Table 4.

Guidelines: premises and recommendations for the inclusion of race or ethnicity in non-polar clinical predictions

Item	Statement
16	The hallmark of a non-polar prediction is that it is used only to optimize an individual’s outcomes or align a decision with a patient’s own values and preferences. (P9)
17	Race or ethnicity may be included in non-polar models if (and only if) predictive effects are independent from other ascertainable attributes, statistically robust, and clinically meaningful (i.e., can alter decision-making in some patients). (R11)a

P denotes premise, R recommendation.

^aThis recommendation reflects the fact that, in some circumstances, inclusion of race predictive effects can make predictions more accurate, particularly in groups comprising a smaller proportion of the population, since omission of a race variable will yield “average” predictions more reflective of the majority population. The benefits of more accurate prediction (Box 4) may need to be balanced against other considerations, as discussed in Box 3.