Step 4
Identify cases where any outcome date has the year entered as ‘9999’ (birth year not
known/not answered). In these cases, there is no sound basis for imputation, and it is
impossible to determine by machine algorithm what order to place that baby in.
These cohort members are therefore flagged to be examined on a case-by-case basis.
See Appendix 6.
Cohort members with no outcome data are also flagged to be by-passed (see (vi)
above, and Appendix 9).
Step 5
Excepting cases eliminated in step 4, flag up all cases where outcomes more than nine
months apart have been bunched in adjacent birth slots as though they were part of
one pregnancy, and cases where outcomes have been entered in the wrong
chronological order (i.e. not putting the most recent pregnancy first).
Step 6
Set up a variable TOTBABY to log which birth slots contain any data at all (i.e. any
outcome date, even if entered as ‘9999’, or any outcome code (e.g. livebirth,
miscarriage, even if entered as ‘9’), so that we have a check, after the re-ordering
process, to ensure no baby was missed out.
Step 7
Implement an iterative procedure to compare all forty possible outcome dates
(BIRDAT1-40), producing up to eight ‘KEY’ variables indicating which of the forty
slots was the most recent date found (KEY1), the next most recent (KEY2), and so
on. For instance, if the third most recent event was found in slot 7 of the forty, KEY3
would be assigned the value 7. We know from a cursory analysis that only one
1999/2000 cohort member ever carried more than eight babies (BCS70 15610009),
and a manual analysis confirmed that case had all its pregnancy history data correctly
ordered, so it was bypassed by this algorithm.
Step 8
Using this information, compare each successfully-ordered birth date with the
adjacent one, to flag up all cases with true multiple births (i.e. births on same day or at
least within 3 days), and all cases with a dubious multiple birth (i.e. more than 3 days,
but less than nine months apart). As with Step 4, the intention is to look at these
manually on a case-by-case basis.