A global systematic analysis and statistical evaluation of prognostic prediction models for non-small cell lung cancer

Authors:
  • YCYuan Chen
    Yuan Chen

Abstract
Prognostic prediction models can aid clinical decision-making for high-risk non-small cell lung cancer (NSCLC) patients. We conducted a systematic search across multiple databases and evaluated eligible studies using PRISMA and PROBAST checklists. Of 28,833 references screened, 233 studies describing 268 models were included. Among them, 89 underwent external validation; 67 (75.28%) were classified with high risk for bias. Most models were developed in North America (45.06%). The median area under the receiver operating characteristic curve (AUC) for 1-, 3-, and 5-year predictions were 0.759, 0.720, and 0.691, respectively. The most common predictors were age (54.85%) and stage (58.21%). And, 47.01% of models were sex-specific. Recently, easily accessible radiomic features have been increasingly used in statistical modeling compared to molecular omics. Models integrating radiomic and clinical features showed potential for improved performance (1-, 3-, and 5-year AUCs: 0.931, 0.985, 0.942). Model discrimination varied across different studies, and there is a lack of model calibration and clinical utility. This review highlights advances and limitations in NSCLC prognostic models. To enhance model reliability and generalizability, it remains crucial to adhere to the TRIPOD guideline to perform prediction model study, and to emphasize comprehensive model validation and bias-reduction strategies. Following a step-by-step guide to develop and validate clinical prediction models by Efthimiou, et al. (BMJ, 2024), along with a multifaceted, multidisciplinary and multi-regional approach, is the key way to facilitate the development and clinical application of qualified models.

This publication has 66 references indexed in Scilit: