SMS messaging to improve retention and viral suppression in prevention of mother-to-child HIV transmission (PMTCT) programs in Kenya: A 3-arm randomized clinical trial

Background Pregnant and postpartum women living with HIV (WLWH) need support for HIV and maternal child health (MCH) care, which could be provided using short message service (SMS). Methods and findings We compared 2-way (interactive) and 1-way SMS messaging to no SMS in a 3-arm randomized trial in 6 MCH clinics in Kenya. Messages were developed using the Health Belief Model and Social Cognitive Theory; HIV messages were integrated into an existing MCH SMS platform. Intervention participants received visit reminder and prespecified weekly SMS on antiretroviral therapy (ART) adherence and MCH, tailored to their characteristics and timing. Twoway participants could message nurses as needed. Clinic attendance, viral load (VL), and infant HIV results were abstracted from program records. Primary outcomes were viral nonsuppression (VL ?1,000 c/ml), on-time clinic attendance, loss to follow-up from clinical care, and infant HIV-free survival. Among 824 pregnant women randomized between November 2015 and May 2017, median age was 27 years, gestational age was 24.3 weeks, and time since initiation of ART was 1.0 year. During follow-up to 2 years postpartum, 9.8% of 3,150 VL assessments and 19.6% of women were ever nonsuppressed, with no significant difference in 1-way versus control (11.2% versus 9.6%, adjusted risk ratio (aRR) 1.02 [95% confidence interval (CI) 0.67 to 1.54], p = 0.94) or 2-way versus control (8.5% versus 9.6%, aRR 0.80 [95% CI 0.52 to 1.23], p = 0.31). Median ART adherence and incident ART resistance did not significantly differ by arm. Overall, 88.9% (95% CI 76.5 to 95.7) of visits were on time, with no significant differences between arms (88.2% in control versus 88.6% in 1-way and 88.8% in 2-way). Incidence of infant HIV or death was 3.01/100 person-years (py), with no significant difference between arms; risk of infant HIV infection was 0.94%. Time to postpartum contraception was significantly shorter in the 2-way arm than control. Study limitations include limited ability to detect improvement due to high viral suppression and visit attendance and imperfect synchronization of SMS reminders to clinic visits. Conclusions Integrated HIV/MCH messaging did not improve HIV outcomes but was associated with improved initiation of postpartum contraception. In programs where most women are virally suppressed, targeted SMS informed by VL data may improve effectiveness. Rigorous evaluation remains important to optimize mobile health (mHealth) interventions This was an individually randomized nonblinded 3-armed trial, conducted at 6 public MCH clinics in Kenya: 2 in peri-urban Nairobi County and 4 in rural Siaya, Kisumu, and Homa Bay Counties. Ethical approval was obtained from the University of Washington and the Kenyatta National Hospital–University of Nairobi. The trial protocol was previously published and registered at ClinicalTrials.gov as {“type”:”clinical-trial”,”attrs”:{“text”:”NCT02400671″,”term_id”:”NCT02400671″}}NCT02400671 [14]. Participants were recruited from antenatal care (ANC) clinics. Eligible participants were pregnant WLWH, ≥14 years old, had daily access to a mobile phone (own or shared) with a Safaricom SIM card, were willing to receive SMS, planned to reside in the area for 2 years postpartum, planned to receive both MCH and HIV care at the facility, and were not enrolled in other studies. During the first 6 months of enrollment, an additional eligibility criterion of ≤36 weeks gestational age was used; this criterion was removed to expedite recruitment. To maximize generalizability, women who were illiterate but had another person to help them read and write SMS were eligible. Study staff obtained verbal consent to assess eligibility using a tablet-based questionnaire, using Open Data Kit (ODK). Eligible women provided written informed consent to participate. Consent materials were provided in English, Kiswahili, and Dholuo. In accordance with Kenyan regulations, adolescent participants age 14 to 17 years were considered emancipated by pregnancy and provided consent without the need for parental permission. The target sample size for the RCT was 825 participants. The target was reached, but 1 individual was later found to have been enrolled twice. It was determined that this participant was not pregnant at her first enrollment; thus, only data from the second enrollment were retained. Participants were individually randomized by study nurses 1:1:1 to 1 of 3 arms: control (no SMS), 1-way SMS (participants received SMS but were unable to send messages to the study), and 2-way SMS (participants received SMS and could message the study). Randomization was stratified by site with no more than 399 women from any 1 site. A randomization list was generated by a statistician not involved with the study, using variable block sizes (blocks of 5 different sizes ranging from 3 to 15 each were randomly selected) using Stata 12.1 ralloc. ado v.3.5.2. Allocation codes were placed in sequentially numbered, sealed, opaque envelopes by site, which were sequentially distributed by study nurses to participants and opened by participants. Study investigators were blinded to block number, size, and sequence. Participants and study investigators were not masked to group assignments. The study intervention has been previously described [10,14]. Participants randomized to 1-way and 2-way SMS arms received weekly, automated messages from enrollment to 2 years after delivery. Additionally, participants received SMS clinic visit reminders. SMS were sent at the participant’s preferred time of day (8 AM, 2 PM, or 8 PM), day of week (Monday to Friday), and language (English, Kiswahili, or Dholuo). Two-way automated messages ended with a question related to the message topic to promote engagement. Participants in the 2-way arm could respond to these or initiate a message to the study nurse at any time. Participant messages were answered by a study nurse within 1 business day. Formative interviews were conducted with peripartum WLWH, healthcare workers, and partners to inform development of the SMS intervention [8,10,15]. SMS content addressed varied topics, including ART adherence, infant HIV prophylaxis, pregnancy education, birth preparedness, pregnancy and delivery complications, infant health, family planning, and clinic appointment reminders. For every 3 messages related to HIV, 1 message was sent related to non-HIV MCH topics (automated SMS bank available on request) [14]. SMS messages were composed based on the Health Belief Model and Social Cognitive Theory [16,17], designed to provide tailored and actionable education, support, and reminder messages to reinforce health behaviors such as clinic attendance and ART adherence. SMS topics were scheduled according to antenatal/postnatal timing and ART experience, with tailored messaging tracks for adolescents, participants newly initiating ART, and participants who experienced pregnancy loss or infant death [14,18]. Participants who had disclosed their HIV status or had their own phone were given the option to receive SMS containing overt HIV-related language; all other participants received SMS with covert references to HIV [10]. Clinic appointment schedules were abstracted from clinical records and entered into the SMS system by the study team. Visit reminder SMS were sent 3 days before the scheduled appointment date, and congratulatory SMS were sent when visits were attended. SMS were sent 3 and 6 days after scheduled visits were missed. All SMS were delivered free of charge through a reverse-billed short code and managed by study nurses through a custom, semiautomated, open-source web application [11,14]. The system was hosted on a password-protected virtual private server. Data collection occurred at in-person study visits at enrollment in pregnancy, 6 weeks postpartum, and 6, 12, 18, and 24 months postpartum. At each study visit, questionnaire data were collected using ODK, blood samples were collected and archived, and infant dried blood spots (DBSs) were collected. Study staff abstracted viral load (VL) results, appointments, deliveries, clinic visits, and medication refills from clinical records. All clinical care was provided by clinic staff. Since retention in care was a study outcome, study visit attendance was not optimized by the study team, except at the exit visit. If participants did not attend a study visit between 21 and 25.5 months postpartum, study staff traced the participant by phone and home visit. If participants did not wish to attend their exit visit at the clinic, study staff completed the visit at participants’ homes. At visits where no programmatic HIV VL data were available, the study performed VL assays on archived plasma samples. These VL assays were performed at the Kenya Medical Research Institute (KEMRI)/Centers for Disease Control and Prevention (CDC) in Kisumu or Nairobi, Kenya, using the Roche COBAS TaqMan Analyzer or COBAS TaqMan Version 2.0 (CAP/CTMv2.0) platform. Infant HIV DNA polymerase chain reaction (PCR) testing were abstracted from routine PMTCT program at 6 weeks, 6 months, and 12 months. The study performed an additional infant test at study exit using fourth-generation rapid or fourth-generation ELISA tests using DBS. Maternal plasma ART resistance was assessed using an oligonucleotide ligation assay (OLA) on archived enrollment plasma samples and subsequent samples where VL was nonsuppressed. [19] OLA is designed to detect mutations at HIV-1 pol reverse transcriptase codons 103, 181, 184, and 190 that can confer resistance to non-nucleoside reverse transcriptase inhibitors (NNRTIs) and nucleoside/tide reverse transcriptase inhibitors (NRTIs). If samples yielded indeterminate results by OLA, the amplicon from RNA also underwent consensus sequencing. Primary trial outcomes were maternal virologic nonsuppression, on-time visit attendance, loss to follow-up, and infant HIV infection or death. Secondary outcomes were maternal ART adherence and ART resistance. MCH outcomes previously noted to be improved by the Mobile WACh platform (timing of postpartum contraception and EBF duration) were also assessed. Qualitative evaluation of participant experiences will be reported elsewhere. Virologic nonsuppression was defined as plasma VL ≥1,000 copies/mL. A secondary analysis evaluated a cutoff of the assay limit of detection (20 copies/mL for plasma, 839 copies/mL, or 550 copies/mL for DBS depending on assay type). In order to probe potential differences in effect in particular groups of women, subgroup analyses for nonsuppression based on participant age, phone sharing, pregnancy history, employment, HIV status disclosure, education, VL, and ART resistance at baseline were conducted. On-time attendance was defined as attending clinic within 2 weeks of (before or after) a scheduled appointment. Loss to follow-up was defined as not attending any clinic visits for at least 6 months. A combined outcome of infant HIV infection or death was based on clinic records, HIV testing, and participant reports of death. ART adherence was defined as the proportion of days “covered” by ART between pharmacy refills. ART resistance was defined as detection of any resistance mutations by OLA at an abundance of 10% or more. Adverse events, including pregnancy loss, hospitalization, death, and social harms such as HIV status disclosure, intimate partner violence, or loss of housing, were monitored. Serious adverse events (maternal or infant hospitalization or death) were reported to the Kenyatta National Hospital/University of Nairobi ethics review committee. All events were reported to the University of Washington institutional review board annually and to the trial Data Safety Monitoring Board (DSMB) at annual meetings. Our analysis compared outcomes between each intervention arm and the control arm and between the combination of either intervention arm and control. Trial sample size was calculated for 80% power to detect a hazard ratio (HR) of <0.65 for either viral nonsuppression or loss to follow-up, assuming a 25/100 py incidence of each outcome during 2-year follow-up [14]. This sample size was estimated to have power to detect an HR of <0.55 for drug resistance (assuming incidence of 15/100 py) and HR >2.0 for HIV-free survival (assuming incidence of 10/100 py). The initial analysis approach (time to event) only utilized data until first viral nonsuppression and required initial viral suppression at baseline. Because WLWH may have episodic nonadherence and viral nonsuppression and because we had serial VL data, generalized estimating equations (GEE) that incorporate all VL data increased study power to detect effect of the intervention. We adapted our prespecified analysis approach ({“type”:”clinical-trial”,”attrs”:{“text”:”NCT02400671″,”term_id”:”NCT02400671″}}NCT02400671) accordingly. Using repeated measures approaches, the study would have power to detect a change from baseline of 70% to 85% prevalence of viral suppression to 95% suppression in intervention arms with <168 women per arm, with fewer women per arm depending on the number of repeated measures per woman. All analyses were intent to treat, adjusted for baseline differences between arms in participant characteristics (primigravida and employment status). Prevalence of viral nonsuppression during follow-up was defined as the proportion of VL measurements that were nonsuppressed at any time in follow-up and compared using GEE clustered by participant, with log-binomial link, exchangeable correlation structure, and robust standard errors. Data were used after enrollment and ≥4 months since ART initiation, to include only data after VL could have been suppressed by ART. Visits with nonsuppressed VL within 30 days of a previous nonsuppressed visit were excluded. Secondary analyses were conducted comparing the incidence of first virologic nonsuppression by Cox regression, comparing the incidence of virologic suppression as a recurrent event by Andersen–Gill regression with robust error estimation clustered by participant, and comparing the cumulative incidence of nonsuppression among women with any post-enrollment VL data by delivery, 6, 12, and 24 months postpartum by log-binomial regression. Time of event was calculated as the midpoint between the last suppressed and the first unsuppressed VL. The prespecified primary analysis was modified between publication of the trial protocol [14] and conducting the analysis: GEE was used as the primary analysis in place of Cox regression, as explained above. Additional exploratory analyses included comparisons of the 2-way versus 1-way arm and analyses using different VL thresholds (undetectable versus detectable using limit of detection). On-time clinic visit attendance during follow-up to 12 and 24 months postpartum was defined as the proportion of scheduled clinic visits attended on time and compared using GEE clustered by participant, with Poisson link, robust standard errors, and exchangeable correlation structure. The proportion of women lost to follow-up at 12 and 24 months postpartum were compared between arms using log-binomial regression. Time to loss to follow-up was compared between arms using Cox regression for the 0 to 365 day and 0 to 730 day intervals. An exploratory analysis was conducted among intervention participants, comparing on-time attendance of visits that successfully received an SMS visit reminder within the preceding 2 weeks versus visits that did not due to system errors. Time to infant HIV infection or death after live birth in the first 850 days postpartum was compared using Cox regression. We classified ART adherence as high (≥95%) versus low (<95%) based on 2 approaches: pharmacy refills and self-report. Pharmacy refill data were abstracted from records of ART doses administered during HIV care visits. Longitudinal ART adherence was calculated as the proportion of days between visits that were covered by doses dispensed. Self-reported data were based on the number of doses missed in the last 30 days at study visits. For each method of ascertainment, GEE log-binomial regression with exchangeable correlation structure and robust standard error was used to compare the prevalence of high (≥95%) versus low (<95%) adherence between randomization arms. Time to development of ART resistance was compared between randomization arms using Cox regression on data from enrollment until 850 days postpartum. Plasma samples with VL <1,000 copies/mL were not tested for resistance and were assumed to not have resistance mutations. Since Mobile WACh-X messaging was derived from Mobile WACh messaging, which focused on MCH topics and was shown to prolong EBF and increase contraception uptake in HIV–negative women [12,13], we explored whether efficacy of MCH messaging persisted with addition of HIV messages. Time to contraception initiation by 6, 12, and 24 months postpartum and time to introduction of complementary foods by 6 months postpartum were compared between arms by Cox regression.