社區老年人日常活動功能與醫療服務利用之探討

研究目的：目前國內大型社區追蹤研究在執行上有許多困難，且很少研究同時探討橫斷及縱貫日常活動功能及醫療利用之情形。故本研究主要透過橫斷及縱貫分析，建構日常活動功能障礙及醫療利用模式，並嘗試以資料探勘的方式，驗證與統計分析的結果是否相同，建立變項之間的關聯性，提供另外一種不同的資訊，供未來研究做進一步探討。研究方法：本研究來自邱亨嘉等人於1999年，由國家衛生研究院與高雄醫學大學合作執行之老化貫時性研究計畫，係利用2001年EPESE城鄉老人追蹤調查資料庫。研究樣本來自2002年高雄縣市社區老人共1,858人，兩年後共1,414人完成第二波追蹤調查，不含死亡之追蹤率為82.3%。研究設計分為橫斷及縱貫研究，此外利用資料探勘方法中的約略集合理論(Rough Set Theory, RST)，驗證資料探勘方法與統計分析所做出來之相關因子是否相同，並建構橫斷分析之PADL障礙及兩年後急診使用之關聯規則，提供變項之間的關聯性作為參考。研究結果：在橫斷分析方面，預測PADL及IADL障礙之因子為年齡愈高、髖關節骨折、中風、自覺健康愈差及憂鬱。控制人口學特質及其他相關因素後，影響老人醫療服務利用之影響因素仍然以自身的慢性疾病、自覺健康為主，尤其在急診及住院使用方面，心臟病、糖尿病均具有顯著預測力；在門診使用上都市地區、心臟病、高血壓、糖尿病、睡眠品質均是顯著影響因素。自覺健康對於醫療服務利用是一很強的預測因子，自覺健康狀況愈差急診、住院、門診使用均會愈多(OR=5.66, p<0.001)。在縱貫分析方面，多分類複對數迴歸(multinomial logistic regression)結果顯示，控制人口學特質及其他相關因素後，相較於ADL均無衰退或有進步者，預測兩年後ADL衰退之影響因子為，社經地位愈低、社會支持愈不滿意、及慢性疾病。預測入住長照機構之影響因子為，社會支持愈不滿意及自覺健康愈差。預測死亡之影響因子為，社經地位愈低、慢性疾病、自覺健康愈差、及心理健康。在兩年後醫療服務利用中，慢性疾病、ADL衰退、基期醫療服務利用能顯著預測兩年後的醫療使用。資料探勘實證分析結果方面，由統計分析所選擇出來的相關因子，在資料探勘中同樣為核心屬性(即不可或缺的變項)，顯示兩種工具所分析出來的重要相關因子結果一致。結論：在橫斷分析中，PADL及IADL障礙主要受到身體及心理健康影響，在醫療服務利用方面，主要受到本身慢性疾病及較差的自覺健康影響。自覺健康愈差對ADL障礙及醫療服務利用均為重要危險因子。在縱貫分析中，老人社會、身體、心理健康能顯著預測兩年後ADL變化或健康惡化，其中憂鬱甚至能顯著預測兩年後死亡情形。此外研究也發現，高教育程度為兩年後ADL衰退與死亡之重要保護因子。ADL衰退為兩年後醫療服務利用之重要危險因子，而基期醫療服務利用容易造成兩年後重複使用的情況，尤其以門診重複利用情況最多。全民健保實施後大大提昇民眾醫療可近性，但城鄉之間的醫療利用仍有差異，以都市地區利用情形較高，可能由於城鄉之間存在的地理、經濟、社會環境及醫療資源分配不均等有關。最後，透過資料探勘所建出來的規則，可瞭解變項之間的關聯性，亦能提供資訊人員建立決策支援系統，未來以衛教及預防醫學的角度，提供一套評估的工具，進而給予適當治療或介入，以減少老人進入功能衰退、入住長照機構、死亡及降低醫療服務利用。

關鍵字

日常活動功能；醫療服務利用；資料探勘

並列摘要

Objective: To investigate the cross-sectional and longitudinal study of the activity of daily life and health service use in the community elderly. Furthermore, we try to test the results between Data-mining and statistics analysis. We can get more information about the relationship between the factors and provide another concept for further researches. Methods: The data came from the project of panel study for the aging, which was executed by the National Health Research Institutes and Kaohsiung Medical University in 1999. We take the urban-rural survey of the EPESE in Kaohsiung in 2001 for study. Study samples were 1,858 in 2002 and 1,414 followed-up after two years. Excluding death, the rate of follow-up was 82.3%. We can divide the study into two sections: the cross-section and longitudinal one. We also use the Rough Set Theory (RST) of Data-mining to compare the outcome with the statistic analysis. Results: In the cross-sectional study, older, hip fracture, stroke, worse self-report health and depression can predict the disability of PADL and IADL. After controlling demography and other factors, chronic diseases and self-report health can affect the health service use. Especially, the cardiovascular disease and diabetes can predict the use of the emergency department visit and the hospitalization. For the outpatient department visit, urban, cardiovascular disease, hypertension, diabetes and quality of sleep are significant factors. The self-report health is a strong predict factor for the health service use. When the self-report health is worse, it has more opportunities of the emergency, hospitalization and outpatient department visit. (OR=5.66, p<0.001) In the longitudinal study, the results of the multinomial logistic regression show that after controlling demography and other factors, compare with one has ADL improve or hasn’t ADL decline, lower socio-economic, worse social support and chronic disease can predict the two-year ADL decline. Worse social support and self-report health can predict the two-year long-term care. Lower socio-economic, chronic diseases, worse self-report health and mental health can predict the two-year death. Finally, chronic diseases, ADL decline and health service use at baseline can predict the two-year health service use. For the Data-mining, the relational factors selected by the statistics analysis are the same as the Data-mining. It means that the results from the two analysis methods are the same. Conclusion: In the cross-sectional study, the disability of PADL and IADL are affected by the physical and mental health. For the health service use, it is affected by chronic diseases and worse self-report health. Worse self-report health is an important risk factor for the ADL disability and the health service use. In the longitudinal study, social, physical and mental health can predict two-year ADL or health change. The depression even can predict two-year death. Additionally, we find that higher education is an important protective factor for two-year ADL decline and death. The ADL decline and health service use at baseline can predict two-year health service use. The National Health Insurance improves the accessibility of the medical care in Taiwan, but we find that there are still differences between urban and rural. The urban elderly has much health service use, which maybe due to the differences of the geography, economic, social environment and medical resources between the urban and the rural. Finally, the rules established by Data-mining provide the relationship between factors. It also can establish the decision system and make a point of view of patient instructions and the preventive medicine for assessing and providing appropriate treatments for elderly. In order to reduce entering functional decline, long-term care, death and health service use.