Our Take
The argument is sound: clinicians already see the data, they lack tools to spot slow patterns across time in under-resourced settings. What's missing is evidence that deployed systems actually reduce maternal mortality or adverse outcomes.
Why it matters
Maternal mortality in the US remains stubbornly high, and the care infrastructure is fraying: L&D units are closing, experienced nurses are leaving, and patients are older with more comorbidities. AI-enabled decision support could address the experience gap in under-staffed and resource-limited settings, but only if the systems are designed to reduce cognitive load rather than add noise.
Do this week
L&D operations leaders: audit your current monitoring stack this month to identify which alerts your team routinely ignores or overrides, then specify to vendors which patterns (fetal heart rate variability trends, not isolated readings) must trigger a single consolidated signal so clinicians can act with confidence.
The maternal care crisis isn't dramatic emergencies—it's slow-building risk
Labor and delivery clinicians are stretched thin and getting thinner. L&D units are closing across the country, experienced nurses are leaving faster than they can be replaced, and the patients themselves are older with more chronic conditions. Meanwhile, the tools meant to support care haven't kept pace with these pressures.
The core problem, according to Matthew Sappern (CEO of Perigen, a maternal health software company), is not that clinicians lack data. Most adverse outcomes in childbirth develop gradually through subtle, cumulative warning signs: fetal heart rate variability that decreases, baselines that creep up, decelerations that lengthen and deepen. Individually, each signal is unremarkable. In aggregate, they signal danger. But humans are poor at tracking these small, evolving patterns over time, especially in under-staffed settings where cognitive load is already high.
The distinction Sappern draws is sharp: the wrong technology adds fragmentation and noise; the right technology provides pattern recognition, context, and clarity so clinicians can act earlier and with confidence.
Consistency in care delivery depends on tools that reduce variability, not amplify it
Three specific dynamics make this urgent. First, the experience gap. Less experienced nurses have fewer opportunities to build the intuition and pattern recognition that comes with repetition. In overnight shifts or low-volume facilities, this makes it harder to catch early warning signs consistently. Decision-support tools that apply the same criteria across all patients can bridge that gap.
Second, equity. AI models trained on intentionally curated data—specifically omitting racial or ethnic factors—could help reduce implicit bias in risk interpretation. Human judgment is shaped by experience and fatigue; algorithmic assessment is not. Consistency in how risk is recognized and escalated across diverse populations could narrow disparities in care outcomes, though this remains a stated opportunity, not a demonstrated result.
Third, scale. Experienced clinical talent is scarce and getting scarcer. Algorithmic triage and virtual care models can stretch that talent further by identifying which patients across multiple sites need attention most urgently, allowing specialist clinicians to focus on highest-risk cases and support bedside teams from a distance.
Sappern's core claim: improving maternal safety depends not on adding more devices or more data, but on building systems that help clinicians recognize patterns earlier, communicate risk clearly, and act with confidence in the moment.
Deploy with clarity or stay out of the delivery suite
The risk of getting this wrong is real. Technology that generates isolated signals or fragmented data forces clinicians to synthesize information across multiple streams while staying hyper-vigilant for incremental changes. In already-strained environments, this increases cognitive load and amplifies care variability instead of reducing it.
The design imperative is stark: tools must provide situational awareness and context, not just more alerts. They must be flexible enough to work in bare-bones environments, not just well-resourced academic centers. And they must fit into care delivery workflows without creating additional documentation burden.
One gap worth noting: the article presents the value proposition but does not cite independent studies showing that deployed maternal health AI systems actually reduce adverse outcomes, maternal mortality, or cost per delivery. The logic is compelling, but practitioners need published evidence before betting operational dollars and clinical workflow redesign on these systems.