CSAIL & IMES, Massachusetts Institute of Technology, Cambridge, MA.
Harvard Medical School, Boston, MA.
JCO Clin Cancer Inform. 2021 May;5:550-560. doi: 10.1200/CCI.20.00139.
Key oncology end points are not routinely encoded into electronic medical records (EMRs). We assessed whether natural language processing (NLP) can abstract treatment discontinuation rationale from unstructured EMR notes to estimate toxicity incidence and progression-free survival (PFS).
We constructed a retrospective cohort of 6,115 patients with early-stage and 701 patients with metastatic breast cancer initiating care at Memorial Sloan Kettering Cancer Center from 2008 to 2019. Each cohort was divided into training (70%), validation (15%), and test (15%) subsets. Human abstractors identified the clinical rationale associated with treatment discontinuation events. Concatenated EMR notes were used to train high-dimensional logistic regression and convolutional neural network models. Kaplan-Meier analyses were used to compare toxicity incidence and PFS estimated by our NLP models to estimates generated by manual labeling and time-to-treatment discontinuation (TTD).
Our best high-dimensional logistic regression models identified toxicity events in early-stage patients with an area under the curve of the receiver-operator characteristic of 0.857 ± 0.014 (standard deviation) and progression events in metastatic patients with an area under the curve of 0.752 ± 0.027 (standard deviation). NLP-extracted toxicity incidence and PFS curves were not significantly different from manually extracted curves ( = .95 and = .67, respectively). By contrast, TTD overestimated toxicity in early-stage patients ( < .001) and underestimated PFS in metastatic patients ( < .001). Additionally, we tested an extrapolation approach in which 20% of the metastatic cohort were labeled manually, and NLP algorithms were used to abstract the remaining 80%. This extrapolated outcomes approach resolved PFS differences between receptor subtypes ( < .001 for hormone receptor+/human epidermal growth factor receptor 2- human epidermal growth factor receptor 2+ triple-negative) that could not be resolved with TTD.
NLP models are capable of abstracting treatment discontinuation rationale with minimal manual labeling.
关键的肿瘤学终点通常不会被编码到电子病历（EMR）中。我们评估了自然语言处理（NLP）是否可以从非结构化的 EMR 记录中提取治疗中断的基本原理，以估计毒性发生率和无进展生存期（PFS）。
我们构建了一个回顾性队列，包括 2008 年至 2019 年在纪念斯隆凯特琳癌症中心接受治疗的 6115 例早期乳腺癌患者和 701 例转移性乳腺癌患者。每个队列分为训练（70%）、验证（15%）和测试（15%）子集。人工摘要员确定与治疗中断事件相关的临床基本原理。串联 EMR 记录用于训练高维逻辑回归和卷积神经网络模型。Kaplan-Meier 分析用于比较我们的 NLP 模型估计的毒性发生率和 PFS 与手动标记和治疗中断时间（TTD）生成的估计值。
我们最好的高维逻辑回归模型在早期患者中识别出毒性事件的曲线下面积为 0.857±0.014（标准差），在转移性患者中识别出进展事件的曲线下面积为 0.752±0.027（标准差）。NLP 提取的毒性发生率和 PFS 曲线与手动提取的曲线没有显著差异（=0.95 和=0.67）。相比之下，TTD 高估了早期患者的毒性（<0.001），低估了转移性患者的 PFS（<0.001）。此外，我们测试了一种外推方法，其中转移性队列的 20%手动标记，其余 80%使用 NLP 算法提取。这种外推方法解决了 TTD 无法解决的受体亚型之间的 PFS 差异（激素受体+/人表皮生长因子受体 2-人表皮生长因子受体 2+三阴性受体<0.001）。
NLP 模型能够在最小的人工标记下提取治疗中断的基本原理。
