Forecasting lung cancer burden using multicenter real-world data
Pattern Recognition Letters · 2026-12 · 210 · 8-19
DOI: 10.1016/j.patrec.2026.09.021 ↗Kurz gefasst
Fachartikel in der Zeitschrift Pattern Recognition Letters, 2026-12. Verfasst von Omer Aydin und 6 weiteren Autor:innen. Die Arbeit nennt 21 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Background – Lung cancer continues to place a significant clinical and operational burden on healthcare infrastructure and systems. While essential epidemiological data can be obtained from population-based registries, these systems often fail to provide information on long-term follow-up and healthcare utilization dynamics. Real-world data collected from healthcare institutions offer valuable insights; however, in-depth and careful analysis is necessary to distinguish true diagnostic trends from changes in service utilization and access. Objectives – The primary objective of this study is to characterize and forecast the prevalence of lung cancer diagnoses at the specified hospitals for future years. The study aims to predict prevalence for the 2026–2030 period by comparing three selected time-series forecasting methods. Additionally, it examines how demographic characteristics, service utilization, regional variations, and the burden of comorbidities relate to the observed temporal trends. Methods – This study included 82,402 lung cancer patients treated at 33 hospitals in Türkiye between 2000 and 2025. Hospital-based prevalence data and annual diagnosis counts were obtained from year-specific admission indicators collected at the hospitals. To account for temporal variations in healthcare utilization, annual diagnosis counts were normalized by the corresponding annual hospital application volume recorded within the lung cancer cohort. Time-series datasets based on annual data were constructed using aggregated clinical, demographic, healthcare utilization, and comorbidity characteristics. ARIMA, Linear Regression, and Exponential Smoothing methods were used as primary forecasting models. XGBoost was included as a machine learning benchmark model. Model performance was evaluated using RMSE, MAE, and MAPE metrics. Results – Hospital-based lung cancer prevalence showed a steady increase between 2000 and 2024. This rise became pronounced after 2019 and persisted beyond 2021. In contrast, annual diagnosis counts experienced a marked but temporary surge in 2021, followed by a partial normalization; normalized diagnosis rates followed a similar trajectory. The 2021 spike reflects a temporary intensification of diagnostic activity rather than a shift in the underlying prevalence trend. Sensitivity analyses excluding the year 2021 attenuated the positive temporal trend but did not eliminate it entirely. All three forecasting approaches applied indicate that hospital-based prevalence will continue to rise through 2030. XGBoost yielded substantially poorer performance on the hold-out dataset compared to the ARIMA and Exponential Smoothing methods; consequently, it was included as a comparative benchmark rather than as a long-term forecasting model. Among the evaluated ARIMA specifications, ARIMA(1,1,2) provided the lowest AIC (287.41), with a corresponding BIC of 291.39, and was therefore selected for subsequent forecasting. Conclusion – Multicenter real-world data used in this study show that the hospital-based burden of lung cancer has increased substantially over the past two decades, along with greater healthcare utilization among patients in the study cohort. The results also indicate a growing operational demand on oncology services due to a substantial increase in hospital-based lung cancer burden and emphasize the importance of consistent forecasts in healthcare systems for capacity planning, early-detection strategies, and efficient resource allocation.
Zitieren
APA
Aydin, O., Korkmaz, L., Selim, A., Kuzlu, M., Catak, F. O., Kusetogullari, H., & Cali, U. (2026). Forecasting lung cancer burden using multicenter real-world data. Pattern Recognition Letters, 210, 8-19. https://doi.org/10.1016/j.patrec.2026.09.021
Vancouver
Aydin O, Korkmaz L, Selim A, Kuzlu M, Catak FO, Kusetogullari H, et al. Forecasting lung cancer burden using multicenter real-world data. Pattern Recognition Letters. 2026;210:8-19. doi:10.1016/j.patrec.2026.09.021
BibTeX
@article{Aydin2026,
author = {Aydin, Omer and Korkmaz, Levent and Selim, Aybeyan and Kuzlu, Murat and Catak, Ferhat Ozgur and Kusetogullari, Huseyin and Cali, Umit},
title = {Forecasting lung cancer burden using multicenter real-world data},
journal = {Pattern Recognition Letters},
year = {2026},
volume = {210},
pages = {8--19},
doi = {10.1016/j.patrec.2026.09.021}
}
EUVIMED