Fig. 3

Internal calibration-in-the-large ratio (mean predicted risk divided by mean observed risk) vs stability calibration-in-the-large ratio when the models are developed and validated in the same database (top three rows) or different database (bottom four rows). The columns “ > = 1000 Outcomes,” “ > = 500 Outcomes,” and “All results” represent only including models and validations where the dataset contained > = 1000 patients with the outcome, only including models and validations where the dataset contained > = 500 patients with the outcome, and all models and validations, respectively. Each row corresponds to a different validation data permutation compared to the model development data (e.g., the population, outcome definition, and/or the database is changed)