Caresoft eICU pulls live data from the monitors and ventilators your hospital already owns, puts every ICU bed on a single board, and — the part most systems skip — proves the data actually arrived.
Most tele-ICU projects stall in the same place — not on the dashboard, on the cable. Here's what we built the platform around.
Nobody needs a universal connector on day one. A fleet survey usually shows a handful of models across 70–80% of beds. We start there and expand.
Including bench testing against a physical unit. We're specific about this because anyone promising a week hasn't tested against a real ventilator.
Numerics every second, buffered locally, backfilled after outages. Waveforms stay at the edge and are pulled on demand, not streamed at your bandwidth.
Data that can't be attributed to a patient goes to a quarantine queue for a human to resolve. It is never discarded and never guessed.
Five hops. Each one is instrumented, so when a number stops updating you know which hop broke and who owns fixing it.
Monitors, ventilators, pumps. Serial, network or vendor gateway — whichever port the device actually has.
A small box on the unit running a driver per device type, normalising every reading to one clinical vocabulary before it leaves the floor.
The link drops, the ICU doesn't. Data queues on the gateway and backfills automatically when the network returns.
Bed mapping from your HIS decides whose data this is. Every sample is stamped with the device clock, the server clock and the offset between them.
Beds ranked by acuity, drill-down to any patient, and a visible link state so nobody reads a frozen number as a live one.
This ordering is a privilege hierarchy, not a list. Each level can do everything the level below it can, within its own hospital.
The lights come on here. Everything above happens on a dark board at 3am. Everything below happens in daylight — in an incident review, an NABH audit, or a conversation with a family. Both halves are the same product.
Responsibilities are configured per hospital, with response and resolution targets. Nothing waits for someone to notice.
| What happens | Who is notified first | Respond | What gets recorded |
|---|---|---|---|
| Gateway offline | Biomedical engineer | 15 min | Downtime opened automatically from last heartbeat, cause category and clinical impact recorded on close |
| Device link down | Named engineer on that device | 30 min | Which bed, which driver, how long, whether data was recovered by backfill |
| Gateway clock drift | Hospital administrator | 60 min | Offset, drift rate and NTP source; affected samples flagged clock-suspect rather than deleted |
| Unattributed data | ICU in-charge | 60 min | Quarantined samples held intact until a human assigns them to the right encounter |
| Nursing chart overdue | ICU in-charge | 30 min | Which bed, how late, who was rostered to that bed at the time |
| Tele advice unacknowledged | Bedside doctor on duty | 15 min | Raised by, directed to, acknowledged at, and the action that was actually taken |
Device time and server time are both stored, with the gateway offset in force at that moment. A trend line you can't defend on timing is not evidence.
Bed mapping comes from your HIS admission feed. Bed swaps, transfers and off-unit trips end the association automatically instead of quietly mislabelling data.
A signed observation is immutable. An amendment creates a new entry linked to the original with a reason. Both stay visible.
Not just changes. Who opened which chart, from which address, at what time — retained and exportable for audit.
Threshold flags here are advisory. The bedside monitor remains the alarm of record. We are explicit about this in the product, the manual and the contract.
Each channel is separately enabled for identified data, and it's off by default. Alerts carry a bed and a reference, not a diagnosis.
Drivers in the catalogue today. Send us your fleet list and we'll tell you honestly what's supported now, what needs building, and how long it takes.
Device support depends on the interface option fitted to your specific unit — some models need a vendor gateway or an export licence that the hospital owns separately. Where a device has no data port at all, we can fall back to screen capture with nurse confirmation, clearly marked as such. Every driver carries a validation state in the product, and we don't describe a driver as clinically validated until it has been tested against that physical model.
Not a brochure — a per-device assessment of what connects today, what needs a driver built, and what it takes to run a pilot in one unit.