- 學位論文
疾病率長期趨勢之年齡、年代、世代效應分析及未來推估
Long-term Trends in Disease Rates: Age-Period-Cohort Analysis and Future Projection
摘要
發生率與死亡率的年齡、年代及世代分析方法，一直是研究者與流行病學家關心的議題。年齡、年代及世代分析方法在求得三個時序效應與未來預測上皆有持續發展，並廣泛應用在各項疾病與健康事件中。年齡、年代及世代模式會遭受不可甄別問題，研究者只能引入額外的假設與條件限制式來求得特定一組估計值，然而此問題不會影響年齡、年代及世代模式進行未來預測的分析。在本論文中，我們提出新的公平離差率限制式。此限制式反映了三時序因子的不同性質，假設年齡效應為決定性機制，並假設年代與世代效應為隨機性機制。我們以各式模擬情境並以美國白人前列腺癌發生率為實例應用測試此方法，展示了良好的統計特性與合理的分析結果。此方法能自動偵測年代或世代效應其中之一或兩者為虛無的情境，並提供正確的結果。公平離差率法並非年齡、年代及世代分析的唯一解，我們建議研究者應當檢視資料本身的特性是否符合上述的兩項假設，才能決定是否應用公平離差率法。另一方面，我們也提出新的年齡、年代及世代分析之系統化預測程序。此方法統整了過去多個年齡、年代及世代分析預測方法，並以交叉驗證的程序由眾多年齡、年代及世代分析預測模式中找出最佳預測模式以進行未來預測。我們採用1997年到2014年間的台灣肝癌發生率資料進行分析並預測到2035年。此方法在實例應用中，展示了極具公共衛生價值性的預測結果。
並列摘要
Age-period-cohort analysis of incidence and/or mortality data has received much attention in the literature for identifying the three temporal effects and predicting the trend into the future, respectively. To circumvent the non-identifiability problem inherent in the age-period-cohort model, additional constraints are necessary on the parameters estimates. However, this problem does not hamper an age-period-cohort prediction. In this paper, we propose a constant-relative-variation constraint to reflect the different nature of the three temporal variables. This constraint assumes the age effects to be deterministic, and the period and cohort effects to be stochastic. We conducted Monte-Carlo simulations and analyzed the data of prostate cancer incidence for whites Americans from 1973-2012 and demonstrated the desirable statistical properties and reasonable results of our method. The proposed method can automatically produce an unbiased age effect and no period and/or cohort effect when a driver for the period and/or cohort effect is lacking in populations. If the assumptions we make are deemed reasonable for the study context, researchers can use our method for an apportionment of period and cohort slopes. We also proposed a systematic age-period-cohort prediction procedure, which integrated many prediction methods from previous studies. The cross-validation process was used to examine an ensemble of age-period-cohort models and pick up one best model for predicting the trend into the future. We used the liver cancer incidence data in Taiwan from 1997 to 2014 for application and predicted the rates to the years of 2035. This procedure was demonstrated for an appropriate prediction results and of contribution to public health.