Appendix 4.4 The control of anaemia in pregnancy
10. Usefulness of hospital routine measurements for further studies
2.1. Overview of methods
2.2.3 Choice of methods for the present study
2.3.1.7 Choosing where to go in Colombo for the urban study
Colombo was chosen by UNICEF as the site for the urban study. In order to obtain representative women from the widest area o f residence in Colombo, Castle Street Hospital Antenatal Clinic was used as the sampling frame. Women came to this clinic
from many miles around. The hospital was chosen because o f the fact that most
"ordinary" people go there for antenatal care. In addition a number o f other Colombo Antenatal and Child Welfare clinics were also visited, including Kirula maternity home (antenatal clinic and well baby clinic), Kirillapone maternity home (antenatal clinic), and also Hennamuila camp for home visits. Numbers seen at these other sites were very small compared to those seen at Castle Street Hospital.
2.3.1.8 Selection of women in Colombo
Women were chosen randomly to be included in the study, and a tally was kept of gestation so that 50 women were seen for each month of pregnancy from 4-9. Near the end only women in relatively unrepresented gestational groups were included.
2.3.1.9 Checking the Colombo population was representative
The FHW (Family Health Worker) clinic area o f origin o f each person was identified and then with the hospital FHW s, coded as near, middle, or far from Castle St Hospital in Colombo. Proportions o f each were checked each day to make sure that the sample was indeed continuing to be representative o f Greater Colombo as a whole. We worked closely with an experienced demographer at UNICEF who was very happy with the representativeness in terms of random procedure and actual area o f origin of both the
T able 2.3.2 Colom bo, selection of subjects fo r cross sectional u rb a n study
Sampling frame pregnant women attending Castle St antenatal
clinic (from all over Greater Colombo)
Stratification by stage (month) o f pregnancy
Selection random selection and allocation to stage of
pregnancy group until there were at least 50 in each group, months 4-9 (see Table 3.2.4, page 187)
+ a few random home visits in Hennamuila camp (a poor area)
+ seeing 100% o f those attending the two community clinics (small attendance only)
Identification o f representativeness inspection o f geographical area o f origin,
2.3.2 Selection o f th e Sri L anka K andy co h o rt longitudinal pop u latio n
2.3 .2 .1 Issues in selection of subjects in coh o rt studies
In cohort (or follow up, longitudinal) studies such as the Kandy population in the present study, the probability of the event o f interest occurring may be strongly related to how the sample was obtained. Population based studies are likely to show a different pattern (e.g. a reduced prevalence o f severe outcome) compared with a clinic study where only the more serious cases are seen. (The extent o f the bias will be variable according to the local referring patterns and facilities available). Clinic based studies are also likely to show higher and more variable re-visit rates than population based studies. Population studies are difficult and expensive to carry out but studies o f highly selected subjects can give misleading results especially about the natural history o f a disease. When most o f the population both attend antenatal care and deliver in hospital (as in Sri Lanka) a hospital based study can be a highly effective way to conduct a study o f late pregnancy and birth (see section 2.3.2.4 below).
Loss to follow up is the main difficulty specifically encountered in cohort studies. Losses to follow up reduce the numbers supplying information and thus weaken the analysis slightly. The main concern though is that subjects may be lost to follow up for some reason that is related to the outcomes being studied or to pre-defined risk categories. Hence the efforts that are made to track subjects lost for some reason. Some losses are inevitable and comparison of characteristics is useful to find out who are those people.
Altman (1991) describes a study by Martin and Bracken (1987) which identified 6219 pregnant women in New Haven for possible inclusion in a study to investigate the relationship between caffeine consumption and birthweight. The number yielding data was reduced to 3858 for the following reasons:
6219 identified for possible inclusion
5331 agreed to be contacted
4926 eligible and willing to be in study
473 refused to be interviewed
263 could not be reached
4 unreliable interviews
4186 valid interviews obtained
76 pregnancy outcome not ascertained
56 delivered at a different hospital
116 not a live birth
46 not singleton deliveries
33 birthweight not recorded
3858 study data obtained for caffeine
consumption and birthweight
Examination o f the reason for loss did not suggest there was a likely link between loss to follow up and caffeine consumption or birthweight, but the possibility o f bias always needs to be considered.
For the current pregnancy and neonatal longitudinal study, Kandy Hospital was chosen in order to be somewhat similar to the rural Mawanella population studied six months earlier. It was known that mothers at Kandy Hospital were not the affluent (they went to Peradeniya, the teaching hospital, or privately) yet would provide large number of Sinhala women across the socioeconomic status spectrum. Professor Aponso was able to arrange permission for the study without undue difficulty.
Table 2.3.3 Kandy, selection of subjects for longitudinal study
Sampling frame Pregnant women attending Kandy hospital
antenatal clinics
Selection Random
Follow up Those attending for delivery at Kandy Hospital
who had been measured antenatally
Identification o f representativeness a) by comparison o f loss to follow up with those who were seen
b) by comparison o f those who were seen with the total population o f births at Kandy hospital during the study months
2.3.2.2 What proportion of the Kandy Hospital births were measured?
To estimate what proportion o f Kandy hospital births were measured, every live birth (N = 2257) for nearly 4 months (April 1 - 22 July inclusive) and stillbirths 1 April - 15 June (N = 4 7 ) was recorded separately as a "total baby population". The mother’s name, the hospital number, sex of the baby, the date o f birth, the baby identification disc number, the consultant identification number, the time o f birth, and whether the baby was live or stillborn. These data were collected from "the hospital baby book" and from the separate record on Frazer Ward (private patients) for " Classes 1 and 2 " private. For the births from 4 May - 22 July the age o f the mother, her parity and the birthweight were also recorded. From 1 May to 15 June a separate stillbirth list was available. In April stillbirths were included in the list o f live births. For the 624 recorded for April the hospital measured birthweight, length and head circumference were also recorded. Altogether 2304 births were recorded as taking place at Kandy Hospital in the study period 1 April - 22 July. The 525 who were measured in that
2.3.2.3 Were the babies measured (N=525) different from the babies not