Rate and associated factors of non-retention of mother-baby pairs in HIV care in the elimination of mother-to-child transmission programme, Gulu-Uganda: a cohort study

This was a retrospective single cohort of HIV-positive mothers and their HIV-exposed babies born between January 2010 and December 2012 at either Lacor Hospital (LH) or Gulu Regional Referral Hospital (GRRH), both located in Gulu district, Northern Uganda. We also collected additional data on possible reasons for non-retention, in a cross-sectional manner. We identified some of the mothers at the hospital, while others were traced using telephone contacts or sketch maps in their hospital care cards, for face-to-face interviews. The study was conducted between June and August 2014. The entire cohort had at least 18 months of being at risk of experiencing non-retention by June 2014.

Lacor hospital and GRRH are the two largest hospitals in Acholi sub-region, serving patients from the entire sub-region and beyond. Gulu district is located at, 02 45 N, 32 00E, approximately 340 kilometres North of Uganda’s capital city, Kampala. The population of Gulu district is 443,733 [16].

Only mothers aged 15 years and above and who were permanent residents of the study area were eligible to take part in the study. Mothers aged 15–17 years are emancipated minors who were included in their own rights, as provided for in the Research Ethics Committee (REC) guidelines [17]. Mothers who had delivered from other hospitals but enrolled in either GRRH or LH and those who had transferred care to other facilities were excluded from the study so as to minimise data incompleteness but to also maintain study eligibility.

This study had two outcome variables, i) rate and ii) time to non-retention. We used the online StatsToDo software [18] to determine the sample size for the time to-non-retention, with 5% type-I error and power of 80%, giving a sample size of 393 mother-baby pairs. For the second outcome variable, rate non-retained within 18 months, a Kish Leslie [19] formula was used assuming a 53.4% rate of non-retention in Northern Uganda from a previous study [10], 5% margin of error. A 10% non-response rate was factored into the calculations. The largest sample size was considered for both outcomes because the study population was the same [20]. We sampled the study participants from the hospital ART registers. A total of 476 mother-baby pairs were eligible for the study (Fig. 1). Half of the study participants were recruited from each of the hospitals, and simple random sampling was conducted to enrol the study participants. The patient hospital number from the register was written on a piece of paper. The pieces of paper were put in a basin. We then randomly picked one at a time until the required sample size was attained. Probability sampling was used to avoid selection bias.

The variables for this study were derived from Andersen’s model of health care utilisation [21]. The outcome variables of this study were non-retention at 18 months, and time to non-retention of mother-baby pair in HIV care. Non-retention was defined as a mother and/or baby not returning at the facility within a period of 30 days after missing a scheduled visit. The rate of non-retention was calculated as the number of mother-baby pairs not retained in HIV care over the 18 month period divided by the total number of mother-baby pairs enrolled in the study. This was categorised as (“0” = retained and “1” = non-retained).

Time to non-retention was calculated as the duration from mother-baby pair enrolment into HIV care and when the mother and/or baby missed a scheduled visit and did not return for care within a period of 30 days. The unit of measurement was months. The period considered was from delivery to 18 months, thus entry into this study was from the date of birth. The event of interest was non-retention. Once the event of interest occurred, no re-entry in the study was allowed for this analysis.

Infant data collected included sex (male/female), feeding method (exclusive breastfeeding/mixed feeding); EID at 4–6 weeks (done/not done), and EID result (positive/negative). Maternal data collected included; age, categorised as 15–19, 20–29, ≥30 to reflect the health seeking behaviours among pregnant women in those categories, which also affect healthcare utilisation; marital status (categorised as never married, married, previously married), education (categorised on the basis of education levels in Uganda -none, primary, secondary, tertiary), religion (categorised on the basis of the main religious denominations in Uganda - Protestant, Catholic, Muslim, Pentecostal, plus the category Others, which incorporated religion or beliefs such as atheism, Hinduism, and Seventh Day Adventism, which are less widely practiced in Uganda). Occupation (categorised as housewife, farmer, wage/salaried workers, manual labour, others). This categorisation was on the basis of the widely recognised income earning sources of women in Uganda. Transport was categorised on the basis of the means of transport commonly used in Uganda (walk/bicycle/motorcycle/vehicle). Other maternal demographic data collected were residence (urban/rural), distance to the health facility (recoded as 0-5 km, 6-10 km, >10 km). Wealth index as determined by monthly income was categorised as [lowest = <UGX. 50,000 (~$15), second lowest = UGX. 50,000 - ≤100,000 ($15.1 - $30), middle = UGX. 100,000 - < 300,000 ($30.1 - $90), second highest = UGX. 300,000 - <500,000 ($90.1 - $151.5), highest = UGX. ≥ 500,000 (≥$151.5)]. The cut-off points for categorisation of wealth index were refined after pre-testing the study tools. Other data collected were the number of times delivered while HIV positive (1, 2+), partner disclosure (yes/no), spouse testing for HIV (yes/no), and the mother’s opinion about keeping all the hospital appointments (not important/important).

The quality of health care variables included privacy during counselling session (categorised as (yes/no), and waiting time, which was categorised as not considered long by the respondent versus considered long.

We abstracted data on HIV diagnosis, eMTCT enrolment, infant feeding, mother’s age, residence, marital status, distance to health facility, delivery, and follow-up outcomes from ART/EID registers, mother/baby appointment books kept at the health facilities, and exposed infant clinical charts. Records from the ART and EID registers and appointment books were used to determine the rate of non-retention of mother-baby pairs in care during the first 18 months of infant life. A structured questionnaire was developed to collect additional data on possible reasons for non-retention of mother-baby pair sampled from the clinic records. These were data not captured in the medical records but are considered important in understanding non-retention. The additional data included; the highest level of education attained, religion, occupation, means of transport to the hospital, wealth index, number of previous deliveries while HIV positive, whether the spouse had tested for HIV or not, and the mother’s opinion on the need to keep all appointments together with the baby, and the quality of care as determined by privacy and waiting time at the health facility. Some of the mothers were interviewed at the hospitals during their routine visits for care, while others were traced and interviewed from a place of their choice. The questionnaire and data abstraction form were pretested in a health facility setting. They were then refined to improve their validity and reliability. Trained research assistants with experience of working in HIV/AIDS units in hospitals collected data.

Data were checked for completeness and accuracy at the end of each day of data collection. Double entry of data in an Excel spreadsheet was performed to detect inconsistent and missing entries. Data were then cleaned to eliminate errors in data entry and backed up regularly. We then exported data to Stata version 12 for statistical analysis.

A total of 410 mother-baby pairs took part in this study, providing a 95% response rate. Just over a half (52.7%) were from GRRH. The mothers were aged 15–44 years with a mean (SD) age of 28.6 (5.9) years. The majority of eligible participants were aged 20–29 years (53%), married (84.6%), and had attained some level of education (primary 38.1% or secondary 33.9%). Exclusive breastfeeding during the first six months was common (54.2%). EID was absent in 11 infants (2.7%), whilst 23 infants (5.8%) were detected as HIV positive by six weeks. Up to 83.2% of mothers disclosed their HIV-status (Table 1).

Nearly a third (30.5%) of the mother-baby pairs were not retained in care by end of month 18; higher in GRRH (34.7%) than in LH (25.8%), p = 0.049. The probability of non-retention was 0.12 at six months, 0.19 at twelve months and 0.3 at 18 months. The 18 month follow-up period was not long enough for the overall median time to non-retention to be ascertained (Fig. 2). It was calculated to be two months for babies with no EID, 15 months for mothers not disclosing their serostatus, while that for mothers who reported lack of privacy during counselling session it was 16 months (Fig. 3). The median time for those with EID, who disclosed their HIV serostatus, and those who reported that there was privacy during counselling could not be ascertained during the 18 month follow-up time.

Early infant diagnosis, education level attained, religion, occupation, transport, disclosure of HIV serostatus, the number of times delivered while HIV positive, spouse testing for HIV, quality of counselling, and waiting time were all significant factors for time to non-retention at the bivariable level (Table 2). Factors independently associated with time to non-retention were; absence of EID, adj. HR = 5.8; CI (2.55, 13.24), disclosing HIV status to a sexual partner adj. HR 1.86; CI (1.22, 2.85), and quality of counselling session adj. HR 1.86; CI (1.26, 2.85). Knowledge of the importance of adherence to all appointments together with the baby was associated with lower risk of non-retention adj. HR 0.43; CI (0.20, 0.92) (Table 3).

Data were primarily abstracted from routine care records, and so missed some important variables like education and transport. However, identification and follow-up of the participants were conducted to generate the missing variables. However, there could have been changes in some cases from the time the mothers were first enrolled into care.