Background
Despite its virtual elimination in the developed world, vertical transmission of HIV remains a significant challenge in resource-limited settings today. According to WHO estimates, 50% of HIV-positive pregnant women worldwide did not receive prevention of mother-to-child transmission (PMTCT) services in 2010, which resulted in 1000 new pediatric infections daily [
1]. The risk of MTCT was shown to be 15-30% among non-breastfeeding populations, and 20-45% among breastfeeding populations in the absence of any PMTCT intervention [
2,
3]. As a result, the number of children living with HIV currently exceeds 3 million, most of them living in Sub-Saharan Africa [
1]. This alarming figure represents 90% of all HIV-infected children globally, with less than a tenth of them being reached with basic health services [
4]. Women in low-resource settings are also disproportionately affected, further intensifying the severity of vertical transmission, with fewer health services reaching women in rural areas [
5]. The estimated antiretroviral therapy (ART) coverage for PMTCT was only 53% in Kenya for 2012 [
6]. Few HIV-exposed infants are being placed on antiretroviral prophylaxis due to high loss to follow-up of women who do not return to the health facilities for antenatal care (ANC) or following birth. ART coverage among eligible children was estimated at only 38% in 2012 for those aged 14 and under, compared to 80.7% for Kenyans aged 15+ years [
6]. Cultural and societal factors may influence utilization, including: HIV-related stigmas, fears of status disclosure, lack of financial resources and motivation to access PMTCT services, and poor education or knowledge about PMTCT [
7,
8].
A health systems approach is essential to heed the call for the virtual elimination of mother-to-child transmission of HIV and reduction of AIDS-related maternal mortality by half by 2015 [
9-
13]. Integrating PMTCT efforts with community-based approaches can be highly effective at ensuring acceptance and uptake of healthcare services. Additionally, there is an increasing interest in exploring the use of mobile phones and mobile technology to enhance PMTCT efforts [
14-
16]. The Millennium Villages Project (MVP
www.millenniumvillages.org) in Sauri, Western Kenya, mainstreams this integration in PMTCT programs by working directly with Community Health Workers (CHWs). The CHWs follow pregnant women during the entire “PMTCT cascade” from the 1
st ANC appointment through to the end of breastfeeding. However, despite the rigorous implementation of a well-defined PMTCT curriculum, CHW activities in Sauri were hampered by a number of challenges. At the inception of MVP in the Kenyan site in 2005, CHWs were unable to follow-up adequately with every woman in the households they cover. Pregnant women were tracked using paper-based methods of data recording, the national norm in Kenya, which proved to be laborious and prone to error. CHWs would often either forget to remind women of their upcoming ANC appointments or not be aware of the next appointment.
The ANC/PMTCT Adherence System (APAS, informally referred to as “PMTCT Module”), which is a mobile Health (mHealth) tool that uses text messages (SMS) to facilitate and coordinate CHW activities around ANC and PMTCT, was implemented in the MVP cluster in October 2010 to help alleviate some of the issues linked with reduced performances. Using any standard mobile phone, readily available in Sub-Saharan Africa [
17], CHWs are able to use SMS to register patients during ANC and report their health status to a central system that provides a real-time view of the health of a community. Powerful messaging features help facilitate communication between the members of the health system and an automated alert system helps reduce gaps in treatment. The APAS works by enabling nurses at participating MVP clinics to schedule the next appointment of pregnant woman or their children, and reminders are generated by the system and sent to the corresponding CHW to visit the household and prompt the patient of their upcoming appointment. It also allows for adequate follow-up of those who have defaulted from care.
Since the introduction of the APAS in Sauri in late 2010, over 800 pregnant women have been registered into the system over a period of 2 years and have been followed by CHWs. This study aims to provide a detailed analysis of the impact of the system on retention in the PMTCT cascade and corresponding MTCT rates, and the end-user perceptions of the module by CHWs and pregnant mothers. Since we did not want to single out HIV positive women through our system and since the PMTCT cascade includes ANC visits and baby follow-ups, these outcomes were studied in addition to vertical HIV transmission rates. Overall, we aimed to show that a CHW-centered mHealth technology reminder system can improve PMTCT efforts by increasing uptake of health services, decreasing loss to follow up and decreasing vertical HIV transmission rates in Western Kenya. This study also provides recommendations on how the APAS can be further improved to satisfy the needs of both pregnant women and CHWs.
Results
Program set up
The duration of set up from concept and design to implementation of the system in the Sauri cluster was 3 months and the pilot was launched in October 2010. The system continued to operate fully until November 2012, and then operated minimally until January 2013. The toll-free SMS service contract was terminated in April 2011, but the disruption in service did not significantly affect the use of the system as most CHWs resorted to using their own phone credit to send text messages to the system and were subsequently reimbursed by the MVP office for this expense.
Costs
Investment costs were only $7,000 for the local program manager time. Programmer time, expert technical assistance and all equipment, including cell phones and server were provided by the Earth Institute. Training costs were $100 per training. All SMS to and from the system were provided free of charge to CHWs throughout the pilot. Each cell phone cost $40, the SIM card $2.50 and each SMS cost 1 cent.
Network coverage in the pilot area
With the exception of only a few households in the community, all of the district homes and clinics were within coverage areas and the network coverage was reliable. Service experienced less than 10% downtime. A major disruption in local network services halted the program for a few weeks in April and May 2011 following the end of the toll-free SMS service contract, which was resolved by reimbursing costs of SMS to CHWs.
Interviews with CHWs
The MVP site in Kenya is divided into 12 sub-locations [Figure
2], each sub-location with its own network of CHWs. Overall, there were 109 CHWs employed at the MVP site (approximately 9 within each sub-location) at the time of the launch of the pilot. Interviews were conducted 10 months after the start of the project in 4 separate focus group sessions with 20 CHWs i.e. 18% of the total CHW population. Sessions were arranged in the following 3 sub-locations: Sauri (9 CHWs), Nyamninia (4 CHWs) and Gongo (7 CHWs); each session lasting approximately 2 hours. Data collected shows 13 out of 20 participants were female.
Before the APAS was set up, CHWs recorded appointment dates during household visits using paper forms issued by health facilities. CHWs then followed up after the appointment date to check if each pregnant woman attended her appointment. CHWs stated that the paper-based tracking forms were ineffective at helping them remember every woman’s appointment.
After the launch of the APAS, reminders were sent directly to CHWs on their cell phones, which helped CHWs give appointment reminders to women up to 3 days prior to the ANC or follow up visit. All CHWs interviewed agreed that the ability to receive SMS updates on appointments was among the major benefits of the APAS. Based on focus group discussions, CHWs performed on average between 0 and 7 appointment reminders every week.
CHWs were also included in a Closed User Group (CUG), which allowed free phone calls and the ability to send text messages to the system and to each other. Poor battery life of the phones and occasional network failure sometimes hindered CHWs ability to send or receive text messages. CHWs recommended that mothers be integrated into the CUG to help reduce the potential cost of calling outside the CUG, and the provision of solar chargers for the phones.
Interviews with mothers
A sample size of 67 pregnant women and new mothersc were selected for interviews. Of these, 54 were married and 13 were single. The average number of pregnancies was 3, and the average year of the 1st pregnancy was 2003. Of the 67 women, 38 were interviewed individually and 29 in focus group sessions. Interviews were conducted in the following 7 sub-locations: Anyiko, Gongo, Marenyo, Nyamninia, Nyawara, Ramula and Sauri. Given that the annual estimated number of pregnancies within a sub-location is 4% of a catchment population of 10,000d, the women in the sample contributed towards approximately 1.7% of total pregnancies in the site.
Of the participants interviewed individually, 19 out of 38 women were issued yellow ID cards, indicating that approximately 50% of individual interview participants were registered in the APAS at the time of their pregnancy with a unique identifier number. Of the 38 women, 27, including all the 19 women registered in APAS, said that their CHW reminded them of appointments during their regular visit, and therefore they rarely missed their ANC appointments. In addition, all women interviewed were able to verify the next date of their clinic appointment in their MCH booklets, and record dates in their personal diaries or calendars.
In focus group sessions, all participants (29 women) mentioned that they had been issued yellow ID cards, and were registered in the APAS at the time of their pregnancy. Only 4 women reported never receiving appointment reminders from CHWS during their regular visit. Of all participants, 23 women noted that they had a mobile phone, and most were able to send, receive and read text messages. All agreed that they would be willing to opt-in to a service that allowed them to have reminders sent directly to their phones.
The women were also asked to share what messages their CHW communicated to them during their regular visits. Common responses were that CHWs educated them on malaria and HIV prevention methods. Furthermore, 55 out of all 67 women in the sample (82% of participants) said that their CHW visited them regularly i.e. every 1 or 2 months. All women agreed that CHWs were reminding them of clinic appointments more regularly than before the introduction of the APAS. Those who rarely saw their CHW shared that they had not seen any changes.
Routine data
ANC visits
There was, however, no statistically significant relationship between intervention 1 compared to 0 in both the univariate and multivariate models (OR = 0.78, 95% CI [0.41-1.49], Adjusted OR = 0.86, 95% CI [0.45-1.69] respectively), meaning residence inside or outside the cluster did not affect the number of ANC visits made (power = 0.05). The relationship between intervention 2 compared to 0 was nearly significant in both the crude and adjusted analysis (OR = 2.30, 95% CI [0.97-5.48], Adjusted OR = 2.37, 95% CI [0.99-5.67] respectively), hence among women who had their 1st ANC visit in the 2nd trimester of their pregnancy, women who were in the APAS and resided in the MVP cluster had twice the odds of going for more, rather than less, ANC visits compared to women who were not in the APAS and resided outside the MVP cluster (power = 0.58).
Baby follow-ups
Study population
The total sample investigated for baby follow-ups consisted of 650 women aged 14–48. The median age was 24 (IQR 20–28). The biggest age category was the 20–24 year olds (34.9% of participants), while the smallest was the 45–49 year olds category (0.2% of participants) [See Table
3]. Lihanda Health Center had the smallest representation of women (50 participants) while the rest of the facilities contributed an equal number of participants (75 each). The intervention group 0 was made up of 200 women, while the intervention groups 1 and 2 each had 225 women. Most women in this sample made their 1
st ANC visit in the 2
nd trimester of their pregnancy (348 in total). The HIV prevalence in the investigated population was 27.1%, with 176 being HIV positive and 474 being HIV negative. Most women were married (91.2%) and had 1–3 children prior to the current child (60.3%). These characteristics are also illustrated graphically in Additional file
5.
Table 3
Population characteristics by number of baby follow-ups
Age (Years)
| | | | | | | |
<15 | 3 | 0.5 | 3 | 0.5 | 0 | 0.0 | 0.0922 |
15-19 | 127 | 19.5 | 108 | 19.4 | 19 | 20.2 | |
20-24 | 227 | 34.9 | 201 | 36.2 | 26 | 27.7 | |
25-29 | 148 | 22.8 | 123 | 22.1 | 25 | 26.6 | |
30-34 | 91 | 14.0 | 73 | 13.1 | 18 | 19.2 | |
35-39 | 47 | 7.2 | 43 | 7.7 | 4 | 4.3 | |
40-44 | 6 | 0.9 | 5 | 0.9 | 1 | 1.1 | |
45-49 | 1 | 0.2 | 0 | 0.0 | 1 | 1.1 | |
Facility
| | | | | | | |
Gongo HC | 75 | 11.5 | 64 | 11.5 | 11 | 11.7 |
<0.0001
|
Lihanda HC | 50 | 7.7 | 46 | 8.3 | 4 | 4.3 | |
Marenyo HC | 75 | 11.5 | 72 | 13.0 | 3 | 3.2 | |
Masogo Disp | 75 | 11.5 | 58 | 10.4 | 17 | 18.1 | |
Mindhine Disp | 75 | 11.5 | 66 | 11.9 | 9 | 9.6 | |
Nyawara HC | 75 | 11.5 | 72 | 13.0 | 3 | 3.2 | |
Ramula HC | 75 | 11.5 | 71 | 12.8 | 4 | 4.3 | |
Sauri HC | 75 | 11.5 | 49 | 8.8 | 26 | 27.7 | |
Yala SDH | 75 | 11.5 | 58 | 10.4 | 17 | 18.1 | |
Intervention*
| | | | | | | |
0 | 200 | 30.8 | 150 | 27.0 | 50 | 53.2 |
<0.0001
|
1 | 225 | 34.6 | 189 | 34.0 | 36 | 38.3 | |
2 | 225 | 34.6 | 217 | 39.0 | 8 | 8.5 | |
Trimester of 1st ANC Visit
| | | | | | | |
1st | 4 | 0.7 | 4 | 0.8 | 0 | 0.0 |
0.9517
|
2nd | 348 | 56.9 | 297 | 56.9 | 51 | 56.7 | |
3rd | 260 | 42.5 | 221 | 42.3 | 39 | 43.3 | |
Missing
| 38 | | | | | | |
Mother's HIV Status
| | | | | | | |
Negative | 474 | 72.9 | 411 | 73.9 | 63 | 67.0 | 0.1638 |
Positive | 176 | 27.1 | 145 | 26.1 | 31 | 33.0 | |
Marital Status
| | | | | | | |
Married | 593 | 91.2 | 504 | 90.7 | 89 | 94.7 |
0.5720
|
Widowed | 2 | 0.3 | 2 | 0.4 | 0 | 0.0 | |
Single | 54 | 8.3 | 49 | 8.8 | 5 | 5.3 | |
Separated | 1 | 0.2 | 1 | 0.2 | 0 | 0.0 | |
Parity
| | | | | | | |
0 | 153 | 23.5 | 129 | 23.2 | 24 | 25.5 | 0.4654 |
1-3 | 392 | 60.3 | 339 | 61.0 | 53 | 56.4 | |
4-6 | 97 | 14.9 | 80 | 14.4 | 17 | 18.1 | |
≥7 | 8 | 1.2 | 8 | 1.4 | 0 | 0.0 | |
Key
| | | | | | | |
Chi-square and Fisher tests
Characteristics of the study population against number of baby follow-ups are presented in Table
3. The chi-square or Fisher’s exact test show that there were no statistically significant differences in the distribution of most population-level characteristics between women who made the nationally recommended 6 or more baby follow-ups and those who made less than 6 baby follow-ups (p-values >0.05) except for the facility at which they sought health care and the intervention group. The p-values for the facility at which they sought health care and the intervention group (both < 0.0001) show that there was a strong statistically significant association between the number of baby follow-ups made and the intervention group, and between the number of baby follow-ups made and facility at which the women sought health care.
Ordinal logistic regression analysis
Table
4 summarizes the univariate and multivariate logistic regression analyses of the intervention group against number of baby follow-ups. The crude (univariate) ordinal logistic regression analyses showed a statistically significant positive association between higher ordered interventions (1 vs 0, 2 vs 0 and 2 vs 1) and higher number of baby follow-ups (OR = 1.75, 95% CI [1.09-2.82], OR = 9.00, 95% CI [4.15-19.51], OR = 5.14, 95% CI [2.34-11.33] respectively). In other words, women who resided inside the MVP cluster had twice the odds of going for more, rather than less, baby follow-ups compared to women who resided outside the MVP cluster; women who were in the APAS and resided in the MVP cluster had 9 times the odds of going for more, rather than less, baby follow-ups compared to women who were not in the APAS and resided outside the MVP cluster; among women who resided in the MVP cluster, women who were in the APAS had 5 times the odds of going for more, rather than less, baby follow ups compared to women who were not in the APAS.
Table 4
Ordinal logistic regression analysis of intervention with baby follow-ups (all women)
Intervention*
| | | | | | |
1 vs 0 |
1.75
|
(1.09-2.82)
|
1.84
|
(1.13-2.98)
|
1.87
|
(1.15-3.05)
|
2 vs 0 |
9.00
|
(4.15-19.51)
|
8.96
|
(4.13-19.41)
|
8.99
|
(4.15-19.49)
|
2 vs 1 |
5.14
|
(2.34-11.33)
|
4.26
|
(1.91-9.48)
|
4.00
|
(1.79-8.93)
|
Key
| | | | | | |
The multivariate models were similar to the univariate models, showing strong statistically significant positive associations between higher ordered interventions (1 vs 0, 2 vs 0 and 2 vs 1) and higher number of baby follow-ups (Adjusted OR = 1.84, 95% CI [1.13-2.98], Adjusted OR = 8.96, 95% CI [4.13-19.41], Adjusted OR = 4.26, 95% CI [1.91-9.48] respectively in the model adjusting for mother’s HIV status), and (Adjusted OR = 1.87, 95% CI [1.15-3.05], Adjusted OR = 8.99, 95% CI [4.15-19.49], Adjusted OR = 4.00, 95% CI [1.79-8.93] respectively in the model adjusting for mother’s HIV status and parity).
Vertical HIV transmission
Study population
Among HIV positive women with data on baby’s HIV status at 18 months, the biggest age category was the 25–29 year olds (33% of participants), while the smallest was the 40–44 year olds category (1.1% of participants) [see Table
5]. Yala Sub-District Hospital had the most participants (29.6%) while Masogo dispensary had the smallest representation of women (1.1%). The intervention group 0 was made up of 33 women; the intervention group 1 had 35 women; and the intervention group 2 had 20 women. Most women in this sample made their 1
st ANC visit in the 2
nd trimester of their pregnancy (57.1%). Most women made 6 or more baby follow-ups (85.2%), and less than 4 ANC visits (51.4%). Most women were married (96.6%) and had 1–3 children prior to the current child (77.3%). The majority of the HIV positive women received ART administration (69.1%). These characteristics are also illustrated graphically in Additional file
6.
Table 5
Population characteristics by baby's HIV status at 18 months
Age (Years)
| | | | | | | |
15-19 | 4 | 4.6 | 0 | 0.0 | 4 | 4.9 |
0.9204
|
20-24 | 23 | 26.1 | 1 | 16.7 | 22 | 26.8 | |
25-29 | 29 | 33.0 | 2 | 33.3 | 27 | 32.9 | |
30-34 | 21 | 23.9 | 2 | 33.3 | 19 | 23.2 | |
35-39 | 10 | 11.3 | 1 | 16.7 | 9 | 11.0 | |
40-44 | 1 | 1.1 | 0 | 0.0 | 1 | 1.2 | |
Missing
| 88 | | | | | | |
Facility
| | | | | | | |
Gongo HC | 3 | 3.4 | 0 | 0.0 | 3 | 3.7 |
0.0914
|
Lihanda HC | 2 | 2.3 | 0 | 0.0 | 2 | 2.4 | |
Marenyo HC | 17 | 19.3 | 4 | 66.7 | 13 | 15.9 | |
Masogo Disp | 1 | 1.1 | 0 | 0.0 | 1 | 1.2 | |
Mindhine Disp | 3 | 3.4 | 0 | 0.0 | 3 | 3.7 | |
Nyawara HC | 19 | 21.6 | 0 | 0.0 | 19 | 23.2 | |
Ramula HC | 13 | 14.8 | 2 | 33.3 | 11 | 13.4 | |
Sauri HC | 4 | 4.6 | 0 | 0.0 | 4 | 4.9 | |
Yala SDH | 26 | 29.6 | 0 | 0.0 | 26 | 31.7 | |
Missing
| 88 | | | | | | |
Intervention*
| | | | | | | |
0 | 33 | 37.5 | 3 | 50.0 | 30 | 36.6 |
0.4998
|
1 | 35 | 39.8 | 3 | 50.0 | 32 | 39.0 | |
2 | 20 | 22.7 | 0 | 0.0 | 20 | 24.4 | |
Missing
| 88 | | | | | | |
Trimester of 1st Visit
| | | | | | | |
1st | 0 | 0.0 | 0 | 0.0 | 0 | 0.0 |
0.0677
|
2nd | 48 | 57.1 | 5 | 100.0 | 43 | 54.4 | |
3rd | 36 | 42.9 | 0 | 0.0 | 36 | 45.6 | |
Missing
| 92 | | | | | | |
Baby Follow Up's
| | | | | | | |
<6 | 13 | 14.8 | 2 | 13.3 | 11 | 13.4 |
0.2144
|
≥6 | 75 | 85.2 | 4 | 66.7 | 71 | 86.6 | |
Missing
| 88 | | | | | | |
ANC Visits
| | | | | | | |
<4 | 45 | 51.4 | 0 | 0.0 | 45 | 54.9 |
0.0112
|
≥4 | 43 | 48.9 | 6 | 100.0 | 37 | 45.1 | |
Missing
| 88 | | | | | | |
Marital Status
| | | | | | | |
Married | 85 | 96.6 | 6 | 100.0 | 79 | 96.3 |
1.0000
|
Widowed | 1 | 1.1 | 0 | 0.0 | 1 | 1.2 | |
Single | 2 | 2.3 | 0 | 0.0 | 2 | 76.8 | |
Separated | 0 | 0.0 | 0 | 0.0 | 0 | 0.0 | |
Missing
| 88 | | | | | | |
Parity
| | | | | | | |
0 | 8 | 9.1 | 0 | 0.0 | 8 | 9.8 | 0.1064 |
1-3 | 68 | 77.3 | 5 | 83.3 | 63 | 76.8 | |
4-6 | 11 | 12.5 | 0 | 0.0 | 11 | 13.4 | |
≥7 | 1 | 1.1 | 1 | 16.7 | 0 | 0.0 | |
Missing
| 88 | | | | | | |
ART Administration
| | | | | | | |
No | 58 | 69.1 | 5 | 83.3 | 53 | 68.0 |
0.6608
|
Yes | 26 | 31.0 | 1 | 16.7 | 25 | 32.1 | |
Missing
| 92 | | | | | | |
Key
| | | | | | | |
Chi-square and Fisher tests
Among the 176 women who were HIV positive, vertical HIV transmission was first studied using a chi-square analysis and Fisher’s exact tests. Characteristics of the study population against the baby’s HIV status at 18 months are presented in Table
5. In this sample of HIV positive women, 88 did not have data on the baby’s HIV status at 18 months. The chi-square or Fisher’s exact tests show that there were no statistically significant differences in the distribution of most population-level characteristics between women who had an HIV positive baby at 18 months and those who had an HIV negative baby at 18 months (p-values >0.05) except for the number of ANC visits (p-value <0.05).
Transmission rates
The chi-square analysis and Fisher’s exact tests did not show a statistically significant relationship between the intervention levels and baby’s HIV status at 18 months [see Table
5]. However, an exact binomial test for proportions showed that the vertical HIV transmission rate at 18 months for women registered in the APAS was significantly different from that of women not registered in the APAS and was significantly different from the global vertical HIV transmission rate of 30% (data not shown). The intervention groups 1 and 0 each had 3 HIV positive babies at 18 months (9% transmission rate) while intervention group 2 had a 0% transmission rate [see Table
5], hence none of the HIV positive women who both resided in the MVP cluster and were in the APAS transmitted the HIV virus to their babies. Although 18 months is a more definite time point to investigate vertical HIV transmission as most babies are weaned from breastfeeding, transmission rates were also measured at an earlier time point (when the babies were 9 months) with a bigger sample (only 10 women missing data as opposed to 88), and the transmission rates for intervention groups 0, 1 and 2 were 14.4%, 8.3% and 0% respectively (data not shown).
Discussion
Mobile telephony has become ubiquitous in Africa and more people have access to mobile phones in Sub-Saharan Africa than to safe drinking water or electricity [
21]. Increasingly, development programs are harnessing the potential of mobile phones and mobile technology for health system improvement, including for nutrition, tuberculosis, malaria [
22] and maternal and child health [
14-
16,
23-
26]. In 2010, the Joint United Nations Programme on HIV/AIDS (UNAIDS) called for the virtual elimination of mother-to-child transmission of HIV and reduction of AIDS-related maternal mortality by half by 2015 [
9-
13]. Our paper shows that the implementation of the APAS in MVP’s Sauri cluster in Western Kenya (formerly Nyanza Province) has proven effective in tracking and improving the treatment-seeking behavior of pregnant women, during pregnancy and up to 18 months following delivery, and could therefore effectively complement on-going PMTCT efforts to reach the goal of virtual elimination by 2015.
In order to ensure a safe pregnancy, 4 ANC visits are standardly recommended, including in Kenya [
27,
28]. Additionally, the 1
st visit is recommended during the 1
st trimester (around 15 weeks) to allow for an appropriate spacing of the ANC visits throughout the pregnancy [
27]. In our study however, most women presented during the 2
nd trimester, as is the norm in rural sub-Saharan Africa settings [
27]. Based on routine based data from the sample of women who had their 1
st ANC visit in the 2
nd trimester of their pregnancy and resided in the MVP cluster, women who were enrolled in the APAS had 3 times the odds of going for more ANC visits rather than less, compared to women not enrolled in the APAS after adjusting for the mother’s HIV status. Therefore, the strength of our proposed mHealth solution lies in the span of its impact to all women, regardless of their HIV status. As shown by our study, women enrolled in the APAS were more likely to undergo the 4 recommended ANC visits compared to women not enrolled in the APAS. In our sample, residence inside or outside the cluster did not seem to affect the number of ANC visits made, but this could be explained by low statistical power in the analysis comparing groups 1 and 0, and by the fact that women from outside the cluster attend MVP clinics, which all focus strongly on the need for the 4-recommended visits for a safe pregnancy. Also, there are CHWs in other areas around MVP that are part of the national program, but they are not structured in the same way as MVP and their curriculum is different (less complete and does not include PMTCT).
The interesting caveat of our mHealth system is that it can only effectively work if women present to the ANC clinic earlier rather than later. From data collected during interviews and focus groups, women shared that family planning and HIV prevention were among the main educational messages disseminated by CHWs during their regular visits. While this data suggests that women in the community are aware of the importance of fulfilling their clinic appointments, only 4 women (0.6%) in the sample used for routine based data analysis made their 1
st ANC appointment in their 1
st trimester, underlining a stark difference between knowledge and practice. Additionally, there may be cultural reasons that further explain or justify why pregnant women tend to wait until the 2
nd trimester before presenting to a health clinic [
8]. The method of counseling that CHWs adopt in the community to convey messages around ANC and PMTCT are as important as the content of the message itself and these methods empower pregnant women to realize the importance of clinic visits for both themselves and their children. CHWs are therefore a critical piece in the system and their role is not just focused on the adherence component, rather is continuous throughout the year as they also allow the mHealth system to be more effective by convincing pregnant women to present to clinics earlier than the 2
nd trimester. We believe that, whenever possible, CHWs should complement any mHealth tool designed to increase adherence, whether it be focused on HIV, TB or even chronic diseases such as diabetes.
In order to ensure appropriate and complete postnatal care, 6 baby follow-up visits are recommended, covering all mandatory vaccinations, nutritional assessments and Early Infant Diagnosis for HIV-exposed infants [
29]. In our study, we found that women who resided inside the MVP cluster had twice the odds of going for more baby follow-ups rather than less, compared to women who resided outside the MVP cluster. Additionally, among women residing in the MVP cluster, women who were registered in the APAS had 4 times the odds of going for more baby follow-ups rather than less, compared to women not registered in the APAS, after adjusting for mother’s HIV status and parity. The APAS therefore greatly increases the likelihood of women making the recommended 6 baby follow-ups. This could lead to improved infant and young child health outcomes. Interestingly, we also showed that there was a statistically significant association between number of baby follow-ups and the facility at which women sough health services (a similar association was not uncovered for ANC visits). This could be linked to a variety of reasons, ranging from the perceived quality of services provided, the friendliness of the staff or more simply, the wait time during special days organized around infant and young child care (vaccination, nutritional assessment etc.). Additionally, in our sample, residence inside or outside the cluster did not seem to affect the number of ANC visits made but there was a statistical difference when looking at baby follow-ups. This could possibly indicate that distance to the health center does not seem to impact access to ANC, but does have a negative influence on the future follow-ups the mother undertakes with her baby. Any program interested in improving their follow-up rates should therefore also look into the improvement of these important factors.
One of the goals envisioned for the APAS upon its inception was to help alleviate vertical HIV transmission rates while also providing additional functionalities to reach all pregnant women, regardless of HIV status. Our routine based data analysis shows that while the APAS does not disclose to CHWs a woman’s HIV status at any point during her pregnancy, through registration in the system, CHW activities are proving to be particularly effective in reaching HIV positive mothers. While we could not show a statistically significant relationship between enrollment in the APAS and a baby’s HIV status at 18 months using a chi-square analysis or Fisher’s exact test, an exact binomial test for proportions showed that the vertical HIV transmission rate at 18 months for women registered in the APAS was significantly different from that of women not registered in the system as well as significantly different from the global vertical HIV transmission rate of 30%.
All 20 HIV-positive women in the randomly selected sample of registered APAS users living in the MVP cluster with complete data on baby’s HIV status at 18 months gave birth to HIV negative babies, compared to the 9% vertical HIV transmission rate for women not registered in the APAS but living in the cluster, and a similar rate for women outside the cluster, suggesting that the system is alleviating MTCT in the community. Transmission rates at 9 months following birth were 14.4% for women not registered in the APAS and residing outside the MVP cluster, 8.3% for women not registered in the APAS but residing inside the MVP cluster, and 0% for women both registered in the APAS and residing inside the MVP cluster. This suggests that the efforts undertaken by MVP to reduce MTCT within the cluster have effectively close-to halved the vertical HIV transmission rate. However, only through the use of the APAS was it possible to eliminate transmission altogether, in the sample we analyzed. Interestingly, we also showed that the number of ANC visits during pregnancy was directly correlated to the transmission rate, which confirms what was intuitively anticipated.
Our study had several limitations that were independent from our work. PMTCT indicators improve as women attend their 1st ANC appointment during the earliest stages of pregnancy i.e. within the 1st trimester. If a pregnant woman makes her first appointment during later trimesters, she will have fewer opportunities to make the 4 ANC visits required before delivery. As such, our study could only effectively study women who made their 1st ANC visit in the 2nd trimester of their pregnancy (as only 4 women visited the clinic in the 1st trimester). When all women were considered regardless of trimester of 1st ANC visit, the statistical relationship between number of ANC visits made and registration in the APAS was not valid. This could be explained by a reduced opportunity for the health system to adequately provide effective services when the women presented later in their pregnancy. Interestingly, even when women were presenting later, they were still more likely to attend up to 4 ANC appointments during the course of their pregnancy if they were registered in the APAS. Overall, this stresses the importance for nurses and CHWs to continue to counsel women and build awareness around PMTCT to encourage more women to present themselves at the clinic as early as their 1st trimester.
Another concern is that HIV positive women were oversampled in this study in an effort to match the regional HIV prevalence [15.1% in Nyanza Province in 2012] [
30] and sample a population large enough to study vertical HIV transmission rates. These efforts, however, resulted in an increased overall HIV prevalence in our sample to 27.3% and possibly led to a bias in our results. We attempted to decrease this bias by adjusting for mother’s HIV status as a confounder in our final models. A separate analysis conducted with only HIV negative women did not significantly change the results (data not shown).
Since this study was a post-intervention analysis, a post-hoc power analysis for the main outcome was conducted. This analysis showed that the power for the relationship between intervention groups 1 and 0 (women not registered in APAS and residing inside and outside the MVP cluster respectively) was very low (0.05). This particular analysis was not powered enough to detect a significant relationship between the 2 groups. However, the study may have overall been over powered since a power of 0.91 was observed in the analysis of the relationship between intervention groups 2 and 1 (women residing in the MVP cluster but registered and not registered in APAS respectively). Since a large effect size (OR of 2.91 in the crude analysis for ANC visits) was observed, however, the findings are still an important indicator of the impact of an mHealth intervention on aiding PMTCT efforts.
Another limitation is that the study only included women with complete information on baby follow-ups in order to be able to accurately assess the total number of baby follow-ups each woman made and trace the impact of the intervention all the way to the time the babies were 18 months old for HIV positive women. This could have introduced a bias if there was something characteristic about women with missing data points on baby follow-ups. The MCH nurses however attribute missing data points on baby follow-ups in the MCH register to merely operational failure. Additionally, since this selection process was applied to all three groups, this possible bias is not likely to impact the difference between groups.
It is also important to note that operationally, CHWs were supposed to register all pregnant women in their sub-location into the APAS, hence “group 1” (women who resided inside the MVP cluster but were not registered in the APAS) was a result of CHWs or clinical staff not adhering to the MVP protocol. The fact that CHWs continued to follow up with this group of women during pregnancy to provide appointment reminders (which they would record in their MCH booklets) and also educate them on HIV and family planning as they did before the introduction of APAS could potentially have biased our results, but this group was different from “group 2” (women who both resided in the MVP cluster and were registered in the APAS) in the fact that group 2 received more systematic and more frequent reminders as demonstrated in Figure
1.
The MVP team also encountered several problems with the system, mainly due to the suspension of the agreement with the network provider (which provided a toll free SMS service) after April 2011. This challenge brought to light the influence providers can have on the success or failure of an SMS-based project. Following the suspension, CHWs were unable to work with the system for a few consecutive weeks; and based on data from interviews, many shared that they were receiving backlogged messages, that is, appointment reminders for women who had already delivered. Furthermore, CHWs were unable to register newly pregnant women to the system, and returned to using Pregnancy Tracking forms. Most CHWs resorted to using their own phone credit to send text messages to the system and were reimbursed by the MVP office for this expense. Future SMS-projects piloted, either in MVP program sites or by any group or government considering mHealth interventions, will need to ensure that agreements with network providers are secure, based on long-term and mutually beneficial relationships.
Our system was easily set-up on the already existing ChildCount + platform rolled out across all MVP sites, including Sauri, Kenya. A small initial investment in equipment and programmer time was sufficient to set up the SMS-based alert system, and running costs during the project were minimal. Other programs interested in setting a similar reminder system may not have access to the CC+ backbone that has simplified registration of patients in our case. This should not however detract other groups from exploring similar SMS-based tracking systems. A complete standalone system that does not rely on CC+ can be devised, and our study only serves to demonstrate the impact of the tool following its implementation.
Based on results found, one improvement to the system would be to gradually complement the CHWs component within the workflow by having text messages sent directly to women using their mobile phone number as the unique identifier. Currently, as described previously, CHWs are a key component of the system, and its effectiveness is reliant on their performance. This option will remove the system’s complete dependence on CHWs, and pregnant women will be included in the Closed User Group (CUG) and contract agreement with the network provider. A challenge with this option is that not all women in households in Sub-Saharan Africa have access to a phone-only 20% of women interviewed mentioned that they owned a phone. Most phones in the households belong to the husband or partner, which would raise issues of confidentiality. Further, there will be additional costs to the project should women be included in the CUG. Hence this option will be feasible should mobile phone penetration rates increase and more women opt-in into the idea of having reminders sent to their phones.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
IM contributed to the design of the study, acquisition of data, interpretation and statistical analysis of data, drafting and critical revision of the manuscript. CL contributed to the design of the study, acquisition and interpretation of data, drafting and critical revision of the manuscript. CIH contributed to the design and implementation of the study and critical revision of the manuscript. YBA contributed to the original idea of the study, its conception, design, drafting, the interpretation of data and finalizing the manuscript. All authors read and approved the final manuscript.
This work was completed when the authors Ivy Mushamiri, Chibulu Luo and Casey Iiams-Hauser were affiliated with the Millennium Villages Project, the Earth Institute, Columbia University.