Study design and settings
This community-based analytical cross-sectional study was conducted in a rural and an urban district in the Dodoma region, Central Tanzania, from August to November 2019. In this context, a rural district has a population density of fewer than 45 people per square kilometre; a lack of modern infrastructure, such as roads and railways; and a low density of shops, reflective of low household income. An urban district has a population density of 45 or more people per square kilometre together; modern infrastructure, such as roads and railways; and a high density of shops, reflective of high household income.
The Dodoma region is divided into 7 districts [13], which in this study were divided into rural (Chemba, Bahi, Mpwapwa, and Chamwino Districts) and urban (Dodoma City Council, Kondoa and Kongwa) categories. One representative district from each category was then randomly selected. Dodoma City Council and Chemba District Council represented the urban and rural settings, respectively. Dodoma City Council (urban) is located between 06°10′32″S and 35°44′19″E and has a population of approximately 460,000, and Chemba District Council (rural) is located between 05°14′34″S and 35°53′24″E and has a population of approximately 250,000 [14]. Dodoma City Council covers approximately 2769 square kilometres and has 4 hospitals, 13 health centres, 48 dispensaries, 364 accredited drug dispensing outlets (ADDOs), and 62 pharmacist-operated pharmacies [13]. Chemba District Council covers a total of 7653 square kilometres and has 4 health centres, 35 dispensaries and 87 ADDOs [13].
The average household size is 4.6 persons in Chemba District Council (rural) and 4.2 persons in Dodoma City Council (urban); over 70% of those residing in Chemba District Council are employed in agriculture, while only 57% of those in Dodoma City Council practise agriculture [13]. Moreover, 93% of the rural residents live in privately owned houses compared to only 50% of those in the urban district, and 71% of the urban residents have access to piped water compared to only 15% of those in the rural district [13]. The distribution of electricity is such that 32% of the households in Dodoma City Council are electrified compared to only 3.8% of the households in Chemba District Council [13].
Study population, inclusion and exclusion criteria
The sampling frame of this study was households in both the rural and urban districts. All adults who were present during the interview and who reported that they were permanent occupants of the household were deemed fit to participate.
Sample size calculation
The sample size calculation was based on the unknown prevalence (P) of 50% with a 95% confidence level and was calculated using the survey formula described by Kothari [15]. The degree of precision was 5%, with a design effect of 1. The response rate was 90%. There was no population correction factor since the total population in the sampling frame was well above 5000. The final sample size was 427.
Sampling technique
The sampling technique employed was multistage stratified random sampling involving a rural and an urban district as representative strata. Rural (Chemba, Bahi, Mpwapwa, and Chamwino) and urban (Dodoma City Council, Kondoa and Kongwa) district names were each written on a piece of paper and placed in the corresponding rural or urban basket. Chemba District Council was randomly drawn from the rural strata and Dodoma City Council from the urban strata. In the second stage, probability proportionate to size sampling was used to decide the total number of households to be included in the study between the two districts. In 2016, there were 93,339 households in Dodoma City Council and approximately 47,100 households in Chemba District Council [13]. Proportionately, we needed twice as many households in Dodoma City Council as in Chemba District Council; thus, we visited 269 households in Dodoma City Council and 161 in Chemba District Council.
Households were selected according to the ICF International DHS Toolkit [16]. The full list of all occupied households was obtained from the Council offices, and households were selected using systematic selection: a random starting point was selected, and a die with the first four faces representing the four cardinal directions (1 = east, 2 = west, 3 = north and 4 = south) was tossed to determine the direction. In the selected direction, 10 households were selected, after which the die was tossed again to determine the next direction. From each household, an adult aged 18 years or more was randomly selected for the interview. When more than one able-bodied adult was present in the household, priority was given to the head of the household or any other person acting in a similar capacity at the moment, regardless of sex.
Data collection
Data were collected using an Open Data Kit (ODK) digital questionnaire (Additional file 1). The questionnaire was adopted from a similar study [17] and modified to incorporate the antibiotics commonly used for self-medication and the perceived disease conditions that prompt SMA as well as rural and urban variables. The modified English questionnaire was translated into Swahili (Additional file 2) for easy comprehension among respondents and data collectors, and its suitability was tested in a pilot study before it was adopted for use in this study.
The questionnaire had both closed-ended and open-ended questions that inquired about the sociodemographic information of the respondents (age, sex, education, district of residence and occupation), their history of SMA in the previous year, conditions prompting SMA, sources of antibiotics, awareness of antibiotic resistance, perceived household distance from health care facilities (hospital/health centre/dispensary) and community drug outlets and types of antibiotics used. For the purposes of the study, participants who had used antibiotics to treat self-diagnosed disorders or symptoms or used leftover prescribed antibiotics without consulting a prescriber in the previous 12 months were considered to have practised SMA.
Data collectors were recruited and oriented to the digital data collection questionnaires using supplied Android smartphones. A detailed description of the Swahili translated consent form was given to all potential participants. The consenting participant in each household was recruited and interviewed for up to 15 min; antibiotics that are commonly available in the Dodoma region were shown to the respondents to guide them in remembering the antibiotics they had used or bought in the previous year.
Data analysis
The collected data were downloaded from KoBo Collect software in Excel format, cleaned and then exported to R statistical software version 3.4.4 [15] for analysis. SMA was used as an outcome variable, and univariate and multivariate analyses were performed using various predictor variables. Univariate analysis involved the calculation of measures of central tendency, and multivariate analyses were performed to determine the independent effect of predictor variables on the probability of SMA across the rural and urban participants.
A descriptive analysis of the demographic information and various predictors is presented in tabular format, while the frequency of complaints leading to SMA is presented graphically. Finally, regression models to determine the odds of SMA with various predictors are presented for both crude odds ratios (univariate analyses) and adjusted odds ratios (multivariate analyses).