A heart monitor can detect an abnormal rhythm while the server behind it delays the alert. Boston Scientific described that problem in an August 2026 customer letter. The delay could occur with a specific alert setting.
As a tester, you could check that the monitor collects data and still miss a delay in the clinic's alert. To test that delay, you need to follow the data through the daily check with that setting in place.
A Baxter pump recall describes another software issue tied to a specific sequence: changing a syringe before the planned infusion is complete. You can use state transition testing to explore both cases. With this technique, you test how software responds to an event while it is in a given state, such as paused or running.
The setting behind the delayed alert
Clinics use LATITUDE Clarity server software with Boston Scientific's LUX-Dx family of implanted heart monitors. One alert tracks AF burden, the time spent in atrial fibrillation, a type of abnormal heart rhythm. The clinic can choose the daily amount that triggers an alert.
According to the letter, the delay needs 3 conditions. The threshold must be set to ">0 hours per day", which is not the default. The stored trend data must meet the alert criteria. And no other event must trigger a full device interrogation before the next scheduled follow-up.
A full interrogation is a more complete check of the device data. With the affected setting, the daily check fails to trigger that step as intended. The server therefore cannot assess the alert until a full check occurs. Another alert or a check started by the patient or clinic can trigger it.
The implant continues to detect, store, and send data as intended. The fault described in the letter is in the server's alert process. The FDA record, Z-3081-2026, was posted on 2 September 2026 and lists the cause as software design.
The daily check should move the server to the next state. With the fault, it does not, so the test fails. Illustrative test only, based on the August 2026 customer letter.
What happens when the syringe runs out?
Baxter's case involves a Volume/Time infusion that needs more than 1 syringe. The user sets a volume of fluid and the time over which to deliver it. A software issue may occur when an empty syringe is replaced while some of that volume remains to be infused. The FDA posted this recall event's records on 31 August, 2 days before the Boston Scientific record.
Baxter's correction tells users to keep the same compatible syringe brand and size within that workflow. It also addresses prompts shown when a replacement syringe cannot support the intended flow rate. The public records do not explain how the software fails or how that affects the dose.
To design a test, you still need to know what the pump should do next. A tester needs the product requirements to define the correct result after the syringe change. Assuming the pump should always resume at the old rate would miss the case where the new syringe cannot support it.
Model the sequence before writing the test
The ISTQB Foundation syllabus, section 4.2.4, describes state transition testing using states, events, and guard conditions. A guard is a condition that must be true for a transition to occur. A state diagram or table helps you turn those rules into tests.
For the pump, a useful starting state is an infusion in progress. An empty syringe triggers a change in state, and the volume still to deliver affects what should happen next. The replacement syringe's supported rates add another condition to test. This is a proposed test model; the public records do not include Baxter's internal design.
For the server, model the daily check and the move to a full check. Include the threshold setting and whether another event has already triggered that full check. A test that ignores those conditions could pass without exposing the reported delay.
So what did the manufacturers test? Did their tests include these sequences? The public notices do not contain their test records. They describe defects, but they do not show why those defects were present in released software. We can use those descriptions to improve our tests without guessing about the teams that built the products.
Check the settings as well as the transitions
A test that finishes an infusion with a single syringe can verify useful behavior. But it provides no direct evidence about a syringe change during that infusion. Testing the change means reaching the empty-syringe state while fluid remains to be delivered, then following the next steps.
An alert test also needs to record the settings it used. Suppose a feature has 5 threshold options and your tests use only 1. You have exercised 20% of those options. That does not tell you what share of the feature's behavior or code paths you covered. Different settings may share code, and each setting may interact with several events.
For a small set of safety-related settings, I would start by checking each option. Then I would tailor the added tests to the risks and the rules that differ between options. Include values at the threshold and on either side where the requirement defines different results. Record the reason for any groups of settings you treat as equivalent.
The ISTQB syllabus distinguishes coverage of states, valid transitions, and all transitions in a state table. All-transition coverage includes attempts to make invalid transitions, such as resuming when a required condition is absent. It calls for at least this level of coverage for safety-critical software. Even then, a model can omit a setting or a sequence that needs its own test.
Turn the alert notice into a test outline
The Boston Scientific letter gives enough detail to outline a test for the delayed alert. In a test environment, I would set it up as follows:
Use the affected ">0 hours per day" threshold with the AF Burden alert enabled.
Supply trend data that meets the alert criteria. Record the values and timing so another tester can repeat the case.
Prevent other events from triggering a full check before the scheduled follow-up.
Perform the daily device check and capture the server's actions.
Check that it starts a full interrogation and shows the alert within the time required by the product specification.
That final time limit must come from the requirements. The letter alone does not supply every pass/fail rule or the setup needed for a repeatable test.
I would also run a separate case in which another event triggers the full check. That case checks the alternate route described in the letter. With separate cases, you can see whether another event hid the daily-check fault.
What belongs in the verification records
Under IEC 62304, including its 2015 amendment, software requirements must be tested or otherwise verified. The manufacturer must record the link between each requirement and that evidence. Test records also include the software version and relevant test setup.
This does not create a blanket rule to test every possible combination of settings. You still need to justify how the chosen tests cover the requirements. A result recorded only as “alert test passed” gives a reviewer little basis to assess the settings and event sequences tested.
When an alert serves as a risk control, there is a further check. ISO 14971:2019, section 7.2, requires evidence that the control is implemented and effective. Here, effective means that it reduces the risk as intended. Risk may remain after the control is in place.
For an alert intended to prompt timely review, a test could check that the alert appears within the required time. You can use that result to assess the control. It does not, by itself, prove that a clinician will act in time or that harm cannot occur.
FAQ
Does state transition testing apply to server software?
Yes. The technique applies whenever behavior depends on the current state and an event. The LATITUDE Clarity case shows why the server belongs in the test scope alongside the implanted device.
Were patients harmed in these cases?
Boston Scientific's August letter reported 1 complaint and no patient injuries or deaths. The cited Baxter records do not provide an injury count. A missing count does not mean nobody was harmed.
Does a missing state diagram prove a gap in testing?
No. Test cases can cover transitions without a separate diagram. During review, look for evidence of the states, events, and conditions covered, in whatever form the team records them.
Use this in your next test review
Choose a workflow that can pause, resume, or use a setting other than the default. Trace it through the test cases and check the expected result at each change. If the evidence is unclear, add the missing setup details or design another test. The next reviewer should be able to see which sequence was checked and repeat it.
For training on the wider process, Nopmark's Medical Device Bundle includes courses in software validation, risk management, and human factors.
Correction, 22 August 2026. This article originally said Edition 2 was in the final stage of standardization with publication targeted for August 2026. Both statements were wrong. Edition 2 is…