Estimating public health risks of infectious disease events

Initiation phase: setting the stage

Risk framing

The steering committee conducts a risk framing (problem formulation) exercise to determine whether a RRA is needed and outline the scope and key objectives of the assessment. This includes defining the following:

• The public health hazard (pathogen or other threat) that poses a potential risk
• The public health concerns related to the hazard
• The adverse event of concern (the event to be avoided or mitigated); e.g., introduction of an infectious individual with disease X into Canada
• The source of the hazard
• The at-risk population(s) of interest
• The timeframe over which the risk should be assessed
• The contextual factors that can influence the likelihood or impact of the event; e.g., conditions affecting exposure or transmission, available countermeasures and resources for risk mitigation
• The relevant stakeholders (including those whose expertise is required to conduct the assessment and those to whom the results of the assessment should be communicated)
• The risk management decisions that should be informed by the RRA; e.g., border health measures, infection prevention and control guidance

This risk framing aids in the formulation of the specific risk question(s) to be answered during the RRA (see Table 1 for more examples).

Execution phase: conducting the assessment

Risk pathway

The risk framing informs the development of a risk pathway: a diagram describing the sequence of events leading from the hazard's source to the adverse event of concern and its resultant impacts. Each box (node) in the diagram represents a step along the risk pathway; arrows (edges) depict causal relationships, linking each event to its consequences (Figure 1). The likelihood of a specific event occurring is, therefore, conditional on preceding events. Typically, a risk pathway for an infectious disease hazard includes components describing the importation, if relevant, of the hazard (pathogen) from the source country, exposure to the hazard within Canada, human infection, the most likely spread scenario should infection occur and the resulting impacts. The adverse event of concern (the event to be avoided or mitigated) should be clearly defined, since this is the event for which the overall likelihood will be assessed. Other components such as potential interventions (e.g., vaccination, treatment) and monitoring points (e.g., surveillance systems) can be added to the pathway where relevant. Mapping out the risk pathway helps to formulate the risk question that is of key concern and the types of information that will be needed to address it.

Figure 1 - Text description

The figure includes three risk pathways for different rapid risk assessments of: viral hemorrhagic fever disease (e.g., Ebola virus) importation into Canada in panel A; infection of un/under-vaccinated close contacts of a poliovirus infectious individual in Canada in panel B; and zoonotic infection with influenza A(H5N1) clade 2.3.4.4b virus in Canada in panel C. The source(s) of infection (black boxes) and pathway steps (white boxes) to the adverse event of concern (orange boxes) are demonstrated, along with each concurrent likelihood sub-question to be addressed in the assessment. Downstream impacts have been omitted.

In panel A, the following pathway source(s), steps, and risk sub-questions are outlined for viral hemorrhagic fever disease (e.g., Ebola virus) importation into Canada:

• The pathway source identified is an infectious person in a given country. The risk sub-question: What is the likelihood that an individual is infected with Ebola virus in the source country during the assessment period?
• Pathway connector from infectious person in source country leads to infectious person bound for Canada. The risk sub-question: What is the likelihood that an infectious individual travels to Canada in the next 4 weeks?
• Pathway connector from infectious person bound for Canada leads to other traveller(s) being exposed in transit and infectious person not detected by travel screening. This leads to adverse event of concern: infectious person enters Canada. The risk sub-question: What is the likelihood that an infectious traveller is not detected by pre-travel or entry screening?

In panel B, the following pathway source(s), steps, and risk sub-questions are outlined for poliovirus importation and infection in Canada:

• The pathway source identified is a poliovirus infectious person in a given country. The risk sub-question: What is the likelihood that an individual is infected with poliovirus in the source country during the assessment period? 
• Pathway connector from infectious person in source country leads to infectious person entering Canada. The risk sub-question: What is the likelihood that at least one infectious individual from the source country enters Canada?
• Pathway connector from infectious person enters Canada leads to infectious person having close contact with an un/under-vaccinated person. The risk sub-question: What is the likelihood that an infectious individual will have close contact with an un/under-vaccinated individual in Canada?
• Pathway connector from close contact with un/under-vaccinated person(s) leads to adverse event of concern: un/under-vaccinated close contact infected. The risk sub-question: If an un/under-vaccinated individual has close contact with an infectious person, what is the likelihood that they will be infected with poliovirus?

In panel C, the following pathway source(s), steps, and risk sub-questions are outlined for human infection with influenza A(H5N1) clade 2.3.4.4b in Canada:

• The pathway sources identified include infected wild birds, infected poultry flock, infected wild mammals or infected domestic mammals. The risk sub-question: What is the likelihood that an individual animal is infected with avian influenza A(H5N1) clade 2.3.4.4b during the assessment period? 
• Pathway connector from infection in wild birds/poultry/wild mammals/domestic mammals leads to exposure in humans. The risk sub-question: What is the likelihood that a human is exposed to a sufficient amount of virus to potentially cause infection following either (a) a higher-level exposure or (b) a lower-level exposure?
• Pathway connector from human exposure leads to human susceptibility to infection. This leads to adverse event of concern: human infected in Canada. The risk sub-question: What is the likelihood that a person who had exposure to sufficient amount of virus will develop an infection?

Risk pathways can vary in complexity depending on the hazard and risk question and individual components of the pathway can be expanded and assessed in greater detail or simplified as needed. For example, in considering the potential for importation of a non-endemic disease into Canada (e.g., Ebola virus) into Canada, measures such as pre-travel health screening could reduce the likelihood of importation by preventing symptomatic individuals in the source country from travelling (Figure 1A). With infectious diseases that mostly present as asymptomatic (e.g., poliovirus), such measures are unlikely to meaningfully influence the likelihood of importation; thus, the importation component of the risk pathway can be simplified, assuming that pre-travel health screening would not detect poliovirus infections (Figure 1B). The level of detail in a risk pathway are dependent on the time and resources, availability of information, complexity of the risk questions, risk management needs and sensitivity of the risk to specific steps in the pathway.

Similarly, the risk pathway approach allows for flexibility in incorporating components at the human-animal-ecosystem interface. For example, when assessing the risk of human infection with influenza A(H5N1) clade 2.3.4.4b viruses, the risk pathway can include the likelihood of infection in relevant animal species and the likelihood of human exposure and infection (Figure 1C); depending on the scope, the assessment could consider impacts on human health, the economy, wildlife and agriculture. Risk pathways are, therefore, useful for incorporating multi-sectoral perspectives within a One Health approach.

The risk pathway is used to develop specific sub-questions to be answered during the risk assessment. These sub-questions correspond to individual steps (nodes) in the pathway influencing the likelihood of the event of concern (likelihood sub-questions) and steps leading from the event of concern to the impacts being assessed (impact sub-questions). For infectious hazards, assessing the impacts typically requires an assessment of the most likely spread scenario(s) should the event of concern occur. It should be noted that the event of concern can differ depending on the context and the specific objectives of the assessment (Figure 1, orange nodes).

Risk pathways can be adapted to consider at-risk populations or settings within the RRA (Figure 2). For example, the risk pathway can capture the likelihood and impact of infection in defined sub-populations of concern, such as specific occupational, demographic or other relevant high-risk groups (Figure 2A). A further consideration is that the unit of analysis may differ depending on the context. Figure 1 and Figure 2A depict risk pathways related to the likelihood of an individual importing, transmitting or acquiring a pathogen. Figure 2B depicts a risk pathway in which the event of concern is the spread of a multidrug-resistant organism between healthcare facilities, making a healthcare facility a more appropriate unit of analysis.

Figure 2 - Text description

This figure includes two risk pathways for different rapid risk assessments of: mpox transmission in Canada in panel A; and multidrug-resistant organism (MDRO) outbreaks in Canada in panel B. The source of infection (black boxes), adverse event of concern (orange boxes), pathway steps and downstream impacts (white boxes) are demonstrated. Panel A illustrates the use of the risk pathway to assess impacts at an individual level, on a specific sub-population (the impact of mpox transmission within the gay, bisexual and other men who have sex with men community), and on the population of Canada as a whole. Panel B illustrates the use of healthcare facilities as the unit of analysis in considering spread of an MDRO between healthcare institutions.

Abbreviations: gbMSM, gay, bisexual, and other men who have sex with men; MDRO, multidrug-resistant organism

Estimation of likelihood and impact

Once the risk pathway is complete, evidence to address each pathway sub-question is compiled, reviewed and appraised to produce qualitative estimates of likelihood and/or impact together with levels of uncertainty (see more on uncertainty below). Estimates are informed by a rapid review of relevant evidence, which can include scientific literature, published and unpublished technical reports and epidemiological investigations, event and case-based surveillance data, intelligence obtained through international networks and reporting systems and scientific expertise. As time and evidence can be limited during the early stages of an event, RRAs rely on expert knowledge and opinion. Subject matter experts in relevant areas can guide the estimation process by providing contextual or privileged information about the hazard being assessed, a nuanced interpretation of the evidence and expert judgement on the event of interest and surrounding context, such as relevant sociocultural factors and industry practices. Uncertainty is estimated based on the availability and quality of relevant evidence, SME opinion and degree of expert agreement. For each pathway sub-question, a qualitative estimate of the likelihood or impact is assigned using pre-defined, standardized scales describing how likely an event is to occur and what impact it is expected to have, both among directly affected individuals and the wider population. Each estimate is accompanied by a brief, focused rationale summarizing the evidence that supports the level assigned. Scales for likelihood and impact estimation available in existing risk assessment frameworks^{Footnote 2Footnote 8} can be adapted to suit the local and situational context (see RRAs of measles in Canada and influenza A(H5N1) clade 2.3.4.4b virus for current likelihood, magnitude of effect, impact and uncertainty scales^{Footnote 11Footnote 12}.

Likelihood considerations

The overall likelihood is a qualitative statement of probability that the adverse event of concern will occur within the time period of interest. The overall likelihood for the adverse event is conditional on the likelihood estimates for preceding steps in the risk pathway and is derived in a manner analogous to the quantitative multiplication of probabilities. When multiplying conditional probabilities, the overall probability can never be higher than the lowest individual probability in the pathway. In the qualitative equivalent, the overall likelihood should not be higher than (and is thus determined by) the lowest likelihood estimate in the pathway (Figure 3)^{Footnote 13}.

Figure 3 - Text description

This figure includes one risk pathway for the rapid risk assessment of viral hemorrhagic fever disease (e.g., Ebola virus) importation into Canada. The source of infection (black box), adverse event of concern (orange box), pathway steps and downstream impacts (white boxes) are demonstrated. The likelihood and impact sub-questions, with corresponding estimates, are based on internal exercises during the testing of different rapid risk assessment methods. The diagram illustrates how the overall likelihood of the adverse event of concern (an infectious person entering Canada) is determined by the lowest conditional likelihood estimate in the pathway leading up to the event. Estimates provided in this figure serve only as an example for the derivation of an overall likelihood, individual-level impacts, and the impact on the general population.

Importation into Canada:

• The pathway source identified is an infectious person in a given country. The risk sub-question: What is the likelihood that an individual is infected with Ebola virus in the source country? The estimate is low.
• Pathway connector from infectious person in source country leads to infectious person bound for Canada. The risk sub-question: What is the likelihood that an infectious individual travels to Canada in the next 4 weeks? The estimate is very low.
• Pathway connector from infectious person bound for Canada leads to other traveller(s) being exposed in transit and infectious person not detected by travel screening. The risk sub-question: What is the likelihood that an infectious traveller is not detected by pre-travel or entry screening? The estimate is low.
• Pathway connector from either infectious person not detected by screening or other traveller exposed in transit leads to adverse event of concern: infectious person enters Canada. The overall likelihood of Ebola virus introduction into Canada within the next 4 weeks is estimated as very low.

Spread within Canada:

• Pathway connector from infectious person enters Canada leads to transmission to close contacts. The risk sub-question: If an infected person enters Canada, what is the most likely spread scenario?

Impacts:

• Pathway connector from either infectious person in Canada or infectious close contacts leads to impact on health of directly affected individuals. The risk sub-question: What would be the impact on directly affected individuals? The estimate is severe.
• Pathway connector from transmission to close contacts leads to impact on the general Canadian population. The risk sub-question: What would be the population-level impacts on the general population in the next 4 weeks? The estimate is minor.

Considerations for assessing the likelihood of an infectious disease event depend on the context and the pathway sub-question (examples in Appendix, Table A1). For example, when considering the likelihood of importation of a disease, relevant factors include the prevalence of infection and epidemic trajectory in the source country, the volume of incoming travellers, health screening measures, the potential for transmission of infection during transit, the case-to-infection ratio and incubation period (which influence the likelihood of infected individuals being detected by surveillance or screening) and the duration of infectiousness (which influences the likelihood of an individual being infectious at the time of travel). Similarly, an assessment of the likelihood of infection in Canada should consider the potential for exposure to infectious individuals, the intensity of exposure, pathogen infectivity and demographic, medical, social and other factors that may influence susceptibility to infection.

Impact considerations

The estimation of impact conveys the severity of consequences resulting from the adverse event of concern, should it occur. The PHAC RRA approach assesses impacts at the individual and population levels. Impacts on individuals affected by the hazard are informed by evidence regarding disease severity and associated sequelae, availability and efficacy of prophylaxis and treatment and intrusiveness of control measures (e.g., isolation, quarantine). Population-level impacts are additionally dependent on the most likely scenario for the extent and duration of spread.

The most likely spread scenario is influenced by the pathogen transmissibility (e.g., the reproduction number, R), speed of transmission (e.g., serial interval, epidemic doubling time) and effectiveness of public health measures (e.g., case detection, case isolation, contact tracing) and medical countermeasures (e.g., antimicrobials, vaccines). It is not necessarily a description of what will happen; in some situations, all spread scenarios may be unlikely to occur if they are contingent upon earlier steps in the risk pathway that are themselves unlikely. Conveying how likely a chosen spread scenario is to occur is useful for providing appropriate context for estimated impacts.

Public Health Agency of Canada's assessments typically focus on direct health impacts, but also consider additional impacts on wellbeing (e.g., mental health, long-term disability), impacts on the health system, impacts arising from implementation of public health measures or wider impacts using a social, technological, economic, environmental, political and regulatory and population and health system (STEEPP) framework depending on the risk framing^{Footnote 2Footnote 14}. Balancing these different impacts is challenging and the prioritization of impacts to be assessed should be clarified at the outset, based on the needs of decision-makers and the risk framing exercise. For example, Ebola virus is likely to have severe consequences for infected individuals because of the high case fatality and the need for case isolation but could have minimal population consequences if little onward transmission is expected to occur. Conversely, seasonal influenza is not expected to cause severe illness in most infected individuals but could have major population consequences due to the large number of cases and resultant pressure on the health system.

Sub-populations with disproportionate impacts

The effects of an event may not be uniformly distributed across the population. Certain population sub-groups may be disproportionately affected due to shared risk factors, occupational exposures, demographics, medical vulnerabilities or socioeconomic circumstances. Infectious disease outbreaks that are highly concentrated in certain sub-groups may have minimal impact on the general population. Assessing only the general population impact may mask significant impacts on specific sub-groups and may have downstream health equity implications. Furthermore, it is important to distinguish between sub-groups that are disproportionately affected because of shared risk factors for infection and sub-groups that are more susceptible to severe consequences of infection, as these may not always be the same groups. For example, during the ongoing global outbreak of mpox, transmission has predominantly occurred among gay, bisexual and other men who have sex with men^{Footnote 15} despite high susceptibility to mpox in the general population. Consequently, the health impacts have overwhelmingly affected this sub-population, while the impact on the general population has been minimal (notwithstanding potential elevated risk of severe outcomes in certain sub-groups such as pregnant women)^{Footnote 16}. In contrast, generalized community transmission of SARS-CoV-2 has been the norm throughout the COVID-19 pandemic, but the health impacts have been disproportionately high among the elderly, those with co-morbidities and under-vaccinated individuals, because of their higher susceptibility to severe illness. Assessing the differential impact on specific population sub-groups may, therefore, be necessary for certain hazards and this should be considered in the risk framing, risk pathway and assessment of likelihood, spread and impact.

Integrating evidence and expert opinion

Evidence used to estimate likelihood and impact is triangulated with expert knowledge to support the overall assessment. Expert opinion is particularly valuable during a RRA when evidence is limited or conflicting, as well as to provide contextual information about the event based on prior experience. Additionally, expert input can help to identify key uncertainties and knowledge gaps in relation to the risk question. The overall timeframe and number of SMEs involved in a RRA can vary depending on the event and the complexity of the issue. For events with less complexity, such as the risk associated with Ebola virus importation, a smaller group of SMEs may suffice and discussion with the aim to build consensus on risk estimates may be feasible. For more complex events, such as the risk associated with human infection with an influenza A(H5N1) clade 2.3.4.4b virus, experts from multiple sectors may be needed, including human and veterinary medicine, public health, virology, immunology, agriculture and environmental science. In such cases, different options can be considered to obtain balanced input from experts. Strategies can vary from requesting targeted input on sections most relevant to experts' field of knowledge to obtaining initial estimates of risk from experts through surveys ahead of group discussions, to help minimize biases and ensure that all relevant views are represented.

Levels and drivers of uncertainty and knowledge gaps

For each likelihood and impact level assessed, a level of uncertainty is assigned based on the availability and strength of relevant evidence, as well as expert opinion. As RRAs are typically conducted in the context of limited data, outlining the level of uncertainty for the likelihood or impact of different steps of the risk pathway is crucial for delineating the weight of evidence supporting individual estimates and provides important contextual information for decision-makers to guide appropriate actions^{Footnote 17}. In addition to uncertainty levels, identifying drivers of uncertainty and variability is important for determining when actions based on the precautionary principle might be warranted and for defining triggers for re-evaluation of the risk, such as changes in epidemiology. For example, it is not possible to pinpoint which human-adaptive mutations will occur in influenza A(H5) strains within a given timeframe; this is inherently unknowable and will always be highly uncertain (i.e., has high variability). The uncertainty in the likelihood of a viral hemorrhagic fever disease importation, on the other hand, is influenced by availability of information on the extent of transmission, the specific groups in which transmission is occurring, the effectiveness of control measures and the expected volume of inbound travel from the source country. Identifying information gaps can inform surveillance and research recommendations. For example, during an assessment of the risk of human infection with avian influenza A(H5N1) clade 2.3.4.4b virus, gaps identified included lack of evidence regarding the infectious dose in humans and the types of exposures necessary for infection. Consequently, recommended actions included enhancing and integrating surveillance activities for avian influenza across the One Health spectrum in Canada to understand infection risk in human population groups with higher exposure (e.g., agricultural workers) and rapid information sharing of case detections^{Footnote 11}.

Assumptions and limitations

During the assessment, certain assumptions may be necessary to make estimation possible in the face of limited information. For example, data on the frequency of and risk factors for severe illness following human infection with avian influenza A(H5N1) clade 2.3.4.4b virus are currently limited, given the small number of human infections that have been identified to date. Some similarity between this and other influenza A(H5) viruses in the propensity to cause severe illness may, therefore, need to be assumed. Any assumptions made during the assessment are described and any relevant limitations that could influence the outcome or limit the scope of the assessment are listed.

Closure phase: recommendations, communication and update

Summary statement and recommendations

The key findings of the RRA are described in a risk statement that summarizes the likelihood and impact estimates, main drivers influencing the estimates and key sources of uncertainty in the assessment. The risk statement, identified knowledge gaps and recommendations form the main outputs of the RRA report. Risk management decisions are outside the scope of RRA and may be based on factors other than assessed risk, including the level of risk tolerance, resource availability, cost-benefit analyses or acceptability of different control measures. However, providing recommended actions helps inform decision-makers, risk managers and relevant stakeholders on risk management options that are proportionate to the risk posed by a given public health hazard. These can include specific actions for response, such as surveillance, implementation of control measures or risk communication to mitigate risk at different levels (e.g., federal, provincial or territorial levels in Canada), as well as research to address knowledge gaps.

If there is considerable uncertainty regarding the likelihood and potential impact of an adverse event beyond the timeframe of the RRA (e.g., for pathogens with pandemic potential or for which the epidemic trajectory is highly uncertain), a description of plausible future scenarios or considerations influencing future risk can be included in the assessment to guide preparedness planning^{Footnote 11Footnote 12}.

Updating the rapid risk assessment

As a public health event evolves and more information becomes available, a re-assessment of the risk and associated uncertainties may be required to ensure that ongoing risk management activities are appropriate. As part of the PHAC RRA process, monitoring indicators are defined that, if met, would indicate a worsening of the situation and trigger a reassessment of whether an updated RRA is needed. For example, increase in case counts, increased disease severity or case detections in new countries or regions of a given infectious disease outbreak could all trigger a re-evaluation of the risk. In future iterations of an RRA, the risk pathway and risk question(s) may need to be revised if the epidemiological situation changes significantly.

Discussion

We have demonstrated the application of a coordinated approach to public health RRA in the Canadian context, with a focus on infectious disease events. This method uses the risk pathway as a flexible framework to characterize the likelihood and impact of a public health event of concern, aligned with established international RRA frameworks^{Footnote 2Footnote 8}. As part of the RRA, estimates of likelihood and impact are reported separately to adequately inform risk management decisions.

Alternative RRA frameworks, such as those used by the ECDC^{Footnote 1} and the UK Human Animal Infections and Risk Surveillance group^{Footnote 7}, use algorithms to guide risk assessors through a pre-determined decision process to derive estimates of likelihood and impact. Algorithm approaches have the advantage of using a standardized set of questions for every risk assessment and of being intuitively easier to understand for both risk assessors and decision-makers. In our experience, however, the risk pathway approach provides greater flexibility than a binary decision process when a more nuanced assessment is required and this approach may be easier to adapt for a broad range of hazards. This is primarily made possible with the ability to craft risk questions specific to the public health event being assessed to ensure that the RRA outputs are practical and relevant for the required risk management decisions under consideration by the steering committee.

Ongoing developments to PHAC's RRA methodology include enhancing approaches for rapid expert elicitation, broadening and improving assessment of impacts beyond health, potentially integrating qualitative and quantitative approaches to inform assessments, strengthening the assessment of quality of evidence, including variability in estimating uncertainty and exploring the possibility of expanding the methodology to hazards other than infectious diseases. Expert judgement, while critical for qualitative RRAs, is known to be prone to various biases^{Footnote 18}. These can be mitigated through rigorous elicitation protocols and training of experts in subjective probability judgements, neither of which are easily implemented within the timeframe of a RRA. More work in rapid expert elicitation is required to develop flexible protocols and training material that can be implemented during RRAs.

Although the human health impacts of a public health event may be of primary concern in many risk assessments, in some situations, the economic and social impacts may be substantial. Examples include the economic impacts of highly pathogenic avian influenza on the agricultural sector and the wide-ranging impacts of the COVID-19 pandemic. Commonly used qualitative impact scales, such as that recommended in the JRA OT^{Footnote 8}, incorporate some of these dimensions, but can be difficult to use in practice because they require judgements about the relative importance of health and other impacts. The development of separate scales to capture impacts in different domains, such as impacts on health, the health system, the environment and wider society, could provide a more specific characterization of the types of impacts expected from both infectious and non-infectious hazards^{Footnote 14Footnote 19}.

Increasingly, developments in mathematical modelling are being used to provide simulations and timely forecasts of likelihood of importation, epidemic spread and the impact of control measures to aid decision-making. Exploring how to integrate quantitative approaches into RRAs may help provide additional understanding of the potential impacts of an event, the key factors influencing those impacts and what public health actions should be prioritized to minimize impacts.

Risk assessment, an evolving field within public health, is important for informing timely and evidence-based decision-making. Given the broad range and complexity of public health hazards, RRA provides a coordinated approach to characterizing and communicating the level of risk to public health posed by an event of concern, to help prioritize and inform risk management activities. When the public health risks fall at the intersection of human-animal-plant-ecosystem health, a multisectoral One Health approach to risk assessment can reduce duplication of effort, improve timely sharing of information across sectors, enhance focus on upstream drivers of health risks and impacts across sectors and facilitate engagement of multiple sectors in risk management measures.

Authors' statement

• SPA — Conceptualization, methodology, writing–original draft, writing–review & editing
• CCT — Conceptualization, methodology, writing–original draft, writing–review & editing
• SC — Conceptualization, methodology, writing–review & editing
• DA — Methodology, writing–review & editing
• LS — Methodology, writing–review & editing
• IL — Methodology, writing–review & editing
• RA — Supervision, writing–review & editing
• JTW — Conceptualization, methodology, supervision, writing–review & editing
• EG — Supervision, writing–review & editing
• LV — Conceptualization, methodology, supervision, writing–review & editing
• SPA and CCT contributed equally to this article.

Competing interests

None.

Acknowledgements

The authors would like to thank Victoria L Edge for contributions to methodology discussion; and steering committee and technical team members involved in the rapid risk assessment of the ongoing epizootic of avian influenza A(H5N1) clade 2.3.4.4b virus.

Funding

This work was supported by the Public Health Agency of Canada.

Appendix