Evaluative Epidemiology Unit, Department of Epidemiology and Data Science, Fondazione IRCCS Istituto Nazionale dei Tumori, Milan, Italy.
National Centre for Healthcare Research and Pharmacoepidemiology, University of Milano-Bicocca, Milan, Italy.
Cancer Med. 2023 Apr;12(8):9849-9856. doi: 10.1002/cam4.5540. Epub 2022 Dec 20.
A strong relationship has been observed between comorbidities and the risk of severe/fatal COVID-19 manifestations, but no score is available to evaluate their association in cancer patients. To make up for this lacuna, we aimed to develop a comorbidity score for cancer patients, based on the Lombardy Region healthcare databases.
We used hospital discharge records to identify patients with a new diagnosis of solid cancer between February and December 2019; 61 comorbidities were retrieved within 2 years before cancer diagnosis. This cohort was split into training and validation sets. In the training set, we used a LASSO-logistic model to identify comorbidities associated with the risk of developing a severe/fatal form of COVID-19 during the first pandemic wave (March-May 2020). We used a logistic model to estimate comorbidity score weights and then we divided the score into five classes (<=-1, 0, 1, 2-4, >=5). In the validation set, we assessed score performance by areas under the receiver operating characteristic curve (AUC) and calibration plots. We repeated the process on second pandemic wave (October-December 2020) data.
We identified 55,425 patients with an incident solid cancer. We selected 21 comorbidities as independent predictors. The first four score classes showed similar probability of experiencing the outcome (0.2% to 0.5%), while the last showed a probability equal to 5.8%. The score performed well in both the first and second pandemic waves: AUC 0.85 and 0.82, respectively. Our results were robust for major cancer sites too (i.e., colorectal, lung, female breast, and prostate).
We developed a high performance comorbidity score for cancer patients and COVID-19. Being based on administrative databases, this score will be useful for adjusting for comorbidity confounding in epidemiological studies on COVID-19 and cancer impact.
在合并症和 COVID-19 严重/致命表现风险之间观察到很强的相关性，但尚无评分可用于评估癌症患者的相关性。为了弥补这一空白，我们旨在基于伦巴第地区医疗保健数据库为癌症患者开发一种合并症评分。
我们使用住院记录来识别 2019 年 2 月至 12 月期间新发实体癌的患者；在癌症诊断前 2 年内检索到 61 种合并症。该队列分为训练集和验证集。在训练集中，我们使用 LASSO-逻辑模型来识别与在第一次大流行浪潮（2020 年 3 月至 5 月）期间发展为严重/致命 COVID-19 形式相关的合并症。我们使用逻辑模型估计合并症评分权重，然后将评分分为五类（<=-1、0、1、2-4、>=5）。在验证集中，我们通过接收者操作特征曲线（AUC）和校准图评估评分性能。我们在第二次大流行浪潮（2020 年 10 月至 12 月）的数据上重复了该过程。
我们确定了 55425 名患有新发实体癌的患者。我们选择了 21 种合并症作为独立预测因子。前四个评分类别显示出相似的经历结局的概率（0.2%至 0.5%），而最后一个类别则为 5.8%。该评分在第一次和第二次大流行浪潮中表现良好：AUC 分别为 0.85 和 0.82。我们的结果对于主要癌症部位（即结直肠癌、肺癌、女性乳腺癌和前列腺癌）也是稳健的。
我们为癌症患者和 COVID-19 开发了一种高性能的合并症评分。该评分基于行政数据库，将有助于调整 COVID-19 和癌症影响的流行病学研究中的合并症混杂因素。
