Comparison of Saliva and Mid-Turbinate Swabs for Detection of COVID-19

Background: Saliva is an attractive sample for detecting SARS-CoV-2 because it is easy to collect and minimally invasive. However, contradictory reports exist concerning the sensitivity of saliva versus nasal swabs. Methods: We recruited and followed close contacts of COVID-19 cases for up to 14 days from their last exposure and collected self-reported symptoms, mid-turbinate swabs (MTS) and saliva every two or three days. Ct values and frequency of viral detection by MTS and saliva were compared. Logistic regression was used to estimate the probability of detection by days since symptom onset for the two sample types. Results: We enrolled 58 contacts who provided a total of 200 saliva and MTS sample pairs; 14 contacts (13 with symptoms) had one or more positive samples. Overall, saliva and MTS had similar rates of viral detection (p=0.78). Although Ct values for saliva were significantly greater than for MTS (p=0.014), Cohen's Kappa demonstrated substantial agreement ({kappa}=0.83). However, sensitivity varied significantly with time relative to symptom onset. Early in the course of infection (days -3 to 2), saliva had 12 times (95%CI: 1.2, 130) greater likelihood of detecting viral RNA compared to MTS. After day 2, there was a non-significant trend to greater sensitivity using MTS samples. Conclusion: Saliva and MTS specimens demonstrated high agreement, making saliva a suitable alternative to MTS nasal swabs for COVID-19 detection. Furthermore, saliva was more sensitive than MTS early in the course of infection, suggesting that it may be a superior and cost-effective screening tool for COVID-19.


39
The U.S. Centers for Disease Control and Prevention (CDC) recommend the use of upper 40 respiratory specimens, including but not limited to nasopharyngeal, mid-turbinate nasal, anterior 41 nasal and saliva specimens for the initial diagnosis of COVID-19 1 . Although nasopharyngeal 42 swabs (NPS) are considered to be the standard for the detection of COVID-19 by most 43 researchers, collection requires the use of trained professionals, can cause discomfort to the 44 patients, and may pose greater risks to healthcare workers during sample collection 1-4 . Mid-45 turbinate swabs (MTS) are sometimes used as an alternative to NPS in an effort to reduce patient 46 discomfort and occupational exposures to healthcare workers 4-6 . Compared to swab-based 47 collection, saliva is even less invasive, more affordable, and can be self-collected with minimal 48 or no supervision 1,7,8 . 49 50 Existing studies focusing on the sensitivity of NPS compared to MTS, and NPS compared to 51 saliva have produced contradictory results 2,4,[9][10][11] . Few studies directly compare saliva and MTS 52 specimens. Furthermore, there is strong evidence that pre-symptomatic transmission results in 53 higher secondary attack rates for both symptomatic and asymptomatic transmission 12,13 . 54 However, most of the existing studies only looked at detection of symptomatic cases after 55 symptom onset 2,4,9,10 and few looked at detection sensitivity starting with the pre-symptomatic 56 period. Therefore, research that conducts a direct comparison of MTS and saliva, including an 57 assessment of sensitivity over time (starting during the pre-symptomatic period) is critical to 58 identifying optimally sensitive methods for early detection and effective control of SARS-CoV-2 59 transmission. 60 61 The purpose of this study was to compare the sensitivity of MTS and saliva specimens for 62 detecting SARS-CoV-2 by actively following close contacts of COVID-19 cases and collecting 63 MTS and saliva samples for real-time reverse transcription polymerase chain reaction (RT-PCR) 64 during their post-exposure quarantine period. 65 66

67
We enrolled 58 individuals with known close contact with an active COVID-19 case. Contacts 68 provided a total of 200 saliva and MTS pairs. The number of days of sample collection per 69 participant ranged from one to seven. Among the contacts, 14 (24%) had at least one positive 70 sample including 11 with both positive saliva and MTS samples over the course of follow-up. 71 One contact had only positive saliva on 3 out of 3 samples (on days -3, 0, and 1 post symptom 72 onset) and 2 had only positive MTS samples; one was positive on 2 of 2 swabs (on days 7 and 73 10) and one on 1 of 5 swabs (day 21, negative on days 14, 17, 19, and 24 all were in the positive group. No other significant differences were identified between the 80 positive and negative groups (Table 1). 81 82 . CC-BY-NC 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 January 6, 2022.  Figure 1a). The Ct values for saliva were on average slightly but significantly greater than 99 for MTS samples (mean difference = 0.64, p=0.01) among all 58 participants (see Figure 1b and 100 Supplementary Fig. S1 online), partially reflecting the difference in their input amounts. 101 102 Relationship between days since symptom onset, probability of detection, and viral RNA copy 103 numbers 104 The Ct values among the positive symptomatic participants increased over time (days -3 through  105 24), along with decreasing viral RNA copy numbers. Saliva tended to have lower Ct values and 106 higher viral RNA copy numbers compared to MTS from days -3 to 1.5, whereas MTS samples 107 had lower Ct values and higher viral load thereafter ( Figure 2A and Figure 2B). 108 109 Among symptomatic participants who had one or more positive saliva or MTS samples, the 110 probability (sensitivity) of detecting viral RNA in saliva samples was 91% (10/11) from day -3 111 to day 2 (Table 3), was 89% (16/18) from day 3 through 8, and declined significantly thereafter 112 (see Figure 2C and Supplementary Fig. S2 online). The probability of detecting virus in MTS 113 samples from day -3 through day 2 was 45% (5/11), was 94% (17/18) from day 3 through 8, and 114 then declined. 115 116 Early in the course of infection (days -3 through 2) saliva had 12 times the odds of being positive 117 (95% CI: 1.2, 130) and 3.2 times higher viral RNA copy numbers (95% CI: 2.8, 3.8) compared 118 to MTS. There was a trend toward greater sensitivity and higher viral RNA copy numbers in 119 MTS than saliva samples after day 2 post onset of symptoms (Table 3). 120 121 Asymptomatic case 122 Only one participant from our study population was an asymptomatic case. They provided one 123 pair of saliva and MTS samples, both of which were positive, with an average Ct value of 25.8 124 for MTS and 34.7 for saliva (see Supplementary Table S4 online). 125 126

127
. CC-BY-NC 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 January 6, 2022. ; https://doi.org/10. 1101/2021 Early in the course of infection, saliva was significantly more sensitive than mid-turbinate nasal 128 swabs (MTS). We found that the optimal performance of saliva was in the pre-symptomatic 129 period and was more sensitive than MTS before symptom onset. Several studies have shown that 130 pre-symptomatic transmission plays a more important role than symptomatic and asymptomatic 131 transmission in the spread of SARS-CoV-2 12,13 . Furthermore, saliva tended to have lower Ct 132 values and higher viral load compared to MTS from the pre-symptomatic period through the first 133 days post symptom onset. Together, these findings suggest that saliva may be the most effective 134 method for detecting SARS-CoV-2 early during the course of infection. 135 136 The CDC and the Infectious Disease Society of American recommendations for COVID-19 137 testing allow MTS, NPS, oral swabs, anterior nasal swabs, and saliva swabs as well as saliva. 1,14 138 Some studies have shown differences in the sensitivity between NPS and MTS. In older, more 139 acutely ill populations, NPS appears to be more sensitive than MTS, especially later in the course 140 of illness (greater than 7 days) 4,9 . In a study of ambulatory and symptomatic participants whose 141 ages were more evenly distributed, NPS and MTS swabs were highly correlated with a mean of 7 142 days since onset of symptoms 15 . Congrave-Wilson et al., in agreement with the current study, 143 found that saliva had the highest sensitivity in the first seven days post COVID-19 onset when 144 using NPS as the reference 2 . Similarly, Savela et al. noticed that although the peak viral load of 145 SARS-CoV-2 in nasal swabs were higher than saliva, the latter was more likely to be positive in 146 the first six days since the participants' first positive sample was detected 16 . Becker et al. 147 compared the sensitivity of saliva and NPS for detecting COVID-19 in a convalescent cohort 8-148 56 days since first symptom and found that NPS performed better 17 . They also showed that 149 saliva was about 30% less sensitive than NPS in a separate diagnostic cohort, however, days 150 since symptom onset were not reported, so we cannot make direct comparison with our findings. 151 Finally, a systematic review by Bastos et al. found that saliva had similar sensitivity to NPS and 152 costs less 11 . 153 154 Our findings have implications for improving public acceptance of COVID-19 testing, reducing 155 the cost of mass COVID-19 screening, and improving the safety of healthcare workers who 156 conduct testing. These findings are extremely important when considering large-scale screening 157 of COVID-19 in schools and workplaces. In addition to its higher sensitivity in the early stage of 158 the disease as demonstrated in our data, saliva has quite a few other advantages that make it an 159 appealing screening tool. Saliva collection is less invasive and more acceptable to the general 160 population 7,18 . One of the barriers hindering COVID-19 testing is people's fear of nasal swabs 161 due to misinformation 19 . Also, the discomfort brought by nasal swabs may also reduce people's 162 willingness to get tested regularly, especially among children 20,21 . With the use of saliva, 163 screening large groups with increased frequency may be more practicable. Saliva is cheaper than 164 swab-based methods, especially if pooled samples are used 11,22 . Bastos et al. estimated that 165 when sampling 100,000 individuals, using saliva saved more than $600,000 in comparison to 166 using NPS 11 , using a method that is more expensive than the SalivaDirect method used here. 167 These cost savings are especially important in the context of low resource settings. 168 169 Saliva collection is also safer for healthcare workers (HCWs). Amid the pandemic, one of the 170 key concerns among HCWs is the occupational exposure to SARS-CoV-2 aerosols during some 171 medical procedures 23 . The collection of nasal swabs introduces such exposure via the close 172 interaction between patients and HCWs and by patients' coughing and sneezing as a result of the 173 . CC-BY-NC 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 January 6, 2022.  196 197 In conclusion, the use of saliva is preferable for testing pre-symptomatic populations. It is more 198 acceptable to people, which reduces barriers to testing. It is also more cost effective for 199 individuals to collect their own saliva rather than using highly trained professionals to collect 200 NPS and/or MTS. Finally, self-collected saliva samples eliminate the exposure to aerosols 201 produced by sneezing, coughing and gagging of patients undergoing NPS/MTS. 202 203

204
Study population 205 We analyzed MTS and saliva sample data from individuals who reported close contact with 206 confirmed COVID-19 cases as part of the University of Maryland StopCOVID study 26 from 207 May 2020 to April 2021. 208 209 Questionnaire and sample collection 210 Participants were followed every two or three days for up to 14 days from their last exposure or 211 until SARS-CoV-2 was detected in their samples. If one or more of their screening samples 212 became positive, results were confirmed by an appropriate clinical diagnostic test and they were 213 recruited to participate in the exhaled breath aspect of the study that also involved the collection 214 of saliva and MTS 26 . On each day of sample collection, participants answered an online 215 questionnaire to update their current symptoms and medications. For those who reported having 216 any symptom, they also reported their symptom onset date (i.e., "When did you begin to feel 217 sick?"). 218 219 . CC-BY-NC 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 January 6, 2022. ; https://doi.org/10. 1101/2021 The symptoms checklist in the baseline and follow-up questionnaires, as previously described 26 , 220 included runny nose, stuffy nose, sneezing, sore throat, earache, malaise, headache, muscle 221 and/or joint ache, sweat/feverish/chills, nausea, loss of appetite, vomiting, abdominal pain or 222 diarrhea, chest tightness, shortness of breath, and cough. Participants self-reported for each of 223 these 16 symptoms on a scale of 0 to 3 (0 = "no symptoms," 1 = "just noticeable," 2 = "clearly 224 bothersome from time to time, but didn't stop me from participating in activities," 3 = "quite 225 bothersome most or all of the time and stopped me from participating in activities"). 226 227 For saliva collection, participants were instructed to not eat or drink 30 minutes prior to the visit 228 and then collect approximately 0. reported per mL for saliva and per sample for MTS. The limit of detection was 75 copies per 248 sample and the limit of quantification was 250 copies per sample. A positive sample was defined 249 as having Ct values < 40 for at least two out of three SARS-CoV-2 targets (ORF1ab, N gene, and 250 S gene) 27 . The average Ct values of all positive targets were used in the following analyses. 251 252 Statistical analyses 253 We analyzed only paired same-day saliva and MTS samples to ensure the comparability of the 254 two samples. Group comparisons were made between participants having a positive result for 255 either sample and those with both samples being negative. Continuous variables (age and BMI) 256 were compared using t-test, and categorical variables were compared using Chi-square test (sex 257 and chronic respiratory illness) and Fisher's exact test (age group, race, and ever smoker). 258 259 To compare the Ct values from saliva and MTS, we conducted paired t-test and Bland-Altman 260 analysis, and calculated the coefficient of determination (i.e., R squared from linear regression) 261 and Pearson correlation coefficient. The Chi-square test was used to explore the relationship 262 between detection and sample types. Cohen's Kappa was calculated to demonstrate the degree of 263 agreement between the two sample types. 264 265 . CC-BY-NC 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 January 6, 2022. . CC-BY-NC 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 January 6, 2022.  . CC-BY-NC 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 January 6, 2022. in the community setting. medRxiv 2020. 05.11.20092338 (2020) 335 doi:10.1101/2020.05.11.20092338. 336 . CC-BY-NC 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 January 6, 2022. 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 January 6, 2022. ;https://doi.org/10.1101https://doi.org/10. /2021 28. Hastie, T. & Tibshirani, R. Generalized additive models for medical research. Stat Methods 360 Med Res 4, 187-196 (1995 CC-BY-NC 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 January 6, 2022. ; https://doi.org/10.1101/2021.12.01.21267147 doi: medRxiv preprint c Symptoms at the time of each sample collection visit. Sixteen individual symptoms were rated 409 from 0 to 3. Systemic (max score of 12) = malaise + headache + muscle/joint ache + 410 sweats/fever/chills; Gastrointestinal (max score of 12) = loss of appetite + nausea + vomit + 411 diarrhea; Lower Respiratory (max score of 9) = chest tightness + shortness of breath + cough; 412 Upper Respiratory (max score of 15) = runny nose + stuffy nose + sneeze + earache + sore 413 throat. 414 415 . CC-BY-NC 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 January 6, 2022. ; https://doi.org/10.1101/2021.12.01.21267147 doi: medRxiv preprint . CC-BY-NC 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 January 6, 2022. ; https://doi.org/10.1101/2021.12.01.21267147 doi: medRxiv preprint a. Days since symptom onset inclusive of the start and end day 422 b. Odds ratios and their 95% confidence intervals were estimated using logistic regression. 423 c. Effect estimates and their 95% confidence intervals are shown as the ratio of RNA copy 424 numbers of saliva to MTS. Analyses were controlled for random effects of subjects and sample 425 nested within subjects and for censoring by the limit of detection using a linear mixed-effects 426 model for censored responses (R Project lmec-package). 427 d. All samples from the 13 mildly symptomatic contacts of known cases with days since symptom 428 onset from day -3 through day 24. 429 430 . CC-BY-NC 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 January 6, 2022. ; https://doi.org/10.1101/2021.12.01.21267147 doi: medRxiv preprint . CC-BY-NC 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 January 6, 2022. C. Probability of being tested positive by days since symptom onset estimated from a generalized 447 additive logistic model. 448 449 . CC-BY-NC 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 January 6, 2022. ; https://doi.org/10.1101/2021.12.01.21267147 doi: medRxiv preprint