How Google and ARMMAN's innovative partnership is transforming maternal healthcare in India

In the heart of India, a quiet revolution is underway, powered by the duo of artificial intelligence and the humble mobile phone. Dr Aparna Hegde, a passionate advocate for maternal health and the Founder, Chairperson and Managing Trustee of ARMMAN, and Dr Aparna Taneja, Research Lead at Google DeepMind India, have joined forces to tackle one of India's most pressing challenges: reducing preventable deaths during pregnancy and childbirth.

Their weapon of choice? An innovative AI-powered system that predicts which expectant mothers are most at risk of disengaging from essential healthcare information programs, allowing ARMMAN, Dr Hegde's non-profit organisation, to provide timely and targeted support.

Business Today delves into this groundbreaking partnership, exploring how AI is not just enhancing outreach efforts but also revealing unexpected insights into the lives of mothers and their families, paving the way for a future where technology and human compassion work hand-in-hand to save lives.

PD: Could you describe the specific technical challenges you faced in developing this AI-powered predictive modelling technology for maternal health? How did Google's pro bono support and resources help overcome these challenges? 

Aparna Hegde: We found that almost 30%-40% of women were not consistently listening to the calls; they weren’t picking up the calls or not listening completely. With limited staff, it was very difficult to reach out to all these women. We needed a method to preempt which of these women were at high risk of dropping out and prioritise them. 

I knew that AI could possibly point us in the right direction. The model developed with pro bono support from Google’s research team helped us retain approximately 30% of the new and expectant mothers at the highest risk of dropping out from the program. 

We have seen that when women listen to the information, the health outcomes are phenomenal. We are able to reach out to more and more women each week, get them back into the fold, and save lives because of AI.

PD: Can you elaborate on the specific types of data used in the AI model and how they are collected ethically and responsibly, ensuring privacy?

Aparna Taneja: Using AI analysis of millions of anonymised call records, the program helps ARMMAN identify participants most at risk of disengaging from the program, prioritising them for additional personalised outreach by ARMMAN’s call centre staff and community partners, including for in-person support and resources for specific concerns.

The model has been developed in adherence with Responsible AI principles and underwent rigorous review by several ethics boards.

PD: Can you quantify the impact of this AI technology on maternal health outcomes so far? Are there any specific metrics you track, like reduction in dropout rates, improvement in antenatal care adherence, etc.?

Aparna Hegde: At ARMMAN, we had been offering our free mobile voice call service for over a decade, bringing timely and targeted preventive health information to expectant and new mothers in underserved communities in languages and at time slots of their choice. We saw that regular use of the service was associated with improved behavioural and health outcomes for women and their babies, with 17% more infants reaching healthy birth weights within their first year, 25% more pregnant women taking vital iron and folic acid (IFA) tablets, and a nearly 48% increase in women’s awareness about family planning methods.

By integrating the AI model developed with pro bono support from Google’s research team, we were able to retain approximately 30% of the new and expectant mothers at the highest risk of dropping out from the program. This meant more women were receiving and acting on healthcare information they received through our program, translating into better health outcomes for them and their babies. 

Initially piloted with 175,000 women in 2019, the AI powering this program was eventually scaled across the database of approximately 350,000 women who had used the service.

PD: How does this technology complement and scale the impact of existing maternal health programs in India?

Aparna Hegde: ARMMAN is now also implementing the Kilkari program, a mobile health information program in collaboration with the Government of India that has reached over 49 million women and their children across 20 Indian states and Union Territories. Google’s research team is providing pro bono support to help ARMMAN develop AI models to help improve Kilkari participation. 

What type of machine learning algorithms are used in predictive modelling? Why were these specific approaches chosen?

Aparna Taneja: The Machine Learning solution was modelled on a resource optimisation framework called Restless Multi-Armed Bandits (RMABs), which is ideal for optimising resource allocation under budget constraints. Our studies discovered that RMABs demonstrated statistically significant superior performance when compared with other approaches, while also allowing for computational efficiencies.

The model analysed a participant’s behaviour and responsiveness on the automated, anonymised calls over a period of time, and provided statistically significant predictions about their continued engagement, allowing ARMMAN’s call centre staff and community partners to prioritise participants most at risk of disengaging from the program for additional personalised outreach, including in-person support and resources for specific concerns.

This prioritisation helped ARMMAN optimise its resources to maximise engagement with the expectant and new mothers. 

PD: Are there any interesting anecdotal stories about mothers whose lives have been positively impacted by this technology?

Aparna Hegde: Attesting to the effectiveness of the AI-assisted support, a program participant has said, “I was unable to listen to the calls earlier. Then the ARMMAN worker reached out and explained the benefits of listening to the messages. Now I listen to the calls regularly. It feels like someone from your own family is looking after you. I follow all the advice and take good care of my baby.”

We’ve also noticed that women’s heightened awareness about their health has facilitated important social shifts across communities, with mothers becoming more vocal about better rights and care for their girl children, while also receiving more support from their families for their children and for themselves.

PD: What are some of the unexpected challenges and successes you've encountered while working on this project? 

Aparna Taneja: The Machine Learning model evolved during the development lifecycle as the research team worked with ARMMAN, initially only identifying women at risk of disengaging from ARMMAN’s program, to eventually ranking them by those who would potentially benefit the most from additional intervention. This was a significant conceptual shift that helped ARMMAN optimise resources.

The model also accounts for the dynamic nature of human behaviour and approaches that may be counterproductive, discouraging repeated calls if a woman’s listening behaviour remains unchanged. This additional consideration makes the model more aligned with real-world human behaviour, ultimately leading to a more effective and respectful intervention strategy.

Furthermore, an unintended design outcome was discovered by ARMMAN workers, who noticed that the AI model was also somehow identifying mothers who may be experiencing difficulties during their pregnancies, and was able to prioritise them for additional support from community workers.

PD: How do you see AI playing a broader role in improving access to healthcare and addressing social issues in India and beyond?

Aparna Hegde: We have seen the effectiveness of the tech + touch approach firsthand, with not just phenomenal health outcomes following the integration of AI models, but also intergenerational transformations that are strengthening families and communities alike. Our quest is to adopt multimedia approaches, and given the massive amounts of data we have, use the power of AI and predictive analytics to better serve even more mothers and children.

Aparna Taneja: The possibilities for AI-assisted interventions are immense. Just the one framework we used to address participant retention in ARMMAN’s healthcare information program potentially has multiple applications - in the healthcare sector, but also beyond in sectors like agriculture and education.

At Google, we are deeply committed to the belief that AI can be a powerful force for not just health, but also social equity. While this has been the driving force for our AI for Social Good team as we provided pro bono support to ARMMAN to develop the AI model, other teams across the company are developing high-impact solutions to global challenges in close collaboration with changemakers, entrepreneurs and stakeholders from across the world, bringing our technologies, capabilities and resources to support and enable impact at scale.