Background
Many countries have developed pandemic preparedness plans in response to the threat from pandemic influenza [
1], to attempt containment of the virus or to reduce the pandemic's impact. The influenza A (H1N1-2009) pandemic has underscored the importance of such plans, with the World Health Organization (WHO) calling for the activation of pandemic plans worldwide [
2]. Although the WHO has made public guidelines for developing pandemic plans [
3], the comprehensiveness and standards of pandemic plans differ widely across different countries and continents [
4‐
6]. To ensure the success of these plans, it is necessary to adopt a combination of different strategies.
Although there are existing historical data on the possible success of strategies used in previous pandemics such as personal hygiene, school and workplace closures, and social distancing, these are often anecdotal and difficult to interpret [
7,
8]. Mathematical models provide a platform for the assessment of multiple interventions in an environment where individual parameters can be altered. The recent increase in mathematical modeling studies on pandemic interventions suggests the effectiveness of these strategies and provides guidance for policy makers. Although the 2009 pandemic has spread rapidly, these combination strategies can be applied in populations yet to be severely affected, for the second wave, or for the next pandemic [
9,
10]. This systematic review aims to determine the individual components that constitute combination strategies, and the quantitative impact of these combination strategies in reducing pandemic spread and morbidity.
Methods
This study explored available mathematical modeling publications on the effectiveness of combination strategies for an influenza pandemic. To obtain papers on the effectiveness of combination strategies, data for this review were identified by the authors through searches of the PubMed search engine for English language articles and articles translated into the English language. The authors used the following search terms to focus on modeling studies, and those which had a focus on pandemic preparedness and strategies - influenza and pandemic and (preparedness or strateg* or model*); influenza and modeling or modelling. The search included all published articles listed on PubMed from 1990 to 30 September 2009 - there were few articles on influenza pandemic planning or modeling before this period.
Abstracts were reviewed where available to determine if a study met the inclusion criteria and the full manuscript was obtained for further scrutiny. Snowball searches by hand were performed on the reference lists of articles meeting the inclusion criteria to find additional studies.
The inclusion criteria were primary mathematical modeling papers that compared and reported the quantitative effectiveness of combination strategies (two or more strategies used together) versus individual strategies for human pandemic influenza. Mathematical modeling papers were those which used quantitative predictive methods to determine the likely impact of strategies, and had descriptions of these methods which could be reproduced or verified. All influenza preparedness strategies were considered, including pharmaceutical and non-pharmaceutical public health strategies. These articles would allow clear comparison on the advantages of combination strategies over and above the impact of individual strategies. An explanation of some of the key strategies are found in the appendix.
Mathematical modeling articles that described the effectiveness of multiple singular strategies but did not analyze the quantitative effect of combination strategies were excluded. Articles that referred to general pandemic preparedness without quantitative evidence, or provided only qualitative discussion were also excluded. Reviews without primary data, articles in abstracts without full publication, and unpublished studies were excluded as their methodology and results could not be verified.
Mathematical models are based on input variables which are assumptions made based on available evidence in specific scenarios. One important assumption is the reproductive number (Ro), which is the average number of secondary infections generated by a single case in a completely susceptible population. No attempt was made in this review to homogenize data across studies for comparison; on the contrary, the heterogeneity of data provides public health professionals with evidence of the effectiveness of strategies across a wide range of assumptions and scenarios. We have instead listed the different types of models used, and the scenarios, interventions, and countries where they were applied.
Discussion
Many modeling studies were performed as a result of H5N1 influenza threat and an impending pandemic, but all have used parameters based on historical pandemics and existing studies on the influenza transmission. In addition, these studies provided sensitivity analyses across a wide range of influenza parameters. As such, they are directly relevant to the 2009 influenza pandemic which has an Ro of between 1.2 to 1.6 [
28], similar to the 1957 and 1968 influenza pandemic [
16], and for future pandemics. At the same time, the 2009 influenza pandemic provides the opportunity to study unknown variables to validate and refine these models.
All of these modeling studies in various settings, and using different models and assumptions, consistently show that combination strategies are more effective compared to individual strategies. Given the lack of good experimental, observation or controlled studies on these strategies, and the difficulties of performing trials during a pandemic, it is difficult for policy makers to know the effectiveness of their policies. These modeling studies provide policy makers with a suggestion of the effectiveness of different combination strategies. At the same time, new models will have to be developed using local data to provide realistic outcomes for local settings. The diverse methodology available from these studies provides sufficient information for countries to build and validate their results locally.
Although the use of individual-based and other stochastic models provide better data resolution, deterministic models mentioned in this review show similar outcomes [
18,
22,
23,
27]. These deterministic or simple stochastic compartmental models are much easier to build and may provide rapid results for policy making. This is especially true in countries where the vast amounts of data required for individual-based and complex stochastic models may not be available compared with high-income countries where most sophisticated models were built.
The use of combination strategies necessitates the availability of resources and feasibility for each individual component. For example, stockpiling of pharmaceutical agents is an integral part of preparedness plans and currently widely adopted in well-resourced countries. The increase in anti-viral drug resistance underscores the importance of combination drug use and provides policy makers with recommendations for their stockpiles [
15]. Combination stockpiles of sufficient amounts of different antiviral drugs such as oseltamivir, zanamivir and adamantanes will allow for early combination chemotherapy or sequential multidrug therapy which was modeled to be effective against antiviral resistance when a small secondary stockpile was used to augment a primary stockpile [
15]. The United States Federal stockpile is composed of 80% oseltamivir and 20% zanamivir, and several million doses of rimantadine from previous stockpiles [
29]. The United Kingdom has purchased additional antiviral drugs to ensure it has a total stockpile for 50% of its population, comprising 68% oseltamivir and 32% zanamivir [
30]. Bacterial pneumonia results in substantial morbidity and mortality among pandemic influenza cases [
31,
32]. Antibiotics should therefore be considered for stockpiling [
31]. Stockpiles should take into account common locally circulating bacteria, and recommended amounts range from 10 to 25% of the population [
33]. In contrast to antiviral drugs that are not widely used, antibiotics can be part of a rolling stockpile which ensures sufficient stockpiles without expiry issues. Vaccination against bacterial infections should likewise be considered.
From the effectiveness of combination strategies in reducing global spread of influenza or resistant viruses [
12‐
15], resource-rich countries should consider redistributing their resources for the greater global benefit and their own benefit if they have yet to be affected by the pandemic. Controlling local outbreaks through combination strategies can reduce global spread, and countries affected early during the pandemic should be provided with assistance [
13].
Vaccines are part of many combination strategies and modeling has shown that introduction of a vaccine four months after the pandemic virus has arrived has limited effectiveness, while stockpiling prototype pandemic vaccines could reduce overall AR [
16]. Therefore countries were stockpiling H5N1 vaccines as candidate pandemic vaccines [
34,
35]. However, if the pandemic influenza virus is totally different from the vaccine virus, the vaccines would be of negligible effectiveness. Investments are needed to develop new vaccines with greater cross-protection against conserved viral regions; vaccine libraries to quickly produce candidate vaccines; better adjuvants and antigen-sparing strategies to increase production capacity; and modes of administration for improved immunogenicity and cross- protection [
36,
37].
Although some individual strategies may seem very effective, they may not be feasible and models assist policy makers in avoiding potentially disastrous decisions. Social distancing has been widely used in epidemics [
7] but their impact remains unclear and highly dependent on disease severity, transmission, and risk groups affected. Local interventions such as school closures may be effective if done early, decisively, and for prolonged periods [
20,
38‐
40]. A United Kingdom model based on a 1957-like pandemic showed more than 20% case reduction if the Ro were low (<2) and schools were closed early, but less than 10% case reduction in pandemics with high Ro [
38]. A French study showed that prolonged closure and limiting contact among children outside school may reduce cases by 17% and peak AR by 45% [
39]. However, school closures and limiting social contact may be socio-economically difficult to achieve. Another study found that total closure of schools and workplaces reduced AR by 95%. However, the socio-economic impact would be unimaginable [
20]. Similarly, most modeling studies found that travel restrictions alone did not impact overall AR [
13,
16,
19,
23]. Reducing air travel has been modeled to be effective in delaying pandemic spread if nearly 100% reduction can be achieved [
13,
16], and will be difficult if not impossible to achieve [
41]. If used alone, local epidemic severity may increase because restriction-induced travel delays can push local outbreaks into high epidemic season [
14].
Although combination strategies are more effective than individual measures, not all combination strategies may be feasible. Active surveillance, isolation of cases, and quarantine of close contacts are important interventions during epicenter containment. These interventions may reduce the Ro of the disease to below one and contain the outbreak. However, it is often difficult to ensure total compliance with these measures and if used alone, will result in missed cases due to surveillance failures, isolation facility exposures, and quarantine failures as shown in the SARS experience [
42]. A Hong Kong modeling study found that although contact tracing and quarantine of all contacts was effective, it was not feasible because the number of people under quarantine would be excessive [
24]. Therefore combination strategies enable policy makers to leverage on the effectiveness of some measures and reduce potential negative impact of others.
For combination strategies to work, they have to be tailored for each scenario at organizational, community, national, and international levels. To facilitate integration of interventions into effective combination strategies, more evidence is needed through targeted research, for example, the effectiveness of non-pharmaceutical interventions (e.g. personnel cohorting, school closures or reduction in air travel). In the absence of definitive studies, mathematical modeling studies provide an effective means of assessing the effectiveness of these strategies.
A limitation of this study is the restriction of our searches to the PubMed database. While we have made attempts to include additional articles from snowball searches, there is the potential for other published or unpublished studies to be missed from other databases and private sources. Other intrinsic limitations of modeling studies exist, and include the fact that they are based on theoretical epidemiology and not fully based on clinical or epidemiological evidence. For example, widespread use of pandemic vaccines raises safety concerns, and widespread use of antiviral drugs raises concern for antiviral resistance. Viral transmission during treatment with anti-viral drugs is also not well understood. It is therefore important to perform clinical and epidemiological studies during pandemic or seasonal influenza to understand the effectiveness and impact of these interventions. Models are also highly dependent on the assumptions and input variables, and are specific for a local context. However, if these limitations are understood by decision makers, modeling provides a reflection of the possible outcomes, helps to delineate possible strategies for inclusion, and avoids costly errors.
Competing interests
VJL has received unrelated research support from GlaxoSmithKline. AWS has received honoraria and reimbursements to attend conferences by GlaxoSmithKline, Novartis, and Sanofi Pasteur.
Authors' contributions
VJL contributed to the design, data collection, and manuscript writing. DCL and AWS contributed to the design and manuscript writing of this study.