We are setting up the PMS plan and according to the MDR Article 88 we need to report trends. We also need to do that based on statistically significant increases. What do we need to do? How do we do trending if we do not have any market data yet? Do I need to hire a statistician now?
No need to hire a statistician. Structure the complaints, plot them over time and describe what you see.
According to the MDR, Article 88 non-serious incidents and expected undesirable side-effects are subject to trend analysis. Statistically significant increases in the frequency or severity will be reported.
That means we need to do two things:
- We need to find a trend in our data
- We need to report them to the competent authority
The ISO/TR 20416 PMS for manufacturers is in some cases the playbook for your PMS activities and I can really recommend taking a look into it. While the technical report describes several ways to identify trends in your PMS data, it boils down to these three types of changes that can be indicated by a trend analysis:
- a sudden significant deviation like an outlier or spike, respectively;
- significant trends, i.e. repetitive deviations or continuous drifting away from the history of earlier values;
- detection by visual inspection, whether the data are subject to cyclic effects, e.g. of calendar events like summer holidays or end of budget periods.
The first step is to document all feedback and complaints that you receive and evaluate if they include serious or non-serious incidents. Trending is only applicable to non-serious incidents. Serious incidents will be covered by your SOP vigilance and you need to initiate a CAPA, FSN and stuff like that.
Afterwards, you cluster the non-serious incidents into categories. These categories could be “device was delivered non-sterile”, “software crashes” or “software made false positive diagnosis”. The individual cluster depends on your software of course.
Now you normalize the data. Ask your sales department for the number of sold devices for each month. Divide the number of complaints by the number of sold or installed devices. Voila, you can now tell the relative number of complaints per month.
Now that you categorized and normalized your data, you can plot them over time. This allows you to inspect the trends visually.
- Is there a peak in complaint rates after you shipped a new software version?
- Did you change your supplier in March and in April your device has an increased number of complaints? This visualization is already a mathematical method to analyze data and helps you to keep an overview.
Coming back to the ISO/TR 20416, you can detect a trend where six or more points in a row exhibit a continous upward trajectory.
In normal people’s language: If you have a steady increase of complaints over the course of 6 months, congratulations that is a trend. This is called Rule 3 of the Nelson Rules, a set of rules to detect trends. You can use other trending rules as well but Rule 3 is by far the easiest and explicitly mentioned by the technical report.
Trend detected – What now?
Once you have identified the trend, you open a CAPA and evaluate the next steps. What changes have been made to the product? What are the incidents? How can they be rectified? Is there a negative impact on the risk-benefit evaluation?
If you decide that there is a change in the risk benefit evaluation you need to report the trend to your competent authority (Article 92). They will investigate the trend with their own methods and provide you with feedback on your counter measures.