«上一篇/Previous Article|本期目錄/Table of Contents|下一篇/Next Article»

《福建醫(yī)藥雜志》[ISSN:1002-2600/CN:35-1071/R]

卷:

45

期數:

2023年04期

頁碼:

107-111

欄目:

基礎研究

出版日期:

2023-08-15

文章信息/Info

Title:

Development and validation of a machine learning model to predict the symptomatic status of COVID-19 patients

文章編號:

1002-2600(2023)04-0107-05

作者:

朱麗萍; 王存澤¹; 王; 凌²

福建省立醫(yī)院藥學部(福州 350001)

Author(s):

ZHU Liping;  WANG Cunze;  WANG Ling

Department of Pharmacy, Fujian Provincial Hospital, Fuzhou, Fujian 350001, China

關鍵詞:

新冠病毒感染;  無癥狀;  機器學習;  GBM;  預測

Keywords:

COVID-19;  asymptomatic;  machine learning;  GBM;  prediction

分類號:

R181.8

文獻標志碼:

B

摘要:

目的 為科學合理分配醫(yī)療資源,提高救治率,探討機器學習算法用于預測新冠病毒感染后是否出現癥狀的效果。方法 回顧性收集 2022年12月至2023年2月在某三甲醫(yī)院確診為新冠病毒感染患者的臨床信息,并隨機分為訓練集(75%)和測試集(25%)。采用單因素logistic分析及最小絕對收縮和選擇算子(LASSO)算法篩選出特征變量。采用fully connected deep neural network(FCDNN)、distributed random forest(DRF)、gradient boosting machine(GBM)以及generalized linear model(GLM)4種機器學習分類器,在訓練集中進行模型的構建,并在驗證集中驗證最佳模型。采用受試者工作特征(ROC)曲線下面積(AUC)、邏輯回歸損失(Logloss)、均方根誤差(RMSE)和均方誤差(MSE)評價機器學習的模型效能。應用基尼指數評價最優(yōu)模型特征變量的重要性。結果 共251例患者納入分析,其中訓練集154例,驗證集97例。經單因素logistic分析和LASSO計算后,篩選出年齡、長期飲酒史、睡眠欠佳比率、進食欠佳比率、糖尿病患病率、高血壓患病率、其他疾病患病率、基礎用藥率、其他用藥率、呼吸頻率以及新冠病毒N基因的CT值等11個特征變量構建機器學習預測模型。4個機器學習模型中,GBM模型的AUC最高,而Logloss、RMSE、MSE最低,GBM模型在訓練集和驗證集中的 AUC 分別為 0.878 0、0.793 3。采用基尼指數評價特征變量的重要性,結果顯示變量的重要性依次為N基因CT值、年齡、患其他疾病、呼吸頻率、患高血壓或糖尿病、長期飲酒史、進食欠佳和睡眠欠佳。結論 本研究開發(fā)并驗證了一個GBM預測模型,在預測新冠病毒感染后有無癥狀上具有良好效能,能為患者后續(xù)的診療策略制定和醫(yī)療資源的分配提供重要參考。

Abstract:

Objective To scientifically and reasonably allocate medical resources,improve the treatment rate,and explore the effectiveness of machine learning algorithms in predicting symptoms after COVID-19 infection. Methods Clinical information of confirmed COVID-19 patients in a tertiary hospital from December 2022 to February 2023 was analyzed retrospectively.All patients were randomly divided into a training set(75%)and a test set(25%). Univariate logistic analysis and the least absolute shrinkage and selection operator(LASSO)algorithm were used to select feature variables. Four machine learning classifiers,including fully connected deep neural network(FCDNN),distributed random forest(DRF),gradient boosting machine(GBM),and generalized linear model(GLM),were used to construct models in the training set and validated the best model in the validation set. Receiver operating characteristic(ROC)curve area under the curve(AUC),logistic regression loss(Logloss),root mean square error(RMSE),and mean square error(MSE)were used to evaluate the performance of the machine learning models. The Gini index was used to evaluate the importance of the optimal model's feature variables. Results A total of 251 patients were included in the analysis,with 154 patients in the training set and 97 patients in the validation set. After univariate logistic analysis and LASSO calculation,11 feature variables were selected,including age,long-term alcohol drinking history,poor sleep ratio,poor eating ratio,diabetes prevalence,hypertension prevalence,other disease prevalence,baseline medication rate,other medication rate,respiratory rate,and CT value of the COVID-19 N gene. Among the four machine learning models,the GBM model had the highest AUC and the lowest Logloss,RMSE,and MSE. The AUC of the GBM model in the training set and validation set were 0.878 0 and 0.793 3,respectively. The importance of the feature variables evaluated by the Gini index was as follows: CT value of N gene,age,other diseases,respiratory rate,hypertension or diabetes,long-term alcohol drinking history,poor eating,and poor sleep. Conclusion This study developed and validated a GBM prediction model that demonstrated good performance in predicting symptoms after COVID-19 infection. It can provide important reference for subsequent diagnosis and treatment strategies as well as the allocation of medical resources for patients.

備注/Memo

備注/Memo:

基金項目:福建省自然科學基金面上項目(2020J011094)
1 福建醫(yī)科大學藥學院; 2 通信作者,Email:[email protected]

常用功能

更新日期/Last Update: 2023-08-15