PeriGen

Split product, and the split is real. The FDA cleared algorithms, Patterns for fetal heart rate interpretation and Curve for labor progress, are the differentiation and the source of every substantive claim the company makes. But PeriWatch is also a fetal surveillance, archiving, EMR interfacing and command centre platform, and the company acquired WatchChild, a conventional fetal monitoring system, which is a substantial business that does not depend on the algorithms at all.

Strip the AI and a hospital still has a working central surveillance system. Not lower than B because the algorithms carry the regulatory clearances and the independent validation, and because they are what a buyer is choosing this vendor for.

Adjunctive by explicit stated policy, with the company publishing a position that AI is meant to complement clinical expertise and never replace it, alongside a stated approach to responsible use of AI in obstetrics. Nothing is auto actioned and there is no de escalation or low risk output, so this record avoids the invisible failure surface that CLEW, Etiometry and Healthplus.ai all carry.

The institution sets the trigger, which is the property this index rewards. Alerts fire when values cross hospital defined safety limits rather than vendor set ones, and escalation paths to designated hospital leaders are configured by the organisation.

Held below A because no false negative rate, abstention behaviour or per pattern miss rate is published, and because the escalation feature raises a governance question the vendor does not address about what an automatically escalated alert means for the clinician who did not act on it first.

Names its own mechanism precisely rather than gesturing at AI, which is more than most of this market manages. The company states that Patterns uses supervised machine learning to interpret fetal heart rate patterns and detect uterine contractions, describes what Curve assesses in labor progression including contraction frequency and epidural use, and third party analysis has reported the use of Google TensorFlow in developing some algorithms.

The hybrid architecture is also disclosed and it matters. The signal interpretation is learned, while the alerting thresholds are hospital defined, so the institution owns the decision about when a pattern becomes an alert.

Held at B because no model card, feature list, training population description or per pattern operating characteristics are published, and the published validations report agreement rather than sensitivity and specificity by pattern type.

Nothing identifies any party in the chain: no model or model family, no hosting arrangement and no sub processor list was located in two passes, and no privacy policy, retention period or training use disclosure was found. What makes that consequential here is not the usual training question but custody over an unusually long horizon.

One product in the suite is a centralised surveillance and documentation system whose described function includes capturing fetal tracings, with a collection architecture built to survive network and system downtime, so the vendor's software is the archive rather than a viewer onto someone else's.

Fetal monitoring tracings are the primary evidence in birth injury litigation, and the limitation period for a claim brought on behalf of a child typically does not begin until they reach majority, so a hospital may need to produce a legible tracing two decades after the birth. That is a retention obligation measured against a child's lifetime and it sits inside a commercial software relationship that will not last that long.

Establish who holds the archive over that horizon, in what format, whether it remains readable if the vendor relationship ends, and what happens on a platform change. A secondary question sits alongside: the vendor markets consolidating and analysing labour and delivery data across multiple births, and whether that runs inside the hospital's own instance or crosses into vendor held datasets is unstated.

Independent federal validation is rare, and it is credited properly here. Three experts at the Eunice Kennedy Shriver National Institute of Child Health and Human Development independently assessed the software across 100 tracings, agreeing with its analysis in over 97 percent of assessments and concluding that computerised interpretation shows substantial agreement with expert evaluation and can screen in real time when an expert is not continuously watching. This was part of a research collaboration with named academic principal investigators at UT Medical Branch Galveston and UT Houston. Separate peer reviewed work in the American Journal of Obstetrics and Gynecology scored 769 segments across 30 tracings against five expert clinicians using a strict five tier framework and found the computer results not statistically different from the clinicians.

Two precisions cap the grade. The 2013 NICHD validation was published as a conference supplement abstract rather than a full paper, the same distinction that applies to Etiometry's length of stay figure.

More importantly, all of this evidence measures agreement with expert interpretation rather than patient outcomes. No study was located showing that deploying PeriWatch changes maternal or neonatal outcomes, which is what an early warning system exists to do. An outcome study would earn A.

No privacy policy, data processing statement, retention period or training use disclosure was located.

The question the earlier assessment identified is confirmed as the central one, and the second pass shows the vendor is not merely adjacent to it. One product in the suite is a centralised surveillance and documentation system whose described function includes capturing fetal tracings with a data collection architecture built to survive network and system downtime. So the vendor's software is the archive rather than a viewer onto someone else's.

That matters because of how long these records must live. Fetal monitoring tracings are the primary evidence in birth injury litigation, and the limitation period for a claim brought on behalf of a child typically does not begin until they reach majority. A hospital may therefore need to produce a tracing two decades after the birth. Establish who holds that archive over that horizon, in what format, whether it remains readable if the vendor relationship ends, and what happens on a change of platform.

The second pass adds a secondary use question the earlier note did not have. The vendor markets consolidating and analysing labour and delivery data across multiple births to identify modifiable factors for quality improvement. That is aggregate analysis beyond the individual patient, and it is a legitimate and valuable use. Establish whether it runs inside the hospital's own instance on its own data, or whether anything crosses into vendor held datasets or algorithm development.

Ask for the retention schedule, the export format, and the boundary on cross birth analysis.

No public statement on business associate agreements, execution terms or subprocessor disclosure was located. Named deployments across academic, community and public health systems mean agreements plainly exist, and nothing about them is published.

Two features of this deployment make scope worth establishing rather than assuming.

The first is what flows. The platform draws from the record system and from the central fetal monitoring system, and one product retains the tracings themselves. Continuous device output attributable to a patient is protected health information, and an agreement written around record system integration may not describe a monitoring feed or a long lived tracing archive. Establish that all three are covered, and that the agreement addresses what happens to the archive at termination given the retention horizon this data carries.

The second is corporate and international. The company is a subsidiary of a listed group headquartered outside the United States, and its software has been deployed beyond North America, including in sub Saharan Africa through a partnership with a United States children's hospital. The domestic privacy rule does not govern a Malawian deployment; whatever the host country and the partnering institution agreed does. A buyer should establish which entity contracts in each market, and any organisation partnering on an international deployment should be clear which framework applies to the data generated there and whether it flows anywhere else.

Ask for the agreement, its coverage of device data and the tracing archive, and the contracting entity.

A second pass again located no attestation, trust centre or report request path, and the earlier assessment's two framings both hold. Parent group reporting sits at group level and says nothing verifiable about this product's information security. Clearance addresses safety and effectiveness rather than information security.

What the second pass adds is what an examination would need to cover, and it is the same architecture this index has now identified across this lane. The early warning product continuously monitors data from the record system and from the hospital's central fetal monitoring system. That second source sits on the clinical device network, which hospitals segment deliberately because monitoring equipment is hard to patch and directly attached to patients. A vendor drawing from both crosses a boundary the organisation maintains on purpose, and in obstetrics the device feed is continuous and simultaneous across an entire unit.

One disclosure deserves crediting because it is unusually honest. Remote access through the mobile product is described as available from anywhere a clinician can reach the hospital network through connectivity methods the hospital's own technology team has approved. Many vendors would present remote access as a secure capability of their own; this one defers the control to the customer and says so, which is both accurate and the right allocation.

Ask which report is held, what network zones the deployment touches, and how the archive component is secured given its retention horizon.

The deepest and longest regulatory history in this category. Multiple cleared components across years: Patterns for automated fetal heart rate pattern recognition, Curve for labor progress assessment, and PeriWatch Surveillance separately 510(k) cleared. Patterns 3.0 was cleared in February 2025, extending the indicated population from 36 weeks of gestational age down to 32 weeks, which is a genuine scope expansion into preterm high risk pregnancies rather than a relabelling.

The company states these are the only FDA cleared algorithms for fetal heart rate interpretation and labor progress assessment commercially available in the United States, a specific and checkable competitive claim rather than a vague one. Gestational age limits are published per clearance, which is a negative scope statement of the kind this index asks for.

No subgroup performance, calibration or fairness analysis is published, and this is the clinical domain where that gap costs most. US maternal mortality carries among the starkest racial disparities in medicine, and the mechanism documented in the literature is substantially differential recognition and response rather than biology.

The argument cuts both ways here, and both directions belong on the record, which is unusual for this axis. Positively, an automated early warning system that fires on objective, hospital defined thresholds removes discretion from the recognition step, which is precisely the intervention health equity researchers propose for differential recognition, and the company's own materials make a version of this argument in describing the product as addressing human inconsistency in reading fetal strips. Negatively, the fetal heart rate model is supervised and trained toward expert interpretation, so whatever is embedded in that interpretation is inherited, and the published validations measure agreement with experts rather than accuracy against outcomes.

Which effect dominates is an empirical question nobody has answered publicly. Publishing subgroup performance and, better, subgroup outcome data would be the single highest value disclosure available to this vendor. The same ask applies to Birth Model.

The hybrid architecture is disclosed and it places the consequential decision in the right hands, which is the basis for this grade. Signal interpretation is learned, using supervised machine learning to interpret fetal heart rate patterns and detect uterine contractions, while the alerting thresholds are hospital defined, so the institution owns the decision about when a pattern becomes an alert rather than inheriting a vendor's judgement about it.

In a setting where alarm burden directly determines whether staff respond, letting the unit set its own threshold is a meaningful allocation of responsibility and it makes the alerting behaviour auditable by the people accountable for it. Held at C because the published validations report agreement rather than sensitivity and specificity by pattern type, and in fetal monitoring the pattern type is the whole question.

Agreement with expert readers tells a buyer the system reads tracings roughly as clinicians do, including where clinicians disagree with each other, which is well documented in this field. It does not tell them how often a pattern requiring intervention is missed, and the categories of tracing that matter clinically are rare by construction.

No model card, feature list, training population description or per pattern operating characteristics were located, and no warranty, indemnity or remediation commitment. Ask for sensitivity by pattern category, and the false negative rate on the categories that trigger delivery.

Designed to sit alongside legacy infrastructure rather than displace it, drawing continuously from the hospital's existing electronic fetal monitoring system and EMR without requiring replacement of either, and the company describes maintaining interfacing libraries following the WatchChild acquisition. That coexistence design is a real interoperability property in a segment where fetal monitoring systems are long lived capital purchases.

Held at B because no EHR vendor is named anywhere, no marketplace listing or partner certification was located, and no FHIR or SMART on FHIR capability is described. The integration surface here is also unusual and worth a buyer's attention: this product depends on the central fetal monitoring system as much as on the EHR, so integration complexity is a function of the monitoring estate rather than the record system.

Named deployments span an unusually wide range of institution types, which is the strongest thing on this axis: Mount Sinai Medical Center, Avera Health across three named South Dakota hospitals, and NYC Health + Hospitals, which went live at North Central Bronx with a stated rollout to all 11 of its hospitals. The last matters because it is the largest US public health system and serves a population that private academic deployments do not represent.

The company describes hundreds of installations spanning academic medical centres, tertiary centres, community hospitals and critical access facilities. The no replacement design lowers the deployment bar materially, since a hospital keeps its existing fetal monitoring estate.

Held at B because no implementation timeline, resourcing requirement, hosting architecture or data residency commitment is published.

No pricing published at any level: no rate card, no unit of pricing such as per delivery, per bed or per unit, no band, and no implementation fee.

One commercial mechanism is published and it is unusual enough to record, though it is not price transparency. IronHealth, the healthcare professional liability and stop loss unit of Ironshore, entered a preferred partnership under which its insured customers could apply risk management reimbursement dollars toward PeriWatch Vigilance.

A malpractice insurer subsidising a clinical safety tool is a meaningful third party signal, because an insurer's actuaries have judged the expected reduction in obstetric liability loss to exceed the cost. It is also a materially different route to affordability for a hospital than a licence fee. Recorded as of the 2020 announcement, so a buyer should confirm the arrangement is still current.

Single specialty and deliberately so: labor and delivery, covering both antepartum and intrapartum monitoring. Coverage within that specialty is genuinely broad, spanning maternal vital signs, fetal heart rate pattern interpretation and labor progress assessment in one system, where most competitors address only the fetal signal.

The gestational age limit is published and was recently extended, from 36 weeks down to 32 with the Patterns 3.0 clearance. That is both a scope expansion and a clear negative scope statement, since below 32 weeks the algorithms are not indicated.

Institution coverage is unusually wide for this lane, reaching critical access and public safety net hospitals rather than academic centres alone. Held at B rather than A because the specialty is singular and no paediatric, neonatal or general inpatient application exists outside obstetrics.