建立新的糖尿病分級模型

Develop a New Staging Model of Diabetes Mellitus

蔡國麟(Kwok-Lun Choi)

指導教授 ： 林文德

長榮大學/健康科學學院/醫務管理學研究所/碩士(2008年)

https://doi.org/10.6833/CJCU.2008.00015

摘要

研究背景及目的
病患疾病嚴重度在醫療過程中的變異，可用作醫療品質評估之依據，而疾病的嚴重程度的級別，能反映出病患在疾病治療前、中、後之健康狀況。為能達到資源有效運用，及提升糖尿病照護品質之監控能力，建立糖尿病分級模型是有須要。

材料及方法
本研究所建立的新分級模型包含下列資料: 1) 病患之診斷資料；2) 病患之處方； 3) 檢驗項目；4) 診療處置及手術資料。為便於比較，新分級模式以及簡易型(單以人口因子:年齡、性別作分級)之分級模式依臨床經驗均歸類為8級。自健保資料庫中，170253位在2001至2002年間有就醫記錄的病患，共截取7174位40歲以上之糖尿病患作為分級對象，分別以新、舊及簡易型三種分級平均住院模式分級，分級結果算出後，再計算其各級之個別平均年度門診醫療支出、全年天數、及其與各評核指標(需急診照護之風險、住院之風險、及在來年死亡之風險)的相關性，並以邏輯回歸計算新舊兩分級模式各等級之勝算比。然後再以C-statistic之統計方式，以為對來年需急診照護之風險、住院之風險、及在來年死亡之風險的預測。而在針對平均年度門診醫療支出及全年平均住院天數之預測上，則以predictive R2來評估。

結果
本研究以二年度需急診照護之風險、住院之風險、來年死亡之風險、2年度門診醫療支出、平均全年住院天數為嚴重度評核指標，在加入更多臨床資料後，新分級結果在上述各指標之相關性上均優於其他兩模式，在上述前三項指標為基準，在對病情嚴重度之解釋力與預測力上，新分級模式之表現較其他兩模式進步，在2001年掛急診及住院的危險預測上,新分級模式表現優於其他兩分級模式,在住院的危險預測上達76.7%；舊分級模式為75.1%，而簡易型分級模式則只有62%。在2002年掛急診及住院的危險預測上,雖然新分級模式的預測力依序也只有62.6%及67.5%,但表現仍優於其他兩分級模式。而在2002年死亡危險的預測上, 新分級模式表現亦優於其他兩分級模式,預測力達74.89%，舊分級模式亦達74.3%。但在預測病患來年之年度門診醫療支出及平均全年住院天數上，經去除極端值後，新模式只略勝於舊模式，且R2值均小於0.1，十分不理想。

結論
新分級模式在對來年急診、住院、死亡、年度門診醫療支出、以及全年住院天數之預測上略有改善，且新分級模式在與上述各指標的相關性上均優於舊型及簡易型，而兩者的效度均不錯，省時又成本低，可使用在大型糖尿病嚴重度監控及風險預測。

關鍵字

健保資料庫 ； 極端值 ； 嚴重度監控 ； 風險預測.

並列摘要

Objective
Health care quality can be measured by the changes in the severity of the disease of patients throughout the course of medical management. Disease severity scales ensure more accurate measurement of health status before, during and after medical intervention. In order to assure efficient usage of resources and promote diabetes care quality monitoring, an effective severity evaluation system is needed.

Method and Materials
Our new model includes: 1. diagnostic details. 2. prescription details. 3. laboratory checkup items being ordered. 4. interventions or operations being performed. A total of 7,174 of DM patients with age ≧40 are selected from a population of 170,253 who have had health care consultation records in the year 2001 to 2002. Claim data from the database of the Bureau of National Health Insurance in Taiwan is used. The old, the new and the simple staging model (just using the demographic criteria: age and gender for staging) are used to stage those patients. In order to make comparison easier, modification and restaging of the new staging model is performed. Bases on clinical experience, a total of eight stages are developed. The average annual out patient department (OPD) expenditures and average annual days of admission (due to DM or related disorders), the correlations between stages in different models with those parameters: risk of E/R consultation, risk of admission, and risk of death in the coming year, of each stage of the three models are calculated. The odd ratios of individual stages of three models are measured by logistic regression. The three models are then used to predict the risk of admission, risk of E/R consultation and risk of death in the coming year by C-statistic. The predictive ability of those models in average annual OPD expenditure and admission days is measured by using predictive R2.

Result
In this study, E/R consultation and admission risks, annual admission days and OPD expenditures in the years 2001 and 2002, and mortality in the coming year (2002) are used as parameters for severity evaluation. After the addition of more clinical chart details, the new staging model works better than the other two, the old and the simple model, in the correlations with the above five parameters. In the explanation and prediction of the risk of E/R consultation, risk of admission in 2001, the new model works superior to the other two that the area under the ROC curve is 67.2% and 76.7% respectively. The new model also works better than the other two models in the prediction of the risk of death in the coming year (2002) that the prediction ability is 74.9%. The old model gets 74.3% and the simple model is only 68.2%. However, after the exclusion of extreme cases, it gets only little or minimal improvement in the new model in prediction of the days of admission and the annual OPD expenditure in the coming year and the R2 value are very unsatisfied.

Conclusion
The new model gets some improvement in the ability in the prediction of risk of E/R consultation, risk of admission, risk of death; annual admission days and annual OPD expenditures at the coming year can be achieved and gets better correlations with those parameters being described above. The validity of both the old and the new model is fairly acceptable and can be applied in large scaled studies on DM severity monitoring and risk prediction. Using the two models for disease severity staging is time saving and is very low in cost.

並列關鍵字

National Health Insurance(NHI) ； Diabetes Mellitus ； co-morbidities ； Out-patient Department(OPD) ； International Code of Disease, 9th version, Clinical Modification (ICD 9–CM) ； utilization analysis ； case-mix analysis ； epidemiologic analysis ； Free For Service(FFS) ； Global Budget payment system.

參考文獻

6. Lin CC, Lai MS, Syu CY, Chang SC, Tseng FY. Accuracy of Diabetes Diagnosis in Health Insurance Claims Data in Taiwan, J Formos Med Assoc 2005; 104(3): 157-163

19. Wei JN, Chuang LM, Lin RS, Chao CL, Sung FC. Prevalance and hospitalization rates of diabetes mellitus in Taiwan, Taiwan Journal of Public Health 1996-2000, 2002; 21(3):173-180.

1. Gonnella JS, Hornbrook MC, Louis DZ. Staging of Disease A Case-Mix Measurement, JAMA 1984; 251:637-644.

2. Starfield B, Lemke KW, Bernbardt T Foldes SS, Forrest CB, Weiner JP. Comorbidity: Implications for the importance of Primary Care in ’Case’ Management, Ann Fam Med 2003; 1:8-14.

4. Charlson ME, Pompei P, Ales KL, Mackenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and Validation, J Chron Dis 1987; 40(5): 373-383.