Study population and setting
We included in the study pregnant women enrolled in Malawi’s national PMTCT programme between 10 August 2018 and 30 November 2019 at four health facilities in the Southern region of Malawi. These health facilities are part of the DREAM (Disease Relied through Excellent and Advanced Means) programme sub recipient of Global Fund, a public health program run by the Community of Sant’Egidio in 11 African Countries. This program focuses on HIV/AIDS, TB and NCDs [19–21]. The facilities are located in Blantyre (urban), Machinjiri (peri-urban), Chileka (peri-urban) and Balaka (rural).
The pregnant women were eligible if they were aged 18 years or older, lived with a male partner, enrolled for the first time in a DREAM facility and willing to sign a written informed consent. We have decided to include only women living with a male partner, assuming that a cohabitant partner could impact more on their behaviours.
This study is part of larger research program that aims to investigate the impact of male participation on women’s ART adherence in PMTCT in Malawi.
We have administered a Knowledge, Attitude and Practices (KAP) survey [22] to women and their male partners attending the health facilities. KAP surveys assess knowledge gaps, cultural beliefs, and behavior of people with HIV/AIDS. It identifies misconceptions or misunderstandings that may represent barriers to access the health care, uptake of interventions and improving of adherence to ART. We have designed the survey questionnaire following the guidelines for conducting a KAP survey [22,23]. Previous literature has been taking into account in the questionnaire designing phase and to analyse the responses [24–26]. The final questionnaire consisted of 27 questions or statements (see Additional File 1,2,3): 15 on knowledge of HIV transmission (1-8), prevention and anti-retroviral therapy (9-15), 7 on attitudes towards HIV/AIDS, and 5 on sexual, alcohol or drug use behaviour. The answer to each question had three possible answers (yes, no, do not know).
To better adapt the questions to the Malawian context, we discussed the readability and clarity of the questionnaire with two healthcare workers and a medical doctor. We then pilot-tested it in a small group of randomly selected women (n=22). The KAP survey was translated in the local language and administered to a woman and her male partner by a community health worker of DREAM facility.
If the couple went together to the first visit, the survey was administered separately to the women and their partners. If the woman was alone, the healthcare worker would give her the invitation card (recommended by Minister of Health in Malawi) asking to go with the partner to the next visit. During the second visit, the healthcare worker would interview the woman (KAP survey). If the male partner did not attend any clinical visit during the six weeks after the first one, the questionnaire was to be completed only by the woman. If the male partner did attend, he has been interviewed, and this survey was not included in the paper due to small sample size. It is recommended that women have at least three PMTC visits after the first one. All the surveys administered to the women were included in the analysis. We adopted a convenience sampling method since this is a pilot study.
Outcomes
Male attendance to health facilities at least once in the six weeks follow-up period was the primary outcome. Among the not-accompanied women, we collected secondary male partner specific outcomes such as the proportion of women who i) disclosed their status to their male partner, ii) reported to their male partners that they had a clinical appointment, iii) asked the male partner’s permission before going to the facilities, iv) delivered the invitation slip and v) the proportion of women were given money by their partner for transport.
In addition to the KAP survey we also used the data from the electronic medical record system on the women enrolled in the study. These data included age, educational level, type of job, number of people in the family, socio-economic condition (e.g. availability of electricity), owning a means of transport, means of transport to get to facility and travel time from household to facility. A woman was considered lost to follow-up (LTFU) when she missed to collect the doses of antiretroviral drugs for more than two months.
Statistical analysis
The characteristics of the women were described using Fisher test or chi- square tests for categorical variables, and Wilcoxon rank sum tests for continuous variables.
To determinate the level of knowledge, attitude and practice of respondents, we asked women to answer ‘‘yes’’, ‘‘no’’ or ‘‘do not know” to every questions. Following literature [24,25,27] a score of ‘1’ was assigned for each correct answer and ‘0’ for each wrong answer. As it was previously done by other researchers [24–29], overall knowledge was determined by aggregating correct answers from all questions according to other researchers. The maximum attainable score was 15 and the minimum score was 0. For the seven attitude related questions, each positive response was assigned a score of ‘1’, and each negative response a score of ‘0’. In the same manner, for the five practice related questions, each safe response was assigned a score of ‘1’, and each risky response as a score of ‘0’. We defined the overall level of attitude and practice by aggregating positive attitude answers and safe practise answers. The scoring range of the attitude section was from a 0 to 7 and of the practice section was from 0 to 5. Then, we determined the percentage. Based on previous research [25,26], levels of knowledge, attitude and practice were categorized in two segments depending on their median score, as data were not normally distributed. Low and high level of knowledge, positive and negative attitude and safe and risky practice presented the two categories. To evaluate how our KAP scoring method impacted the analysis of results, we also run the analysis considering an alternative scoring system: score of ‘1’ for each correct, ‘0’ for each uncertain, ‘-1’ for each wrong answers.
We compared the single answer score using Mann-Whitney tests and the overall level of knowledge, attitude and practice using Mann-Whitney tests with 95% confidence intervals (CI) in order to determine if there was any difference in knowledge, attitude and practice among women accompanied and not accompanied by their male partner.
We conducted univariable logistic regressions to study the association between male attendance (yes, no) and explanatory variables. In the univariable logistic regression the explanatory variables included age, education (no education/primary school, secondary school/pre-University), employment (employed, unemployed), owning a means of transport such as cars, motor cycles and bikes (yes, no), means of transport to get to facility (minibus, motorbike, car or bike, by foot, other), time to reach the facility (0-89 minutes, >89 minutes), economic condition (availability of electricity in household, yes, no), and knowledge (low, high), attitude (positive, negative), practice (safe, risky) toward HIV. We included in the multivariable model our variables of interest (knowledge, attitude and practice) and all the variables that resulted to be significantly associated with male attendance in the univariable logistic regression. The “unemployed” category includes houseworkers (stay-at-home) and unemployed women, the “employed” category includes employee (formal job with contract) and temporary job (informal job with no contract).
We imputed missing values of explanatory variables using multiple imputation with chained equations (MICE). To improve the imputation we added the following variables [30]: mother alive (yes/no), owns a phone (yes/no), gestational age (1-3, 4-6, >7), piped water available in the dwelling (yes/no), owned means of transport ( bike, motorbike, car, bus, others). Furthermore, we considered the outcome in the imputation. We ran the model on 20 imputed datasets for each analysis and combined the estimates with Rubin’s rule [31].
All analysis were performed using STATA 13.