Effects of an mHealth voice message service (mMitra) on maternal health knowledge and practices of low-income women in India

Background

• Mobile Health (mHealth) is becoming an important tool to improve health outcomes in maternal, newborn and child health (MNCH).
• Studies have demonstrated the effectiveness of mHealth interventions in improving the uptake of recommended maternal services, such as antenatal visits.
• Evidence of the impact on maternal health outcomes is still limited.
• Urban slums are home to some of the poorest and most underserved populations.
• Women and children in these settings often die from preventable causes due to living conditions and limited access to care.
• India has some of the world’s largest slums with high burdens of disease and increasing mobile phone penetration rates.
• Simple measures like vaccinations, antibiotics, improved nutrition and/or supplementation can improve health outcomes, but generating demand and promoting healthy behaviors are challenging.
• Mobile health (mHealth) is becoming an important mechanism to improve maternal, neonatal and child health (MNCH) globally.
• Mobile phones enable pregnant women to access/receive information, potentially leading to improved uptake of MNCH services and health outcomes.
• Mobile phones have also helped reach individuals who do not normally engage with the health system.
• Systematic reviews show mHealth can improve maternal health service uptake, retention in care, adherence to treatment, and use of facility-based services such as antenatal care (ANC) visits.
• More research is needed to assess the impact of mHealth on clinical outcomes.
• A pseudo-randomized controlled trial tested the hypothesis that mMitra, a mobile phone-based voice messaging service in India, would lead to improved antenatal care (ANC) practices and maternal self-care knowledge, resulting in better maternal health outcomes.
• The study focused on lower levels of anemia in women, a priority maternal health outcome in India.
• Despite progress between 1990 and 2017, India’s maternal mortality was still high at 145 per 100,000.

Methods

• A pseudo-randomized controlled trial was conducted from January 2015 to December 2017 to evaluate the impact of mMitra.
• mMitra is the India country program affiliated with the Mobile Alliance for Maternal Action (MAMA).
• MAMA is a 4-year global initiative focused on improving the health and wellbeing of pregnant women and their newborns/infants through mobile phone-delivered age- and stage-based tailored voice or text messages.
• ARMMAN, a non-profit, designed and implemented mMitra with support from MAMA.
• mMitra aimed to help pregnant women improve self-care and uptake of key MNCH services during pregnancy through the first year of their children’s lives, primarily through behavior change.
• The program has been implemented in nine states in India and has reached over 2 million women.
• A series of three questionnaires were administered to assess changes in knowledge and uptake of practices known to improve maternal health outcomes, in addition to acceptance of the mMitra intervention, with a targeted focus on urban slums in Mumbai, India.
• This impact evaluation is registered with the ISRCTN (Trial# 88968111).
• Mumbai has 27 municipal wards, each with a population of roughly 800,000-900,000.
• Each ward is typically served by one maternity home and five or six health posts that provide pregnancy and infant health care services.
• Each ward appoints approximately 100 Angan Wadi Workers (AWWs) who make home visits, register pregnant women, and motivate them to seek mother and child care.
• Two wards (F North and M East) were purposely selected for having a large area, high population density, and lack of prior exposure to mMitra.
• Researchers canvassed the wards to identify pregnant women for enrollment in the study.
• Pregnant women in the two wards who spoke Hindi or Marathi were identified by research team members.
• The team members followed up with eligible women to inform them of the study, obtain voluntary consent, and enroll and assign them to the intervention or control group.
• Women without access to a mobile phone at home and not likely to be in Mumbai for 4–5 months during pregnancy and after delivery were not enrolled in the study.
• The sample selection was single-blind.
• Women gave informed written consent prior to inclusion in the study and were not made aware of whether they were included in the intervention or the control group.
• Assignment to the intervention or control group was based on time of enrollment.
• For every 4 women consecutively enrolled in the study, the first 3 were assigned to the intervention group and the remaining woman was assigned to the control group.
• This approach was used in an attempt to ensure sufficient numbers of women were enrolled based on gestational age or pregnancy trimester to test the hypothesis that longer exposure to the messages would have an effect on outcomes (dose response) as the messages are designed to be delivered to coincide with the specific stage of the pregnancy.
• No other interventions were introduced alongside the messages.
• There was no known information contamination that was identified in the study.
• Women assigned to the intervention and control groups were followed by the investigators throughout their pregnancy until their infants turned one year of age.
• The intervention group received the mMitra voice messages in Hindi or Marathi, and the control group did not receive messages at all.
• One hundred and forty-five audio messages comprised the mMitra call package.
• The messages were designed by BabyCenter, adapted to local practices in partnership with ARMMAN and representatives from the Federation of Obstetric and Gynecological Societies of India and the Indian Academy of Pediatrics, and based on global (World Health Organization) and local (National Health Mission) guidelines.
• The audio messages were timed to the gestational age and developmental stage of the fetus and infant, respectively, and covered content relevant to women from when they were 6 weeks pregnant until their child reached 1 year of age.
• Specifically, this included nutrition, iron/folic acid supplementation, calcium supplements, development of the fetus, ANC reminders, anemia, rest during pregnancy, blood tests, HIV testing, sonography, danger signs during pregnancy, birth preparedness, identifying labor pains, cutting the umbilical cord/dressing the stump, family planning, sanitation and hygiene, breastfeeding practices, colostrum, and sex determination.
• The messages were designed to be delivered two times per week during pregnancy but were clustered at one message per day immediately post-partum for a week and then reduced back to two messages per week until the child completed 1 year.
• In case women missed messages, they could initiate a free call-back service within 2 days of the original call.
• The messages, as well as their audio translations into Hindi and Marathi, were tested for appropriateness and cultural nuances with local health experts and through in-field focus group studies.
• The voice and tone of the female recording artist were field-tested.
• The final message product was roughly 2 min long, began with a recognizable ‘jingle’ to alert the woman and her family, and ended by summarizing the key learning point.
• The sampling approach and sample size estimates were selected to ensure representativeness of the study population to the target population.
• The sample size was determined using a two-sample z proportion test.
• Sample size estimates were calculated for the primary outcome of interest: reduction of anemia during pregnancy.
• The District Level Household and Facility Survey 4 for urban poor in Maharashtra estimates the prevalence of anemia in pregnancy at 68.5%.
• The sample size needed to detect a 10% reduction in anemia, assuming 30% attrition at an alpha of 0.05 and 80% power, was 392 per arm.
• There were three rounds of data collection during this study: Pregnancy (baseline/Time 1); Post-partum (Time 2) and when the infant turned 1 year old (Time 3).
• There were no maternal health outcomes or indicators collected at Time 3.
• Satisfaction with the mMitra service for women receiving the service was assessed through a series of questions using a 5-point Likert scale.
• Data were collected by six female investigators who were part of the research team and went from house-to-house.
• During the baseline period (Time 1), the investigators visited 100–110 homes per day to enroll 4–5 women.
• Women who agreed to enroll in the study signed a consent form and were then allocated either to the intervention or control group based on their time of enrollment.
• They then completed a survey administered orally by the investigator in either Hindi or Marathi – based on participant preference.
• The Time 1 Survey Questionnaire has been included as Supplementary Material S1-First interview mMitra.
• Their response was recorded on an Android tablet using the Kobo Collect platform.
• In the event of connectivity issues or low battery, the investigator filled out a paper-based form and entered the data into the tablet later that same day.
• At the end of each day, the investigators submitted their tablets to their supervisors.
• The supervisors reviewed the surveys and uploaded the data onto the central server.
• Every day, a data manager examined the uploaded data for completeness and consistency in responses, and any problems identified were discussed in daily morning briefings with the research team and addressed.
• The subsequent rounds of interviews were completed in the same manner.
• In addition to the interviews, the investigators copied clinical information from the Mother and Child Health (MCH) card into the Kobo Collect platform.
• The MCH card is routinely issued to each woman at the health facility and contains information on services provided and clinical/laboratory findings (e.g., weight, blood pressure, hemoglobin level) of the woman and the child until the child reaches 1 year of age.
• Information on maternal health knowledge and attitudes were collected at Times 1 and 2 through surveys. Data on maternal health outcomes were collected at Time 2 through surveys and MCH cards.
• In the intervention group, experience with mMitra was assessed at Time 2 through a semi-structured questionnaire.
• The Time 2 Survey Questionnaire has been included as Supplementary Material S2- Second interview mMitra.
• In terms of study compliance, of the original 2016 women enrolled, 1760 women (87%) were reached for Time 2 interviews.
• 256 (13%) women could not be interviewed because some had gone to their mother’s house for delivery and had not returned home at the time of the interview held within 2 weeks after delivery (many of them had subsequently returned home and participated in Time 3 interviews when child health information was sought).
• Five women were reached but refused to be interviewed at Time 3.
• The primary outcome of interest was anemia reduction.
• The data for this outcome were obtained at Time 2 and sourced from both the survey and triangulating with their MCH cards.
• It was selected for its clinical importance (e.g., low birth weight, premature birth, and maternal mortality) and prioritization by the Government of India for intervention.
• The hemoglobin level measured closest and prior to her date of delivery was used and obtained during Time 2; the data source was the MCH card.
• Hemoglobin levels greater than 110 g per Liter (g/L) were considered normal and indicative of the absence of anemia.
• There were several additional outcomes of interest, including maternal health seeking knowledge, attitudes and practices.
• Most of the indicators related to maternal health seeking and care were included in the first two survey rounds (Times 1 and 2).
• Women’s responses to questions on proven ANC and post-natal care practices were compared at Times 1 and 2 and then in the intervention group versus control group.
• First, descriptive analyses were conducted to obtain a general sense of the data.
• Continuous data were compared between the intervention and control groups using two-sample t-tests.
• When comparing continuous data from Times 1 and 2 across groups, the two-paired t-test was used.
• Categorical outcomes were assessed using chi-square tests.
• For the outcomes data, simple and binary logistic regressions were conducted.
• Intention to treat and per protocol analyses were conducted and compared.
• All data were analyzed using SPSS version 18.0.
• The study was approved by the Foundation for Research in Health Systems’ Institutional Review Board under Protocol Number HHS00009235, and written informed consent was obtained from all participants.

Results

• The evaluation of mMitra took place from June 2015 to January 2017 where over 23,500 households were visited and 2050 pregnant women were identified as eligible, and a total of 2016 pregnant women were enrolled in the study.
• 1516 women were allocated to the intervention group and 500 to the control group.
• Timeline Enrollment and baseline surveys (Time 1) lasted from June–October 2015, and post-partum surveys (Time 2) started in November 2015 and ended in March 2016 when all women enrolled had delivered their babies.
• At baseline (Time 1), the 1516 women in the intervention group and 500 women in the control group were similar in age, parity, education, family type, and membership in a social community.
• Women in the intervention group were more likely to be employed, own a mobile phone, listen to a radio, read the newspaper, and have literate and employed husbands.
• This changed at Time 2: women in the intervention group were more likely to be older (p = 0.001) and watch TV (p = 0.013) yet have no difference in employment status, newspaper reading, or their husband’s literacy or employment status as compared to controls.
• Of the women enrolled, 1750 (87%) were reached at Time 2 where 174 women (11.4%) were lost to follow-up in the intervention group compared to 92 (18.4%) in the control group.
• Reasons for attrition included moving out of the area, addresses not locatable, mobile phones switched off or women refusing to be interviewed.
• One woman in the control group died during delivery.
• 16 (3.9%) of the 418 women in the control group reported receiving mMitra calls, and 229 (17%) of the 1342 women in the intervention group reported never receiving mMitra phone calls.
• An NGO working in the neighboring area had inadvertently enlisted a few women in the control group to receive mMitra messages.
• The women in the intervention group who reported that they did not receive mMitra calls was due to Do-not-Disturb (DND) option being activated on their phones so the messages could not be delivered.
• The data for these women were discarded, resulting in the data of 1515 women being analyzed (intervention n = 1113, control n = 392).
• Women in both groups did equally well in retaining their MCH cards.
• There were 1734 women (87.5% in the intervention group, 81.6% in the control group) who showed their MCH cards at Time 1.
• At that time, 12% of women in their first trimester reported that they were yet to receive their cards, and 19% had the cards but refused to show them to the investigators.
• More women in both groups at Time 2 showed their cards to investigators (93.4% in the intervention group, 94.8% in the control group).
• There was a small but significant difference in the recording of health information in the MCH cards, with women in the intervention group having more complete cards compared to controls for ANC visits completed, weight and blood pressure measurements during ANC visits, and receiving a tetanus shot.
• There were discrepancies found between women’s self-reports (via the surveys) and documentation in the MCH cards for both groups.
• For example, over 90% of women self-reported making at least 3 ANC visits, but only 20% of cards had a record of three or more ANC visits.
• Similarly, 96% of women self-reported receiving a tetanus toxoid injection while only 25% of cards had recorded administration of a tetanus toxoid injection.
• None of the MCH cards showed women being provided or prescribed calcium tablets, though over 67% of women reported taking calcium tablets during pregnancy.
• The practice levels at the baseline ranged from 28% (early registration of pregnancy) to 99% (took rest/nap in the afternoon), which showed consistency between practice and knowledge.
• The intervention group performed significantly better on the following 4 practice indicators: receiving the tetanus toxoid injection (OR: 1.6, 95% Confidence Interval (CI): 1.05–2.4, p = 0.028), consulting a doctor if spotting or bleeding (OR: 1.72, 95%CI: 1.07–2.75, p = 0.025), saving money for delivery expenses (OR: 1.79, 95%CI: 1.38–2.33, p = 0.0001), and delivering in hospital (OR: 2.5, 95%CI: 1.49–4.35, p = 0.001).
• Women in the intervention group had 28.5% increased odds of registering their pregnancy in the first trimester compared to controls (p = 0.077).
• The control group performed significantly better than the intervention group on 2 practices indicators: resting regularly during pregnancy (OR: 0.7, 95%CI: 0.54–0.88, p = 0.002) and having at-home deliveries attended by a skilled birth attendant (OR: 0.46, 95%CI: 0.23–0.91, p = 0.027).
• The groups performed similarly for the remaining practices during pregnancy, especially those related to anemia control, such as consuming iron and folic acid supplements, eating more frequently, eating nutritious food during pregnancy, and consulting the doctor for symptoms like tiredness (when experienced).
• Clinical data on hemoglobin levels were only available for 24.3% of women in the control group and 33.8% of women in the intervention group, so anemia status at or near the time of delivery could not be assessed.
• In the study population, correct responses to various knowledge questions at the baseline ranged from 13 to 98% with the lowest performing knowledge question being “During which month of pregnancy should a woman first see a doctor? (first trimester: 13%).
• Responses to nutrition-related questions were equivocal, like As compared to her usual intake, should a woman eat more/less/usual quantity of food during pregnancy? (More = 46%) and which tablets should women take for supplementation? (Iron =56%). However, near-universal awareness was recorded to the question “pregnant women need to take more rest/nap in the afternoon (92%).
• For 13 of the 15 knowledge indicators compared at Times 1 and 2, there were improvements within both groups, but only one knowledge indicator was significantly increased among women in the intervention group: women should see a doctor in the first trimester of pregnancy.
• Women in the intervention group had 2.13 times increased odds of agreeing that medical support is required during pregnancy compared to controls (95%CI: 1.17–3.86, p = 0.013).
• There were no indications of differential knowledge on the other health topics, including other reasons for medical support during pregnancy, nutrition and supplements, and birth spacing.
• There was no discernable association between background demographic characteristics and knowledge levels.
• 975 (88%) report receiving the mMitra calls, 43% of women reported listening to the calls by themselves, roughly 79% of the women who reported receiving the calls said that they often to always listened to the entire message and close to 91% of women were satisfied to very satisfied with the mMitra calls and 92% found that the calls were useful.
• Only 28.7% knew about the “missed call” function, and 19.2% reported using it.
• Approximately two thirds of respondents reported discussing information contained in messages with family members and friends (60% with other family members, 61% with mother-in-law, 65.2% with friends), and 86% reported discussing the information with their husband.
• Over 60% of women reported that mMitra had improved their health awareness, and by sharing that information with family members, they felt comfortable seeking assistance in daily chores and in taking care of themselves; however, only one third of women said that they felt comfortable discussing pregnancy-related problems with the health staff.

Discussion

• Tailored mobile phone-based voice messages can improve key maternal behaviors and maternal knowledge among women in low-resource settings.
• Women exposed to mMitra messages improved adoption of both home- and health facility-based practices that can positively impact maternal and child health outcomes.
• Exposure to a voice messaging service, like mMitra, could potentially improve access to and use of recommended services for routine and high-risk pregnancies, thereby improving maternal outcomes.
• Women in the intervention group were more likely to receive their tetanus toxoid injection and deliver in hospital.
• There was a trend towards registering their pregnancy in the first trimester, and there were also slightly more women who completed three or more ANC visits.
• These findings align with previous reports where mobile phone text-based messages increased attendance of ANC visits in low and middle-income countries and recent studies that report that mobile phone text-based interventions improved skilled birth attendance and uptake of facility-based maternal care.
• While it is reasonable to assume change in knowledge levels will lead to behavior change and practices, additional research is needed to map out the pathways through which mHealth interventions contribute to behavior change for MNCH, and to what degree those depend on change in knowledge alone as well as exposure to the intervention.
• Since voice messages are accessible to women of any literacy level and low techno-literacy levels, it will be useful to compare voice and text on ‘softer’ outcomes such as emotional connection, relatability and tailoring.
• Clinical data on hemoglobin levels were only available for 20–30% of cards, limiting the sample size needed to detect a statistically significant difference in the outcome of interest and minimizing confidence in the data.
• The fact that significant increases in the information recorded on the number of ANC visits and patient-level data among women in the intervention cohort was an important positive result of the intervention.
• Significantly fewer women in the intervention group reported taking rest and/or taking naps compared to the control group, as fewer women in the control group were employed and were more likely to have an opportunity to take naps/rest regularly.
• Women in the intervention group appeared to take a more active control of their health during pregnancy and prepare for delivery, including saving money for delivery expenses and consulting a doctor in case of spotting/bleeding.
• Women receiving the mMitra messages shared and discussed the information they received with others, especially family members, and felt more confident asking for assistance around the house.
• This is especially important in communities where behavior change depends on buy-in and support within an individual’s social-ecosystem and underscores the potential of tailored voice messages to create shared emotional connections to the pregnancy and to strengthen such social support systems.
• Future studies could test whether sharing such information is directly correlated with increased adoption of maternal care practices.
• The study did not show many significant changes in increased knowledge on maternal care topics in the intervention group because not all knowledge indicators measured in this study were directly correlated to specific messages or practices, as the study instead aimed to gain a general understanding of knowledge on a broad set of topics related to maternal, newborn, and child care among poor, urban women.
• Knowledge levels did increase in both the control and intervention groups as pregnancy continued suggesting that women had access to other quality sources of information (peers, health workers, TV, radio media etc.) possibly due to being in urban settings, making it difficult in this context to attribute any increase in knowledge to the mMitra intervention alone.
• The level of knowledge on some topics was already high, and two-thirds of women in both control and intervention groups had been pregnant before, potentially making them extremely familiar with these topics already.

Limitations

• The study relied primarily on self-reported data obtained by survey from women, which could be subject to recall bias and even inaccurate or false reporting by women.
• Loss to follow-up and compliance differences in the control and intervention groups led to reductions in data that was included in the analyses.
• Clinical records like MCH cards can provide valuable information on actual ANC services received and help overcome self-reporting bias, but issues related to data quality in official health records can be difficult to overcome, especially when clinics are crowded, and healthcare providers are very busy providing service or women forget or lose their cards.
• There were discrepancies between data on the MCH cards and what was reported by women.
• There was a missed opportunity to obtain and use system data, which is objective and not subject to the same biases as self-reported data.
• Another limitation of the study was the inability for the data to be linked to specific behavioral practices as there was not a one-to-one matching of messages to health outcomes or indicators.
• Several mMitra messages address the same outcome, and single messages can also be linked to multiple outcomes and indicators.
• The contamination of the control group was due to mMitra implementations outside of the impact evaluation being conducted in nearby areas or women in the intervention group informing their friends of the mMitra messages.
• The randomization method is susceptible to misclassification, so the study may not have been as well balanced as a traditional randomized controlled trial.
• Intention-to-treat analyses could not be conducted nor compared to the per protocol analyses as part of sensitivity analyses.

Conclusion

• This study adds to the growing body of evidence on the impact of mHealth interventions on behaviors proven to improve MNCH outcomes.
• This study is one of the first voice interventions for MNCH to be evaluated systematically by a pseudo-randomized controlled trial design in a low–resource setting.
• The findings suggest that the voice messages may make women more inclined to act on information, even though big differences in knowledge were not apparent.
• Further research is recommended to assess the relationship between changes in knowledge and behavior as well as to compare voice and text message interventions ‘head-to-head’ on MNCH outcomes and to systematically explore the differential impacts of tailored voice messages compared to text messages on behavior and practice changes among women’s family members.