How to Read Survival Decision Curve Analysis

What is Survival DCA?

Survival DCA extends standard Decision Curve Analysis to time-to-event data. Instead of a fixed binary outcome, it evaluates whether a prediction model provides net benefit at a specific time point t, accounting for censoring via Inverse Probability of Censoring Weighting (IPCW).

Reading the Main Plot

• X-axis (Threshold Probability, Pt): The predicted risk threshold at which you would decide to treat. Pt = 0.2 means treating when predicted risk of event by time t is >= 20%.
• Y-axis (Net Benefit): The IPCW-weighted net benefit of using the model at that threshold. Higher = better.
• Model curves: Each colored line shows a single predictor (Cox model) or the Combined Model.
• Treat All (gray dotted): Net benefit if everyone received treatment (IPCW-weighted).
• Treat None (gray dashed): Net benefit if no one received treatment (always 0).

Key Interpretation Rules

• A model is clinically useful when its curve is above both 'Treat All' AND 'Treat None' at clinically relevant thresholds.
• If a curve falls below 'Treat None', the model is worse than doing nothing.
• If a curve overlaps 'Treat All', the model offers no advantage over treating everyone.
• The wider the gap between a model and the gray reference lines, the greater the clinical value.

Time Point Selection

Use the slider in Section 3 to choose the evaluation time point. Each time point produces a separate DCA. For example, at t = 5 years, the analysis evaluates the model's ability to predict 5-year event risk. The default is the median survival time.

Combined Model Detail Plot

When multiple predictors are selected, a multivariable Cox model is automatically fitted. The bottom plot zooms in on the Combined Model (red line) against Treat All / Treat None references.

Multi-Time DCA

The 'Multi-Time DCA' tab compares the Combined Model's net benefit curve across multiple time points (e.g., t = 1, 3, 5, 7 years). This reveals how the model's clinical utility evolves over follow-up time. Only the Combined Model is plotted; each colored line represents a different evaluation time. The gray dashed line at NB = 0 represents Treat None.

Threshold Probe Tool

Use the 'Threshold Probability' input + 'Set Probe' button to mark a specific Pt on the Combined Model plot. This draws an orange dashed line showing:

• The exact Net Benefit at your chosen threshold
• A horizontal line to the Y-axis showing NB value
• The Pt value labeled below the X-axis

IPCW Methodology

Censoring is handled via Inverse Probability of Censoring Weighting (Vickers et al., 2008):

• G(t) = Kaplan-Meier estimate of censoring survival (status reversed).
• Known cases (event by t): weight = 1/G(T_i) - accounts for subjects censored before the event.
• Known controls (survived past t, NOT censored): weight = 1/G(t).
• Censored subjects: excluded from direct counting; their information is embedded in the weights of known subjects.
• Net Benefit = (weighted TP - weighted FP * Pt/(1-Pt)) / N_total

Data Requirements

• Time: Survival time (numeric, >= 0).
• Status: Event indicator (0 = censored, 1 = event).
• Predictors: Continuous numeric variables. Categorical variables are not supported.
• ID columns: Automatically detected and excluded.
• Missing values: Rows with any NA in selected variables are removed before analysis.

Models Fitted

• Individual markers: Univariate Cox model: Surv(time, status) ~ marker. Predicted risk = 1 - S(t|marker).
• Combined Model: Multivariable Cox: Surv(time, status) ~ marker1 + marker2 + ... Predicted risk = 1 - S(t|all markers).

Export & Download

• PNG: High-resolution plot (300 DPI) for publications.
• CSV (Summary): BOM header for direct opening in Windows Excel.
• CSV (Optimal Threshold): Threshold ranges where each model beats Treat All / Treat None.
• CSV (Raw DCA): Complete point-by-point DCA data for custom plotting.

Demo Data Source

The demo dataset simulates 200 subjects with survival time, event status, and 5 biomarkers (creatinine, glucose, WBC, age, BMI). High-risk group (n=100) has shorter survival and higher biomarker values. Some missing values are included for realism.

Reference

Vickers AJ, Cronin AM, Elkin EB, Gonen M. Extensions to decision curve analysis, a novel method for evaluating diagnostic tests, prediction models and molecular markers. BMC Medical Informatics and Decision Making 2008; 8:51.