基層醫療診所病患流失之預警－倒傳遞神經網路之運用

Patient Churn Forecast in Primary Health Care -using a Back Propagation Network System

范牧蘭(Mu-Lan Fan) ； 楊瓊珠(Chiung-Chu Yang) ； 劉崇祥(Chung-Hsiang Liu) ； 趙崇翎(Chang-Ling Chao)

《醫務管理期刊》 9卷4期 (2008/12) Pp. 286-298

https://doi.org/10.6174/JHM2008.9(4).286

摘要

目的：本研究是以資料探勘技術之人工類神經網路(Artificial Neural Networks, ANNs)建立顧客關係管理議題中病患流失分析模式，以及病患組成結構及流失特徵，並比較羅吉斯迴歸分析方法，針對可能流失病患給予醫療機構「預警」，提供適切的行銷策略以供決策者參考。
方法：是以基層醫療診所之門診掛號病患資料庫之基本資料進行分析，利用人工類神經網路之倒傳遞神經網路(Back-Propagation Network; BPN)方法，進行資料之前置處理、「探勘」作業，及網路模型建置。
結果：研究發現1.羅吉斯迴歸分析中，只有年齡變數具有顯著性，其餘變數均呈現不顯著；測試樣本的整體正確判別率：66.2%；2.倒傳遞網路結果分析：正確判別率爲71.30%，相關性分析中除「病思身份別」變數呈現負相關外其餘均呈現正相關；敏感度係數分析：是以「病患性別」變數影響程度爲最重要，其中又以男性病思流失最多。了解變數之影響程度後，管理者可依流失最多者爲優先管理之重點。
結論：兩種預警模式之比較在整體正確判別率是以倒傳遞類神經網路模式優於羅吉斯廻歸模式。本研究所提出之倒傳遞神經網路模式確實能有效找出潛在流失病患並具實務應用價值。

關鍵字

顧客關係管理 ； 病患流失 ； 資料探勘 ； 倒傳遞神經網路

並列摘要

Objectives: This study compares a data mining technique, the Artificial Neural Networks (ANNs) method, with Logistic Regression in order to analyze patient churn patterns and the characteristics of patient construction in primary health care patients. It is aimed at establishing a forecast model for the prediction of possible patient churn patterns in primary health care clinics, and could provide pertinent information to enable the development of marketing strategies, serving as an important reference for decision makers.
Methods: Through analysis of the patient database of the out-patient service in a primary health care clinic using an ANNs method, the Back-Propagation Network (BPN) method, data were used in preprocessing and data mining, and the network model establishes.
Results: The findings of this study include: (1) Logistic Regression analysis-of the variables used, only age shows a significant difference, and the correct classification rate was 66.2%. (2) BPN analysis-the correct classification rate was up to 71.3%. Correlation analysis shows that only the patient category parameter shows negative correlation; all other parameters show positive correlation. Patient’s gender was shown to be the most important variable in sensitivity analysis, and the male gender has a higher churn rate. After the influence of each variable has been indentified, primary health care managers are then able to understand which factors should be made priority.
Conclusion: Comparison of the two forecast models shows that BPN has a higher correct classification rate on the whole than has Logistic Regression. This study demonstrates that the BPN model can be used to identify the potential churn of patients and is important in practical applications.

並列關鍵字

Customer Relationship Management ； Patient Churn ； Data Mining ； Back-Propagation Network

參考文獻

黃隆虎(2006)。醫療社團法人對基層醫療院所競爭力之影響－以透析基層診所爲例(碩士論文)。逢甲大學。

Davids M.(1999).How to avoid the 10 Biggest Mistake in CRM.Journal of Business Strategy.22-26.

Reichheld Frederick F.,W. Earl Sasser(1990).Zero Defection: Quality Comes to Services.Harvard Business Review.68(5),105-111.

Reichheld Frederick F,Thomas Teal.(1996).The Loyalty Effect.Boston:Harvard Business School Press.

Fayyad Usama M （editors）,Piatetsky-Shapiro G （editors）,Smyth P （editors）,Uthurusamy R. （editors）(1996).Advances in knowledge discovery and data mining.Menlo Park, CA:AAAI Press/MIT Press.