A simplified nomogram for predicting deep vein thrombosis: development in NHANES and retrospective single-center external validation

Figures

• 

  Figure 1.. Study flowchart depicting participant selection and cohort establishment. The research framework integrated a development cohort from the National Health and Nutrition Examination Survey (NHANES) and an external validation cohort from a tertiary hospital in China. For the development cohort, NHANES participants (2005–2018) were initially screened (N = 70,190), with exclusions based on age (< 20 years, n = 30,441) and missing primary outcome (deep vein thrombosis [DVT]) status (n = 57), resulting in a final analytical cohort of 39,692 individuals (825 DVT and 38,867 non-DVT cases). LASSO regression was then employed within this development cohort to select optimal predictors. Concurrently, an external validation cohort was established by identifying patients from the electronic medical records of The Affiliated Huaian No.1 People’s Hospital of Nanjing Medical University between February 2016 and July 2023 using predefined inclusion and exclusion criteria. After variable harmonization, the final 4-variable model was validated in 3,136 patients (1,036 DVT and 2,100 non-DVT cases). Abbreviations: DVT, deep vein thrombosis; NHANES, National Health and Nutrition Examination Survey;…

• 

  Figure 2.. Variable selection and multivariable logistic regression analysis for DVT risk factors. The figure illustrates the process of predictor selection and model construction. (A) LASSO coefficient paths. Each curve represents the coefficient of a potential predictor as a function of the L1 penalty (Log Lambda). As the penalty increases, non-essential variables are shrunk to zero, mitigating overfitting. (B) LASSO parameter selection via cross-validation. The binomial deviance is plotted against Log(λ). The two vertical dashed lines indicate the λ at minimum deviance and the 1-standard-error (1-SE) criterion, with the latter used to select a parsimonious set of four final predictors. (C) Forest plot of multivariable logistic regression results. Odds ratios (ORs) and 95% confidence intervals (CIs) are shown for the four variables selected via LASSO. Age (per year), diabetes, and smoking history were significantly and independently associated with DVT risk (all p <.001), whereas physical activity did not reach statistical significance (p =.166). Abbreviations: CI, confidence interval; DVT, deep vein thrombosis; LASSO, Least Absolute Shrinkage and Selection Operator; OR, odds ratio.

• 

  Figure 3.. Predictive nomogram for individualized DVT risk assessment. The nomogram was constructed based on the four independent predictors identified in the multivariable logistic regression model from the NHANES development cohort. To estimate the risk of deep vein thrombosis (DVT), a vertical line is drawn upward from each variable value (Age, Diabetes Mellitus, Smoking Status, and Physical Activity) to the ‘Points’ scale to obtain the score for each factor. The sum of these individual scores is then located on the ‘Total Points’ axis. Finally, a vertical line is drawn downward to the ‘Predicted Probability of DVT’ axis to determine the individualized risk. Advanced age contributes the most to the cumulative risk score, followed by diabetes, smoking, and physical inactivity. Abbreviations: DVT, deep vein thrombosis; NHANES, National Health and Nutrition Examination Survey.

• 

  Figure 4.. Performance evaluation of the DVT prediction models in the development and external validation cohorts. Development cohort (NHANES): (A1) ROC curves illustrating the discrimination of the logistic and XGBoost models, with the logistic model achieving an AUC of 0.816 (XGBoost, 0.820); (A2) Calibration plot for the logistic model, showing close agreement between predicted and observed DVT probabilities (MAE = 0.003; calibration slope, 0.998; intercept, 0.002); (A3) Decision curve analysis (DCA) demonstrating net clinical benefit across a wide range of threshold probabilities. External validation cohort (The Affiliated Huaian No.1 People’s Hospital of Nanjing Medical University): (B1) ROC curves in the clinical cohort (Logistic AUC = 0.792; XGBoost = 0.804); (B2) Calibration plot for the external cohort, indicating a mean absolute error (MAE) of 0.022, a calibration slope of 0.963, and an intercept of 0.014; (B3) DCA confirming the clinical utility of the predictive model in a real-world hospital setting. Abbreviations: AUC, area under the curve; DCA, decision curve analysis; DVT, deep vein thrombosis; MAE, mean absolute error; NHANES, National Health and Nutrition Examination…