Tammy Liu ties signal ROI to EHR

Tammy Liu’s May 19 posts laid out a narrow standard for sensor-based health products: passive signals such as heart-rate variability and galvanic skin response need to fit directly into existing clinical software and produce a fast, measurable action. In Liu’s framing, the problem is not whether a signal is interesting in a model or a paper, but whether it changes work inside the electronic health record without adding another screen or another queue. She cited deployed environments including ICUs and sleep labs as examples where physiological data can already be acted on and where time savings can be counted. The thread amounts to a commercial test as much as a technical one: founders need to prove return on investment before asking health systems to absorb integration work. ### Why is Liu drawing a line between “interesting signal” and “useful product”? Tammy Liu’s May 19 posts focused on workflow, not sensing hardware. She argued that passive physiological signals only become valuable when they are embedded in the tools clinicians already use and when the resulting action can be completed in less than two minutes, according to her posts on X. The distinction matters because HRV, GSR — also called electrodermal activity — and related signals already have a large research literature behind them. Reviews and recent studies describe those signals as useful for stress detection, autonomic monitoring and other classification tasks, but they also note uneven methods and reporting across the field. (ieeexplore.ieee.org) ### Why do ICUs and sleep labs keep coming up in this argument? ICUs and sleep labs are settings where physiological monitoring is already part of routine care. A Scientific Reports study found HRV measured early in ICU admission was associated with outcomes, using signals captured from standard monitoring systems. Sleep research shows a similar pattern. A medRxiv study examining ICU and sleep-lab patients reported that heart-rate variability and respiratory signals could be used to estimate sleep state and generate automated information from data that can already be measured in care settings. (royalsocietypublishing.org) Liu’s point, as described in the source briefing, is that those environments matter because they are already instrumented and operationally constrained. (nature.com) In that setup, a product can be judged on whether it saves clinician time or moves a decision faster, rather than on whether it produces another standalone score. ### What does “no extra dashboards” mean in practice? “No extra dashboards” is a product requirement, not a design preference. Liu’s posts, as summarized in the source briefing, argue that clinicians are unlikely to adopt passive-signal tools if they must leave the EHR, monitor a separate interface or learn a new review workflow. (medrxiv.org) That view lines up with how hospital software is typically bought and deployed. Integration work is expensive, clinical attention is limited, and any new screen competes with existing documentation, messaging and order-entry tasks. Liu’s standard therefore pushes vendors toward lightweight hooks — alerts, summaries, flags or task suggestions inside the record — instead of standalone analytics portals. ### Why does she call this an “ROI gap pre-integration”? The “ROI gap pre-integration” refers to a common health-tech problem: a company asks a provider to fund or tolerate integration before the company has shown enough operational benefit to justify the effort. Liu’s argument is that founders need evidence of measurable time savings before scaling commercially, according to the source briefing. That is a harder bar than showing model accuracy. A sensor feature may classify stress or sleep state well in testing, but a hospital buyer still has to ask who sees it, where it appears, what action it triggers and how many staff minutes it saves. Liu’s posts place those questions before expansion, not after it. ### What should founders and product teams take from this now? May 19’s posts point product teams toward a short list of proof points: EHR-native workflow placement, actions that take less than two minutes, and time-savings metrics from real deployments. Those are the facts Liu emphasized in the source briefing. The next step is visible in Liu’s own thread history on May 19, which includes related posts on integration and commercialization sequencing. For companies building around passive data, the immediate evidence to collect is not another dashboard screenshot but deployment data showing who acted, where they acted and how much time the workflow saved.