Assessing the Sensitivity of Decision-Analytic Results to Unobserved Markers of Risk: Defining the Effects of Heterogeneity Bias