We included all adult patients who underwent orthotopic heart transplantation at our center between 1 January 2015 and 31 December 2018. Patients were excluded for: (1) Re-transplantation or multi-organ transplantation; (2) Immediate death within the first postoperative day; (3) Extreme body weight (<40 kg or >130 kg); (4) Lack of sufficient data on vasoactive-inotropic agents. After exclusion, 325 patients qualified for further analyses (Supplementary Figure S1). The donor hearts were all procured from voluntary donations after brain death and allocated using the China Organ Transplant Response System. The organs of executed prisoners were not used. Our research work conformed to the Declarations of Helsinki and Istanbul, and was approved by the Institutional Review Boards of Tongji Medical College. The requirement for patient consent was waived because the study’s nature was retrospective.

We acquired patient data from the electronic medical record system. Among them, advanced heart failure treatment was defined as the preoperative administration of levosimendan or a recombinant human brain natriuretic peptide. The VIS was calculated using the formula modified from the inotrope score formula in the PGD consensus definition [16]: VIS = dopamine + dobutamine + 15 × milrinone + 100 × epinephrine + 100 × norepinephrine. Each item denotes the quotient of the drug dose (μg/min) divided by body weight (kg). Within the first 24 postoperative hours, the VIS at each hour was calculated and the maximum VIS (VIS_max) [12] was obtained. Survival information was obtained through follow-up with the recipients and consultation with the related responsible doctors.

The primary outcome was the severe adverse outcome, a composite of early outcomes including early death, neurological complications, renal injury, septic shock and cardiac reoperation, which are commonly studied in cardiac patients [5, 12, 13]. The development of at least one of the above early outcomes was defined as the severe adverse outcome. Secondary outcomes were 90-day, 1-year and 6-year survival.

Early death was defined as in-hospital death or out-of-hospital death within 30 days of discharge [13]. Other complications all occurred in the hospital. Neurological complications were defined as the combination of stroke, as demonstrated by new cerebral deficits on radiological imaging, and seizure episodes requiring intervention. Renal injury was defined as newly initiated continuous renal replacement therapy (CRRT). Septic shock was defined as hypotension or hypoperfusion status with an infectious etiology. Cardiac reoperation was defined as a second thoracotomy after the initial transplantation.

Descriptive data were presented as “median (interquartile range)” or “mean (standard deviation)” for continuous variables, and as “number (percentage)” for categorical variables. Comparisons were performed by t-test or Mann-Whitney U-test for continuous data, and by Pearson χ² test, continuity-adjusted χ² test or Fisher’s exact probability test for categorical data. Survival curves were generated by the Kaplan-Meier method and their differences were examined using the Log rank test. Landmark analysis was undertaken for crossed survival curves. A logistic regression model was used to determine the independent risk factors for the severe adverse outcome. Clinical variables were selected according to clinical importance and the significance level in the univariate analysis of p < 0.1. All predictors were preoperative or intraoperative except VIS_max. A correlation matrix was generated to assess all the continuous variables for collinearity. A forward stepwise method was used to screen variables for the multivariate model. The missing values for each variable were imputed using the multiple imputation method. A nomogram was constructed based on the multivariate logistic model. The regression coefficients in the model were used to derive linear predictors and allocate points in the nomogram.

The model’s performance was evaluated by calibration, discrimination and clinical utility. The calibration was assessed using a calibration plot and the Hosmer-Lemeshow test. The discrimination was assessed using the C-index or area under the curve (AUC) in the receiver operating characteristics (ROC) plot. The difference between the two AUCs was examined using DeLong’s method. The net reclassification index (NRI) and the integrated discrimination index (IDI) were calculated to determine whether the addition of a new index to the original model would improve the prediction. A decision curve analysis was performed to evaluate the clinical utility of the nomogram. Statistical analyses were conducted using SPSS v22.0 (SPSS, Chicago, IL, United States) and R v4.2.1 (The R Foundation for Statistical Computing, Vienna, Austria ¹ ). Figure plotting was completed using the same R software and GraphPad Prism v8.3.0 (GraphPad Software, San Diego, CA, United States). A p-value <0.05 was required for statistical significance.

The median age of our cohort was 50 years (IQR, 39.5–57 years), and the proportion of male patients was 78.46% (255/325). The median BMI was 22.81 kg/m^2 (IQR, 19.86–25.35 kg/m^2). After transplantation, the median VIS_max was 17.50 (12.92–24.90), and the rates for postoperative IABP and ECMO use were 37.23% (121/325) and 4.94% (16/325) respectively. Other demographic and clinical characteristics of the total cohort are summarized in Table 1. To explore the clinical value of VIS_max, the cohort was divided into two groups according to its median. The high VIS_max group (VIS_max >17.5) had baseline variables that were overall comparable with the low VIS_max group (VIS_max ≤17.5) except for the ratios of lung disease history and preoperative dopamine usage (Table 1).

The survival curves of the two VIS_max groups intersected at approximately day 20 within a 90-day and 1-year follow-up (Figures 1A, B). In the landmark analysis, no significant survival difference was observed before the intersection, while the survival of the low VIS_max group was evidently higher than that of the high VIS_max group after the intersection within a 90-day (p = 0.005) and 1-year follow-up (p = 0.039) (Figures 1D, E). Subsequently within a 6-year follow-up, the intersection became negligible and the survival difference between groups became not significant (Figure 1C). These results show that VIS_max is useful in predicting post-transplant survival in the short term rather than the long term.

High VIS_max was significantly associated with various early post-transplant outcomes such as intra-aortic balloon pump (IABP), extracorporeal membrane oxygenation (ECMO), cardiac reoperation, secondary intubation, CRRT, respiratory system syndrome, early death, prolonged duration of mechanical ventilation, ICU stay and hospital stay (Table 1). We further divided our cohort into 5 groups with different VIS_max grades. Grades 1 to 5 corresponded to a VIS_max of: <=10, 10–15, 15–20, 20–25 and >25 respectively [13]. Significant increasing trends along with VIS_max grade existed in the rates of CRRT, mechanical circulatory support (IABP or ECMO), prolonged mechanical ventilation, ICU stay and hospital stay (p < 0.05), while a tendency for this trend existed for other outcomes such as early death, septic shock and cardiac reoperation (p > 0.05) (Figure 2). The above results show that VIS_max is associated with extensive early outcomes in a grade-dependent manner, indicating its predictive ability for an early composite outcome. The severe adverse outcome occurred in 19.4% of our patients and was also significantly associated with VIS_max in a grade-dependent manner (p < 0.05) (Figure 2).

For model establishment, a set of candidate variables included common preoperative variables such as recipient age, sex, BMI, diagnosis, and donor age, sex, BMI and cold ischemia time; intraoperative variables such as CPB duration and operation length; and VIS_max (Details are in Supplementary Table S1). The univariate logistic regression analyses were conducted to determine whether each candidate variable had a potential association with the severe adverse outcome (Supplementary Table S1). Forward stepwise selection in multivariate logistic modeling identified the following 4 variables independently related to the severe adverse outcome: VIS_max (OR: 1.055; 95%CI: 1.027–1.084; p < 0.001), hemoglobin (OR: 0.981; 95%CI: 0.967–0.996; p = 0.013), serum creatinine (OR: 1.012; 95%CI: 1.005–1.019; p = 0.001) and advanced heart failure treatment (OR: 2.499; 95%CI: 1.265–4.939; p = 0.008) (Table 2). This model established from the complete variable set was called the “complete model”. Next, by excluding VIS_max from the variable set of the complete model, a simplified set was generated and used to construct a control model. Similarly, in multivariate modeling, we identified 3 independent variables for the same outcome: hemoglobin (OR: 0.982; 95%CI: 0.968–0.997; p = 0.015), serum creatinine (OR: 1.012; 95%CI: 1.005–1.019; p = 0.001), advanced heart failure treatment (OR: 2.318; 95%CI: 1.208–4.448; p = 0.011) (Supplementary Table S2).

The VIS-based nomogram for the severe adverse outcome in heart transplant recipients is shown in Figure 3. The points for each variable were summed up to generate a total score. A higher total score was related to a higher risk of the severe adverse outcome after heart transplantation. For example, a patient with a VIS_max of 7.62, hemoglobin of 96 g/L, serum creatinine of 70.7 μmol/L and no advanced heart failure treatment, would have 61.5 points (6.5 points for VIS_max, 42 points for hemoglobin, 13 points for serum creatinine and 0 points for advanced heart failure treatment), for a predicted risk of the severe adverse outcome of 10.7%.

The calibration curve of the VIS-based nomogram was near the diagonal line (Figure 4A). The Hosmer-Lemeshow test yielded a χ² of 8.094 (p = 0.424). There was a good agreement between the predicted and observed probabilities. The C-index was 0.745 (95%CI: 0.672–0.817) (Figure 4B), indicating moderate to strong discrimination. The prediction model after the removal of VIS_max is shown in Supplementary Table S2. The C-index for the control model was 0.708 (95%CI: 0.629–0.786), which was inferior to that of the complete model (Figure 4B). The addition of VIS_max to the control model resulted in a positive categorical NRI of 0.136 (p = 0.065), a significantly positive continuous NRI of 0.398 (p = 0.004), and a significantly positive IDI of 0.0485 (p = 0.006), suggesting a significant improvement in the risk classification ability of the model. The decision curve showed that when the selected interference threshold was >10%, using the VIS-based nomogram to predict the severe adverse outcome created more net clinical benefit than using a treat-all, a treat-none, and the control models (Figure 4C).