| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1 Department of Diagnostic Radiology and Organ Imaging, Chinese University of
Hong Kong, Prince of Wales Hospital, Shatin, Hong Kong, China.
2 Statistics and Research Unit, Hong Kong Hospital Authority, Hong Kong,
China.
3 Department of Medicine and Therapeutics, Chinese University of Hong Kong,
Prince of Wales Hospital, Shatin, Hong Kong, China.
4 Professional Services and Medical Development Division, Hong Kong Hospital
Authority, Hong Kong, China.
Received February 27, 2004;
accepted after revision August 10, 2004.
Authors' note.—The chest radiographs used in this study—the
initial radiographs of the first 138 patients— were previously analyzed
in a study for the chest radiographic changes in SARS
[1]. The present study differs
from the former in that it is a complete evaluation of all available
radiographic information available for the Prince of Wales Hospital (including
later radiographs of the first 138 patients). This study focuses on the
radiographic profile differences between the patients who were discharged and
those who died and on the use of radiographic scores early in the disease as
potential prognostic indicators.
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. Chest radiographs that had been obtained from presentation until the death or discharge of 313 patients with SARS were scored on the basis of the percentage area and location of lung opacification. Profile analysis and univariable logistic regression were performed on these radiographic scores.
RESULTS. Despite the increased mortality risks of advanced age and
male sex, no significant difference was seen in the percentage area of
opacification (AO%) between the sexes in either the group of patients with
fatal outcomes or the group of patients who were discharged. No difference
existed between age groups (< 65 years vs 
65 years), except for the
radiograph showing the peak lung opacification in the deceased group in which
the lungs of older patients had less opacification than those of younger
patients. The radiographic scores obtained by day 7 were the earliest ones
with good performance in prognostic prediction. The model showed good
discriminatory performance, indicated by high C-indexes for receiver operator
characteristic curves (0.86 for AO% and 0.90 for the number of opacified
zones). The predicted proportion of patients with fatal outcomes showed high
agreement with percentage of patients who died (goodness-of-fit statistic
p = 0.18 for AO%, 0.73 for the number of opacified zones). By day 7,
crude odds ratio of death was 1.73 per 5% of AO% (p < 0.0001) or
2.93 per lung zone opacified (p < 0.0001).
CONCLUSION. Chest radiographic scores (percentage of lung or the number of zones opacified) by day 7 could be used as fatal prognostic indicators.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
At the Prince of Wales Hospital in Hong Kong, 336 patients were admitted for SARS, with 288 patients who were eventually discharged and 48 who died. This study was a continuation of previous efforts by this institution to analyze the radiographic findings in SARS [1–3]. In this retrospective study, all available serial radiographs for every patient with SARS who was treated at the hospital were scored and used for statistical analysis. In the end, complete inpatient radiographic records were available for 313 patients. The patients in this study were treated with broad-spectrum antibiotics, a combination of ribavirin and low-dose corticosteroid, and then IV high-dose methylprednisolone, depending on responses [4].
This study was aimed at analyzing the trends in the major radiographic abnormality, air-space opacification [5–10], using a percentage area score and the number of radiographic lung zones affected. A retrospective analysis of radiographic scores was performed from two perspectives: first, a profile analysis of the radiographic scores at different stages of disease between the discharged patients and those who died, and second, a detailed analysis of the scores from the first day of symptom onset to the peak radiographic opacification point. The latter was conducted to examine the utility and sensitivity of the chest radiographic scores for predicting prognosis in isolation and to identify the earliest radiograph potentially capable of doing so.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Image Evaluation
Chest radiographs were obtained at initial clinical presentation and during
treatment. Only frontal chest radiographs were assessed (posteroanterior for
patients who could stand, anteroposterior for those who could not). All
radiographic examinations were performed with computed radiography equipment
(Mobilett Plus, Siemens Medical Solutions) using a standardized technique (75
kV, 4 mAs, 180-cm film-focus distance for posteroanterior; 70 kV, 4 mAs,
100-cm film-focus distance for anteroposterior; broad tube focus for both).
The images were assessed using a PACS (Magicview, model VA22E, Siemens Medical
Solutions) viewer (2K monitor).
The chest radiographs obtained at clinical presentation and during treatment were retrospectively reviewed by seven radiologists working in pairs. Regular discussions among the radiologists were held for problematic cases and to reduce observer bias in the scoring system. The radiologists were blinded to the clinical progress or final outcomes of the patients. The method for radiographic evaluation was identical to that used in an earlier study by our institution [1].
Each lung was divided into three zones (upper, middle, and lower) for each side. Each of the three zones spanned one third of the craniocaudal distance of the lung on a frontal radiograph. Each of the six zones was evaluated separately for opacification of the lung parenchyma, and the findings were reached by consensus. Two sets of radiographic scores were obtained from each radiograph: one in terms of the percentage area of lung opacification (AO%) and one for the number of zones with opacification. The size of the lesion was assessed by visually estimating the percentage area occupied (at 5% intervals from 0–100%) within each zone. The overall AO% was obtained by averaging the percentage of involvement in the six lung zones. The number of zones involved was obtained by counting the zones with nonzero involvement. For each subsequent radiograph, the extent of lung parenchyma involvement was assessed by the same method.
Statistical Analysis
The daily radiographic scores of all 313 patients were merged with their
demographic data, date of symptom onset, and key treatment dates for
statistical analysis. The data on symptom onset and treatment dates were
retrieved from the Hong Kong Hospital Authority central clinical database of
SARS patients. The analysis was performed using Statistical Analysis System
(version 8.0, SAS Institute). The chi-square test was performed to assess if
the distribution of baseline characteristics of the study subjects was
comparable to the overall 1,755 patients with a clinical diagnosis of SARS in
Hong Kong [12,
13].
Profile analysis.—The AO% at five milestones during each patient's hospital stay was plotted (giving an individual radiographic profile) and analyzed. These milestones included radiographs obtained at presentation; after initiation of treatment with ribavirin; after initiation of treatment with pulsed corticosteroid; at peak lung opacification; and at discharge or death. If no radiograph was acquired on the day of the second and third milestones, the scores for the radiograph obtained the day before (or if that was also unavailable, the radiograph obtained the day after) were used. This model analysis of variance (ANOVA) with a repeated-measures (at five milestones) design, was performed to look for differences in the AO% in the discharged patients and those who died; the AO% shown at the five serial milestones; and radiographic profiles over the five milestones for the two groups. For the third analysis (a profile analysis), the individual AO% profile of each patient (in each group) for the five milestones was plotted. These were collectively compared with the profiles obtained for the other group, looking for any difference between the two groups. This analysis was also done to look for profile differences between sex and age subgroups within each group of the deceased and discharged patients.
Prognostic indicator selection.—The risk of death (crude odds ratio) with respect to the radiographic score by a specified day (after onset of symptoms) was estimated using a univariable logistic regression model. This was repeated for each day from the first day of the initial radiograph to the day with peak radiographic opacification, using first the AO% separately and then the number of opacified zones (each as a predictor variable of continuous value).
To assess the performance of the model for each day, we performed the Hosmer-Lemeshow goodness-of-fit test [14] for calibration evaluation, and the area under the receiver operating characteristic (ROC) curve was computed with the use of the C-index [15] for discrimination evaluation.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
The number of patients who survived and were discharged was 265, which
consisted of 106 males (40.0%) and 159 females (60.0%). There were 48 deaths
(mortality rate of 15.3% as compared with the overall rate in Hong Kong of
17%), which consisted of 31 male (64.6%) and 17 female (35.4%) patients. The
mean age was 36.8 years for the discharged group and 73.6 years for the group
who died. Those patients who died were mainly older (81.3% 65 years vs
55.1% < 35 years for the discharged patients).
Radiograph at Presentation
The initial chest radiograph was obtained an average 4.4 (SD, ± 3.4)
days from symptom onset, and 231 (73.8%) of the 313 patients had abnormal
findings on the initial chest radiograph. Air-space opacification with an
ill-defined margin was the radiographic abnormality observed in all these
patients.
Profile Analysis
Discharged versus deceased groups.— The AO% for each zone at
each milestone is shown in Figure
1. At all five milestones, the radiographs showed that lung
opacification had a predilection for the lower zones in both discharged and
deceased groups. The right lung scores were slightly higher than those of the
left. At all five milestones, the mean AO% of the deceased group was
severalfold that of the discharged group
(Fig. 1).
|
The mean AO% for the deceased and discharged groups over the five milestones is shown in Figure 2A. According to repeated-measures ANOVA model, there was a significant difference (all p < 0.0001) in the AO% between the discharged and deceased groups; AO% among the five milestones; and the radiographic profiles over the five milestones between the two groups.
|
Significant differences in the percentage of lung opacification (p < 0.005) were found in multiple pair-wise comparisons between milestones in each group of patients (discharged and deceased) with one exception: In the discharged group, the percentage of lung opacification at initial presentation was not significantly different from that of the opacification seen on the radiograph obtained before discharge (p = 0.67).
Sex and age difference.—The mean AO% for the discharged and
deceased groups was plotted against the five milestones in terms of sex
(Fig. 2B) and age subgroups
(divided patients into groups < 65 years and those 65 years)
(Fig. 2C). No difference was
seen in the radiographic profiles over the five milestones between the sexes
in each group (p = 0.22 for discharged and p = 0.7 for
deceased groups).
|
|
A significant difference was found in the radiographic profiles between the two age subgroups within the deceased group (p = 0.004). At the peak lung opacification point, the older group scored lower than the younger patients (p = 0.02). No difference (p > 0.05) was found in the other four milestones. In the discharged group, no difference was seen in the radiographic profiles between the two age subgroups (p = 0.15).
Peak lung opacification.—Among the discharged patients, the mean and median numbers of days from symptom onset to peak lung opacification were 12.2 and 11 days, respectively. The corresponding mean and median values for patients with a fatal outcome were 16.7 and 12 days, respectively. In more than half (57%) of all patients, the worst radiographic lung opacification developed by day 12. Therefore, day 12 was further analyzed; the analysis showed a diametric concentration of frequencies (Table 2) within the discharged and the deceased groups. Most (63%) of the discharged patients had less than 10% lung opacification, and most (61%) of the patients who died had more than 25% lung opacification. Similarly, 89.1% of the patients who died had four or more zones with opacification.
|
Prognostic Indicator Selection
Univariable logistic regression analysis on patients with fatal outcomes
was performed for each day from day 1 to day 12, using the overall AO% as a
predictor variable of continuous value first and then the number of zones
opacified separately (Table 3). These two parameters were factored into the model separately due to their high
correlation (r2 = 0.76).
|
The Hosmer-Lemeshow goodness-of-fit test statistic was computed for each of the first 12 days from symptom onset (Table 3). The resultant p values suggested that the model adequately fitted data on AO% from day 7 onward (p = 0.18–0.47) and data on the number of zones opacified from day 3 onward (p = 0.15–0.86). In terms of the C-index (equivalent to the area under ROC curve), the model of day 7 achieved the best discriminatory performance (0.86 for AO%, 0.90 for the number of opacified zones). Therefore, day 7 was considered the earliest day showing good performance as a prognostic predictor in terms of both calibration and discrimination.
Day-7 Radiograph as Prognostic Indicator
We found polarization of radiographic scores when comparing the discharged
group with the deceased group (Table
4). Of the discharged patients, 86% had less than 10% of total
lung opacification by day 7. Of the patients with fatal outcomes, 55% had lung
opacification of 20% or more by day 7. We found a similar distribution when
considering the number of zones opacified. None of the patients who died had
normal findings on chest radiographs by day 7.
|
For calibration evaluation of our models, we divided the patients into the three subgroups using the two sets of dividing points to compare the actual and predicted proportion of deaths. The first set of dividing points was 5% and 20% of lung opacification (Fig. 3A), and the second set was one and four opacified lung zones (Fig. 3B). With day-7 radiographs, each set showed good agreement between the actual and predicted proportion of patients who died.
|
|
The ROC curves for the outcome of death with respect to the range of dividing points using either AO% or the number of opacified zones are shown in Figure 4. The dividing point achieving maximal sensitivity and specificity was 10% for AO% (sensitivity of 0.82 and specificity of 0.86) and three zones of opacification (sensitivity of 0.84 and specificity of 0.84). The crude odds ratio of death was more than 26.3 (95% confidence interval [CI], 10.7–64.5) if a patient had 10% or more of the total lung opacified. The crude odds ratio was 27.8 (95% CI, 10.9–71.0) if a patient had three or more lung zones opacified.
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
To our knowledge, this is the first study of serial chest radiographic scores to evaluate for an independent, early prognostic indicator of fatal outcome in infectious pneumonia.
The patient cohort of this study consisted of all the patients who stayed in the same university hospital for the complete duration of their SARS treatment. The total mortality rate was 15.3%, with a predilection for male patients (64.6%) compared with the slight female majority (56.2%) of the cohort. The latter is probably a reflection of the patient population of this cohort, in which health care workers (mostly women) formed a substantial proportion (52.1%). The male predominance in the fatal cases in our study was similar to the results of other studies that showed that males were more likely to have a poor outcome [5, 7, 10, 18, 19]. However, our study found no difference between the sexes in the radiographic profile over the five milestones within the deceased or discharged patient groups (p = 0.7 and p = 0.22, respectively). Males and females in both groups appear to develop similar degrees of chest infiltrates despite the difference in mortality risk.
Other studies have consistently shown that advanced age was a significant
risk factor in SARS
[5–7,
10,
18,
19]. This was confirmed in our
study, in which there was a predominance of older patients in the group with
fatal outcomes (81.3% 65 years). However, there was no significant
difference in the degree of lung opacification between the older and younger
patients ( 65 years and older vs < 65 years) in the deceased or
discharged cases. The exception was at the peak lung opacification milestone
(p = 0.02), at which the older patients who died actually had a lower
average peak percentage of lung opacification than their younger counterparts
(Fig. 2C). Younger patients may
have mounted a more florid immune response and developed more lung infiltrates
before their death.
Radiographic Profile
The mean AO% for the deceased and discharged patients, when plotted against
five milestones, showed that lung opacification favored the lower zones. This
lower zone predilection is consistent with findings of studies on the
radiographic pattern of SARS by other investigators
[8–10,
20,
21]. At each of the five
milestones, the overall mean AO% for patients who died was severalfold (range,
2.7–11.8) greater than the mean AO% of the discharged cases. These
results are similar to those from other studies
[5–7,
10,
18,
19]. In a study of 38 patients
with SARS who were admitted to the ICU, Fowler et al.
[19] found that patients who
died had more extensive radiographic abnormalities than those who survived.
Similarly, Ooi et al. [6] have
shown that there is a relationship between the radiographic score (which also
was based on the percentage of lung opacification) and treatment response. A
study of 267 patients by Choi et al.
[5] showed that radiographic
progression of lung infiltrates coincided with or sometimes preceded clinical
deterioration. The significant difference in radiographic abnormalities
between the deceased and discharged groups and the significant relationship
between outcome and chest radiographic progression are most likely a
reflection of the fact that SARS-CoV caused single-organ failure, which
contributed to death in most patients with fatal outcomes
[19,
22]. These results highlight
the importance of radiographs in monitoring patients' progress during their
illness.
Prognostic Indicator
Speed is of great importance in an epidemic such as SARS, both in terms of
diagnosis and of assessing progress during treatment
[23]. The high infectivity of
SARS-CoV contributed to the rapid spread through the affected communities, and
resources are stretched at times of an epidemic. A semi-quantitative
parameter, such as a radiographic score that may act as an early prognostic
indicator, would be valuable in guiding treatment decisions and resource
allocation. If patients could be stratified according to risk early in their
illness, different arms of a treatment protocol could be devised and
individual treatment could be tailored. Inasmuch as the treatment protocol
(combination of ribavirin and corticosteroids)
[22,
24] used on the patients in
our institution and others has not been tested in a randomized controlled
trial, future treatment may have to show better results (radiographic
progression and final outcome) than those provided here to claim improved
efficacy.
Because of these reasons and the relationship between prognosis and radiographic progression, the next step in our analysis was to find the earliest radiograph that could predict a fatal clinical outcome. We began by identifying the most frequent day on which the chest radiograph showing the worst findings was obtained in all the patients (day 12 was thus identified). Then we analyzed the potential prognostic predictive utility of chest radiographs obtained from symptom onset up until that day. Evaluating univariate logistic regression models for the radiographic scores for each of the first 12 days (after symptom onset), we identified day 7 as the earliest day on which the radiographic scores had the best performance in potential prognostic prediction. Further analysis of the scores by day 7 showed polarization of radiographic scores between the discharged group and the deceased group, similar to that seen by day 12. For the discharged patients, more than half (61%) had less than 5% of total lung opacification (and 86% had < 10% lung opacification) by day 7. For patients with fatal outcomes, more than half (55%) of the patients had 20% or more of lung opacification by day 7. A similar polarized distribution was observed using the number of opacified zones, in which approximately two thirds (63%) of discharged patients had no or only one zone involved, whereas approximately two thirds (71%) of patients who died had four or more zones involved.
Predictive Accuracy
With the model for day-7 radiographic scores, we found a good correlation
between the predicted and actual proportion of deaths using either the AO% or
the number of opacified zones. Comparison of the test statistics between the
two parameters has shown that the number of opacified zones was a more
discriminatory and predictive potential prognostic indicator.
Use in Practice
Any noncomputerized radiographic scoring may be subjective to the
individual observers. Computer-aided diagnostic or scoring software programs
are, however, of limited value if the radiographic images show great variation
in quality, which is still a problem with portable radiographs. The results of
this study have shown that simply by scoring the number of zones affected by
day 7 from symptom onset, one may obtain a prediction of fatal outcome for
patients with a good degree of accuracy (area under ROC curve = 0.90). In
fact, this crude assessment is more discriminatory than using a more precise,
but more subjective, AO% score. This simplification makes this scoring
adaptable for real-time clinical use where its effect, in terms of suggesting
change in treatment due to a rise in the odds of death, could be realized.
It was interesting to note that the updated guidance on the identification and evaluation of possible SARS-CoV disease issued January 8, 2003 by the U.S. CDC uses day 6 as a cutoff point [25]. The CDC document advises physicians to obtain a CT scan 6 days from the day of symptom onset to look for occult lung opacification in patients with suspected SARS who previously have had negative findings on chest radiographs. From the results of our study, if this group of patients requiring CT continues to have normal or marginally abnormal findings on radiographs on day 7 (with < 5% or one zone of opacification), they have a low probability of death. A radiograph showing normal results by the end of the first week may therefore have significant diagnostic, prognostic, and treatment implications, aside from being a cutoff date to perform CT.
Our study has some limitations. First, it was a derivational study, and no validation analysis was done. Thus, the results are tentative; confirmation will require a prospective independent study. Second, there was no interobserver reliability evaluation for the radiographic scoring. Given the number of radiographs to be scored and the number of radiologists involved, we tried to minimize interobserver variability by having the radiologists work in pairs and to simplify the scoring system. Third, one potential bias is that the radiographic appearance is a factor for consideration in treatment. We sought to limit this bias with the profile analysis that only included milestones at the commencement of therapy rather than exact days after symptom onset, so that patients with similar clinical status were considered and compared. For the potential prognostic indicator, we tried to find the earliest date for such an indicator to limit the effect of treatment on the radiographic appearance.
Our study has shown that despite the increased mortality risk of advanced age and male sex, there was no significant difference in the radiographic profile between age groups or sexes within the deceased or discharged patient groups. This study has also shown that the degree of opacification by day 7, either using the overall percentage of lung opacification or the number of zones opacified, may be useful as a potential prognostic predictor of fatal outcome.
Acknowledgments
We thank the Hospital Authority SARS Collaborative Group for their advice
and support on the use of the SARS central clinical database and the
Statistics and Research Unit of the Hospital Authority Head Office for their
help in preparing this article.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  G. E. Antonio, C. G. C. Ooi, K. T. Wong, E. L. H. Tsui, J. S. W. Wong, A. N. L. Sy, J. Y. H. Hui, C. Y. Chan, H. Y. H. Huang, Y. F. Chan, et al. Radiographic-Clinical Correlation in Severe Acute Respiratory Syndrome: Study of 1373 Patients in Hong Kong Radiology, December 1, 2005; 237(3): 1081 - 1090. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
with solid line) and discharged
patients (
with dotted line).
with dotted line.
.