T2681/16: is your data good enough?

T2681/16: is your data good enough?

While this Board of Appeal decision is not concerned with AI or machine learning as such, it is highly relevant in that the decision turned on whether the claim was limited to using enough data points to be supported by the description and produce a technical effect supporting an inventive step. This is, of course, of potential relevance to machine learning, where data is paramount. The Board also found that calculating a diabetic risk range index was an algorithmic feature that produced a technical effect serving a technical purpose and is to be considered in assessing inventive step. 

The application in question relates to a system for measuring blood glucose variability. This system comprises an acquisition module for acquiring a plurality of blood glucose data points over a plurality of predetermined time periods, for example, over a number M of days. Based on the acquired blood glucose data points, the system calculates a risk index (an average daily risk range ADRR index). It classifies it into a risk category that is then displayed. Various requests were presented, including varying degrees of detail on the risk calculations and the time periods used in the calculations.  

The first two claims considered by the Board required several data points but left it open when they were collected so that the calculation could be based on a single period, for example, one of the days for which data was acquired. The Board found that, based on the disclosure of the description, the skilled person would understand that summation over a sufficiently large number M of periods is essential to compute a composite risk range value that is meaningful and useful. This was, therefore, an essential feature and the claims lacking this feature were not supported by the description. The remaining claim sets required at least two data points spanning a day, which was found to be sufficient to be supported by the description. 

The Board then turned to the question of inventive step. In agreement with the applicant, the Board found that the difference over the closest prior art related solely to the algorithm used for the risk calculation, which, taken in isolation, cannot contribute to an inventive step only if they produced a technical effect serving a technical purpose. The applicant submitted that the algorithmic difference produced the technical effect of providing an overall measure of the glucose variability (i.e. equally sensitive to both hypo- and hyperglycaemic events) and a prediction of glycaemic events that were better than, or at least alternative to, those used in the closest prior art. The Board accepted that this effect was produced for more extended observation periods.  

Considering the following two claim sets that respectively required a) two data points spanning a day and b) two or more data points for each of two or more days, the Board found it not credible that this effect would indeed be achieved for some of the shorter periods of observation covered by the claims. In light of the disclosure of the application and the skilled person's general knowledge, the Board did not find it credible that such a limited time of data collection would enable the calculation of a meaningful ADRR risk range. Whether the difference over the closest prior art contributed to a technical effect for other parts of the claimed subject matter when more data points were used was irrelevant. Since the claims in question covered embodiments for which no technical effect could be acknowledged, the claims lacked an inventive step.  

The final set of claims considered by the Board required the use of at least 21 data points spanning seven days. While this was a smaller set of data points than that used to produce the experimental validation studies reported in the description of the application, the Board did find it credible that this was a sufficiently large sample to be statistically significant. The Board was therefore satisfied that the technical effect in these conditions was achieved over the whole scope of the claim. Therefore, the algorithmic features were, in principle, capable of supporting the presence of an inventive step. And so, the story ended well for the appellant since the Board also agreed that the difference over the closest prior art was not obvious and sent the application back to the first instance to be granted.  

Some valuable points can be gleaned from this decision. It is an interesting illustration of how the Board will get right down to the technical detail to assess the breadth of claim and whether it is in line with what the applicant has disclosed and the technical effect that is argued to support the presence of an inventive step. In my view, the Board did this in a way fair to the applicant, not holding the applicant to the number of data points on which the validation data was based but taking a common-sense approach as to what can plausibly work.  

It is also interesting that the Board acknowledged, at least implicitly, that calculating a diabetes risk score serves a technical purpose so that purely algorithmic features could contribute to a corresponding technical effect and hence support the presence of an inventive step. This is also in line with my experience of healthcare-related cases in my practice. It seems reasonable to extrapolate from this that calculating (and, e.g. displaying) a risk score that can help a patient or clinician to manage a condition is a technical purpose. This case, therefore, might come in handy when prosecution of similar cases runs into trouble with technical effect/purpose during examination.  

Follow Alexander Korenberg’s AI musings

AI musings

Cookies improve the way our website works. By using this website you are agreeing to our use of cookies. For more information see our cookie policy I accept