About the power of biostatistics in sibling analysis—comparison of empirical and simulated data

Abstract

The determination of potential sibship is a common task in routine kinship analysis, but often the putative parents are not available for analysis anymore. Then, a sibling analysis has to be conducted investigating only the potential siblings, thus reducing the power of the conclusion. In an attempt to determine meaningfulness of biostatistical calculations, 346 dizygotic twin pairs, 30 confirmed half siblings, and 112 unrelated people (to generate 6216 pair comparisons) were studied, all genetically typed using at least the Powerplex® 16 STRs. From every pair, the probabilities for a full sibship (identical parents) and half sibship (different fathers) were calculated using a commercially available computer program. Additionally, we simulated marker data for one million pairs of full sibs, half sibs, and unrelated persons each. Ninety-five percent of full sibling pairs demonstrated a likelihood ratio (LR) > 9 (W-value > 90 %) and less than 4 % of these showed a LR < 3 (W-value < 75 %) for full sibship after analysis of 15 STRs. The results for half siblings are less unambiguous. Here, only 57 % achieved a LR > 9 and 23 % a LR < 3. Regarding the unrelated pairs, more than 90 % had a LR < 1/9 and only 2 % reached a LR > 9. All in all, our results show that 15 to 20 STRs have sufficient power for analyses in kinship. Moreover, our data provide a statistical basis for the determination of the information content of a LR/W-value in a sibship case. Investigating an identical number of full siblings and unrelated pairs, it could be shown that 92 % of pairs with a LR > 9 for full sibship probability really are full siblings. So, setting a cutoff level for full sibship at LR > 9, less than 10 % of pairs will be wrongly assumed as full siblings even though they are unrelated.