Search results

One of the consequences of owning a farm in Central Brazil is that I frequently meet people who do things and have life narratives that we are unlikely to come across in texts on sociology or epidemiology. In the nearby town they live in streets that have postal codes that would place all the residents in the same cell of a contingency table.
But in each residence lives a family with a separate and unique story. I believe that it is worth asking whether it is really the perceived similarities that determine their health outcomes. Yes, perhaps when sanitation is involved, but there are other health outcomes that would not be centered on the postcode.
I have spent the last two years helping to link notifications of interpersonal violence with birth and mortality records. The idea is to find some way to stop men harming their partners. If this can be achieved by changing the way the law reacts to the violence then these linkages may prove useful, especially when legal and penal records are included in the studies.
But what if what needs to be changed is the way a boy is treated in his first year of life? It is unlikely that information collected at the time of the violence would be accurate about events in early childhood. How could record linkages tell us that we should be looking elsewhere? I have no idea. But I believe it to be the most important question in population studies today.
Statisticians are always pleased to tell us that we have failed to prove something. We need a methodology that tells us where to look. Also it must be based on something with more possibilities than those currently offered by diluted Marxism.

ABSTRACT
ObjectiveBloom Filters have been used in a number of studies conducted for the Ministry of Health. They are usually recommended because of the possibility that they may participate in secure protocols for the exchange of data. In our case the speed of the program, once the filters have been prepared, is so high that that itself is sufficient motive for their adoption.
Nevertheless if two calendar dates differ by one character this may merit more attention than a similar difference in personal names. This became evident in a large linkage between mortality records and hospital separations where the patient had died. Higher scores were obtained when the date fields differed by only one character, but when that character represented a year there would no reason to notice the pair. When the character difference was compatible with a difference of a few days this would be more interesting because in studies like the one just cited it would be reasonable to admit differences of a few days or even, perhaps, weeks between the events ( recording of the death of the patient ).ApproachHow then to represent the difference between dates in a Bloom Filter? A date can be represented as a Boolean vector where the day (or week) is set to '1'. It may be represented by several contiguous '1's to admit admissible uncertainty in comparisons. The similarity between two dates can then just be the Dice Coefficient of the corresponding vectors.
ResultBut a vector representing a date may then be very large. It could be as much as 365 bits per year, far more than is usually used for the other fields. The number of logical word comparisons would go up and the program would become slower. Knowing that the admissible range is presented by contiguous '1's means that we can obtain the effect of constructing the Bloom Filter and calculating the Dice Coefficient more directly. Starting with the two dates we can obtain the number of bits that are shared, which will depend on the admissible range. The Dice Coefficient can then be calculated directly without the need to construct the Filter.
ConclusionWe are then left with the decision on how to add the result to the value obtained from the other variables, and this will depend on what importance it is felt the date should have.

ABSTRACT
BackgroundIn Profile Matching the term "Profile" means that the information available about an individual could be the same as that available for some other individuals.
Reduced versions of birth and mortality records are made publicly available by the Brazilian Ministry of Health and this study covers the years 1996 to 2013. Date of birth, sex and municipality of residence are available in both types of record, so their combination could be studied as a pseudo-identifier. The municipality could be different at the two events so the results outlined here must be seen as a first cut.
MethodThe ICD 10 code for microcephaly is Q02. Among the mortality records where Q02 was recorded either as a basic cause or in one of the other lines there were 1676 combinations of the three variables. All but three of them were unique and the remaining three each occcurred with only two records. For 1324 of these records the year of birth was between 1996 and 2013 so that we could reasonably expect these combinations of values among the birth records. However, there were only 204 birth records with combinations that matched and where Q02 was recorded. Possible reasons for the difference could include i) underreporting of births, ii) the failure to detect microcephaly at birth registration and iii) the possibility that microcephaly can only be securely identified sometime after a birth would normally be registered.
ResultsAmong the birth records there were 2349 combinations where Q02 was mentioned as a birth anomaly. All but seven of them were unique and the remaining seven each occcurred with only two records. Corresponding to these combinations there were 1416 combinations found among the mortality records. These were found in 2149 records. However in only 541 of the mortality records was there a mention of Q02. It appears as the basic cause of death in 204 of these 541 records. It could be that such a small number of deaths attributed to Q02 is consistent with best medical opinion.
ConclusionIt is easy to dismiss these apparent discrepancies as poor recording. Some of it may be, but it could be that doctors were faithful to what they observed. The next step is have a clear picture of what was recorded, like other types of birth defect, when Q02 was not observed in both records of a match.