Study design and setting
We designed a retrospective cohort study using available data within the EMTCT program in which exposure to breastfeeding had occurred at some time in the past. The cohort consisted of all HIV infected postpartum mothers enrolled on EMTCT program between June 2014 and June 2018 at Ndejje Health Center IV, a large peri-urban health facility in Makindye Division, Wakiso district, Uganda. We chose this health facility because it serves a greater proportion of HIV infected mothers in the division and it has a high number of patient load. Ndejje Health Center (HC) IV implements the Uganda national and World Health Organization’s Option B Plus(7), a policy we have fully described elsewhere.(8, 9) Under this policy, all HIV infected pregnant mothers are started on Anti-retroviral Therapy (ART for life irrespective of their immunological and clinical status while an HEI receive Nevirapine syrup as prophylaxis from birth until six weeks adjusted according to weight and age bands. Once Nevirapine syrup is stopped at six weeks, Cotrimoxazole prophylaxis is introduced, a dry blood spot is obtained for HIV test using Deoxyribonucleic Acid-Polymerase Chain Reaction (DNA-PCR). At one year, breastfeeding ceases and another dry blood spot is obtained at exactly six weeks after breastfeeding cessation for the second DNA-PCR test. A final HIV test is performed at 18 months using a rapid HIV test.
However, should the HEI test HIV positive at any of the testing time points, ART is started and Cotrimoxazole prophylaxis is continued for life. It is important to note that HEIs receive exclusive breastfeeding for the first six months of life and complementary feeding commences thereafter.
Study population and sample size
The study population consisted of all HIV infected postpartum mothers enrolled on EMTCT program for one year or more, and all of them were still receiving HIV care at the health facility at the time of data abstraction. We excluded HIV infected postpartum mothers with an HEI below one year of age because it would be erroneous to measure breastfeeding cessation in such mother-baby pairs, and those transferred to other health facilities because it was infeasible to obtain data on breastfeeding cessation. We did not calculate a sample size but used census sampling as a retrospective cohort study that consisted of records review was conducted. Accordingly, exposures had occurred at a time in the past and outcomes were determined at the time of data analysis. The cohort consisted of all HIV infected postpartum mothers enrolled on EMTCT program in the period June 2014 to June 2018, followed for at least 12 months to establish exposures and outcome.
Study variables
Our outcome variable was breastfeeding cessation at one year, measured on a binary scale (yes and no). We defined breastfeeding cessation at one year as the proportion of HIV infected postpartum mothers documented in the early infant diagnosis of HIV (EID) register to have stopped breastfeeding at one year. The independent variables included: maternal variables such as age in years but later dichotomized as below 25 and 25 years and beyond, monthly income measured in Ugandan Shillings, marital status measured as single or never married, currently married and divorced, stigma and discrimination measured independently as yes and no, disclosure of HIV sero-status measured as yes and no, knowledge of MTCT during pregnancy, labour, and delivery measured as yes and no, antenatal care attendance at last pregnancy measured as yes and no, number of antenatal care attendances at last pregnancy, ART regimen measured as Tenofovir (TDF) or Zidovudine (AZT) containing regimen, parity measured as nulliparous, secundiparous, and multiparous, nutrition status measured by maternal body mass index (BMI) as weight in kilograms per height in meters squared and later categorized as malnourished (BMI≤18.1 kg/m2), well nourished (BMI 18.2-24.9 kg/m2), overweight (BMI 25.0-30.0 kg/m2), mild obese (BMI 30.1-34.9 kg/m2) and over obese (BMI ≥35.0 kg/m2), and place of delivery measured as health facility and home. The infant variables we studied included age in months, sex measured as male and female, and nutritional status measured using mid-upper arm circumference (MUAC) categorized as malnourished (MUAC <11.5 cm), at risk of malnutrition (MUAC 11.5-12.4 cm), and well-nourished (MUAC ≥12.5 cm). We questioned HIV infected postpartum mothers on modes of mother to child transmission of HIV and the likely preventive measures, the importance of breastfeeding, when to stop breastfeeding in an HEI, and the reasons for stopping to breastfeeding and HEI before one year of age or after one year of age.
Data collection
We used two methods of data collection: quantitative and qualitative. For quantitative data, we reviewed the PMTCT and EID registers and abstracted data using a standardized checklist. We corroborated all entries in the registers with that in the electronic database (Open-MRS) to ensure data integrity. To enrich and triangulate the quantitative results, we conducted qualitative interviews with HIV infected postpartum mothers and healthcare providers. In particular, we held four focus group discussions (FGDs), each consisting of eight to 12 HIV infected postpartum mothers selected randomly from amongst those attending the EMTCT clinic. The FGDs were held within the premises of the health facility in the local language “Luganda” by two research assistants (JGP and MN, both female MPH postgraduate students trained in qualitative research methods).
One research assistant (JGP) moderated all the FGDs while the other (MN) audio-recorded the responses and probed where necessary. Each FGD lasted for about 40-60 minutes on average. For key informant interviews (KII), four healthcare providers namely, two Midwives and two nursing officers engaged in the provision of EMTCT services at the health facility were interviewed to elicit their expert opinions on practices of breastfeeding cessation among HIV infected mothers. The KII lasted for 30-45 minutes, also conducted within the health facility premises in English language, but in a quiet and convenient room. Both FGDs and KIIs were held until saturation was reached.(10)
Data processing and analysis
Qualitative data
We audio-recorded all interviews and transcribed them verbatim. To ensure accuracy in transcription, we correlated the audio-recordings with the transcripts by replaying the audio-recordings. We exported the transcripts to Nvivo, a qualitative data analysis software, for thematic analysis where two reviewers (JGP and SO) read the transcripts thoroughly and coded them independently for common patterns. The independent coding prevented selective perception and interpretive biases in the coding process. The codes were then compared and discrepancies resolved by consensus and a final codebook was developed. The codes were used to enrich and triangulate the quantitative results.
Quantitative data
We single-entered quantitative data in Epi-Data version 3.1 (11) with quality control measures namely skip patterns, alerts, range and legal values, and then exported the data to Stata version 15 for analysis.(12) We analyzed numeric data using descriptive statistics of means and standard deviations, and categorical data using frequencies and percentages.
To determine breastfeeding cessation at one year, the numerator was the number of HIV infected postpartum mothers who stopped breastfeeding at one year expressed as a percentage of the sample size, coded as “1” and “0” to denoted “yes” and “no” respectively.
We tested differences in proportions of breastfeeding cessation at one year with categorical variables using the Chi-squared test for large cell counts (five and more counts) and the Fisher’s exact test for smaller cell counts (less than five counts). To test for differences in means of breastfeeding cessation at one year with numerical variables such as age, we used the student’s t-test. We considered variables with probability values (p values) less than five percent as statistically significant for univariable and multivariable analyses. Our data showed that the outcome variable was frequent (more than 10%). Accordingly, the odds ratio (OR) was not an appropriate measure of association because of overestimation.(13, 14) We hence used risk ratios (RR) for both unadjusted and adjusted analysis computed using a modified Poisson regression analysis with robust error variance to control for mild violations of the assumptions of Poisson regression analysis. We reported each RR with subsequent 95% confidence intervals (CI). We noted that five (2.1%) HEIs had missing data on sex but we did not imputed them at multivariate analysis because the missing observations were fewer than 10%.
Human subject issues
Clarke International University Research Ethics Committee, CIU-REC (reference #CIU-REC/0136), approved this study. We received administrative approval from the Health Department of Wakiso district (reference # CR: MSMC 220/1). Key informants and focus group participants provided a written informed consent after explaining the purpose, benefits and risks of the research. Participation in the study was voluntary and participants were free to withdraw at any stage if they so wished. All participant data were handled with confidentiality and privacy, and individual identifiers were anonymized.