制定篩檢政策時，常面臨篩檢頻率究竟應該多長的問題。一般而言，疾病的臨床症前可偵測期(PCDP, Preclinical Detectable Phase)愈短、疾病病程愈快者，其篩檢頻率應該要愈頻繁。因此如何估計臨床症前期之平均長短(亦稱Mean Sojourn Time, MST)，對提供篩檢政策的擬定扮演著很重要之角色。本研究自1995年至1996年10月止利用臺灣地區12家大型醫院，鎖定出2629名高危險群婦女接受乳癌篩檢，第一年篩檢到的31名病例和第二年篩檢的3名病例之實證資料，推估得婦女乳癌之臨床症前期發生率(preclinical incidence rate)為每年每千人5.7人(0.0057/年)。依據這些 資料，利用模擬追蹤六年之技術，以馬可夫三階段前進模式及五階段模式進而推估病理分期之間隔。結果發現婦女乳癌臨床症前期平均長短約為1.90年(95% CI:1.18-4.86)。根據乳癌之病理分期，第一期者由臨床症前期至臨床期約為8年，第二期及以上者則約為3個月。應用此結果模擬不同篩檢間隔之成效發現篩檢間隔個案(interval cases)之比例隨間隔增加而增加(1年間隔為22.8%；3年間隔為50%)，病理分期第二期及以上者之比例亦有此趨勢(3年間隔為6.5%；1年間隔為3.1%)。配合存活資料可得現行一年之篩檢間隔政策可以降低乳癌死亡率36% (RR=0.64, 95% CI=0.32-0.97) 。本研究所發展之馬可夫鏈模式(Markov chain model)亦可應用於缺乏篩檢間隔發現個案(interval cases)之其他篩檢計畫評估。

　　　　　The optimal screening frequency is highly dependent on the duration of pre-clinical detectable phase (PCDP, also called sojourn time). This parameter is difficult to estimate partly because the progression from PCDP to clinical phase is unobservable and partly because data on interval cases is hardly available from screening project. To tackle these problems, 2629 women of high risk group aged 35 and above identified until October 1996 from 12 large hospitals in Taiwan received their first screening exams, and 31 individuals were detected with breast cancer. Among 575 women who had returned for the second year screening exam, three persons were found with positive results. The progress intervals between stages of the disease were estimated using left-censored and interval-censored Markov chain models with a 6-year follow-up simulation. Results showed an annual pre-clinical incidence rate of 5.7 per 1000 for the high-risk group. The mean sojourn time (MST) was 1.90 years (95% CI=1.18-4.86). The proportion of interval case increased with the screening interval. A similar situation was also observed for the proportion of stage II+. Applying the Swedish Two-County trial experience, the one-year screening regimen would be able to reduce the mortality from breast cancers for 36% (RR=0.64, 95% CI=0.32-0.97). The breast cancer screening policy for high risk group initiated by department of health is justified by a high pre-clinical incidence rate estimated in this study. According to the estimated MST and the relati onship between screening interval and the proportion of both interval case and mortality reduction, it is advisable that the screening interval for this high-risk group be no longer than two years. Finally, a left-censored and interval-censored Markov chain developed in this study could be applied to other screening projects short of data on interval cases.