Maternal socio-economic indices for prenatal care research in rural China

Abstract

Background: The conceptualization and measurement of socio-economic status (SES) is difficult in developing settings. In the absence of SES indices for women in rural China, we constructed SES indices for prenatal care research, and examined their relation to perinatal care and outcomes. Methods: This study utilized data of 4364 rural women having recently given birth, collected by a cross-sectional survey in three rural Chinese provinces in 2007. Principal component analysis (PCA) was used to construct the SES indices and multilevel logistic regression was use to relate the indices to low birthweight, short exclusive breastfeeding (≤4 months), childbirth at the county or higher level health facility, caesarean section, inadequate prenatal care and no postnatal care. Results: Three separate SES indices (wealth, occupational and educational indices) were obtained from the PCA analysis, capturing maternal, paternal and household SES characteristics. After adjusting for individual level factors, village and township wealth, higher levels of the indices were inversely associated with inadequate prenatal care. Higher occupational status was positively associated with short exclusive breastfeeding and childbirth at the county or higher level health facility, but inversely associated with no postnatal care. Higher educational status was positively associated with no postnatal care. Conclusion: Three SES indices (wealth, occupational and educational) were obtained from this study for prenatal care research. The indices gave mostly varying results on their associations with perinatal care and outcomes, indicating that SES measures may be outcome-specific.

Introduction

Socio-economic status (SES) is an important health determinant, and in studies not focusing on it, it is an important counfounder.^1–3 However, the conceptualization and measurement of SES remain difficult.^2,4 There are standardized measures of SES in developed countries,¹ even though they have been critiqued.^1,4 Defining indicators for measuring SES in developing settings poses a great challenge.² The assessment and classification of the frequently used measures, income, occupation and education, have inherent challenges, and their meaning may not be the same as in developed countries. Furthermore, objective and reliable data sources for these indicators are usually lacking in rural developing settings. In these situations, researchers have resorted to proxy measures to represent SES.^2–5

SES is a multidimensional measure, integrating the individual as well as the aggregated characteristics of the household and context where a person resides.^1–4,6,7 It has also been suggested that SES is context- and outcome specific, implying that a particular definition or indicator may fail to be applicable across contexts and studies.^1,3 Accordingly, a challenge in defining SES, whether in the developed or developing world, has been the limited understanding of these perspectives.¹

Often researchers have used a single SES indicator or used some SES indicators as proxies for others.^1,2 A single SES indicator may fail to capture the multidimensional nature of SES and the use of one SES indicator as a proxy for another may fail to capture the actual constructs measured by the targeted indicator.¹ Without taking into account the different dimensions of SES, it may be inaccurate to conclude that the associations between an exposure and health outcome is independent of SES.^1,6 Amidst these challenges, it behoves on researchers, within the context of their research, to develop appropriate SES indices that have contextual relevance to their research topic and setting.^1,2,3

In developing settings as well as in some developed countries, measuring the SES of women poses greater challenge than that of men.⁸ Many women are not in the paid labour force and data to measure their SES is lacking.⁸ During pregnancy women's SES may influence perinatal outcomes and use of care.^9,10 Currently, there are no reference indices developed to measure women's SES in rural China, making it difficult to select appropriate SES indicators to use in prenatal care research. Constructing SES indices for rural women will not only enable a valid consideration of SES in predicting health outcomes in prenatal care research in these settings, but will also provide a ready reference option for future studies.

Using the opportunity provided by a recent representative population-based cross-sectional survey in rural China, this study aimed (i) to construct maternal pregnancy SES indices that will capture maternal, paternal and household SES characteristics, and (ii) to investigate the associations between the constructed indices and perinatal care and outcomes (hereafter called perinatal outcomes).

Methods

Study design and subjects

This study is based on the maternal health-care survey in six counties in three provinces (Anhui, Chongqing and Shaan'xi) in rural China, conducted in 2007. Targeted women were those who had given birth during 2005–06. The survey was carried out to provide baseline data in preparation for various interventions to improve health care in rural China, under the framework of the CHIMACA (Structural hinders to and promoters of good maternal care in rural China) project. The six counties were selected based on their poverty levels (according to China national classification) and comprised of 55 townships. In total, 485 randomly selected villages from these townships were included in the survey. In rural China, maternal health care is organized at the village, township and county levels. While antenatal care is usually offered at the three levels, delivery is mainly carried out at the township and county health facilities.¹¹

The survey gathered data on women's background characteristics, the use of maternal health care during the last pregnancy, child birth and after delivery, as well as barriers for use of care. Women were identified through doctors and family planning workers, birth registers of township hospitals and by snowballing. Altogether, 4364 were interviewed. Before the interview, the women were contacted in advance either through telephone call or home visit by the township doctor or family planning worker. The interview was conducted by trained interviewers and lasted for 20–30 min. The interview instrument was a 64-item structured questionnaire, which was first piloted and modified accordingly. The questionnaire was made in Chinese, translated into English and checked against the Chinese version by a bilingual researcher. Ethical approval for the project was obtained from the International Center for Reproductive Health, ICRH, Ghent University, Belgium. Local approvals were obtained from Anhui Medical University, Chongqing University of Medical Sciences and Xi'an Jiatong University in Shaan'xi.

Outcome measures

The outcomes studied were low birthweight (<2500 g), short exclusive breastfeeding (≤4 months), childbirth at county or higher level hospital, caesarean section, inadequate use of prenatal care (classified using the Adequacy of Prenatal Care Utilization index),¹² and no postnatal care within 42 days after birth.

Statistical analysis

Maternal SES indices

Women's SES indices were constructed by considering variables describing the women [education (≤primary school, middle school, ≥high school) and occupation (farmer, others)]; their husbands (education and occupation); and the household [household ownership of television, telephone, type of house (building, soil, concrete, others) and family total income during the previous 1 year (as a continuous variable)]. These questionnaire items were subjected to principal component analysis (PCA). Varimax (orthogonal) rotation with Kaiser Normalization was used. Eigenvalues >1 was used as the criteria for extraction, and confirmed by a scree plot test.

In interpreting the rotated components, an item was said to load on a given component if the factor loading was greater than or equal to ±0.50 for that component, and was less than ±0.50 for the other components. Factor-based scores were created by summing the values of a subject on the variables loading on a component and weighted by the factor scores of the corresponding component. The created factor-based scores were then categorized into thirds for further analyses. SPSS version 15 (SPSS, Inc., IL, USA) was used to compute the PCA.

Village and township level wealth indices

The village and township level wealth indices were computed using the household ownership of television, telephone and type of house. First, scores (0–3) were assigned to the responses on each of these variables, with the highest score being assigned to the most affluent category of each indicator. Then a total score was computed for each woman by summing the scores. The maximum total score that could be obtained was 9. The village and township wealth indices were aggregated from the individual scores and classified as thirds. The thirds were named as follows: 1st third—‘Disadvantage villages/townships’; 2nd third—‘Middle advantage villages/townships’; 3rd third—‘Advantage villages/townships’.

Associations between SES indices and outcome measures

Due to the hierarchical nature of the data, the associations between maternal SES indices and the perinatal outcomes were investigated using multilevel logistic regression. The data structure involved three levels: the individual level (level 1); the village level (level 2); and the township level (level 3). Three-part groups of three multilevel logistic models were fitted: Model 1 examined the association between each of the SES indices and each outcome measure. Model 2, in addition to Model 1, adjusted for maternal age (centred), sex of child (0 = boy, 1 = girl), parity (0 = one child, 1 = ≥2 children). Model 3, in addition to Model 2, adjusted for the village and township wealth indices (0 disadvantaged neighbourhoods, 1 middle advantaged neighbourhoods, 2 advantaged neighbourhoods). The odds ratios for the fixed part of the models were computed with their 95% confidence intervals. The random part of the models was estimated by computing the intra-class correlation coefficients (ICC) accompanied by their respective median odds ratios. All models were estimated using the second order penalized quasi-likelihood (PQL2). The R² was used to assess the performance of the models: higher values indicate better fit of the models. MLwiN version 2.19 was used to estimate the multilevel models.

RESULTS

Characteristics of the study participants

The mean age of the women at birth was 27.8 (SD 5.5) years. Of the women, 38% had two or more children; 35% had inadequate prenatal care; 4% had low birthweight babies; 41% caesarean section; 51% had short exclusive breastfed; and 24% gave birth at the county or higher level hospitals. Twenty-nine per cent of the villages fell into the disadvantaged neighbourhood category; 21% into the middle advantaged neighbourhoods; and 50% into the village advantaged neighbourhoods. The corresponding figures for the townships were 31, 27 and 42%.

Maternal SES indices

Supplementary table S1 shows the initial correlations between each of the SES indicators before the PCA. Overall, the correlation coefficients between the indicators ranged from low to moderate, with the highest correlation seen between maternal and paternal occupation (r = 0.46) and lowest between paternal education and type of house (r = 0.14). In the PCA, three components were extracted, which in total explained 56% of the variance in women's SES (table 1). Household income, ownership of television, telephone and housing type loaded on the first component and was named ‘wealth index.’ The second component comprised of maternal and paternal occupation, and was named ‘occupational index.’ The last component was loaded by maternal and paternal education, thus was named ‘maternal educational index’ (table 1).

Associations between maternal SES indices and perinatal outcomes

The associations between the maternal SES indices and the perinatal outcomes are presented in tables 2 and 3 and Supplementary tables S2 and Supplementary Data. Overall, the results from the three models (Models 1, 2 and 3) did not differ from each other. The associations between each of the social class indices and the outcomes were mostly different from each other. None of the SES indices was statistically significantly associated with low birthweight (Supplementary table S2) or caesarean section (Supplementary table S3). Only the occupational index was significantly associated with short exclusive breastfeeding and childbirth at the county or higher level health facility (table 2): being on the highest third of the occupational index was associated with an increased risk of short exclusive breastfeeding, whereas being on the second third was associated with a decreased risk of child birth at the higher level health facility. On the other hand, the three social class indices were similarly associated with inadequate prenatal care, showing decreased risks of inadequate care with the higher thirds of the indices (table 3). In regard to use of postnatal care, the three indices showed different results: wealth index was not associated, occupational index was positively associated and educational index was negatively associated with the use of postnatal care (table 3).

Based on the random part of the models, the results show that the associations between the SES indices and low birthweight seemed not to have any contextual variation, giving an interclass correlation of 0 for both the village and township levels and a median odds ratio of 1 after accounting for the confounding factors and the village and township wealth indices (Tables 2 and Supplementary table S2). On the other hand, the associations between each of the indices and short exclusive breastfeeding (tables 2 and Supplementary table S2), caesarean section (Supplementary table S3), childbirth at the county or higher level hospital (tables 2 and Supplementary table S3), inadequate prenatal care (table 3) and no postnatal care (table 3) had some contextual variations at both the village and township levels. The variation was highest for no postnatal care than for the other outcomes. The variations in most outcomes were higher for the township level than for the village level.

Discussion

We observed low to moderate correlations between the different single SES indicators studied (ranging from 0.14 to 0.46), with the highest correlation seen between maternal and paternal occupation. Three separate SES indices (wealth index, occupational index and educational index) were derived to classify women's SES during pregnancy in the study population. The associations between each of the indices and the studied perinatal outcomes were mostly different from each other, after adjusting for individual and contextual-level variables.

Most previous studies have utilized mainly individual level indicators in developing SES indices for health-care research and in examining their associations with health outcomes.¹ However, the multidimensional nature of SES implies that the use of individual level indicators alone may fail to capture its contextual distribution.^1,2,4,6 In many developing rural settings, in particular, many women rely on their husbands for their economic power and social status. Consequently, incorporating the partner's and household's indicators into women’s individual-level indicators in defining their SES presents a more valid option.² Therefore, in the present study, using the indicators on these several levels, three separate uncorrelated indices were derived to describe women's SES during pregnancy: wealth index, occupational index and educational index pointing to the three main general aspects of SES, namely income, occupation and education.^1,13,14

Often researchers have used some SES indicators as proxies for others. In our study, the correlations between each of the single SES showed low to moderate correlations. This is consistent with the suggestion from other studies.^1,13,14 This finding suggests that, in our study population, the level of education or type of occupation may not be proportionate to household income and ownership of household commodities. Ownership of household commodities such as television, telephone and housing type may depend, besides wealth, on personal preference and the availability of electric utilities in the community. The level of correlations between the single indicators in the present study suggests that each indicator may in fact be measuring peculiar constructs, as has been previously suggested.¹ Thus, using one indicator as a proxy for another may fail to capture the underlying measure of the intended indicator.

A previous study, based on the Demographic and Health Survey (DHS) data from five African countries, used several household and community indicators to construct indices that captured both household and community attributes for women aged 15–49 years.² Three SES indices were obtained in that study, and named as household wealth index, household social index and community endowment index. Regardless that the same indices were obtained from each of the five countries, the proportions of explained variance of the women’s SES varied across the countries. That study differed from ours in terms of the age of the participants: while the women in the African study were restricted to women 15–49 years, no age restriction was made in our study. Our study included only women who had recently given birth whereas no reference was made to the time of birth in the previous study. Furthermore, our study did not have community level indicators (such as access to roads, schools, banks, hospitals, etc) as these items were not measured while the African study included them. As has been suggested, SES indicators may differ from one context to the other both in their conceptualization and their impact on health.¹

Only inadequate prenatal care had similar association with each of the SES indices, while the other outcomes were differentially associated with each of the indices. It has been suggested that SES measures may be outcome-specific, so that different SES measures may have different associations with outcomes.^1,3 The decreased likelihood of having inadequate care with higher levels of the SES indices in our study is consistent with earlier studies^15,16 In our study, higher occupational status was related to more use of postnatal care, while higher educational status was related to less use of postnatal care. Other studies have suggested that working mothers may be less likely to uptake adequate health care after childbirth, while more educated women may be more likely to undergo postnatal care.^17–19

Only occupational index, but not educational or wealth indices, was positively related to short exclusive breastfeeding. An opposite finding by social class was reported from the Millennium Cohort Study in UK, with women in routine jobs being less likely to initiate breastfeeding and exclusively breastfeed compared with women in professional and managerial occupations.²⁰ On the other hand, Skafida (2009) reported that in Scotland maternal education was a more important predictor of breastfeeding than occupation-related social class.²¹ A study from the USA among low-income women concluded that the women had some difficulty in combining work and breastfeeding.²² The different findings from these studies indicate that the relation between social class and breastfeeding may be context-specific.

The occupational index, and not educational and wealth indices, was also associated with childbirth at the county or higher level health facility. Better maternal or paternal occupational status and higher socio-economic class, measured by wealth and education have been associated with more likelihood of child delivery at a higher level health facility.^23–26 In our study, this was found only for occupational index. One possible explanation for our finding may be that the women at the higher occupational status may more likely live and work in the towns, thus more likely to give birth at the county or higher level health facilities, where they live.

The application of a multilevel modelling to this data allowed us to account for the contextual variations in the outcomes of our study and in the associations between the SES indices and the outcomes. While there was a null variation regarding low birthweight across the study villages and townships, varying levels of contextual variations were observed for short exclusive breastfeeding, caesarean section, childbirth at a higher level hospital, inadequate prenatal care and no postnatal care. Such leftover variations, even after adjusting for the village and township wealth, may probably signify cultural differences between the contexts of study or differences in beliefs and habits regarding to the use of health-care and child-care practices. There are no in-depth studies on women's beliefs of maternity or maternity care in the study areas. Studies elsewhere in China^27,28 and our own experiences during the study suggest that the importance of time before birth has been medically induced; local traditions have centred on birth and the time after it (‘zuo yue zi’—‘doing the month’).²⁹ However, births have rapidly transferred to hospitals, possibly facilitated by the greater value put into the only child (or two children) after the one child policy. ²⁹ ‘Doing the month’ with confinement to the house and special balanced diets has lived on; this tradition does not include care by health professionals. Although we did not take a detailed examination of these cultural and traditional practices in our survey, the robustness of the multilevel modelling is that such latent unobservable variables are accounted for, through the random part of the models as we have reported.

Conclusion

Using available data on women's, their partners’ and household SES indicators, this study derived three separate SES indices (wealth index, occupational index and educational index) to classify women for prenatal care research. Consistent with previous suggestions, the associations between the indices and the perinatal outcomes studied were mostly different between the indices, indicating distinctive effect of different SES measures on health outcomes.

Supplementary Data

Supplementary Data are available at Eurpub online.

Funding

This survey, as part of the CHIMACA project, was financed by the European Commission INCO (international co-operation) Programme and coordinated by the National Institute for Health and Welfare, Helsinki.

Conflicts of interest: None declared

Acknowledgements

The authors thank all local health-care workers who compiled the list of women and contacted them in advance. Thanks are due to staff at the local universities (Anhui Medical University, Chongqing University of Medical Sciences and Xi’an Jiatong University College of Medicine) and the coordinating universities (Fudan University and Peking University). The authors also express their gratitude to all the partners for their professional and technical support during the course of this study. The present article was designed, analyzed and written by Bright Nwaru under the supervision of Elina Hemminki. Wang Hong, Shen Yuan, Huang Kun, Zhuochun Wu did the surveys in the three provinces and commented the manuscript. Zhuochun Wu was responsible for the survey instrument. Reija Klemetti and Elina Hemminki were responsible for the survey design and participated in writing the manuscript.

References