Fig. 1: The ComCor survey combines closed-ended question answers with free text responses to open-ended questions. | Nature Communications

Fig. 1: The ComCor survey combines closed-ended question answers with free text responses to open-ended questions.

From: Extracting circumstances of Covid-19 transmission from free text with large language models

Fig. 1

In the ComCor survey, individuals tested positive for SARS-CoV-2 were asked to fill out an online questionnaire comprising a number of closed-ended questions and an open-ended question. The questionnaire begins with between 70 and 100 closed-ended questions (the number varies because some answers trigger follow-up questions that are not asked systematically). These questions concern various sociodemographic aspects (age, gender, household size, occupation, degrees,...), and aspects of potential epidemiological relevance, such as details on vaccination (number and date of shots), recent behavior (social distancing, wearing masks, smoking habits, washing hands,...) health status, symptoms, recent mode of transportation, type of accommodation, type, size and duration of gatherings (e.g. concerts) and more. Respondents were divided into 4 groups (only 3 of which are shown), depending on whether they know who infected them or not, in the first case whether or not the person who infected them lived under the same roof, and in the second case whether or not they suspect one or more situations of transmission: (i) those infected by a household member (“intra-household” cases; n = 119,162), (ii) those infected by a known person outside of the household (“extra-household” cases; n = 131,125), (iii) those who did not know who infected them but suspected one single specific situation (“suspected situation”; n = 88,955), (iv) those who did not know who infected them but suspected either no specific situation or multiple situations (n = 193,777; not shown). In each group, an open-ended question was asked, which led to free text responses as shown by the examples in bold. We focus our analyses on the n = 79,444 cases in the “suspected situation” group who provided free text responses and on the closed-ended question about the infection context on the bottom right, which allowed seven possible answers: Work, Family, Friends, Sports, Cultural, Religious, Other. Questions and responses were translated from French. See Supplementary Fig. 1 for the original French version. (*) Please note that 12,939 cases (2.4% of the total of n = 545,958) were not asked “Do you know who infected you” and hence do not show up in the downstream categories. (**) Also note that 158 individuals (<0.03%) who responded “Yes, but I have doubts” to the question “Do you know who infected you?” were not asked “Do they live under the same roof?”, accounting for minor discrepancies in the reported numbers. (***) Note that the open-ended question was simplified here for brevity—see the main text for details.

Back to article page