A practical solution to estimate the sample size required for clinical prediction models generated from observational...

Area of Science:

• Biostatistics
• Clinical Epidemiology
• Health Informatics

Background:

• Estimating sample size is critical for clinical prediction model development and validation.
• A lack of consensus exists regarding appropriate sample size determination methods.
• This study addresses this gap by comparing existing methods for sample size estimation in clinical prediction models.

Purpose of the Study:

• To compare available sample size estimation methods for clinical predictive models.
• To define a practical solution for sample size estimation in this context.
• To apply these methods to the Horizon 2020 PRIMAGE project as a case study.

Main Methods:

• Employed three distinct sample size calculation methods: Riley's method and "rules of thumb" (10 and 5 events per predictor).
• Analyzed the variation in required sample size based on different parameters.
• Estimated the necessary sample size for both model development and model validation.

Main Results:

• For model development, sample sizes required were 1397 (neuroblastoma), 1060 (high-risk neuroblastoma), and 1345 (diffuse intrinsic pontine glioma - DIPG).
• Sample sizes for validation were estimated at 326 (neuroblastoma), 246 (high-risk neuroblastoma), and 592 (DIPG).
• Sample size could be reduced by limiting predictors, using direct outcome measures, or extending follow-up.

Conclusions:

• Recommends sample size methods based on epidemiological data and specific clinical problem characteristics due to variability in results.
• Emphasizes tailoring sample size estimation to the unique aspects of the clinical prediction task.
• Suggests strategies to reduce sample size, including optimizing predictor variables and incorporating direct outcome measures.