Relationship between In-person Instruction and COVID-19 incidence among University Students: A Prospective Cohort Study

Whether university teaching on campus with infection control measures in place is associated with higher risk of COVID-19 than online instruction, is unknown. We will assess this by conducting repeated surveys among students at universities and university colleges in Norway, where some instruction is given in-person, and some is provided online (hybrid model). We will ask about the students’ COVID-19 status, and how much in-person and online instruction they are getting. We will estimate the association between in-person instruction and COVID-19-risk using multivariate regression, controlling for likely confounders. We will also assess whether type of instruction is associated with how satisfied the students are with the instruction, their quality of life, and learning outcomes. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted September 2, 2020. ; https://doi.org/10.1101/2020.08.30.20182139 doi: medRxiv preprint NOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice. STUDY PROTOCOL, 27 August 2020 2 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted September 2, 2020. ; https://doi.org/10.1101/2020.08.30.20182139 doi: medRxiv preprint


Background:
Social distancing measures, including encouraging people to maintain a distance of 1-2 meters from each other and/or to work from home, are likely to reduce the spread of COVID-19, based on observational evidence and logical arguments [1][2][3].
Policy makers need to balance the effectiveness of infection control measures against other considerations, including psychosocial and societal consequences.This balance is difficult to strike, especially for institutions where social interactions are of key importance, such as universities.Lack of knowledge about the relative effectiveness of various social distancing measures makes this judgement particularly difficult.As Gressman and Peck put it: "In the absence of relevant prior experience, these institutions are largely in the dark about how one might expect a COVID-19 outbreak to evolve in the unique environment of a college campus and how much of an effect the many possible mitigation strategies should be expected to produce."[4] Some attempts have been made at modelling the risk of offering instruction on campus, e.g. a group at Cornell University arrived at the counter intuitive conclusion that shifting to online instruction only would lead to more COVID-19 cases than a full return of students.However, this was premised on "aggressive asymptomatic surveillance where every member of the campus community is tested every 5 days", as well as sufficient capacity for quarantining, and a series of other assumptions [5].
While universities and colleges in the United States have opted for different models, ranging from completely online to only in-person instruction [6], the higher education institutions in Norway have decided to offer a hybrid model, with some online, and some in-person instruction.This decision enables us to compare students with predominantly in-person instruction and students who receive more online instruction, and to assess the association between campus presence and COVID-19 incidence.
We are not aware of other studies on the association of in-person instruction, or the effect of offering online instruction, on COVID-19 risk.We therefore believe that this study is important to carry out as it has the potential to generate findings that can inform infection control policies at our universities, and in similar environments.

Methods:
We are inviting all universities and university colleges in Norway to take part.All students at institutions that agree to participate will receive an SMS (alternatively an e-mail) inviting them to take part in the study.The invitation includes a link that directs them to a web-based informed consent-form and questionnaire.
We will ask the participants if they have been tested for COVID-19, the results of such a test, how much in-person instruction they have been offered, and how much online instruction they have been offered.We will also inquire about other risk factors for COVID-19 and background variables that may be included as potential adjustment factors (confounders) in the analyses (see Attachment 1 -Questionnaire).
We will survey the students every two weeks by new invitations by SMS or e-mail.The study period will last as long as the universities maintain their arrangements with in-person instruction for only select groups of students.We plan for a study period that lasts for the remainder of 2020.
Students will be asked for consent to link the survey results to information on study programme, basis for admission, study status, academic results, sex and age from the Common Student System (FS) (see Attachment 2 for full list of variables).Obtaining this information through data linkage will reduce the survey burden for the students and increase the accuracy and quality of the data.The impact of the intervention on exam results and completion will be of significant interest for the institutions and for society at large, as such impacts must be balanced against the needs for disease control.Information on academic performance the previous semester and from upper secondary school will be needed to adjust for these variables as important possible confounders.
We developed a questionnaire through an iterative process, partly based on existing items from existing questionnaires (see Attachment 1 -Questionnaire).Pilot testing with a group of 10 students at Oslo Metropolitan University showed that the questionnaire could be completed with little use of time (around 5 minutes), and that some questions needed to be amended.
We prepared an English version of the questionnaire for students who prefer English over Norwegian.A native English speaker translated the questionnaire to English, and another person not familiar with the questionnaire translated the English version back to Norwegian.We compared the original Norwegian version and the back-translated version, and made minor adjustments.
We have prepared a communication plan, which includes several measures to ensure a high response rate among the students.
Student involvement has taken place at two levels during the project-planning period: We have informed the Student Parliament about the project and they have offered their support, and the pilot testing described above.
Main outcome: Secondary outcomes: • Quality of life ("Overall, how satisfied are you with life right now?") • Teaching satisfaction ("Overall, how satisfied have you been with the teaching you have received in the past 14 days?") We will run multivariate regressions to test whether there is an association between in-person instruction and the outcomes.

In-person instruction is a continuous variable, defined as (Number of days offered in-person instruction) (Number of days offered in-person instruction + Number of days offered online instruction)
We will control for potentially confounding variables, including year of study program, field of study, age, and gender.For further details, see Attachment 3 -Data analysis plan.
We plan for separate analyses for each participating university/university college, and a pooled analysis across institutions.
We will collect directly identifiable data, probably the participants' e-mail addresses.This will enable us to delete responses from participants who wish to withdraw from the study, to avoid sending messages to students who do not wish to take part, and to link different responses from the same respondent.The latter is of scientific importance as it allows monitoring of each participant over time, thus making this a cohort study.
We do not plan to transfer the collected data out of Norway.If we do, all data will be fully anonymized.
We will ensure highly secure data collection and data storage in collaboration with the data handler, USIT at the University of Oslo.Data management will be in accordance with GDPR-regulations.Our Data Protection Officer at the Norwegian Centre for Research Data, has assessed the project plan, and found it satisfactory (19 August 2020).
We see few ethical dilemmas related to this project, apart from the need to ensure secure handling of the collected person identifiable data.The only burden for the participants concerns the time spent completing the survey.The Regional Ethical Committee has assessed and approved the project plan (24 August, REK sør-øst A, reference number 172155).

Dissemination of findings
We will likely publish the findings in the format of scientific article in an open access medical journal.

Power analysis and sample size
We have carried out a power analysis to estimate the necessary sample size needed to detect differences between students who are offered in-person instruction and students who are offered online instruction.We make a number of assumptions in estimating the necessary sample size.
We assume an incidence of COVID-19 of 0.23 % over a 10-week intervention period for students who are assigned to online instruction.This corresponds to the current level of disease in the age group 20-29 in Norway.If we wish to detect effects of in-person instruction that doubles the risk of disease, we would need 21,000 respondents to be 80% certain to detect an effect at 5% significance level.With a 50% response rate, we would need 42,000 students to be invited to the study.Oslo Metropolitan University alone has around 20,000 students.
Several of these assumptions can be challenged.It is highly uncertain whether the spread of COVID-19 will remain at its current level in the Norwegian student population, especially since the incidence has been rising for the past four weeks.Furthermore, there are few if any other studies of COVID-19 interventions in university settings, and thus it is difficult to estimate a likely effect size.Finally, the number of students who will be invited to this study remains to be seen, as it depends on the number of universities that agree to participate in the study.
As there is a large information need on the impacts of COVID-19 interventions and a scarcity of relevant studies, we furthermore believe the study can be beneficial even if underpowered and since a randomized assignment of participants was not possible.It may provide a solid foundation for more rigid, further intervention studies, and also possibly provide some evidence for the plausible range of effect sizes to be expected from moving to online instruction.A further benefit of the study is to map other important consequences of an online instruction intervention, such as effects on teaching satisfaction and life satisfaction.
We have submitted a registration form for the study to ClinicalTrials.gov.Scale from 0 to 10, where "0" is not satisfied at all and "10" is very satisfied 0 (not satisfied at all) A main concern in using a non-random design is that the treatment and comparison groups vary in observable and unobservable characteristics predictive of the outcome (infection).The main measure that safeguards against the comparison group holding more characteristics predictive of the outcome than the treatment group, is that the treatment (in-person instruction) was prescribed to groups of students, i.e. students did not themselves individually select in-person or online instruction.Whole classes where offered in-person instruction, and our main estimator is thus the impact on infection of in-person instruction for everyone offered in-person instruction (intention to treat, ITT [4]), not only those actually receiving in-person instruction.
It is still possible, of course, that the students being offered in-person instructions are more (or less) prone to become infected than the students referred to online instructions, in which case the estimates cannot be given a causal interpretation.The same will hold if the students who respond to the questionnaire (compared to those note responding) are more (or less) prone to infection and non-responding students are more prevalent in the group of students who are offered in-person instruction than in the group referred to online instructions (or v.v.).To explore this empirically, we will collect information that allows us to test the extent to which characteristics of the students differed in the group offered in-person instruction vs. online instruction at the outset.We will undertake balancing tests illustrated as follows: Pre-determined_characteristic i,t-k = a + b Offered_on_campus_instruction i,t + c X i,t-m where i represents individual and t is time, k≥m>0.Even if balanced, we would expect about 1 in 20 tests to be significant at the 5% level of significance, and if more tests are significant this strongly suggest that estimates should not be given a causal interpretation.If fewer are significant, we can still not conclude that the intervention is "as good as randomly" distributed and we should be very careful in making causal inference (1).
The most important pre-determined characteristic is likely the outcome variable before the intervention, i.e. already having been infected.We will collect data on previous infection as well as other (secondary) pre-determined outcome variables and characteristics that may be correlated with infection.These are age, gender, birthplace, parents' birth place, socio economic status (parents' education level), GPA points at admission, admission basis, exam results previous semester (where applicable), planned study credits, social behavior, living conditions, and use of public transport.
Importantly, we know a priori that some pre-determined variables are not balanced, by construction of the intervention.Most clearly, it was a stated goal for many universities that first-year students should be offered more in-person instruction than more senior students.Also, some fields of study were more dependent on in-person instruction than others, and were given priority.All balancing tests will therefore be performed with and without controlling for these "pre-stratification" variables (the vector X).Specifically, we will consider including the following -Dummies for year of study program -Dummies for field of study -Dummies for courses -Age and gender We will also consider models where two or all these variables are interacted.Interactions will reduce the power to detect imbalance, but since the point is to run the exact same regression on the actual outcome variable, including many interactions will also reduce the power to detect effects.This is thus a classical trade-off between consistency and power.Since each university made different considerations, in what groups of students where offered more in-person instruction, these "prestratification" variables will be adapted to the policy, as described by each university before the intervention took place (or was changed).
Our population parameter (latent) of main interest is the effect of the intervention (offering inperson instruction) on the outcome (e.g.disease among the students).This ITT effect may be attempted estimated in a model illustrated as follows: where k≥m≥0.X will include the same "pre-stratification" variables as in the balance test.With the sole purpose to improve precision, we will also consider including more pre-determined variables (i.e. the vector Z, like sex, age, place of birth, socioeconomic background, given that these variables are not included in X).When assessing learning outcomes, GPA points at admission (by cohort and admission basis) will be included as control variables.
The treatment variable will be operationalized both as two dummy variables: 80% or more of offered instruction time (excluding practice-time) is in-person vs. less than 80%; from 75 th percentile upward with in-person instruction vs. from 25 th downward with in-person instruction; and as a continuous variable, i.e. percent of teaching time offered in-person.Because the virus may spread more and more as the campus becomes more and more crowded, we will also estimate models allowing increasing marginal effects on disease as the percent with in-person instruction increases (several dummies).
Our primary outcome is (self-reported) infection, and secondary outcomes are -COVID-19 test taken (dummy, logistic regression) -Teaching satisfaction (scale, linear regression or multinomial logistic regression) -Quality of life (scale, linear regression or multinomial logistic regression) -Learning outcomes (completion and exam performance, linear regression or multinomial logistic regression) In addition to ITT, we will also estimate the following "endogenous" model (using same estimation methods as described above): Outcome i,t = a + b Actual_on_campus_time i,t-m + cX i,t-k + d Z i,t-k Moreover, we will estimate the local average treatment effect (LATE) in the following two stage model [4] Actual_on_campus_time i,t-m = a + b Offered_in_person_instruction i,t-m + cX i,t-k + d Z i,t-k Outcome i,t = a + b Actual_on_campus_tıme i,t-m + cX i,t-k + d Z i,t-k Where Actual_on_campus_tıme in the second equation can be considered as the estimate from the first stage (using e.g.2SLS (1)).LATE is thus providing the effect of actually participating on campus for the students being moved from online to in-person by the offer.There are several strong assumptions necessary to give this a causal interpretation, where the exclusion restriction is typically considered the most important in practice: There is no conditional (on X and Z) direct effect of the

Parent/guardian 1 Parent/guardian 2
Primary/elementary school ( ) Secondary or high school ( ) College/university ( )Next pageHave you been tested for the corona virus previously this year, before the beginning of the fall 14 days, have you been tested for the corona virus?Yes [the Norwegian version used No then yes this time, but the responses should always be in the same order so I am writing yes then no!] No Next page To what extent are you worried about being infected by the corona virus?To a very little extent [Or perhaps it would be better: not at all, slightly, somewhat, moderately, extremely] To a small extent To a moderate extent To a great extent To a very great extent Next page In the past 14 days, have you been to a social gathering where you would guess that there were 20 or more people?Yes [Yes first again] No Do not know Next page How many times have you participated in-person at a social gathering with your "faddergruppe" (buddy group) this fall semester 2020?

2
corona virus Would you like to receive a receipt via email?ATTACHMENT 3 -Data analysis plan