Reuniting El Salvador families

The problem

From about 1980 to 1992 many small children were taken to orphanages when their when families were disrupted and separated in the civil war (El Salvador human rights practices, 1993). The organization Pro Busqueda (see links) attempts to reunite the children, many of whom were adopted into the United States, with what remains of their biological families. Among the efforts in support of Pro Busqueda is a volunteer DNA identification program led by a group of DNA analysts of the California DOJ, Jan Bashinski DNA laboratory in Richmond, Ca.

Role

I volunteered to assist, feeling that the problems of matching children with their relatives is very similar to the problem of identifying mass disaster victims. I've participated in or advised many related projects and developed various kinds of software – especially as part of the World Trade Center identification project – which I believe will be helpful with the family associations in El Salvador.

Update – July 2006 visit to San Salvador Four days in San Salvador gave me a chance to meet the dedicated staff of ProBusqueda and to learn a lot more about the history – the nature of the civil war that caused the desaparecidos – the technical and political difficulties and process involved in finding and reuniting children families, and the process of human rights work. Our contingent included Physicians for Human Rights, Human Rights Center (UC Berkeley), and DNA folks. 14,000 missing children? Among other things, data given to me extrapolates that the total number of children adopted out of the country in the 80's may be about 14,000.

These software facilities are

• nice simulation graphs
• screening
• kinship
The main reason for this web page is to present some simulations that we have done.

Scope of the problem – estimates

There are about 500 families with reported missing children, a total of about 700 children. For our present estimates, we simplify that there are merely N=500 children, for now mostly ignoring the extra 200.

We tentatively assume that three factors will help narrow the search:

1. sex
2. approximate year of birth
3. approximate year of entering the orphanage
Assuming that both the biological and the adoptive families know those three data, then perhaps that typically reduces the possible (original) identities for a particular adoptee down to 10 choices (factors of 1/2, 1/5, 1/5). Subject to revision, we've made some calculations taking 1/10 as the prior probability.

We've also decided, as a working criterion, to aim for 99.9% probability for each identification. Since

Posterior odds = (likelihood ratio)(prior odds),

99.9% is posterior odds of 1000, and the prior odds are about 1/10 (easier to deal with than 1/9), this means that we need roughly

likelihood ratio > 10,000.

The workers in El Salvador are in the process of locating the biological relatives. Sometimes extra relatives are available but hard to get to. We would like to give them guidelines as to what combination of relatives is likely to be adequate.

DNA modalities

• Identifiler – 15 STR loci (plus Amelogenin) – is the principal method of testing
• Y-haplotypes are expected to be used sometimes
• mtDNA may also be used, but expense is an issue

Simulations

The simulations are based on Identifiler, and "Hispanic" populations for allele frequency estimates.

The simulation procedure is to

• posit some collection of relatives
  • simulate alleles for each relative and the child
  • compute the LR that the child is who he is, versus unrelated
  • compute the posterior probability that he is the right child
• repeat the simulation a large number (30-50) times
• report the average posterior probability
The average posterior probability seems like a sensible summary statistic. Usually the simulated likelihood ratios have a very large range, such as from 100 to billions. The worry is how often do the small values occur? If 10% of simulations give a relatively small likelihood ratio, then that represents a 10% chance to be uncertain about the identification. Whether the large values are in the millions or the trillions doesn't much matter. And that is the way the average posterior probability works – it is mostly a measure of the less successful simulations.

The chart below summarizes the results for the various combinations of relatives that we have considered.

Family references	Probability of identity (Mean posterior probability at 10% prior)	comment
one full sibling	92.1%
sibling and aunt	94.4%	or sibling and uncle
sibling and two sister aunts	97.8%	i.e. the aunts or uncles are either both maternal or both paternal
sibling, aunt, uncle different sides	99.8%	e.g. maternal aunt, paternal uncle
sibling and half sibling	98%
sibling and two half siblings	99.4%	all children share mother; half-siblings share father with one another
two siblings	99.91%
one parent	99.9%	ignoring mutation
sibling and parent	99.996%
father and one half sibling	99.95%	maternal half-sibling
father and two half siblings	99.996%	maternal half-siblings of course
father and aunt	99.993%	maternal aunt (else why bother)
three grandparents	96.7%
four grandparents	99.99%	surprise – big bonus for all four
three grandparents and sibling	99.994%

The children were generally at most 12 when taken, so none of them left children of their own in El Salvador that can be considered as references.

Conclusions

We would like to have a very simple set of rules to help the caseworkers decide how many and which relatives they need. Based on the above chart, the rules might look something like –
• need 10 points worth of relatives
• prefer 15 points worth
  • parent=10
  • sibling=5
  • aunts, uncles, half-siblings=1.5
  • grandparents=2
  • bonus for all four grandparents=4
• Should it not be possible to reach even 10 points (no relatives available anywhere to reach that score), then in order to at least assure us of either Y-STR or mtDNA information:
  • If Missing person is a male:
    • a full brother (Y-STR and mtDNA), or if that's not possible
    • paternal uncle (Y-STR) plus maternal aunt (mtDNA)
  • If Missing person is a female:
    • full sib or half sib (mtDNA, from shared mother), or maternal aunt.

---

Go to top
Neruda