根據衛生福利部歷年癌症死因統計，「氣管、支氣管和肺癌」長年位居癌症死因首位。另一方面，台灣鄉鎮層級的肺癌死亡率資料存在著高度複雜的統計特性，本文死亡率資料以比例形式呈現，其數值範圍限制於 [0, 1]，且資料在下界 0 處存在大量觀測集中之現象，形成不可忽視的零值膨脹（Zero Inflation）問題。此外傳統時空回歸模型在處理鄉鎮級長期時空資料時，可能面臨較高的運算負擔，若進一步採用以常態誤差為主的觀測假設，則難以同時維持預測準確度並有效描述邊界值大量集中之非典型分布結構。

本研究以貝氏神經場（BayesNF）建構具膨脹結構之混合模型。首先針對比例型死亡率資料，本研究透過特定轉換方式將其映射至較適合常態建模的連續空間，使模型能以較彈性的函數形式捕捉資料中隱含的非線性時空關係。其次針對點膨脹現象，本文設計了具膨脹結構之觀測概似函數，使模型能有效將膨脹觀測值與非膨脹連續觀測分開描述，避免邊界集中現象干擾整體死亡率趨勢之估計。最後本研究將 BayesNF 所萃取之時空特徵與其他協變量納入線性迴歸架構，藉由迴歸模型評估各協變量與死亡率之間的關聯，並進一步說明時空效應在死亡率變異解釋中所扮演的角色。

Lung cancer consistently ranks as the leading cause of cancer-related deaths in Taiwan. However, analyzing township-level lung cancer mortality data poses significant statistical challenges, as the data are proportions bounded in [0, 1] with severe zero-inflation. Traditional spatiotemporal models often struggle with heavy computational burdens and fail to accurately fit this atypical distribution.

To address these issues, this study proposes a zero-inflated mixed model based on Bayesian Neural Fields (BayesNF). First, the proportional data is transformed into a continuous space to capture non-linear spatiotemporal dynamics. Second, a zero-inflated likelihood function is applied to effectively separate excess zeros from continuous observations. Finally, spatiotemporal features extracted by BayesNF are integrated with covariates into a linear regression framework to evaluate their associations with mortality rates and interpret the underlying spatiotemporal effects.

L. Bernardinelli, D. G. Clayton, C. Pascutto, C. Montomoli, M. Ghislandi, and M. Songini. Bayesian analysis of space–time variation in disease risk. Statistics in Medicine, 14(21–22):2433–2443, 1995. doi: 10.1002/sim.4780142112.
David M. Blei, Alp Kucukelbir, and Jon D. McAuliffe. Variational Inference: A Review for Statisticians. Journal of the American Statistical Association, 112(518):859–877, 2017. doi: 10.1080/01621459.2017.1285773.
George E. P. Box, Gwilym M. Jenkins, Gregory C. Reinsel, and Greta M. Ljung. Time Series Analysis: Forecasting and Control. John Wiley & Sons, Hoboken, NJ, 5 edition, 2015.
Noel Cressie and Christopher K. Wikle. Statistics for Spatio-Temporal Data. Wiley Series in Probability and Statistics. John Wiley & Sons, 2011.
Diederik P. Kingma and Max Welling. Auto-Encoding Variational Bayes. In International Conference on Learning Representations, 2014.
Leonhard Knorr-Held and Günter Raßer. Bayesian detection of clusters and discontinuities in disease maps. Biometrics, 56(1):13–21, 2000. doi: 10.1111/j.0006-341X.2000.00013.x.
Hsin-I Liu, Chun-Ju Chiang, Shih-Yung Su, Jing-Rong Jhuang, Dai-Rong Tsai, Ya-Wen Yang, Li-Ju Lin, Yu-Chen Wang, and Wen-Chung Lee. Incidence trends and spatial distributions of lung adenocarcinoma and squamous cell carcinoma in taiwan. Scientific Reports, 13:1655, 2023. doi: 10.1038/s41598-023-28253-4.
Peter McCullagh and John A. Nelder. Generalized Linear Models. Chapman and Hall, London, 2 edition, 1989.
Victoire Michal, Alexandra M. Schmidt, Laís Picinini Freitas, and Oswaldo Gonçalves Cruz. A bayesian hierarchical model for disease mapping that accounts for scaling and heavy-tailed latent effects. Statistical Methods in Medical Research, 34(2):307–321, 2025. doi: 10.1177/09622802241293776.
Radford M. Neal. Bayesian Learning for Neural Networks. PhD thesis, University of Toronto, 1996.
Carl Edward Rasmussen and Christopher K. I. Williams. Gaussian Processes for Machine Learning. Adaptive Computation and Machine Learning. MIT Press, Cambridge, MA, 2006.
Feras Saad, Jacob Burnim, Colin Carroll, Brian Patton, Urs Köster, Rif A. Saurous, and Matthew Hoffman. Scalable spatiotemporal prediction with bayesian neural fields. Nature Communications, 15:7942, 2024. doi: 10.1038/s41467-024-51477-5.
衛生福利部國民健康署. 肺癌危險因子. 衛生福利部國民健康署衛生教育網頁, 2026. URL https://www.hpa.gov.tw/Pages/List.aspx?nodeid=4053.
衛生福利部統計處. 114年死因統計結果新聞稿 (含簡報). 衛生福利部新聞稿與簡報, 6 2026. URL https://www.mohw.gov.tw/dl-100896-73bd5dff-71ba-439d-8ee3-87ebf2789912.html.