How do I trend medical device complaints?
Count complaints for each product version, and for each problem within it, and compare a recent window, such as the last 30 days, with that version's own history instead of a fixed number. Where you know how many units are in the field, use them as the denominator, so a version shipped in growing numbers does not look like a rising trend for that reason alone. A person then reviews every rise and decides what it needs.
Two questions, two checks
A count rule asks whether several complaints about the same problem arrived close together, for example 3 within 30 days. A rate check asks whether a version's complaint rate is now higher than its own history. You need both: a slow rise never forms a cluster, and a cluster on its own can be chance.
Trend at the level where problems show
Trending all complaints of a product together, or only by a code assigned when a complaint is closed, can hide a rise in one version or one problem. Group by product version first, then by problem code or complaint type, and count complaints from when they are received, not only once they are closed.
Choose the denominator
A complaint count grows when more devices are in use, even if nothing about the device changed. With lot quantities and release dates you can compare complaints per unit in the field. Without them, compare by calendar time and keep in mind that a growing installed base can raise the count by itself.
A new version has no history yet
The first months of a new version are when a design or supplier change shows, yet the version has too little history to compare with. Compare it with the version it replaced over the same length of time.
Where the rules ask for it
- ISO 13485:2016 clause 8.4 asks for the analysis of data, including feedback and complaints.
- EU MDR Article 88 asks manufacturers to report statistically significant increases in the frequency or severity of incidents that are not serious incidents, and of expected side-effects, that could have a significant impact on the benefit-risk analysis.
What MDARIX does
- A detection rule opens a signal when enough complaints about the same problem arrive within a set number of days: 3 within 30 days unless you change it.
- The trend check compares each product version's last 30 days with its own history (up to 180 days, never before the version's release), grouped by version, by problem code or by complaint type.
- Each row says Significant rise (a chance under 1% if nothing had changed), Higher than usual (under 5%), In line with history, or that there are too few complaints or too little history to assess.
- When every lot has a quantity, the expected number follows the units in the field; without quantities the check counts by calendar days and says so. A version too young for its own history is compared with the version it replaced.
- The trend check is a check, never a verdict: it raises no signal on its own. A person raises a signal from a rising row and triages it.