One of the great promises of applying machine learning to clinical data is the possibility of learning optimal per-patient treatment rules. The goal is to use data collected in clinical settings such as a patient’s hospital record, or data collected in clinical trials, to learn an individualized treatment rule that will have greater benefit than existing treatment policies if applied to the entire population. Concretely, we use this running example: choosing the best blood pressure-lowering medication “A” or “B” for a patient with hypertension, measured in terms of lower 6-week prospective blood pressure. We note that hypertension is a condition with substantial variation in how patients are actually treated (Hripsak and others, 2016). Learning such individualized treatment rules involves at least two distinct challenges—a statistical one and a causal one.

中文翻译：

---

我们可以从临床数据中学习个体水平的治疗策略吗？

将机器学习应用于临床数据的最大希望之一就是有可能学习最佳的每位患者治疗规则。目的是使用在临床环境中收集的数据（例如患者的医院记录）或在临床试验中收集的数据，以学习个性化的治疗规则，如果将其应用于整个人群，则其收益将比现有治疗政策更大。具体来说，我们使用这个运行的示例：为高血压患者选择最佳的降血压药物“ A”或“ B”，以较低的6周预期血压来衡量。我们注意到，高血压是患者实际治疗方式差异很大的疾病（Hripsak等，2016）。