Research is showing strong evidence of the adverse impacts of climate change on maternal, neonatal, and child health (MNCH) outcomes. Events such as rising temperatures and heat waves have been associated with adverse outcomes such as preterm births, stillbirths, low-birth weights, and other adverse maternal and child outcomes [1–3]. Low-income and low-resource nations, such as The Republic of Tajikistan, are grappling with a lack of adequate resources and effective mitigation strategies against the escalating adverse impacts posed by climate change. The rising incidence of extreme weather events, inadequate healthcare infrastructure, and limited resources necessitate an urgency to address climate-related health risks using innovative approaches [4]. Given this context, investigating the impacts of climate change on the health sector in Tajikistan is urgent.

💡 There is strong evidence associating preterm births, stillbirths, low-birth weights, and other adverse maternal and child outcomes with exposure to various climate events

Palindrome Data, along with Abt Global (formerly Abt Associates), a global implementing partner mandated to improve the quality and services of maternal, newborn, and child health (MNCH) and nutrition through the Healthy Mother Healthy Baby (HMHB) program in Tajikistan, worked on an innovative solution to uncover climate-driven adverse health outcomes and develop prototype digital health tools from machine learning models. This project included investigating the effect of climate on MNCH outcomes to better understand its impact in light of changing climate patterns. We used predictive machine learning insights to develop practical tools that healthcare workers can use to proactively intervene and prevent adverse health outcomes in response to exposure to extreme climate or weather events.

Local climate and weather data coupled with routine clinical health data can uncover patterns of climate-driven adverse MNCH outcomes

The project used routinely collected demographic, clinic, and visit data from the HMHB program and climate/weather data from a governmental meteorological agency in Tajikistan. Health and climate-related outcomes of concern were identified through a systematic literature review, workshops with local stakeholders, science and climate experts, and project sponsors. These efforts yielded a list of outcomes which were then evaluated for their modeling potential. Selected health outcomes showed sufficient data coverage during exploratory data analysis and were chosen as candidates for investigating their relationship with climate. Machine learning models were used to develop predictions for being at risk for an adverse health outcome at the next visit.

Machine learning models are capable of predicting the likelihood of preeclampsia, anaemia, and missed visits among pregnant women at their next visit

According to the World Health Organization (WHO), global incidence rates of preeclampsia range between 2% — 10% [5]. Although there is a significant lack of extensive literature on preeclampsia in Tajikistan, a study found the prevalence of preeclampsia to be 1.6% in pregnant women [6], a slightly lower estimate compared to global incidence rates. Preeclampsia, if unmanaged, leads to maternal and/or foetal morbidity and mortality [5’]. During the modelling phase, we were able to reliably predict whether a patient will be diagnosed with preeclampsia at their next visit. We were also able to identify climate-related drivers of a patient being diagnosed with preeclampsia which included lower minimum temperatures in the past week. which contributed to patient’s being more likely to have a higher risk of being diagnosed with preeclampsia. This finding highlights the predictive capability of machine learning models in identifying the risk of preeclampsia and the significant impact of climate-related factors, such as lower minimum temperatures, on MNCH outcomes in Tajikistan.

Anaemia was identified as another significant health outcome of concern in Tajikistan, especially in pregnant women and children. According to the 2017 Tajikistan Demographic and Health Survey, over 41% of women and, similarly, 42% of children aged 6–59 months are anaemic [8]. These rates are higher compared to global estimates of anaemia which range from 30% — 37% per WHO estimates [7]. Similarly, we were able to reliably predict whether a woman will be diagnosed with anaemia at their next visit in this project. We identified elevated temperatures over the past week, prolonged heat over the past month, and reduced precipitation over a three-month period as climate-driven factors associated with a higher likelihood of an anaemia diagnosis in the next visit. Similarly, this finding adds to the strong evidence of the impact climate and weather-related factors are having on MNCH outcomes in Tajikistan.

Finally, another important outcome studied was missed visits. Attending visits on time is an indicator of adherence to prescribed antenatal care (ANC) policies and engagement in care. Although we weren’t able to see strong climate-related drivers affecting a patient’s visits, prior behavioural patterns were strong factors that can predict whether a woman will attend her next visit as scheduled. Altogether, these findings indicate that short or long-term sustained periods of exposure to climate and weather factors can have a significant impact on MNCH outcomes — particularly preeclampsia and anaemia.

Using Insights from Machine Learning Modelling to Develop Prototype Digital Health Solutions

Leveraging insights from machine learning models, a prototype for a risk-score predicting pre-eclampsia, anaemia, and missed visits was developed. The solution was integrated into the MNCH Commcare application used by the local program team in Tajikistan [Image 1]. The MNCH risk scorecard is a digital health tool designed to evaluate individual patient risk of being diagnosed with an adverse climate-driven health outcome. The integration of the climate risk score to the MNCH commcare app will contribute towards mitigating health impacts of climate change and create a proactive approach to healthcare delivery in the event of climate or environmental challenges.