Result of the systematic review
A total of 2742 papers were found. Removing duplicates left 1653. We retained 622 based on title, 137 based on abstract, and 39 based on full text. We found 23 additional studies via reference lists and PubMed-related article searches (eFigure in the Supplement). Therefore, a total of 62 studies were included (Table 1). Randomized trial evidence was only available to support the effectiveness of hand-washing and face masks [5, 7,8,9,10]. For other interventions, only lower-quality (observational and modelling) evidence was available.
Cost-effectiveness of interventions
Pasquini-Descomps et al.  conducted a systematic review of the cost-effectiveness of interventions in H1N1 influenza. They found 18 studies covering 12 interventions: disease surveillance networks (very cost-effective), contact tracing and case isolation (very cost-effective), face masks (very cost-effective), preventive measures in hospitals (cost-effective), antiviral treatment (cost-effective), antiviral prophylaxis (cost-effective), low efficiency vaccination (cost-effective if timed before cases peak), high efficiency vaccination (cost-effective if timed before cases peak), stockpiling antiviral medicine (cost-effective for high-income countries), quarantining confirmed cases at home (cost-effective for viruses with a case fatality rate of 1%, not cost-effective for viruses with a case fatality rate of 0.25%), self-isolation at home (cost-effective with a case fatality rate of 1%, not cost-effective with a case fatality rate of 0.25%), and school closure (not cost-effective). Based on these findings, Madhav et al.  estimated that for H1N1 influenza, contact tracing was 4363 times more cost-effective than school closure ($2260 vs. $9,860,000 per death prevented). Other systematic reviews found that school closures did not help control of the 2003 SARS epidemic in China, Hong Kong, and Singapore and would prevent only 2–4% of COVID-19 deaths ; reduced the peak of epidemics by 29.65% on average and were more effective when timed early ; are most effective when they cause large reductions in contact, when the basic reproduction number is below 2, and when attack rates are higher in children than in adults ; and appeared to be moderately effective in reducing the transmission of influenza and in delaying the peak of an epidemic, but were associated with very high costs . Differences in publication date, virus transmissibility, and study selection may explain the discrepancies among these reviews.
Using data from Wang et al.  and Pasquini-Descomps et al.  found that contact tracing and case isolation was one of the most cost-effective interventions to control H1N1 in Hubei, China (less than $1000 per disability-adjusted life year). In a simulation study, Hellewell et al.  found that in most scenarios, highly effective contact tracing and case isolation would be enough to control a new outbreak of COVID-19 within 3 months. Transmissibility was an important factor, i.e. when Ro = 2.5, 80% of contacts needed to be traced and isolated. Timing was another important factor—with five initial cases, there was a greater than 50% chance of achieving control, even at lower contact-tracing levels. However, at 40 initial cases, control was much less likely. Similarly, any delay from symptom onset to isolation decreased the probability of control, highlighting the need for swift action. Furthermore, Armbruster and Brandeau  found that contact tracing is cost-effective only when population prevalence is still low (e.g. under 8% for tuberculosis). In a systematic review, Halton et al.  found that contact tracing and progressively earlier isolation of probable SARS cases were associated with the control of SARS outbreaks in Southeast Asia. In another review, Bell et al.  recommended contact tracing and case isolation at the start of an outbreak, but not in the late phase, when there is increased and sustained transmission in the general population. In a modelling study, Zhang et al.  found that voluntary self-isolation at symptom onset can achieve the same level of effectiveness as antiviral prophylaxis, but that this strategy had a limited effect on reducing transmission when delayed by 2 days. Young et al.  also found that delays could prevent case isolation from stopping incipient outbreaks. Li et al.  found that in the 2009 H1N1, quarantine of close contacts in Beijing reduced confirmed cases by a factor of 5.6. However, since H1N1 was mild, they concluded that this was not an economically effective measure. In another modelling study, Tuncer et al.  found that social distancing had the most impact on the 2014 Ebola epidemic in Liberia, followed by isolation and quarantine. Case isolation, household quarantine, and contact tracing were the most effective interventions in four other modelling studies [25,26,27,28]. Collectively, in the context of COVID-19, these studies suggest that these interventions can be effective and cost-effective, and highly so when implemented early and executed swiftly.
Saunders-Hastings et al. [9, 10] carried out a systematic review and meta-analysis of personal protective measures to reduce pandemic influenza transmission. Meta-analyses suggested that regular hand hygiene provided a significant protective effect (OR = 0.62; 95% CI 0.52–0.73). Face masks had a non-significant protective effect (OR = 0.53; 95% CI 0.16–1.71), which became significant (OR = 0.41; 95% CI 0.18–0.92) when randomized control trials and cohort studies were pooled with case–control studies (this also decreased heterogeneity). In an earlier systematic review, Jefferson et al.  also found a protective effect of masks. Overall, they were the best performing intervention across populations, settings, and threats. Similarly, in a narrative review, MacIntyre and Chughtai  drew on evidence from randomized community trials to conclude that face masks do provide protection against infection in various community settings, subject to compliance and early use. Differences in publication date, search strategy, and study selection criteria may explain the discrepancies among these reviews. Tracht et al.  estimated savings of $573 billion if 50% of the US population used masks in an unmitigated H1N1 epidemic. For hand-washing, Townsend et al.  estimated that a national behaviour change programme in India would net $5.6 billion (3.4–8.6), a 92-fold return on investment. A similar programme in China would net $2.64 billion (2.08–5.57), a 35-fold return on investment.
Preventive measures in hospitals include the use of personal protective equipment for healthcare workers in direct contact with suspected patients. Dan et al.  estimated that this measure was cost-effective for H1N1 ($23,600 per death prevented). However, adopting a wider set of measures (full personal protective equipment, restricting visitors, and cancelling elective procedures) was much less cost-effective ($2,500,000 per death prevented). Similarly, Lee et al.  found that increasing hand hygiene, use of protective apparel, and disinfection are the most cost-saving interventions to control a hospital outbreak of norovirus. If they are not adequately protected, healthcare workers can contribute disproportionately to the transmission of the infection .
Suphanchaimat et al.  found that influenza vaccination for prisoners in Thailand was cost-effective. The incremental cost-effectiveness ratio of vaccination (compared with routine outbreak control) was $1282 to $1990 per disability-adjusted life year. Shiell et al.  also found that vaccination (for measles) was cost-effective ($32.90 marginal cost per case prevented). Prosser et al.  also found that H1N1 vaccination in the US was cost-effective under many assumptions if initiated prior to the outbreak. Incremental cost-effectiveness ratios ranged from $8000 to $52,000 per quality-adjusted life year for persons aged 6 months to 64 years without high-risk conditions. The authors noted that all doses (two for some children, one for adults) should be delivered before the epidemic peak. Similarly, in a modelling study, Nguyen et al.  found that vaccination should be administered 5 months before to 1 week after the start of an epidemic to be cost-effective. If vaccine supplies are limited, Lee et al.  found that priority should be given to at-risk individuals and to children within high-risk groups. Likewise, Van Genugten et al.  estimated similar results from vaccinating the entire population versus only at-risk groups. Herrera-Diestra and Meyers  found that vaccinating based on the number of infected acquaintances is expected to prevent the most infections while requiring the fewest intervention resources. Optimal control modelling studies also suggest that early intervention and vaccination are more cost-effective and that interventions before vaccines are available need to be balanced with the potential gains of future vaccines or the potential for multiple outbreaks [41,42,43].
In another systematic review of economic evaluations, Pérez Velasco et al.  examined 44 studies and found that combinations of pharmaceutical and non-pharmaceutical interventions were more cost-effective than vaccines and/or antivirals alone. Reducing non-essential contacts, using pharmaceutical prophylaxis, and closing schools was the most cost-effective combination for all countries. However, quarantine for household contacts was not cost-effective, even in low- and middle-income countries. A modelling study by Day et al.  suggested that quarantine (of all individuals who have had contact with an infected individual) would be beneficial only when case isolation is ineffective, when there is significant asymptomatic transmission, and when the asymptomatic period is neither very long nor very short.
Perlroth et al.  estimated the health outcomes and costs of combinations of 4 social distancing strategies and 2 antiviral medication strategies. For a virus with a case fatality rate of 1% and a reproduction number of 2.1 or greater, school closure alone was the least cost-effective intervention and cost $32,100 per case averted. Antiviral treatment ($18,200), quarantine of infected individuals ($15,300), and adult and child social distancing ($5600) had increasing levels of cost-effectiveness. However, combining interventions was more cost-effective, and the most cost-effective combination included adult and child social distancing, school closure, and antiviral treatment and prophylaxis ($2700 per case). However, the same combination without school closure was more cost-effective for milder viruses (case fatality rate below 1%, reproduction number 1.6 or lower). If antivirals are not available, the combination of adult and child social distancing and school closure was most effective. Similarly, in another modelling study, Bolton et al.  found that a combination of non-pharmaceutical interventions proved as effective as the targeted use of antivirals.
In a similar study of cost-effectiveness, Saunders-Hastings et al.  examined a range of interventions (school closure, community-contract reduction, hand hygiene, face mask, voluntary isolation, quarantine, vaccination, antiviral prophylaxis, antiviral treatment) in response to a simulated pandemic similar to the 1957 H2N2. In a population of 1.2 million, with no intervention, 9421 life-years were lost. Vaccination plus antiviral treatment was the most cost-effective intervention (cost per life-year saved: $2581). However, it still led to 3026 life-years lost. Only 1607 life-years were lost at a marginally higher cost ($6752 per life-year) with a combination of interventions including community-contact reduction, hand hygiene, face masks, voluntary isolation, and antiviral therapy. Combining all interventions saved the most lives (only 267 life-years lost), but was very costly ($199,888 per life-year saved) due to school closure and workdays lost.
Halder et al.  aimed to determine the most cost-effective interventions for a pandemic similar to H1N1. They found that a combination of interventions was the most cost-effective. This combination included treatment and household prophylaxis using antiviral drugs and limited duration school closure ($632 to $777 per case prevented). If antiviral drugs are not available, limited duration school closure was significantly more cost-effective compared to continuous school closure. Other social distancing strategies, such as reduced workplace attendance, were found to be costly due to productivity losses. Closing school for 2 to 4 weeks without other interventions did not cost much more than doing nothing but gave a significant 34 to 37% reduction in cases, if optimally timed.
Studies on intervention effectiveness without cost-effectiveness analysis
Smith et al.  carried out a systematic review of non-pharmaceutical interventions to reduce the transmission of influenza in adults. Only randomized trials were included, and 7 studies met all selection criteria. The authors found that positive significant interventions included professional oral hygiene intervention in the elderly and hand-washing, and noted that home quarantine may be useful, but required further assessment.
Jefferson et al.  conducted a Cochrane systematic review of physical interventions to interrupt or reduce the spread of respiratory viruses. They found that the highest quality randomized cluster trials suggested this could be achieved by hygienic measures such as hand-washing, especially around younger children. They recommended that the following effective interventions be implemented, preferably in a combined fashion, to reduce transmission of viral respiratory disease: frequent hand-washing with or without adjunct antiseptics; barrier measures such as gloves, gowns, and masks with filtration apparatus; and suspicion diagnosis with isolation of likely cases.
Lee et al.  carried out a systematic review of modelling studies quantifying the effectiveness of strategies for pandemic influenza response. They found that combinations of strategies increased the effectiveness of individual strategies and could reduce their potential negative impact. Combinations delayed spread, reduced the overall number of cases, and delayed and reduced peak attack rate more than individual strategies. Similar results were found by Martinez and Das . In another systematic review of 12 modelling and three epidemiological studies, Ahmed et al.  found that workplace social distancing reduced cumulative influenza attack rate by 23%. It also delayed and reduced the peak attack rate.
Pan et al.  examined associations between public health interventions and the epidemiology of COVID-19 in Wuhan, China. Traffic restrictions, cancellation of social gatherings, and home quarantines were associated with reduced transmission, but were not sufficient to prevent increases in confirmed cases. These were reduced and estimates of the effective reproduction number fell below 1 only when additional interventions were implemented. Those included hospital-based measures (designated hospitals and wards, use of personal protective equipment, increased testing capacity, accelerated reporting, and timely medical treatment) and community-based interventions (quarantine of presumptive cases and quarantine of confirmed cases of their close contacts in designated facilities).
Markel et al.  examined non-pharmaceutical interventions in US cities during the 1918–1919 influenza pandemic (isolation or quarantine, school closure, public gathering ban). They found that all 43 cities in the study adopted at least one of these interventions and that 15 cities applied all three. The most common combination (school closure and public gathering bans) was implemented in 34 cities (79%) for a median duration of 4 weeks and was significantly associated with reductions in weekly excess death rate. Cities that implemented interventions earlier had greater delays in reaching peak mortality (Spearman r=−0.74, P<0.001), lower peak mortality rates (Spearman r=0.31, P=.02), and lower total mortality (Spearman r=0.37, P=.008). There was a significant association between increased duration of interventions and a reduced total mortality burden (Spearman r=−0.39, P=.005). Another similar, historical study of US cities found that early intervention was associated with lower mortality (R2=0.69, P<0.01) .
Ishola and Phin  reviewed the literature on mass gatherings. They found 24 studies and cautiously concluded that there is some evidence to indicate that mass gatherings may be associated with an increased risk of influenza transmission. In a more recent systematic review, Rainey et al.  found that mass gathering-related respiratory disease outbreaks were relatively rare between 2005 and 2014 in the US. They concluded that this could suggest—perhaps surprisingly—low transmission at most types of gatherings, even during pandemics. Similarly, in a US survey of 50 State Health Departments and 31 large local Health Departments, Figueroa et al.  found that outbreaks at mass gatherings were uncommon, even during the 2009 H1N1 pandemic. In a modelling study, Shi et al.  found that mass gatherings that occur within 10 days before the epidemic peak can result in a 10% relative increase in peak prevalence and total attack rate. Conversely, they found that mass gatherings may have little effect when occurring more than 40 days earlier or 20 days after the infection peak (when initial Ro = 1.5). Thus, the timing of mass gatherings might explain the apparent lack of evidence in support of their ban.
Recently, Zhao et al.  quantified the association between domestic travel out of Wuhan, China, and the spread of SARS-CoV-2. Using location-based data, they estimated that each increase of 100 in daily new cases and daily passengers departing from Wuhan was associated with an increase of 16.25% (95% CI: 14.86–17.66%) in daily new cases outside of Wuhan. Ryu et al.  conducted a systematic review of international travel restrictions, screening of travellers, and border closure. They examined 15 studies and concluded that the evidence did not support entry screening as an effective measure and that travel restrictions and border closures would have limited effectiveness in controlling pandemic influenza. In another systematic review, Mateu et al.  concluded that the evidence did not support travel restrictions as an isolated intervention for the containment of influenza and that restrictions would be extremely limited in containing the emergence of a pandemic virus. Chong and Ying Zee  modelled the impact of travel restrictions on the 2009 H1N1 pandemic in Hong Kong. They estimated that restricting air travel from infected regions by 99% would have delayed the epidemic peak by up to 2 weeks. Restricting both air and land travel (from China) delayed the peak by about 3.5 weeks. However, neither 90% nor 99% travel restrictions reduced the epidemic magnitude by more than 10%, and antiviral treatment and hospitalization of infectious subjects were found to be more effective than travel restrictions. Chinazzi et al.  modelled the impact of travel limitations on the spread of COVID-19. They estimated that the travel quarantine of Wuhan delayed the overall epidemic progression by 3 to 5 days in mainland China and reduced international case importations by nearly 80% until mid-February. In addition, sustained 90% travel restrictions to and from China only modestly affected the epidemic trajectory, unless combined with a 50% or higher reduction of transmission in the community. Bell et al.  point out that screening international travellers who depart infected countries (instead of all travellers entering all countries) would be a better use of resources. Case in point: Zhang et al.  reported that in the 2009 H1N1, only 132 of the 600,000 travellers who underwent border entry screening in Beijing were infected (0.02%). Travel limitations may be more effective when neighbouring countries fail to implement adequate outbreak control efforts [66, 67].
We found little evidence to support the following interventions: (1) communicating health risk and promoting disease control measures in low- and middle-income countries (evidence not conclusive according to a review by ); (2) screening to contain spread, at the borders or locally (even under best-case assumptions, more than half of infected people would be missed, according to a modelling study by ).
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.