Related Experiment Video
Updated: Sep 13, 2025

Establishing a Competing Risk Regression Nomogram Model for Survival Data
Published on: October 23, 2020
When would I be surprised? Variability in predicted probability of survival for being "surprised" and "not surprised"
David Hui1,2, Amy Ontai3, Clark Andersen4
1Department of Palliative, Rehabilitation and Integrative Medicine, The University of Texas MD Anderson Cancer Center, Unit 1414, 1515 Holcombe Boulevard, Houston, TX, 77030, USA. dhui@mdanderson.org.
Purpose:
The surprise question is commonly used in clinical practice; however, the variability in how "surprise" is defined is unclear. We examined the variability in the predicted probability of survival for being "surprised" and "not surprised" among palliative care physicians across seven timeframes.
Methods:
Palliative care specialists completed the surprise question for 3 days, 1 week, 2 weeks, 1 month, 2 months, 3 months, and 6 months for patients with advanced cancer at an outpatient clinic and concurrently predicted the probability of survival (0-100%). The primary outcome, coefficient of variation (CV) for the predicted probability of survival, was computed for "surprised" and "not surprised" answers for each physician and each timeframe. A mixed-effect logistic regression assessed the probability threshold for being "surprised."
Results:
Twenty physicians provided 3024 survival estimates for 216 patients (mean age 61, 50% female). When "surprised," physicians consistently predicted a probability of survival > 50%, with low variation among physicians (mean CV 6-23%) and across timeframes (mean CV 5-20%). In contrast, physicians who answered "not surprised" predicted a probability of survival from 2 to 100%, with high variation among physicians (mean CV 15-83%) and across timeframes (mean CV 22-69%). Variability increased with longer timeframes. The probability of survival thresholds for being "surprised" were ≥ 74%, ≥ 62%, ≥ 68%, ≥ 83%, and ≥ 84% for 1 week, 2 weeks, 1 month, 2 months, and 3 months, respectively.
Conclusion:
We found low variability for predicted probability of survival when clinicians were "surprised" but high variability when they were "not surprised."
Related Concept Videos
Assumptions of Survival Analysis
Comparing the Survival Analysis of Two or More Groups
Censoring Survival Data
Parametric Survival Analysis: Weibull and Exponential Methods
Weibull Distribution
The Weibull distribution is a flexible model used in parametric survival analysis. It can handle both increasing and decreasing hazard rates, depending on its shape parameter...
Survival Tree
Building a Survival Tree
Constructing a...
Survival Curves
The Kaplan-Meier estimator is the most common method for constructing survival curves. This...

