医学研究中常见动态预测模型方法介绍

An introduction of common dynamic predictive modeling methods in medical research

风险预测模型可预测个体的疾病风险,对疾病预防、治疗和预后管理等决策有重要价值。现有风险预测模型多仅使用单一时间截面的变量数据的静态模型,未考虑疾病进展过程中的诸多变化而导致预测准确性受限。动态预测模型可纳入随访期间同一指标的重复测量的纵向数据,捕捉个体特征随时间的纵向变化趋势,描述个体疾病风险的动态轨迹并提高模型的预测精度,然而其目前在医学研究中的应用仍然较少。本文总结目前常用的风险预测动态模型:联合模型、界标模型和贝叶斯动态模型,介绍各自应用场景、优缺点和软件实现并进行比较,以期为未来动态预测模型在医学研究中的应用提供方法学参考。

The risk prediction model(RPM) can be used to predict the risks of disease for individuals, playing an extremely important role in decision-making regarding disease prevention, treatment, and prognostic management. Most of the existing RPMs only utilize a single-time cross-section of variable data, so-called static models, which fail to consider the many changes during disease progression and lead to limited prediction accuracy. Dynamic prediction models can incorporate longitudinal data such as repeated measurements of variables during follow-up to capture the longitudinal changes in individual characteristics over time, describe the dynamic trajectory of individual disease risk and improve the prediction accuracy of the models;however, their application in medical research is still relatively small. In this paper, we conducted a systematic literature search to summarize the commonly used dynamic models: joint model, landmark model,and Bayesian dynamic model. By introducing their application scenarios, advantages and disadvantages, and software implementations and conducting comparisons, we aimed to provide methodological references for the future application of dynamic prediction models in medical research.