The analyses in this paper have confirmed the association between ANC and perinatal outcomes (pregnancy outcomes and birthweight) in the Kwale region of Kenya, although the percentage of women accessing ANC (32%) is considerably below that found in other surveys [1, 16]. This low utilisation of ANC may be expected given that Kwale is a particularly deprived rural area of Kenya. The informal encouragement by enumerators was insufficient to convince many women to attend, although this encouragement was not tested as an intervention to improve attendance. However our data reveal an interesting pattern, whereby two ANC visits is associated with a better outcome than three or more visits. This may be related to general low attendance and hence circumstances where women tend to attend more frequently if they are having problems with their pregnancy.
It is possible that the relationship between ANC and birthweight is affected by the length of pregnancy, since mothers of babies born prematurely would have had less time in which to attend for ANC. However we do not have sufficient data on gestational age in order to test this hypothesis and ascertain the extent of any bias arising from prematurity.
One of the reasons ANC may have a positive association with perinatal outcomes is the association between attending ANC and behavioural decisions, including TT and SP doses, use of an ITMN and decisions regarding place of delivery and who assists the delivery. The associations between ANC and TT/SP doses and use of an ITMN are not surprising as these interventions are part of the ANC care package, and it is unlikely that women would have access to these interventions outside of ANC services. ANC clients are also routinely given iron, folic acid and vitamin supplements, which may impact perinatal outcomes. As noted in previous research [1, 15], women who have had ANC are more likely to deliver in a formal health facility and/or in the presence of a doctor or nurse-midwife. Given the low level of attendance for ANC, it is not surprising that the proportion of women delivering in a formal health facility and/or in the presence of a doctor or nurse-midwife is roughly half the 40% found in the 2003 Demographic and Health Survey in Kenya . However deciphering the links between ANC, behavioural decisions and outcomes is always difficult in non-experimental studies. For example, it is possible that differences in the incidence of malaria between dispensaries affected the proportion of low birthweight babies across the dispensaries, but data on the incidence of malaria are inadequate for us to test this hypothesis.
Given the low attendance at ANC by women in the study areas, there is a need for qualitative research to investigate why women do or do not attend for ANC and could help to identify the effect of ANC on behavioural decisions. Qualitative research in a rural area of Zambia, for example, revealed that long distances, lack of transport, user fees, lack of health education and poor quality care deterred women from delivering in a clinic . Three potential factors deserve particular attention. First, there appears to be a significant proportion of women who live some way from the dispensary and who have no ANC. Hence lack of good roads or transport may be the barrier to attendance and this factor should be explored in qualitative research. If this is the case, women in distant areas could be encouraged to attend for ANC by providing transport or arranging an out-reach ANC service in certain areas. Alternatively, an attempt could be made to provide more health facilities to reduce the need for women to travel long distances while pregnant or to consider the use of maternity waiting homes .
Second, variations in the content and quality of care received across dispensaries should be investigated. Such variations may help to explain differences in the number of ANC visits between dispensaries: women receiving poor care would be less likely to return for further visits. For example, there is some evidence that women in Samburu are more likely to attend ANC than women in Matuga and this may reflect the previous work of Aga Khan in strengthening health systems in Samburu. The quality of care should be assessed both within the dispensary and also through a survey of women who have received care: it may be the case that women in one area are less satisfied with a given level of care than women in another area, due to differences in expectations .
Third, the analyses identified a positive association between secondary education and attendance for ANC (although the association was only statistically significant in one of the multivariate analyses). Campbell et al.  suggest that the provision of advice on seeking ANC is an important component of the care package for reducing maternal mortality and this advice can be delivered by a variety of methods. Efforts to encourage women to attend ANC could therefore be targeted at less educated women and could include formal (e.g. employment or school based) or informal education sessions for younger women which include information on ANC and childbirth . Any programs used to encourage use of ANC should be evaluated to identify their effectiveness (or otherwise).
In contrast to previous studies in Kenya [1, 15], the analyses reported here did not identify any association between income or SES and attending ANC or perinatal outcomes. Even though two different measures of SES were used in the analyses, it may be the case that another financial variable is affecting decisions of whether to attend ANC and again qualitative research may help to uncover this variable.
The data in this study are limited to two areas within the Kwale region of Kenya, and hence the results may not be generalisable to other regions. The response rate was very high (almost 100%), indicating high precision in the data. However, the data collection process was unable to identify the timing of women's ANC since few knew their dates of conception or ANC visits and therefore it has not been possible to compare ANC timing with previous research in Kenya, or to identify the effect of timing on perinatal outcomes. There were a few sets of twins in the sample (the exact number is not known), and the birthweights of each twin were recorded separately. This may have created a small bias in the data, particularly in terms of birthweight. Furthermore, the explanatory power of the multivariate models was low, indicating that there are a number of other factors affecting ANC attendance and perinatal outcomes. Qualitative research may help to identify these factors and hence identify possible interventions to improve ANC attendance and perinatal outcomes.