Clearance is not the finish line for a cloud-connected device. After launch, the software and the cloud services it depends on keep changing.
Two kinds of change hit your cloud after launch. There are the changes you make: features, bug fixes, security patches and maintenance. And there are the changes in your environment: the services, runtimes and dependencies under your device keep evolving.
IEC 62304 covers medical device software development and maintenance, and AAMI TIR45 gives guidance on doing that work with Agile practices. AAMI TIR115:2026, which builds on AAMI CR510:2021, gives more detailed guidance on using public cloud computing for medical device functions. This post shows how to control both kinds of change without slowing your team, and which part a platform can support.
Hardware is mostly locked at design freeze. Software keeps evolving after clearance, and that is expected. Four sources of change show up in every software device we see:
The volume is the real problem. Every change, however small, needs an impact assessment, appropriate verification and a documentation trail. That is before a single new feature.
Then add the cloud itself. Services, runtimes and dependencies change often. Some changes come with long notice, such as runtime end of support. Others are routine updates. Either way, you need a process to identify the changes relevant to your device, assess the risk and verify the system.
Six documents govern a software change after launch:
| Document | What it asks of you |
|---|---|
| ISO 13485 §7.3.9 / 21 CFR 820 | Review, verify and approve design changes before release |
| IEC 62304 §6 and §8 | A maintenance plan, change requests and configuration control |
| ISO 14971 §10 | Feed post-production information back into the risk file |
| FDA software change guidance (2017) | Decide when a change needs a new 510(k) |
| FDA PCCP guidance | Pre-agree planned AI changes so they ship without a new submission |
| AAMI TIR45 | Guidance on using Agile practices in medical device software development |
In practice, every change runs through the same four questions. Does it affect risk (safety, security, privacy, usability)? Does it affect the intended use? How much of the design does it touch? Does it cross a regulatory threshold for reporting or submission?
The answers set the tier. Clear tier definitions let you scale the work to the risk:
| Tier | Typical change | What it usually triggers |
|---|---|---|
| Patch | Bug fix that restores the spec | Focused verification, regression of the affected area, letter to file |
| Minor | Changed behavior of an existing feature | Risk file update, targeted usability check, updated requirements |
| Major | New feature or expanded intended use | Full design change package, possible new submission |
TIR45 adds a useful discipline. A change is "ready" only when its impact assessment is done. It is "done" only when the code, the tests and the documentation changes are all complete. The product is not just code. It is the code plus every record that supports it.
Two failure modes are common. Over-process: a one-day fix waits three weeks for paperwork, and engineers start routing around the QMS. Under-process: "just a bug fix" breaks production, and the design history file drifts away from what is actually running.
AAMI TIR115:2026 gives detailed guidance here. It builds on AAMI CR510:2021, which addressed the use of public cloud computing for medical devices. TIR115 is a Technical Information Report, not a standard. It is voluntary, and nobody will audit you against it. Its starting point is practical: you cannot freeze a public cloud.
So TIR115 moves you from direct control of every change to indirect control of the environment. You validate and control what you build, the Medical Device System (MDS). You qualify, configure and monitor what you run on, the Medical Device Digital Environment (MDDE). The cloud provider becomes a purchased supplier under ISO 13485 §7.4, not a piece of SOUP. We covered the boundary itself in our AAMI TIR115 practical guide.
After launch, what matters is the operating model. Design for continuous change:
You do not need a record of every provider change. Routine updates are covered by the staging mirror and regression. Announced deprecations and breaking changes go through your formal change control, like any change you make. Your risk analysis decides which services deserve tighter monitoring.
Whoever runs your cloud, these ten capabilities have to exist after launch:
| Capability | Why you need it |
|---|---|
| Documented MDS and MDDE boundary | Know which components you validate and which you monitor |
| Written change tier definitions | Scale the process to the risk (IEC 62304) |
| Production-like staging environment, defined as code | Preview changes before they reach patients |
| Automated regression on a schedule | Catch the cumulative effect of changes you did not make |
| Versioned releases with release notes and breaking-change notice | Configuration control (IEC 62304 §8) |
| SBOM per release and vulnerability triage | FDA cybersecurity expectations |
| Audit trail of every change | Show an auditor who changed what, and when |
| Post-deployment verification and rollback | Some issues only appear with production data |
| Monitoring that feeds complaints and CAPA | Close the loop to the risk file (ISO 14971 §10) |
| Supplier qualification of the cloud provider | Purchasing controls (ISO 13485 §7.4) |
Built in-house, every row is yours to build, document and maintain for the life of the device.
A platform sits on both sides of the TIR115 line. Its connectivity, storage and authentication are part of your MDDE. Your plugin code, configured data model and portals remain your MDS. The value is in an explicit split:
| Area | BioT provides at the platform layer | You still own |
|---|---|---|
| Cloud provider changes | Tracks AWS service and runtime changes and ships them as versioned BioT releases, with release notes, breaking changes and upgrade notes | Testing your plugins and configuration against each release |
| Evidence | ISO 13485 certified QMS, IEC 62304-compliant Design History File, SBOM per release, quality agreement | Your DHF, risk file and submission |
| Security | HITRUST r2, SOC 2 Type II, ISO 27001 and ISO 27799 certified platform layer | Security of your application layer and device |
| Environments | Dedicated environments for development, studies and production; Terraform provider and Python SDK to define configuration as code | Your test plan and acceptance criteria |
| Traceability | Audit log of entity changes: who, what, when, values before and after | Linking changes to your QMS records |
| Device updates | OTA module: approved software versions per device group, with installation status per device | Your firmware and the release decision |
| Regulatory | Platform documentation to support your impact assessment | The decision on a new submission |
A concrete example. When an AWS Lambda runtime approaches end of support, teams need to assess and migrate the affected functions. On BioT, the platform-side change was documented in the BioT release notes, with the breaking change called out for plugin owners. Customers still updated and re-tested their own plugin code.
The cost difference adds up. Our 3-year cost comparison puts an in-house build at $2,288,875 against $438,100 on BioT.
No. TIR115 is a voluntary Technical Information Report, not a standard. Your obligations still come from ISO 13485, IEC 62304, ISO 14971 and the regulations. TIR115 is useful as a reference for what good looks like when your environment is a public cloud.
No. Group provider changes by type. Test and log the ones that affect the interface or the performance of a service your device depends on. Your risk analysis decides which ones those are.
Rarely on its own. The FDA software change guidance looks at whether safety, effectiveness or intended use are affected. You still need a written assessment that supports the decision, usually as a letter to file.
Often enough to fix bugs and vulnerabilities quickly, and rarely enough that the fixed cost of each release stays reasonable. Many teams settle on a regular sprint cadence for planned changes, plus a fast path for urgent security fixes.
No. The manufacturer stays responsible for the device. A platform reduces how much of the cloud you have to change, test and document yourself, and gives you the evidence for the part it provides.
Your device will change after launch. Your cloud environment will change even when your code does not. You are expected to control both, with evidence.
You can build and staff that machinery yourself. Or you can start from a platform layer that already runs it, and spend your engineering time on the part that is truly your device.
If you are setting up your post-launch change process for a connected device, we are happy to go through it with you: your change tiers, your MDS and MDDE split, and what evidence you would need from whoever supplies your cloud. That conversation is worth having whether or not BioT turns out to be the right fit for you.
Related reading: AAMI TIR115: A Practical Guide for Medical Device Companies Using the Cloud and Build vs Buy: Medical Device Cloud Platform Cost Comparison.
This article is commentary, not a substitute for the referenced standards and guidance or for regulatory advice.