RAPID AWARENESS AND TRANSMISSION OF SEVERE ACUTE RESPIRATORY SYNDROME IN HANOI FRENCH HOSPITAL, VIETNAM

INTRODUCTION

Notwithstanding the announcement of containment by the World Health Organization (WHO) in 2003,¹ severe acute respiratory syndrome (SARS) has remained a matter of concern worldwide, and it is not surprising that several cases of SARS have reemerged, for example, in China in April 2004.² Although the mode of transmission remains partially unclear, especially with regard to airborne transmission³ and super-spreading events,^4,5 it appears to occur predominantly by large droplets, direct contact with infectious material, or contact with fomites contaminated with infectious material.^6,7 The most effective containment measures identified to date include the tracing of contacts,⁸ quarantine,⁹ triage and early case detection,^10,11 and isolation.¹² Further, because the close contact required for transmission easily occurs in hospital settings,^13–15 nosocomial spread was determined as one of the major epidemiologic features of SARS.^7,16,17 The elimination of hospital transmission through enhanced infection control practices is therefore a crucial control measure.

An early study in Hong Kong showed that the practice of droplet and contact precautions was adequate in most clinical settings in significantly reducing the risk of infection after exposure to patients with SARS,¹⁸ and if practiced by a high proportion of susceptible individuals, precautionary measures are expected to significantly reduce transmission.¹⁹ The adoption of routine preventive behaviors based on appropriate training and control among health care workers (HCWs), undertaken prior to the isolation of SARS patients, was shown to be one of the most crucial control measures.^20–22

In this context, Vietnam is considered to have achieved the first highly successful containment of SARS during the early phase of the outbreak.²³ One reason for this rapid containment is thought to be the prevention of infection leakage from hospitals back into the general community.²⁴ A second is the successful discontinuation of the chain of nosocomial transmission several days after onset based on the radical control measures of the Ministry of Health, Vietnam.²⁵ Although several nosocomial transmissions were observed in Hanoi French Hospital (HFH) in the early days of the outbreak,^26,27 none were identified in HFH or another local hospital in the latter phase.²⁸ In both hospitals, staff instituted stringent precautions, strict isolations, and quarantines under the encouragement of Dr. Carlo Urbani (Dr. Urbani died of SARS before seeing the success of the containment).²⁹ We therefore consider that a comprehensive understanding of the successful containment measures adopted by HFH and their theoretical underpinnings are crucial to the success of control strategies for any future recurrence. Here, we use a case-control study design to time-dependently examine the relationship between SARS and the precautionary behaviors undertaken by those exposed in HFH. We then use mathematical approaches to develop intuitive analyses of the impact of individual behaviors on the control of a SARS epidemic.

MATERIALS AND METHODS

RESULTS

Table 2 shows the univariate association between the pre-cautionary behaviors taken (SARS and non-SARS [control] cases) in Stage 1 and SARS. The use of masks (P = 0.011) and gowns (P = 0.012) appeared to prevent infection, whereas handwashing and the use of gloves were less likely to provide protection. Only two subjects who performed all protective measures developed symptomatic infections (P = 0.059). Forward stepwise logistic regression of the five protective measures (0.05 for entry and 0.10 for removal probability) showed that only the use of masks was significant in the final model (OR, 0.29, 95% CI; 0.11–0.73, P = 0.009). In Stages 2 and 3, the use of masks (P = 0.001) and gowns (P = 0.010) was significantly associated with non-infection among doctors and nurses still not infected after Stage 1 (Table 3). Most performed all the personal protective measures recommended, and only one individual who wore masks was infected. The comparative results of the behaviors of all participants at Stage 1 and after entering Stage 2 are shown in Figure 1a. The proportions of individuals who performed the investigated protective behaviors increased after entering Stage 2. However, these behavioral changes were not significantly different between the two phases (P = 0.960). The behaviors performed by the doctors and nurses (N = 48; Figure 1b) who had the closest contact with the SARS patients drastically and significantly improved after entering Stage 2 (χ² = 9.855, P = 0.043).

The univariate associations between sociodemographic variables and SARS throughout the epidemic are shown in Table 4. Females were more likely to become infected than males (P = 0.011), and a significant association of SARS with nurses (P = 0.008) was observed. In HFH, infection was frequent in the 40–49 age strata (P < 0.015). Among all study subject, relatives of patients (P < 0.001) appeared to be the least frequently infected. Table 5 shows the interaction between the use of masks and other significantly associated variables in univariate analyses. Even though we saw no significant difference in the OR of using masks versus the use of gowns, females (OR = 0.2) and nurses (OR = 0.1) were more effectively protected by the use of masks than others in Stage 1. In Stages 2 and 3, the use of gowns showed overall reasonable OR (= 0.2), whereas most other interactions could not be calculated due to the scarcity of cases.

Figure 2a shows the mean and corresponding 95% CI of the trajectory (shown as prevalence) of an epidemic from 250 simulation runs which hypothetically assumed unchanged coverage as well as the protective effects of the precautionary measures observed in Stage 1. The precautionary measure in this simulation was based on a multivariate logistic regression which included all variables showing significant associations in univariate analyses, and focused on the impact of the use of masks, given the identification of this behavior as the most important protective measure (β = 0.6 obtained from OR = 0.4, P = 0.020). The coverage of masks was obtained as 52.0% from Table 2. If an outbreak was simply allowed to continue growing under these conditions, the results showed that approximately 50 to 90 symptomatic cases would occur by Day 30. The reproduction number (R) was estimated as 4.1 (95% CI; 1.9–6.4), and from this estimate the basic reproduction number was estimated as 6.0. Sensitivity of the cumulative number of cases to the coverage of masks, in the mean field, is shown in Figure 2b. Certain reduction in the cumulative number of cases was observed with significant improvements in coverage.

Figures 2c and 2d shows the outbreak trajectory of 250 simulations assuming improved coverage (from 52.0 to 81.5%) among susceptible individuals on Day 8 and restriction of contact with symptomatic individuals to health care workers on Day 13. The protective effect obtained from multivariage regression was 0.9 (OR = 0.1, P = 0.955). The reproduction number in Stage 2 was estimated as 0.7 (95% CI; 0.0–2.3). The number of incubating individuals began to show a decreasing trend after these events (Figure 2c), followed by a declining trend in the number of symptomatic cases (Figure 2d). Most of the simulated outbreaks eventually declined to extinction before Day 120. The sensitivity of the final size of an epidemic, evaluated through observations of the cumulative numbers of cases, to the timing of drastic changes in protective behaviors accompanied by strict isolation is shown in Figure 2e. When the stochastic effects are taken into account together with the effects of single precautionary measures and isolation, the rapid implementation of combined measures reduces the number of transmissions and increases the probability of extinction.

DISCUSSION

The findings of this case-control study indicate that the use of masks was significantly associated with the prevention of SARS transmission and that precautions against droplet contamination and contact were adequate in preventing transmission; this implies mainly to in-hospitals. The results are roughly consistent with those of previous reports.^18,20,22 Although a number of exceptions were seen with regard to protective effects during patient intubation, during which transmission to staff occurred even when droplet and contact precautions were taken,^7,34 one of the most important lessons from the SARS outbreak is the need to enhance infection control programs in hospitals.^13,35 Even though the use of masks was the most effective precautionary measure, masks alone together with the observed coverage did not reduce the reproduction number below unity (R₀ = 6.0 and R with the protective effects of masks = 4.1). Put simply, the use of masks alone was shown to be insufficient to contain the epidemic. Further, it was shown that the coverage of precautionary behaviors among the study subjects increased with the progression of the outbreak, and this was especially obvious among doctors and nurses. In HFH, remarkable changes occurred in the very early phase of the outbreak before detailed information about SARS was available. According to the stochastic simulations, an increased probability of extinction would be observed if the combined measures of precaution and isolation were rapidly implemented.

With regard to sociodemographic variables, females were more frequently infected than males. Given that transmission was most frequently observed among nurses, a plausible explanation for this finding would be occupational background. Although the 40–49 age group was frequently infected, we have no persuasive explanation for this apart from occupation: 61.9% of this stratum was medical doctors or nurses. Considering that nurses were more effectively protected from transmission by the use of masks, the control measures taken by them within HFH from early in the epidemic were admirable. The lowest frequency of infection was seen in relatives of patients, showing that our study included many relatives who remained uninfected but were nevertheless believed to have had contact. Because nonmatched case-control designs such as this are vulnerable to selection bias, we obtained estimates of the protective effect of masks by means of multivariate logistic regression analysis which entered all other variables significantly associated with infection in univariate analysis. After adjustment for internal confounding variables, the estimated reproduction number was given as 0.7 in Stages 2 and 3. Previous studies have shown that the (effective) reproduction number, defined as the average number of secondary cases generated by one index case in a susceptible population under certain restrictions and interventions, decreases with increasing awareness of the epidemic combined with several public health measures.^36,37 Using reasonable estimation procedures, another study showed that R significantly decreased after a global alert in most affected countries.³⁸ The current study showed that the estimated R decreased below unity after notification of a hospital outbreak, although the estimates were obtained using rough assumptions and the process of estimation was biased by various factors.

In HFH, the rapid increase in awareness, which led to not only strengthened precautionary measures and isolation but also quarantining of health care workers, seems to have been the greatest contributor to successful containment. One reason for this quick response could be attributed to the background of secondary cases that arose mainly from health care workers who had close contact with the index case. Almost all staff members working or on duty in the earliest days of the outbreak (in Stage 1) were severely infected.^39,40 Another reason might be due to the efforts led mainly by Dr. Carlo Urbani, who suggested quick improvements in the precautionary measures taken and isolation.²⁹ As a result, transmission leakage into the community was prevented, thus having a huge impact on the chains of transmission.²⁴ In HFH, those who were exposed implemented precautionary and other controlling measures quickly and efficiently, and the epidemic consequently declined to extinction.

In the interests of objective interpretation, the limitations of our study design must be addressed, as follows:

1. A study such as ours in which exposure has a strong intuitive causal link with outcome (i.e., mask usage) is vulnerable to recall bias. Even though we limited our subjects in Stages 2 and 3 to medical doctors and nurses, and cases were appropriately selected according to the probable date of infection and incubation period, our estimates are likely less accurate than would be obtained by blinded or matched case-control study. In addition to this directional bias, further bias may have been introduced by random misclassification, as our records were completed 1 year after the outbreak, and it is therefore possible that some of the precautions were uncertain exposures. The frequent use of masks among controls may have reduced the strength of the associations.

2. Model-generated results must be interpreted cautiously. Although the simulations shown here included only the effect of masks and were considered according to the results of multivariate logistic regression adjusted for internal factors, unknown external confounding factors likely exist. For example, in Stages 2 and 3, although multivariate logistic regression was performed with other variables, the P value obtained was 0.955, and overall the model was weak. Owing to the scarcity of case records, stratification in this stage failed to separate the effects of masks. Thus, the estimates of the protective effect of masks and reproduction number in this stage may include the effects of other concomitant changes, such as the reduced frequency of contacts and quarantine.

3. There are limitations concerning the simplicity of our model; for example, we neglected the possible differential susceptibility of humans to asymptomatic infections,^41,42 individual variance in severity and/or prognosis,^23,43,44 and the highly heterogenous transmission of SARS.^4,5,45 Theoretical exercises never replace reality.

4. Finally, because our model was based on a case-control study, the estimates of coverage were biased; principally, coverage in a case-control design is taken from a nonrepresentative sample. Although this study was conducted as a first attempt to incorporate the effect of behavioral factors, which change time-dependently, to model building strategies for the control of directly transmitted airborne diseases, further studies incorporating a number of methodological improvements are required.

In conclusion, given that early recognition that leads to the implementation of protective behaviors and effective control strategies is crucial in hospitals,⁴⁶ we believe our model provides intuitive results that at least partly satisfy the need to evaluate outbreak trajectories based on individual behaviors.

APPENDIX

Each simulation starts with one index case and is based on a model constructed as follows:

1. The expected number of people who used protection on each subsequent day was determined by the number of susceptible individuals (S), number of contacts per day (κ), proportion of individuals who performed the protective behavior (p), and the protective effect of the precautionary measure (β), which were obtained based on our survey. The number of infectious contacts, denoted by the product of the number of susceptible individuals (S) and the mean number of contacts (κ), was divided into two subgroups: one that represents protection due to precautionary behaviors against infection with SARS-CoV (SARS-associated coronavirus) and another that does not, according to (1 - pβ). However, these groups were not permanently fixed. The mean of the number of contacts based on our survey was approximated by:

   $\text{κ}={\text{κ}}_1{\text{π}}_1+{\text{κ}}_2{\text{π}}_2={\text{κ}}_1{\text{π}}_1+{\text{κ}}_1\ln \left({\text{OR}}_{\text{close}}\right){\text{π}}_2$(A1)

   where κ₁, κ₂, π₁, and π₂ denote the number of casual and close contacts and the fraction of individuals who had casual and close contacts, respectively, while the odds ratio of getting infected with close contact is represented by OR_close and N, respectively.

2. Both the incubation (E) and symptomatic (I) periods were assumed to be independently and identically distributed following an approximated probability density function with gamma distributions³³ (denoted by γ_k and c_l for the discretized stages [days] k and l, respectively). We divided the probability density functions into k (i = 14) and l (j = 12) stages; the methodology of approximation by date was previously reported.²⁴ The relative measure of infectiousness for the incubation (E) period (q) was assumed to be 0.1.¹²

3. Based on realistic settings in Vietnam, it was assumed that all individuals were isolated with the onset of early signs of clinical symptoms under the isolation measures; and for simplicity, the effect of quarantine was neglected. When considering strict isolation, the number of susceptible individuals having contact with SARS patients was limited to 20 (which is the approximate number of ward workers); the number of susceptible individuals was treated as being stable (always S = 20) so that S would not be exhausted thereafter; without isolation there were assumed to be 300 susceptible individuals (which is roughly the total number of people involved in possible contacts in HFH). N = S + E + I + R, and background mortality was neglected. The resulting simplest difference equations were formulated as follows:

   $\begin{array}{c}S(t+1)=\exp \left[-\text{κ}(1-p\text{β})\frac{I+qE}{N}\right]S(t)\\ E_1(t+1)=\left\{1-\exp \left[-\text{κ}(1-p\text{β})\frac{I+qE}{N}\right]\right\}S(t)\\ E_k(t+1)=\left(1-{\text{γ}}_{k-1}\right)E_{k-1}(t)\\ I_1(t+1)={\displaystyle \sum _{k=1}^i{\text{γ}}_k}{E}_k(t)\\ I_l(t+1)=\left(1-c_{l-1}\right)I_{l-1}(t)\\ R(t+1)=R(t)+{\displaystyle \sum _{l=1}^j{c}_l}{I}_l(t)\end{array}$(A2)

   Based on the forward stepwise logistic regression result in the case-control study, and to facilitate understanding, p and β were used only to represent the use of masks. However, the protective effect, β, was obtained from the result of further multiple logistic regression which entered all other significantly associated variables (in univariate analysis). All terms shown here as products of a probability and a state variable were generated in our simulations by using random variables with binomial distributions. Under these assumptions and using mean length of incubation and symptomatic periods, the reproduction number (R) is given by: q 1

   $R=\text{κ}(1-p\text{β})\left(\frac{q}{\text{γ}}+\frac{1}{c}\right)$(A3)

   where γ⁻¹ and c⁻¹ are the means of the incubation and symptomatic periods in days, respectively. The basic reproduction number was estimated by

   $R_0=\frac{R}{(1-p\text{β})}$(A4)

   For the purpose of mathematical convenience, although unrealistic, our model assumed homogenous mixing as well as all infectious individuals being equally infectious.