摘要
一位新闻记者兼机器人与机器学习的新闻编辑每日新闻-一篇关于癌症的新研究是一篇报道的主题。根据NewsRx Jo Urnalists在新加坡的新闻报道，研究表明：“预先识别生命最后一年有助于采取积极的姑息治疗方法。在电子健康记录(EHR)上训练的机器学习模型显示出在癌症预测方面的良好表现。”这项研究的财政支持者包括国家医学研究委员会、Lien姑息治疗中心。新闻记者从新加坡国家癌症中心获得了一段研究的引文，“然而，文献中的空白包括对模型性能的报道不完整，模型制定与实施用例不一致，以及解释不足阻碍了临床AL环境中的信任和采用。因此，”我们的目标是开发一个基于机器学习的EHR模型，通过预测门诊晚期癌症患者365天死亡风险来提示姑息治疗过程。我们的COHO RT包括5926名在2017年7月1日和2006年6月30日被诊断为3或4期实体器官癌的成年人。分类问题是使用极端梯度t Boosting(XGBoost)建模的，并与我们设想的用例相一致：“给定一个与门诊癌症遭遇相对应的预测点，使用EHR数据预测从预测点开始365天内的死亡率，直到365天前。”该模型使用75%的数据集进行训练（n=39416）门诊就诊)，并在25%的坚持数据集(n=13122 o utpatient就诊)上进行验证。为了解释模型输出，我们使用了Shapley加法解释(SHAP)值。使用临床特征、实验室测试和治疗数据来训练模型。使用受试者操作特征曲线下面积(AUROC)和精确呼叫曲线下面积(AUPRC)评估性能。在52538个预测点中，17149个(32.6%)在365天预测窗内发生了MORT事件，模型显示AURO C为0.861(95%CI 0.856-0.867)，AUPRC为0.771.，Brier SCOR E为0.147.这表明对死亡风险的估计略微过高。利用SHAP值的解释性Di agrams允许可视化特征对预测离子在全球和个人层面上的影响。我们的机器学习模型在预测晚期癌症患者365天死亡风险方面表现出良好的辨别能力和精确回忆能力。
Abstract
By a News Reporter-Staff News Editor at Robotics & Machine Learning Daily News Daily News – New research on Cancer is the subject of a report. According to news reporting from Singapore, Singapore, by NewsRx jo urnalists, research stated, “Ex-ante identification of the last year in life fac ilitates a proactive palliative approach. Machine learning models trained on ele ctronic health records (EHR) demonstrate promising performance in cancer prognos tication.” Financial supporters for this research include National Medical Research Council , Lien Centre for Palliative Care. The news correspondents obtained a quote from the research from National Cancer Centre Singapore, “However, gaps in literature include incomplete reporting of m odel performance, inadequate alignment of model formulation with implementation use-case, and insufficient explainability hindering trust and adoption in clinic al settings. Hence, we aim to develop an explainable machine learning EHR-based model that prompts palliative care processes by predicting for 365-day mortality risk among patients with advanced cancer within an outpatient setting. Our coho rt consisted of 5,926 adults diagnosed with Stage 3 or 4 solid organ cancer betw een July 1, 2017, and June 30, 2020 and receiving ambulatory cancer care within a tertiary center. The classification problem was modelled using Extreme Gradien t Boosting (XGBoost) and aligned to our envisioned use-case: ‘Given a prediction point that corresponds to an outpatient cancer encounter, predict for mortality within 365-days from prediction point, using EHR data up to 365-days prior.’ Th e model was trained with 75% of the dataset (n = 39,416 outpatient encounters) and validated on a 25% hold-out dataset (n = 13,122 o utpatient encounters). To explain model outputs, we used Shapley Additive Explan ations (SHAP) values. Clinical characteristics, laboratory tests and treatment d ata were used to train the model. Performance was evaluated using area under the receiver operating characteristic curve (AUROC) and area under the precision-re call curve (AUPRC), while model calibration was assessed using the Brier score. In total, 17,149 of the 52,538 prediction points (32.6%) had a mort ality event within the 365-day prediction window. The model demonstrated an AURO C of 0.861 (95% CI 0.856-0.867) and AUPRC of 0.771. The Brier scor e was 0.147, indicating slight overestimations of mortality risk. Explanatory di agrams utilizing SHAP values allowed visualization of feature impacts on predict ions at both the global and individual levels. Our machine learning model demons trated good discrimination and precision-recall in predicting 365-day mortality risk among individuals with advanced cancer.”
NETL
NSTL