Luminare

The mechanism as described is a structured SIRS screening checklist combined with the nurse's own assessment, driving the hospital's existing sepsis protocol. That is rule based screening plus workflow automation rather than machine learning, and the grade describes the mechanism rather than the quality of the product.

The mechanism is the point here rather than a shortfall. The company argues explicitly that machine learning prediction has not improved sepsis outcomes because staff fail to intervene, and it built a response layer instead of a better model. Held at C rather than lower because the product does process record data in real time against criteria and is squarely doing this category's job.

One inconsistency recorded: marketing carries some AI framing, including a claim of commitment to saving lives with AI, that the described mechanism does not support. Ask which component, if any, is learned.

A structural property worth naming, and it is different from every other vendor in this category. The human is an input, not an overseer. Every competitor computes a risk score from data and presents it to a clinician who may or may not read it. Luminare requires the nurse's assessment as part of the screen, so the system cannot produce an output that a clinician passively ignores. That inverts the automation bias problem this category otherwise struggles with, and a nurse quoted in the company's own materials makes the point that it does not override her judgment. Nothing is auto closed, auto escalated or suppressed, and there is no low risk output creating the invisible failure mode seen elsewhere in this lane.

The honest counterweight belongs on the record too. This design adds work rather than removing it, its value depends entirely on nurses completing screens every shift, and a checklist completed under time pressure can be completed carelessly. That is why the company's headline metric is compliance.

Held below A because no false negative behaviour, threshold or override analysis is published.

You can tell exactly what this product does, which places it at the transparent end of a category whose anchor record scores D on this axis. The screening basis is a named, published clinical standard in SIRS criteria, the second input is the nurse's documented assessment, and the action taken is the hospital's own sepsis protocol rendered in SBAR format.

A clinician can trace any output back to a criterion they already know and to their own institution's protocol, and can disagree with either. Same reasoning that carried Elaborate and VieCure to B on rules based mechanisms. Held below A because the enhancement to standard SIRS screening is asserted rather than specified, so the one component that differentiates it from a paper checklist is undisclosed, and because no operating characteristics of the screen are published anywhere.

One condition is disclosed publicly that most vendors would leave in a contract, and it is the substance of this record. The published grant programme states plainly that qualifying hospitals receiving free deployments must share data with the company, alongside focus group participation and testimonials.

Naming the price of a free deployment in public is more candid than this market usually manages, and it lets a hospital evaluate the trade before entering a sales process rather than discovering it in a redline. What is absent is any definition of what that means.

No privacy policy, data processing statement, retention period, de identification method or training use disclosure was located, so the scope, duration and purpose of the sharing are unestablished, as is whether the obligation survives the grant period or whether data already contributed can be withdrawn.

A hospital on the grant tier is in a weaker negotiating position than a paying customer by construction, and it is likely to be a smaller or more constrained institution, which is precisely the buyer least able to press for terms. Nothing else is named either: no model, hosting arrangement or sub processor list. Ask exactly which data, at what identifiability, for how long, for what purpose, whether it trains models served to other hospitals, and what happens at the end of the grant.

Precision matters on what the publication actually is. Ferreira LD, McCants D, Velamuri S, Using machine learning for process improvement in sepsis management, Journal of Healthcare Quality Research 2023;38(5):304-311, published online 29 October 2022. It is a review article synthesising PubMed literature from January 2017 to February 2022, not original research and not an evaluation of Luminare's product. The paper states of itself that it does not present novel findings. Conflicts are disclosed: Velamuri is co founder and chief executive of Luminare and a co author holds an intern position there. So it is a well founded, peer reviewed position paper arguing the thesis the company sells against, and it should be credited as an intellectual contribution rather than as product evidence.

Everything else is vendor reported without derivation: sepsis detected 15 times faster than a national average that is never defined, sepsis alert compliance improved more than fivefold over an unstated baseline, and a single site report of 12 percent lower mortality and 27 percent fewer life threatening emergencies in a first year. No denominator, no comparison group, no measurement method and no time period accompanies any of them. No sensitivity, positive predictive value, alert burden or lead time figure is published anywhere.

One disclosure of substance and no governing documentation around it. The published grant programme states plainly that qualifying hospitals receiving free deployments must share data with Luminare, alongside focus group participation and testimonials. Disclosing that condition publicly is more candid than most vendors manage.

But no privacy policy, data processing statement, retention period, de identification method or training use disclosure was located, so what data sharing means in scope and duration is unestablished. A buyer on the grant tier should establish exactly which data, for how long, for what purpose, and whether the arrangement survives the grant period.

No statement on business associate agreements, execution terms, cost or subprocessor disclosure was located. Named hospital deployments including a major academic medical centre mean agreements plainly exist, and nothing about them is published.

The second pass surfaces a structural point the record should carry. The company has run two quite different lines of business, and they sit under different relationships. Sepsis detection inside a hospital is ordinary business associate territory: the hospital is the covered entity and the vendor processes its records. The public health products, screening and vaccination management deployed for city and county authorities, are not the same. A public health department operates under its own statutory authority, state public health reporting law applies alongside the federal privacy rule, and records generated there may be public health records rather than treatment records.

A buyer should establish which relationship applies to them and not assume terms negotiated for one line govern the other.

One practical caution for a company of this size. Public company databases carry inconsistent signals about its corporate status while its own materials and recent programme participation indicate continuing activity. Nothing here suggests otherwise, and it is a reminder that for a small vendor holding a business associate agreement the identity and continuity of the counterparty matters. Establish which legal entity signs, and what happens to data and to the agreement on a change of control.

Ask for the agreement, its scope across both product lines, and the subprocessor list.

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

The public sector exposure the earlier assessment flagged is confirmed and it is the most productive lead on this record, because those environments generate assurance artefacts by requirement rather than by choice. The company deployed screening and vaccination management products for municipal and county public health customers, and is itself based in Texas serving Texas cities. Texas state agencies and institutions operate a state cloud security authorisation programme modelled on the federal one, and any federal work of the kind referenced would carry its own authorisation requirements. Where such an authorisation exists, a documented system security plan and defined boundary exist with it.

So the ask is specific rather than general: which public sector authorisations has the company held, for which products, and can the boundary documentation be shared. A publication gap of that kind is straightforward to close.

One further question follows from the company's own description of itself. The team is distributed, with staff outside the United States including in Europe. That is unremarkable for a software company and it does mean personnel security, access location and offshore support arrangements are live questions for a vendor whose product reads patient records inside hospitals. Establish who can reach a customer environment and from where.

The company is small and early stage, and this grade records what a counterparty can verify rather than a judgement that controls are absent.

Not FDA cleared, and the company has said it operates without clearance on the basis that it makes no claims about lives saved and issues no treatment recommendations, which is in substance a clinical decision support exclusion argument.

Two things are true and both belong on the record. First, its exclusion argument is the strongest in this category, and for exactly the reason the Epic Sepsis Model's is the weakest. The exclusion turns on whether a clinician can independently review the basis for the output, and here the basis is a published screening standard plus the hospital's own protocol, both fully reviewable. Second, there is visible tension between the stated basis and the marketing, since the product's advertised value is providing clear actionable intervention steps, and the site does publish mortality and outcome claims.

Stated as a question for the vendor rather than as a legal conclusion, which is how this index handles the Cures Act throughout.

No subgroup performance, calibration or fairness analysis of any kind is published.

The specific governance exposure here is not algorithmic, it is an incentive structure, and the company states it openly enough to assess. Luminare markets a documented financial case built on increased sepsis capture: a 400 bed hospital coding 80 to 100 sepsis patients monthly is described as likely missing another 20, at 10,000 to 15,000 dollars average reimbursement, framed as up to 3.6 million dollars a year, alongside a clinical documentation improvement programme said to lift case mix index by around 10 percent.

Identifying genuinely missed sepsis is the clinical goal, and being paid appropriately for care delivered is legitimate, so this is recorded as a structure rather than an accusation. But a system that prompts staff to screen for and document sepsis, sold on the revenue from documenting more sepsis, points its clinical and commercial incentives in the same direction with no independent check. OmniMD and Droxi Unlock carry the same tight loop. Sepsis coding is already an area of established payer scrutiny, which sharpens rather than softens the question.

A clinician can trace every visible part of this product back to something they can already argue with, which is the basis for the grade. The screening basis is a named published criteria set rather than a proprietary score, the second input is the nurse's own documented assessment, and the action is the hospital's own sepsis protocol rendered in a standard clinical handoff format.

So an output points at a criterion the clinician knows, an assessment they wrote, and a protocol their institution authored, and any of the three can be disputed on its own terms. Against the closed sepsis model graded elsewhere in this index, that is the opposite posture. Held at C for the same reason recorded on another product in this lane: the published parts are the borrowed parts.

The enhancement to the standard screen is asserted rather than specified, so the one component that differentiates this from a paper checklist is the one nobody can inspect, and if the enhancement is what suppresses or promotes an alert then it is also the component that decides what a clinician sees. No operating characteristics of the screen are published anywhere, and no warranty, indemnity or remediation commitment was located. Ask what the enhancement does, sensitivity and specificity against the unenhanced criteria, and what proportion of alerts it changes.

Named partners rather than a generic integration claim: Microsoft, Cerner, CPSI and the American Heart Association Get With The Guidelines registry are all listed as interoperability partners, and the company presented at the HIMSS Epic Interoperability Showcase in 2021. Bedside data is written directly back into the EMR rather than held in a separate system, which is a genuine two way integration and materially more than most of this lane offers.

CPSI matters more than its profile suggests here, because it serves community and rural hospitals, which is the segment this product is aimed at. Held at B because Epic presence is showcase participation rather than a marketplace listing or partner review, no FHIR or SMART on FHIR capability is described, and the standalone tier explicitly requires no integration at all, so integration depth varies by tier in a way the pricing structure makes clear but the integration claims do not.

Publishes a genuine implementation ladder with timelines attached, which almost nothing else in this category does: a standalone product needing no integration and offered on a 30 day trial, an Enterprise Lite tier deployable in a few weeks, and a full enterprise platform implemented over months.

For a small hospital, the standalone no integration option is the most practically useful thing on this record. Every other vendor in this lane requires a real time data feed before anything works at all, which excludes facilities without the IT capacity to build one.

Named deployments include an enterprise agreement with Cedars-Sinai from 2023, two CHG hospitals and Cornerstone Specialty Hospitals Clear Lake. A separate onsite consulting workshop offering is described in three stages. Held below A because no hosting architecture, data residency commitment or technical prerequisite detail is published.

The best commercial disclosure in this category, in a lane where every other vendor publishes nothing at all. Three tiers are named with their contract structure and, crucially, the unit of pricing is stated: a 30 day trial on the standalone product, a flat annual fee for Enterprise Lite, and a per patient day rate on an annual contract for the enterprise platform. Naming the pricing unit lets a hospital model its own cost envelope before any conversation, which is the property that earns credit here.

It also publishes a grant programme offering three, six or twelve month deployments at no financial cost to qualifying smaller and acute care facilities, with the conditions stated: screening compliance thresholds, best practice implementation, data sharing, focus groups, surveys and testimonials.

Held at B rather than A because no actual figures or bands accompany any tier, so the ladder gives structure without magnitude.

Deliberately narrow: one condition, sepsis, in adult inpatient acute care. No paediatric or neonatal indication is stated, no ICU specific configuration is described, and no other clinical condition is covered. Narrowness is a strategic choice here rather than a gap, and specialisation is arguably why the workflow is as specific as it is, but this axis measures coverage and the coverage is single condition.

Recorded separately for scoping clarity: the wider company also sells products unrelated to this record, including a vaccination management platform, an emergency response system, a syndromic screening tool and a record abstraction product named Leonardo, with public sector customers including state health departments, FEMA and Department of Defense. This record is scoped to the sepsis product only, per this index's convention of indexing by AI product rather than by company.