本文目的是以統計模式之建立為出發點，對每天之平均及累積住院費用進行分析，探討腦中風病患向急性醫院「轉診」至長期照護機構之臨界日。在病患住院初期，進行大部份的檢驗及急性治療，住院費用最高、待病情較穩定，不需要急性醫院的醫療設備後，費用便會下降。因此，基於病情能反映於費用上的假設，對住院費用及住院天數進行兩段迴歸分析，前後兩段之迴歸線斜率差異為最大、或前後兩段平均費用差距最大，便可視為轉診的臨界日。除此之外亦可以累積住院費用達某百分比（如75%，80%或90%）時作為切點討論。

A few statistical models were constructed to investigate the most appropriate time to refer hospitalized CVA patients from a medical center to long term care institutes. Based on the assumption that patients' physical conditions can be reflected in their hospital fees, the cumulative charges were examined by a two-piece regression. It is perhaps most efficient, from hospital management's point of view, to transfer patients at the time where the difference between the slopes of the regression lines before and after this point is maximized. Alternatively, one can also maximize the difference in average daily charges before and after such point, or when the cumulative charges reach, say, 75%, 80% or 90%. The studied sample consisted of 1357 CVA inpatients randomly drawn from a medical center in northern Taiwan. Their average length of stay (LOS) was 22.9 days, with a mean hospital fee of NT$76325. Results of the piece-wise regression analysis show that most patients can be referred in 3 to 5 days. On the other hand, we also found that 75%-90% of the total fee was accumulated in the first 12-13 days. Should patients be transferred within the derived days, operating efficiency can increase by 12%-34%. Since only fees are used in the modeling process as the dependent variable, but physical conditions and outcomes of the patients have been largely ignored, results of this research can by no means be considered complete. Yet this still seems to be one of the pioneer projects investigating ”referral date”, instead of the reasonably well studied ”length of stay”. Therefore, it is suggested that further research be carried out from a more medical point of view, so that the combined results will be more comprehensive.

中風 ； 轉診 ； 臨界日 ； 長期照護 ； 統計模式

CVA, referral day, long term care, statistical model