Connecting to a hospital system: what connects, what does not
Updated 2026-11-16
Separate three layers first — patient index, report write-back, model viewing — then discuss fields and protocols. The same word covers three jobs with completely different effort and ownership.
"Can it connect to our hospital system" is the most asked and least answered question at purchase. The vagueness is not technical. It comes from not separating the layers: the same word can mean bringing patient details in automatically, writing the report back into the electronic medical record, or letting a doctor view the 3D model inside the hospital system. Three different jobs, three different efforts and owners.
Separate the layers first, then discuss interfaces, and integration stops being a bottomless pit.
1. Three layers, not one word
Layer
Typical requirement
Layer one: patient index
Patient number and basic details carried in automatically at capture, avoiding duplicate entry and mismatches. Smallest effort, most immediate benefit.
Layer two: report write-back
Structured results and the report file written back into the hospital system so the doctor sees them in the existing interface. Medium effort, depending on how open the hospital interface is.
Layer three: model viewing
The 3D model viewable and rotatable inside the hospital system. Largest effort: storage, format and client performance.
The layers can be delivered in stages. Most institutions do layer one first, then two, and decide on three by department need. Attempting all three at once usually stalls at layer three and drags the benefits of the first two down with it.
2. What connects: fields, protocol, trigger
Layer one needs four fields: patient number, name, sex, date of birth. Layer two needs report identifier, report time, structured indicators and the report file. Layer three needs the model file address and format.
Protocol follows the hospital's existing interface method — commonly standard message exchange and file exchange. The hospital IT department provides the interface document; the device side implements against it.
Trigger points must be written into the interface list: carried in at registration, written back after the report is generated, pulled when the doctor opens the patient view. "Real time" is not a specification.
Failure handling. When the network or interface times out, the device must keep working locally and backfill on recovery. This is where sites fail most often — if the interface is down and capture stops, the device is simply out of service.
3. What does not connect: three boundaries
No diagnostic conclusions. The device outputs measurements and risk indications, not diagnoses. Write-back contains no diagnostic codes or conclusion fields, so the data cannot be read as a diagnosis.
No automatic ordering. Results do not trigger medical orders, open examination requests or start billing. That is hospital business logic, and the device side should not enter it.
No cross-hospital sharing by default. Data stays inside the hospital. Regional sharing is designed separately against the regional platform's requirements; it is not a default capability.
Put these three in the technical specification rather than the contract annex — the engineers building the interface are the people who read the specification.
4. Who signs the interface list
Integration fails most often not on technology but because nobody owns field semantics. The remedy is an interface list signed by three parties.
Hospital IT: confirms the interface is usable, network policy is open, accounts and permissions are in place.
Clinical department: confirms the fields suffice, the report view matches reading habits, and exception prompts are visible.
Device side: confirms the implementation scope, retry strategy and version compatibility commitment.
For every field, state name, type, value range, whether required, and what happens when it cannot be obtained. These columns remove most of the argument during joint testing.
5. Five joint-testing steps
Environment. Test environment, test account, test patient. Never use real patient data for integration testing.
Single field end to end. Start with the patient number only, proving network and authentication work. Nothing later is meaningful until this passes.
All fields and exceptions. Complete the fields, then test exceptions deliberately: empty field, unknown number, duplicate registration.
Report write-back. Verify against how the hospital system displays it; watch long-text truncation and special characters.
Load. Group screening produces continuous write-backs. Test at peak once.
6. Acceptance after go-live, and how to retreat
Acceptance covers four points: the normal path works, exceptions produce prompts, capture continues without a network, and backfill works after recovery.
Write the rollback plan before go-live: on interface failure, switch to local mode so capture and reporting are unaffected, and backfill on recovery. Without this, the interface turns from an efficiency gain into a single point of failure. During the first week, compare the write-back count against local records daily; investigate any difference, even one.
7. Four recurring blockers
Field mapping. The hospital's sex and department codes differ from the device's. Build a mapping table before go-live and confirm it item by item.
Network zoning. The device sits on a different segment from the hospital system and the policy is not open. Have IT open it in advance, not on testing day.
Clock drift. A few minutes' difference between device and server reorders write-back records. Time synchronisation is cheap and high value.
Missing audit log. No record of who viewed whose model. Where personal information is involved this is a hard requirement; confirm the log fields before go-live.
8. Scheduling the project and where the effort goes
Layer one is usually fastest. Few fields, simple logic; most of the cost is IT opening network and accounts. Once it works, the department feels the difference that same day — no more duplicate patient entry.
Layer two depends on the hospital system: whether the interface is open, whether documentation is complete, whether it must go through the hospital integration platform. The waiting time here is scheduling, not engineering.
Layer three needs a client check. Model viewing has performance requirements; old client machines will stutter. Test on a typical machine, not the engineer's laptop.
Effort concentrates in three places: interface development and joint testing, changes to the hospital integration platform, and dual running during the first weeks (manual and interface checked in parallel). The third is most often left out of the budget, and it decides whether the first month is calm.
9. The boundary in one sentence
The goal of integration is to save one manual trip for the data, not to let the device make judgements for the hospital. Anything involving diagnosis, medical orders or billing stays inside the hospital system; the device is responsible for delivering accurate, traceable measurements to where they belong. Hold that line and the integration scope stays discussable.
Next step
If you are planning an integration, send the hospital's interface document and the department's requirements. The account manager will map the three layers against your timeline. Cooperation runs as direct purchase or through a dealer arrangement; details are discussed directly.
Note: interface protocols vary between hospital systems; the scope described here is the general boundary, not a specification for a specific hospital. This article contains no pricing.
Measurement conditions and what the result covers
The accuracy figures and posture data discussed here all rest on the same measurement conditions: the person wears close-fitting clothing or shorts, is barefoot, stands naturally, stays as still as possible during the scan (about 3–5 seconds), and has hair tied back or pinned up so it does not cover the head and shoulder outline.
When these conditions are met, the data is used for preliminary screening, risk indication and follow-up comparison of posture and spinal status; when they are not met (loose clothing, movement, hair covering the outline), the result is an estimate for reference only. The device does not provide a medical diagnosis — any abnormal finding needs review at a medical institution.
Accuracy is millimetre-level (vendor nominal ±1 mm). A living body is not a calibration artefact: millimetre-level variation across three consecutive scans of the same person is normal, so read the trend from the range of several measurements rather than from a single number.
Where the data comes from
Specifications here come from the device datasheet and user manual — nothing is estimated or inferred.
Case data comes from client project records; organization names can be verified by phone. We do not use client names without authorization.
Qualification records can be looked up by certificate number in the National Medical Products Administration database.
FAQ
We only want patient details filled in automatically. Which layer is that?
Layer one, the patient master index, and it is the one worth doing first: few fields, immediate benefit, and duplicate entry plus mis-matching disappear on day one. For most institutions this layer is enough; decide the other two against actual departmental need.
If the interface is down, does the device still work?
It must. Capture and local reports do not depend on the hospital interface; on failure the device switches to local mode and uploads later. Write this into the plan as a requirement, otherwise integration turns an efficiency gain into a single point of failure.
Will writing reports back into the medical record make them look like a diagnostic basis?
Not if what is written back is measurement data and risk notes without diagnostic conclusions or diagnostic codes. List the write-back fields explicitly in the technical plan and exclude any conclusion-type field. Settling this before integration testing is far easier than explaining it after go-live.
How long does integration take, and where does it usually stall?
Layer one is usually measured in weeks, and the usual blocker is the IT department's schedule for opening network access and accounts, not the implementation itself. Layer two depends on whether the hospital system exposes interfaces and whether traffic must go through an integration platform. Approve the three layers as separate items so one cannot hold up the rest.