Utilisation of health services fails to meet the needs of pregnancy-related illnesses in rural southern Ethiopia: A prospective cohort study

Although maternal survival has improved in the last decades, evidence on illnesses and the use of health services during pregnancy remains scarce. Therefore, we aimed to assess the incidence and risk factors for illnesses among pregnant women and measure the use of health services. A prospective cohort study was conducted in three kebeles in rural southern Ethiopia among 794 pregnant women from May 2017 to July 2018. Each woman was followed every two weeks at home. Poisson and survival regression models were used for analysis. The incidence rate of episodes of illnesses was 93 per 100 pregnant-woman-weeks (95%CI: 90.6, 94.2), with an average of eight episodes of illnesses per woman. Anaemia accounted for 22% (177 of 794 women), and hypertension 3% (21 women of 794 women). However, utilization of health services for any illness episodes was only 8% (95% CI: 7.6%, 8.9%). The main reasons for not using health services were that the women thought the illness would heal by itself, women thought the illness was not serious, women could not afford to visit the health institutions, or women lacked confidence in the health institutions. The risk factors for illnesses are having many previous pregnancies in life time (ARR = 1.42; 95%CI = 1.02, 1.96), having history of stillbirth (ARR = 1.30; 95%CI = 1.03, 1.64), having history of abortion (AHR = 1.06; 95%CI = 1.02, 1.11), and walking more than 60 minutes to access the nearest hospital (AHR = 1.08; 95%CI = 1.03, 1.14). The risk factors for low use of health services are also having history of abortion (AHR = 2.50; 95%CI = 1.00, 6.01) and walking more than 60 minutes to access the nearest hospital (AHR = 1.91; 95%CI = 1.00, 3.63). Rural Ethiopian pregnant women experience a high burden of illness during pregnancy. Unfortunately, very few of these women utilize health services. A prospective cohort study was carried out among 794 pregnant women attending antenatal care (ANC). In 2016, the proportion of pregnant women attending at least one antenatal care in Ethiopia was 62% [19]. The healthcare system in Ethiopia is organized based on the type of care provided. For example, primary healthcare comprises one primary care unit (health centre) with five health posts, secondary healthcare includes zonal and regional hospitals, and tertiary care includes a central and teaching hospital. Primary healthcare utilizes the health extension package as an approach that focuses on communicable diseases, maternal health, common nutritional disorders, hygiene, and environmental health. Maternal and child healthcare, immunization against childhood illnesses, and family planning and reproductive health, infectious diseases, including tuberculosis, malaria and control of sexually transmitted infections and HIV/AIDS, are also critical areas to address. Pregnant women primarily use health extension workers at health posts. If the case is serious, the health post may refer them to a health centre. According to the national antenatal care guidelines in Ethiopia, a focused ANC visit is advised to take place at least four times during a pregnancy. The main components of antenatal care include protection at birth from tetanus, blood pressure measurement, nutritional counselling, iron-folate supplementation, and information about the danger signs of pregnancy complications [19]. In 2013, more than 80% of the population lived in rural Ethiopia, 26% of the rural residents lived on less than $1 per day, and 77% of rural women needed to travel more than 20 km to get to a hospital [20]. The Wonago district (zone) is located 420 km far away from the capital, Addis Ababa. It has 15 rural and four urban kebeles. A kebele is part of a district (wereda), which is the lowest administrative unit in Ethiopia and comprises approximately 1000–1500 households (an average of 5000–7500 people) [21]. In 2017, the total population of the district was estimated to be 145,000 people [22]. The major ethnic group was the Gedeo people, and the population density was 980 persons per km2. Agriculture is the dominant means of livelihood. The district has six health centres, 20 health posts, and two private clinics. Three kebeles (i.e., Hase-Haro, Mekonisa, and Tumata-Chiricha) were randomly selected by the lottery method from Wonago district, which is located in the Gedeo zone in southern Ethiopia. The recruitment of pregnant women started in May 2017 and follow-up ended in July 2018. The three kebeles were similar in socio-demographic and economic features to most rural areas in Ethiopia. Mekonisa kebele has one health centre and two health posts, Tumata-Chiricha kebele has one health post, and Hase-Haro kebele has one health centre and one health post. In 2017, the estimated total population of the three kebeles was 29,780 people [22], and the crude birth rate of rural people in Ethiopia was 33.2 per 1000 population [19]. The proportion of observed (number of pregnant women included in our study) to expected (total number of pregnant women) antenatal care visits was 80.3% from the three kebeles, and varied from 70.6% to 85.2%. There is no significant difference between the expected and actual births from the three kebeles, x2 = 4.8618(2), p-value = 0.09 (Table 1). *The number of expected pregnancy calculated based on EDHS 2016 [19] Pregnant women attending ANC at health posts formed the study population. All women in the reproductive age group in the selected kebeles were the source population. The pregnant women were recruited based on ANC visits mostly in the second trimester, which is estimated to be within 24–28 weeks of gestation according to EDHS 2016 [19]. All women were followed at regular intervals (every two weeks) at home based on a scheduled visit. The inclusion criteria were a pregnant woman who had attended two or more antenatal care visits to a health post. Exclusion criteria were a woman not living in the study area, or not found at home at the scheduled visit. Unfortunately, we did not include women with less than two antenatal visits, and this may have caused a selection bias, as we have discussed. The criteria for identifying illnesses among pregnat women were decided prior to initiating the study. The concepts of illness have been previously used to indicate personal ailments (subjective undesirable state of health) [23]. A pregnant woman with an illness was identified using the illness category and/or by the recording of an associated disability (S1 Table). Illness identification criteria were based on general symptoms and screening of anaemia and hypertension. The symptoms and subsequent use of health services were recorded. Our assumption was that pregnant women would seek healthcare primarily due to medical problems. The primary outcome variable was the incidence of illnesses among pregnant women and measured among all participants. Subsequent use of health services was measured among those who had an illness during pregnancy. Pregnant women were followed over time to assess the occurrence of illnesses among pregnant women and subsequent use of health services. Women’s basic characteristics were defined as women’s age, women’s age at first marriage and at first birth (increase vs. decrease), marital status (ever married vs. not married), educational status (had formal education vs. had no formal education), occupation (others (daily labourer, farming, etc.) vs. domestic service), wealth index (rich vs. poor), total monthly household expenditure ($30+ vs. <$30), gravidity (multigravida vs. primigravida), parity (multipara vs. nullipara) prior viable pregnancy, birth interval (2+ years vs. <2 years), and history of abortion (yes vs. no) and stillbirth (yes vs. no). Community or kebele level exposure status of the pregnant women was defined as the type of road to the nearest health facility (asphalt vs. others), and walking distance to the nearest health post (30+ minutes vs. <30 minutes). Questionnaires were used to assess illnesses amnog pregnat women and the use of health services. Blood pressure was measured by the Riester ri-champion®N digital apparatus (www.riester.de), and haemoglobin (Hgb) was measured by the HemoCue analyzer®Hb 301 System (www.hemocue.com). Hgb values for pregnant women were categorized into no anaemia (> = 11 g/dl), mild anaemia (10–10.9 g/dl), moderate anaemia (7–9.9 g/dl), severe anaemia (4–6.9 g/dl), and very severe anaemia (< = 3.9 g/dl). As the concentration of Hgb declines during the first trimester, reaches its lowest point in the second trimester, and begins to rise again in the third trimester, due to physiological changes, Hgb values for pregnant women were determined at around the third trimester, or 27 gestational weeks or later [24]. Hypertension during pregnancy was classified as either a systolic blood pressure greater than 140 mmHg, or diastolic blood pressure greater than 90 mmHg, or both [25]. At the time of each visit, blood pressure (BP) readings were taken in at least one-minute intervals [26] between two consecutive readings, and their mean was recorded. The sample size was determined by Openepi software Version 3.03 (www.openepi.com). To obtain the maximum sample size, we used different socio-demographic factors as exposure variables and the incidence of illnesses among pregnant women as outcome variables. The following assumptions were made to assess the sample size: 15.5% of the incidence of illnesses among pregnant women, and 1.65 relative risk [7] among poor compared with rich women, with 95% confidence level, 80% power, and 1:1 ratio of unexposed to exposed. The sample size was estimated to be 898 after adding 10% for non-response (Fig 1). Continuous variables were assessed for symmetry, and parametric tests were used for normally distributed variables. For example, women’s age was categorised into the following groups (15–19, 20–24, 25–29, 30–34, and 35+ years), women’s age at first marriage (10–14, 15–19, 20–24, and 25–29), and women’s age at first birth (15–19, 20–24, and 25–29). Walking distance to the nearest health post was classified based on the mean of the sample in minutes (<30 and 30+), and household total daily expenditure categorized based on the $1 a day poverty line for developing countries [27] (<$30 and $30+ a month). At baseline and during the follow-up period, we collected information on variables that are important for the study. We then assessed illness among pregnant women. The illnesses during pregnancy may occur once or multiple times. The follow-up time ranged from four weeks to 14 weeks. The data were collected at home every two weeks, and only one woman per household was included in the study. At each visit, data on the use of health services and reasons why they did not seek healthcare were collected from the women. Data were collected using a validated interviewer-administered questionnaire, which was adapted from the “WHO maternal morbidity measurement tool pilot: study protocol” [2] by piloting the questionnaire. The pregnancy questionnaires were prepared in English, translated into the local languages, i.e., Gedeo language (S1 Questionnaire.rar) and Amharic (S2 Questionnaire.rar), and then back-translated into English (S3 Questionnaire.rar). A pre-test was conducted in a neighbouring kebele. The data collectors read the symptoms, and women indicated whether they had any of the symptoms before the current visit (in the past two weeks). Pregnant women were also asked to report any other health-related problems that they had experienced. The six data collectors were women, residents of the selected kebeles, and had completed grade 10. Stillbirth: a baby born with no signs of life at or after 28 weeks' gestation [28]. Abortion: is the natural death of an embryo or fetus before it is able to survive independently before 28 weeks of gestation. Gravidity: all number of pregnancies in a lifetime (includes complete or incomplete). Parity: number of children previously borne by a woman (excludes abortions, but it includes stillbirths). Utilisation of health services: defined as the number of healthcare services used by persons for the purpose of curing illnesses. Recurrent events: an event (i.e., illness and use of health services) experienced repeatedly by pregnant women. These events could all be of the same type or different types. Repeated measures: a research design that involves multiple measures of the same variable taken on the same subjects either under different conditions or over two or more time periods. Therefore, in this study, we measured repeatedly the type of illnesses and use of health services during pregnancy. Multiple responses: refers to the situation in which people are allowed to tick or respond with more than one answer option for a question. The data were entered in EpiData version 3.1 software (EpiData Association Odense, Denmark). Principal components analysis (PCA) was utilized to construct a wealth index of households based on 35 household assets and facilities. For this study we used two categories of quintiles (i.e. from 1st to 3rd quintiles categorized as poor, and the 4th and 5th quintiles categorized as rich). Descriptive statistical analysis was used to determine the distribution of the incidence of illnesses and the use of health services. In this recurrent events analysis, the pregnant woman was at risk for the same or different events throughout the follow-up period, regardless of whether or not an event has occurred. Different pregnant women, of course, could have different numbers of events; some women had no illness or did not use health services, whereas others had many or did use health services. However, these different numbers of events that were observed across different pregnant women tended to follow a certain pattern that can be described using a Poisson distribution [29]. The recurrent event was described by estimating the mean cumulative function, which was the average number of cumulative events experienced by a pregnant woman in the study since the start of follow-up. The outcome (an event) was analysed as a count variable, and illnesses as exposure for use of health services were also analysed. In this paper, two classes of statistical models were used to analyze recurrent event data: Poisson regression model and longitudinal techniques, which are an extension of the Cox-proportional hazards regression survival model. Cumulative number of events (counts) and event rates (total number of events divided by total follow-up period) by the end of the study were assessed using these two models. Poisson regression is a technique that models the number of events (i.e., illnesses and use of health services) and the length of follow-up time. In the Poisson analysis method, all events were assumed to be independent. The Poisson model does not utilize all available data, but rather uses only one summary observation for each pregnant woman. The Poisson analysis method was done for the count data, including an offset for the time in the study, and event rates. Poisson logistic regression analysis was performed to analyse the difference in the proportion of pregnant women with illnesses at the end of the study. The data structure for fitting the Poisson regression model was composed of a column labelled “number of events” which was used as an outcome, and a column labelled “the length of follow-up time” which was used as an offset term. To account for the different lengths of follow-up between pregnant women, we included an offset term denoting the length of follow-up. In the dataset, each pregnant woman contributed one record. In longitudinal techniques, the model does not use one observation for each pregnant woman, but instead all observations for each pregnant woman are used in the analysis. In such a long data structure, there was more than one record present for each pregnant woman. The time at risk is defined as the total time approach, in which the starting point for each period is the beginning of the study. The Prentice-Williams-Peterson (PWP) total time model is an extension of the Cox-proportional hazards regression survival model, which was utilized to analyze the repeated occurrence of events (recurrent events) over time, as there is a dependency of observations within a pregnant woman [30]. The PWP total time model considers each sequential event (first, second, etc.) separately because the time scale used in this model is the time from study entry. The model was stratified by event sequence, so that the baseline hazard function can differ between the sequential events. The PWP total time model allows any covariate to have different associations with different sequential events. The data structure for fitting the PWP total time model was composed of a column labelled “sequence number” which represents the order of time intervals for each pregnant woman which is unique to each sequential event and can be used to define the strata, the columns labelled “start-time” and “end-time” represents the start and end time of each interval, respectively, and the column “event indicator” represents whether an event occurs at the end of the time interval. A multivariable regression model was carried out to identify independent predictors of the incidence of illnesses. The interaction effect of 20 exposure variables was examined, and no significant interaction effects were observed. The Hosmer and Lemeshow recommendations were used in the selection of the factors, which were P-values ≤0.2 in univariate analysis for multivariate analysis [31]. P-values ≤0.05 were used as cut-off points to determine statistical significance. Poisson regression and the PWP total time model survival model was fitted in STATA software version 15 (Stata Corp., LLC. College Station, Texas U.S.A.). This study was approved by the Institutional Ethical Review Board at Hawassa University, College of Medicine and Health Sciences (IRB/100/08), and by the Regional Committees for Medical and Health Research Ethics (REC) of western Norway (2016/1626/REK vest). Written permission letters remain from the Gedeo Zone Health Department and the Wonago Wereda (district) health office. Written informed consent was obtained from each mother after she had received an explanation of the purpose of the study. The privacy, anonymity, and confidentiality of study participants were maintained. If a woman was found to have an illness during pregnancy, the data collectors linked the patient with health extension workers in the kebele.