Data
We analyzed de-identified data from the 2015 Afghanistan Demographic and Health Survey (AfDHS). AfDHS is a nationally-representative cross-sectional household survey, which used a two-stage stratified sampling technique to collect data on nutrition, mortality, fertility, maternal, and reproductive health of ever-married women. For the survey, 25,741 households were initially selected, and 24,395 of them were finally interviewed. The response rate was 98%. Data were collected between June 15, 2015 and February 23, 2016. Data entry was done twice for 100% accuracy and verification. The AfDHS survey sought informed consent from each respondent. In this study, the analysis was limited to the most recent singleton live birth. After excluding observations with missing values, the final analytic sample included 15,581 women aged between 15 and 49 years.
Outcome variables
The analysis included two binary outcomes in this paper – delivery at a health facility and delivery assisted by a skilled birth attendant. Delivery at a health facility was coded 1 if the delivery took place at a health facility (e.g., government or private hospital/health center/health post, maternal and child welfare center, NGO static clinic, and sub-district health center) and 0 if delivery was at home. Likewise, delivery assisted by a skilled birth attendant was coded 1 if a skilled birth attendee assisted the delivery (e.g., doctor, nurse, community skilled birth attendant, family welfare visitor, and midwife) and 0 otherwise.
Key explanatory variables
ANC visit: The key exposure variable in the analysis was ANC visits during pregnancy. Women were asked about the number of their ANC visits during pregnancy. The response was continuous and ranged between 0 and 20. The following categories were used in the analysis: 0, 1, 2, 3, and ≥ 4 ANC visits.
Covariates: The analysis adjusted for women’s demographic factors, socioeconomic factors, and women’s autonomy index. Demographic variables included women’s current age (in years) and place of residence (rural/urban). Socioeconomic variables included education (in years), wealth index, and employment (occupied/not occupied), while other covariates were decision-making autonomy and wife beating not justified.
Wealth index: Wealth index is a measure of a cumulative living standard of a household. This index consists of the household’s possession of various household assets, dwelling characteristics, and farm characteristics. For instance, assets and dwelling characteristics include radio, television, bicycles, mobile phones, computers, refrigerators, housing construction materials, furniture, farm animals, agricultural land, sanitation facilities, water access, etc. Using the principle component analysis, households were assigned scores based on their ownership of type and amount of the above assets and household items. Finally, a continuous asset score was assigned to each household, and then they were categorized into five wealth quintiles [24].
Decision-making autonomy index: Decision-making autonomy index was a composite measure of three household decision-making questions, including a) “Person who usually decides on respondent’s health care”, b) “Person who usually decides on large household purchases”, and c) “Person who usually decides on visits to family or relatives”. For this index, the response “respondent alone” was coded 2, “respondent and husband/partner” was coded 1, and other responses such as “respondent and other person”, “husband/partner alone”, and “someone else” were coded 0. The summed score of three items ranged from 0 to 6, where a higher score referred to women’s higher autonomy in the household decision-making process.
Wife beating not justified index: Wife beating not justified index was a summative index composed of the following five items representing women’s perception about violence perpetrated by their husbands/partners against them: a) “Beating justified if wife goes out without telling husband”, b) “Beating justified if wife neglects the children”, c) “Beating justified if wife argues with husband”, d) “Beating justified if wife refuses to have sex with husband”, and e) “Beating justified if wife burns the food”. The response “no” for all items was coded 0, and the response “yes” was coded 1, while the response “don’t know” was coded missing. Then, the scores of all items were summed, and the total score ranged between 0 and 5, where score 0 indicated beating not justified, and score 1-5 indicated that women perceived the beating as justified.
Statistical analyses
In the beginning, we presented sample characteristics of the respondents. Then, we assessed the associations between ANC visits and two outcomes, using logistic regression analysis adjusting for the potential covariates. In regression analysis, we estimated three models. In the first model, we minimally adjusted for respondents’ age and place of residence. The second model further adjusted for socioeconomic factors, including education (in years), wealth index, and employment status. In the third model, we further adjusted for women’s household decision-making autonomy and beating not justified in addition to the covariates adjusted in the second model. We estimated model fit statistics using AIC (Akaike information criterion) and BIC (Bayesian information criterion). The model with minimum AIC and BIC values was selected for final regression. The regression analysis also incorporated complex survey design factors, including survey strata, clusters, and weights.
Finally, the intersecting effects of ANC visits with SES factors and household decision-making autonomy were examined using a series of two-way and three-way statistical interactions. We used Stata’s margins command (at means) to compute adjusted predicted probabilities from the models with interactions for ease of interpretation. We presented the adjusted predicted probabilities in Figures 1-4. Stata 15.1 (Stata Corp LP, College Station, TX) was used for all data analyses.