Compared with COVID-19, the influenza epidemics is a persistent public health challenge that had occurred many times in human history, but there has been a lack of guidance on formulating NPI strategies to cope with influenza epidemics. To the best of our knowledge, the current study is the first to use a machine learning algorithm combined with explainable artificial intelligence tool to quantify the effectiveness of 14 NPIs targeted to COVID-19 in suppressing influenza transmission in 33 countries. We found that the gathering limitation made the biggest contribution in suppressing influenza transmission. Additionally, we also estimated the effectiveness of each level of NPIs, the intensity and time threshold for each NPI to take effect, and the interaction between each pair of NPIs.
We found that gathering limitation, school closure, contact tracing, and health education promotion contributed the most in suppressing influenza transmission. These results partially agree with previous studies on estimating the effectiveness of NPIs against COVID-19.16,17 We found that school closure was more effective than workplace closure. The reason might be that the transmission of the influenza virus mainly occurs in children rather than adults. Previous studies also found that closing school played an important role in the 2009 H1N1 pandemic.18–20
We found that contact tracing generated effects on suppressing influenza transmission even though this NPI was specific to the contact of COVID-19 patients. It might be because the disease symptoms of the two viruses are very similar (e.g., fever, cough), and both viruses are airborne transmitted. Therefore, patients with COVID-19 exposure history might also be previously or currently exposed to the influenza virus, and those patients with suspected COVID-19 symptoms under close tracing and monitoring might actually be infected with the influenza virus. Another reason might be the high co-infection rate of SARS-CoV-2 and influenza virus. A single-centered retrospective study conducted in Wuhan, China reported that among 307 COVID-19 patients, 57.3% of them were also positive for influenza viruses.21 A recent experimental study found that influenza A virus pre-infection could significantly enhance the infectivity of SARS-CoV-2.22 Similarly, other NPIs such as workplace closure and health education promotion were also effective in suppressing influenza transmission regardless of whether the NPI was restricted to regions suspected of COVID-19 outbreak or implemented nationwide. These results could partly be explained by the same transmission route of the two pathogens and further indicate that there exist interactions between influenza virus and SARS-CoV-2. Although our results show that the main effect of health education promotion was moderate and the lag time was the longest one among the 14 NPIs, this NPI's contribution in suppressing influenza in 2020 was sizeable (11.47%). This might be because health education promotion is easy to use and therefore frequently implemented by many governments worldwide.
We for the first time quantify the intensity for each NPI to start to generate effect and that to reach maximum effect. As for the NPI of the international travel restriction, we found that only banning arrivals from some foreign regions rather than all regions or total border closure could not produce the effect in suppressing influenza transmission. The reason for this might be that for viruses that could cause global pandemic, such as influenza virus and SARS-CoV-2, it is crucial to strictly restrict their transmission across country borders. For example, at the early age of COVID-19, the US only banned traveling from China23 even if there were some emerging cases in Europe. But this incomplete restriction of international movement resulted in a surge in COVID-19 cases in the US due to the imported COVID-19 cases also from Europe.24 We found that for school and workplace closure, the recommendation alone was not sufficient to generate a significant effect in suppressing the influenza transmission. This is might because leaders of schools and factories have inadequate knowledge about the threat of infectious diseases, thereby continue the normal operation of organizations. This means that the government needs to enforce the requirement of closing schools and workplace. In terms of the NPI of health education promotion, our results show that official notice alone was not sufficient to generate a significant effect in suppressing the transmission of influenza, unless it was combined with social media. Therefore, the key role of social media in public health education and disease prevention should be recognized and deployed by policymakers. However, false information and rumors can easily spread via the social media25, so the authority of official notification is indispensable.
We found that there existed saturation points of NPIs. Finding the saturation points indicates that maximize suppressing effect could be achieved while minimizing the social and economic costs of their interventions. The reasons behind this phenomenon might because enforcing NPIs at a certain level could reach saturation. However, we lacked data regarding the public's actual behaviors in reaction to NPIs, which is more closely relevant to the influenza transmission. Therefore, another possible reason could be the deterrent effect of a certain level of NPIs. For example, restricting 1000 gatherings might make people think twice before attending social activities with less than 1000 participants. However, our analysis was restricted to the earlier stage of the pandemic, while the deterrent effect might decline with the prolonging of the pandemic period because of the growing pandemic-policy fatigue26 among the public.
We took into consideration and calculated the lag time of each NPI. Results show that the lag time of health education is the longest. It is reasonable given that it takes a longer time for the public to receive and internalize public health information compared with other NPIs. We also found that the lag time of some NPIs (e.g., domestic movement restriction, testing policy) was less than three days, which is not reasonable in theory because the series interval (i.e., the time between two successive cases) of influenza is around three days.27 This might be mainly because the fact that the impact of these NPIs might be too negligible to calculate a reasonable lag time. This could also be supported by the low contribution rank of these NPIs.
We found a positive interaction effect when the intensity of mask wearing requirement and other NPIs (e.g., gathering limitation) are both strong, whereas merely implementing mask wearing requirement at high intensity shows a negative effect. This could be explained by risk compensation.28 In other words, the public might assume that only wearing masks could fully protect them from respiratory infection, thereby increasing the frequency and time of contacting with others. However, influenza virus is not only transmitted by air, but also by contact transmission (e.g., a person touches the surface accumulated with droplets from infected patients and then touches his/her face).1 Therefore, influenza cases could rise in those people who have close contact with others despite wearing masks.
A negative interaction was shown when both the intensity of international travel restriction and the intensity of contact tracing were strong, while a positive interaction effect was shown when the intensity of contact tracing was weak. This might be because when the intensity of international travel restriction is strong enough to suppress influenza transmission, contact tracing might no longer be necessary with the reduction of cases. Additionally, during the early decline period of influenza epidemic, we observed a positive interaction effect when both the intensity of contact tracing and that of testing policy were strong (Figure S4C), which may indicate other NPI factors may be involved.
Our study has some strengths. First, compared with previous studies that simply used the time of introducing NPIs as a surrogate for the effect of NPIs, we used a more reliable, detailed, and comprehensive NPI database, which allowed us to quantify different levels of each NPI. Second, most previous studies6,8−10 simply investigated one country or region, whereas we for the first time included and compared the impact of various NPIs on influenza transmission across a total of 33 countries in the northern hemisphere. Third, the previous studies5,16,17 evaluating the effectiveness of NPIs in suppressing COVID-19 lack accurate case data, because COVID-19 is an emerging infectious disease that lacks detection kits of high sensitivity and specificity, and the testing rate is very likely to be influenced by the intensity of testing policies vary by country. In comparison, the well-established influenza surveillance system used in this study (i.e., GISRS) has consistently and reliably monitored influenza activity since 1952.29 Lastly, from the perspective of methodology, we used the machine learning model and explainable machine learning method to capture the complex relationship between the intensity of NPIs and influenza suppression index. Hence, we could obtain a range of more detailed and informative results.
Nevertheless, our study has several limitations. First, due to the lack of influenza incidence in our dataset, we only used the influenza-positive rate to approximate actual influenza activity. Although previous studies30,31 used influenza positive rate multiple influenza-like illness rate to approximate the incidence, the majority of countries (20/33, 60.6%) in our dataset did not report influenza-like illness rate. Second, the influenza positive rate might be underestimated because under the impact of COVID-19 pandemic, the health seeking behavior of influenza patients might be reduced and medical resources tend to be inadequate. Nevertheless, the number of specimens tested in 2020 was comparable with that in previous year (1,701,758 vs 1,675,945).32 Lastly, we used an ecological study design, so the effectiveness of NPIs might be impacted by a range of uncontrolled confounding factors specific to the country in which NPIs were implemented, such as country demographic structure, climate, and the presence of other NPIs. Due to the same reason, we were unable to adjust for specific implementation details of each NPI which may vary by country. And the interpretation for interaction between NPIs should be careful.
In conclusion, our results estimated the effectiveness of various NPIs in suppressing influenza transmission and provided detailed scientific evidence from different aspects. These results could provide reference to policymakers to deal with potential influenza pandemics in the future. Nevertheless, more detailed information from other aspects such as other unincluded NPIs in our analysis and the cost effectiveness of implementing NPIs could be further explored.