The sensitivity and specificity of the RT-PCR and the real time LightCycler assays have been reported (9,11,12). A total of 3,611 respiratory, fecal, and urine specimens and 1,699 serum samples were tested for SARS-CoV RNA and antibody, respectively, from 1,048 patients for whom an initial clinical suspicion of SARS was considered. The laboratory results were retrospectively correlated with the clinical diagnoses of these patients. Clinically, 590 of these patients were considered to have clinical SARS, 79 to have suspected SARS, and 379 not to have SARS. The third group included patients hospitalized with febrile respiratory illnesses, many with radiologic changes, in whom SARS had been initially considered in the differential diagnosis.

Overall, 948 (91%) of the patients had one or more specimens tested for SARS-CoV RNA by RT-PCR, and 454 (43%) had acute- and convalescent-phase serum samples available for serologic analysis, with a convalescent-phase serum taken at least 21 days after onset of illness. While specimens for RT-PCR were available from similar proportions (89%–91%) of patients in each clinical category, paired sera were more frequently available from patients clinically categorized as having SARS (417 [71%] from 590) than from patients in the not SARS category (25 [7%] from 379) (Table 1).

Of the patients clinically diagnosed as having SARS, 322 (60%) of 537 patients had evidence of SARS-CoV RNA in clinical specimens. In contrast, 2 (0.6%) of 341 of those clinically diagnosed as the “not SARS” category had RT-PCR evidence of SARS-CoV infection (Table 1). To assess the extent of circulation of SARS-CoV in the general population, 184 fecal specimens (submitted for investigation of diarrheal illnesses from patients thought not to have SARS) and 148 nose and throat swabs (from patients visiting a general practice for nonrespiratory illnesses) were tested for viral RNA by RT-PCR. None of 148 control throat swab specimens and 1 of 184 control stool specimens had evidence of detectable SARS-CoV RNA.

Of 417 patients with clinical SARS for whom paired sera were available, 383 (92%) had a >4-fold rise in antibody titer to SARS-CoV. None of 45 controls had seroconversion to SARS-CoV. Two (8%) of 25 patients clinically diagnosed as the “not SARS” category seroconverted (Table 2), but a further 47 convalescent-phase sera from patients in this group failed to show any more seropositive patients (data not shown). Neither of these two patients had a history of contact with other patients with SARS. However, one had a left mid-zone consolidation confirmed by high-resolution computed tomography scan and had a discharge diagnosis of pneumonia of unknown cause. The other had a mild febrile illness of unknown cause without radiologic evidence of consolidation. None of 200 blood donor serum samples collected in Hong Kong during March 2003 and 2,200 additional serum samples collected in May 2003 had evidence of antibody to SARS.

The profile of SARS-CoV RNA detection in the 386 patients with serologically confirmed SARS-CoV infection was analyzed (Figure). Viral RNA was detectable in the respiratory tract of a proportion (11%–42%) of patients within the first 4 days of illness but was not detectable in stool or urine specimens until days 5 and 7 of the illness, respectively. The proportion of respiratory and stool specimens positive for viral RNA progressively increased and then peaked at approximately day 11 of the illness. While the nasopharyngeal aspirates and throat and nose swabs were the most productive specimens in the first 4 days of disease, stool samples were more useful after the 5th day of illness. Although the rate of detection in clinical specimens gradually decreased from day 16 onward, viral RNA could still be detected after 30 days of illness in samples from the nasopharynx, feces, and urine in a small proportion of patients (Figure). Smaller numbers of saliva, endotracheal aspirate, and sputum specimens were available for testing (Table 3).

Since confirmation of a laboratory diagnosis of SARS within the first 5 days of illness is the greatest clinical need, we studied the diagnostic yield from different specimens in patients with serologically confirmed SARS-CoV infection during this period (Table 4). Sputum appeared to be a good clinical specimen in the early stage of the disease, although the number of specimens tested was small. Nasopharyngeal aspirates and throat and nose swabs appear to be of comparable sensitivity (30% and 28%, respectively), while stool specimens are less useful specimens in the first 5 days of illness (sensitivity 20%). Saliva and endotracheal aspirates are alternative specimens (Table 3), but we could not assess their usefulness because of the lack of specimens collected in the early stage of the illness. In patients whose first specimen tested negative, 25 had a second specimen (of any type) collected within the first 5 days of illness. Three of these 25 were positive; the additional diagnostic yield from a second specimen was approximately 12% (data not shown).

Virus was isolated retrospectively from stored clinical specimens that were RT-PCR positive for viral RNA (Table 5). Virus was more readily isolated from the respiratory tract than from stool specimens. Furthermore, virus isolation was most successful during the first 2 weeks of the illness and was generally negative after day 22 of illness, even though virus was detectable in these specimens by RT-PCR.

In April 2003, the first-generation diagnostic tests for the SARS-CoV became available to clinicians caring for patients in whom SARS was considered in the differential diagnosis. Normally, new laboratory diagnostic tests are extensively evaluated and validated before they are introduced in routine clinical practice. However, in the case of SARS, a new and poorly understood disease, these first-generation test results were provided to clinicians on the understanding that the tests had not been validated and results had to be interpreted with caution.

Continued improvement of the sensitivity of RT-PCR methods (12) makes an analysis of the sensitivity of these first-generation diagnostic methods less relevant. However, these results provide useful information on the best specimens for detection of virus at different stages of illness, the tissue tropism of the virus, and the duration of virus excretion.

Culture of SARS-CoV for preparing the virus-infected cell smears and for virus isolation was carried out under BSL3 conditions, but routine clinical specimens were processed in the clinical virology laboratory under BSL2 conditions after enhanced and reinforced education on safety and good laboratory practice. Given that up to 250 specimens per day were being processed for RT-PCR detection and serologic testing during peak periods, the workload could not be managed in a BSL3 laboratory. None of the laboratory staff became ill with SARS symptoms, indicating that clinical specimens for serologic testing and RT-PCR can be processed safely in BSL2 level conditions.

The association of SARS-CoV with the clinical syndrome of SARS is illustrated by the detection rates of viral RNA in clinical specimens (60% in patients with SARS, 0.6% in the non-SARS group, and 0.3% of controls). Viral RNA detection by these first-generation RT-PCR tests is less sensitive than serologic testing for diagnosing SARS. Correspondingly, 92% of 417 patients with clinically diagnosed SARS and none of the paired sera from 45 unrelated controls seroconverted to SARS-CoV. However, 2 of 25 patients designated as “not SARS” category from whom paired sera were available also seroconverted. Paired sera were available from only a few (25 of 379) patients in the “not SARS” group. At a time of intense pressure on the clinical front-line staff, there was little incentive to obtain convalescent-phase sera from patients believed not to have SARS. These 25 patients may represent a biased sample of the larger group of non-SARS patients. This contention is supported by the fact that a further 47 convalescent-phase sera subsequently obtained from this group of “not SARS” patients failed to show any additional antibodies to SARS. Even patients in the “not SARS” category had a febrile, respiratory, often pneumonic, illness; one of the two patients in the “not SARS” category who had evidence of seroconversion had an undiagnosed pneumonic illness, while the other had an undiagnosed febrile illness without radiologic consolidation of the lung. Overall, a clinical diagnosis of SARS is closely correlated with detection of viral RNA by RT-PCR and seroconversion supporting the etiologic association of SARS-CoV and SARS.

None of 2,400 blood donor sera collected in Hong Kong during the height of the SARS outbreak has any evidence of antibody to the virus. This finding suggests that the spread of SARS-CoV infection in the general community was minimal, with most of the infection associated with clusters and hospital outbreaks (13).

The RT-PCR detection rates for SARS-CoV in respiratory, stool, and urine specimens in the 383 patients with seroconversion to SARS-CoV show that viral shedding progressively increased from onset of the illness until approximately day 11 after onset. Since the first-generation RT-PCR test has relatively low sensitivity, these results reflect the increasing viral load at different clinical sites during the illness. Whereas these data are cross-sectional, in a previous study viral load in nasopharyngeal aspirates was followed up longitudinally in nasopharyngeal specimens collected at days 5, 10, and 15 after illness onset; results of this study also indicated that viral load peaks at day 10 of illness (4). Such a profile of a progressive increase in viral load is unusual for respiratory viral infections. Most other infections (e.g., respiratory syncytial virus, influenza) have peak viral titers in the respiratory secretions at or soon after the onset of clinical illness, after which viral titers and laboratory diagnostic yield decrease progressively (14). This “crescendo” pattern in SARS-CoV detection rates and viral load in clinical specimens has a number of implications. The pattern explains the poor sensitivity of the first-generation diagnostic tests during the first 5 days of the illness and emphasizes the challenge in making laboratory diagnosis early in the disease. These results may also suggest a fundamental difference in the efficacy of the innate immune response in controlling SARS-CoV infection, in contrast, for example, with influenza infection. Innate immune mechanisms are the earliest host defenses that control viral replication and, in the case of many respiratory viruses, do so within the first few days of illness, even before the specific adaptive immune responses have been activated. This response does not appear to occur with SARS, and viral load in the respiratory tract (4) begins to fall only when the antibody response appears, i.e., at approximately day 10 after onset of illness (4,5). This finding may suggest that SARS-CoV is able to evade the host innate response and requires the adaptive immune response to bring the infection under control. Finally, the peak viral load in the 2nd week of illness would predict that virus is more likely to be transmitted later in the course of the illness. This result indeed accords with epidemiologic observations (15). With regard to observations of viral load, the frequent use of steroid therapy in hospitals (16) is a confounding factor that may contribute to the increase in virus load later in the illness.

The relative virus detection rates from different specimens during the illness suggests that respiratory specimens (nasopharyngeal aspirate, throat swab) are more useful in the first 4 days of the illness, while fecal samples are better later in the illness. Urine samples, on the other hand, are not useful at any stage of the illness. A productive cough is not common in the early stage of illness, but in patients who do produce sputum, this specimen provides a high diagnostic yield. Thus, nasopharyngeal aspirates, throat swabs, and sputum, if available, are the best specimens in the first 5 days of the illness.

Detecting virus in the fecal and urine samples, in addition to the respiratory tract, suggests that SARS is not restricted to the respiratory tract. The finding of diarrhea unrelated to antimicrobial drug use in a number of patients supports evidence that the disease is not a purely respiratory one (4). A number of animal coronaviruses (e.g., mouse hepatitis virus and feline coronavirus) have tropism for multiple organs (17). Viral shedding is detectable by RT-PCR in the respiratory, gastrointestinal, and urinary tracts for many weeks after onset of illness, reflecting continued virus replication at these sites. However, SARS-CoV cannot be readily cultured from any of these sites after week 3 of illness. The viral RNA detected by RT-PCR after week 3 of illness is unlikely to represent persistence of viral RNA in the absence of ongoing viral replication. The apparent dissociation between virus isolation and RT-PCR may reflect the mucosal antibody’s neutralizing the virus and rendering it less infectious. This observation also accords with the apparent absence of transmission of infection after week 2 of illness. The fact that virus isolation was done retrospectively may have affected the overall isolation rate. However, SARS-CoV appears relatively stable to freezing and thawing and is stable for many weeks in clinical specimens at 4°C or frozen at –70°C (K.H. Chan and J.S.M. Peiris, unpub. data). In any event, such a bias would be expected to be uniform both early and late in the disease.

In summary, SARS is closely associated epidemiologically with the novel SARS-CoV. The unusual profile of viral shedding from the respiratory tract may explain some of the observed transmission pattern of this disease, including the predilection for affecting healthcare workers.