利用多重預測因子發展乳癌個案之存活預測模式

Abstract

研究背景: 雖然影響乳癌病程進展之因子在許多研究中已多所著墨，但隨著醫療科技之進展下對於乳癌之治療方法日趨進步以及乳癌早期偵測之影響下，乳癌疾病對於婦女之威脅與影響與過往相較已有相當程度之差異。 研究目的: 本研究旨在運用包含乳房攝影表徵、乳癌病灶分子生物標記，以及乳癌傳統病灶特徵發展對於乳癌病患之危險辨識方法。 材料與方法。 資料來源: 本研究以回溯式研究設計納入1996到2014年間共計2540位於瑞典法倫中央醫院診斷之侵襲性乳癌個案，收集其乳癌傳統病灶特徵、荷爾蒙受器表現等之分子特徵，以及病灶之乳房攝影表現特徵等多種類之乳癌風險預測因子。追蹤至2015年底為止，在此乳癌個案族群中，共計有251位乳癌死亡個案發生。除上述乳癌預後風險因子外，其他如治療方式與乳癌偵測模式之變項亦收集於觀察資料中。 統計方法: 本研究運用包含發生乳癌死亡事件之觀察時間以及治癒模型(cured model)發展預測乳癌存活與治癒比例之方法。本研究首先逐一評估上述乳癌預後之多重因子影響程度。對於乳房攝影表現特徵與乳癌病灶分子表現對於乳癌個案預後之影響，除了不同類別預後因子對於乳癌存活之影響，本研究亦運用多變項分析對於各因子，特別是乳房攝影表現特徵與乳癌病灶分子表現，對乳癌存活造成之影響進行整合性評估。本研究繼而運用多變項增速模型評估對於上述多重預後因子對於乳癌死亡之影響，並以估計結果繪製乳癌存活預測曲線。對於乳癌治癒機率之考量上，本研究運用貝氏治癒模型(Bayesian cured model)在考慮乳癌治癒比例之情形下評估各因子對於乳癌存活之影響並預測30年存活機率以及建構乳癌個案危險分層。 研究結果: 在考慮乳房攝影表徵對於乳癌死亡之影響下，粉狀以及碎石狀之乳房攝影表徵之乳癌死亡風險極低，表示此兩類乳房攝影表徵之過度診斷之可能較高；相對而言，管狀鈣化與結構扭曲之乳房攝影表徵則具有高乳癌死亡風險，須採用更有效與積極之治療方式。 在乳癌病灶分子標記表現對於乳癌死亡風險之影響分析方面，管腔A型與管腔B型之病灶進展以及死亡風險較低，表示具有此兩類分子標記表現之病灶其過度診斷可能較高；若病灶之分子生表現屬於基層型或三陰性表現形態，則其乳癌死亡風險較高，其治療方式亦須以較有效積極之方式為之。 對於治癒率而言，本研究結果顯示在乳癌病患中，在經過初始之療程後，治癒比例為55.7%。對於粉狀乳房攝影表現之病灶，其治癒率最高(66%)，其次為圓形乳房攝影表現(治癒率為62%)、碎石狀(58%)以及星狀(56%)。管狀鈣化與結構扭曲之乳房攝影表徵者之治癒率極低(19%)。對於乳癌病灶分子標記表現而言，管腔A型(56%)、管腔B型(46%)以及HER-2(56%)之治癒率皆較高，但基層型(29%)與三陰性(31%)之治癒率皆較低。 結論: 本研究在納入多重乳癌死亡預後因子下建構了乳癌個案危險預測模型。借助於本研究所建構之風險預測模型不僅可辨識乳癌死亡高風險者(管狀鈣化/結構扭曲以及基層型分子標記表現)對其提供密切之臨床介入與監測，亦可辨識乳癌病患中屬於低風險者(粉狀以及碎石型，管腔A型與B型之低度惡性者)，以避免對此一低風險族群進行過度之監測與治療。

Background: Prognosis of breast cancer has been well documented but epidemiological and clinical profiles have been updated due to early detection and the advent of new therapy and treatments guided by new molecular biomarkers and imaging techniques. Aims: We aimed to develop a risk prediction model based on a constellation of mammographic appearances, molecular biomarkers, and clinical tumour attributes in order to classify different risk profiles of breast cancer. Data and Methods Data Sources: A retrospective cohort was designed by enrolling 2540 patients diagnosed with invasive breast cancer at Falun Central Hospital of Dalarna County between year 1996 to 2014 with information on the three main disciplines of predictors of conventional tumour attributes, expression of hormonal receptors, and mammographic appearance. Among the study population, 251 event of breast cancer death were ascertained till the year 2015. Factors associated with breast cancer survival such as modality of therapy and detection mode were also collected. Statistical Methods: The time-to-event design in conjunction with the cured model for the building of the prediction of the breast cancer survival and the cure rate was applied. The effect of each discipline of breast cancer predictors on the risk of breast cancer death were first evaluated separately. In addition to the main effect, their interactive influence on breast cancer survival, focusing on the expression of hormonal receptors and mammographic appearance were also assessed based on the multivariate accelerated failure time model. Based on the joint results of the assessment on each discipline of predictors, a multi-disciplinary prediction model for the risk of breast cancer survival was constructed. With the consideration of both cured and survival probability, a cured model with Bayesian approach was developed to predict 30-yer survival of breast cancer to provide risk stratification of breast cancer. Results: As far as mammographic appearances are concerned, powdery and crushed had a very low rate of death, suggesting high possibility of over-diagnosis, whereas casting type and architectural distortion has a higher death rate probably requiring aggressive treatment and therapies. Regarding molecular biomarkers, luminal A and B with low grade had lower death rate suggesting high possibility of over-diagnosis, but basal-like phenotype or triple negative breast cancer had a higher death rate probably requiring aggressive treatment and therapies. We also estimated 55.7% breast cancer with completely cured after initial treatment therapy or overdiagnosis. The cured rate was highest for powdery (66%), followed by circular (62%), crushed-stone (58%), stellate (56%). The cure rate was very low for casting type and architecture distortion (19%). The cured probability for the molecular phenotype of luminal A (56%), luminal B (46%), and HER-2 (56%)were higher compared with basal phenotype (29%) and triple negative (31%). Conclusion: We developed a risk prediction model for breast cancer by using multidisciplinary factors. Such a risk prediction model is not only useful for the identification of high risk group (casting type /architecture and basal phenotype) so as to provide adequate intensive surveillance aggressive medical regime but also for the identification of low risk (powdery and crushed stone and luminal A and B with low grade) in order to avoid unnecessary surveillance and treatment and therapy.