%)33 (22.8%)0.001 High-energy injury mechanism (n, %)275 (45.5%)180 (39.1%)95 (65.5%)< 0.001 Injury side (Right, n, %)315 (52.1%)240 (52.2%)75 (51.7%)0.923 History of previous ankle injury (n, %)68 (11.2%)41 (8.9%)27 (18.6%)0.001 Heavy physical labor/high daily activity level (n, %)208 (34.4%)145 (31.5%)63 (43.4%)0.01 ASA classification (≥ III, n, %)72 (11.9%)48 (10.4%)24 (16.6%)0.045Preoperative Biochemical Indices WBC (×10⁹/L)8.2 ± 2.57.9 ± 2.29.1 ± 3.1< 0.001 Neutrophil ratio (NEU%)68.5 ± 8.467.2 ± 8.172.6 ± 7.9< 0.001 CRP (mg/L)16.3 [8.5–25.4]12.5 [6.8–19.2]28.6 [15.4–42.1]< 0.001 ESR (mm/h)22.4 ± 9.520.8 ± 8.227.5 ± 11.4< 0.001 Fasting blood glucose (mmol/L)5.8 ± 1.45.6 ± 1.16.4 ± 1.9< 0.001 Glycosylated hemoglobin (HbA1c, %)5.9 ± 0.85.8 ± 0.76.3 ± 1.0< 0.001 Albumin (g/L)41.2 ± 4.541.8 ± 4.239.3 ± 4.8< 0.001 D-dimer (mg/L)1.8 [0.9–3.2]1.5 [0.8–2.6]2.9 [1.4–4.5]< 0.001 Total cholesterol (TC, mmol/L)4.6 ± 1.14.5 ± 1.04.8 ± 1.20.006Nursing and Perioperative Status Time from injury to admission (days)1 [0–2]1 [0–2]1 [0–3]0.214 Time from injury to surgery (days)6 [4–9]5 [3–8]8 [5–12]< 0.001 Poor preoperative soft tissue condition (severe swelling/blisters, n, %)215 (35.5%)140 (30.4%)75 (51.7%)< 0.001 Open fracture (n, %)45 (7.4%)20 (4.3%)25 (17.2%)< 0.001 NRS-2002 nutritional risk screening ≥ 3 points (n, %)122 (20.2%)75 (16.3%)47 (32.4%)< 0.001 CFS clinical frailty scale ≥ 4 points (n, %)88 (14.5%)55 (12.0%)33 (22.8%)0.001 Preoperative pain score (VAS)6.5 ± 1.26.3 ± 1.17.1 ± 1.3< 0.001Fracture and Radiographic CharacteristicsDanis-Weber classification (n, %)< 0.001 - Type A110 (18.2%)100 (21.7%)10 (6.9%) - Type B345 (57.0%)280 (60.9%)65 (44.8%) - Type C150 (24.8%)80 (17.4%)70 (48.3%) Lauge-Hansen pronation-type injury (PER/PAD, n, %)165 (27.3%)105 (22.8%)60 (41.4%)< 0.001 Combined with ankle dislocation/subluxation (n, %)170 (28.1%)90 (19.6%)80 (55.2%)< 0.001 Fracture involving medial malleolus (n, %)380 (62.8%)285 (62.0%)95 (65.5%)0.443 Fracture involving posterior malleolus (n, %)295 (48.8%)205 (44.6%)90 (62.1%)< 0.001 Posterior malleolar fragment involving > 25% of articular surface (n, %)125 (20.7%)65 (14.1%)60 (41.4%)< 0.001 Syndesmotic injury (n, %)185 (30.6%)105 (22.8%)80 (55.2%)< 0.001 Talar shift > 2 mm (n, %)148 (24.5%)85 (18.5%)63 (43.4%)< 0.001 Widening of medial clear space (n, %)162 (26.8%)102 (22.2%)60 (41.4%)< 0.001 Severe articular comminution (n, %)115 (19.0%)60 (13.0%)55 (37.9%)< 0.001Surgical Details and Reduction Quality Surgical approach (Minimally invasive/percutaneous, n, %)85 (14.0%)72 (15.7%)13 (9.0%)0.042 Specific fixation of posterior malleolus (n, %)198 (32.7%)145 (31.5%)53 (36.6%)0.254 Syndesmotic fixation (n, %)172 (28.4%)100 (21.7%)72 (49.7%)< 0.001 Operative time (minutes)85.4 ± 25.680.2 ± 22.4101.8 ± 28.5< 0.001 Intraoperative blood loss (mL)100 [50–150]80 [50–120]150 [100–200]< 0.001 Postoperative residual articular step-off/gap > 2 mm (n, %)55 (9.1%)20 (4.3%)35 (24.1%)< 0.001 Postoperative incomplete syndesmotic reduction (n, %)42 (6.9%)18 (3.9%)24 (16.6%)< 0.001 Early weight-bearing protocol (< 6 weeks, n, %)195 (32.2%)165 (35.9%)30 (20.7%)0.001Traditional Early Postoperative Clinical Evaluation Models (for comparative validation) Postoperative 3-month AOFAS ankle-hindfoot score82.5 ± 8.484.6 ± 7.275.8 ± 9.1< 0.001 Postoperative 3-month OMAS score78.2 ± 9.580.5 ± 8.170.9 ± 10.2< 0.001 Postoperative 3-month VAS pain score2.1 ± 1.01.8 ± 0.83.1 ± 1.2< 0.001Continuous variables are presented as mean ± standard deviation (SD) or median [interquartile range (IQR)], as appropriate; categorical variables are presented as n (%). P values were calculated using the independent-samples t test, Mann–Whitney U test, chi-square test, or Fisher’s exact test, as appropriate Distribution of candidate predictor variables and early postoperative clinical evaluation indicators in the training cohort, stratified by structural PTOA status Continuous variables are presented as mean ± standard deviation (SD) or median [interquartile range (IQR)], as appropriate; categorical variables are presented as n (%). P values were calculated using the independent-samples t test, Mann–Whitney U test, chi-square test, or Fisher’s exact test, as appropriate When the two cohorts were compared, most selected candidate predictors and early postoperative clinical evaluation indicators were broadly comparable between Center I and Center II, with no significant between-center differences in age, sex, BMI, diabetes history, high-energy injury mechanism, CRP, D-dimer, time from injury to surgery, Danis-Weber type C distribution, dislocation/subluxation, posterior malleolar fragment > 25%, talar shift, residual articular step-off/gap > 2 mm, or 2-year structural PTOA incidence. The only significant difference was that the actual follow-up duration was slightly longer in the training cohort than in the validation cohort (38.6 ± 10.5 vs. 36.2 ± 9.8 months, P = 0.003), whereas the 2-year incidence of structural PTOA remained comparable between the two cohorts (24.0% vs. 29.0%, P = 0.154). These findings indicate that the training and external validation cohorts were generally comparable with respect to selected candidate predictors, early postoperative clinical indicators, and 2-year outcome distribution (Table 2). Table 2Comparison of selected candidate predictor variables, early postoperative clinical evaluation indicators, follow-up duration, and outcome distribution between the training and external validation cohortsVariableCenter I(Training Cohort, N = 605)Center II(Validation Cohort, N = 207)P ValueDemographics and Medical History Age (years)46.1 ± 12.247.8 ± 11.50.076 Male (n, %)335 (55.4%)120 (58.0%)0.512 BMI (kg/m²)24.9 ± 3.625.3 ± 3.50.165 History of diabetes (n, %)92 (15.2%)36 (17.4%)0.442 High-energy injury mechanism (n, %)275 (45.5%)105 (50.7%)0.188 Heavy physical labor/high daily activity level (n, %)208 (34.4%)82 (39.6%)0.168 Biochemical and Perioperative Indicators CRP (mg/L)16.3 [8.5–25.4]18.1 [9.2–27.5]0.114 D-dimer (mg/L)1.8 [0.9–3.2]2.0 [1.0-3.5]0.089 Time from injury to surgery (days)6 [4–9]7 [4–10]0.082 NRS-2002 nutritional risk screening ≥ 3 points (n, %)122 (20.2%)48 (23.2%)0.354Radiographic and Reduction Indicators Danis-Weber Type C (n, %)150 (24.8%)62 (30.0%)0.138 Combined with ankle dislocation/subluxation (n, %)170 (28.1%)68 (32.9%)0.187 Posterior malleolar fragment involving > 25% of articular surface (n, %)125 (20.7%)51 (24.6%)0.231 Talar shift > 2 mm (n, %)148 (24.5%)58 (28.0%)0.315 Postoperative residual articular step-off/gap > 2 mm (n, %)55 (9.1%)23 (11.1%)0.384 Traditional Evaluation Indicators Postoperative 3-month AOFAS ankle-hindfoot score82.5 ± 8.481.2 ± 8.90.062 Postoperative 3-month VAS pain score2.1 ± 1.02.3 ± 1.10.051Follow-up and Outcome Endpoints Actual follow-up duration (months)38.6 ± 10.536.2 ± 9.80.003 2-year incidence of structural PTOA (n, %)145 (24.0%)60 (29.0%)0.154Continuous variables are presented as mean ± SD or median [IQR], as appropriate; categorical variables are presented as n (%). P values were calculated using the independent-samples t test, Mann–Whitney U test, chi-square test, or Fisher’s exact test, as appropriate Table 2Comparison of selected candidate predictor variables, early postoperative clinical evaluation indicators, follow-up duration, and outcome distribution between the training and external validation cohortsVariableCenter I(Training Cohort, N = 605)Center II(Validation Cohort, N = 207)P ValueDemographics and Medical History Age (years)46.1 ± 12.247.8 ± 11.50.076 Male (n, %)335 (55.4%)120 (58.0%)0.512 BMI (kg/m²)24.9 ± 3.625.3 ± 3.50.165 History of diabetes (n, %)92 (15.2%)36 (17.4%)0.442 High-energy injury mechanism (n, %)275 (45.5%)105 (50.7%)0.188 Heavy physical labor/high daily activity level (n, %)208 (34.4%)82 (39.6%)0.168 Biochemical and Perioperative Indicators CRP (mg/L)16.3 [8.5–25.4]18.1 [9.2–27.5]0.114 D-dimer (mg/L)1.8 [0.9–3.2]2.0 [1.0-3.5]0.089 Time from injury to surgery (days)6 [4–9]7 [4–10]0.082 NRS-2002 nutritional risk screening ≥ 3 points (n, %)122 (20.2%)48 (23.2%)0.354Radiographic and Reduction Indicators Danis-Weber Type C (n, %)150 (24.8%)62 (30.0%)0.138 Combined with ankle dislocation/subluxation (n, %)170 (28.1%)68 (32.9%)0.187 Posterior malleolar fragment involving > 25% of articular surface (n, %)125 (20.7%)51 (24.6%)0.231 Talar shift > 2 mm (n, %)148 (24.5%)58 (28.0%)0.315 Postoperative residual articular step-off/gap > 2 mm (n, %)55 (9.1%)23 (11.1%)0.384 Traditional Evaluation Indicators Postoperative 3-month AOFAS ankle-hindfoot score82.5 ± 8.481.2 ± 8.90.062 Postoperative 3-month VAS pain score2.1 ± 1.02.3 ± 1.10.051Follow-up and Outcome Endpoints Actual follow-up duration (months)38.6 ± 10.536.2 ± 9.80.003 2-year incidence of structural PTOA (n, %)145 (24.0%)60 (29.0%)0.154Continuous variables are presented as mean ± SD or median [IQR], as appropriate; categorical variables are presented as n (%). P values were calculated using the independent-samples t test, Mann–Whitney U test, chi-square test, or Fisher’s exact test, as appropriate Comparison of selected candidate predictor variables, early postoperative clinical evaluation indicators, follow-up duration, and outcome distribution between the training and external validation cohorts Continuous variables are presented as mean ± SD or median [IQR], as appropriate; categorical variables are presented as n (%). P values were calculated using the independent-samples t test, Mann–Whitney U test, chi-square test, or Fisher’s exact test, as appropriate Interobserver reliability of radiographic assessment Interobserver reliability analyses demonstrated good reproducibility for the radiographic assessments used in this study. Agreement was high for the primary endpoint, with Cohen’s κ of 0.86 (95% CI 0.81–0.91) for structural PTOA classification and weighted κ of 0.83 (95% CI 0.78–0.88) for the overall Kellgren–Lawrence grading scale. For the main radiographic variables, Cohen’s κ values ranged from 0.79 to 0.93, and the ICC for posterior malleolar fragment size was 0.91 (95% CI 0.87–0.94), indicating good overall interobserver agreement (Supplementary Table S1). Univariable associations with structural PTOA and Kaplan–Meier analysis In univariable Cox regression analysis performed in the training cohort, a wide range of baseline, biochemical, nursing/perioperative, radiographic, surgical, and early postoperative clinical variables were significantly associated with structural PTOA. Among the most notable predictors were age, BMI, diabetes, osteoporosis, high-energy injury, previous ankle injury, CRP, ESR, HbA1c, albumin, D-dimer, time from injury to surgery, poor preoperative soft-tissue condition, open fracture, nutritional risk, frailty, Danis-Weber type C classification, fracture-dislocation, posterior malleolar involvement, syndesmotic injury, talar shift, medial clear space widening, severe articular comminution, syndesmotic fixation, operative time, blood loss, postoperative residual step-off/gap > 2 mm, incomplete syndesmotic reduction, and delayed early weight-bearing. In contrast, sex, alcohol consumption, injury side, time from injury to admission, fracture involving the medial malleolus, and specific posterior malleolar fixation were not significantly associated with PTOA risk (Table 3). Table 3Univariate Cox proportional hazards regression analysis of candidate predictors for structural PTOA in the training cohortVariablesHazard Ratio (HR)95% Confidence Interval (CI)P ValueBasic Clinical Characteristics Age (per 1-year increase)1.051.03–1.07< 0.001 Male1.180.82–1.680.365 BMI (per 1 kg/m² increase)1.151.09–1.21< 0.001 Smoking history1.481.05–2.080.024 Alcohol consumption1.120.78–1.610.538 History of diabetes2.151.45–3.18< 0.001 History of hypertension1.420.95–2.110.086 History of osteoporosis1.881.25–2.820.002 High-energy injury mechanism2.681.85–3.88< 0.001 Injury side (Right)0.980.70–1.380.925 History of previous ankle injury1.951.28–2.980.002 Heavy physical labor/high daily activity1.621.15–2.280.006 ASA classification (≥ III)1.681.06–2.650.028Preoperative Biochemical Indices WBC (per 1 × 10⁹/L increase)1.181.10–1.26< 0.001 Neutrophil ratio (NEU%, per 1% increase)1.061.04–1.08< 0.001 CRP (per 1 mg/L increase)1.041.03–1.06< 0.001 ESR (per 1 mm/h increase)1.051.03–1.07< 0.001 Fasting blood glucose (per 1 mmol/L increase)1.221.12–1.34< 0.001 Glycosylated hemoglobin (HbA1c, per 1% increase)1.451.22–1.72< 0.001 Albumin (per 1 g/L decrease)1.121.06–1.18< 0.001 D-dimer (per 1 mg/L increase)1.281.16–1.42< 0.001 Total cholesterol (TC, per 1 mmol/L increase)1.21.05–1.380.008Nursing and Perioperative Status Time from injury to admission (per 1-day increase)1.050.96–1.150.264 Time from injury to surgery (per 1-day increase)1.11.05–1.16< 0.001 Poor preoperative soft tissue condition2.251.62–3.12< 0.001 Open fracture2.851.80–4.52< 0.001 NRS-2002 ≥ 3 points2.31.60–3.32< 0.001 CFS ≥ 4 points1.921.28–2.880.002 Preoperative pain score (VAS, per 1-point increase)1.451.25–1.68< 0.001Fracture and Radiographic Characteristics Danis-Weber classification (Reference: Type A) - Type B2.151.12–4.150.022 - Type C4.852.50–9.42< 0.001 Lauge-Hansen pronation-type injury (PER/PAD)2.181.55–3.06< 0.001 Combined with ankle dislocation/subluxation3.252.32–4.55< 0.001 Fracture involving medial malleolus1.150.81–1.620.445 Fracture involving posterior malleolus2.051.46–2.88< 0.001 Posterior malleolar fragment > 25%3.152.22–4.48< 0.001 Syndesmotic injury3.422.45–4.78< 0.001 Talar shift > 2 mm2.882.05–4.05< 0.001 Widening of medial clear space2.451.75–3.42< 0.001 Severe articular comminution3.452.30–5.18< 0.001Surgical Details and Reduction Quality Surgical approach (Minimally invasive/percutaneous)0.550.32–0.950.032 Specific fixation of posterior malleolus1.220.86–1.720.258 Syndesmotic fixation2.651.88–3.72< 0.001 Operative time (per 10-minute increase)1.181.10–1.26< 0.001 Intraoperative blood loss (per 50 mL increase)1.251.12–1.38< 0.001 Postoperative residual step-off/gap > 2 mm4.122.75–6.15< 0.001 Postoperative incomplete syndesmotic reduction3.852.42–6.12< 0.001 Early weight-bearing protocol (< 6 weeks)0.480.32–0.72< 0.001Traditional early postoperative clinical evaluation models Postoperative 3-month AOFAS (per 1-point decrease)1.121.08–1.16< 0.001 Postoperative 3-month OMAS (per 1-point decrease)1.11.07–1.14< 0.001 Postoperative 3-month VAS (per 1-point increase)1.651.42–1.92< 0.001Hazard ratios (HRs) with 95% confidence intervals (CIs) were estimated using univariate Cox proportional hazards models in the training cohort. Unless otherwise specified, HRs for continuous variables represent the risk associated with a 1-unit increase. For albumin, postoperative 3-month AOFAS, and postoperative 3-month OMAS, HRs were calculated per 1-unit decrease. For Danis–Weber classification, type A served as the reference category. Binary variables were coded as yes versus no Table 3Univariate Cox proportional hazards regression analysis of candidate predictors for structural PTOA in the training cohortVariablesHazard Ratio (HR)95% Confidence Interval (CI)P ValueBasic Clinical Characteristics Age (per 1-year increase)1.051.03–1.07< 0.001 Male1.180.82–1.680.365 BMI (per 1 kg/m² increase)1.151.09–1.21< 0.001 Smoking history1.481.05–2.080.024 Alcohol consumption1.120.78–1.610.538 History of diabetes2.151.45–3.18< 0.001 History of hypertension1.420.95–2.110.086 History of osteoporosis1.881.25–2.820.002 High-energy injury mechanism2.681.85–3.88< 0.001 Injury side (Right)0.980.70–1.380.925 History of previous ankle injury1.951.28–2.980.002 Heavy physical labor/high daily activity1.621.15–2.280.006 ASA classification (≥ III)1.681.06–2.650.028Preoperative Biochemical Indices WBC (per 1 × 10⁹/L increase)1.181.10–1.26< 0.001 Neutrophil ratio (NEU%, per 1% increase)1.061.04–1.08< 0.001 CRP (per 1 mg/L increase)1.041.03–1.06< 0.001 ESR (per 1 mm/h increase)1.051.03–1.07< 0.001 Fasting blood glucose (per 1 mmol/L increase)1.221.12–1.34< 0.001 Glycosylated hemoglobin (HbA1c, per 1% increase)1.451.22–1.72< 0.001 Albumin (per 1 g/L decrease)1.121.06–1.18< 0.001 D-dimer (per 1 mg/L increase)1.281.16–1.42< 0.001 Total cholesterol (TC, per 1 mmol/L increase)1.21.05–1.380.008Nursing and Perioperative Status Time from injury to admission (per 1-day increase)1.050.96–1.150.264 Time from injury to surgery (per 1-day increase)1.11.05–1.16< 0.001 Poor preoperative soft tissue condition2.251.62–3.12< 0.001 Open fracture2.851.80–4.52< 0.001 NRS-2002 ≥ 3 points2.31.60–3.32< 0.001 CFS ≥ 4 points1.921.28–2.880.002 Preoperative pain score (VAS, per 1-point increase)1.451.25–1.68< 0.001Fracture and Radiographic Characteristics Danis-Weber classification (Reference: Type A) - Type B2.151.12–4.150.022 - Type C4.852.50–9.42< 0.001 Lauge-Hansen pronation-type injury (PER/PAD)2.181.55–3.06< 0.001 Combined with ankle dislocation/subluxation3.252.32–4.55< 0.001 Fracture involving medial malleolus1.150.81–1.620.445 Fracture involving posterior malleolus2.051.46–2.88< 0.001 Posterior malleolar fragment > 25%3.152.22–4.48< 0.001 Syndesmotic injury3.422.45–4.78< 0.001 Talar shift > 2 mm2.882.05–4.05< 0.001 Widening of medial clear space2.451.75–3.42< 0.001 Severe articular comminution3.452.30–5.18< 0.001Surgical Details and Reduction Quality Surgical approach (Minimally invasive/percutaneous)0.550.32–0.950.032 Specific fixation of posterior malleolus1.220.86–1.720.258 Syndesmotic fixation2.651.88–3.72< 0.001 Operative time (per 10-minute increase)1.181.10–1.26< 0.001 Intraoperative blood loss (per 50 mL increase)1.251.12–1.38< 0.001 Postoperative residual step-off/gap > 2 mm4.122.75–6.15< 0.001 Postoperative incomplete syndesmotic reduction3.852.42–6.12< 0.001 Early weight-bearing protocol (< 6 weeks)0.480.32–0.72< 0.001Traditional early postoperative clinical evaluation models Postoperative 3-month AOFAS (per 1-point decrease)1.121.08–1.16< 0.001 Postoperative 3-month OMAS (per 1-point decrease)1.11.07–1.14< 0.001 Postoperative 3-month VAS (per 1-point increase)1.651.42–1.92< 0.001Hazard ratios (HRs) with 95% confidence intervals (CIs) were estimated using univariate Cox proportional hazards models in the training cohort. Unless otherwise specified, HRs for continuous variables represent the risk associated with a 1-unit increase. For albumin, postoperative 3-month AOFAS, and postoperative 3-month OMAS, HRs were calculated per 1-unit decrease. For Danis–Weber classification, type A served as the reference category. Binary variables were coded as yes versus no Univariate Cox proportional hazards regression analysis of candidate predictors for structural PTOA in the training cohort Hazard ratios (HRs) with 95% confidence intervals (CIs) were estimated using univariate Cox proportional hazards models in the training cohort. Unless otherwise specified, HRs for continuous variables represent the risk associated with a 1-unit increase. For albumin, postoperative 3-month AOFAS, and postoperative 3-month OMAS, HRs were calculated per 1-unit decrease. For Danis–Weber classification, type A served as the reference category. Binary variables were coded as yes versus no Kaplan–Meier analyses further demonstrated clear separation of PTOA-free survival curves according to representative risk strata. Patients with higher BMI, longer time to surgery, elevated preoperative CRP, fracture-dislocation, syndesmotic injury, and postoperative residual articular step-off/gap > 2 mm and incomplete syndesmotic reduction showed significantly lower PTOA-free survival during follow-up. All six representative curves showed statistically significant differences by the log-rank test, supporting the clinical relevance of these predictors in time-to-event analysis (Fig. 2). Fig. 2Kaplan–Meier curves for PTOA-free survival according to representative risk strata in the training cohort. A–F Kaplan–Meier analyses were performed for representative predictors identified from the training cohort, including BMI group, time to surgery, preoperative CRP level, fracture-dislocation status, syndesmotic injury, and reduction quality. For descriptive stratification, poor reduction quality was defined as postoperative residual articular step-off/gap > 2 mm and/or postoperative incomplete syndesmotic reduction. P values were calculated using the log-rank test. Shaded bands indicate 95% confidence intervals Kaplan–Meier curves for PTOA-free survival according to representative risk strata in the training cohort. A–F Kaplan–Meier analyses were performed for representative predictors identified from the training cohort, including BMI group, time to surgery, preoperative CRP level, fracture-dislocation status, syndesmotic injury, and reduction quality. For descriptive stratification, poor reduction quality was defined as postoperative residual articular step-off/gap > 2 mm and/or postoperative incomplete syndesmotic reduction. P values were calculated using the log-rank test. Shaded bands indicate 95% confidence intervals Feature selection and multivariable Cox modeling LASSO regression was applied in the training cohort to reduce dimensionality and select the most informative predictors for subsequent multivariable modeling. The coefficient profiles gradually shrank toward zero as the penalization parameter increased, and cross-validation identified the optimal λ values for variable retention. On this basis, 14 candidate predictors were retained and entered into the full multivariable Cox regression model (Fig. 3). Fig. 3LASSO-based feature selection in the training cohort. A LASSO coefficient profiles of candidate predictors plotted against log(λ). B Cross-validation curve used to identify the optimal penalization parameter, with λ_min and λ_1se indicated by the dotted vertical lines LASSO-based feature selection in the training cohort. A LASSO coefficient profiles of candidate predictors plotted against log(λ). B Cross-validation curve used to identify the optimal penalization parameter, with λ_min and λ_1se indicated by the dotted vertical lines In the full multivariable Cox model, 11 variables remained independently associated with structural PTOA: age (HR 1.04, 95% CI 1.01–1.07, P = 0.012), BMI ≥ 24 kg/m² (HR 1.45, 95% CI 1.12–1.88, P = 0.005), high-energy injury mechanism (HR 1.58, 95% CI 1.15–2.18, P = 0.005), preoperative CRP (HR 1.03 per 1 mg/L increase, 95% CI 1.01–1.05, P = 0.008), time to surgery > 7 days (HR 1.52, 95% CI 1.10–2.10, P = 0.011), fracture-dislocation/subluxation (HR 2.15, 95% CI 1.48–3.12, P < 0.001), syndesmotic injury (HR 1.85, 95% CI 1.32–2.58, P < 0.001), posterior malleolar fragment > 25% (HR 1.72, 95% CI 1.25–2.38, P = 0.001), severe articular comminution (HR 2.05, 95% CI 1.42–2.95, P < 0.001), postoperative residual articular step-off/gap > 2 mm (HR 2.85, 95% CI 1.95–4.18, P < 0.001), and postoperative incomplete syndesmotic reduction (HR 2.45, 95% CI 1.65–3.62, P < 0.001). By contrast, D-dimer, poor preoperative soft-tissue condition, and operative time were included in the full multivariable model but did not retain statistical significance after adjustment (Table 4; Fig. 4). Table 4Full multivariable Cox regression model for structural PTOA in the training cohortVariablesHazard Ratio (HR)95% Confidence Interval (CI)P ValueAge (per 1-year increase)1.041.01–1.070.012BMI Group (≥ 24 kg/m²)1.451.12–1.880.005High-energy injury mechanism1.581.15–2.180.005Preoperative CRP (per 1 mg/L)1.031.01–1.050.008D-dimer (per 1 mg/L)1.080.95–1.220.246Time to Surgery (> 7 days)1.521.10–2.100.011Poor preoperative soft tissue condition1.250.88–1.780.215Combined with ankle dislocation/subluxation2.151.48–3.12< 0.001Syndesmotic injury1.851.32–2.58< 0.001Posterior malleolar fragment > 25%1.721.25–2.380.001Severe articular comminution2.051.42–2.95< 0.001Operative time (per 10-min increase)1.050.98–1.120.182Postoperative residual step-off/gap > 2 mm2.851.95–4.18< 0.001Postoperative incomplete syndesmotic reduction2.451.65–3.62< 0.001Variables retained after LASSO-based feature selection were entered into the multivariable Cox proportional hazards model. Adjusted HRs and 95% CIs are presented. BMI group was dichotomized as ≥ 24 kg/m² versus < 24 kg/m², and time to surgery was dichotomized as > 7 days versus ≤ 7 days Table 4Full multivariable Cox regression model for structural PTOA in the training cohortVariablesHazard Ratio (HR)95% Confidence Interval (CI)P ValueAge (per 1-year increase)1.041.01–1.070.012BMI Group (≥ 24 kg/m²)1.451.12–1.880.005High-energy injury mechanism1.581.15–2.180.005Preoperative CRP (per 1 mg/L)1.031.01–1.050.008D-dimer (per 1 mg/L)1.080.95–1.220.246Time to Surgery (> 7 days)1.521.10–2.100.011Poor preoperative soft tissue condition1.250.88–1.780.215Combined with ankle dislocation/subluxation2.151.48–3.12< 0.001Syndesmotic injury1.851.32–2.58< 0.001Posterior malleolar fragment > 25%1.721.25–2.380.001Severe articular comminution2.051.42–2.95< 0.001Operative time (per 10-min increase)1.050.98–1.120.182Postoperative residual step-off/gap > 2 mm2.851.95–4.18< 0.001Postoperative incomplete syndesmotic reduction2.451.65–3.62< 0.001Variables retained after LASSO-based feature selection were entered into the multivariable Cox proportional hazards model. Adjusted HRs and 95% CIs are presented. BMI group was dichotomized as ≥ 24 kg/m² versus < 24 kg/m², and time to surgery was dichotomized as > 7 days versus ≤ 7 days Full multivariable Cox regression model for structural PTOA in the training cohort Variables retained after LASSO-based feature selection were entered into the multivariable Cox proportional hazards model. Adjusted HRs and 95% CIs are presented. BMI group was dichotomized as ≥ 24 kg/m² versus < 24 kg/m², and time to surgery was dichotomized as > 7 days versus ≤ 7 days Fig. 4Forest plot of the full multivariable Cox regression model for structural PTOA in the training cohort. Adjusted hazard ratios (HRs) and 95% confidence intervals (CIs) for predictors retained in the final multivariable Cox model are shown. The vertical reference line indicates HR = 1.0 Forest plot of the full multivariable Cox regression model for structural PTOA in the training cohort. Adjusted hazard ratios (HRs) and 95% confidence intervals (CIs) for predictors retained in the final multivariable Cox model are shown. The vertical reference line indicates HR = 1.0 These 11 statistically significant predictors were then retained to construct the final parsimonious prediction model. Spearman correlation analysis showed that pairwise correlations among the 11 predictors were generally weak to modest, suggesting no major concern regarding severe collinearity in the final model (Fig. 5). Fig. 5Spearman correlation heatmap of the 11 predictors included in the final parsimonious prediction model. The lower triangle displays pairwise Spearman correlation coefficients among predictors retained in the final model. Color intensity indicates the strength and direction of the correlation, and the diagonal values are fixed at 1.00 Spearman correlation heatmap of the 11 predictors included in the final parsimonious prediction model. The lower triangle displays pairwise Spearman correlation coefficients among predictors retained in the final model. Color intensity indicates the strength and direction of the correlation, and the diagonal values are fixed at 1.00 Incremental predictive performance of the stepwise models The discrimination performance of the traditional early postoperative clinical evaluation indicators and the five stepwise Cox models is summarized in Table 5. Among the traditional indicators, the postoperative 3-month AOFAS score showed the best discrimination, with an apparent C-index of 0.685 in the training cohort and 0.662 in the validation cohort, whereas OMAS and VAS showed lower performance. In comparison, the incremental models showed a stepwise improvement in discrimination as additional data blocks were incorporated. Model 1, which included basic clinical variables only, had relatively limited discrimination, whereas Model 4, which incorporated radiographic features, yielded a marked increase in performance. The final comprehensive model (Model 5) achieved the best discrimination, with an apparent C-index of 0.892 in the training cohort, a bootstrap-corrected C-index of 0.887, and an external validation C-index of 0.865. Table 5Predictive performance and incremental value of the traditional early postoperative clinical evaluation indicators and the five incremental Cox models in the apparent, bootstrap-corrected, and external validation cohortsPredictive ModelsApparent C-index (Training Cohort)Bootstrap-corrected C-index (Training)External Validation C-indexContinuous NRI (95% CI)IDI (95% CI)Traditional early postoperative clinical evaluation indicators Traditional 1: Postop AOFAS score0.685 (0.622–0.748)0.681 (0.619–0.743)0.662 (0.595–0.729)ReferenceReference Traditional 2: Postop OMAS score0.668 (0.605–0.731)0.664 (0.602–0.726)0.645 (0.582–0.708)-- Traditional 3: Postop VAS pain score0.635 (0.572–0.698)0.631 (0.569–0.693)0.618 (0.555–0.681)--Incremental Predictive Models Model 1: Basic Clinical0.662 (0.601–0.723)0.658 (0.598–0.718)0.648 (0.585–0.711)-0.045 (-0.125–0.035)-0.012 (-0.035–0.011) Model 2: Model 1 + Biochemical0.695 (0.635–0.755)0.690 (0.631–0.749)0.672 (0.608–0.736)0.125 (0.045–0.205)0.038 (0.012–0.064) Model 3: Model 2 + Nursing/Peri.0.728 (0.668–0.788)0.722 (0.663–0.781)0.698 (0.636–0.760)0.215 (0.125–0.305)0.085 (0.045–0.125) Model 4: Model 3 + Radiographic0.835 (0.782–0.888)0.828 (0.776–0.880)0.805 (0.748–0.862)0.545 (0.415–0.675)0.198 (0.135–0.261) Model 5: Comprehensive Model0.892 (0.855–0.929)0.887 (0.849–0.925)0.865 (0.818–0.912)0.785 (0.635–0.935)0.285 (0.212–0.358)Apparent and bootstrap-corrected C-index values were derived from the training cohort, whereas external validation C-index values were obtained from the independent validation cohort. Continuous net reclassification improvement (NRI) and integrated discrimination improvement (IDI) were used to evaluate incremental value. Continuous NRI and IDI were calculated using the postoperative 3-month AOFAS score as the reference comparator. The final comprehensive model was based on the 11 independent predictors that remained statistically significant in the full multivariable Cox regression model. The postoperative 3-month AOFAS, OMAS, and VAS scores were used as clinically relevant reference comparators. These single postoperative clinical indicators are not fully equivalent to a multivariable structural prediction model and were included to provide comparative clinical benchmarking Table 5Predictive performance and incremental value of the traditional early postoperative clinical evaluation indicators and the five incremental Cox models in the apparent, bootstrap-corrected, and external validation cohortsPredictive ModelsApparent C-index (Training Cohort)Bootstrap-corrected C-index (Training)External Validation C-indexContinuous NRI (95% CI)IDI (95% CI)Traditional early postoperative clinical evaluation indicators Traditional 1: Postop AOFAS score0.685 (0.622–0.748)0.681 (0.619–0.743)0.662 (0.595–0.729)ReferenceReference Traditional 2: Postop OMAS score0.668 (0.605–0.731)0.664 (0.602–0.726)0.645 (0.582–0.708)-- Traditional 3: Postop VAS pain score0.635 (0.572–0.698)0.631 (0.569–0.693)0.618 (0.555–0.681)--Incremental Predictive Models Model 1: Basic Clinical0.662 (0.601–0.723)0.658 (0.598–0.718)0.648 (0.585–0.711)-0.045 (-0.125–0.035)-0.012 (-0.035–0.011) Model 2: Model 1 + Biochemical0.695 (0.635–0.755)0.690 (0.631–0.749)0.672 (0.608–0.736)0.125 (0.045–0.205)0.038 (0.012–0.064) Model 3: Model 2 + Nursing/Peri.0.728 (0.668–0.788)0.722 (0.663–0.781)0.698 (0.636–0.760)0.215 (0.125–0.305)0.085 (0.045–0.125) Model 4: Model 3 + Radiographic0.835 (0.782–0.888)0.828 (0.776–0.880)0.805 (0.748–0.862)0.545 (0.415–0.675)0.198 (0.135–0.261) Model 5: Comprehensive Model0.892 (0.855–0.929)0.887 (0.849–0.925)0.865 (0.818–0.912)0.785 (0.635–0.935)0.285 (0.212–0.358)Apparent and bootstrap-corrected C-index values were derived from the training cohort, whereas external validation C-index values were obtained from the independent validation cohort. Continuous net reclassification improvement (NRI) and integrated discrimination improvement (IDI) were used to evaluate incremental value. Continuous NRI and IDI were calculated using the postoperative 3-month AOFAS score as the reference comparator. The final comprehensive model was based on the 11 independent predictors that remained statistically significant in the full multivariable Cox regression model. The postoperative 3-month AOFAS, OMAS, and VAS scores were used as clinically relevant reference comparators. These single postoperative clinical indicators are not fully equivalent to a multivariable structural prediction model and were included to provide comparative clinical benchmarking Predictive performance and incremental value of the traditional early postoperative clinical evaluation indicators and the five incremental Cox models in the apparent, bootstrap-corrected, and external validation cohorts Apparent and bootstrap-corrected C-index values were derived from the training cohort, whereas external validation C-index values were obtained from the independent validation cohort. Continuous net reclassification improvement (NRI) and integrated discrimination improvement (IDI) were used to evaluate incremental value. Continuous NRI and IDI were calculated using the postoperative 3-month AOFAS score as the reference comparator. The final comprehensive model was based on the 11 independent predictors that remained statistically significant in the full multivariable Cox regression model. The postoperative 3-month AOFAS, OMAS, and VAS scores were used as clinically relevant reference comparators. These single postoperative clinical indicators are not fully equivalent to a multivariable structural prediction model and were included to provide comparative clinical benchmarking Incremental value analysis further showed that the comprehensive model had higher predictive performance than the reference postoperative 3-month clinical indicator model, although these comparators are not fully equivalent in construct and intended use. Relative to the AOFAS-based model, the comprehensive model yielded a continuous net reclassification improvement of 0.785 (95% CI 0.635–0.935) and an integrated discrimination improvement of 0.285 (95% CI 0.212–0.358).This comparison was intended to provide clinically relevant benchmarking against commonly used postoperative indicators rather than to imply direct interchangeability between the approaches. Model 4 also showed substantial improvement, whereas Model 1 did not improve classification performance over the reference model. Overall ROC comparisons in the training and validation cohorts consistently showed that the comprehensive model had the best overall discriminative performance among all candidate models (Fig. 6). Fig. 6Overall discriminative performance of the traditional early postoperative clinical evaluation indicators and incremental prediction models in the training and external validation cohorts. Receiver operating characteristic (ROC) curves showing the overall discriminative performance of the traditional early postoperative clinical evaluation indicators and the five incremental Cox-based prediction models in the training cohort (A) and the external validation cohort (B) Overall discriminative performance of the traditional early postoperative clinical evaluation indicators and incremental prediction models in the training and external validation cohorts. Receiver operating characteristic (ROC) curves showing the overall discriminative performance of the traditional early postoperative clinical evaluation indicators and the five incremental Cox-based prediction models in the training cohort (A) and the external validation cohort (B) Time-dependent discrimination and bootstrap internal validation Time-dependent ROC analysis of the final comprehensive model demonstrated favorable discrimination across clinically relevant prediction horizons in both cohorts. In the training cohort, the time-specific AUCs were 0.898 at 6 months, 0.885 at 12 months, and 0.857 at 24 months. In the external validation cohort, the corresponding AUCs were 0.867, 0.876, and 0.845, respectively. These results indicate that the model maintained good discrimination over time, although some attenuation was observed at the longer follow-up horizon (Fig. 7). Fig. 7Time-dependent ROC curves of the final comprehensive model at 6, 12, and 24 months in the training and external validation cohorts. Time-dependent ROC analyses were performed to assess the discrimination of the final comprehensive model for predicting structural PTOA at 6, 12, and 24 months in the training cohort (A) and the external validation cohort (B) Time-dependent ROC curves of the final comprehensive model at 6, 12, and 24 months in the training and external validation cohorts. Time-dependent ROC analyses were performed to assess the discrimination of the final comprehensive model for predicting structural PTOA at 6, 12, and 24 months in the training cohort (A) and the external validation cohort (B) Bootstrap internal validation of the final Cox prediction model was performed using 1,000 resamples. The optimism was low (0.005), with only a minimal decrease from the apparent C-index (0.892) to the bootstrap-corrected C-index (0.887). The calibration slope was 0.985 (95% CI 0.912–1.058), and the calibration-in-the-large was − 0.012 (95% CI -0.085 to 0.061), indicating good overall agreement between predicted and observed risk. The Brier score was 0.115 (95% CI 0.095–0.135), supporting acceptable overall predictive accuracy (Table 6). Table 6Bootstrap internal validation of the final Cox prediction model in the training cohortMetricValue (95% CI)C-index (Apparent)0.892 (0.855–0.929)C-index (Bootstrap-corrected)0.887 (0.849–0.925)Optimism0.005Calibration Slope0.985 (0.912–1.058)Calibration-in-the-large / Intercept-0.012 (-0.085–0.061)Brier Score0.115 (0.095–0.135)Bootstrap Resamples (B)1000Internal validation was performed using 1000 bootstrap resamples. Optimism was calculated as the difference between the apparent and bootstrap-corrected C-index. A calibration slope close to 1.0 and a calibration-in-the-large (intercept) close to 0 indicate good agreement between predicted and observed risks. Lower Brier scores indicate better overall model accuracy Table 6Bootstrap internal validation of the final Cox prediction model in the training cohortMetricValue (95% CI)C-index (Apparent)0.892 (0.855–0.929)C-index (Bootstrap-corrected)0.887 (0.849–0.925)Optimism0.005Calibration Slope0.985 (0.912–1.058)Calibration-in-the-large / Intercept-0.012 (-0.085–0.061)Brier Score0.115 (0.095–0.135)Bootstrap Resamples (B)1000Internal validation was performed using 1000 bootstrap resamples. Optimism was calculated as the difference between the apparent and bootstrap-corrected C-index. A calibration slope close to 1.0 and a calibration-in-the-large (intercept) close to 0 indicate good agreement between predicted and observed risks. Lower Brier scores indicate better overall model accuracy Bootstrap internal validation of the final Cox prediction model in the training cohort Internal validation was performed using 1000 bootstrap resamples. Optimism was calculated as the difference between the apparent and bootstrap-corrected C-index. A calibration slope close to 1.0 and a calibration-in-the-large (intercept) close to 0 indicate good agreement between predicted and observed risks. Lower Brier scores indicate better overall model accuracy Nomogram development, calibration, and clinical utility A nomogram was constructed based on the 11 independent predictors retained in the final parsimonious model, including age, BMI ≥ 24 kg/m², high-energy injury mechanism, preoperative CRP, time to surgery > 7 days, ankle dislocation/subluxation, syndesmotic injury, posterior malleolar fragment involving > 25% of the articular surface, severe articular comminution, postoperative residual articular step-off/gap > 2 mm, and postoperative incomplete syndesmotic reduction. Among these predictors, postoperative residual articular step-off/gap > 2 mm, postoperative incomplete syndesmotic reduction, ankle dislocation/subluxation, and severe articular comminution showed the strongest associations with structural PTOA, whereas age and preoperative CRP had relatively smaller effect sizes (Fig. 8). Fig. 8The nomogram integrates the 11 independent predictors retained for the final parsimonious prediction model for individualized prediction of 6-, 12-, and 24-month structural PTOA risk. The nomogram integrates the predictors retained in the final multivariable Cox model to estimate individual probabilities of structural PTOA at 6, 12, and 24 months after ankle fracture surgery. Total points correspond to the predicted risk at each time horizon The nomogram integrates the 11 independent predictors retained for the final parsimonious prediction model for individualized prediction of 6-, 12-, and 24-month structural PTOA risk. The nomogram integrates the predictors retained in the final multivariable Cox model to estimate individual probabilities of structural PTOA at 6, 12, and 24 months after ankle fracture surgery. Total points correspond to the predicted risk at each time horizon Calibration analysis demonstrated good agreement between predicted and observed PTOA risk. In the training cohort, the 12- and 24-month calibration curves showed close alignment with the ideal reference line after bootstrap correction. In the external validation cohort, the corresponding calibration curves also showed acceptable agreement, although minor deviation was present in the middle risk range. Decision curve analysis showed that the nomogram-based comprehensive model consistently provided greater net benefit than the treat-all and treat-none strategies across a clinically relevant range of threshold probabilities, and also showed greater net benefit than the AOFAS-based reference comparator across the evaluated threshold range. These findings support the potential clinical utility of the nomogram for individualized postoperative risk stratification (Fig. 9). Fig. 9Calibration and decision curve analysis of the nomogram in the training and external validation cohorts. A Bootstrap-corrected calibration curves for the 12- and 24-month nomogram predictions in the training cohort. B Calibration curves for the 12- and 24-month nomogram predictions in the external validation cohort. C Decision curve analysis (DCA) in the training cohort. D Decision curve analysis in the external validation cohort. In the training cohort, calibration was internally validated using 1000 bootstrap resamples to reduce overfitting bias Calibration and decision curve analysis of the nomogram in the training and external validation cohorts. A Bootstrap-corrected calibration curves for the 12- and 24-month nomogram predictions in the training cohort. B Calibration curves for the 12- and 24-month nomogram predictions in the external validation cohort. C Decision curve analysis (DCA) in the training cohort. D Decision curve analysis in the external validation cohort. In the training cohort, calibration was internally validated using 1000 bootstrap resamples to reduce overfitting bias Distributional characteristics of continuous predictors and nonlinear risk relationships Before inclusion in the model, the distributions of continuous variables were assessed in the training cohort. Age and BMI were approximately normally distributed, whereas CRP and time from injury to surgery were markedly right-skewed. After log10 transformation, the distributions of CRP and time to surgery improved to some extent, particularly for CRP, although the transformation was used primarily to explore distributional characteristics rather than to redefine the final analytic model. These supplementary findings supported the need to further examine possible nonlinear effects of continuous predictors (Figure S1, S2). Restricted cubic spline analyses were performed as supplementary exploratory analyses to examine the continuous dose–response relationships of age, BMI, preoperative CRP, and time to surgery with structural PTOA risk. These analyses were intended to characterize potential nonlinearity and were not used to derive the categorical thresholds applied in the main Cox model. Age, BMI, and time to surgery all showed significant overall and nonlinear associations with structural PTOA risk (all P_overall = 0.005; P_nonlinear = 0.015 for age, 0.002 for BMI, and 0.009 for time to surgery), indicating that the risk relationship for these predictors was not adequately captured by a simple linear assumption. In contrast, preoperative CRP showed a significant overall association with structural PTOA risk (P_overall = 0.005) but no evidence of significant nonlinearity (P_nonlinear = 0.801), suggesting a predominantly linear increase in risk across its observed range (Fig. 10). Fig. 10Restricted cubic spline curves showing the dose–response relationships between continuous predictors and structural PTOA risk in the training cohort. A CRP, B BMI, C age, and D time to surgery with structural PTOA risk. The solid red lines indicate adjusted hazard ratios, and the shaded areas represent 95% confidence intervals. The histograms along the x-axes show the distribution of the corresponding variables. P_overall and P_nonlinear values are displayed in each panel Restricted cubic spline curves showing the dose–response relationships between continuous predictors and structural PTOA risk in the training cohort. A CRP, B BMI, C age, and D time to surgery with structural PTOA risk. The solid red lines indicate adjusted hazard ratios, and the shaded areas represent 95% confidence intervals. The histograms along the x-axes show the distribution of the corresponding variables. P_overall and P_nonlinear values are displayed in each panel Discussion Principal findings This prospective two-center study developed and externally validated a Cox-based prediction model for structural PTOA after ankle fracture surgery. The 2-year incidence of structural PTOA reached 24.0% in the training cohort and 29.0% in the external validation cohort, indicating that early structural degeneration is already common within the first postoperative years. The final parsimonious model retained 11 predictors spanning patient characteristics, injury severity, inflammatory burden, fracture morphology, and postoperative reduction quality. The comprehensive model showed strong discrimination, with an apparent C-index of 0.892, a bootstrap-corrected C-index of 0.887, and an external validation C-index of 0.865, and also outperformed traditional early postoperative clinical evaluation indicators. Importantly, this study also has practical clinical implications. By integrating multi-source perioperative data into an individualized risk prediction framework, the model may enable early identification of patients at high risk for structural PTOA at a stage when closer surveillance and targeted intervention may still be clinically beneficial. This may facilitate more personalized follow-up, timely optimization of rehabilitation and weight-bearing strategies, and better-informed counseling regarding long-term joint outcomes. The inclusion of an independent external validation cohort further strengthens the potential clinical utility of the model. The present model should therefore be interpreted as applicable to surgically treated isolated malleolar ankle fractures, rather than to talar fractures, pilon fractures, or other peri-ankle injuries. Fracture severity, instability, and reduction quality Our study demonstrated that the strongest predictors in the present model were mechanical and structural rather than demographic. Postoperative residual articular step-off/gap > 2 mm (HR 2.85), postoperative incomplete syndesmotic reduction (HR 2.45), ankle dislocation/subluxation (HR 2.15), and severe articular comminution (HR 2.05) had the largest effect sizes, followed by syndesmotic injury (HR 1.85) and posterior malleolar fragment involvement > 25% (HR 1.72). This pattern suggests that PTOA risk is driven mainly by the magnitude of the initial articular injury and the quality of postoperative restoration of congruity and stability. This interpretation is consistent with previous reports. Lübbeke et al. [9]identified Weber C fractures, associated medial malleolar fractures, fracture-dislocation, higher BMI, age ≥ 30 years, and longer time since surgery as significant long-term risk factors, and estimated that the probability of developing PTOA increased to 60%–70% when three or more risk factors were present. Beak et al. [16] likewise reported that obesity, fracture-dislocation injury, concomitant large posterior malleolar fracture, and articular incongruence were significant risk factors for PTOA after ankle fracture surgery. Wang et al. [10] further showed that posterior malleolar involvement was not only important in itself, but also graded by fragment size, with odds ratios of 2.024 for posterior malleolar involvement ≤ 25% and 3.533 for involvement > 25%, while postoperative articular incongruity independently increased risk (OR 2.011, 95% CI 1.138–4.058). Our results are broadly concordant with these reports, but place even greater emphasis on postoperative residual step-off and syndesmotic malreduction, both of which remained among the most influential predictors after multivariable adjustment. From a biomechanical perspective, this pattern is expected. Persistent incongruity alters tibiotalar contact mechanics and promotes focal overload of cartilage and subchondral bone. Review articles on ankle PTOA consistently identify malalignment, instability, articular incongruity, and concomitant soft-tissue damage as central mechanisms of degeneration after trauma [17–19]. The present data align with that framework and suggest that the long-term consequence of ankle fracture is determined not only by whether the fracture is fixed, but by how precisely joint congruity and syndesmotic alignment are restored [20, 21]. Patient-related and biologic contributors Several host-related and biologic factors also contributed independently to risk heterogeneity. Age, BMI ≥ 24 kg/m², high-energy injury mechanism, preoperative CRP, and time to surgery > 7 days all remained significant in the final model. Their effect sizes were smaller than those of postoperative residual incongruity or syndesmotic malreduction, but they remained relevant after adjustment for structural injury markers. This suggests that PTOA after ankle fracture is not a purely mechanical disorder and that patient-specific susceptibility may influence the transition from injury to radiographic degeneration [22–24]. Our age- and BMI-related findings are concordant with prior literature on ankle fracture prognosis. Verhage et al. [25] reported that higher age independently increased the risk of post-traumatic osteoarthritis after trimalleolar fracture, whereas BMI was independently associated with worse long-term functional outcome. Similarly, Cardoso et al. [26] demonstrated that, after operative treatment of rotational ankle fractures, obese patients had significantly greater pain and functional limitation at long-term follow-up, with a graded decline in outcome across increasing BMI levels. Building on these observations, our model indicates that the adverse effect of BMI persists even after accounting for major injury-pattern and reduction-related variables. This supports the view that obesity is not merely a marker of fracture severity, but may act as an independent contributor to joint degeneration by increasing chronic mechanical demand and amplifying the adverse biological milieu within the injured ankle [27, 28]. Preoperative CRP is another noteworthy signal. In the restricted cubic spline analysis, CRP showed a significant overall association with PTOA risk without clear evidence of nonlinearity, suggesting a largely monotonic relationship across the observed range. CRP is not a joint-specific biomarker, but it is a reasonable surrogate for peri-traumatic inflammatory burden. Experimental and translational work has increasingly emphasized the role of post-injury inflammation in PTOA, including synovial activation, matrix degradation, chondrocyte apoptosis, and abnormal subchondral remodeling. The retention of CRP in the final model supports the view that early biologic response adds prognostic information beyond fracture morphology alone [29]. Timing of surgery and early structural progression Time to surgery > 7 days remained independently associated with PTOA in the final model, and the spline analysis suggested a nonlinear increase in risk with longer delay. This finding should be interpreted cautiously. It does not mean that every ankle fracture should undergo immediate surgery regardless of soft-tissue condition; rather, it suggests that prolonged delay is likely a marker of greater overall risk, potentially reflecting severe soft-tissue swelling, more complex injury patterns, referral delay, or a longer period of unreduced or poorly reduced joint incongruity. There is some support for this interpretation in prior literature. In ankle fracture-dislocations, Fonkoué et al. [11] reported that early PTOA developed in 19 of 44 patients (43.2%) after a mean follow-up of 27.2 ± 18.3 months, and delay since trauma was associated with PTOA. Their study also showed that anatomic reduction was achieved in only 50% of patients and that the talus was centered in the mortise in only 68.2%, illustrating how delayed or difficult management may coexist with incomplete restoration of joint mechanics. Our data are not directly comparable because the present cohort was broader and included all surgically treated ankle fractures, but both studies support the idea that delayed definitive management may identify a subgroup with heightened structural risk. The present study may therefore have practical implications for postoperative surveillance. Patients with delayed surgery, especially when combined with fracture-dislocation, posterior malleolar involvement, syndesmotic injury, or poor reduction quality, may warrant more intensive radiographic follow-up during the first 2 years after surgery, even if early clinical recovery appears acceptable. Predictive performance and incremental value A major strength of this study lies in the stepwise multimodal modeling strategy. The apparent C-index increased from 0.662 in the basic clinical model to 0.695 after adding biochemical markers, 0.728 after adding nursing/perioperative variables, 0.835 after adding fracture and radiographic variables, and 0.892 in the final comprehensive model. The largest gain occurred when radiographic variables were introduced, which is consistent with the strong etiologic role of fracture severity, instability, and reduction quality. At the same time, the incremental improvements from biochemical and perioperative information indicate that structural descriptors alone do not fully explain risk [30, 31]. The comparison with traditional early postoperative clinical evaluation indicators is also informative. In the present dataset, the 3-month AOFAS score performed better than OMAS and VAS among the conventional indicators, but its apparent C-index was 0.685, which was lower than that of the comprehensive model (0.892). This difference should be interpreted with caution, because the 3-month postoperative clinical scores are single postoperative status measures, whereas the proposed model is a multivariable structural risk prediction model. Relative to AOFAS, the final model improved risk classification substantially, with a continuous NRI of 0.785 and an IDI of 0.285. This difference is clinically reasonable. Early symptom and function scores are influenced by pain control, rehabilitation progress, and short-term mobility, whereas structural PTOA is the result of a longer interplay among joint injury, instability, inflammation, and biomechanical overload. A single postoperative functional score cannot fully substitute for that multidimensional information [32]. Model robustness was also supported by validation analyses. Bootstrap internal validation showed minimal optimism (0.005), and the external validation C-index remained high at 0.865. Calibration and decision curve analyses further suggested that the model may have practical clinical utility. From a translational perspective, that may be one of the key contributions of this work. The ankle fracture PTOA literature contains many studies on isolated risk factors, but comparatively few clinically oriented multivariable prediction tools with independent external validation. A model that remains stable across two centers is more likely to be useful for individualized surveillance, counseling, and follow-up planning after ankle fracture surgery [33, 34]. Limitations and future directions Several limitations should be acknowledged. First, although this was a prospective two-center study with geographic external validation, both cohorts were derived from the same regional setting, which may limit the generalizability of the model to other populations, healthcare systems, and surgical practices. Second, the minimum follow-up duration was 2 years; while appropriate for evaluating early structural PTOA, it may underestimate the full long-term burden of post-traumatic degeneration. Third, the primary endpoint was based on radiographic structural PTOA. Future studies should further integrate patient-reported outcomes and advanced imaging markers to improve clinical relevance. Fourth, some potentially important factors, such as detailed rehabilitation adherence, postoperative loading behavior, implant-related variables, and joint-specific biochemical or MRI-based markers, were not available for inclusion. In addition, although bootstrap internal validation and external validation supported model robustness, the nomogram still requires validation in larger multicenter cohorts and in populations with different fracture patterns and treatment strategies. Future studies should therefore focus on broader multicenter validation, longer follow-up, dynamic risk updating during recovery, and integration of imaging, biologic, and rehabilitation-related data to further refine individualized prediction of PTOA after ankle fracture surgery. Conclusion In conclusion, we developed and externally validated a Cox-based prediction model for structural PTOA after ankle fracture surgery. The final 11-variable model showed good discrimination, satisfactory calibration, and clear incremental value over traditional early postoperative clinical evaluation indicators. This model may help support individualized risk stratification and postoperative follow-up after ankle fracture surgery. Supplementary Information Supplementary Material 1: Figure S1. Distribution plots of age, BMI, CRP, and time from injury to surgery in the training cohort. Histograms with kernel density curves and theoretical normal distribution curves are shown for age (A), BMI (B), CRP (C), and time from injury to surgery (D). Shapiro–Wilk P values are displayed to describe distributional characteristics. Supplementary Material 1: Figure S1. Distribution plots of age, BMI, CRP, and time from injury to surgery in the training cohort. Histograms with kernel density curves and theoretical normal distribution curves are shown for age (A), BMI (B), CRP (C), and time from injury to surgery (D). Shapiro–Wilk P values are displayed to describe distributional characteristics. Supplementary Material 2: Figure S2. Distribution plots of log10-transformed CRP and time from injury to surgery in the training cohort. Histograms with kernel density curves and theoretical normal distribution curves are shown for log10(CRP + 1) (A) and log10(time from injury to surgery + 1) (B). The transformation was used to explore distributional improvement before modeling continuous variables. Supplementary Material 2: Figure S2. Distribution plots of log10-transformed CRP and time from injury to surgery in the training cohort. Histograms with kernel density curves and theoretical normal distribution curves are shown for log10(CRP + 1) (A) and log10(time from injury to surgery + 1) (B). The transformation was used to explore distributional improvement before modeling continuous variables. Supplementary Material 3: Table S1. Interobserver reliability for the primary radiographic endpoint and main radiographic variables. Supplementary Material 3: Table S1. Interobserver reliability for the primary radiographic endpoint and main radiographic variables.