Parity and institutional delivery in rural Tanzania: a multilevel analysis and policy implications

Abstract

Objectives We assess the extent to which the use of healthcare facilities for childbirth varies by parity, conditional on socio-economic, psychological and health characteristics. We also assess differences in the determinants of institutionalized delivery for first-time mothers and multiparous, and explore village-level variations in observed relationships.

Methods Survey data from a three-stage cross-sectional cluster sample of 1205 women from a rural district of Tanzania were analysed using random-intercept multilevel models.

Results Use of health facilities for delivery was low (39%), with odds of institutionalized delivery three times higher among nulliparous women (0 children prior to current delivery) compared with women with one to four children; and 30% lower among women with five or more children compared with those with one to four children. In parity group analyses, women with at least some education and women with more than three antenatal care visits had higher odds of institutionalized delivery among nulliparous. Belief in the importance of institutionalized delivery increased the odds of delivering in a facility among multiparous women; so did health insurance for women with five or more children. We found a significant variation in institutionalized delivery among multiparous women based on their village of residence (one to four and five or more children), but these variations were not observed among nulliparous women.

Conclusion Parity is a pivotal determinant of the use of health facilities for delivery, and its significance varies by village of residence; hence, interventions targeting women according to parity may increase the use of facilities for delivery in rural Tanzania. Future research should focus on the village-level characteristics that influence institutionalized delivery in multiparous.

KEY MESSAGES.

• The relationships between individual characteristics, place of residence and institutionalized delivery vary significantly by parity group.

• For first-time mothers, education and antenatal care were strongly associated with institutionalized delivery; however, institutionalized delivery did not vary by village of residence in this group.

• Attitude towards the importance of institutionalized delivery for health was the most important covariate for multiparous women, with significant village-level variations in the outcome.

• These findings call for modelling policies to women’s reproductive life-stage and residential context.

Introduction

Women in rural Africa bear the heaviest maternal mortality tolls (WHO 2009). Unfavourable social status, coupled with geographic isolation and lower access to life-saving health services during childbirth puts rural African women at great risk of adverse maternal outcomes (Walraven et al. 2000; Glei et al. 2003; Gage and Calixte 2006). In Tanzania, maternal mortality was estimated at 950 per 100 000 live births in 2004. And, only 47% of all births in rural areas in Tanzania were attended by a skilled health worker, compared with 83% in urban areas (WHO 2009). In the region of Kigoma, in remote western Tanzania, 76% of women reported major problems accessing health care compared with 62.1% nationwide (National Bureau of Statistics 2005).

Access and utilization of medical care for delivery can avert maternal death, as most pregnancy-related complications occur at time of delivery or soon thereafter (AbouZahr 1998). Therefore, understanding the reasons why Tanzanian women do not use health care facilities for delivery is an important policy question.

According to the ecological perspective on the production of health and illness, different factors in an ecological hierarchy affect both human behaviours and health outcomes. These include the ‘micro’ or individual level, the ‘meso’ or family and community level and the ‘macro’ or higher order levels of organization where, general economic, social and political decisions are made (county, country, etc.). In this study, we focus on the individual and village levels.

Individual level

At the individual level, the effect of socio-demographic characteristics, such as education and age, on institutionalized delivery has been extensively studied in developing nations (AbouZahr and Wardlaw 2003; Magadi et al. 2007). Parity, however, has had little policy attention, although studies in various nations suggest a strong and consistent relationship with the use of maternal health services. In general, women are more likely to use maternal health services with their first-born child than they are with subsequent pregnancies (Wong et al. 1987; Obermeyer and Potter 1991; Elo 1992; Obermeyer 1993; Bhatia and Cleland 1995; Raghupathy 1996; Magadi et al. 2000, 2007; Stephenson and Tsui 2002; Ekele and Tunau 2007; Gabrysch and Campbell 2009). Parity affects health-seeking behaviours by shaping perceived risk associated with first pregnancies, and by creating time and financial constraints for families with large numbers of children (Gabrysch and Campbell 2009).

In addition to a direct effect on utilization, parity may affect other individual determinants of maternal health services utilization. The life-course perspective proposes that health and health decisions depend on individuals’ life stage, which is shaped by exposures to physical, social, environmental and historical factors from birth to old age (Ryder 1965; Riley 1987; Elder 1998a,b; Ben-Shlomo and Kuh 2002; Halfon and Hochstein 2002). Women’s life-stage pacing and spacing are defined by chronological transitions into social roles as wives, mothers, students, workers and caregivers (Moen 2006). Childbearing and childrearing define life-stages within which any given social determinant is differently relevant. The reasons why women use health care facilities for delivery are expected to change by parity because the availability and significance of health protective resources vary based on women’s reproductive career (Gabrysch and Campbell 2009). For instance, in low-income areas, while motivation and education can positively affect institutionalized delivery for first-time mothers, they are less likely to matter for women who have already had children, and have formed a worldview on the delivery system based on their own experience (Navaneetham and Dharmalingam 2002). First-time mothers may rely heavily on external resources (financial and knowledge) to access and utilize health services, given the value placed on the first-born child. In India, for instance, women’s natal families get involved to improve access to care for their daughters’ first-born child (Navaneetham and Dharmalingam 2002). Finally, context that facilitates access to health services may be more relevant for multiparous because of the low financial and instrumental support they receive from family members compared with nulliparous.

Village level

Beyond individual-level factors, the community context, including the compositional, economic, physical and social aspects of the areas within which individuals operate, affects their decisions regarding their lives and health (Sampson 2001, 2003; Morenoff 2003), and there is evidence this effect is greater for non-discretionary care such as childbirth. In rural Mali, community factors account for ∼42% of the variance in institutionalized delivery, compared with 17% for antenatal care (Gage 2007). Lack of availability of and distance from delivery facility or trained personnel, adverse physical environment, such as mountainous areas, poor road conditions and low economic development at the community level have been repeatedly, although not consistently, identified as determinants of maternal health services use in the literature (Magadi 2003; Montgomery and Hewitt 2005; Gage and Calixte 2006; Pebley et al. 2006; Gage 2007). Social processes such as social norms and cultures favouring traditional birth also can hinder institutionalized delivery (Adamu and Salihu 2002; Van Den Broek et al. 2003; Stephenson et al. 2006; Gage 2007). Neighbourhood composition, including ethnicity and the proportion of women in the community using healthcare facility to deliver, has also been identified as a determinant (Stephenson et al. 2006). The extent to which these community factors affect women at different stages in their childbearing career has not been explored.

In this study, we test the role of parity as a determinant of the use of healthcare facilities for delivery in rural Tanzania. And because women’s decisions about their health are made within a social context, we further assess the role of village-level factors in delivery decisions for women with different parities.

Methods

Study setting

We used data from a cross-sectional survey of a representative sample of Tanzanian women living in the District of Kasulu in Tanzania (population of 630 000). The District of Kasulu is an isolated rural area, in the administrative Region of Kigoma, which nears the western border of the Democratic Republic of Congo. Most residents of the district are farmers, from the Muha tribe, and speak Kiha, the local language, and Swahili, the national language. The main town in the district of Kasulu, with ∼33 000 residents, was omitted from the study to reduce population heterogeneity. More information about the setting can be found in Kruk et al. (2008).

Sampling and data collection

The sampling methods relied on a three-stage representative cluster sampling of women living in rural Kasulu District. In the first stage, 50 villages were randomly selected from a total of 89 villages, with probability proportional to village size, determined using the 2002 Tanzania census. The villages in rural Kasulu are divided into subvillages approximately similar in size. In the second stage, one subvillage was randomly selected among subvillages in each village. In the final stage, subvillage leaders provided an exhaustive list of households, and 35 households were randomly selected from each subvillage, out of a total of 100. From each household, women aged 18 or older who had a child within the 5 years prior to study recruitment were invited to participate. The study received approval from both the National Institute for Medical Research in Tanzania and the Institutional Review Board at the University of Michigan.

Face-to-face surveys translated into Swahili and back-translated into English for accuracy were administered to a total of 1205 (91% response rate) eligible women between June and mid-July of 2007. The questionnaires were administered by two teams of interviewers fluent in Swahili, with at least one interviewer fluent in Kiha. Each interview took ∼30 min, and an observer checked the reliability of the survey administration through daily field observations. The surveys asked questions about maternal health services utilization, childbirth history, household composition and assets, perceptions and knowledge about the local health system and barriers to health services utilization.

Measures and data analysis

The use of healthcare facility for delivery was coded as a dichotomous variable (1 = used a government, mission or private healthcare facility for delivery; 0 = delivered at a friend’s home or own home). The main predictor, ‘prior parity’ was assessed via a question about the number of children respondents had prior to the current delivery being investigated. Prior parity was categorized as no child (nulliparous), one to four children (multiparous) and five or more children (grand multiparous). This categorization is medically and socially relevant to the study of health services utilization. Since Solomons’ study of pregnancy outcome, it is widely accepted that women with more than four children are at increased risk of adverse maternal outcome (Solomons 1934); and thus high parity is likely to affect women’s self-appraisal of the risks and benefits of childbirth. Socially, however, women with a high number of children and uncomplicated deliveries may be attempted to bypass institutionalized delivery, especially if they did not encounter any complications (Stephenson and Tsui 2002). Women pregnant with their first-born child, on the other hand, may feel anxious to have a healthy child, and consequently more frequently use healthcare facility for delivery (Gabrysch and Campbell 2009). Women’s beliefs were measured with a self-assessed question on respondents’ beliefs of the importance of delivering in a healthcare facility for the health of the child and mother (1 = very important; 0 = important to not important). This categorization was necessary as most women’s response choices fell between the first two categories.

In addition to parity, other socio-demographic characteristics relevant to the study of maternal health service utilization were included (World Bank 1999; Bloom et al. 2001; Van Den Broek 2003; Smith et al. 2004; Stekelenburg et al. 2004; Montgomery and Hewitt 2005; Magadi et al. 2007). These were age at childbirth (1 = 35 or older, 0 = younger than 35), marital status (1 = married; 0 = other), education (1 = some formal education; 0 = no formal education) and household poverty (1 = in the poorest household wealth index quintile). Education was used as a dichotomous variable for ease of interpretation, and because preliminary analyses showed similar associations for having less than high school education and high school education with the outcome, without substantive changes to the coefficients of the other covariates. The household wealth index was obtained by performing a principal components analysis based on 10 household assets (radio, bicycle, number of bed-nets, etc.). Households were then allocated into wealth quintiles. More information on the household wealth index can be found in Kruk et al. (2008).

The use of a healthcare facility for delivery may also be influenced by women’s appraisal of threat, which is shaped by their knowledge, previous experiences such as complicated pregnancies, behaviour/treatment of attendants and personal biases and beliefs (Corbin 1987; Patterson 1993; Lazarus and Folkman 1984). One item asking whether respondents had at least one stillborn child or infant death was included in the analysis (1 = yes; 0 = no). We also included perceived quality of the nearest health facility, such as dispensary, health centre, hospital (1 = excellent, very good, good; 0 = fair or poor); and satisfaction with antenatal care (1 = very satisfied, fairly satisfied; 0 = fairly dissatisfied, very dissatisfied). In preliminary analyses, however, the latter two variables were not significantly related to the outcome, and did not significantly contribute to the estimation models. They were, therefore, not included in the final analyses.

Access to resources was measured as insurance coverage (1 = insured; 0 = not insured) and use of antenatal care services (1 = at least four visits; 0 = less than four visits), whereas health status was measured as self-rated health on a five-point Likert scale (1 = very good to 5 = very bad). Four antenatal visits were used based on findings from a World Health Organization (WHO) randomized controlled trial. In this study, a minimum of four visits prior to childbirth had the optimal effect on positive birth outcomes (WHO 2002).

We included four measures of community characteristics (community defined here as village), often cited in the literature: village physical characteristics (accessibility: distance from facility, road network), economic characteristics (village poverty) and social characteristics (female empowerment and freedom of choice: female literacy). Individual responses were aggregated at the village level and corresponding village estimates were computed for each variable, and assigned to the corresponding respondents. These were as follows: distance to the nearest facility in kilometres; percent female literacy in the village; type of road network (mostly primary, mostly secondary or mostly tertiary roads) and village poverty measured as the percent residents in the lowest quintiles within each village.

Descriptive statistics were computed taking the complex design and nested-structure of the data into account. Rao–Scott chi-square tests for categorical variables and univariate linear combinations comparing variable means were performed to describe the total sample and parity groups.

Because respondents living in the same village are more alike to each other (in observed and non-observable ways) than they are alike to respondents in other villages, and as some determinants of utilization may function at an area level, we used a random-intercept multilevel model (Raudenbush and Bryk 2002). First, the data were fitted to an empty random-intercept model to describe the total variance in the outcome attributable to context. Subsequently, parity was entered into the model, followed by other individual characteristics and the four village characteristics of interest. To test our second set of hypotheses of parity as a moderator, we ran similar models as described earlier within parity groups. Multilevel analyses were performed using the command xtlogit STATA 11® (Rabe-Hesketh and Skrondal 2008).

Results

In our sample of 1205 women surveyed, 1183 had complete parity information (98%) and were included in the analysis. The sample mean age was 28.7 (Standard Error (SE) = 0.24), and most women were married, with some education and lacked health insurance. Overall, the majority believed the healthcare system to be important for the health of the mother and child. More than one of four women had experienced a stillborn baby or infant death prior to their current delivery. Use of facility for delivery was low (39%), and only one of five respondents had an optimal number of antenatal visits according to WHO’s recommendation. More information about the sample characteristics is presented in Table 1.

Table 1.

Descriptive statistics for the sample and by parity

	N	Total sample		Nulliparous (0 previous children)		Multiparous (1–4 previous children)		Grand multiparous (5+ previous children)		Design-based P-value
		Mean or %	SE	Mean or %	SE	Mean or %	SE	Mean or %	SE
Main predictor
Parity distribution (% respondents in each group)	1183	–		11.00		54.00		33.00		***
Number of children (mean ± SE)	1183	3.5	0.09	0.00		2.44	0.04	6.46	0.08	***a
Individual characteristics
Age at last childbirth (mean ± SE)	1180	28.73	0.24	20.80	0.30	26.42	0.20	35.34	0.36	***a
Age at last childbirth (% 35+)	1173	22.40		–		7.80		54.60		***
Marital status (% married)	1182	95.80		93.80		96.08		95.89		***
Perceived importance of delivering in a facility for the mother and child’s health (% very important)	1184	74.30		78.30		74.30		72.80
Household poverty (% in the lowest wealth index quintile)	1168	18.58		27.20		18.54		15.84		*
Education (% with some education)	1183	72.36		71.32		74.70		68.72
Health insurance (% insured)	1182	12.44		13.18		11.45		13.88
Optimal antenatal care visits (% with 4+ visits)	1182	44.35		44.96		42.02		51.16		**
Ever had a stillborn or deceased infant (% yes)	1183	22.40		0.78		15.66		41.03		***
Self-rated health (1 = very bad to 5 = very good; mean ± SE)	1173	4.17	0.09	4.10	0.06	4.29	0.53	4.00	0.05	***b
Village characteristics
Distance to nearest facility (km; mean ± SE)	1183	1.68	0.46	1.96	0.61	1.80	0.50	1.45	0.43
Female literacy (% women in the village with some education)	1183	0.72	0.02	0.72	0.03	0.73	0.02	0.73	0.02
Road network or quality (1 = primary to 3 = tertiary; mean ± SE)	1183	2.32	0.12	2.33	0.14	2.31	0.12	2.35	0.13
Primary (%)	1183	25.36		25.58		25.45		25.13
Secondary (%)		16.65		16.28		17.92		14.62
Tertiary (%)		57.99		58.14		56.63		60.26
Village poverty (% village residents in the lowest wealth quintile)	1183	0.25	0.02	0.26	0.02	0.25	0.02	0.26	0.02
Outcome
Delivery in facility (% delivery in government, mission, other)	1183	39.48		62.02		38.55		33.59		***

Design-based analyses are based on Rao–Scott chi-square tests for categorical variables and univariate linear combinations comparing variable means by parity group.

^aMean difference is significant for the three categories at 0.05.

^bMean difference between the first and the third group is significantly different at 0.05.

*P < 0.10, **P < 0.05, ***P < 0.001.

Bivariate analyses adjusted for the multistage survey design show that parity was associated with age, household wealth, optimal number of antenatal visits, adverse maternal health outcomes and self-rated health (Table 1).

Multiparous with one to four children were older than first-time mothers, and marginally less likely to live in households in the lowest wealth quintile. They were at greater risk for not completing at least four antenatal visits; they also more frequently reported adverse infant health (stillborn or infant death). Women with one to four children were significantly more likely to have used a healthcare facility for their latest delivery and to live in villages nearest to healthcare facilities.

When compared with respondents with five or more children, women with one to four children were younger, lived in poorer households and reported better self-rated health. Although they were significantly less likely to have completed at least four antenatal visits, they were also significantly more likely to deliver in facility than multiparous with five or more children.

Women in the different parity groups lived in villages that were not statistically different with regard to their distance from the nearest facility, female literacy, road condition or village poverty (Table 2, Model 3).

Table 2.

Random intercept models for institutionalized delivery in rural Tanzania

Variables in the model	Model 1		Model 2		Model 3
	OR	P	OR	P	OR	P
Individual-level characteristics
Parity prior to current child (1–4 children)
Nulliparous (0)	3.39	0.000	3.36	0.000	3.39	0.000
Grand multiparous (5 or more)	0.79	0.139	0.70	0.043	0.70	0.041
Marital status (married)			1.84	0.108	1.86	0.103
Perceived importance of delivering in a facility for the mother and child’s health (very important)			2.06	0.000	2.05	0.000
Household poverty (lowest wealth index quintile)			1.20	0.361	1.22	0.318
Education (some education)			1.22	0.255	1.21	0.271
Health insurance (insured)			1.71	0.032	1.65	0.044
Optimal antenatal care visits (4+ visits)			1.35	0.051	1.34	0.060
Ever had a stillborn or deceased infant (yes)			1.11	0.593	1.11	0.584
Self-rated health (1 = very bad to 5 = very good; mean ± SE)			0.81	0.045	0.81	0.046
Village-level characteristics
Distance to nearest facility (km; mean ± SE)					0.89	0.085
Female literacy (% women in the village with some education)					1.47	0.782
Road network or quality (1 = primary to 3 = tertiary; mean ± SE)					0.13	0.251
Village poverty (% village residents in the lowest wealth quintile)					1.22	0.430
Number of observations	1183		1153		1153
Number of villages	50		50		50
Log likelihood of the model	−664.7	0.000	−631.0	0.000	−628.2	0.000
Random effects
Village-level variance	1.87	<0.05	1.88	<0.05	1.69	<0.05
Variance explained by unobserved village characteristics (ICC)	0.36	<0.05	0.36	<0.05	0.34	<0.05
Percent variance explained by observed village characteristics	–		Baseline		6.98
Log likelihood test for the random intercept	225.2	0.000	210.8	0.000	188.7	0.000

Multiparous with 1–4 children is the reference; OR, odds ratio. ICC describes the amount of outcome variance among women from different villages that cannot be explained by village characteristics.

As expected, we found a strong main effect of parity on the use of healthcare facility for delivery (Table 2). This relationship remained significant and unchanged upon adding individual and community characteristics for nulliparous (Odds Ratio (OR) = 3.39, P = 0.000).

To determine the extent to which differences in local context explain individual-level variations in institutionalized delivery, we fitted the data to a series of random-intercept models (Table 2). In the empty model, which is not shown here, we found a significant village-level variance of 1.71 (SE = 0.03), with an intra-class correlation (ICC) coefficient of 0.341. In other words, village-level differences in unobserved characteristics account for about one-third of individual level variations in the use of facility for delivery in rural Tanzanian women. This contextual effect is barely modified by the introduction of the four village characteristics frequently identified in the literature as determinants of facility for delivery (from 34.1 to 33.9%, corresponding to <1% change). Except for distance from facility, which had a marginal negative association with institutionalized delivery (OR = 0.89, P = 0.085), none of the observed village variables were significant.

Within parity categories (Table 3), individual-level determinants of institutionalized delivery included education (OR = 4.54, P = 0.005) and four or more prenatal visits (OR = 2.57, P = 0.042) for nulliparous; beliefs of the importance of health care for the health of the mother and child was significant only for multiparous (OR = 2.03, P = 0.004) and grand-multiparous (OR = 2.79, P = 0.004), possibly because of earlier experiences with the health system combined with self-assessed risk, related to age and parity. To further understand the link between previous experiences with the health system, we tested the relationship between satisfaction with the nearest healthcare facility on use of facility for delivery, including dispensary (P = 0.537), health centre (P = 0.336), hospital (P = 0.994) and satisfaction with antenatal care (P = 0.583). Neither were significant in bivariate analyses, nor contributed to the multilevel models. They were, therefore, not included in the final models.

Table 3.

Random intercept models for institutionalized delivery by parity group in rural Tanzania

	Nulliparous (0)		Multiparous (1–4)		Grand multiparous (5 and more)
	OR	P	OR	P	OR	P
Individual-level characteristics
Marital status (married)	2.64	0.320	1.52	0.422	1.88	0.421
Perceived importance of delivering in a facility for the mother and child’s health (very important)	2.07	0.174	2.03	0.004	2.79	0.004
Household poverty (lowest wealth index quintile)	1.46	0.518	1.00	0.995	1.85	0.124
Education (some)	4.54	0.005	1.09	0.711	1.15	0.661
Health insurance (insured)	1.54	0.562	1.34	0.411	2.74	0.022
Optimal antenatal care visits (4+ visits)	2.57	0.042	1.18	0.442	1.35	0.305
Ever had a stillborn or deceased infant (yes)	(omitted)		1.53	0.119	0.92	0.785
Self-rated health (1 = very bad to 5 = very good; mean ± SE)	0.51	0.090	0.97	0.784	0.70	0.054
Village-level characteristics
Distance to nearest facility (km)	0.96	0.528	0.88	0.079	0.88	0.125
Female literacy (% women in the village with some education)	0.25	0.409	1.09	0.954	2.57	0.581
Road network or quality (1 = primary to 3 = tertiary; mean ± SE)	1.52	0.847	0.13	0.275	0.18	0.427
Village poverty (% village residents in the lowest wealth quintile)	1.06	0.860	1.11	0.687	1.39	0.279
Number of women	125		649		379
Number of villages	47		50		50
Log likelihood of the model	−73.4	0.210	−364.6	0.065	−203.6	0.048
Random effects
Village-level variance	0.31	>0.05	1.80	<0.05	1.94	<0.05
Variance explained by unobserved village characteristics (ICC)	0.09	>0.05	0.35	<0.05	0.37	<0.05
Percent variance explained by observed village characteristics	22.45		8.15		5.56
Log likelihood test for the random intercept	0.36	0.274	93.45	0.000	47.11	0.000

Multiparous with 1–4 children is the reference; OR: odds ratio. ICC describes the amount of outcome variance among women from different villages that cannot be explained by village characteristics.

Determinants of institutionalized delivery differ in the latter two groups in that insurance increased almost 3-fold the odds of using facility for delivery among women with five or more children, and self-rated health had a negative relationship with institutionalized delivery for women in this group only, most likely related to the correlation between parity, health and age.

In the empty random-intercept models by parity group, 21.7, 38.6 and 34.4% of the outcome variance was attributable to unobserved village differences for nulliparous, and multiparous with one to four and with five or more children, respectively. Context accounts for a significant part of the individual-level variations in institutionalized delivery, except for nulliparous, where both the ICC and the test of model significance failed to support a significant village-variation. In the final models with individual and village characteristics, only 11% of the variance of the outcome was attributable to village differences among nulliparous; however, we found that context significantly explained 39% of the variance for multiparous. Given small changes in the model log-likelihood statistics between Models 1 and 2 for nulliparous and grand-multiparous, one can conclude that the village-level variables in Model 2 explained little of the village variation on institutionalized delivery for these two groups. On the other hand, models for women with one to four children showed a significant change in the log-likelihood statistic, suggesting that the same village-level variables explained 4% of the outcome variance (ρ = 0.39–0.35). Of these, only distance from nearest facility approached significance.

With the high correlation between age and parity (correlation coefficient = 0.77, P < 0.001), the effects of age were largely explained by parity, thus these results are not discussed here, but are presented in the Appendix. When one was substituted for the other, age and parity seemed to behave similarly, except the relationship with parity was of greater magnitude on health services utilization than that of age, and it was slightly stronger for the covariates when compared with age (Appendix). Parity explained the effect of beliefs and insurance on health services use to a greater extent than age. Finally, while age reduced the effect of antenatal care, parity had the opposite effect.

Discussion

We set out to investigate the role of parity on the use of healthcare facility for delivery in rural Africa, and the extent to which the relationship with individual and village-level characteristics varies by parity group.

We confirmed that parity is an important determinant of institutionalized delivery in rural Tanzania, as it was in other low-income countries (Obermeyer and Potter 1991; Obermeyer 1993; Bhatia and Cleland 1995; Raghupathy 1996; Magadi et al. 2000, 2007; Ekele and Tunau 2007; Gabrysch and Campbell 2009). We further demonstrate that individual- and village-level factors played different roles in delivery care use at different stages of women’s reproductive career.

First-time mothers’ use of healthcare facilities for delivery was almost exclusively determined by individual characteristics, including four or more antenatal visits and formal education, with little village-level variability in facility use. In rural Tanzania, four or more antenatal care visits can increase the odds of using healthcare for delivery 2.57-fold for the first delivery. We further found that first-time mothers did not use antenatal care services in greater frequencies than multiparous mothers. But parity-disaggregated models showed that having four or more visits was associated with three times greater odds of institutionalized delivery for nulliparous women, whereas it had no discernible association on multiparous women. This suggests as previously reported in the literature that women may be receiving different advice during antenatal care (Rockers et al. 2009). In this instance, nulliparous women could be provided with more appropriate advice to deliver in facilities during their antenatal care visits.

In addition to antenatal care visits, first-time mothers’ education increases institutionalized delivery almost 4-fold, a magnitude greater than any other individual characteristic. By contrast, education is not a significant predictor of facility use by multiparous. This differential role of education by parity has not been previously described. It suggests that school-based health education may under emphasize childbirth risks for multiparous women. It may also be a consequence of cohort differences in educational content promoting institutionalized childbirth in more recent years. Future studies should further explore the pathways through which education impacts delivery choices by parity group.

For women who have had children in the past, the use of healthcare facility at childbirth is associated with a different set of individual characteristics, including believing delivery care is important, having health insurance, and to a lower degree, good health. Among women with very high parity (five children or more), who were the least likely to use facility for delivery, health insurance was associated with a 3-fold increase in the odds of using facility for delivery when compared with women with one to four children, and this relationship was independent of household poverty. The higher frequencies of antenatal care visits among women with five or more children had no bearing on their use of facility at the moment of delivery. Neither did education, despite levels similar to first-time mothers’. These findings support previous results that antenatal care is no guarantee for delivery in a healthcare facility among multiparous; rather, it is affordability in the form of health insurance, and past experience with the health system that constitute the drivers of institutionalized delivery in this group. Both qualitative and quantitative studies have found that women’s perception of quality of care can affect where they deliver their child. In a review article, Gabrysch and Campbell (2009) identify the dimensions of care shown to influence institutionalized delivery, including interpersonal relationships with the medical staff; cultural appropriateness of delivery services; hygiene, access to water, availability of drugs and aseptic practices; and the concentration of medical staff in a geographic area. Although these measures were not included in this study, the findings that multiparous use antenatal care at the same rate as nulliparous, yet use healthcare facilities for childbirth less often suggest, as already proposed by others, that improving the quality of care and thus women’s satisfaction with institutionalized delivery of the first child may motivate women to return for care (Kruk et al. 2009).

This is the first study to report variable effects of health insurance by parity. It is possible that instrumental and financial support from the extended family is more readily available to nulliparous given the greater value placed on first-born children (Navaneetham and Dharmalingam 2002). Therefore, financial accessibility through health insurance may be more relevant for multiparous, but this hypothesis needs further investigation.

Women in this study lived in villages that were similar in terms of available health facilities, road conditions, economic standing and female literacy. But we found significant village-level variation in the outcome variance, which remains largely unexplained (35 and 37% unexplained variance for women with one to four, and five and more children, respectively). This study is the first, to the authors’ knowledge, to demonstrate that village-level effects vary only for women with prior childbirth. For these women, especially when childbirth was uneventful, individual motivation for institutionalized delivery might be relatively low. Therefore, external factors, including village social and physical characteristics may be of greater importance for their decisions about where to deliver their child. As the village physical and economic characteristics in this study did not significantly explain village variations in institutionalized delivery, it is plausible that social processes, not included in this assessment, are more relevant to institutionalized delivery among multiparous in rural areas. For instance, using the Demographic and Health Survey, Stephenson et al. (2006) illustrates how community-level misconceptions, fears of medical institutions and admiration for silent birth determine medical decisions in India and Benin. Qualitative studies further found that on top of structural factors, alignment in the belief systems between traditional birth attendants (TBA) and rural residents, rivalry between TBA and healthcare workers, and cultural views of childbirth as women’s personal battle also emphasize the role of community belief systems on women’s use of healthcare delivery in rural Benin, Nigeria, Uganda and Indonesia (Okafor and Rizzuto 1994; Grossman-Kendall et al. 2001; Kuyomuhendo 2003; Titaley et al. 2010). It is plausible that such traditional rules and expectations are embodied to a greater extent by multiparous, because of their extended exposures to cultural norms (related to the correlation between age and parity), and economic dependence. Alternatively, measurement issues could be at play. Aggregates of individual measures, such as the ones used in the study, have been criticized for their low variability, with risks of high standard errors, and failure to detect significance (Diez-Roux 2007). Future work should explore a broader range of village-level characteristics, including aspects of the dominant social discourse on birth practices and their variance based on parity and place.

In addition to village social and physical characteristics, village-level variations in the quality of maternal health services could explain observed differences in delivery care. To the authors’ knowledge, variations in the quality of services at the village level have not been studied, and need further investigation.

The results of this study should be interpreted considering its limitations. First, because of its cross-sectional design, the associations identified here do not imply causality. Second, as noted earlier, aggregate village-level measures were unable to explain a substantial portion of variance at that level. Third, our sampling technique did not allow for household-level analysis, although household composition, dynamics and cultures do influence the use of facility for delivery. For instance, household poverty does not measure the extent to which families are willing to mobilize resources to cover the cost of delivery in a facility; neither does it measure familial beliefs about the relative value of TBA, but both are significant predictors of institutionalized delivery. If these data were available, we expect that the gap by parity group would be wider, as most household resources are likely to be mobilized for first-time mothers (Gabrysch and Campbell 2009). Fourth, the estimates may suffer from omitted variable bias. For instance, the relationship between distance to facility and usage can be mediated by length of labour, which is usually shorter among multiparous. Last but not least, the findings are limited to a rural region in Tanzania, therefore are not generalizable to other settings.

Policy implications

We confirm that parity plays a pivotal role as a determinant of institutional delivery. We further found that widely recognized individual and contextual factors vary in their relevance for first-time mothers and multiparous. Given the slow progress in efforts to increase the rate of delivery by skilled attendants in sub-Saharan Africa, refocusing research and policy attention on parity and the context within which women live is warranted.

For first-time mothers, policies aimed at increasing institutional delivery should focus on individual characteristics. Considering that 71.3% had some education, and fewer than half had four or more antenatal visits, improving the educational outputs of young girls, and maintaining optimal prenatal visits according to the WHO’s current recommendations are relevant policy options. These interventions can be implemented across village, as village-level effects on facility use for childbirth did not vary.

The policy options for multiparous and particularly women with more than four children are less straightforward. Reducing the costs of care through comprehensive insurance may reduce an important barrier to access, but as found in other parts of the world, financial access is only a partial fix of a much larger problem. Better communication during antenatal care about the unpredictability of birth complications and the importance of delivery in the health system may be particularly important in countering the notion that uneventful deliveries inoculate women against risk. In addition, some of the substantial unmeasured village-level effect that we found for multiparous may be due to variation in advice on the necessity of institutionalized delivery. Future research should explore this and other sources of village-level effects on institutionalized delivery decisions among multiparous prior to policy development. Given that the ultimate aim of institutionalized delivery is saving maternal and newborn lives, this latter objective could also be advanced by expanding family planning to assist women in having their desired number of children, which in turn can reduce the number of grand multiparous.

Conclusion

In conclusion, the study findings support the need for parity-sensitive interventions, which tailor programs and incentives to women’s reproductive life-stage. These should further account for the variety of village-level and individual factors among women of different parity in rural settings. For example, while formal education may improve delivery care usage for first-time mothers, a more reproductive health-specific education should be tailored to women who already have children to sustain institutionalized delivery over the reproductive life-course. To this end, good quality advice on delivery risk during antenatal care visits, along with other village-level interventions will likely be needed to encourage multiparous to give birth within the health system.

Funding

This study was funded by the Averting Maternal Death and Disability at the Mailman School of Public Health at Columbia University; the William Davidson Institute of the Ross School of Business at the University of Michigan; grants from the National Institutes of Health (DA017642; DA 022720); and the University of Michigan School of Public Health.

Appendix

Table A1.

Random intercept models of institutionalized delivery with age as the main predictor

Variables in the model	Model A		Model B		Model C		Model D
	OR	P	OR	P	OR	P	OR	P
Individual-level characteristics
Age at childbirth	0.67	0.021	0.73	0.087	0.73	0.085	1.03	0.885
Parity prior to current child
Nulliparous (0)			–	–	–	–	3.56	0.000
Multiparous (5 or more)			–	–	–	–	0.70	0.074
Marital status (married)			1.78	0.124	1.79	0.121	1.91	0.091
Perceived importance of delivering in a facility for the mother and child’s health (very important)			2.14	0.000	2.13	0.000	2.09	0.000
Household poverty (lowest wealth index quintile)			1.27	0.214	1.29	0.187	1.21	0.350
Education (some)			1.15	0.424	1.14	0.448	1.20	0.301
Health insurance (insured)			1.72	0.025	1.67	0.034	1.62	0.054
Optimal antenatal care visits (4+ visits)			1.26	0.134	1.24	0.155	1.30	0.097
Ever had a stillborn or deceased infant (yes)			0.87	0.423	0.87	0.426	1.13	0.520
Self-rated health (1 = very bad to 5 = very good; mean ± SE)			0.79	0.021	0.79	0.023	0.80	0.041
Village-level characteristics
Distance to nearest facility (km; mean ± SE)					0.90	0.092	0.89	0.082
Female literacy (% women in the village with some education)					1.59	0.728	1.44	0.793
Road network or quality (1 = primary to 3 = tertiary; mean ± SE)					0.17	0.288	0.13	0.251
Village poverty (% village residents in the lowest wealth quintile)					1.20	0.449	1.22	0.425
Number of observations	1193		1161		1161		1144
Number of villages	50		50		50		50
Log likelihood of the model	−688.3	0.000	−652.8	0.000	−650.2	0.000	−621.9	0.000
Random effects
Village-level variance	1.757	<0.05	1.763	<0.05	1.585	<0.05	1.713	<0.05
Variance explained by unobserved village characteristics (ICC)	0.35	<0.05	0.35	<0.05	0.33	<0.05	0.34	<0.05
Percent variance explained by observed village characteristics	–		Baseline		6.80		–
Log likelihood test for the random intercept	223.86	0.000	209.76	0.000	189.22	0.000	190.54	0.000

Multiparous with 1–4 children is the reference; OR: odds ratio. ICC describes the amount of outcome variance among women from different villages that cannot be explained by village characteristics.