世代追蹤研究中選擇偏差的嚴重度

Abstract

世代追蹤研究中，研究者收案時必須排除已得病及疾病前期個案。然而疾病前期者通常無法診斷，而已得病者亦不盡可被診斷出有病。此兩種未被診斷出的個案沒有排除，即造成選擇偏差。本研究進行電腦模擬，探討研究者進行世代追蹤研究時，在收案時以及收案後各追蹤時間點上選擇偏差的大小。我們由模擬結果得知選擇偏差的方向趨向虛無。我們也發現稀有疾病的選擇偏差可忽略不計，但常見疾病的選擇偏差不應被忽略（約 ）。此外，收案時若可排除疾病前期者，此選擇偏差在收案時將減小，然而無法完全被排除。且此偏差仍需要很長的追蹤時間才會漸趨於零。我們建議研究者在收案後進行疾病主動診查。疾病主動診查可大幅縮短排除選擇偏差所需的資料年數。收案後的疾病主動診查以及排除收案後第一年的資料，兩者並行應可消除掉絕大多數世代追蹤研究的選擇偏差。

In cohort studies, researchers need to exclude prevalent cases, as well as cases in the preclinical period of the disease under study. However, these two kinds of cases are often difficult to be diagnosed, and therefor cannot be excluded from the study. This will create selection bias. In this study, we perform computer simulation to examine the magnitude of selection bias at each time point after the inception of a cohort study. We found that the direction of selection bias is toward the null. We also found that the selection bias can be ignored for rare diseases but not for common diseases (the bias is approximately for common diseases). Excluding the preclinical subjects at the inception of a cohort study will decrease the magnitude of the selection bias, but cannot completely eliminate it. In addition, we found that the selection bias needs a long follow-up time to dissipate. We encourage researchers to conduct active follow-up during the follow up, which can significantly reduce the follow-up time needed for dissipation of the selection bias. Active follow-up and excluding the data of the first year of follow-up can eliminate most selection bias of a cohort study.