During the epidemic period of SARS from March to July 2003 in Taiwan, 28 patients with probable cases of SARS were treated in our hospital. The 28 subjects include four males and 24 females with ages ranging from 13 to 87 years (mean ± SD = 40.78 ± 18.02). Two patients were less than 15 years old, 20 patients were 18 to 49 years old, three patients were 50 to 59 years old, and three patients were older than 60 years. All subjects experienced fevers exceeding 38℃ (100.4°F) and fulfilled the WHO criteria for probable SARS (9). All 28 patients were subsequently shown to have significant seroconversion of the SARS-associated coronavirus (SARS-CoV) antibody indicating arecent infection by the coronavirus. In addition to having a respiratory illness of unknown cause, during the 10 days prior to onset of symptoms, 12 patients had either lived in an area in which community transmission of SARS was documented, eight patients had traveled to such an area, and eight patients had come into close contact with a person known to have SARS. The institutional review board of our hospital approved the study, and informed consent was not required from the patients.

In total, 481 chest radiographs were obtained for the 28 patients to monitor the disease process. The chest radiographs included 379 in the anteroposterior view, 94 in the posteroanterior view, and eight in the lateral view. The studies were obtained using either conventional radiography (274 examinations, AMX-4, General Electric Medical Systems, Milwaukee, WI) or computed radiography (187 studies, Fuji Medical Systems, Tokyo, Japan) with images archived through a picture archiving and communications system (PACS, Centricity Workstation, version 1.0, General Electric Medical Systems). Hard-copy images from both the conventional and computed radiographies were viewed for all reports. The settings for portable radiography were 72-80 kVp at 1.25-1.5 mAs. The number of radiographs obtained for each patient ranged from 4 to 50 (mean ± SD = 17.18 ± 9.84). The hospitalized days ranged from 11 to 29 with a mean of 18.17 days; thus, an average of approximately one radiograph was obtained each day for each patient, particularly during the critical period. The frequency of radiography decreased as patient symptoms abated.

Two radiologists who had 15 and 26 years of professional experience, respectively, reviewed the radiographs together. The radiographic findings were recorded without the knowledge of the clinical data but with the knowledge that the patients were victims of suspicious or probable cases of SARS. Any differences in opinion were resolved by consensus. On each radiograph, each lung was divided into upper, middle and lower lung zones transversely, with each zone spanning one-third of the craniocaudal length of the lung. The severity of the SARS-related lesions within each lung zone was evaluated by scoring the radiographs with a four-point scale based on visual assessment, as follows: 0 = normal, 1 = up to one-third of lung zone involved, 2 = between one-third and two-thirds of lung zone involved, and 3 = more than two-thirds of lung zone involved. The scores for all 6 zones on each radiograph were added to provide a cumulative score that had a range from zero to 18 (Figs. 1, 2). In addition, data collected from all available chest radiographs included the following: whether the SARS-related lesions were unilateral or bilateral; were associated with radiographically identifiable pleural effusion (defined as increased pleural density with obscuration of the costophrenic sinuses and the hemidiaphragm with meniscus-shaped or horizontal upper border); mediastinal or hilar lymphadenopathy (defined as widening or increased opacity of the mediastinum and pulmonary hila); cavitary lung lesions. The mean radiographic score for each patient during hospitalization was obtained by summation of the radiographic score on each radiograph divided by the total number of radiograph obtained. The maximal radiographic score for each patient represented the worst condition of lung opacities during the clinical course.

For each patient, we retrospectively recorded the age, severity of fever (degree of elevated body temperature above 38℃ or 100.4°F), leukocyte count, decreased percentage (< 20%) of lymphocytes in the leukocyte count, lymphopenia (absolute lymphocyte count < 1,000/mm³), thrombocytopenia (platelet count < 15,000/mm³), elevated levels of aspartate aminotransferase (AST), alanine aminotransferase (ALT), lactate dehydrogenase (LDH), creatine kinase (CK), C-reactive protein (CRP), the number of hospitalized days, the duration of endotracheal intubation, the severity of hypoxemia (PaO₂), the level of retained CO₂ in arterial blood (PaCO₂), death, and the presence of comorbid disease. The patient status, mean value of the body temperature and laboratory data during hospitalization were analyzed if the data statistically correlated with the mean and maximal radiographic scores. The normal values, range, median, mean and standard deviation of each variable are shown in Table 1. Seven of our 28 patients had associated comorbid diseases; these include two cases of malignancy including one associated with hypertension, and one case each of hypertension, chronic lung disease, valvular heart disease, thalassemia and diabetes mellitus associated with hypothyroidism. Three of 28 patients expired due to SARS: these included two women without comorbidity, aged 47 and 72 years, and one 41-year-old woman with diabetes mellitus. The onset of illness was based on the clinical manifestation of fever or body temperature above 38℃.

A correlation among continuous variables including the clinical parameters and radiographic scores was assessed with the Pearson correlation coefficient. A forward stepwise multiple linear regression analysis was performed for all variables with a p-value less than < 0.1 in an univariate analyses. The Kruskal Wallis test was used to analyze whether the radiographic scores were related to the categorical variables including death, survival without comorbidity, and survival with comorbidity. Post hoc comparison using the Bonferroni-corrected Mann-Whitney U tests was performed for further analysis of the categorical variables. A p-value less than 0.05 was considered statistically significant.

No mediastinal abnormality, pleural effusion and cavitation in the parenchymal lesion were observed on the chest radiographs. By univariate analysis, both the mean and maximal radiographic scores significantly correlated with fever, age, elevated white blood cell count, a decreased percentage (< 20%) of lymphocytes in the leukocyte count, lymphopenia, elevated LDH, the number of hospitalized days, the number of days of endotracheal intubation, and the level of retained arterial CO₂ (Table 2). In addition, the maximal radiographic score also correlated with the severity of hypoxemia (p = 0.050) (Table 2).

Using the Kruskal Wallis test, the mean as well as maximal radiographic scores significantly correlated with death and the presence of comorbid disease. The mean radiographic score of the patients who died, patients with comorbidities and without comorbidity were 11.12 ± 2.56 (mean ± SD) and 6.31 ± 5.69 and 2.89 ± 2.55, respectively (p = 0.032). The corresponding values for the maximal radiographic score were 17.67 ± 0.58, 9.67 ± 8.09 and 6.00 ± 5.00, respectively (p = 0.033) (Table 3). A multiple comparison by the Bonferroni-corrected Mann-Whitney U test revealed that there was a significantly difference between the number of deaths and survival without comorbidity with respect to the mean as well as maximal radiographic score (p = 0.003 and p = 0.005, respectively) (Table 4).

A forward stepwise multiple linear regression showed that the mean radiographic score correlated most significantly with the number of hospitalized days (p < 0.001), while the second most significant factor was the absolute lymphocyte count (p < 0.001) and the third most significant factor was the number of days of intubation (p = 0.025). The maximal radiographic score correlated best with the percentage of lymphocytes in the leukocyte count (p < 0.001). The second most significant factor was the number of hospitalized days (p < 0.001), and the third most significant factor was the absolute lymphocyte count (p = 0.13) (Table 5).

Several series have described the clinical prognostic factors of SARS. These factors include the severity of hypoxemia and platelet count (17, 38), an age over 60 years and elevated LDH (18, 20), comorbidities (19, 27, 38), and a high neutrophil count on admission (20). Poor prognostic factors for SARS in radiological findings comprise the number of involved lobes during the acute phase, multifocal opacities with progression to diffuse airspaces opacification or an initial radiographic presentation with multifocal or diffuse airspaces opacities (17, 30-32, 38).