Officials and clinicians in charge of participating clinics provided written documentation of permission. A waiver of informed consent was obtained to abstract routine clinical information from patient obstetric and maternity records. All women participating in validation interviews by study staff provided written informed consent. The study was reviewed and approved by the Botswana Health Research Development Committee and the Institutional Review Board of the Harvard School of Public Health. The trial clinicaltrials.gov registration number is NCT01836003. Registration occurred after trial completion due to misunderstanding as to whether a programmatic intervention should be registered and future trials would be registered. The protocol for this trial and supporting CONSORT checklist are available as supporting information; see S1 CONSORT Checklist and S1 Protocol.

The 20 highest volume antenatal clinics in greater Gaborone were included in this project, which was named Tokafatso (‘improvement’ in Setswana, the local language). At the start of the project in July 2011, each of these clinics provided antenatal care, onsite HIV testing and counseling, and phlebotomy (or referral for phlebotomy) for CD4 enumeration. CD4 testing and most deliveries occurred at the national referral hospital in Gaborone. At study start, clinics received CD4 results via hand-delivered paper reports (9 clinics), electronic medical record (EMR, 3 clinics), or travel to and transcription from an EMR-enabled clinic (8 clinics). Six clinics had onsite ART services (not integrated with antenatal care activities), and 14 clinics referred offsite for ART initiation.

At the start of the project, women with a CD4 count of 250 cells/μL or less (or WHO stage III/IV) were eligible for ART. Women with a CD4 count greater than 250 cells/μL received zidovudine from 28 weeks gestation and single-dose nevirapine in labor [26]. A pilot program was in place in 5 study clinics, which expanded ART eligibility to all pregnant women regardless of CD4 count. In this program, CD4 testing was required to direct ART regimen choice and determine whether treatment would be continued after delivery. During the course of the project, new national guidelines were released and gradually adopted by clinic staff beginning in late 2012. These revised guidelines recommended ART for all HIV-infected women during pregnancy and indefinite ART for women with a CD4 count of 350 cells/μL or less (or WHO stage III/IV) [27,28]. While a CD4 result was not required for ART initiation following this revision, clinicians generally still performed and waited for CD4 count results before starting ART.

We conducted a cluster randomized study utilizing a ‘stepped wedge’ design [29,30]. This design was selected to ensure that all clinics could benefit from the Tokafatso intervention, permit adjustment for time effects given evolving prevention of MTCT programs, and enable effective implementation by a single study team. During the course of ten 4-week blocks between July 2011 and April 2012, all antenatal care clinics (ANCs) received the SMS-based intervention, but the order of implementation was randomized. Clinics and the study team were blinded to randomized order. One week prior to the start of a block, a randomization envelope was opened revealing the names of the 2 clinics to receive the intervention during that block. Implementation was scheduled as close to the first day of the block as practicable. Once implementation occurred, the intervention remained in effect.

A 12-week lead-in period prior to implementation at the first clinics was planned. However, a generalized public service strike leading to restricted health services limited this baseline period to 3 weeks. A follow-up period of 4-weeks was included after implementation in all clinics for capture of additional pregnancies (Figs. 1 and 2).

Two randomly selected clinics (one pair) received the SMS-based intervention every 4 weeks from Period 2 to Period 11. Note: The sample sizes for each period and each clinic pair are provided along the horizontal and vertical axes, respectively. The sample sizes for each clinic pair are recorded for Usual Care, Transition, and Intervention in the colored bars.

ART-naïve, HIV-infected Botswana citizens registering at a participating antenatal clinic prior to 26 weeks gestation were included in the primary analysis. The primary study endpoints were the proportion of HIV-infected, ART-naïve pregnant women a) completing phlebotomy for CD4 enumeration by 26 weeks gestation and b) initiating ART by 30 weeks gestation. For analyses related to CD4 testing, women with a CD4 result within 6 months of antenatal clinic registration were excluded as these women may not have been eligible for retesting.

Study endpoints were assessed through the abstraction of data from routine obstetric, maternity, and laboratory records at Princess Marina Hospital, the principal delivery ward for the study antenatal clinics. Records of deliveries from June 2011 through October 2012 were reviewed. Gestational age was estimated from the date of the last normal menstrual period as recorded in the obstetric record. Gestational age at delivery was imputed from the population median of eligible women for records with missing or unlikely (gestational age at delivery less than 22 or more than 45 weeks gestation) dates.

For purposes of the analysis, a women was defined as being in the Intervention or Usual Care period on the basis of whether the clinic was under usual care the intervention or usual care at the time of her registration for antenatal care. Women first presenting for antenatal care during the initial 4 weeks of the clinic receiving the intervention were assigned to Transition period and not included in the primary analysis.

To assess accuracy and completeness of the abstracted information and to better understand barriers to timely initiation of ART, a convenience sample on consenting HIV-infected women delivering without a recorded CD4 result, or with a CD4 count of 250 cells/μL or less but not on ART, were interviewed.

As designed, the study had estimated 90% power to detect an absolute 15% improvement in CD4 testing by 26 weeks gestation (estimated baseline 36%) and a 35% improvement in rate of ART initiation before 30 weeks gestation (estimated baseline 33%).

The Tokafatso intervention included an improvement package of augmented services around a novel SMS-based platform for CD4 result distribution to antenatal clinics (Table 1). The package included the following:

The operational cost of delivering a CD4 result was estimated for the Tokafatso SMS-based platform and for the three other modalities utilized for result distribution to the greater Gaborone antenatal clinics. Operational costs of the Tokafatso platform were estimated through review of time logs, transportation costs, and expenditures for SMS messages. Through interviews with clinic staff we determined, for each cadre of the clinic staff, the monthly time spent in CD4 result retrieval (reviewing paper or electronic records to find results), transport to locations to retrieve CD4 results, and support of computerized systems. Salaries for each cadre were estimated using representative salaries from the Botswana Harvard AIDS Institute in Botswana. Fuel and vehicle costs were estimated at 1.75 pula per kilometer. Fixed infrastructure costs, including licensing and support fees to the provider of the government electronic medical record system (MEDITECH, Westwood, Massachusetts), were not included. Cost per CD4 result delivered was estimated from the total monthly cost related to CD4 delivery, total number of CD4 results delivered, and the proportion of these results that were antenatal CD4 results.

Comparisons of the study endpoints were performed utilizing general estimating equations with adjustment for temporal effects and clustering at the clinic level following standard methodology for stepped-wedge trials [30]. Small sample size precluded inclusion of individual 4-week blocks, consequently temporal efforts were modeled using 12-week periods. The primary models (Model A) accounted for clustering of participants within clinics and were adjusted for the time that a clinic started the intervention. Findings in the primary models were assessed for sensitivity to imputation strategy used for missing gestational ages, alternate group assignment definitions, significant univariate predictors of endpoints, exclusion of individual clinics, and different methods of modeling temporal effects. In addition, a secondary model (Model B) including all factors identified as significantly associated with the endpoint was also fit to permit adjustment for possible confounding. Analyses of time to testing or ART initiation were conducted using marginal Cox models that incorporated the clinic-level clustering through the use of robust sandwich covariance estimates [31].

Statistical analyses were performed with the use of the SAS statistical package, version 9.2 (SAS Institute, Cary, North Carolina). All tests were two-tailed and P-values of less than 0.05 were considered statistically significant.

A total of 19 of the 20 randomized antenatal clinics received the intervention. One antenatal clinic had begun sending CD4 specimens to a different laboratory for analysis just prior to receipt of the intervention. As the clinic could not receive the SMS and patient tracing components of the intervention, data from patients registering at this clinic after the change in CD4 testing were excluded.

During the study period, 2502 Botswana citizens registering at one of the study antenatal clinics delivered at the principal maternity ward for these clinics. Of these, 726 were HIV-infected (29.0%), 1769 (70.7%) were HIV-negative, and 7 (0.3%) had unknown HIV status. A total of 239 (32.9%) HIV-infected women were already on ART prior to receiving antenatal care. Of the remaining 487 ART-naïve women, 397 presenting for antenatal care prior to 26 weeks gestation without a CD4 count in the prior 6 months were included in the analyses. Of these women 177 (44.6%) registered under Usual Care, 189 (47.6%) under Intervention, and 31 (7.8%) during Transition. The median gestational age at antenatal clinic registration was 18 weeks (Intervention 18 weeks, Usual Care 17 weeks, P = 0.30). The comparison groups were well balanced in baseline characteristics (Table 2), except that women registering under Intervention had less education and were more likely to receive care at a clinic with an on-site ART clinic and early adoption of universal ART (WHO option B).

CD4 draw. Of ART-naïve, CD4-eligible women, 100 (59.2%) women registering at an ANC during Intervention and 79 (50.6%) women registering during Usual Care completed CD4 phlebotomy prior to 26 weeks gestation. There was no significant difference between groups in the primary analysis model (Model A) adjusting for clustering and temporal effects (adjusted odds ratio (aOR) 0.87, 95% confidence interval [CI] 0.47–1.63, P = 0.67). Findings were similar after additional adjustment (Model B) for possible confounders (Table 3). There were significant temporal effects in both models (P<0.001) with overall improvement in the proportion of women undergoing timely CD4 testing during the study period. During the first 3 months of the study period—immediately following a protracted public employee strike—performance under Usual Care clinics was poorer with 26.5% of women having CD4 by 26 weeks compared with 69.3% during later periods. Registration to antenatal care later in pregnancy, unmarried status, new HIV diagnosis in pregnancy, and presentation to a clinic with early implementation of universal ART were each significantly associated with decreased odds of CD4 by 26 weeks gestation.

ART Initiation. Among ART-naïve women, 56 (36.4%) women registering under Intervention and 37 (24.2%) women registering under Usual Care initiated ART prior to 30 weeks gestation. There were no significant differences in ART initiation prior to 30 weeks between Intervention and Usual Care in the primary analysis model (Model A)—aOR 1.06 (95% CI 0.53–2.13, P = 0.87). Additional adjustment for possible confounders (Model B) produced similar results. There was also no significant difference in ART initiation prior to delivery between the study groups, aOR 1.43 (95% CI 0.85–2.41, P = 0.18).

Impact on Specific Elements of the Prevention of MTCT Cascade. Rates of HIV testing were high under both Intervention (99.7%) and Usual Care (99.6%), however rates of CD4 testing and ART initiation remained low.

Timing of CD4 Phlebotomy. HIV-infected women eligible for CD4 enumeration were no more likely to have a CD4 drawn within 1 day of registration if their clinic was under Intervention or operating under Usual Care, 19.6% and 16.4%, respectively (P = 0.33). Time from antenatal clinic registration to CD4 testing was also not improved under Intervention versus Usual Care (adjusted Hazard Ratio [aHR] 0.69, 95%CI 0.48–1.00, P = 0.05). Women registering under the Intervention actually had significantly decreased odds of having a CD4 draw prior to delivery compared with those registering during Usual Care periods (aOR 0.45, 95% CI 0.21–0.94, P = 0.035).

Receipt of CD4 Results. Results from CD4 testing were obtained faster under the intervention. Prior to implementation of the SMS-based platform for CD4 delivery, clinics obtaining paper CD4 results reported a typical turn-around time of 17 days from sample draw to results receipt at the clinic (range 7–35 days). Clinics with EMR access reported turn-around of 7 days (range 3–14 days), and clinics needing to travel to an EMR-enabled clinic reported 14 days (range 7–28 days). Weighting by proportion of results sent by each method, at baseline results took a median of 16 days to reach clinics after CD4 phlebotomy. Under the SMS-based platform, results were received by the antenatal clinic in a median of 6 days (interquartile range [IQR] 3–7 days), significantly faster that under Usual Care (P<0.001, single sample Wilcoxon). The SMS-based platform enabled more than 90% of ART-eligible women to be contacted by the antenatal clinic and counseled to present for ART within 10 days of CD4 phlebotomy.

CD4 Phlebotomy to ART Initiation. CD4 testing prior to 26 weeks gestation was strongly associated with ART initiation prior to 30 weeks gestation—aOR 7.48 (95% CI 3.69–15.1, P<0.001). Registration under Intervention, however, was not significantly associated with decreased time from CD4 draw to ART initiation compared with Usual Care—aHR 0.89 (95%CI 0.60–1.31, p = 0.55). Time from clinic registration to ART initiation was also not significantly different between groups in adjusted analyses (aHR 0.88 95% CI 0.56–1.37, p = 0.56).

Labor expenses represented the majority of the operational costs for the Tokafatso SMS-based platform and the existing CD4 result delivery methods employed in the study clinics. The number of staff hours (time spent by nurses, data clerks, drivers, computer technicians) required was lower utilizing the Tokafatso system compared with existing methods, 0.26 versus 0.39 staff hours per CD4 result, respectively. The Tokafatso system delivered a CD4 result for an estimated $1.98 (lower $1.82, upper $2.14) per result delivered compared with an overall cost of $2.73 (lower $2.56, upper $2.89) with standard delivery methods.

Twenty postpartum, HIV-infected women who had received care at one of the study antenatal clinics were approached and consented to be interviewed. Fifteen of these women were selected for interview because they had no recorded antenatal CD4 measurement. The interviews confirmed that none of these women had received CD4 results, however 10 women (67%) reported that blood had been drawn for CD4 but that they did not receive the results (clinic couldn’t locate results- 7, patient delivered prior to resulting- 2, patient couldn’t travel to clinic to collect- 1). Five (33%) women reported never having blood drawn for CD4 (verified no electronic record of CD4 draw).

Five women were selected for interview because they were eligible to start ART, but did not initiate ART during pregnancy. Of these, 3 (60%) reported that could not start ART as the clinic required them to have an adherence partner participate in counseling prior to initiation. One woman reported that the ART clinic delayed starting for reasons that were not clear to her and one women reported that she was not ready to accept treatment so soon after learning that she was HIV-infected.

Midwives and HIV lay counselors from all 19 antenatal clinics that received the intervention, and the regional prevention of MTCT coordinator, expressed that the intervention assisted them with caring for HIV-infected pregnant women. There was strongest support for SMS CD4 result platform and for the airtime provided to contact ART-eligible women. All clinics expressed that they would like the project continued.