A total of 1,876 suspected AMI patients with high troponin I levels and chest pain lasting longer than 12 h, who were hospitalized to the First Affiliated Hospital of Soochow University's Department of Cardiology between January 2012 and September 2015 were retrospectively included in this study. 1,007 patients were finally included according to the inclusion criteria. They were divided into a death group (n = 122) and a survival group (n = 885) according to their survival status. Figure 1 displayed the inclusion and exclusion criteria (26).

Clinical baseline data included: 1) Demographic characteristics: age, sex, and lifestyle; 2) Medical history: hypertension, diabetes mellitus, stroke, smoking, and alcohol consumption; 3) Height and weight; 4) ECG characteristics; 5) Coronary angiographic findings: single-vessel, double-vessel, or multi-vessel disease; 6) Whether PCI was performed; 7) Left ventricular ejection fraction; 8) Laboratory parameters: blood lipids, serum albumin, serum creatinine, blood glucose, and troponin I; 9) ACEF score and GRACE score.

Biochemical indicators were obtained from fasting blood samples collected on the second day after admission (usually after at least 8 h of fasting) and analyzed using a Siemens 2,400 fully automated biochemical analyzer (serum albumin was determined using the bromocresol green method). For each enrolled patient, risk scores were calculated: the GRACE score according to established methodology (27), and the ACEF score using the formula: age (years)/left ventricular ejection fraction (%)+creatinine score (1 if serum creatinine ≥2 mg/dL; 0 if <2 mg/dL) (28).

All patients were followed up for 4 years after discharge, with the primary endpoint being all-cause mortality. All-cause mortality was defined according to the standardized Academic Research Consortium-2 (ARC-2) definition as death from any cause during the study period (29). Acute myocardial infarction (AMI) was defined according to the Third Universal Definition of Myocardial Infarction (2012) (30). The A2A Index (A2A Index) is defined as the age to serum albumin ratio, calculated as age (years)/serum albumin (g/L). We acknowledge that the GRACE score was primarily developed and validated for predicting in-hospital and 6-month post-discharge mortality. Nonetheless, our study aimed to evaluate the A2A index for long-term mortality (up to 4 years), a timeframe beyond the GRACE score's original design.

Candidate variables identified by univariate analysis (P < 0.05) were incorporated into multivariate Cox proportional hazards models. The variance inflation factors (VIFs) of all included variables were below 5, suggesting no significant multicollinearity.

Three risk models were defined: Model 1: univariate analysis; Model 2: adjusted for sex, body mass index (BMI), smoking status, and alcohol consumption; Model 3: adjusted for sex, BMI, smoking, alcohol consumption, triglycerides, troponin I, serum creatinine, left ventricular ejection fraction (LVEF), hypertension, diabetes mellitus, prior stroke, number of diseased coronary vessels, and percutaneous coronary intervention (PCI).

Statistical analysis was performed using Stata/MP 17.0 (Stata Corp, Texas, USA). Missing values were handled using multiple imputation (chained equations, 25 iterations). Continuous variables were expressed as mean ± standard deviation or median (interquartile range, IQR), while categorical variables were expressed as frequencies and percentages. Comparisons were performed using the Kruskal–Wallis rank test or Pearson's chi-square test, as appropriate. Kaplan–Meier curves were used to compare survival differences between different A2A Index groups. Candidate variables identified by univariate analysis (P-value <0.05) were incorporated into multivariate Cox proportional hazards model to assess the independent predictive value of the A2A Index for all-cause mortality. Restricted cubic spline was used to analyze non-linear relationships between the A2A Index and all-cause mortality. Subgroup analysis and interaction testing evaluated the consistency of the relationship between the A2A Index and all-cause mortality across different population characteristics. Sensitivity analysis assessed the robustness of results by excluding specific samples. Receiver operating characteristic (ROC) curves were used to evaluate the discriminative ability of the A2A Index compared to ACEF score and GRACE score, while the Hosmer–Lemeshow test assessed the calibration. We have also performed time-dependent ROC analysis and Landmark analysis. Comparative analysis of different ROC curves using DeLong test. All tests were two-sided, a P-value <0.05 was considered statistically significant.

Baseline clinical characteristics were presented in Table 1. Compared with the survival group, patients in the death group were older (median age 75 vs. 65 years), had higher serum creatinine levels and A2A Index values, and exhibited a higher prevalence of hypertension, diabetes mellitus, or prior stroke (all P < 0.05). In contrast, serum albumin levels and left ventricular ejection fraction (LVEF) were significantly lower in the death group (P < 0.05). Unexpectedly, the proportions of both smoking and drinking in the death group were low. The A2A Index was left-skewed in distribution with a smaller proportion exhibiting extremely high values associated with poor prognosis.

The A2A Index was stratified into four quartiles: Q1 (<1.43), Q2 (1.43–1.69), Q3 (1.70–2.02), and Q4 (≥2.02). Correlation analysis indicated a moderate positive association between the A2A Index and the risk of all-cause mortality (Somers' D = 0.55; 95% CI: 0.475–0.622; P < 0.001). As shown in (Figure 2), Kaplan–Meier survival declined with increasing A2A quartiles. Higher A2A quartiles were associated with progressively lower survival probabilities (log-rank P < 0.001), with the Q4 group exhibiting the lowest 4-year survival rate at 56%.

The results of the univariate analysis and multivariate Cox proportional hazards regression were presented in Table 2. Using the lowest A2A quartile (Q1) as the reference, the adjusted hazard ratios (HRs) for all-cause mortality were 5.71 (95% CI: 1.40–23.36) in Q2, 18.59 (95% CI: 4.59–73.39) in Q3, and 27.64 (95% CI: 6.99–109.18) in Q4, demonstrating a significant dose–response effect (P for trend <0.001). we have performed a landmark analysis excluding 6 events occurring within the first 48 h. The results consistently demonstrate that higher A2A quartiles were associated with progressively lower survival probabilities (log-rank P < 0.001) (Supplementary Figure 3A).

The A2A Index was analyzed as a continuous variable in univariate analysis (Table 3), each one-unit increase in the A2A Index was associated with a 7.62-fold higher risk of all-cause mortality (HR = 7.62; 95% CI: 5.43–10.68; P < 0.001). In sex-stratified univariate analyses, the association was significant in both males (HR = 6.40; 95% CI: 4.34–10.16) and females (HR = 8.96; 95% CI: 4.90–16.36), with all P-values < 0.001. Multivariate Cox proportional hazards regression analysis showed that each one-unit increase of A2A Index was associated with a 4.98-fold higher risk of all-cause mortality (HR = 4.98; 95% CI: 3.34–7.43; P < 0.001) (Table 3).

Restricted cubic spline regression illustrated a significant J-shaped dose-response relationship between the A2A Index and all-cause mortality risk (P-nonlinearity < 0.001) (Figure 3).

ROC curves were used to evaluate the prediction performance of the A2A Index. The area under the curve (AUC) for the A2A Index was 0.83(95% CI: 0.79–0.86), which was significantly greater than that of the GRACE score [AUC = 0.80(95% CI: 0.76–0.84); P = 0.041] and comparable to the ACEF score [AUC = 0.83(95% CI: 0.79–0.87); P = 0.656] (Figure 4). The optimal cutoff value for the A2A Index was 1.86, with a sensitivity of 79% and a specificity of 70%. The A2A Index (χ² = 9.08, P = 0.336) demonstrated a better calibration than the GRACE score (χ² = 14.13, P = 0.078) and a comparable one to the ACEF score (χ² = 5.70, P = 0.680) (Figure 5). we also performed time-dependent ROC analysis at 1, 2, 3and 4 years post-discharge. The results consistently demonstrate that the A2A index maintains discriminative ability over time, with AUC values of 0.821, 0.818, 0.831and 0.836 respectively. Within the initial two years, A2A index and GRACE score show equivalent risk; thereafter, A2A index is associated with a higher all-cause mortality risk than GRACE score (Supplementary Figures 1a–d and 2a).

After adjusting for confounding factors, an elevated A2A Index was significantly associated with increased mortality across all subgroups (P < 0.001), except among patients without hypertension (P > 0.05). No significant interactions were observed between the A2A Index and any subgroup variables (all P for interaction > 0.05). Detailed results of the subgroup analyses and interaction terms are shown in Figure 6.

Sensitivity analyses confirmed the robustness of the findings. The association between the A2A Index and all-cause mortality remained statistically significant after excluding patients who died within the first year of follow-up, those with missing BMI data, and those meeting both conditions (all P < 0.001). Data were shown in Supplementary Materials.

This study has several limitations. First, it was a retrospective analysis conducted at a single center, and external validation using multicenter prospective cohorts is needed. Moreover, as the study population included only AMI patients who presented more than 12 h after symptom onset, the findings may not be generalizable to those receiving early interventional therapy. Second, only all-cause mortality was assessed as the clinical endpoint. Future research should incorporate specific cardiovascular outcomes to better evaluate the prognostic value of the A2A Index. Third, the A2A Index was calculated based solely on baseline measurements at admission, without accounting for dynamic changes in serum albumin levels during hospitalization or follow-up. Fourth, bromocresol Green Method (BCG) may overestimate certain parameters compared to direct biochemical assays. We will conduct further studies to externally validate our findings using alternative measurement techniques and diverse populations. Fifth, we must consider potential biases, such as those arising from measurement bias, unmeasured confounders, and sample selection bias.