Time Series & Survival · TOOL GUIDE

Cox Proportional Hazards

Open Quantly ↗

Understand the method. Use it with confidence.

Model relationship between covariates and survival time. Quantly runs the method against your active analysis context and returns structured results for review and interpretation.

01

When to use it

Use this tool for ordered observations, forecasting, or time-to-event outcomes where timing and censoring matter.
Use it only when the selected variables and study design match the method's requirements.

02

How to use it

  1. Open the tool from the Quantly Tools panel.
  2. Confirm the active dataset and analysis session.
  3. Set Time to Event Column (Required): Select the column with duration data.
  4. Set Event Observed Column (Required): Select the column indicating if the event occurred.
  5. Set Event Value (Required): Value indicating an event if the event column is text (e.g., 'Dead').
  6. Set Covariates (Required): Select columns to include as predictors in the model.
  7. Set Strata Column (Optional) (Optional): Stratify by this column to allow non-proportional hazards between groups.
  8. Set Categorical Encoding (Required): Method for encoding categorical covariates.
  9. Set Significance Level (α) (Required): Controls the confidence interval width (1-α).
  10. Set Penalizer (L2 Regularization) (Optional): Add a small penalty to handle collinearity, especially with one-hot encoding. Try 0.01 to start.
  11. Set Generate Assumption Plots (Optional): Generate plots to visually check the proportional hazard assumption. This can be slow.
  12. Review optional or advanced settings and keep defaults unless your design requires a change.
  13. Run the tool, inspect warnings and diagnostics, then save or export the result with its assumptions.
03

How to interpret results

Review time order, stationarity or proportional-hazards assumptions, uncertainty intervals, censoring, and validation over later periods. Forecasts and survival estimates become less certain farther from observed data.