Factors associated with 30-day mortality in patients with post-myocardial infarction ventricular septal rupture: a retrospective cohort study

Abstract

Background

Post-myocardial infarction ventricular septal rupture (VSR) is associated with high mortality, and comparative outcome data across contemporary treatment strategies remain limited. We aimed to identify factors associated with 30-day mortality, compare observed outcomes across four treatment strategies, including surgical repair combining an occluder and a patch (SurCOP), and develop an internally validated multivariable model incorporating the treatment ultimately received for retrospective prognostic assessment.

Methods

Between May 2012 and October 2025, 214 patients with post-myocardial infarction VSR were included and classified according to 30-day survival status. Baseline characteristics, clinical presentations, and treatment strategies, including conservative medical therapy, transcatheter closure, traditional surgical repair, and SurCOP, were compared. Univariable and multivariable logistic regression analyses were performed to identify factors associated with 30-day mortality. The prognostic model was developed in a 187-patient complete-case cohort defined before variable selection. Model discrimination was assessed using receiver operating characteristic (ROC) curve analysis, and internal validation was performed using stratified 10-fold cross-validation and bootstrap optimism correction with 500 resamples.

Results

Among the 214 enrolled patients with VSR, the overall 30-day mortality rate was 51.9%. Notably, observed 30-day mortality was lowest in the SurCOP group (21.8%), although treatment-group comparisons should be interpreted in light of non-random treatment allocation. Among the 123 patients who underwent a definitive invasive intervention, the median interval from VSR diagnosis to treatment differed among the SurCOP, traditional surgical repair, and transcatheter closure groups (6.5, 31.4, and 17.8 days, respectively; P<0.001). Multivariable logistic regression analysis revealed that conservative medical therapy was associated with a higher odds of mortality compared to the SurCOP technique [odds ratio (OR), 27.822]. Additionally, elevated admission creatinine (OR, 1.010) and lactate (OR, 1.337) were independently associated with higher 30-day mortality. Conversely, a history of angina pectoris or myocardial infarction (OR, 0.235) was independently associated with lower 30-day mortality. A retrospective prognostic model incorporating the treatment ultimately received was subsequently constructed. The model demonstrated good discriminatory performance, with an apparent area under the ROC curve (AUC) of 0.906 [95% confidence interval (CI): 0.863–0.948], a sensitivity of 0.768, and a specificity of 0.943. Stratified 10-fold cross-validation yielded an AUC of 0.875 (95% CI: 0.825–0.925), and the bootstrap optimism-corrected AUC was 0.884.

Conclusions

In this retrospective single-center cohort, treatment modality, admission creatinine, admission lactate, and a history of angina or myocardial infarction were associated with 30-day mortality. The combined model showed good apparent discrimination and retained acceptable discrimination after internal validation. Because treatment allocation was non-random and the model incorporated the treatment ultimately received, the findings should be interpreted as prognostic associations and require external validation before clinical application.

Highlight box.

Key findings

• In this retrospective cohort of 214 patients with post-myocardial infarction ventricular septal rupture (VSR), observed 30-day mortality differed across treatment groups and was highest among patients managed conservatively. Conservative medical therapy, higher admission creatinine, and higher admission lactate were associated with increased 30-day mortality, whereas a history of angina or myocardial infarction was associated with lower mortality.

• A multivariable model incorporating the treatment ultimately received showed an apparent area under the receiver operating characteristic curve (AUC) of 0.906; the cross-validated and bootstrap optimism-corrected AUCs were 0.875 and 0.884, respectively.

What is known and what is new?

• Post-myocardial infarction VSR is associated with high short-term mortality, and treatment comparisons are strongly affected by patient selection and treatment timing.

• This study compared four treatment strategies and internally validated a retrospective prognostic model, but did not demonstrate the superiority of surgical repair combining an occluder and a patch over other invasive strategies.

What is the implication, and what should change now?

• The findings support cautious retrospective prognostic assessment and multidisciplinary evaluation but should not be used to select treatment at admission. External validation and prospective multicenter studies are required.

Introduction

Ventricular septal rupture (VSR) is a rare but catastrophic mechanical complication of acute myocardial infarction (AMI) (1,2). Although modern interventional and surgical approaches have substantially improved clinical outcomes for patients with VSR, the overall mortality rate remains high. Management becomes exceptionally challenging when the condition is complicated by cardiogenic shock or multi-organ failure. Currently, the optimal surgical timing and definitive treatment strategies remain subjects of ongoing debate (3). Furthermore, surgical repair of VSR during the acute phase carries a high risk of mortality, whereas conservative medical management yields dismal survival rates (4). Transcatheter closure serves as a minimally invasive alternative for selected patients; however, its application is frequently limited by the anatomical location and size of the rupture (5).

Modified surgical approaches, including surgical repair combining an occluder and a patch (SurCOP), have been developed for selected patients with post-infarction VSR (6). A recent multicenter cohort reported 30-day outcomes, surgical timing, and preoperative risk factors in 60 patients undergoing SurCOP (7). However, comparative outcomes across conservative medical therapy, transcatheter closure, traditional surgical repair, and SurCOP remain uncertain because the available evidence is predominantly observational (5,8). Accurate prognostic assessment in patients with VSR remains clinically important. Baseline variables available at admission, including Killip class, renal function, and markers of tissue hypoperfusion, may help characterize disease severity. The treatment ultimately received may provide additional information for retrospective prognostic assessment, but it cannot be used as an admission-time predictor. Although previous studies have investigated factors associated with mortality after VSR, prognostic models integrating baseline characteristics, acute clinical status, and multiple treatment strategies remain insufficiently validated.

To address these challenges, this study evaluated a cohort of 214 patients admitted with VSR between 2012 and 2025. We assessed the prognostic impact of various treatment strategies and identified independent risk factors for 30-day mortality. Based on these findings, we developed and internally validated a multivariable model incorporating baseline characteristics and the treatment ultimately received. The model was intended for retrospective prognostic assessment rather than for admission-based risk prediction or treatment selection. We present this article in accordance with the STROBE reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1814/rc).

Methods

Study population and research design

This retrospective single-center cohort study included consecutive patients with post-myocardial infarction VSR who were treated at The First Affiliated Hospital of Zhengzhou University between May 2012 and October 2025. During the study period, from May 2012 to October 2025, a total of 26,389 patients with AMI were admitted to our institution, of whom 214 developed VSR, corresponding to an overall incidence of 0.81%. Based on 30-day outcomes, 103 patients were classified as survivors and 111 as non-survivors. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Institutional Ethics Committee of The First Affiliated Hospital of Zhengzhou University (Approval No. 2026-KY-0033-002). Because of the retrospective design, the requirement for individual informed consent was waived by the committee. The primary outcome was all-cause mortality within 30 days after the index hospital admission. Mortality status was ascertained from hospital records and follow-up data. Patients who remained alive for more than 30 days after the index admission were classified as survivors. Of the 64 patients in the SurCOP group, 55 had also been included in a previously published SurCOP cohort from our group (7). The previous study focused on outcomes, surgical timing, and preoperative risk factors among patients undergoing SurCOP, whereas the present study evaluated four treatment strategies in a broader postinfarction (PI)-VSR cohort and developed a retrospective prognostic model for 30-day mortality.

Inclusion and exclusion criteria

Patients were eligible for inclusion if they met the following criteria: (I) a confirmed diagnosis of AMI, defined by electrocardiographic findings (ST-segment elevation ≥0.1 mV in at least two contiguous limb leads or >0.2 mV in two contiguous precordial leads) and elevated cardiac biomarkers; (II) echocardiographic evidence demonstrating a structural discontinuity or defect in the ventricular septum, accompanied by a left-to-right shunt.

The exclusion criteria were as follows: (I) concomitant mechanical complications of AMI, such as ventricular free wall rupture or chordal rupture; (II) an acute cerebrovascular accident; (III) a congenital ventricular septal defect.

Clinical data collection

Clinical data were extracted from the electronic medical records and categorized into five domains: (I) baseline characteristics and medical history; (II) clinical status at admission, including Killip classification (9), heart rate, blood pressure, and cardiogenic shock; (III) laboratory parameters at admission; (IV) morphological and angiographic characteristics; and (V) clinical events occurring during the index hospitalization, including documented in-hospital cardiac arrest.

For patients who underwent SurCOP, traditional surgical repair, or transcatheter closure, the interval from the documented diagnosis of VSR to the definitive intervention was recorded in hours. The hour-based variable was used for analysis and was converted to days by dividing by 24.

Renal insufficiency was defined as pre-existing chronic kidney disease documented in the medical history, with a baseline serum creatinine level >133 μmol/L or an estimated glomerular filtration rate <60 mL/min/1.73 m². This was a pre-existing comorbidity rather than acute kidney injury. Liver dysfunction was defined as pre-existing chronic liver disease with a documented history of chronic hepatitis, cirrhosis, or abnormal liver function tests with ALT or AST greater than 2 times the upper limit of normal prior to the index event. Chronic pulmonary disease was defined as pre-existing chronic obstructive pulmonary disease, asthma, or another chronic lung condition requiring ongoing medical therapy, as documented in the medical history. Cardiogenic shock was defined as systolic blood pressure <90 mmHg for more than 30 minutes or the need for vasopressors to maintain systolic blood pressure ≥90 mmHg, together with evidence of end-organ hypoperfusion, such as oliguria, cool extremities, altered mental status, or an elevated lactate level >2.0 mmol/L, in the setting of AMI (10). Cardiac arrest was defined as a documented in-hospital loss of consciousness with absence of a palpable pulse requiring cardiopulmonary resuscitation during the index hospitalization. The study database captured documented in-hospital cardiac arrest only; out-of-hospital cardiac arrest was not systematically recorded. The exact timing of in-hospital cardiac arrest relative to VSR diagnosis, treatment selection, and the definitive intervention was not available. Heart failure was defined as a clinical syndrome of pulmonary congestion, including dyspnea, pulmonary rales, or hypoxia, or systemic congestion, including peripheral edema or jugular venous distension, requiring medical therapy and documented in the medical record. Valvular regurgitation was defined as moderate-to-severe mitral or tricuspid regurgitation as assessed by echocardiography. Previous angina or myocardial infarction was defined as history of prior angina pectoris or documented MI by ECG changes and biomarker elevation occurring before the index AMI.

The SurCOP procedure

SurCOP, which stands for surgical repair combining an occluder and a patch, is a modified surgical technique for the repair of post-myocardial infarction VSR. The procedure was performed through a median sternotomy under general anesthesia. Cardiopulmonary bypass and moderate systemic hypothermia were used for myocardial protection. The rupture was approached through a longitudinal left ventriculotomy in the infarcted region, approximately 1–2 cm from the left anterior descending coronary artery. After debridement of the surrounding necrotic infarcted myocardium, while preserving the myocardium immediately adjacent to the VSR, an appropriately sized T-shaped patent ductus arteriosus occluder was introduced directly through the ventriculotomy. The cylindrical waist of the occluder was positioned within the rupture, with the aortic retention disc facing the left ventricular side. The aortic retention disc was secured to the surrounding septal myocardium with pledgeted mattress sutures placed at the 3, 6, 9, and 12 o’clock positions. A bovine pericardial patch, trimmed to be slightly larger than the occluder, was then placed over the aortic retention disc and continuously sutured to the adjacent myocardium. Finally, the ventriculotomy was closed with felt-strip-buttressed sutures and reinforced with surgical glue. The principal intraoperative steps of the procedure are illustrated in Figure 1 (6).

Figure 1.

Principal intraoperative steps of the SurCOP procedure. (A) Exposure of the VSR and debridement of the infarcted rim through a left ventriculotomy. (B) Deployment of a patent ductus arteriosus occluder within the septal defect. (C) Fixation of the aortic retention disc using pledgeted mattress sutures. (D) Placement of a bovine pericardial patch over the occluder. (E) Continuous suturing of the bovine pericardial patch to the adjacent myocardium. (F) Felt-strip-buttressed closure of the left ventriculotomy with surgical-glue reinforcement. SurCOP, surgical repair combining an occluder and a patch; VSR, ventricular septal rupture.

Statistical analysis

All statistical analyses were performed using R software (version 4.5.0; R Foundation for Statistical Computing, Vienna, Austria). Continuous variables were initially assessed for normality using the Shapiro-Wilk test. Normally distributed continuous variables were expressed as mean ± standard deviation (SD) and compared using the Student’s t-test. Non-normally distributed continuous variables were presented as medians with interquartile ranges (IQRs) and compared using the Mann-Whitney U test. Categorical variables were reported as frequencies and percentages (%), and intergroup comparisons were conducted using the Pearson Chi-squared test or Fisher’s exact test, as appropriate. A post hoc interaction between treatment modality and Killip classification was evaluated using a likelihood-ratio test comparing models with and without the interaction term.

Treatment-timing analyses were restricted to patients who underwent a definitive invasive intervention, because the interval from VSR diagnosis to intervention was structurally undefined for patients managed conservatively. The interval was calculated from the recorded hour-based variable, converted to days for reporting, and summarized as the median and IQR. The three invasive-treatment groups were compared using the Kruskal-Wallis test. When the overall test was significant, pairwise Mann-Whitney U tests with Holm adjustment for multiple comparisons were performed. For descriptive purposes, intervention timing was operationally categorized as emergency (<24 hours), urgent (24 hours to ≤7 days), or delayed (>7 days). The distribution of these categories across the three invasive-treatment groups was compared using the Fisher-Freeman-Halton exact test because of sparse cell counts. Treatment timing was not imputed or entered into the primary full-cohort prognostic model because it was defined only after receipt of a definitive intervention and was unavailable by design for patients managed conservatively.

The variables excluded from the primary candidate set included intra-aortic balloon pump use, continuous renal replacement therapy, mechanical ventilation, vasoactive-agent use, cardiogenic shock, and cardiac arrest. These variables were excluded to reduce reverse causality, pathway bias, and overlap with the retained measures of clinical severity. Cardiogenic shock was excluded from the primary candidate set because it overlapped substantially with Killip class IV as a measure of severe hemodynamic compromise. Cardiac arrest was excluded because only in-hospital cardiac arrest was recorded and its exact timing relative to VSR diagnosis, treatment selection, and definitive intervention was unavailable. It could therefore represent a downstream event rather than a baseline predictor. Intra-aortic balloon pump (IABP), continuous renal replacement therapy (CRRT), mechanical ventilation, and vasoactive-agent use were excluded because these measures were generally initiated after clinical deterioration and could introduce reverse-causality or pathway bias.

The candidate predictors evaluated in the primary model included treatment modality, age, sex, history of angina or myocardial infarction, Killip classification, admission creatinine, admission lactate, and VSR size. A multivariable logistic regression model was constructed using forward stepwise selection based on likelihood-ratio tests, with P<0.10 specified as the criterion for variable entry. No backward-removal step was applied, and Akaike Information Criterion (AIC) was not used to determine variable entry or retention. In the multivariable model, treatment modality and Killip classification were treated as categorical variables, with SurCOP and Killip class IV as the respective reference categories. Thirty-day mortality was coded as 1 for death and 0 for survival. The overall significance of the final model was assessed using the likelihood-ratio test, and the regression results were reported as odds ratios (ORs) with 95% confidence intervals (CIs). Model performance and diagnostics were evaluated across the following dimensions:

1. Multicollinearity: the variance inflation factor (VIF) and tolerance were calculated to check for multicollinearity among the independent variables, with a VIF <5 indicating the absence of significant collinearity.

2. Goodness of fit and explanatory power: overall model fit was assessed using the Hosmer-Lemeshow goodness-of-fit test; P>0.05 was interpreted as indicating no evidence of lack of fit. The explanatory power of the model was estimated using Cox-Snell R² and Nagelkerke R².

3. Predictive capacity: a classification table was used to determine the sensitivity, specificity, and overall accuracy of the model. Additionally, receiver operating characteristic (ROC) curve analysis was performed to calculate the area under the curve (AUC) of the combined model.

Treatment modality was treated as a nominal categorical variable with four levels: conservative medical therapy, transcatheter closure, traditional surgical repair, and SurCOP. It was entered into both the univariable and multivariable logistic regression models using dummy variables, with SurCOP as the reference category. In the univariable analysis, the overall association between treatment modality and 30-day mortality was assessed using a 3-degree-of-freedom Wald test. Because the treatment categories are unordered and have no clinically meaningful numerical threshold, no single-variable ROC curve or optimal cutoff value was calculated for treatment modality.

Missing-data handling and definition of the analysis cohort: the overall descriptive cohort comprised all 214 eligible patients. For the primary multivariable logistic regression analysis, complete-case restriction was applied before the forward stepwise variable-selection procedure. Thus, complete data were required for all candidate predictors entered into the selection process, including VSR size, even if a candidate variable was not retained in the final model. Among these candidate predictors, admission creatinine was missing in 3 patients (1.4%), admission lactate in 24 patients (11.2%), and VSR size in 2 patients (0.9%). Two patients had missing values for both admission creatinine and admission lactate, whereas the 2 patients with missing VSR size did not overlap with those with missing creatinine or lactate. Consequently, 27 unique patients (12.6%) were excluded from the primary complete-case analysis, leaving 187 patients (87.4%) for variable selection, final model estimation, and ROC analysis. Treatment modality, history of angina or myocardial infarction, and Killip classification were complete in all 214 patients. Although VSR size was not retained in the final multivariable model, its missingness affected inclusion because case-wise deletion was performed before variable selection.

Multiple-imputation sensitivity analysis: missing candidate-predictor data were addressed using multiple imputation by chained equations under a missing-at-random assumption. Twenty imputed datasets were generated using 10 iterations and a fixed master seed of 20260818. Predictive mean matching with a donor pool of five was used for admission creatinine, admission lactate, and VSR size.

Clinically informed sensitivity analysis: as a sensitivity analysis, a clinically informed model was fitted in which age and sex were forced into the fixed final predictor set irrespective of statistical significance. The remaining covariates were treatment modality, history of angina or myocardial infarction, Killip classification, admission creatinine, and admission lactate. The apparent AUC of this sensitivity model was compared descriptively with that of the primary stepwise model.

For estimation of the pooled regression coefficients and apparent model performance, the imputation model included 30-day mortality, treatment modality, history of angina or myocardial infarction, Killip classification, age, sex, admission creatinine, admission lactate, and VSR size. The fixed final logistic regression model, including treatment modality, history of angina or myocardial infarction, Killip classification, admission creatinine, and admission lactate, was fitted in each imputed dataset. Regression coefficients and standard errors were combined using Rubin’s rules, and apparent AUCs were pooled on the logit-AUC scale.

To avoid information leakage during internal validation, multiple imputation was nested within each outer fold of the stratified 10-fold cross-validation. For each fold, the imputation models were estimated using the training observations only, and the held-out observations were excluded from imputation-model fitting and from the predictive-mean-matching donor pool. The 30-day mortality outcome was excluded from the predictor matrix used for cross-validation imputation so that the held-out outcomes did not contribute to imputation of the validation data. Twenty imputed training-validation dataset pairs were generated within each fold. The fixed final logistic regression model was fitted separately in each imputed training dataset, and predicted probabilities for the corresponding held-out patients were averaged across the 20 imputations. Out-of-fold predictions from all 10 folds were then combined to estimate the cross-validated AUC and its 95% CI using the DeLong method. Variable selection was not repeated within the imputed datasets or cross-validation folds; therefore, this analysis evaluated the robustness and internal performance of the fixed final predictor set rather than the entire data-driven variable-selection procedure.

ROC analyses of the combined model and selected individual predictors were performed in the same 187-patient complete-case cohort. The combined-model ROC curve was generated from the predicted probabilities of the final multivariable logistic regression model. Killip classification, admission creatinine, and admission lactate were evaluated in the direction of increasing mortality risk. Because a history of angina or myocardial infarction was inversely associated with mortality, absence of such a history was specified as the positive test direction. Optimal operating points were identified by maximizing the Youden index, and 95% CIs for the AUCs were estimated using the DeLong method.

Internal validation of the fixed final model was performed using stratified 10-fold cross-validation. Model coefficients were re-estimated in each training fold, and predictions for the corresponding held-out fold were pooled to calculate the cross-validated AUC and its 95% CI. Harrell bootstrap optimism correction was additionally performed using 500 resamples. For each bootstrap resample, the model was refitted using the fixed final predictor set, and optimism was estimated as the difference between bootstrap-sample performance and performance when the bootstrap-fitted model was applied to the original cohort. Variable selection was not repeated within the cross-validation folds or bootstrap resamples. Accordingly, these analyses represent internal validation of the fixed final predictor set rather than validation of the entire data-driven variable-selection procedure. Fixed random seeds were used to ensure reproducibility.

Calendar-time sensitivity analyses: calendar time was derived from the year of hospital admission and categorized into three treatment eras—2012–2016, 2017–2021, and 2022–2025. To assess potential temporal confounding, the fixed final multivariable logistic regression model was refitted in the 187-patient complete-case cohort after treatment era was added as a categorical covariate. The model included treatment modality, history of angina or myocardial infarction, Killip classification, admission creatinine, admission lactate, and treatment era. Thirty-day mortality was coded as 1 and survival as 0; SurCOP, Killip class IV, and the 2012–2016 era were used as the respective reference categories. The overall contribution of treatment era was assessed using a 2-degree-of-freedom (df) likelihood-ratio test. As a complementary analysis addressing the absence of SurCOP-treated patients before 2017, the same fixed model was refitted after restricting the cohort to patients admitted during 2017–2025.

Results

Baseline characteristics

Of the 214 patients with VSR, 111 (51.9%) died within 30 days and 103 (48.1%) survived. Non-survivors were older than survivors (median age, 69 vs. 64 years) and included a lower proportion of male patients (43.2% vs. 65.0%). Renal insufficiency (49.5% vs. 13.6%), liver dysfunction (37.8% vs. 8.7%), and chronic pulmonary disease (16.2% vs. 3.9%) were more frequent among non-survivors. In contrast, a history of angina or myocardial infarction (9.0% vs. 30.1%) and previous percutaneous coronary intervention (PCI) (4.5% vs. 22.3%) were less frequent among non-survivors than among survivors. Of the 214 patients, 187 (87.4%) had complete data for all candidate predictors entered into the forward stepwise variable-selection procedure and were included in the primary complete-case analysis. Twenty-seven patients (12.6%) were excluded because of missing candidate-predictor data: 22 had missing admission lactate only, 2 had missing values for both admission lactate and creatinine, 1 had missing creatinine only, and 2 had missing VSR size only. The resulting 187-patient analysis cohort comprised 88 survivors and 99 non-survivors. The 27 excluded patients comprised 15 survivors and 12 non-survivors.

Markers of clinical severity were more common among non-survivors. At admission, non-survivors more frequently had Killip class IV (64.9% vs. 26.2%) and cardiogenic shock (68.5% vs. 18.4%) than survivors. During the index hospitalization, documented in-hospital cardiac arrest occurred in 28.8% of non-survivors and 2.9% of survivors. These severe clinical presentations correlated with worse initial laboratory profiles, characterized by notably higher baseline creatinine [103 (79.25, 178.75) vs. 84 (71.0, 101.0) μmol/L] and blood lactate levels [2.35 (1.575, 5.825) vs. 1.5 (1.100, 2.175) mmol/L]. Observed 30-day mortality differed across treatment groups. Mortality was 90.1% among patients managed conservatively, compared with 21.8% after SurCOP, 25.0% after traditional surgical repair, and 25.7% after transcatheter closure. Because treatment allocation was non-random and influenced by clinical status and treatment timing, these unadjusted differences should not be interpreted as causal treatment effects. The comprehensive baseline characteristics and intergroup comparisons are summarized in Table 1.

Table 1. Baseline characteristics according to 30-day survival status.

Characteristics	Survivors (n=103)	Non-survivors (n=111)	P
Treatment modalities			<0.001***
Conservative medical therapy	9 (8.7)	82 (73.9)
Transcatheter closure	26 (25.2)	9 (8.1)
Traditional surgical repair	18 (17.5)	6 (5.4)
SurCOP	50 (48.5)	14 (12.6)
Male	67 (65.0)	48 (43.2)	0.001**
Age (years)	64 (58.0–71.0)	69 (64.0–75.0)	<0.001***
SBP (mmHg)	110.39±16.268	106.72±18.04	0.12
DBP (mmHg)	70 (63.0–80.0)	69 (62.0–76.0)	0.12
Heart rate (bpm)	88 (78.5–104.0)	96 (80.0–108.0)	0.13
History of smoking	37 (35.9)	25 (22.5)	0.03*
History of alcohol consumption	23 (22.3)	20 (18.0)	0.43
Hypertension	55 (53.4)	64 (57.7)	0.53
Diabetes mellitus	32 (31.1)	37 (33.3)	0.72
Previous cerebrovascular event	9 (8.7)	18 (16.2)	0.10
Chronic lung disease	4 (3.9)	18 (16.2)	0.003**
Renal insufficiency	14 (13.6)	55 (49.5)	<0.001***
Liver dysfunction	9 (8.7)	42 (37.8)	<0.001***
History of angina or myocardial infarction	31 (30.1)	10 (9.0)	<0.001***
History of thrombolytic therapy	3 (2.9)	1 (0.9)	0.56
History of arrhythmia	10 (9.7)	7 (6.3)	0.36
History of valvular disease	2 (1.9)	1 (0.9)	0.95
Previous PCI	23 (22.3)	5 (4.5)	<0.001***
Previous CABG	5 (4.9)	0 (0.0)	0.058
Killip class			<0.001***
I	8 (7.8)	4 (3.6)
II	35 (34.0)	12 (10.8)
III	33 (32.0)	23 (20.7)
IV	27 (26.2)	72 (64.9)
PCI	52 (50.5)	38 (34.2)	0.02*
IABP	23 (22.3)	54 (48.6)	<0.001***
ECMO	5 (4.9)	9 (8.1)	0.34
CRRT	5 (4.9)	27 (24.3)	<0.001***
Mechanical ventilation	5 (4.9)	30 (27.0)	<0.001***
Vasoactive agents	79 (77.5)	109 (98.2)	<0.001***
Cardiogenic shock	19 (18.4)	76 (68.5)	<0.001***
In-hospital cardiac arrest during the index hospitalization	3 (2.9)	32 (28.8)	<0.001***
Time from myocardial infarction to VSR diagnosis (h)	96 (24.0–267.5)	96 (30.5–185)	0.45
Admission creatinine (μmol/L)	84 (71.0–101)	103 (79.25–178.75)	<0.001***
Admission ALT (U/L)	44 (19.0–147.12)	57 (33–275)	0.01*
Admission AST (U/L)	38 (21.0–119.0)	140.5 (40.0–375.25)	<0.001***
Admission lactate (mmol/L)	1.5 (1.100–2.175)	2.35 (1.575–5.825)	<0.001***
Valvular regurgitation	50 (48.5)	43 (39.1)	0.17
LVEF (%)	49 (42.5–58.5)	49 (42.0–55.0)	0.42
Location of rupture			0.15
Apical	69 (67.0)	85 (76.6)
Anterior	6 (5.8)	8 (7.2)
Posterior	28 (27.2)	18 (16.2)
VSR size (mm)	12 (8.0–19.5)	9 (6.0–13.75)	0.001**
Number of culprit vessels			0.44
1	60 (60.0)	31 (50.0)
2	23 (23.0)	19 (30.6)
3	17 (17.0)	12 (19.4)
Heart failure	53 (51.5)	88 (79.3)	<0.001***
Pleural effusion	71 (68.9)	76 (68.5)	0.94

Data are presented as n (%), mean ± standard deviation or median (interquartile range). *, P<0.05; **, P<0.01; ***, P<0.001. The descriptive analysis included all 214 patients and was based on the available observations for each variable. Admission creatinine was missing in 3 patients, admission lactate in 24 patients, and VSR size in 2 patients. Because missingness overlapped, 27 unique patients were excluded from the primary complete-case model. Other variables with missing observations were vasoactive-agent use (n=1), time from myocardial infarction to VSR diagnosis (n=1), admission AST (n=3), admission ALT (n=4), valvular regurgitation (n=1), and LVEF (n=3). The number of culprit vessels was available for 162 patients, including 100 survivors and 62 non-survivors; percentages for this variable were calculated using the corresponding available-case denominators. ALT, alanine aminotransferase; AST, aspartate aminotransferase; CABG, coronary artery bypass grafting; CRRT, continuous renal replacement therapy; DBP, diastolic blood pressure; ECMO, extracorporeal membrane oxygenation; IABP, intra-aortic balloon pump; LVEF, left ventricular ejection fraction; PCI, percutaneous coronary intervention; SBP, systolic blood pressure; SurCOP, surgical repair combining an occluder and a patch; VSR, ventricular septal rupture.

SurCOP was introduced at our institution in 2017 and its use increased over time. Of the 64 SurCOP procedures, 50 (78.1%) were performed after 2020. Treatment allocation differed substantially across treatment eras. In 2012–2016, 29 of 38 patients (76.3%) received conservative medical therapy, 7 (18.4%) underwent transcatheter closure, 2 (5.3%) underwent traditional surgical repair, and none underwent SurCOP. In 2017–2021, the corresponding numbers were 37 of 95 (38.9%), 24 (25.3%), 13 (13.7%), and 21 (22.1%), respectively. In 2022–2025, 25 of 81 patients (30.9%) received conservative medical therapy, 4 (4.9%) underwent transcatheter closure, 9 (11.1%) underwent traditional surgical repair, and 43 (53.1%) underwent SurCOP. The unadjusted 30-day mortality rates were 76.3% in 2012–2016, 52.6% in 2017–2021, and 39.5% in 2022–2025.

The interval from VSR diagnosis to definitive intervention was available for all 123 patients who underwent SurCOP, traditional surgical repair, or transcatheter closure. The median intervals were 6.5 days (IQR, 4.0–12.5 days) in the SurCOP group, 31.4 days (IQR, 18.7–43.2 days) in the traditional surgical-repair group, and 17.8 days (IQR, 10.6–23.0 days) in the transcatheter-closure group (Kruskal-Wallis P<0.001). All pairwise comparisons remained significant after Holm adjustment (SurCOP vs. traditional surgical repair, P<0.001; SurCOP vs. transcatheter closure, P<0.001; traditional surgical repair vs. transcatheter closure, P=0.03). The distribution of the operational timing categories also differed among the three groups (Fisher-Freeman-Halton exact P<0.001). Emergency, urgent, and delayed intervention occurred in 3 (4.7%), 32 (50.0%), and 29 (45.3%) patients in the SurCOP group; 1 (4.2%), 2 (8.3%), and 21 (87.5%) patients in the traditional surgical-repair group; and 0, 4 (11.4%), and 31 (88.6%) patients in the transcatheter-closure group, respectively (Table 2).

Table 2. Timing from VSR diagnosis to definitive intervention among patients who received invasive treatment.

Characteristic	SurCOP (n=64)	Traditional surgical repair (n=24)	Transcatheter closure (n=35)	P
Interval from VSR diagnosis to definitive intervention, days	6.5 (4.0–12.5)	31.4 (18.7–43.2)	17.8 (10.6–23.0)	<0.001
Timing category				<0.001
Emergency (<24 hours)	3 (4.7)	1 (4.2)	0 (0.0)
Urgent (24 hours to ≤7 days)	32 (50.0)	2 (8.3)	4 (11.4)
Delayed (>7 days)	29 (45.3)	21 (87.5)	31 (88.6)

Values are presented as median (interquartile range) or n (%). The hour-based interval was used for analysis and was converted to days by dividing by 24. The P value for the continuous interval was obtained using the Kruskal-Wallis test. Pairwise Mann-Whitney U tests with Holm adjustment yielded P<0.001 for SurCOP versus traditional surgical repair, P<0.001 for SurCOP versus transcatheter closure, and P=0.03 for traditional surgical repair versus transcatheter closure. The overall timing-category distribution was compared using the Fisher-Freeman-Halton exact test. Emergency, urgent, and delayed intervention were operational categories used for descriptive reporting and should not be interpreted as validated clinical urgency standards. SurCOP, surgical repair combining an occluder and a patch; VSR, ventricular septal rupture.

Whether a definitive invasive intervention had been planned before death could not be reliably determined from the retrospective database.

Univariable analysis

Univariable logistic regression analyses were performed to evaluate the associations of demographic characteristics, medical history, admission status, laboratory findings, and treatment strategy with 30-day mortality. The results are summarized in Table 3. Treatment modality was associated with 30-day mortality overall (Wald χ²=68.132, df =3, P<0.001). With SurCOP as the reference category, conservative medical therapy was associated with higher odds of 30-day mortality (OR, 32.540; 95% CI: 13.121–80.698; P<0.001). Neither transcatheter closure (OR, 1.236; 95% CI: 0.472–3.236; P=0.67) nor traditional surgical repair (OR, 1.190; 95% CI: 0.397–3.568; P=0.76) differed significantly from SurCOP.

Table 3. Univariable logistic regression analysis of factors associated with 30-day mortality.

Variables	β	SE	Wald χ2	P	OR (95% CI)
Treatment modality, overall	–	–	68.132 (df =3)	<0.001	–
SurCOP	–	–	–	–	1.000 (Reference)
Conservative medical therapy	3.482	0.463	56.477	<0.001	32.540 (13.121–80.698)
Transcatheter closure	0.212	0.491	0.187	0.67	1.236 (0.472–3.236)
Traditional surgical repair	0.174	0.560	0.097	0.76	1.190 (0.397–3.568)
Sex	0.893	0.282	10.045	0.002	2.443 (1.406–4.244)
Age	0.051	0.016	10.251	0.001	1.052 (1.020–1.086)
History of smoking	−0.657	0.306	4.598	0.03	0.519 (0.284–0.945)
Chronic pulmonary disease	1.567	0.571	7.519	0.006	4.790 (1.563–14.678)
Renal insufficiency	1.832	0.345	28.260	<0.001	6.244 (3.178–12.266)
Liver dysfunction	1.850	0.400	21.375	<0.001	6.357 (2.902–13.926)
History of angina or myocardial infarction	−1.470	0.395	13.845	<0.001	0.230 (0.106–0.499)
Previous PCI	−1.807	0.515	12.309	<0.001	0.164 (0.060–0.450)
Killip class	0.874	0.169	26.565	<0.001	2.395 (1.718–3.339)
PCI	−0.672	0.281	5.732	0.02	0.511 (0.294–0.885)
IABP	1.192	0.303	15.449	<0.001	3.295 (1.818–5.972)
CRRT	1.841	0.509	13.072	<0.001	6.300 (2.323–17.087)
Mechanical ventilation	1.982	0.506	15.356	<0.001	7.259 (2.693–19.565)
Vasoactive agents	2.764	0.752	13.517	<0.001	15.867 (3.635–69.261)
Cardiogenic shock	2.262	0.326	48.139	<0.001	9.600 (5.067–18.187)
In-hospital cardiac arrest	2.603	0.622	17.495	<0.001	13.502 (3.988–45.719)
Admission creatinine	0.011	0.003	15.274	<0.001	1.011 (1.006–1.017)
Admission AST	0.000	0.000	4.133	0.04	1.000 (1.000–1.001)
Admission ALT	0.000	0.000	3.323	0.07	1.000 (1.000–1.001)
Admission lactate	0.398	0.096	17.118	<0.001	1.489 (1.233–1.799)
VSR size	−0.069	0.023	8.681	0.003	0.934 (0.892–0.977)
Heart failure	1.284	0.306	17.582	<0.001	3.610 (1.981–6.577)

Treatment modality was entered as a nominal categorical variable using dummy coding, with SurCOP as the reference category. The overall P value for treatment modality was obtained using a 3-df Wald test. ALT, alanine aminotransferase; AST, aspartate aminotransferase; CI, confidence interval; CRRT, continuous renal replacement therapy; df, degree of freedom; IABP, intra-aortic balloon pump; OR, odds ratio; PCI, percutaneous coronary intervention; SE, standard error; SurCOP, surgical repair combining an occluder and a patch; VSR, ventricular septal rupture.

Other variables associated with 30-day mortality in the univariable analyses included sex, age, smoking history, chronic pulmonary disease, renal insufficiency, liver dysfunction, a history of angina or myocardial infarction, previous PCI, Killip classification, acute PCI, IABP use, CRRT, mechanical ventilation, vasoactive-agent use, cardiogenic shock, cardiac arrest, admission creatinine, admission lactate, VSR size, and heart failure. Because several interventions and acute clinical events were strongly related to illness severity and could introduce reverse causality or pathway bias, they were not considered eligible predictors for the primary multivariable model.

Multivariable logistic regression analysis and model evaluation

A restricted candidate set was evaluated using forward stepwise likelihood-ratio selection with P<0.10 as the criterion for entry. The final model retained treatment modality, history of angina or myocardial infarction, Killip classification, admission creatinine, and admission lactate. Age, sex, and VSR size did not meet the entry criterion and were not retained in the final stepwise model. The overall fit, explanatory power, and discriminative capacity of the model were evaluated to validate its performance. To achieve this, the analysis relied on the likelihood ratio test, pseudo-R² values, the Hosmer-Lemeshow test, and classification accuracy. A forest plot was also generated to visually represent the effect directions and corresponding interval estimates. Post hoc interaction testing between treatment modality and Killip classification did not show evidence of effect modification (P=0.41).

As outlined in Table 4, the inclusion of the selected predictors substantially improved the model’s fit over the baseline intercept-only model. The -2 log-likelihood decreased to 141.328 (likelihood ratio χ²=117.262, df =9, P<0.001). The model also exhibited strong explanatory capacity, yielding a Cox-Snell R² of 0.466 and a Nagelkerke R² of 0.622. The Hosmer-Lemeshow goodness-of-fit test did not indicate evidence of lack of fit (P=0.47).

Table 4. Model fit statistics for the intercept-only and final multivariable logistic regression models.

Model	−2 Log-likelihood	Chi-squared	df	P	AIC	BIC
Intercept-only model	258.590	–	–	–	–	–
Final model	141.328	117.262	9	<0.001	161.328	193.620

AIC, Akaike Information Criterion; BIC, Bayesian Information Criterion; df, degree of freedom.

When treatment era was added as a categorical covariate to the fixed final model in the 187-patient complete-case cohort, conservative medical therapy remained associated with higher odds of 30-day mortality than SurCOP (OR, 18.127; 95% CI: 5.581–58.871; P<0.001). Neither traditional surgical repair (OR, 1.112; 95% CI: 0.277–4.459; P=0.88) nor transcatheter closure (OR, 1.736; 95% CI: 0.452–6.671; P=0.42) differed significantly from SurCOP. The overall contribution of treatment era did not reach statistical significance (likelihood-ratio χ²=5.160, df =2, P=0.08). In the complementary analysis restricted to patients admitted during 2017–2025 (n=160), conservative medical therapy remained associated with higher mortality than SurCOP (OR, 21.855; 95% CI: 6.723–71.052; P<0.001), whereas traditional surgical repair (OR, 1.517; 95% CI: 0.401–5.737; P=0.54) and transcatheter closure (OR, 2.445; 95% CI: 0.645–9.268; P=0.19) were not significantly different from SurCOP. These analyses reduced but did not eliminate potential temporal, learning-curve, and treatment-selection confounding.

Table 5 and Figure 2 summarize the final multivariable regression results. Treatment modality was associated with 30-day mortality overall (Wald χ²=35.281, df =3, P<0.001). Compared with SurCOP, conservative medical therapy was associated with substantially higher odds of 30-day mortality (OR, 27.822; 95% CI: 8.878–87.191; P<0.001). Neither traditional surgical repair (OR, 1.504; 95% CI: 0.397–5.700; P=0.55) nor transcatheter closure (OR, 2.990; 95% CI: 0.864–10.347; P=0.08) differed significantly from SurCOP.

Table 5. Multivariable binary logistic regression analysis of factors associated with 30-day mortality.

Variables	β	SE	Wald χ2	df	P	OR (95% CI)
Treatment modality	–	–	35.281	3	<0.001	–
Conservative medical therapy vs. SurCOP	3.326	0.583	32.566	1	<0.001	27.822 (8.878–87.191)
Traditional surgical repair vs. SurCOP	0.408	0.680	0.361	1	0.55	1.504 (0.397–5.700)
Transcatheter closure vs. SurCOP	1.095	0.633	2.990	1	0.08	2.990 (0.864–10.347)
History of angina or myocardial infarction	−1.448	0.600	5.834	1	0.02	0.235 (0.073–0.761)
Killip class	–	–	6.812	3	0.08	–
I vs. IV	−1.217	0.976	1.554	1	0.21	0.296 (0.044–2.007)
II vs. IV	−1.439	0.583	6.103	1	0.01	0.237 (0.076–0.743)
III vs. IV	−0.410	0.513	0.640	1	0.42	0.664 (0.243–1.812)
Admission creatinine	0.010	0.004	5.172	1	0.02	1.010 (1.001–1.019)
Admission lactate	0.290	0.133	4.799	1	0.03	1.337 (1.031–1.733)
Constant	−2.333	0.733	10.123	1	0.001	0.097 (–)

Cox-Snell R²=0.466; Nagelkerke R²=0.622; Hosmer-Lemeshow goodness-of-fit test, P=0.47. CI, confidence interval; df, degree of freedom; OR, odds ratio; SE, standard error; SurCOP, surgical repair combining an occluder and a patch.

Figure 2.

Forest plot of adjusted odds ratios for predictors of 30-day mortality in the multivariable logistic regression model. CI, confidence interval; OR, odds ratio; SurCOP, surgical repair combining an occluder and a patch.

A history of angina or myocardial infarction was independently associated with lower odds of 30-day mortality (OR, 0.235; 95% CI: 0.073–0.761; P=0.02). Higher admission creatinine (OR per μmol/L increase, 1.010; 95% CI: 1.001–1.019; P=0.02) and lactate (OR per mmol/L increase, 1.337; 95% CI: 1.031–1.733; P=0.03) were independently associated with higher mortality. Killip classification met the likelihood-ratio entry criterion (P=0.07), although its overall Wald test in the final model did not reach the conventional P<0.05 threshold (P=0.08). In the category-specific comparisons, Killip class II was associated with lower odds of mortality than Killip class IV (OR, 0.237; 95% CI: 0.076–0.743; P=0.01).

The model correctly classified 84.09% of survivors and 78.79% of non-survivors, with an overall accuracy of 81.28% (Table 6).

Table 6. Classification performance of the multivariable logistic regression model.

Observed outcome	Predicted		Predictive accuracy (%)	Error rate (%)
	0	1
Observed
0	74	14	84.09	15.91
1	21	78	78.79	21.21
Overall	–	–	81.28	18.72

Thirty-day outcome was coded as 0 for survival and 1 for death.

In the multiple-imputation sensitivity analysis using 20 imputed datasets, the pooled effect estimates were directionally consistent with those of the complete-case analysis. The pooled ORs were 32.078 (95% CI: 10.663–96.498) for conservative medical therapy versus SurCOP, 0.255 (95% CI: 0.084–0.776) for a history of angina or myocardial infarction, 1.010 (95% CI: 1.002–1.018) per μmol/L increase in admission creatinine, and 1.329 (95% CI: 1.025–1.722) per mmol/L increase in admission lactate. The pooled apparent AUC was 0.915 (95% CI: 0.868–0.946). When multiple imputation was nested within each outer training fold to prevent information leakage, stratified 10-fold cross-validation based on averaged out-of-fold predictions across the 20 imputations yielded an AUC of 0.891 (95% CI: 0.846–0.935).

In the clinically informed sensitivity model with age and sex forced into the model, neither variable achieved statistical significance. The OR was 1.023 for age (95% CI: 0.991–1.056; P=0.16) and 1.412 for sex (95% CI: 0.693–2.876; P=0.34). The effect estimates for the remaining predictors were directionally consistent with those of the primary model, and the clinically informed sensitivity model showed broadly comparable apparent discrimination, with an AUC of 0.891 versus 0.906 for the primary model.

ROC analysis

The standalone discriminative performance of selected clinical predictors was evaluated using ROC curve analysis. ROC curves were constructed for Killip classification, history of angina pectoris or myocardial infarction, admission creatinine, and admission lactate. Treatment modality was excluded from the single-variable ROC analysis because it is a nominal categorical variable without a clinically meaningful ordering or cutoff value. Table 7 summarizes the AUC, sensitivity, specificity, and corresponding 95% CI for each predictor.

Table 7. ROC analysis of the combined model and selected individual predictors of 30-day mortality in the 187-patient complete-case cohort.

Model or predictor	AUC	Sensitivity	Specificity	P	95% CI
Combined model	0.906	0.768	0.943	<0.001	0.863–0.948
Killip classification	0.715	0.657	0.727	<0.001	0.645–0.784
Absence of a history of angina or MI	0.614	0.909	0.318	<0.001	0.557–0.670
Admission creatinine	0.644	0.586	0.670	<0.001	0.565–0.723
Admission lactate	0.695	0.333	0.966	<0.001	0.620–0.770

All ROC analyses were performed in the same 187-patient complete-case cohort. Optimal operating points were selected by maximizing the Youden index. Because a history of angina or MI was inversely associated with mortality, absence of such a history was specified as the positive test direction. AUC, area under the curve; CI, confidence interval; MI, myocardial infarction; ROC, receiver operating characteristic.

In the 187-patient complete-case cohort, Killip classification showed an AUC of 0.715 (95% CI: 0.645–0.784), and admission lactate showed an AUC of 0.695 (95% CI: 0.620–0.770). Admission creatinine and absence of a history of angina or myocardial infarction showed lower standalone discrimination, with AUCs of 0.644 (95% CI: 0.565–0.723) and 0.614 (95% CI: 0.557–0.670), respectively. At the operating points that maximized the Youden index, Killip class IV had a sensitivity of 0.657 and a specificity of 0.727; admission creatinine at 95 μmol/L had a sensitivity of 0.586 and a specificity of 0.670; and admission lactate at 4.4 mmol/L had a sensitivity of 0.333 and a specificity of 0.966. No individual predictor demonstrated sufficient discrimination to be used as a standalone prognostic tool.

The combined model demonstrated an apparent AUC of 0.906 (95% CI: 0.863–0.948). At the predicted-probability threshold of 0.639 that maximized the Youden index, the model had a sensitivity of 0.768 and a specificity of 0.943. Stratified 10-fold cross-validation yielded an AUC of 0.875 (95% CI: 0.825–0.925). Harrell bootstrap optimism correction based on 500 resamples estimated a mean optimism of 0.022, resulting in an optimism-corrected AUC of 0.884. The corresponding ROC curves are shown in Figure 3. The combined model had a higher apparent AUC than each individual predictor; however, no formal pairwise comparison of AUCs was performed.

Figure 3.

ROC curves for the combined model and selected individual predictors of 30-day mortality in the 187-patient complete-case cohort. All curves were generated from the same analysis cohort. Because a history of angina or MI was inversely associated with mortality, absence of such a history was specified as the positive test direction. Treatment modality was not evaluated as a standalone ROC predictor because it is a nominal categorical variable. AUC, area under the curve; MI, myocardial infarction; ROC, receiver operating characteristic.

Discussion

This retrospective analysis of a cohort of 214 patients investigated the demographic, clinical, and therapeutic factors influencing 30-day survival outcomes following post-myocardial infarction VSR. These data were subsequently used to develop and internally validate a multivariable model for 30-day mortality. The final model included treatment modality, Killip classification, admission lactate, admission creatinine, and a history of angina or myocardial infarction. Treatment modality, admission creatinine, admission lactate, and a history of angina or myocardial infarction showed statistically significant associations with 30-day mortality, whereas the overall association of Killip classification did not reach statistical significance. By integrating these variables, we developed an internally validated prognostic model with an apparent AUC of 0.906. Internal validation yielded a cross-validated AUC of 0.875 and a bootstrap optimism-corrected AUC of 0.884, indicating some optimism in the apparent performance estimate. However, because treatment modality was included as a predictor, the model should be interpreted as a retrospective prognostic model incorporating the treatment ultimately received rather than an admission-based tool for treatment selection or personalized clinical decision-making.

The management of post-myocardial infarction VSR remains challenging, as reflected by the overall 30-day mortality of 51.9% in this cohort. Observed mortality was highest among patients managed conservatively (90.1%) and lower among patients who underwent an invasive intervention. In the multivariable analysis, conservative medical therapy was associated with substantially higher odds of death than SurCOP. However, treatment allocation was not randomized and was likely influenced by baseline severity, treatment timing, anatomical suitability, and survival to intervention. Moreover, mortality did not differ significantly between SurCOP and traditional surgical repair, and the comparison between SurCOP and transcatheter closure did not reach statistical significance. The findings therefore support an association between treatment strategy and outcome but do not establish the superiority of SurCOP over other invasive approaches. Consequently, given the high mortality associated with post-myocardial infarction VSR, timely multidisciplinary assessment for definitive intervention remains important (2,4). The findings were broadly consistent with previous large registries. Recent observational studies have likewise reported substantial short-term mortality and marked treatment-selection effects across conservative, surgical, and transcatheter strategies (11,12). A large statewide surgical registry also noted that the apparent association between delayed repair and improved survival was probably influenced by survival bias (13). The Society of Thoracic Surgeons National Database reported an operative mortality of 42.9% after surgical repair of post-myocardial infarction VSR (4), whereas the CAUTION study reported a mortality of 38.7% (3). The 30-day mortality of 25.0% in the traditional surgical-repair group in the present cohort was numerically lower, although direct comparisons should be interpreted cautiously because of differences in patient selection, treatment timing, operative technique, and study period. The overall 30-day mortality of 51.9% nevertheless remained consistent with the historically poor prognosis of this condition. The demographic profile of the present cohort was also broadly comparable with those of the Society of Thoracic Surgeons (STS) and CAUTION cohorts.

Treatment timing differed substantially among the three invasive-treatment groups. The median interval from VSR diagnosis to intervention was shortest in the SurCOP group, whereas traditional surgical repair and transcatheter closure were generally performed later. This imbalance may reflect differences in treatment availability, patient selection, anatomical suitability, clinical stability, and institutional practice over time. Because treatment timing was observed only among patients who survived to receive a definitive intervention, these findings cannot establish whether earlier intervention improved survival. Nevertheless, the marked timing imbalance indicates that treatment timing and survival to intervention remain important sources of residual confounding when outcomes are compared across treatment groups. These concerns are consistent with recent observational evidence showing that associations between delayed intervention and survival are vulnerable to survival-selection bias (7,13).

Admission lactate and creatinine were independently associated with 30-day mortality, consistent with their roles as markers of systemic hypoperfusion and end-organ dysfunction (14,15). Killip classification met the likelihood-ratio entry criterion, although its overall Wald test in the final model did not reach the conventional P<0.05 threshold, suggesting that the category-specific estimates should be interpreted cautiously. Earlier registry data characterized the clinical and angiographic patterns and adverse outcomes associated with post-infarction VSR (16). In the present study, the inverse univariable association between measured VSR size and mortality was counterintuitive and did not persist in the final multivariable model. Possible explanations include survivor and treatment-selection bias, incomplete or less reliable echocardiographic measurements in critically ill patients, and unmeasured differences in defect morphology, complexity, or the extent of myocardial necrosis. A smaller measured defect may therefore not represent less severe disease, and the present findings should not be interpreted as indicating a protective effect of larger rupture size.

A history of angina or myocardial infarction was independently associated with lower 30-day mortality in the multivariable model. Ischemic preconditioning is a possible explanation for this association (17,18); however, neither pre-infarction ischemic burden nor collateral coronary circulation was directly measured in the present study. Differences in prior medication use, previous revascularization, healthcare contact, and other unmeasured baseline characteristics may also have contributed. Therefore, this association should not be interpreted as evidence of a causal protective effect and requires confirmation in future studies.

ROC analyses of the individual predictors in the 187-patient complete-case cohort showed only modest standalone discrimination, with AUCs ranging from 0.614 to 0.715. The combined model showed higher apparent discrimination, with an AUC of 0.906, a sensitivity of 0.768, and a specificity of 0.943. Internal validation yielded a stratified 10-fold cross-validated AUC of 0.875 and a bootstrap optimism-corrected AUC of 0.884. The reduction from the apparent AUC to the internally validated estimates indicated some optimism in the apparent performance. In the full 214-patient cohort, fold-wise multiple imputation yielded a similar cross-validated AUC of 0.891 (95% CI: 0.846–0.935), supporting the robustness of model discrimination to the handling of missing predictor data. Because the treatment ultimately received was included in the model, these results reflect retrospective prognostic assessment and should not be interpreted as evidence that the model can predict risk or guide treatment selection at admission.

Several limitations of this study warrant acknowledgment. First, the retrospective, single-center design was susceptible to selection bias, residual confounding, and treatment-selection bias. Treatment allocation was non-random and may have been influenced by baseline severity, anatomical suitability, treatment timing, and survival to intervention.

SurCOP was unavailable before 2017 and became increasingly common in later treatment eras. Although treatment-era adjustment and a contemporary-era sensitivity analysis were performed, residual confounding due to secular improvements in reperfusion therapy, perioperative care, mechanical circulatory support, and institutional experience cannot be excluded. Although the interval from VSR diagnosis to definitive intervention was available for patients who underwent SurCOP, traditional surgical repair, or transcatheter closure, it was structurally undefined for patients managed conservatively and could be observed only among patients who survived long enough to receive an intervention. Consequently, treatment timing was not incorporated into the full-cohort prognostic model, and residual survivor-treatment bias and immortal-time bias could not be eliminated.

Second, follow-up was limited to 30 days, precluding assessment of long-term survival, ventricular remodeling, and structural durability. Third, the sample size was limited relative to the number of potential candidate predictors, and interaction analyses could not be performed comprehensively. Although stratified cross-validation and bootstrap optimism correction provided internal validation, these analyses cannot establish generalizability to external populations. The reduction from the apparent AUC of 0.906 to the internally validated estimates of 0.875–0.884 indicates some optimism in the apparent performance estimate.

Out-of-hospital cardiac arrest was not systematically recorded, and the exact timing of the 35 documented in-hospital cardiac arrests relative to VSR diagnosis, treatment selection, and definitive intervention was unavailable. Therefore, sensitivity analyses excluding out-of-hospital cardiac arrest or cardiac arrest occurring before treatment selection could not be performed. Exact times of death and whether an invasive intervention had been planned before death were also not reliably captured. Consequently, a valid very-early-death sensitivity analysis could not be conducted, and survivor-treatment bias related to the requirement to survive long enough to undergo an invasive intervention cannot be excluded.

Multiple imputation relied on the missing-at-random assumption, which cannot be verified from the observed data, and residual bias due to data that were missing not at random cannot be excluded. Multiple imputation was nested within the outer cross-validation folds so that held-out observations and outcomes did not contribute to imputation-model estimation or donor selection. However, variable selection was not repeated within the imputed datasets or cross-validation folds. Therefore, the imputation-based cross-validation assessed the internal performance of the fixed final predictor set rather than the entire model-development procedure, and some residual optimism may remain. In addition, pre-admission medication use was not systematically available and may have confounded the observed association between a history of angina or myocardial infarction and lower mortality. Finally, because the treatment ultimately received was included as a predictor, the model cannot be used as an admission-based risk-prediction or treatment-selection tool. It should be interpreted only as a retrospective prognostic model. External validation in an independent multicenter cohort is required before clinical application, and future studies should develop a separate baseline-only model using variables available before treatment selection.

Conclusions

In this retrospective single-center cohort of 214 patients with post-myocardial infarction VSR, treatment modality, admission creatinine, admission lactate, and a history of angina or myocardial infarction were independently associated with 30-day mortality. The multivariable model showed good apparent discrimination, with an AUC of 0.906, while stratified 10-fold cross-validation and bootstrap optimism correction yielded AUCs of 0.875 and 0.884, respectively. Because treatment allocation was non-random and the model incorporated the treatment ultimately received, the findings should be interpreted as retrospective prognostic associations rather than causal treatment effects or evidence supporting admission-based treatment selection. External validation in an independent multicenter cohort is required before clinical application.

Supplementary

The article’s supplementary files as

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Institutional Ethics Committee of The First Affiliated Hospital of Zhengzhou University (Approval No. 2026-KY-0033-002). Because of the retrospective design, the requirement for individual informed consent was waived by the committee.

Data Sharing Statement

Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1814/dss

jtd-18-9-1017-dss.pdf ^{(3.5KB, pdf)}

DOI: 10.21037/jtd-2026-1814