应用人工神经网络建立老年肺癌患者化疗期间发生感染的预测模型

目的:建立老年(>65岁)肺癌患者住院化疗期间发生感染的预测模型,评价模型的预测性能,探讨个体发病风险预测的方法。方法:收集本院收治的老年肺癌化疗患者327例,按4∶1分为训练组(n=260)与检验组(n=67),分别用于筛选变量、建立预测模型及对模型的检测和评价。应用logistic回归Logistic Regression(LR)模型对资料进行多因素筛选,将筛选出有统计学意义的指标建立LR模型和人工智能神经网络(artificial neural network,ANN)模型,利用受试者工作特征曲线(receiver operator characteristic,ROC)评价模型的预测性能。结果:医院感染的发生率为21.10%(69/327),进一步分析发现骨髓抑制(≥Ⅱ度,P=0.001)、长时间住院(≥14 d,P=0.001)、糖尿病(P=0.005)、联合化疗(含顺铂,P=0.009),肺癌手术史(P=0.033)以及应用激素(P=0.025)是老年肺癌患者化疗期间发生感染的危险因素。利用两种模型对67名检验集预测,预测正确率分别为87.15%和79.81%;通过ROC曲线下面积比较模型的预测能力:LR模型曲线下面积(0.757±0.036)小于ANN模型(0.905±0.017)。结论:在老年肺癌患者住院化疗期间发生感染的预测方面ANN模型预测性能优于LR模型。

Application of artificial neural network in establishing predictive models with infection during chemotherapy for hospitalized elderly patients with lung cancer

Objective:To establish predictive models with infection during chemotherapy for hospitalized elderly patients (＞ 65years old) with lung cancer and then evaluate the predictive capacity of the models,so as to explore methods to predicate the disease onset risks in individuals.Methods:A total of 327 elderly patients with lung cancer admitted in our hospital for chemotherapy were collected retrospectively,and divided into training group (n =260) and detection group (n =67) by ratio of 4 ∶ 1 to screen variables,establish predictive models,and detect and evaluate those models.Logistic Regression (LR) models were used to conduct multivariate screening to the data,and screened indexes with statistical significance were used to establish LR models and Artificial Neural Network (ANN) models.Receiver Operator Characteristic CUI-Ve (ROC) was used to evaluate the prediction performance of these models.Results:The occurrence rate of nosocomial infection was 21.10％ (69/327),and further analysis indicated that myelosuppression (≥ degree Ⅱ,P =0.001),long length of hospital stay (≥ 14 days,P =0.001),diabetes (P =0.005),combination with chemotherapy (containing cisplatin,P =0.009),history of surgery for lung cancer (P =0.033) and application of hormones (P =0.025) were risk factors for the occurrence of infection during chemotherapy for elderly patients with lung cancer.The two models were used to predict the 67 validation sets,with rates of accurate prediction being 87.15％ and 79.81％,respectively.ROC area under the curve (AUC) was used to compare the prediction capacity of the models,and the results showed that the AUC of LR models was (0.757 ± 0.036),smaller than the (0.905 ± 0.017) of ANN models.Conclusion:ANN models have better prediction performance than LR models in predic ting the occurrence of infection during chemotherapy for elderly patients with lung cancer.