key: cord-0752213-mv4vp27h
authors: Lau, Eric H.Y.; Cowling, Benjamin J.; Muller, Matthew P.; Ho, Lai-Ming; Tsang, Thomas; Lo, Su-Vui; Louie, Marie; Leung, Gabriel M.
title: Effectiveness of Ribavirin and Corticosteroids for Severe Acute Respiratory Syndrome
date: 2009-12-02
journal: Am J Med
DOI: 10.1016/j.amjmed.2009.07.018
sha: f40e19f9fab159093287408f7cfbf24037def8e4
doc_id: 752213
cord_uid: mv4vp27h

OBJECTIVE: Ribavirin and corticosteroids were used widely as front-line treatments for severe acute respiratory syndrome; however, previous evaluations were inconclusive. We assessed the effectiveness of ribavirin and corticosteroids as the initial treatment for severe acute respiratory syndrome using propensity score analysis. METHODS: We analyzed data on 1755 patients in Hong Kong and 191 patients in Toronto with severe acute respiratory syndrome using a generalized propensity score approach. RESULTS: The adjusted excess case fatality ratios of patients with severe acute respiratory syndrome receiving the combined therapy of ribavirin and corticosteroids within 2 days of admission, compared with those receiving neither treatment within 2 days of admission, were 3.8% (95% confidence interval, −1.5 to 9.2) in Hong Kong and 2.1% (95% confidence interval, −44.3 to 48.5) in Toronto. CONCLUSIONS: Our results add strength to the hypothesis that the combination of ribavirin and corticosteroids has no therapeutic benefit when given early during severe acute respiratory syndrome infection. Further studies may investigate the effects of these treatments later in disease course.

The severe acute respiratory syndrome outbreak in 2002 and 2003 infected more than 8000 people in 29 territories with an overall worldwide case fatality ratio of 9.6%. 1 Hong Kong and Toronto bore a significant proportion of the global case burden, together accounting for 25% of all "probable cases." 1 Although there have been only sporadic human infections with the severe acute respiratory syndrome coronavirus since the first global outbreak, 2 the pos-sibility of widespread reemergence cannot be ruled out given the continued trading and consumption of palm civets, a likely animal reservoir. 3, 4 Meanwhile, severe acute respiratory syndrome-like coronaviruses are endemic in bats in Hong Kong and southern China. 5, 6 Ribavirin has broad activity against many DNA and RNA viruses. 7 Corticosteroids can be protective against extensive inflammatory lung damage. 8, 9 Therefore, these 2 drugs were recommended for use as front-line treatments during the outbreak, although not without controversy. 10 Evaluation of the effectiveness of these 2 drugs in the treatment of severe acute respiratory syndrome has consisted largely of uncontrolled studies and small hospital cohorts. A recent meta-analysis confirmed that there are few reliable data on which to base future treatment decisions in severe acute respiratory syndrome, largely because of the lack of adjustment for important confounding variables in evaluative studies published to date. 11 Although randomized controlled trials would provide more conclusive evidence, they cannot be conducted unless large future out-breaks occur, 12 and in the meantime it is critical that current recommendations, both for the treatment of patients and for the identification of those agents that should be prioritized for testing in future trials, are based on the best available data.

We analyzed data on patients with severe acute respiratory syndrome from Hong Kong and Toronto to assess and compare the treatment efficacies, as measured by mortality reduction, of early administration of ribavirin alone, corticosteroids alone, or both in combination. We used propensity score analysis to adjust for potential confounding effects, an approach that has been proposed as the most valid analytic strategy for observational data in the absence of a randomized design 13 and has been used to evaluate drug effectiveness in many observational studies. [14] [15] [16] [17] [18] In addition to providing important evidence-based guidance for future human cases of severe acute respiratory syndrome, our work may serve as a template for the evaluation of treatment efficacies for other emerging infectious diseases.

We analyzed an integrated database containing clinical and epidemiologic details on all 1755 patients in Hong Kong reported to have "probable severe acute respiratory syndrome" according to the World Health Organization definition, 19 derived from the Hong Kong Hospital Authority Electronic Severe Acute Respiratory Syndrome system and the Hong Kong Department of Health's Master list. 19, 20 For the Toronto data, the Canadian Severe Acute Respiratory Syndrome Research Network conducted a retrospective cohort study of all probable severe acute respiratory syndrome cases among adults (age Ն 16 years) admitted to a hospital during the outbreak. 20, 21 Medical charts were reviewed by trained research staff, and clinical, radiographic, and laboratory data were double entered into a database.

Key patient characteristics at hospital admission that were extracted from these 2 sources included gender, age, occupation, preexisting comorbid conditions, calendar date of symptom onset, delay between symptom onset and admission, lactate dehydrogenase level, chest radiograph findings, neutrophil count, platelet count, lymphocyte count, and oxygen saturation. We classified patients into 4 treatment groups based on whether they had started a course of ribavirin, corticosteroids, both ribavirin and corticosteroids, or neither drug within 2 days of admission to hospital.

We included all patients in Hong Kong. In Toronto the outbreak occurred in 2 waves (first wave: March 2 to April 21; second wave: April 22 to June 10), 22 and ribavirin was not used at all during the second wave because of clinical experience in the first wave. 12 In this analysis we require that each patient had the opportunity to be treated with ribavirin or corticosteroids, and therefore we excluded the 94 patients in Toronto who were admitted to hospital during the second wave. We excluded the 12 patients (0.6%) in Hong Kong and 2 patients (1%) in Toronto who died within 2 days of admission.

In randomized controlled trials, the randomization process should on average balance patient characteristics across treatment groups, and comparison of crude event rates across treatment groups should be unconfounded, 23 although adjustment for potential confounders may improve precision. 24 However, in situations in which treatment assignment is not random, but may depend on patients' presenting characteristics and clinical course, failure to appropriately adjust for such factors may lead to misleading findings. Standardization or stratification can be carried out for each characteristic but is typically not feasible when there are a large number of potential confounders. Regression adjustment is often used provided there are sufficient events; a simple rule of thumb is 20 events (eg, deaths) per potential confounder. 25 An alternative approach is the use of propensity score analysis. 13 Under this approach, a model is specified to explain the assignment of treatments in terms of patient characteristics. The predicted probabilities, or "propensity scores" from this model are in turn incorporated in a successive model to assess the association between the treatment and the outcome of interest. In the latter model, the estimate of the treatment effect should reflect adjustment for differences in the observed patient characteristics between treatment groups.

To study the relative effects on mortality of the 4 treatment groups, namely, neither drug, ribavirin alone, corticosteroids alone, or both drugs, we used a generalized propensity score approach that can allow for more than 2 treatment choices. 26 We estimated the propensity scores using a multinomial regression model adjusting for the key patient characteristics at admission as described above and 3 other factors, including calendar date of onset, onset-toadmission delay, and admitting hospital. Because some prognostic factors were only available for a subset of approximately 30% of the patients in Hong Kong and 60% of

• Previous evaluations on the effectiveness of ribavirin and corticosteroids in treating severe acute respiratory syndrome were inconclusive.

• On the basis of a large dataset with 25% of global severe acute respiratory syndrome cases and properly adjusted for treatment selection bias, the combination of ribavirin and corticosteroids was found to have no significant beneficial effect in the initial treatment of patients with severe acute respiratory syndrome.

the patients in Toronto, 20 we used multiple imputation to make best use of all available data in the propensity score model. [27] [28] [29] [30] Ten datasets were constructed on the basis of all clinical, demographic, and outcome variables with imputed values based on a generalized additive model and predictive mean matching (function aregImpute in R library Hmisc). We used chi-square tests to evaluate whether the propensity scores were able to balance observed covariates across treatment groups. We then weighted the observed individual outcome by the inverse estimated propensity scores to infer adjusted case fatality ratios in each treatment arm. 26 We used case-resampling bootstrap with 1000 repetitions in each imputed dataset to estimate 95% confidence intervals (CIs) for the case fatality ratios. 31 For the primary analysis we focused on treatment choice within 2 days of hospital admission, because early treatment decisions were most likely based on patient characteristics at hospital admission as opposed to later treatment decisions that also would have been affected by the effects of initial treatments and the subsequent course of illness. We carried out sensitivity analyses to examine the effectiveness of treatment assignments within 1 to 5 days of admission. Further technical details of the methods are described in the Appendix. All analyses were conducted in R version 2.3. 1. 32 This study received ethics approval from the institutional review boards of the University of Hong Kong/Hospital Authority Hong Kong West Cluster, Hong Kong; McMaster University, Hamilton, Ontario; the University of Toronto, Toronto; and all Toronto-area hospitals where data were collected. Table 1 summarizes the patient characteristics in Hong Kong and Toronto, by treatment within 2 days of admission. More detailed comparison of the characteristics is given in Appendix Table 1 . There were statistically significant differences (P Ͻ .001) in all demographic variables across the 4 treatment groups in Hong Kong, whereas clinical variables were relatively more balanced across groups. For patients in Toronto, the distribution of characteristics was more balanced across treatment groups, especially in demographic variables, but there were greater differences in delay between symptom onset and admission, lactate dehydrogenase level, and neutrophil count (P Յ .05). In Hong Kong and Toronto, 301 (17.3%) and 25 (12.6%) deaths occurred in total, respectively. The crude case fatality ratios were higher in the patients untreated with ribavirin or corticosteroids within 2 days of admission (23.3% and 20.0%, respectively) compared with those treated with ribavirin and corticosteroids within 2 days of admission (12.6% and 12.8%, respectively). The crude combined treatment effect of ribavirin and corticosteroids in Hong Kong and Toronto, in terms of absolute risk reduction, was 10.7% and 7.2%, respectively. In Hong Kong, younger patients, health care workers, those with a longer delay between symptom onset and admission, and those with earlier calendar date of symptom onset were more likely to be treated with both ribavirin and corticosteroids (Appendix Table 2 ). In Toronto, patients with younger age, longer delay between symptom onset and admission, high lactate dehydrogenase level, or hazy chest radiograph were more likely to be treated with both ribavirin and corticosteroids. After adjustment for propensity scores, balance was achieved among all patient characteristics (test for heterogeneity across treatment groups not explained by propensity score, P Ͼ .24 for all covariates in Hong Kong; P Ͼ .10 for all covariates in Toronto). Table 2 shows the estimated case fatality ratio by treatment group in Hong Kong and Toronto based on the generalized propensity score weighting. By adjusting for patient characteristics, the model predicted that the overall case fatality ratio would have been highest (19.2%) if all patients in Hong Kong had been treated with both ribavirin and corticosteroids within 2 days of admission, whereas it would have been lowest (15.4%) if no Hong Kong patients had been treated with ribavirin or with corticosteroids within 2 days of admission. The difference in case fatality ratios was 3.8% (95% CI, Ϫ1.5 to 9.2, P ϭ .16), which suggests at most no effectiveness of combined therapy with ribavirin and corticosteroids, and is inconsistent with even a mildly beneficial effect of these treatments. The results for Toronto were generally consistent albeit with wide confidence bounds because of the relatively small sample size (Table 2) , and the adjusted excess case fatality ratios for ribavirin and corticosteroids versus neither treatment within 2 days of admission was 2.1% (95% CI, Ϫ44.3 to 48.5, P ϭ .93).

In sensitivity analyses, Figures 1 and 2 show the estimated case fatality ratios in Hong Kong and Toronto, respectively, for 5 separate analyses in which patients were grouped by the treatment they had received within days of hospital admission for d ϭ 1, 2, . . . , 5. For Hong Kong, the estimated case fatality ratios were consistent with the main results in Table 2 and generally stable across the different cutoffs, except for the small numbers of patients receiving corticosteroids only. Appendix Table 3 shows the estimated adjusted odds ratio of case fatality ratios among treatments, based on conventional multinomial logistic regression models on 10 imputed datasets. For Hong Kong, the estimated odds ratio is consistent with the main results in Table 2 , where patients treated with both ribavirin and corticosteroids had a higher case fatality ratio and the difference was marginally statistically significant. For Toronto, the results were not statistically significant with wide CIs.

During the 2003 severe acute respiratory syndrome outbreaks in Hong Kong and Toronto, ribavirin and corticosteroids were widely used, although no controlled studies of their effectiveness were performed because of the novelty of the causative agent. This study used propensity score methods to estimate the effectiveness of ribavirin and corticosteroids for severe acute respiratory syndrome early in the course of disease, adjusting for differences in patient characteristics between treatment arms. 11 For example, the most significant factors associated with severe acute respiratory 

The American Journal of Medicine, Vol 122, No 12, December 2009 syndrome fatality in Hong Kong are age more than 60 years old, preexisting comorbid conditions, and non-health care worker status, 19 ,33 all of which were more commonly present for patients not receiving ribavirin or corticosteroids ( Table 1 ). The high crude case fatality ratio for patients receiving neither treatment was likely because of this case mix that was fully adjusted in our propensity score analysis. 12 We did not find any evidence that early treatment with both ribavirin and corticosteroids was beneficial to patients in general, which is consistent with previous studies. [34] [35] [36] [37] By observing that there were higher case fatality ratios for the ever treated patients but lower case fatality ratios in patients treated within 2 days on hospital admission, it is possible that treatment decisions made later during hospitalization depended on whether patients' health status was improving or deteriorating rather than baseline clinical characteristics. Without detailed data on clinical progression, which we do not have, we cannot assess the potential benefits of therapies later in the clinical course and in particular as rescue treatment. It is always easier to act with the benefit of hindsight, and our results should not be taken as a criticism of those early prescribing decision but rather as an evaluation of treatment strategies based on information that has since become available. This also points to the need for common data-collection protocols and specification of a standardized dataset (or minimum dataset), which may facilitate integration of data during the initial period of epidemics and allow more comprehensive retrospective analyses after epidemics.

Early in the severe acute respiratory syndrome epidemic when the causative agent was still unknown, ribavirin, along with other broad-spectrum antibiotics, was used in Hong Kong and Toronto because of its in vitro activity against a broad spectrum of viral agents. On this basis, it was thought that ribavirin had the potential to suppress acute viral replication in the early phase of severe acute respiratory syndrome. 38,39 Early treatment with ribavirin was considered important because the viral load in patients with severe acute respiratory syndrome peaked at day 10 after symptom onset, 9 and the peak level was independently associated with mortality. 40 However, the in vivo inhibitory effect of ribavirin at clinically achievable doses remains controversial, significant adverse events are associated with ribavirin, 37,41 and our findings suggest that it is unlikely that ribavirin had any significant beneficial effect on patient outcome.

Corticosteroids were used to reduce pulmonary damage because of the immune-mediated inflammatory responses that usually started in the second week. 9,42 However, early corticosteroids treatment during the stage of viral replication may have suppressed the immune response and allowed a higher peak viral level. 43, 44 The clinical effectiveness of corticosteroid therapy has not been established for severe acute respiratory syndrome 45,46 or acute respiratory distress syndrome, 47 and corticosteroids use in severe acute respiratory syndrome was associated with aspergillus superinfection, avascular necrosis, reduced bone mineral density, and other long-term adverse outcomes. 48-50 LIMITATIONS A limitation of our analysis is that a propensity score model cannot adjust for unmeasured confounding variables and that the unconfoundedness assumption of propensity score models is inherently untestable. However, given that our analysis included and balanced the most important variables that have been identified as predictive of survival, 20,34,51 the best effort was made to control for potential confounders. Our data on treatment assignments are limited to information on the first date of ribavirin use and the first date of corticosteroid use; therefore, we have been unable to investigate alternative strategies of treatment administration, for example, whether corticosteroids were pulsed or not or variations in dosage or duration of treatment. We do not have data on other treatments, such as lopinavir/ritonavir, which was associated with improved clinical outcomes in a small uncontrolled study, 52 or traditional Chinese medicine. 53 The main analyses were based on treatment choice within 2 days of admission; however, results were consistent when treatment within 1, 3, 4, or 5 days of admission was considered. A further limitation is the presence of substantial proportions of missing data on some clinical variables. We implemented multiple imputation to use all available data in the analysis, which is considered the gold standard approach, 54 and the estimated treatment effects were consistent with those estimated in a complete case analysis (data not shown). Another limitation is that outcomes other than death were not considered; common side effects such as avascular necrosis due to corticosteroid therapy might have a significant impact on long-term quality of life, but such outcomes were not considered in this study. 48,49 A more complete comparison between treatments should consider both nonfatal adverse drug effects and quality of life. A complete dataset including disease progression and all clinical events (eg, ventilation) was not available. Finally, our analyses were based on patients with probable severe acute respiratory syndrome according to the World Health Organization definition. 55 We adopted this definition rather than laboratory confirmation because the latter definition may be potentially biased toward including more survivors, particularly in the early cases. 19

On the basis of a large sample from 2 epicenters, our results add strength to the hypothesis that the combination of ribavirin and corticosteroids has no significant beneficial effect in the treatment of severe acute respiratory syndrome. In the absence of further evidence, clinicians should not use ribavirin and corticosteroids to treat severe acute respiratory syndrome because to the best of our knowledge, they provide no benefit in terms of survival. Investigators should consider other promising therapies in the event of another outbreak of severe acute respiratory syndrome, and ideally protocols would be in place to evaluate alternative treatments in randomized trials. 

For the primary analysis that compares all 4 treatment choices within the first 2 days after admission, we applied a multivariable multinomial regression model to estimate the propensity scores for each of the 4 treatment alternatives, including variables that were likely to be influential in early treatment decisions. The latter included the following known prognostic factors: age, gender, health care worker, presence of comorbid condition, hospital, delay in admission, onset date, lactate dehydrogenase level, chest radiograph, oxygen saturation on room air, neutrophil count, platelets, and lymphocyte count, 20 and 3 other factors that may have been potentially important in treatment assignment, including date of onset, onset-to-admission delay, and hospital. Because some prognostic factors were only available for a subset of up to approximately 30% of the patients in Hong Kong and 60% of the patients in Toronto, 20 we used multiple imputation to facilitate inclusion of these factors in the propensity score model. 27 The resulting propensity scores were then used to weigh the case fatality ratios in each treatment group, allowing estimation of the mean causal effect, 26 that is, for each treatment group we estimated the overall case fatality ratios had all patients been assigned to that particular treatment. Let p it be the estimated propensity score for patient i with treatment t, D i be 1 for a death outcome for patient i, and I be an indicator function; then the propensity score-adjusted case fatality ratio for each treatment t is given by: 

World Health Organization. Summary of probable SARS cases with onset of illness from 1

World Health Organization. Situation updates -SARS. Available at: www

Civet cats found at restaurants again

SARS-CoV infection in a restaurant from palm civet

Coronaviruses in bent-winged bats (Miniopterus spp

Bats are natural reservoirs of SARS-like coronaviruses

The broad-spectrum antiviral ribonucleoside ribavirin is an RNA virus mutagen

Lung pathology of fatal severe acute respiratory syndrome

Clinical progression and viral load in a community outbreak of coronavirus-associated SARS pneumonia: a prospective study

Critics slam treatment for SARS as ineffective and perhaps dangerous

SARS: systematic review of treatment effects

Clinical trials and novel pathogens: lessons learned from SARS

The central role of the propensity score in observational studies for causal effects

Efficacy of oral beta-lactam versus non-beta-lactam treatment of uncomplicated cellulitis

The risk associated with aprotinin in cardiac surgery

Effect of pretreatment with statins on ischemic stroke outcomes

The role of aggressive corticosteroid therapy in patients with juvenile dermatomyositis: a propensity score analysis

Selective cyclooxygenase-2 inhibition and cardiovascular effects: an observational study of a Medicaid population

The epidemiology of severe acute respiratory syndrome in the 2003 Hong Kong epidemic: an analysis of all 1755 patients

Clinical prognostic rules for severe acute respiratory syndrome in low-and high-resource settings

Early diagnosis of SARS: lessons from the Toronto SARS outbreak

Public health measures to control the spread of the severe acute respiratory syndrome during the outbreak in Toronto

Covariate adjustment in randomized controlled trials with dichotomous outcomes increases statistical power and reduces sample size requirements

A simulation study of the number of events per variable in logistic regression analysis

The role of the propensity score in estimating doseresponse functions

Propensity score estimation with missing values using a multiple imputation missingness pattern (MIMP) approach

Using multiple imputation and propensity scores to test the effect of car seats and seat belt usage on injury severity from trauma registry data

A 1-year prospective cost-effectiveness analysis of roflumilast for the treatment of patients with severe chronic obstructive pulmonary disease. Phar-macoEcon

Maternal employment and child development: a fresh look using newer methods

We thank colleagues in the Department of Health and Hospital Authority involved with the public health control of and clinical care associated with the severe acute respiratory syndrome epidemic and for data collection and processing.

OR ϭ odds ratio; CI ϭ confidence interval. a Only 3 patients were treated with corticosteroids only in Toronto and were excluded. b OR represents the relative odds of the corresponding treatment choice versus no treatment, relative to the reference group. An OR of 0.84 for men under "ribavirin only" means that men has 16% less odds than women to be treated with ribavirin only instead of no treatment. c Intercept represents the baseline relative odds of the corresponding treatment choice versus no treatment. Appendix