Discussion

Analysis of ILI activity data from 19 countries revealed important pan-European similarities in pandemic influenza outcomes when compared with seasonal influenza, but also highlighted major and unexpected heterogeneities. In particular, Norway, the UK and Greece experienced increases in total cases compared with prepandemic years that were much larger than those seen elsewhere. In the UK this occurred despite the introduction of the National Pandemic Flu Service in July 2009 and recommendations for most patients with influenza-like illness not to consult their general practitioners, a policy that is likely to have resulted in an anomalously low number of reported ILI cases in the second wave.

We found no evidence that the increased impact of the 2009–2010 pandemic in some countries relative to prepandemic years could be explained by shortcomings in national strategic pandemic plans, nor that a reduced impact could be explained by greater completeness of plans. This was true for aggregate measures of planning, specific planning activities that might plausibly be expected to impact a pandemic, and also for less plausible associations. Where evidence of associations were found, in all cases these pointed to improved planning being associated with worse ILI outcomes. While causal mechanisms for such relationships cannot definitively be ruled out, they are not highly plausible. With 88 tests for correlations we would expect between four and five to show significant associations at the 5% level by chance alone. The five significant associations we found are therefore fully consistent with chance.

The introduction of measures that reduce influenza transmission can, however, under some circumstances, lead to earlier peaks and increasing the effectiveness of transient social-distancing measures can sometimes paradoxically lead to increases in total case numbers.16 However, if such mechanisms were operating positive correlations between the change in ILI cases and the pandemic planning components public health interventions and antiviral use planning might be expected. In both cases we found only weak and non-significant associations, and these only in the 19-country analysis. When the anomalous ILI data from Lithuania and Romania were removed only planning for essential services was found to be significantly associated with influenza outcomes. Since this category would only be likely to affect ILI outcomes in a severe pandemic for which morbidity and mortality were large enough to threaten the maintenance of essential services, this association seems very unlikely to reflect a causal link. Moreover, our analysis suggests that less complete planning for essential services was associated with public health benefit.

Another possible explanation for the correlations is that countries varied in their propensity to identify pandemic ILI cases relative to seasonal cases. While our analysis allows for national variation in ILI surveillance coverage, this variation is assumed to be constant over time. A greater likelihood of detecting ILI cases in pandemic relative to non-pandemic years in well-prepared countries could plausibly account for some of the associations seen. This could happen, for example, if countries with improved public communication systems encouraged possible cases to seek medical attention.

While these findings do not preclude important beneficial effects of planning activities on influenza outcomes, they do indicate that the marked variation in completeness of pandemic plans across Europe is unable to explain most of the diversity in pandemic outcomes. In particular, no evidence was found that improved planning was associated with reductions in ILI cases, reduction in the ILI peak height, or delays in the pandemic peak. Given the body of theoretical and empirical evidence that a number of planning components, if implemented, can substantially reduce transmission,17–20 and the large variation in these planning components across the 17 countries, these results are surprising. However, they hold true both when planning scores are considered in aggregate and also when themes that plausibly impact directly on transmission are analysed. There are a number of possible explanations: first, effective components of pandemic plans may not have been implemented; second, components of pandemic plans may have been implemented but could have been less effective than theoretical studies suggested; third, it is possible that country-wide variation in pandemic outcomes was dominated by factors unrelated to control measures such as demographics and prior immunity. It is also possible that the summary measure of planning completeness we have used fails to reflect important heterogeneities in the effectiveness of specific components. A measure of planning completeness that gives higher weights to more effective components may have yielded different results. Assigning such weights, however, would be problematic.

Our study has some limitations. First, we could retrieve sufficient data from only 19 countries including only 17 of the 27 EU member states. Second, correlation is not causation and, equally, lack of correlation is not lack of causation. A causal link between pandemic planning and H1N1pdm outcomes is possible, but may be masked by variation in environmental, demographic, behavioural and genetic factors that might have had larger impacts.2 ,21 ,22 These merit further investigation. Second, we analysed ILI data rather than laboratory-confirmed H1N1 pandemic cases because the quantity of laboratory-confirmed cases was insufficient to enable meaningful comparisons. Standardised data on related outcomes such as severe pneumonia and attributable mortality are not available across European countries. Although ILI is an imperfect outcome that may be affected by changing patterns of healthcare-seeking behaviour, and case definitions vary across Europe,23 we found that in most countries it peaked at the same time as the proportion of ILI cases that were laboratory-confirmed, a result consistent with prepandemic findings. This provides reassurance that ILI peaks are not simply artefacts of the surveillance system. Additionally, to enable meaningful between-country comparisons robust to the vagaries of national ILI reporting, we considered only summary measures relating to peak timing, or changes in ILI outcomes relative to prepandemic years. This should largely prevent distortions attributable to marked country-level variation in ILI reporting systems. Finally, we considered only pandemic plans and not implementation and therefore our analysis only addresses the stated research question which concerns the relationship between plans and outcomes. It is currently unclear how a documented intention to implement control measures is translated into practice and in the absence of such detailed implementation data we cannot address important questions concerning the relationship between plans and implementation, and implementation and outcomes. Evaluating the implementation of pandemic plans is an important area for future research, one that others are attempting,3 and it will be important to evaluate both effectiveness and cost-effectiveness of specific components of pandemic plans, such as antiviral use,24 vaccination25 and social distancing measure.16 Model-based analyses may help in this matter and also enable us to begin to make quantitative assessments of the importance of contact patterns (and the effect of school closure on these), and atmospheric conditions.26–28 Model-based analysis may also help overcome deficiencies in surveillance data: two reconstructions of the pandemic within the UK suggested a very different pattern to that seen in the crude ILI data (figure 1), and suggested that the second peak was comparable in size to the first.27 ,28 However, a sensitivity analysis (data not shown) found that our overall conclusions would not have changed had we considered the second UK peak to be the major one.

Drawing lessons from the 2009 influenza pandemic is challenging and likely to be the subject of research for years to come. One seems clear, however: for lessons to be drawn through comparative analyses across countries, surveillance data need to be robust, timely and systematic. Our research highlights the lack of complete or consistent influenza surveillance data across Europe,29 and the absence of in-built mechanisms for evaluating the effectiveness or otherwise of pandemic plans. Given that interventions have the potential to make outcomes worse as well as better, and that responses to previous pandemics appear to have been far from optimal,17 this is concerning. A pandemic surveillance model that combines surveillance for non-specific syndromes (such as ILI) with consistent laboratory testing for a subset of cases has been advocated.30 As well as conserving resources when surveillance systems are under considerable stress, this has the potential to facilitate comparative analyses and provide greater lesson-learning opportunities.

Finally, to help ensure the international response to future pandemics has a sound evidence base that effectively draws upon past experience, we suggest that future pandemic plans should consider how data on implementation of planning components can be recorded and shared, and how formal assessments of the effectiveness of interventions can be made.