Risk prediction models for discrete ordinal outcomes: Calibration and the impact of the proportional odds assumption.

"see section 6 and data s1 for information about alternative flexible recalibration models.; the chosen sample sizes are partly arbitrary but see data s1 for further explanation.; having binary predictors or a number of noise predictors did not change the findings4 2 3 we present results for the main scenarios in the main text (figures5 6 7 ) and for all other scenarios in data s1 ( figures s19-s33 ).; see data s1 for example r code to fit models and evaluate performance.; in data s1 we compared this approach to other approaches to assess whether non-mlr prediction models are disadvantaged by this setup.; one may prefer to replace the z ^ j in equation ( 18 ) with logit v ^ j to acknowledge the ordinal nature of the outcome (ie the third approach in data s1 ).; data s1 supporting information click here for additional data file."

"the complete r code is available on github ( https://github com/benvancalster/ordinalcalibration ) 4 2 4 2 1 in the large sample simulations when true predictor means were equidistant (scenarios1-2) risk estimates corresponded almost perfectly with true risk for the mlr ac-po and slm models (figures1 2 ).; the complete r code is available on github ( https://github com/benvancalster/ordinalcalibration ) 5 2 the likelihood ratio tests suggested violations of the proportional odds assumption for the cl-po model mainly for age and hypertension ( table s8 )."

"Funding information Fonds Wetenschappelijk Onderzoek, G0B4716N; Onderzoeksraad, KU Leuven, C24M/20/064"