The statistical fallacy
Correct interpretation of the results from BioAlder will be analogous to the assessment of a diagnostic test, where the positive predictive value of the test will vary according to the prevalence in the population. When testing a low-prevalence population, as in the case of screening, the proportion of false positives can be higher than true positives in certain cases.
Imagine that in a given period of time, 100 asylum seekers claim to be 17 years old. Then assume that ten of these are 18 years old. In order to identify these, all of the 100 are subjected to a BioAlder assessment. At 90 % sensitivity (proportion of 18-year-olds that the test indicates are at least 18 years of age, with 87.5 % as the decision limit), nine out of ten will be intercepted. However, due to natural variation in biological maturation, the test will also identify some of the 17-year-olds as being at least 18 years of age. At 94 % specificity (proportion of 17-year-olds that the test indicates are no more than 17 years old), 6 %, i.e. five of the 17-year-olds, will be classified as at least 18 years old. Consequently, the 87.5 % test will classify 14 asylum seekers as 18 or older, when only nine of them actually are. Thus, the probability that any one of these is 18 years or older is 64 %, not 87.5 %.
In Sweden, dental X-rays and MRI of the femur are used to estimate the age of young asylum seekers. The Swedish Agency for Health Technology Assessment and Assessment of Social Services (SBU) has recently published two reports (9, 10) in which these methods are assessed. SBU argues (11) that it is fundamentally wrong to use the results of age testing on individuals or a group when the age distribution in the tested population is unknown.
It is fundamentally wrong to use the results of age testing on individuals or a group when the age distribution in the tested population is unknown
The confidence intervals yielded in BioAlder, and which the UDI and the legal system use as a basis to assess age, assume a known and even age distribution. If this distribution is not exactly what is assumed – and it seldom will be – the results will be misleading, as illustrated above.