Development and Internal Validation of a Nomogram for Predicting One-Year All-Cause Mortality in Heart Failure Patients Undergoing Non-Cardiac Surgery: A Retrospective Analysis of 308 Cases

Technical Summary: Development and Internal Validation of a Nomogram for Predicting One-Year All-Cause Mortality in Heart Failure Patients Undergoing Non-Cardiac Surgery

Problem Statement
Heart failure (HF) represents a significant global public health burden, with over 15% of patients undergoing non-cardiac surgery (NCS) presenting with comorbid HF. These patients face elevated risks of perioperative hemodynamic instability and long-term postoperative mortality. While surgical interventions often trigger traumatic stress responses in HF patients, there is currently no straightforward, precise instrument available to forecast the probability of one-year postoperative mortality specifically for this population. Existing tools, such as the Revised Cardiac Risk Index (RCRI), lack specific biomarkers and routine clinical indicators necessary for accurate long-term risk stratification in HF cohorts. This study addresses the need for an individualized, precise risk evaluation tool to facilitate early formulation of personalized therapeutic strategies and improve postoperative survival outcomes.

Methodology
This retrospective cohort study analyzed 308 patients with heart failure who underwent non-cardiac surgery at Nanjing Luhe District People's Hospital between January 2023 and June 2024. The study was approved by the local Medical Ethics Committee and registered with the Chinese Clinical Trial Registry.

• Data Collection and Preprocessing: The study collected 18 candidate predictive covariates, including demographic data, surgical characteristics, anesthesia type, and various laboratory biomarkers (e.g., BNP, CRP, lipoprotein(a), hemoglobin, cardiac markers). Missing values (<5%) were handled via multiple imputation (MICE). The cohort was randomly stratified into a training set (n=217) and a validation set (n=91) at a 7:3 ratio.
• Variable Screening: A two-step screening process was employed. First, LASSO regression combined with the Boruta algorithm was used to identify potential predictors and control for multicollinearity. This reduced the initial 18 variables to 9 candidates. Second, a stepwise backward elimination approach within a multivariate logistic regression framework (entry p=0.05, removal p=0.10) identified the final set of independent predictors. Benjamini–Hochberg False Discovery Rate (FDR) correction was applied to mitigate Type I errors.
• Model Construction: A nomogram was constructed based on the six independent predictors identified. The model utilized logistic regression to calculate the probability of one-year all-cause mortality.
• Validation and Evaluation: The model's performance was rigorously evaluated using:
  • Discrimination: Receiver Operating Characteristic (ROC) curves and Area Under the Curve (AUC) with 95% Confidence Intervals (CI).
  • Calibration: Calibration plots and the Hosmer-Lemeshow test.
  • Clinical Utility: Decision Curve Analysis (DCA) and Clinical Impact Curves (CIC).
  • Comparison: The nomogram was compared against the conventional RCRI score using DeLong tests.
  • Sample Size Justification: The sample size (n=308, 69 events) was verified to meet the "10:1" events-per-variable rule and statistical power requirements (80% power, OR=1.6).

Key Contributions

1. Identification of Six Independent Predictors: The study identified six routine clinical indicators as independent predictors of one-year mortality:
   • Postoperative new-onset cardiac insufficiency (strongest predictor).
   • Myocardial biomarker levels (elevated preoperative cTnI).
   • B-type natriuretic peptide (BNP).
   • Lipoprotein(a) [Lp(a)].
   • Body Mass Index (BMI).
   • Hemoglobin (Hb).
2. Development of a High-Performance Nomogram: A predictive model was constructed that integrates these variables into a visual scoring system.
3. Dynamic Clinical Tool: An interactive, web-based dynamic nomogram (R Shiny application) was developed to allow clinicians to input patient data and instantly receive personalized mortality risk probabilities, eliminating manual calculation errors.
4. Superiority Over RCRI: The study demonstrated that the proposed nomogram significantly outperforms the traditional RCRI score in predictive accuracy for this specific patient population.

Results

• Mortality Rate: The one-year postoperative all-cause mortality rate in the cohort was 22.4% (69/308).
• Predictive Performance:
  • Training Set: The nomogram achieved an AUC of 0.933 (95% CI: 0.893–0.974).
  • Validation Set: The model maintained strong discriminative ability with an AUC of 0.892 (95% CI: 0.812–0.971).
  • Comparison with RCRI: The nomogram significantly outperformed the RCRI in both the training set (AUC 0.933 vs. 0.659, P<0.001) and the validation set (AUC 0.892 vs. 0.700, P=0.003).
• Calibration: The model showed excellent calibration. The Hosmer-Lemeshow test yielded P-values of 0.8428 (training) and 0.8973 (validation), indicating no significant lack of fit. Mean absolute calibration errors were low (3.41% and 4.05%, respectively).
• Clinical Utility: Decision Curve Analysis (DCA) confirmed favorable net clinical benefits within a threshold probability of 0.0–0.4. Clinical Impact Curves (CIC) verified the model's ability to accurately screen high-risk patients across varying decision thresholds.
• Variable Importance: Postoperative new-onset cardiac insufficiency (OR=9.434) and Lp(a) (OR=1.005 per unit) were identified as the primary drivers of risk. Conversely, higher BMI and Hemoglobin levels were identified as protective factors.

Significance and Claims
The authors claim that this novel nomogram provides an intuitive, reliable, and precise tool for perioperative risk assessment in heart failure patients undergoing non-cardiac surgery. By integrating routine clinical indicators and postoperative complication data, the model offers superior predictive accuracy compared to existing standard indices.

The study posits that this tool facilitates the identification of high-risk cohorts, enabling the early implementation of personalized therapeutic strategies (e.g., nutritional support for low BMI, anemia management, and close monitoring for new-onset heart failure). The authors emphasize that the model's reliance on routinely available indicators makes it highly feasible for clinical implementation. They acknowledge limitations, including the single-center retrospective design and the lack of long-term follow-up beyond one year, stating that future multi-center prospective studies are necessary to validate the model's generalizability. However, they conclude that the current model represents a significant step toward precise perioperative management and individualized clinical intervention for this vulnerable patient population.