Case-Control Study of Use of Personal Protective Measures and Risk for SARS-CoV 2 Infection, Thailand

We evaluated effectiveness of personal protective measures against severe acute respiratory disease coronavirus 2 (SARS-CoV-2) infection. Our case-control study included 211 cases of coronavirus disease (COVID-19) and 839 controls in Thailand. Cases were defined as asymptomatic contacts of COVID-19 patients who later tested positive for SARS-CoV-2; controls were asymptomatic contacts who never tested positive. Wearing masks all the time during contact was independently associated with lower risk for SARS-CoV-2 infection compared with not wearing masks; wearing a mask sometimes during contact did not lower infection risk. We found the type of mask worn was not independently associated with infection and that contacts who always wore masks were more likely to practice social distancing. Maintaining >1 m distance from a person with COVID-19, having close contact for <15 minutes, and frequent handwashing were independently associated with lower risk for infection. Our findings support consistent wearing of masks, handwashing, and social distancing to protect against COVID-19.

cross-reactivity with a panel of other respiratory viruses including SARS-CoV, MERS-CoV, and hCoV (NL63, OC43, 229E, and HKU1) (3). When an RT-PCR assay result was negative, the result was reported immediately. When a RT-PCR assay result was positive for either or both ORF1b and N gene regions, a confirmation test was conducted by using either a PCR assay targeting different regions or targeting ORF1b and N gene regions with different targets, or nucleotide sequencing of ORF1b or N gene regions.
We telephoned contacts during April 30-May 27, 2020 and asked details about their contact with a COVID-19 index patient, such as date, location, duration, and distance of contact.
We asked whether contacts wore a mask during the contact period with the following possible responses: never; yes, non-medical mask; yes, medical mask; yes, alternately nonmedical and medical masks; unknown or cannot remember. If contacts reported wearing a mask, we asked the frequency of mask-wearing during the contact period with the following possible responses: sometimes; all the time; unknown or cannot remember. We asked if and how frequently they washed their hands during the contact period with the following possible responses: no washing with soap or alcohol-based solutions; yes, sometimes; yes, all the time after any contact (defined below as 'often'); unknown or cannot remember. We asked if they performed social distancing, including type of contact with the COVID-19 index patient or other persons at place of contact, if unable to remember who the patient was with the following possible responses: had physical contact; shortest distance <1 m and no physical contact; shortest distance >1 m and no physical contact; unknown or cannot remember. We asked about the total duration of contact with the following possible responses: >1 hour; at least 15 minutes but not >1 hour; <15 minutes; unknown or cannot remember. We asked whether contacts shared a cup or a cigarette with other persons in the place they had contact or had highest risk for contact with the index patient with the following possible responses: no, yes, unknown or cannot remember. We asked whether the COVID-19 index patient, if known to the respondent, had worn a mask with the following possible responses: never; yes, nonmedical mask; yes, medical mask; yes, alternately nonmedical and medical masks; unknown or cannot remember. We asked if the COVID-19 index patient wore a mask and the frequency of mask-wearing with the following possible responses: sometimes; all the time; unknown or cannot remember. We also asked, and verified using DDC records, whether and when the contacts had symptoms or COVID-19 was diagnosed. The reporting of this study follows the STROBE guidelines (5).

Statistical Analysis
We developed the final multilevel mixed-effect logistic regression models on the basis of previous knowledge and a purposeful selection method (6). In short, we performed the following: 1) fit a multilevel mixed-effect univariable model with each covariate; 2) selected candidate variables with the α level of <0.25; 3) evaluated variables that were not statistically significant in the multivariate model at an α level of 0.10; 4) fit a reduced model and evaluated confounding by change in log odds ratios of any remaining variables compared with the full model; 5) repeated steps 3 and 4 until the model contained statistically significant covariates or confounders; and 6) added back in the model, 1 at a time, any variable not originally selected, kept any that were statistically significant, and reduced the model following steps 3 and 4. We kept sex, age group, and sharing dishes or cups in the mixed-effect multivariable model on the basis of previous knowledge that sex, age group, and sharing dishes or cups were associated with COVID-19 development.
We estimated the direct population attributable fraction (PAF) by using the imputed dataset and the direct method, as previously described (7,8). Direct PAF can be obtained by calculating PAFs directly from subjects' data by using logistic regression (7,8). First, we modified our final logistic regression model by considering each risk factor dichotomously.
Then, irrespective of exposure to each risk factor for each subject, that factor was removed from the population by calculating probability based on all observations as unexposed. The predicted probability of developing COVID-19 for each asymptomatic contact, with the assumption that no exposure to a certain risk factor occurred, was defined by: Pki is representative of predicted probability of COVID-19 in an asymptomatic contact, k, assuming no exposure to a specific risk factor (xi); β j indicates the regression coefficient of risk factor (xj), except risk factor number i (xi). Subsequently, the sum of all predicted probabilities for all subjects in the study would be equal to the adjusted estimate of total cases, which is anticipated in the absence of that specific risk factor (xi).
Then, PAF was estimated by subtraction of the total number of predicted cases from total number of observed cases, divided by the total number of observed cases: PAF = Total number of observed cases − total number of predicted cases total number of observed cases

Characteristics of the Cohort Data
For the nightclub cluster, we identified 11 primary index patients who started having

Characteristics of Cases and Controls
Because all household contacts were tested with RT-PCR assays, we further explored and household contacts who did not have COVID-19 (9 days [IQR 6-12 days]; p = 0.65).

Multivariable Analyses
Wearing masks all the time during contact was independently associated with lower risk for SARS-CoV-2 infection compared with not wearing masks (aOR 0.23; 95% CI 0.09-0.60) (Appendix Table 1), but wearing masks sometimes during contact was not (aOR 0.87; 95%CI 0.41-1.84). We further explored whether the risk for infection was different between those who wore masks all the time and those who wore masks sometimes. We found a negative association between risk for SARS-CoV-2 infection and wearing masks all the time compared with wearing masks sometimes (aOR 0.27; 95% CI 0.11-0.70; p = 0.007).

PAF
We estimated that the proportional reduction in cases that might occur if everyone wore a mask all the time during contact with COVID-19 patients was 0.28; that is, the PAF of not wearing masks all the time (Appendix Table 2 †Crude and adjusted odds ratios were estimated by using logistic regression with random effects for location and for index patient nested within the same location. Missing values were imputed using the imputation model. ‡The state enterprise office was included as a workplace. Others included restaurants, markets, malls, religious places, and households of index patients or other persons but not living together. Location was included in the model as a random effect variable. §Sharing multiserving dishes and using communal serving utensils to portion food individually was not categorized as sharing dishes. ¶Included sharing electronic cigarettes and any vaping devices. #Included washing with soap and water, and by using alcohol-based solutions. **Wearing masks during the contact period. Wearing masks incorrectly, such as not covering both nose and mouth, was considered the same as not wearing a mask for analyses. *Prevalence was estimated by using the imputed dataset. Population attributable fraction was estimated by using the direct method (7,8). PAF, population attributable fraction; Prev, prevalence. †Sharing multiserving dishes and using communal serving utensils to portion food individually was not categorized as sharing dishes. ‡Included sharing electronic cigarettes and any vaping devices. §Included washing with soap and water, and by using alcohol-based solutions. ¶Wearing masks incorrectly, such as not covering both nose and mouth, was considered the same as not wearing a mask for analyses. #Age and gender were considered nonmodifiable risk factors; other risk factors were considered modifiable. Total PAF was directly estimated by using logistic regression in the form of natural logarithm; therefore, total PAF was not equal to the direct summation of PAF of each risk factor.

Appendix
Appendix Figure 1. Timeline and possible transmission of severe acute respiratory syndrome coronavirus 2 from primary index patients of the nightclub cluster, March-April 2020, Thailand.