Valérie Golaz

Test-related indicators accurately illustrate what we know about how the epidemic spreads through across France. However, they are not representative of the population as a whole. In order for those indicators to represent the real magnitude of the virus throughout the population, population categories would have to be tested in the proportion that each represents within the population at large. The fact is that most people who get an RT-PCR test have health risks or symptoms. Moreover, the number of tests done and the number of people tested vary by period. For example, a graph from a recent Santé Publique France weekly report shows that the number of tests done is sensitive to immediate policy and context: after a speech by the Prime Minister in late September 2020 in which he said that contacts indicated by contacts (rather than known infected persons) should not be thought of as contacts (September 21 to October 2; epidemic weeks 39 and 40), the number of tests fell sharply, whereas it rose sharply immediately before and during the end-of-year holidays. Doing more testing necessarily means finding more positive cases and a lower positive test rate. Users of these follow-up data often forget the relations obtaining between them and the fact that at no time do they come from a representative sample of the population. 

Test data are sent electronically through the SI-DEP “screening information” system, a secure platform set up by Santé Publique France to monitor the health crisis. The dataset is exhaustive for some types of tests; at this time it covers all RT-PCR and antigen tests done in France. The data provide information on person tested—place of residence, sex, age—that enable researchers and epidemiologists to map and graph developments with considerable precision. However, as with all data collection, these particular data cannot be perfect. In some cases information is missing or may have been entered too quickly by laboratories or pharmacies. Depending of type of information, some is either discarded or replaced. For example, if the tested person’s place of residence is missing, it is sometimes replaced with the test site postal, and it can happen that such cases are not taken into account in determining the results. Once the data are entered—which may not be the day the test was done—they travel through the SI-Dep system. These heavily publicized daily tallies depend on the day of the week they are posted, and the time elapsed between the day the tests were done and the day information is posted may also vary. Daily statistics are therefore subject to wide variability and need to be interpreted cautiously, as do statistics by town or community, now available on the Santé Publique France website. To correctly interpret results, then, researchers need to be aware of all the ways in which the data may be less than accurate.

For more information: 

Golaz V., Gastineau B., Flahaux M.-L., dos Santos S., 2020. Quatre questions-clés sur sur les indicateurs de suivi de l’épidémie de Covid-19 en France [Four key questions about the indicators used to follow the COVID-19 epidemic in France].