Over time, the performance of clinical prediction models may deteriorate due to changes in clinical management, data quality, disease risk and/or patient mix. Such prediction models must be updated in order to remain useful. Here, we investigate methods for discrete and dynamic model updating of clinical survival prediction models based on refitting, recalibration and Bayesian updating. In contrast to discrete or one-time updating, dynamic updating refers to a process in which a prediction model is repeatedly updated with new data. Motivated by infectious disease settings, our focus was on model performance in rapidly changing environments. We first compared the methods using a simulation study. We simulated scenarios with changing survival rates, the introduction of a new treatment and predictors of survival that are rare in the population. Next, the updating strategies were applied to patient data from the QResearch database, an electronic health records database from general practices in the UK, to study the updating of a model for predicting 70-day covid-19 related mortality. We found that a dynamic updating process outperformed one-time discrete updating in the simulations. Bayesian dynamic updating has the advantages of making use of knowledge from previous updates and requiring less data compared to refitting.

翻译：随着时间的推移，临床管理方式、数据质量、疾病风险及患者构成的变化可能导致临床预测模型性能下降。为保持模型有效性，必须对其进行更新。本研究比较了基于重新拟合、重新校准及贝叶斯更新的离散式与动态式临床生存预测模型更新策略。不同于单次离散更新，动态更新指利用新数据反复迭代更新预测模型的过程。受传染病临床场景启发，我们重点关注快速变化环境中的模型表现。首先通过模拟研究对比各类方法：我们模拟了生存率变化、新疗法引入及人群中出现罕见生存预测因子等场景。继而将各更新策略应用于英国全科医学电子健康记录数据库QResearch的患者数据，检验针对70天新冠相关死亡率预测模型的更新效果。研究发现：动态更新策略在模拟实验中优于单次离散更新；相比重新拟合方法，贝叶斯动态更新既可利用历史更新知识，又需要更少的数据支持。