建立台灣居民缺血性中風復發風險之預測分數

Development of Risk Score for Ischemic Stroke Recurrence in Taiwan

王芝淇(Chih- Chi Wang)

指導教授 ： 邱弘毅

臺北醫學大學/公共衛生暨營養學院/公共衛生學系暨研究所/碩士(2014年)

摘要

中風病人即使經過急性治療後，日後仍有復發的機會，往往造成病人和家屬沉重的負擔；中風復發不但會增加日後殘疾及認知功能下降之發生，亦會增加住院時間、神經功能受損及死亡的風險。造成中風病人復發的可能因素與機制，不僅複雜且原因很多，因此，一個容易評估且具高正確性的預測模式，能夠協助醫師評估病人是否為復發高風險族群，並給予適當且正確的治療及衛教，以達到次級預防的效果。直至目前為止，已發展的中風復發風險預測分數，因為統計上預測效果較差，或使用上具其困難度，而無法廣泛於臨床使用。過去以亞洲族群為主的研究亦相當有限，且尚未有研究針對台灣中風病人建立風險預測模式，因此，本研究將以回溯性世代追蹤研究，針對缺血性中風病人，建立中風復發風險預測分數。
本研究資料來源為台灣中風登錄資料庫 (Taiwan Stroke Registry, TSR) 及全民健康保險資料庫 (National Health Insurance database, NHI)，以新發缺血性中風病人為研究族群，追蹤其發病後一年內的復發事件。符合條件之研究對象，隨機分為模式估計組及效度檢定組，前者用於預測模式的建立及分數和預測風險的估計；後者用於進行內部效度檢定。首先針對模式估計組，以Cox proportional hazard model分析基本人口學、危險因子及中風相關臨床因子與一年內中風復發之相關性；及Purposeful selection及文獻參考作模式選擇；最後，使用Points system估計最終模式之預測分數及預測風險。針對效度檢定組，以適合度檢定評估預測分數的準確性；另以接受器操作特徵曲線底線面積 (Area Under Receiver Operating Characteristic, AUROC) 評估其分立性。最後，以Kaplan- Meier survival及Log- rank test檢定低度、中度及高度風險族群一年內復發風險之差異性。
結果顯示風險預測分數包括五個危險因子：糖尿病、心房顫動、慢性腎臟病、TOAST分類為非小血管阻塞及頭痛症狀。分數介於0至18分，其對應之預測風險由0分的8.82%及1分的9.68%，增加至14分以上的風險大於30.38%。預測風險與觀察發生率之相關係數於模式估計組及效度檢定組，分別為0.98788 (P<.0001) 及0.98333 (P<.0001)；兩組的AUROC則分別為0.564及0.580。相較於低度風險族群，中度及高度風險族群在模式估計組 (HR中= 1.35, 95%C.I.= 1.20- 1.52; HR高= 3.13, 95%C.I.= 2.41- 4.06) 及效度檢定組 (HR中= 1.55, 95%C.I.= 1.22- 1.98; HR高= 2.73, 95%C.I.= 1.52- 4.92)，皆得到顯著較高的復發相對風險，且趨勢檢定達統計上顯著意義。
本研究建立一個介於0至18分的風險預測分數，其具有良好準確性，以五個危險因子，預測缺血性中風病人一年內復發風險。此預測分數可讓醫師於臨床接收到病人時，能夠快速評估其危險因子狀況，並容易計算預測分數，以瞭解病人一年內復發的風險，給予適當且正確治療或衛教，達到次級預防的效果。

關鍵字

缺血性中風 ； 中風復發 ； 風險預測 ； 台灣中風登錄資料庫 ； 全民健康保險資料庫

並列摘要

Stroke patients who received the treatment in acute phase still have the chance to be recurrent. Stroke recurrence increases the probability of disability, cognitive decline, longer hospital days, neurology damage and mortality. The risk factors and mechanism of stroke recurrence are multiple. In order to give stroke patients property and accurate treatment and health teaching to prevent risk of recurrent stroke, doctors need a prediction model with high calibration to identify the high risk patients. Up to now, the available models for recurrent stroke are not used in clinical practice, due to poor prediction ability and hard measurement. There were limited studies focusing on Asian population and there also not a stroke recurrence prediction model for Taiwanese. So, we are going to conduct a retrospective cohort study to develop an ischemic stroke recurrence prediction model.
The study resource is Taiwan Stroke Registry database (TSR) and National Health Insurance database (NHI). We follow the incident ischemic stroke patients 1- year recurrent event. Then we randomize the eligible patients to two groups including model development and validation, respectively. The model development group is used to analyze the association between basic characteristics, risk factors and 1- year stroke recurrence by Cox proportional hazard model. We further calculate the predicting score and estimated risk of the model by Points system. We use the Goodness- of- fit to test the calibration for the model. In addition, Area Under Receiver Operating Characteristic (AUROC) was used to elucidate the discrimination. Final, Kaplan- Meier survival curve and Log- rank test are used to compare the cumulative recurrence rate between the lowest risk, medium risk and highest risk group.
The result is that prediction model's score range from 0 to 18. There are five risk factors in the model. The factors include diabetes, atril fibrillation, chronic kidney disease, non- SVO TOAST subtype and headache symptom. The estimated risk of 0 score is 8.82%; 1 score is 9.68%; greater than 14 score is more than 30.38%. The correlation coefficient for estimated risk and observation incidence of model development group and model validation group is 0.98788 (P<.0001) and 0.98333 (P<.0001). AUROCs are 0.564 in the development group and 0.580 in the validation group. Compared with the lowest risk group, both medium and highest risk group have the significant higher risk for stroke recurrence in the development group (HRmedium= 1.35, 95%C.I.= 1.20- 1.52; HRhigh=3.13, 95%C.I.= 2.41- 4.06) and validation group (HRmedium= 1.55, 95%C.I.= 1.22- 1.98; HRhigh= 2.73, 95%C.I.= 1.52- 4.92), and trend test are significant in two group.
We have derived and validated a simple prediction model with five risk factors to predict the risk for ischemic stroke recurrence within 1 year. The prediction model range from 0 to 18 score with good calibration. Doctors can use this model to assess the patient's situation and expect their risk of stroke recurrence. Through the accurate treatment and health teaching, doctors can help their patients to prevent the stroke recurrence.

並列關鍵字

Ischemic stroke ； Stroke recurrence ； risk prediction ； Taiwan Stroke Registry ； National Health Insurance database

參考文獻

1993. Predictors of major vascular events in patients with a transient ischemic attack or nondisabling stroke. The Dutch TIA Trial Study Group. Stroke. 24, 527-31.

Adams, H.P., Jr., Bendixen, B.H., Kappelle, L.J., Biller, J., Love, B.B., Gordon, D.L. and Marsh, E.E., 3rd, 1993. Classification of subtype of acute ischemic stroke. Definitions for use in a multicenter clinical trial. TOAST. Trial of Org 10172 in Acute Stroke Treatment. Stroke. 24, 35-41.

Aldoori, M.I. and Rahman, S.H., 1998. Smoking and stroke: a causative role. BMJ. 317, 962-963.

Alzamora, M.T., Sorribes, M., Heras, A., Vila, N., Vicheto, M., Fores, R., Sanchez-Ojanguren, J., Sancho, A. and Pera, G., 2008. Ischemic stroke incidence in Santa Coloma de Gramenet (ISISCOG), Spain. A community-based study. BMC Neurol. 8, 5.

Appelros, P., Stegmayr, B. and Terent, A., 2009. Sex differences in stroke epidemiology: a systematic review. Stroke. 40, 1082-90.