To our knowledge, this is the largest study to use qRT-PCR to compare NP and OP swabs for a range of respiratory viruses. We found that the relative performance of specimen type varied by virus. Neither specimen performed uniformly better: NP swabs were more sensitive for some viruses (influenza B virus, PIV 2, and PIV 3), OP swabs were more sensitive for others (overall influenza A virus, 2009 H1N1 virus, and adenovirus), and there was no difference for the rest of the viruses. The large number of patients in this study allowed us to perform comparative statistical analysis with a relatively high degree of precision.
For adenovirus, OP swabs were more sensitive than NP swabs, a finding that was statistically significant in both ILI and SARI patients. This difference may reflect the fact that the major site of initial replication of adenoviruses is the non-ciliated respiratory epithelium of the oropharynx 
. The kappa value between NP and OP swabs for adenovirus was low (κ
0.33). This result is consistent with findings of Lambert et al., in which adenovirus accounted for the highest proportion of discordant paired NPA and nasal-throat swab specimens from children 
. Adenoviruses include over 50 serotypes, and the low concordance between NP and OP specimens for these viruses may reflect different cell tropisms of the adenovirus serotypes for different parts of the respiratory tract 
. Although we did not conduct serotyping, future studies that evaluate specimen performance for specific adenovirus serotypes could test this hypothesis.
For influenza viruses, sensitivities of NP and OP swabs differed by both type and subtype: NP swabs were more sensitive than OP swabs for influenza B virus, while OP swabs were more sensitive than NP swabs for overall influenza A and 2009 H1N1 virus; there was no significant difference between swabs for H3N2 virus or the unsubtypable influenza A viruses. The sensitivities of NP and OP swabs for unsubtypable influenza A viruses were low, and the strength of agreement between the two swabs was poor. However, unsubtypable influenza A viruses were likely a mix of 2009 H1N1, seasonal H1N1, and H3N2 viruses, making it difficult to interpret this finding. Because half the influenza A specimens were 2009 H1N1, the overall influenza A results were biased towards the 2009 H1N1 findings. Previous studies evaluating NP and OP swabs in detecting influenza viruses found NP swabs to be more sensitive than OP swabs, but these studies used combined outcomes for influenza A and B viruses and did not analyze by influenza A subtypes 
. The difference in sensitivities of the two swabs in our study may reflect different affinities of influenza types and subtypes for different locations in the respiratory tract. While all influenza viruses infect the respiratory epithelium from the nasopharynx to the bronchioles, 2009 H1N1 virus (and H5N1 virus, which we did not find in our study) can infect lower parts of the respiratory tract, including the alveoli, more commonly than seasonal influenza 
. This difference could account for the better sensitivity of OP swabs, which reach deeper into the respiratory tract than NP swabs, for 2009 H1N1 virus. Of note, OP swabs have been shown to have superior yield over NP swabs for human cases of avian influenza A (H5N1) 
The OP swab sensitivities and kappa values of the parainfluenza viruses were relatively low, with the sensitivity of the OP swab for PIV 2 being the lowest of any virus in our study. This preference for NP swabs is consistent with reports that nasal washes and nasal aspirates have yielded the highest rates of viral recovery for PIV 
This study had several limitations. First, we compared only NP and OP swabs; although many routine surveillance systems for influenza and respiratory diseases collect NP and/or OP swabs, other surveillance systems collect NPAs, NP washes, or nasal swabs 
. While recent studies have used PCR to compare the relative yield of NPAs with those of nose-throat swabs or nasal swabs for respiratory viruses, no studies have evaluated more than three sampling techniques 
. Head-to-head studies in the future comparing NP swab, OP swab, NPA, NP wash, nose-throat swab, nasal swab, and nasal wash specimen types would provide important information for decisions about which specimens to use for respiratory disease surveillance systems. Second, the NP and OP swabs consisted of different kinds of swab material and used different designs. We used conventional polyester swabs for sampling the nasopharynx and flocked nylon swabs for the oropharynx. Although flocked swabs are superior to conventional swabs for cell recovery, a study comparing different swab material and design (rayon versus nylon flocked swabs) in both the nasopharynx and oropharynx found that the difference in the cycle threshold values between sampling sites was much greater than the difference between swab material and design 
. Because neither specimen type was consistently more sensitive than the other, we think it is unlikely that the difference in swab material and design substantially affected our results. Additionally, while we tested for eight viruses, we did not include some common viruses, such as coronaviruses and rhinoviruses, and we did not test for bacteria. Finally, the number of adults (n
85) in this study was relatively small, accounting for just 3.6% of all patients, and as a result there was limited power to compare the sensitivities of NP and OP swabs for specific viruses in this population.
In summary, NP and OP specimens collected from patients with respiratory illness had variable sensitivities by qRT-PCR for eight viruses. Neither specimen was consistently more sensitive than the other. Collecting both swabs had a complementary effect; even when there was higher sensitivity for one technique over the other, the lower-sensitivity technique still identified a considerable number of cases not identified by the higher-sensitivity one. For respiratory disease surveillance programs using qRT-PCR that aim to maximize sensitivity for a large number of viruses, collecting combined NP and OP specimens would be the ideal approach. However, the enhanced sensitivity of using both swabs comes at a higher cost; this includes not only the expense of the second swab, but further patient discomfort as well as additional time and effort from the person taking the sample. Thus, for surveillance systems with limited resources, a single-swab approach, whether NP or OP, would be the most logistically simple and maintain moderate sensitivity for many of the pathogens we tested in our study.