Re: posthoc step

From: "Gastonguay, Marc" marcg@metrumrg.com Subject: Re: [NMusers] posthoc step Date: Tue, December 7, 2004 7:17 pm Ken, I'd like to suggest an additional point to consider regarding your explanation about shrinkage estimation properties of MAP Bayes estimates. As you point out, the number of data points does affect the balance between the influences of the individual data and population priors, but this is not the only consideration. The relative magnitude of OMEGA vs SIGMA is also an important factor in determining if the MAP Bayes estimate is dominated by the individual's data or the population priors. If inter-individual variability (OMEGA) is very large relative to measurement noise (SIGMA), a situation that is not uncommon in population PKPD, the second term in the MAP Bayes OF is essentially minimized regardless of the magnitude of ETAki. Under such circumstances, it is possible to have a data-dominated individual MAP Bayes estimate, even when sampling is sparse (but greater than zero). We've all observed the situation where POSTHOC IPREDs from a base model fit the individual data very well, even with sparse sampling. The opposite is also theoretically possible (SIMGA>>OMEGA), and could lead to prior-dominated MAP Bayes estimates even when sampling is extensive, but this is rarely the case in population PKPD. Marc Marc R. Gastonguay Metrum Research Group LLC www.metrumrg.com