應用類神經網路建構戒菸預測模型

Using Artificial Neural Network to Establish the Prediction Model of Smoking Cessation

黃芝瑋(Chih-wei Huang)

指導教授 ： 李友專

臺北醫學大學/醫學科技學院/醫學資訊研究所/碩士(2012年)

https://doi.org/10.6831/TMU.2012.00109

摘要

菸害防制已成為全球醫療衛生之重要議題，由於戒菸可大幅減少吸菸者曝露於疾病下之危險因子，且可以省下大筆社會成本及醫療資源，故本研究透過我國行政院衛生署國民健康局所辦理「全國性醫療院所戒菸服務補助計畫」之個案資料內容及其於該醫院之常見慢性病史的有無，進行資料分析及預測模型建立。

本研究資料樣本為2009年1月至2010年12月兩年間接受北部地區某兩家教學醫院戒菸治療補助之18歲以上吸菸者，初診後六個月進行電話訪問追蹤個案在接受門診戒菸治療後有無戒菸成功。位於北部市郊之A醫院102名及位於北部市中心之B醫院84名吸菸者，依照三組變項分類方式分別探討樣本之個人因素、過去病史、戒菸治療藥物與戒菸成功之相關性。本研究利用A醫院吸菸者資料作為推衍組，經由類神經網路(ANN, Artificial Neural Network)及邏輯斯迴歸(LR, Logistic Regression)訓練出模型之後再利用B醫院84名吸菸者資料進行驗證，最後評估ROC曲線下面積 (AUROC, Area under ROC curve)表現結果選擇最佳預測模型。

在資料分析結果中，樣本的年齡分布在18-84歲之間，平均年齡為42.6歲，其中82.8%為男性。所有吸菸者的戒菸成功率為20.38%，與全國平均戒菸成功率相當。而模型建立的表現上，三組不同輸入變項在ANN表現分別為0.86、0.83、0.77，在LR表現分別為0.85、0.84、0.70。兩種演算法在三組模型中皆具有較佳表現的正確率及預測能力，其中ANN的表現稍優於LR且不具有統計上的顯著差異。

從研究結果中我們可以推斷此模型在臨床的運用上是可行的，此外，醫療院所可利用國民健康局所推行實施之全國性戒菸補助計畫，透過本預測模型達到更有效的管理、評估及協助吸菸者戒菸，且能持續有效地進行戒菸行為。

關鍵字

類神經網路 ； 戒菸 ； 預測

並列摘要

The Tobacco Hazards Prevention has become an important issue of global health. Quit smoking can reduce the exposure under disease risk factors and save substantial social costs and medical resources significantly.

Objectives. This study used Artificial Neural Network (ANN) to develop prediction model which can be used in providing appropriate strategy on smoking cessation.

Materials and methods. The data were collected from January 2009 to December 2010 in 2 teaching hospitals in Taipei. The samples in this study were collected for all the smokers over 18 years, who were enrolled in "Smoking Cessation Therapy Project." We used data from one hospital as the training set, and the data from the other hospital as the validation set. Three models were designated based on the use of input variables. Each model was fitted to the data using ANN and Logistic Regression (LR). The performance of each model was shown by the Area under ROC curve (AUROC).

Result. Information from 186 smokers were collected, 102 smokers from hospital A were used for training set, and 84 from hospital B as validation set. The mean age of the smokers was 42.6 years (range from 18 to 84), and 82.8% were male. The overall smoking cessation success rate was 20.38%. The AUROC of ANN for the three models were 0.86, 0.83, 0.77 and the AUROC of LR for the three models were 0.85, 0.84, 0.70 respectively.

Conclusion. Both ANN and LR can predict smoking cessation with good accuracy with an AUROC ranging from 0.86 to 0.70. The performance of ANN was slightly better than LR without statistical significance. The outcomes of this study is expected to be effective in helping physicians to give appropriate and efficient medical advice and smoking cessation services based on case status. We believe that these prediction models can be used for better management of national-level smoking cessation program.

並列關鍵字

Artificial Neural Network ； Smoking Cessation ； Prediction

參考文獻

行政院衛生署國民健康局(2002)。台灣地區臨床戒菸指南-醫師篇。台北市：行政院衛生署國民健康局。

行政院衛生署國民健康局(2010)。菸害防制年報。台北市：行政院衛生署國民健康局。

行政院衛生署國民健康局(2009)。醫療院所戒菸服務補助計畫作業須知。台北市：行政院衛生署國民健康局。

國民健康局戒菸管理中心(2011)。門診戒菸治療服務通訊報告。新北市：行政院衛生署國民健康局戒菸管理中心。

賴裕和(2010)。菸害的流行病學及其衝擊。台灣菸害防治衛教指引99：2-6。台北市：行政院衛生署國民健康局、台灣護理學會。