Dynamic Prediction Models in Cancer Prognosis

Dynamic prediction models represent a shift from static, one‐off risk assessments to continually updated prognostic tools that accommodate evolving patient information. In cancer care, longitudinal biomarkers, imaging findings and treatment milestones often influence a patient’s risk of recurrence or mortality. Dynamic models harness time-dependent covariates and repeated measures to refine individual risk estimates at clinically relevant time points. Approaches include landmark analysis, which recalculates risk from specified “landmark” times using updated predictor values, and joint models, which link the trajectory of a longitudinal biomarker to the hazard of an event within a unified statistical framework. These techniques can also accommodate competing risks—situations where non‐cancer events preclude the outcome of interest—and produce time-dependent discrimination and calibration metrics such as dynamic receiver operating characteristic curves and Brier scores. By integrating new data as patients progress through surveillance or therapy, dynamic prediction supports personalised decision-making, more precise timing of follow-up investigations and adaptive treatment strategies. Practical implementations include user-friendly software and web-based calculators that facilitate clinical uptake. As oncology moves towards precision medicine, dynamic models are increasingly recognised for their global relevance in improving prognostic accuracy, optimising resource allocation and enhancing patient counselling.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Dynamic Prediction Models in Cancer Prognosis publication trend

The graph below shows the total number of articles in dynamic prediction models in cancer prognosis across all publications each year (not limited to Nature Index journals).

Technical terms

Dynamic prediction model: A statistical approach that updates an individual’s risk estimate over time by incorporating new information as it becomes available.

Landmark analysis: A method that defines specific time points (“landmarks”) and builds prediction models using covariate values observed up to each landmark.

Joint model: A unified framework combining longitudinal (repeated measures) and time-to-event submodels to account for the relationship between a biomarker trajectory and an outcome.

Time‐dependent covariate: A predictor variable whose value can change over the course of follow-up.

Competing risks: A scenario in survival analysis where alternative events prevent the occurrence of the primary event of interest.

Time‐dependent ROC curve: An extension of the receiver operating characteristic curve that evaluates the discrimination of a prognostic model at different time horizons.

References

1. Software Application Profile: dynamicLM—a tool for performing dynamic risk prediction using a landmark supermodel for survival data under competing risks. International Journal of Epidemiology (2023).
2. Predicting the risk of a clinical event using longitudinal data: the generalized landmark analysis. BMC Medical Research Methodology (2023).
3. Nonlinear joint models for individual dynamic prediction of risk of death using Hamiltonian Monte Carlo: application to metastatic prostate cancer. BMC Medical Research Methodology (2017).