Knowing that something feels wrong is not always the same as knowing what to do next.
For a pregnant woman with a headache, abdominal pain or an unfamiliar symptom, the decision can be surprisingly difficult. Is it something that can be watched at home? Should a clinic be contacted? Is the situation urgent enough to justify the cost and difficulty of travelling for care?
In South Africa, those decisions can be complicated further by distance, unemployment and the cost of reaching a healthcare facility.
A new study suggests that providing personalised symptom guidance through a platform many mothers already use could influence what happens next.
Researchers integrated a diagnostic decision-support system into MomConnect, South Africa’s government-led maternal-health platform, and studied 968 pregnant women and mothers who used it.
Before using the symptom checker, only 17% planned to seek professional care. At follow-up, 43% had done so. Among participants for whom researchers could directly compare intentions and subsequent behaviour, roughly one-third moved towards a more urgent level of care.
The result does not prove that an AI symptom checker improves maternal health outcomes.
But it suggests something potentially important: better information may alter the point at which people decide that a symptom deserves medical attention.
A different kind of digital health tool
MomConnect is already designed to connect pregnant women and mothers with health information. The system examined in the study went further.
Instead of sending the same general information to everyone, the diagnostic decision-support system asked users about their symptoms and relevant risk factors. It then provided a probable explanation and recommended an appropriate level of care.
That distinction matters.
A general pregnancy message might tell someone which warning signs to watch for.
A symptom-assessment tool attempts to answer a more immediate question:
Given what I am experiencing now, how urgently should I seek help?
To test whether that advice was medically reasonable, researchers did not rely only on users’ impressions.
An independent physician panel reviewed 184 cases.
The panel judged the system’s advice to be safe in 181 of them — 98.4%. In about 45% of cases, the system was more cautious than the physicians considered strictly necessary. That is an important distinction.
A cautious system may sometimes recommend a higher level of care than ultimately proves necessary. But in maternal health, the opposite mistake, telling someone to remain at home when urgent assessment is needed — can carry much greater consequences.
Women changed what they planned to do
Before using the system, 59.4% of participants intended to manage their symptoms at home, while 17% planned to seek care and another 17.3% were unsure.
Follow-up information was available for 785 participants.
By then, 43% reported seeking medical care and only 1.8% remained unsure about what to do.
Among 596 participants with sufficiently clear information for a direct comparison, 33.9% escalated to a higher-urgency response, while 10.7% moved towards a lower level of care.
In other words, the tool did not simply push everyone towards a clinic.
Some people were reassured. Others decided their symptoms required more attention.
That is closer to what a useful triage system is supposed to accomplish: distinguishing between situations rather than treating every symptom as equally serious.
The medically important cases
Some of the individual outcomes make the behavioural shift easier to understand.
Among participants who eventually sought care and received treatment, the researchers reported that half had not originally intended to seek medical attention.
The study also identified 19 participants diagnosed with hypertensive disorders of pregnancy.
Seven were hospitalised, and all 19 received some form of treatment.
Several had originally planned to manage their symptoms at home or were unsure what action to take.
Hypertensive disorders during pregnancy can require timely medical assessment, making these cases particularly relevant to the study’s central question.
But they also illustrate why the findings need to be interpreted carefully.
The research cannot establish what would have happened to these women had they not used the symptom checker.
We know what they initially intended to do. We know what they later did. We cannot observe the alternative version of events.
Information cannot remove every barrier
The study also exposes an important limit to digital healthcare. Information may help someone recognise that medical attention is needed.
It cannot necessarily make that care accessible.
Half of participants with available income information reported monthly household incomes below R1,600, while 64% were unemployed. Among those interviewed by phone, financial barriers prevented 42% from seeking care at least sometimes.
Lower household income was also associated with less safe subsequent behaviour, even when people’s original intentions were appropriate. That distinction is critical.
A woman can receive correct advice telling her to visit a clinic urgently and still face transport costs, lost income, childcare responsibilities or a long journey.
Clinical information addresses one part of healthcare access. It does not replace the healthcare system around it.
The study has important limitations
The research was prospective and observational, not a randomised controlled trial.
That means researchers observed how behaviour changed after participants used the system, but there was no comparable group of women randomly assigned not to use it.
The study therefore demonstrates an association between use of the tool and changes in care-seeking behaviour rather than proving that the system alone caused those changes.
Much of the follow-up information was also self-reported.
The researchers acknowledge that this creates potential recall and social-desirability bias. Non-English-speaking participants were excluded, and women already registered with MomConnect may have been more engaged with healthcare than the broader population.
Follow-up lasted only one week, limiting conclusions about longer-term outcomes.
There is also a commercial consideration worth making visible.
The study was funded by the Rockefeller Foundation, with funding granted to Reach Digital Health and subgranted to Ada Health. Several authors were current or former Ada Health employees, consultants or option holders. The paper states that the funder did not participate in study design, data collection, analysis, interpretation or the publication decision.
These relationships do not invalidate the research.
They are part of the context readers should have when assessing it.
A tool that helps decide when care is needed
The most interesting possibility raised by this study is not that artificial intelligence could replace doctors.
It is almost the opposite. The system’s value may lie in helping people recognise when they need one.
Digital healthcare is often discussed in terms of diagnosis, automation or replacing expensive human interactions. Maternal healthcare presents a different problem.
Sometimes the first challenge is simply helping someone decide whether a symptom warrants professional attention at all.
For South Africa, where mobile platforms can reach people across enormous geographic and economic differences, integrating carefully validated decision support into an existing public-health service could offer one route towards making that decision easier.
But the study also reveals the boundary technology cannot cross.
A symptom checker can recommend that someone seek care.
Only a functioning health system can ensure that care is actually there when she arrives.
Source Information
Study Title: An AI-based symptom checker integrated into MomConnect maternal healthcare platform: a prospective, observational mixed-methods study
Authors: Marcel Schmude, Stephen Gilbert, Caroline Moor Larsson et al.
Journal: Nature Health
Published: 16 June 2026
DOI: 10.1038/s44360-026-00125-x
