VirtuSense VSTOne

The inference is the product, and the distinction from the other sensor vendor in this category is instructive. Biobeat's sensor produces cleared physiological measurements that stand on their own with or without a model. A LiDAR sensor in a hospital room produces nothing a clinician can use until a model interprets it: the value is entirely in inferring intent to stand from subtle postural change, several seconds before the movement itself.

Apply this index's standard test and the answer is unambiguous, since removing the model leaves a depth sensor and no product. Trained on a stated corpus exceeding 1.5 million hours across the product family.

Described as fully automated and autonomously monitoring, but the action taken is always to notify a human. An alert reaches the assigned nurse and the floor console within a second, and a contingency protocol escalates to the wider team if nobody responds. That escalation path is a genuine oversight feature rather than a marketing line, because it addresses what happens when an alert is missed, which most vendors here leave unstated. Sensitivity is configurable by patient risk level and by time of day, with distinct prevention and detection modes, so the institution rather than the vendor sets the posture.

The suppression question applies here more directly than anywhere else in this category, because suppression is the headline claim. A 98 percent reduction in alarms means the system is deciding, tens of times per bed per day, not to notify anyone. The same pattern runs through Affineon auto handling normal labs, CLEW labelling patients low risk and AlertWatch filtering alerts. Ask what proportion of suppressed events preceded an actual fall.

Mechanism is described with unusual specificity for this market. The sensing stack is enumerated as LiDAR with infrared imaging, microphones, real time location sensing and an optical pan tilt zoom element. Computation is stated to run locally at the bedside rather than in a remote cloud. The inference target is named precisely as postural shifts preceding volitional movement rather than the movement itself, and the training corpus is quantified at over 1.5 million hours.

An unresolved contradiction sits in the same material, and it matters because it underpins a privacy claim. The company markets monitoring without cameras and states its sensors do not capture faces or sounds, while the VSTOne product description lists an infrared camera, microphones and an optical pan tilt zoom camera in the sensor stack. These may be reconcilable, since depth and infrared sensing are not conventional video and some components may be optional, but the vendor does not reconcile them.

Held at B also because no model class, validation methodology or accuracy derivation is published.

Local computation bounds this axis structurally and the same unreconciled claim complicates it. Inference runs at the bedside rather than streaming to external servers, which removes the largest exposure a room monitoring product would otherwise create and means what leaves the unit is derived signal rather than raw observation, at least for the components that run locally.

The sensing stack is also enumerated rather than described generically, covering depth sensing with infrared imaging, microphones, real time location sensing and an optical pan tilt zoom element, so a buyer can see what hardware is in the room. What is unresolved is which of those are active, because the marketing states the sensors do not capture faces or sounds while the component list includes a camera and microphones, and nothing reconciles the two.

That determines what exists to be transmitted or retained at all. On enumeration there is nothing beyond the hardware: no model or provider named, no hosting arrangement for whatever tier exists beyond the unit, no sub processor list, no retention statement, and no provenance disclosure for a corpus of more than one and a half million hours of patient room observation. Ask which components are active by default, what leaves the device, where the corpus came from and on what consent, and for a retention schedule.

This vendor publishes the right metrics and none of their derivations, which is a precise finding rather than a generic complaint.

It is the only vendor in this category that publishes all three of the measures this category exists to ask for: a lead time, stated as 31 to 65 seconds before bed exit; an alert burden, stated as 0.5 false alerts per day for the sibling product; and an accuracy figure, stated as 98 percent for VSTAlert and 99.99 percent for VSTOne. The shape of the disclosure is exactly what this index asks of every vendor here.

The substance is unverifiable. No denominator accompanies the false alert figure, no definition explains what accuracy is measuring or over what population, the two accuracy numbers differ across products with no methodology for either, and 99.99 percent on a rare event is close to meaningless without a breakdown by error type. The comparative claims about conventional bed pads, at as low as 15 percent accuracy and 20 to 30 false alarms per bed per day, are asserted without a source. The headline outcome, an average 74 percent reduction in falls with injury, has no comparison group, time period or denominator. No peer reviewed evaluation was located.

Publishing a number invites the methodology question, which is the same paradox recorded against Pieces.

The best privacy position among the sensor vendors in this category, and it is architectural rather than promissory. Computation runs locally at the bedside rather than streaming video to external servers, which removes the largest exposure surface a room monitoring product would otherwise create.

The company states its sensors detect movement without capturing faces or sounds and markets the approach as dignified monitoring for patients who would refuse a camera in the room, which is a real clinical consideration rather than a slogan. HIPAA compliance is asserted.

Held at B rather than A for two reasons: the sensor stack described for VSTOne includes an infrared camera, microphones and an optical pan tilt zoom element, which sits awkwardly against the no cameras and no sounds claim and is nowhere reconciled, and no privacy policy, retention statement or training data provenance disclosure was located for a system trained on over 1.5 million hours of patient room observation.

HIPAA compliance is asserted repeatedly and prominently, described as 100 percent compliant, but no business associate agreement posture is published: no statement of whether a BAA is executed as standard, on what terms, at what cost, or with what subprocessor disclosure. Graded C rather than Not Rated because a claim is made and can be assessed, and a bare compliance assertion without BAA terms is the weak rung of this index's HIPAA ladder. The local processing architecture would materially shape any such agreement, which makes the absence of published terms more noticeable rather than less.

A second pass again located no attestation, trust centre or report request path.

The architecture genuinely reduces the surface and deserves crediting. Processing runs on the device at the bedside, the primary sensor produces a three dimensional point cloud rather than a photorealistic image, and the vendor states that monitoring data does not leave the room. For fall and pressure injury monitoring that is a materially better privacy posture than a camera based system, not merely a different data handling policy.

The same device carries other paths, and the marketing claim does not always travel with the qualification. The sensor stack is described as including an optical camera with pan, tilt and zoom control, and microphones, alongside the depth sensor. The vendor states the camera feed is used only for telehealth and virtual nursing, and those functions necessarily transmit video and audio to a remote clinician. So the statement that no patient data leaves the room is true of one function and cannot be true of the other, while appearing in material a buyer would read as describing the device.

Two claims should also be corrected rather than repeated: true compliance and one hundred per cent compliance with the privacy rule are not properties a product can hold, and no body certifies either.

One fact points at what to ask for. A variant built for veterans' hospitals is described as designed to run off network. That implies the standard product is on network, and that an off network build is achievable. Ask whether it is offered generally, and what the management console and record system integration transmit.

No FDA clearance was located for VSTOne or for the product family, and none is claimed. That is not necessarily a gap, since a system that alerts staff to imminent bed exit may fall outside device regulation depending on how its outputs are framed, and the company does not present the product as diagnosing or predicting a clinical condition.

But the position is entirely undocumented: nothing published states the regulatory basis on which the product operates, and the accuracy and outcome claims made in marketing are of a kind that would ordinarily attract scrutiny if the product were regulated. Establish the regulatory rationale directly, particularly for the pressure injury use case, which sits closer to a clinical assessment than bed exit alerting does.

No subgroup performance, calibration or fairness analysis is published.

One genuine advantage is worth recording, because it distinguishes this vendor from the other sensor products in this category. Depth based LiDAR sensing does not rely on optical absorption through skin, so it does not share the photoplethysmography pigmentation mechanism that affects Biobeat across every parameter and Etiometry through its oxygenation input. That is a real structural difference rather than a claim the vendor makes.

The exposures here are different in kind and equally unexamined. An inference trained on movement patterns may perform differently for patients with contractures, tremor, limb amputation, high body mass, mobility aids in the bed, unusual bedding, or atypical room layouts, and nothing published reports performance across any of these. Given the product is trained on over 1.5 million hours of observation, the composition of that corpus is the question, and it is not described.

Two passes located no operating characteristics, no validation methodology, no accuracy derivation and no warranty, indemnity or remediation commitment, and a contradiction sits in the vendor's own material on the exact point a patient would care about. The company markets monitoring without cameras and states its sensors do not capture faces or sounds, while the product description lists an infrared camera, microphones and an optical pan tilt zoom element in the sensor stack.

These may be reconcilable, since depth and infrared sensing are not conventional video and some components may be optional or disabled by default, and the vendor does not reconcile them anywhere. A privacy claim contradicted by the same company's own component list is worse than silence, because a patient or family told there is no camera in the room has been told something the product documentation does not support, and they have no way to discover it.

Genuine disclosure exists alongside and should be credited: computation runs locally at the bedside rather than in a remote cloud, the training corpus is quantified at more than one and a half million hours, and the inference target is named with unusual precision as postural shifts preceding volitional movement rather than the movement itself, which tells a clinician the system is predicting intent to rise rather than detecting a fall. Ask which components are present and active in the configuration you are buying, and for detection performance on that configuration.

The strongest integration claim among the sensor vendors in this category, and the only one that names an EHR. VSTOne is described as fully integrated with Epic, with each alert logged immediately both in the platform and written to the electronic health record, alongside onsite performance integrations.

Writing alerts back to the record rather than holding them in a parallel system is the property that matters, because it puts the event in the document a clinician and a quality reviewer will actually look at. Unit and site level reporting on fall trends over time extends that into quality workflow. Held at B because no marketplace listing, Showroom entry or partner review credential was retrieved to corroborate the Epic claim, no other EHR is named, and no FHIR or interface detail is published.

Simple by design and that simplicity is the practical argument. One LiDAR based sensor per room, no wearable on the patient, no dependency on the hospital's existing monitoring estate, and computation local to the bedside. Alerting reaches assigned staff on mobile devices, a floor console and a management kiosk, with room assignment, shift tracking and real time room status handled in the same interface, so the operational wrapper is described rather than assumed.

The company publishes sizing guidance for its long term care product, recommending coverage of 20 to 25 percent of beds rather than all of them, which is unusually candid commercial advice and implies a targeting rather than blanket deployment model. Held at B because no acute hospital customer is named beyond a general reference to several hospital systems, no implementation timeline is published, and no data residency statement exists beyond the local processing claim.

No pricing published at any level: no per sensor, per room, per bed or per unit cost, no platform subscription, no band and no implementation fee. The per room hardware model makes the unit of pricing consequential, since cost scales directly with rooms covered and the company's own sizing guidance implies partial rather than full deployment.

One commercial mechanism is worth recording. VirtuSense partners with an underwriting firm to bring the tools to post acute facilities, mirroring the arrangement in which a professional liability insurer subsidises PeriGen's maternal safety product. An insurer participating in distribution is a meaningful third party signal about expected loss reduction, and it is also a route to affordability that a licence fee comparison misses.

This record is scoped to the acute inpatient product, and that scoping is itself the setting statement. VSTOne serves hospital inpatient rooms, while the sibling product VSTAlert serves skilled nursing and assisted and independent living, which fall outside this category's setting boundary, the same boundary that keeps SafelyYou in home care operations rather than here.

Within acute care the coverage is condition agnostic rather than specialty specific, since the system monitors any patient in any monitored room rather than a clinical population, with fall risk and pressure injury named among a stated eight use cases.

Held at B because the remaining use cases are not enumerated anywhere retrieved, no exclusions are published for patients or room types where the sensing does not work, and a product that infers from movement plainly has such limits even though none are stated.