Study design and study settings
The study was conducted in Tanzania, which includes mainland and island. This was a Crossectional study. The study used data from the Tanzania Demographic Health Survey (TDHS), Further details of the survey are available elsewhere(13), but in brief this is a national representative survey done after a period of five years and the main objective is to obtain the current and reliable information on demographic and health indicators with regard to family planning, fertility levels and preferences, maternal mortality, infant and child mortality, nutritional status of mothers and children, ANC, delivery care, and childhood immunizations and diseases. Data were obtained from www.dhsprogram.com, DHS measure website whereby the permission to access the data was first obtained before being authorized to download data from DHS measure website. Data from 2004/04, 2010 and 2015/16 surveys were used.
The population was all women of reproductive age (15–49 years) who had given birth to at least one child within the five years prior the survey and had information on ANC visits. For a woman with more than one child birth during the five-year period, the most recent birth was considered for this analysis. A total of 33,734 women age 15–49 years in Tanzania participated in the three TDHS surveys. Within 2004/5 there were 10,329, 10,139 in 2010, and 13,266 in 2015/16. Of these, 15,668 (46.4%) did not give birth within the preceding five years period and they were excluded from the analysis leaving 18,066 (53.6%) who reported a birth in the previous 5 years. Women who reported a birth in the past 5 years was 5658 in 2004/05, 5358 in 2010 and 7050 in 2015/16. Of the remaining a total of 90(0.5%) were also excluded due to missing information on the outcome and finally a total of 17,976(53.3%) were included for the final analysis. Of the17,976 women enrolled in the study; 4541(77.9%), 4201(76.9%) and 5193(70.1%) for 2004/05, 2010 and 2015/16 surveys respectively (Fig. 1).
The dependent variable was adequate ANC visits, which was categorized as four or more ANC visits and coded 1. Less than four ANC visits as inadequate and coded 0. The independent variables included respondent’s age at last birth (15–19 years, 20–24 years, 25–29 years, 30–34 years, 35 + years), education level (no formal education, primary education, secondary and higher education), employment status (unemployed, employed), marital status (married/cohabiting, single, divorced/widowed/separated), residence (urban, rural), wealth index (poorest, poorer, middle, richer, richest), zones (western zone, northern zone, central zone, southern highlands, southern zone, south west highlands zone, lake zone, eastern zone, Zanzibar), first ANC initiated (later than 1st trimester, within 1st trimester), decision maker of respondent’s health care (respondent alone, respondent and partner, partner alone, someone else), parity (1 child, 2–3 children, 4–5 children, 6 or more children), frequency of listening to radio (not at all, Less than once a week, at least once a week), frequency of watching TV (not at all, less than once a week, at least once a week), desire of last pregnancy (wanted then, wanted later, wanted no more), history of terminated pregnancy (never had, ever had) and distance from health facility (big problem, not a big problem).
Data were analyzed using STATA Corporation, College Station, TX, USA version 13. The trend in adequate ANC visits across the three surveys, zones and initiation of first ANC visit was calculated by dividing the number of women with adequate ANC in the particular survey year divided by total number of pregnant women in that particular survey year multiply by 100, then done for all the three surveys.
Poisson regression was used to identify the factors associated with adequate ANC visits. Poisson regression was used as an alternative to classical logistic regression as the outcome was prevalent. Crude and adjusted prevalence ratios were obtained and presented with their respective 95% confidence intervals (CI). All associations were considered statistically significant at a P-value of less than 0.05. Variables included in the final multivariable model were considered if they were priori confounders or had a p-value less than 5% in the crude analysis.
A multivariable Poisson decomposition analysis was conducted to determine factors associated with changes in adequate ANC visits. The decomposition analysis was conducted to understand whether observed changes in adequate ANC visits could be explained by changes in factors or in the population structure. The decomposition analysis was conducted between two time points, firstly to unpack the changes between the 2004/05 to 2010 surveys and secondly to look at the changes between the 2015/16 to 2010 surveys. Contributions were considered statistically significant at a P-value of less than 0.05.