86.4% Preventable: The Six Reasons Women Are Still Dying from Infection in Childbirth
Maternal Mortality Review Committees identified the contributing factors in exquisite detail. What they found is not a mystery. It is a checklist of failures, an accounting of institutional inaction, and a roadmap for what AI can do that human systems have not.
A woman dies from a pregnancy-related infection in the United States. Her family grieves. A report is filed. A committee meets. The word “preventable” appears in the documentation.
And next year, another woman dies.
A study published in Obstetrics and Gynecology analyzed data from Maternal Mortality Review Committees in 29 states covering pregnancy-related deaths from infection between 2017 and 2019. Of the 88 deaths fully assessed, 86.4% were determined to be preventable. Not possibly preventable. Preventable with existing knowledge and available interventions.(1)
The review catalogued 27 contributing factor classes.
The five most common accounted for 56.2% of all factors identified: clinical skill and quality of care (18.6%), delays (10.1%), knowledge (10.1%), lack of continuity of care (9.6%), and lack of access or financial resources (7.8%).(1) This post addresses each one directly, including where hospitals and ACOG have failed, and where AI can do what human systems have not.
1. Clinical Skill and Quality of Care (18.6%)
This was the single most frequently cited contributor. Nearly 1 in 5 contributing factors pointed to deficiencies in clinical skill or the quality of care delivered. In the context of infection, this means failure to recognize sepsis early, failure to apply evidence-based obstetric sepsis protocols, and failure to escalate care when early warning signs appeared. Maternal sepsis is treatable. Bundle-based care, including early antibiotics, fluid resuscitation, and source control, reduces mortality. The problem is not the absence of a protocol. The problem is inconsistent application of protocols that already exist.
Where hospitals and ACOG have failed:
Hospitals have known for years that sepsis bundles save lives. The Surviving Sepsis Campaign published obstetric adaptations over a decade ago. Yet implementation remains uneven, driven by no mandatory accreditation standard requiring obstetric-specific sepsis protocols. ACOG has issued guidance documents but has stopped well short of mandating protocol adoption or linking quality metrics to credentialing. A guideline that a hospital can ignore without consequence is not a standard of care. It is a suggestion. When 18.6% of maternal infection deaths trace back to clinical skill and quality failures, suggestions are not enough.
AI solution:
AI-powered clinical decision support tools can monitor patient vitals in real time and flag sepsis criteria before a human clinician recognizes the pattern. Systems trained on obstetric-specific parameters, including the modified early obstetric warning score, can alert nurses and physicians when a postpartum patient meets threshold criteria, regardless of how busy the unit is or how experienced the covering clinician. An AI that never gets tired and never overlooks a trending heart rate is not a replacement for clinical judgment. It is a backstop against the lapses that review committees keep documenting. This technology exists today. The barrier is deployment, not invention.
2. Delays (10.1%)
The “three delays” framework has been used in global maternal health for decades: delay in seeking care, delay in reaching care, delay in receiving care once at the facility. In the United States, the third delay, receiving care once inside the hospital, remains a documented killer. The Joseph et al. analysis noted that delays in antibiotic administration were common across reviewed deaths.(1) Antibiotics for sepsis are time-sensitive. Every hour of delay increases mortality. Delays happen for predictable and addressable reasons: understaffed emergency departments, triage systems that do not flag postpartum patients appropriately, and clinicians who do not recognize obstetric warning signs outside a labor and delivery setting.
Where hospitals and ACOG have failed:
Most hospital triage protocols were not designed with postpartum patients in mind. A woman presenting to an emergency department nine days after delivery is often triaged the same way as any adult with a fever, without the obstetric-specific escalation her condition demands. ACOG has not established a national standard requiring hospitals to implement postpartum-specific triage pathways in emergency departments. Without that standard, individual hospitals set their own protocols or none at all. The result is that the speed of care a postpartum woman receives in an emergency room depends largely on whether the triage nurse happens to know that the obstetric rules are different.
AI solution:
AI triage tools can identify postpartum status from registration data and immediately flag the patient for obstetric-specific assessment criteria. A patient who enters her recent delivery date at check-in should trigger an automated alert that modifies her triage pathway before she is ever seen by a nurse. Natural language processing tools can also scan electronic health record notes in real time, identifying language that suggests a clinician is not applying urgency appropriate to the presentation, and prompt a secondary review. Time-to-antibiotic is a measurable outcome. AI systems can track it continuously across an entire hospital system and surface outliers before the next death, not after.
3. Knowledge (10.1%)
Knowledge gaps contributed to roughly 1 in 10 identified factors. Some gaps were on the patient side: a woman who does not know that fever and rapid heart rate three days after delivery are warning signs cannot seek care she does not know she needs. But knowledge gaps also exist on the clinician side. The MMRCs found that clinicians in emergency department and outpatient settings need education on modified obstetric early warning signs.(1) A clinician who sees a postpartum patient in an urgent care clinic and does not apply obstetric-specific criteria is a knowledge gap at the system level.
Where hospitals and ACOG have failed:
ACOG publishes educational materials. It runs continuing medical education programs. But there is no mechanism ensuring that the emergency physician who sees a postpartum woman at day 10 has ever received training on obstetric sepsis recognition. Cross-departmental education, reaching emergency medicine, internal medicine, and urgent care, is not systematically required. Discharge instructions given to postpartum women vary by hospital, by clinician, and by shift. There is no federal or professional society standard mandating that every woman discharged after delivery receives written, plain-language information about infection warning signs and a clear instruction about when and where to seek care. That is a knowledge delivery failure at the institutional level.
AI solution:
AI can close the knowledge gap on both sides simultaneously. For patients, AI-powered postpartum apps can deliver daily check-ins after discharge, ask about symptoms, and trigger escalation pathways when warning signs are reported. The technology to build this exists and is not expensive. For clinicians, AI decision support tools can embed obstetric-specific guidance into the emergency medicine and urgent care workflow, surfacing postpartum sepsis criteria automatically when a patient’s recent delivery is noted in the chart. An AI that knows what a human clinician was never formally taught is not a workaround. It is what the system should have provided in the first place.
[PAYWALL BREAK]
4. Lack of Continuity of Care (9.6%)
Nearly 1 in 10 contributing factors involved a breakdown in continuity. A woman delivers at a hospital. She is discharged at 48 to 72 hours. Her next scheduled appointment is at six weeks. In between, she is alone with a newborn, healing from either a vaginal delivery or major abdominal surgery, navigating a body undergoing the most rapid physiologic change of her adult life. If she develops an infection at day five, there is no standard system designed to catch her. She may call her obstetrician’s office and be told to wait and see. She may go to an urgent care that cannot recognize postpartum sepsis. The failure is not any individual clinician’s wrong decision. The failure is that no connected system was designed to prevent that situation.
Where hospitals and ACOG have failed:
The six-week postpartum visit was the standard of care for generations despite no evidence that six weeks is the right interval for detecting postpartum complications. ACOG updated its guidance in 2018 to recommend earlier contact, ideally within three weeks, and a comprehensive visit by 12 weeks. That update was an improvement. It was not a solution. “Contact” can mean a phone call. It does not ensure that a woman with early infection signs is identified or that her care is coordinated across the obstetric, emergency, and primary care systems she may touch in the weeks after delivery. Hospitals have not built the infrastructure this coordination requires, and ACOG has not made that infrastructure a condition of quality.
AI solution:
AI-powered remote monitoring platforms can maintain active contact with postpartum women from the day of discharge through the first six weeks. Automated symptom check-ins by text, app, or voice can identify warning signs daily rather than waiting for a scheduled visit. When a woman reports fever, increasing pain, or malaise, an AI system can escalate immediately to a human clinician for follow-up, routing her to the appropriate level of care within minutes rather than days. This is not futuristic. Postpartum remote monitoring programs using these tools have already been piloted. The evidence supports their utility. What is missing is the institutional commitment to deploy them at scale and the professional society standard that would make them the expectation rather than the exception.
5. Lack of Access or Financial Resources (7.8%)
Access and financial barriers were cited in nearly 1 in 12 contributing factors. This is likely an undercount. Women who cannot access care are not always captured in committee reviews the same way deaths inside the health system are. But the deeper story here is not just about insurance gaps. It is about what happens when Medicaid pays obstetricians and hospitals at rates that are 30 to 50 percent below commercial insurance rates. Physicians and hospitals respond to that gap the same way any business responds to below-cost reimbursement: they limit exposure. The result is that the 40% or more of pregnant women in the United States who are on Medicaid, disproportionately Black, Hispanic, and rural women, face a two-tiered system in which their coverage exists on paper and their access exists in theory.
Where hospitals and ACOG have failed:
ACOG has advocated for Medicaid expansion and for extending postpartum coverage to 12 months. That advocacy is on the record and is appropriate.
What ACOG has not done is confront the reimbursement discrimination that makes Medicaid a second-class card in a first-class system. When obstetricians limit the number of Medicaid patients they will see, when hospitals close obstetric units in low-income communities because the margin does not pencil out, the professional society responsible for the standard of care has an obligation to say that this is not acceptable. It has not said that with the force the situation demands.
Universal healthcare coverage, or at minimum Medicaid reimbursement parity with commercial rates, is not a political position. It is a precondition for the word “preventable” to mean anything at all.
AI solution:
AI cannot fix a reimbursement structure that devalues the lives of low-income women. But it can reduce the cost of delivering care to underserved populations enough to change the math. AI-powered telehealth platforms dramatically lower the overhead of postpartum follow-up visits, making it economically feasible to see Medicaid patients at higher frequency without the facility costs of in-person care. AI tools that automate Medicaid enrollment verification, coverage extension applications, and connection to community health workers can reduce the administrative burden that drives clinicians away from Medicaid panels. The goal is to make caring for Medicaid patients less costly to deliver until the payment system catches up to what equity requires. That is a bridge strategy, not a solution. The solution is parity.
My Take
These five contributors, and the data behind them, are not a surprise to anyone who has followed maternal mortality review committee findings over the past decade. What they reveal is not a mystery. They reveal a system that has identified its failures in precise detail and has not restructured itself to fix them.
86.4% of pregnancy-related infection deaths are preventable. That figure should be on the wall of every labor and delivery unit in the country. It should drive quarterly quality reviews, accreditation standards, and professional accountability. It should inform how we train emergency physicians, how we design discharge instructions, how we structure postpartum follow-up, and how we deploy technology.
The AI solutions I have described are not hypothetical. The tools exist. Remote monitoring platforms, AI-powered triage alerts, clinical decision support for sepsis recognition, discharge risk screening, postpartum symptom apps: each of these is deployable today. What they require is the institutional will to deploy them and the professional society leadership to make them the standard rather than the experiment.
ACOG issues guidelines. Hospitals implement selectively. Women die preventably. That cycle has continued long enough. AI will not replace clinical skill, eliminate financial barriers by itself, or fix a broken triage system through algorithm alone. But AI can close the gaps that human systems, through inertia, underfunding, and inconsistent accountability, have refused to close themselves.
Patients are not passive in this. Ask your hospital what postpartum monitoring they provide between discharge and your first appointment. Ask whether your discharge instructions include specific infection warning signs. Ask whether your coverage continues after delivery. The answers will tell you more about your safety than any brochure.
Bottom Line
Eight in ten pregnancy-related infection deaths are preventable. Review committees in 29 states documented why they keep happening: clinical skill failures, treatment delays, knowledge gaps, disconnected postpartum care, and financial barriers that block access before infection can be treated. Hospitals have known these failure points for years. ACOG has documented them and stopped short of mandating the fixes. AI-powered tools, deployed at discharge, in triage, in remote monitoring, and in clinical decision support, can do what the current system has failed to do. The technology is ready. The question is whether the institutions responsible for women’s lives are ready to use it.
References
1. Joseph NT, Trost SL, Hollier LM, Perkins KM, Goodman DA, Leonard M, Busacker A. Pregnancy-related mortality due to infection: maternal mortality review committees in 29 U.S. states, 2017-2019. Obstet Gynecol. 2026. doi:10.1097/AOG.0000000000006172
2. Hoyert DL. Maternal mortality rates in the United States, 2024. NCHS Health E-Stats. 2026. Available from: https://www.cdc.gov/nchs/data/hestat/hestat113.htm
3. Hoyert DL. Maternal mortality rates in the United States, 2023. NCHS Health E-Stats. 2025. doi:10.15620/cdc/170564
Amos Grunebaum, MD | ObGyn Intelligence - Evidence Matters | obmd.com