either experiencing no outcome, or any one of the three outcomes).ĭoes this intuition make sense? Do you know of any papers that discuss this? For this assumption to be incorrect, the power of the ordinal logistical model would have to be less than a logistic model that was based on a dichotomisation of the same outcomes (i.e. This is of course not the case, I have plenty information contained within the other more common outcomes and intend to “exploit” the proportional odds assumption to “borrow” information from them.īecause the coefficients are estimated using all events, it seems intuitive to me that the minimal sample size calculated in steps 3 and 4 should be based on the overall proportion of individuals who experience at least one of the three possible outcomes. This is because the calculation assumes I am dichotomously trying to predict this rare outcome alone and assumes I have only this rare information to estimate the coefficients. Because one of my ordinal outcomes is rare, these steps produce a ridiculously high minimal sample size, upwards of 150000 patients. However, I run into problems in steps 3 (What sample size will produce predicted values that have a small mean error across all individuals?) and 4 (What sample size will produce a small optimism in apparent model fit?). I was then going to target the highest minimal sample size obtained from calculating any one of the four steps on any of the three outcomes, my thinking being that a ordinal model will be at least as powerful as a dichotomous logistic model. My simple approach was to follow steps 1-4 for a binary outcome model and calculate the required minimal sample size for each of the three outcomes, for each step. I was therefore going to adhere to the guidance discussed in the paper you linked, and also summarised here: I am not aware of any guidance or methodological papers that specifically address how to do this for an ordinal model. I have a question for you and have been tasked with calculating the minimum required sample size for a multivariable ordinal logistic model with three outcomes.
0 Comments
Leave a Reply. |