There has been an increasing awareness that the perinatal period is a neglected area in recent years, and interventions targeting mothers and newborns have been encouraged [1]. Nearly four million newborns die during the first four weeks of life every year [2, 3] and the rate of neonatal mortality has remained basically unchanged in the past decades [4]. Some improvements can be seen, but still the pace is slow, especially in the early neonatal period [5]. Most of these neonatal deaths occur during the first day of life and complications related to delivery care make up a large proportion of the overall neonatal mortality [3, 6]. Skilled assistance at delivery and access to emergency obstetric care are the most effective interventions to prevent these early and intra-partum related deaths [7]. This requires both the availability of such services as well as the will and possibility for pregnant women to seek this care at delivery [8].

Thaddeus and Maine's conceptual framework of the three delays in care seeking has been widely used when investigating health care utilisation [9]. They developed their thoughts around obstetric emergencies, but the framework is valid for any kind of care seeking behaviour, and Gabrysch and Campbell adapted it to also cover use of preventive services around delivery [8]. Geographical factors such as distance between home and health institutions are part of the first and the second delay and suggested an influence on the choice of delivery place [8] as well as being related to neonatal mortality risks [10]. It has also been demonstrated that the usage of health services decreases with increasing distance between health facilities and families' homes. This phenomena, often labelled by the geographical term distance decay [11], has been used to describe various situations and patient groups [12–14]. It has also been shown for the utilisation of maternal health care and delivery services [15–17]. Geographical Information Systems (GIS) can be used to investigate spatial dimensions of different health outcomes. This has been applied to study infectious diseases such as malaria and acute respiratory infections [18, 19], as well as to study geographical variation in child mortality [20]. So far little has been explicitly done in the field of peri- and neonatal health.

We have initiated a cluster-randomised trial for improved neonatal health and survival in the Quang Ninh province in northern Vietnam. The trial that has been given the acronym NeoKIP (Neonatal Health - Knowledge Into Practice, ISRCTN 44599712), is a collaboration between the Ministry of Health, Uong Bi General Hospital and Hanoi School of Public Health in Vietnam and Uppsala University in Sweden. Preliminary results from the initial part of the trial indicate that ethnic group and health system utilisation before and at delivery were major determinants of neonatal survival [21]. We have also shown a strong negative association at baseline between home delivery rate and the chances of neonatal survival [22] and that a quarter of the mothers losing their baby in the neonatal period did not have any contact with the health system prior to death [23]. In this study we further investigate the determinants of neonatal mortality, including the care seeking behaviour at delivery by the addition of GIS data and analysis techniques. Specifically, we aim to examine the association between distance from the mother's home to the closest health facility and neonatal mortality. In this analysis a special emphasis will be put on different patterns of perinatal health care utilisation.

During the study period from July 2008 to December 2009 there were 11 708 live births and 197 neonatal deaths registered (NMR 17/1000). There was a marked geographical variation in neonatal mortality rate when calculated on district level, with higher rates in the remote and mountainous districts of the province (Figure 1). Home delivery rate for the whole study area was 9.9% (1155/11708). Most deliveries took place at a hospital (8691/11708), with the Vietnam-Sweden General Hospital in Uong Bi being the major delivery service provider with 3340 deliveries from the study area during the time period.

Distance decay

Mothers of cases were less likely to deliver at a health facility (66.4%) compared to referent mothers (85.6%) (p < 0.001). Figure 2 demonstrates the relation between health facility utilisation during pregnancy and at delivery based on distance to the closest health facility, indicating a difference in distance-decay in perinatal care utilisation between cases and referents. Mothers of newborns who died in the neonatal period had a sharper decay in both attendance to antenatal care as well as in delivery at health facilities than referents.

Neonatal mortality

Sixty percent (95/183) of the neonatal death cases died within the first 24 hours after delivery. There was an increasing proportion of very early neonatal deaths the farther away the mother lived from a health facility (Figure 3). There was however no difference in time of death when comparing home deliveries to health facility deliveries (p = 0.21). Neither did the place where mothers primarily sought care for delivery influence the time of death (p = 0.42) nor did the total distance from home to the final place of delivery (2007 meters for deaths on day 0 and 1879 meters for newborns dying during day 1-27, p = 0.75).

Crude odds ratios (COR) for factors associated with neonatal mortality are presented in Table 2. Sex of the newborn, mode of delivery, family structure and parity did not show any association with neonatal mortality (data not shown) and were not considered in a multivariate analysis (p > 0.10). Mother's marital status and age at delivery were associated on a significance level of 10%, and were included in further analysis (p < 0.10). Considering distance to the closest health facility to be in the casual pathway of the health care utilisation variables of antenatal care attendance and place of delivery, as expressed by the exhibited distance decay, these variables were also excluded from the multivariate analysis. After relevant adjustments in a multivariate logistic regression model there was an almost double risk of neonatal mortality for families living farthest away from a health facility (4^th and 5^th quintile, ≥ 1257 meters) compared to the group who resided closest to a health facility (1^st to 3^rd quintile < 1257 meters) (OR 1.96, 95% CI 1.40 - 2.75, adjusted for maternal age at delivery and marital status).

	Cases (n)	Referents (n)	Crude odds ratio	Confidence interval (95%)
Socioeconomic determinants
Maternal ethnicity
Kinh	371	68	Ref
Minority	228	115	2.75	1.94 - 3.91
Maternal education
Tertiary school or higher	40	207	Ref
Secondary school	39	151	1.33	0.82 - 2.18
Primary school	37	134	1.43	0.87 - 2.35
No schooling	65	106	3.17	1.97 - 5.09
Household economy
5th quintile (rich)	12	120	Ref
4th quintile	22	120	1.83	0.86 - 3.89
3rd quintile	23	113	2.04	0.96 - 4.31
2nd quintile	60	124	4.84	2.41 - 9.71
1st quintile (poor)	66	122	5.41	2.69 - 10.9
Proximate determinants
Distance to closest health facility
1st quintile (<401 m)	30	119	Ref
2nd quintile (402 - 740 m)	26	120	1.28	0.68 - 2.44
3rd quintile (741 - 1256 m)	32	119	1.60	0.86 - 2.97
4th quintile (1257 - 2232 m)	48	120	2.38	1.32 - 4.29
5th quintile (>2233 m)	54	119	2.70	1.50 - 4.85
Close (1st to 3rd quintile)	78	358	Ref
Remote (4th to 5th quintile)	102	239	1.96	1.39 - 2.75
Antenatal care visits
Yes	113	507	Ref
No	70	91	3.45	2.35 - 5.07
Place of delivery
Regional hospital	37	152	Ref
District hospital	75	277	1.11	0.72 - 1.73
CHC	13	90	0.59	0.29 - 1.18
Home	58	80	2.98	1.79 - 4.96
Facility delivery	80	519	Ref
Home delivery	58	125	3.01	2.02 - 4.49

When stratifying by socioeconomic factors no association between distance to the closest health facility and neonatal mortality could be found for Kinh mothers, mothers with higher education and mothers who were better off economically (Table 3). There was however a risk elevation for mothers with low education (OR 2.59, 95% CI 1.16 - 5.82, adjusted for maternal age at delivery and marital status) and for mothers from poor households (2.84, 95% CI 1.30 - 6.21, adjusted for maternal age at delivery and marital status). Ethnic minority mothers displayed a near significant risk (Table 3).

		Cases (n)	Ref (n)	Odds ratio*	Confidence interval (95%)
Household economy
Non-poor	Close	68	317	Ref
	Remote	48	159	1.40	0.92 - 2.13
Poor	Close	10	41	Ref
	Remote	54	80	2.84	1.30 - 6.21
Mother's ethnicity
Kinh	Close	45	226	Ref
	Remote	23	104	1.30	0.75 - 2.28
Minority	Close	33	92	Ref
	Remote	79	135	1.61	0.99 - 2.62
Maternal education
Completed 1° school (≥5 y)	Close	68	325	Ref
	Remote	46	165	0.84	0.86 - 2.01
Not completed 1° school (<5 y)	Close	10	33	Ref
	Remote	54	73	2.59	1.16 - 5.82

"Close" representing 1^st to 3^rd distance quintiles (<1257 m), "Remote" representing 4^th and 5^th quintiles.

* Odds ratios adjusted for maternal age and marital status at delivery.

In this study we have examined the association between distance to the nearest health facility and neonatal mortality and found an increased risk of neonatal death for mothers who live farthest away from health facilities. We have also shown distance decay in antenatal care attendance and facility usage at delivery. Mothers who delivered at home live farther away from a health facility than mothers who delivered within the health system and our results demonstrate how the rate of facility use at delivery decreases as the distance increases. This indicates that distance is a factor for the choice or necessity of home delivery and is in concord with previous research showing that an increasing distance from residence to the closest health facility decreases the delivery care utilisation rate [15, 29, 30]. Neonatal mortality was also associated with antenatal care use and the place of delivery. The increased risk for mothers living farthest away persisted when adjusting for place of delivery and antenatal care attendance. Health care utilisation patterns can thus not fully explain the association between distance and neonatal mortality. We also found that there was an earlier distance-decay effect among mothers who suffered a neonatal death than among referents (Figure 2) showing that there are other factors than distance affecting the choice of delivery place. One factor for the choice of delivery place not investigated here is the perceptions of the quality of health facility care. Earlier research has described a lower level of knowledge about perinatal health issues among health staff in the more remote and mountainous areas where distances in general are longer [31]. This could contribute to the demonstrated association between distance to the closest health facility and neonatal mortality.

When stratifying by the socioeconomic determinants the association between neonatal mortality and distance disappeared in the Kinh group and in the groups with higher education and better household economic status. It is well known that the people living in remote areas in Vietnam in general are more disadvantaged when it comes to economy and education [32, 33]. There is also a higher density of ethnic minorities in the remote areas [26, 28]. This association between socio-economic factors and distance has been taken as an argument many times for the disadvantaged position of these vulnerable groups [33]. Our results indicate that distance is also an important factor for neonatal survival among the least privileged, and may be an important piece in explaining the previously described increased mortality risk for newborns of mothers with low education and ethnic minority background [21, 34]. Lack of resources for transportation and costs associated with facility delivery, poor understanding of the dangers of delivery and the importance of preventive delivery services might be factors hindering mothers in overcoming the obstacle of distance to health facilities.

This study is population based, including all neonatal deaths that occurred in the study area during 18 months. By applying a case-referent design we get a good sample to study determinants of neonatal mortality even if not all neonatal mortality cases were identified, since the referents represent the study population. However, we still believe that we have managed to capture and identify most neonatal deaths through the data collection system set up by the research group, not relying primary on faulty health records but on first hand information from local care givers. The primary limitation of this study can be found in the geographical measurement. As discussed above, distance is a difficult entity to pinpoint and it is up for discussion whether straight-line distances are really a trustworthy proxy for the true distance. Straight-line distance has however been frequently used in previous studies. Another limitation in relation to measuring distances is the diversity of geographical traits of the study area, which consists of both mountainous areas as well as flatlands and coastal areas. It can be argued that a straight-line distance in a mountainous area is not equal to the same distance in the flatlands. However, even this would rather strengthen the association between distance and neonatal mortality, since most neonatal deaths were found in the mountainous area, as depicted in figure 1.

We have used linear distances for our analyses. Two-dimensional road distances were manually measured using the ArcGIS software as well. These distances were proportional to straight-line distances with a factor of 1.4. Considering the imperfection of the concept of road distance we chose not to use them in the analysis. In real life there are a lot of components to consider when calculating the true distance, like slope, road quality and temporary obstructions along the road. There is also a tendency in any GIS program to underestimate the true length of a geographic line due to short-cutting and generalizations [35]. Straight-line distances are on the other hand well defined and calculated consequently and may act as a proxy for the true distance. Even if the true distance between home of the respondent and the closest health facility or the facility where delivery care was first sought would be refined by the use of more detailed geographical information, there would still be a question about transportation and other constraints on care seeking. The perception of physical accessibility, of which the true distance is one component only, plays an important part for the first delay in care seeking and the ability to pay and lack of transportation may for example play a crucial part in the second delay [8]. Shannon et al argue that what actually counts is the total effort made to reach a health facility and suggest that perceived travel time could be a good approximation of this effort [11]. In our data there is neither a difference in straight-line distance travelled nor in perceived travel time between cases and referents, further strengthening the use of straight-line distance as an approximation for geographical constraints on delivery care seeking.

The Vietnamese government has for many years prioritized primary health care and managed to achieve a good coverage of health facilities. This is reflected in our results by the relative short distances to a health facility for most of the mothers. Despite this, we can show that there is an association between the distance to the closest health facility and neonatal mortality. To further expand the numbers of CHCs, however, does not seem like a viable option considering the already short distances in this setting. The quality of roads might be an area where improvements may affect facility delivery rates. However this is usually beyond the scope of the health care system. In other settings, where travel time to the health system is long, trials with maternity waiting homes (MWHs) have been made. A Cochrane review has however concluded that there is not enough evidence to support the effectiveness of such facilities [36].

Distance from home to the closest health facility was associated with neonatal mortality risk. Delivery care utilisation could partly explain this risk elevation since distance was an important factor for the decision to deliver at a health facility. To mothers who were poor, had low education and belonged to an ethnic minority distance to the closest health facility was an important determinant for neonatal survival. Mothers, especially from socioeconomically disadvantaged groups, must therefore be facilitated to overcome the constraints of distance and encouraged to deliver at a health facility.