A Comparative Analysis of Methodological Approaches Used for Estimating Risk in Travel Medicine

Authors

  • Karin Leder FRACP,

    Corresponding author
    1. Victorian Infectious Diseases Service, Centre for Clinical Research Excellence, Royal Melbourne Hospital, Victoria, Australia
    2. Department of Epidemiology and Preventive Medicine, Monash University, Victoria, Australia
    Search for more papers by this author
  • Mary E. Wilson MD,

    1. Mount Auburn Hospital, Cambridge, MA, USA
    2. Harvard Medical School, Boston, MA, USA
    3. Department of Global Health and Population, Harvard School of Public Health, Boston, MA, USA
    Search for more papers by this author
  • David O. Freedman MD,

    1. William C. Gorgas Center for Geographic Medicine, Division of Infectious Diseases, University of Alabama at Birmingham, Birmingham, AL, USA
    Search for more papers by this author
  • Joseph Torresi FRACP

    1. Victorian Infectious Diseases Service, Centre for Clinical Research Excellence, Royal Melbourne Hospital, Victoria, Australia
    2. Department of Medicine, University of Melbourne, Victoria, Australia
    Search for more papers by this author

Dr Karin Leder, FRACP, Department of Epidemiology and Preventive Medicine, Monash University, Alfred Hospital, Melbourne 3004, Victoria, Australia. E-mail: karin.leder@med.monash.edu.au

Providing optimal pretravel advice requires travel medicine practitioners to perform an epidemiologic and host-related risk assessment so that preventive measures can be appropriately prioritized for each traveler.1,2 Individual host characteristics and itinerary details must be considered to properly balance the efficacy, side effects, and costs of various interventions against estimates of the incidence and severity of the disease(s) they can prevent. Most preventable travel-related diseases are associated with relatively low risks, generally of the order of 1 of 100,000 to 100 of 100,000. Consequently, decisions regarding intervention measures that minimize risks will depend on the risk threshold, such that diseases with poor or fatal outcomes will undoubtedly be associated with less tolerance of even small risks than diseases leading to only mild morbidity. The patient’s own perception of risk and their attitude toward the reassurance provided by the intervention measure versus its potential side effects or costs is also a vital consideration when tailoring individual pretravel advice. Research providing data on the risk factors for specific adverse health outcomes during travel enables high-risk travelers to be identified and preventive measures to be optimally targeted.1

In addition to benefiting risk management in the individual traveler, improved understanding of the health risks faced according to individual traveler characteristics and itineraries and up-to-date data regarding the epidemiology of specific infections also provide guidance regarding possible differential diagnoses following travel to specific areas. This facilitates assessment and quantification of disease risks if a traveler returns with an illness and helps guide diagnostic measures and rapid empiric treatment interventions. Reliable data regarding the burden of travel-related illness also have the potential for significant public health impact by facilitating the recognition and limitation of disease transmission across international borders.

Until recently, the evidence base for travel recommendations has largely relied on case reports, case series, retrospective chart reviews, or small single-institution cross-sectional or cohort studies. Many reports have been retrospective, often citing opportunistic rather than systemically collected data. These studies have helped to establish the groundwork for travel-related research and to suggest risk factors associated with common adverse health outcomes among travelers. However, there are limitations and biases associated with each research approach that can significantly impact risk estimates, and results of risk estimate studies performed among limited populations or involving limited travel destinations will often not be generalizable to other travelers. Some issues regarding the advantages and limitations of different research approaches are similar in travel medicine as in other fields, but the fact that travelers generally have a defined and identifiable period of risk—namely their trip—results in some unique issues for travel researchers. This paper tracks how different methodological approaches that have evolved over the past three decades have contributed to risk characterization and risk estimates in travel medicine.

Challenges in travel medicine research

The attack rate of illness following travel to a particular region is expressed as the ratio of travelers to that destination who became ill to the total number visiting that destination. In practice, ascertainment of exact numerators of all incident cases of infection over a time period or of denominators reflecting the total numbers of travelers to each region is difficult to obtain for all diseases. Additionally, disease risks are not stable over time and new infections continue to emerge,3–5 yet for many pathogens, data regarding the incidence of infection in travelers are based on figures collected almost 25 years ago. Furthermore, travel-related data are nonexistent or scarce for some existing vaccine-preventable diseases, and new interventions such as vaccines for dengue and hepatitis E are being developed, so establishing evidence-based recommendations for appropriate implementation of these interventions requires more precise risk evaluation.

The short incubation period for many travel-related infections means that travel researchers often focus health outcomes studies on the first few months following a trip. While rapid symptom onset following exposure is advantageous in enabling the period required for follow-up of patients post-travel to be relatively brief, it also means that current approaches to risk identification may miss infections with long incubation periods. Additionally, the fact that many symptoms occur during a trip can make it difficult for researchers situated in the patients’ home country to make an accurate diagnosis of illness. Another problem is that attribution of place of exposure to determine the incremental risks associated with visiting a specific destination can be problematic, particularly if the itinerary has included multiple countries, if several trips have been undertaken within a relatively short period, or if diseases with delayed symptom onset are being examined. Researchers often present results according to regions of exposure in an attempt to overcome the problem of assigning risk to a particular country; however, exposure risks will not be uniform throughout an entire country let alone an entire region. Similarly, the incremental effect of travel duration as a risk factor for disease acquisition can be difficult to assess, particularly if areas with variable disease endemnicity have been visited. There may be a temptation to use surveillance data of disease risks among endemic populations to infer risks to travelers, but such data may not be particularly relevant to the traveling population as vaccination status, behaviors, and exposures may be markedly different. For example, notifications of Japanese encephalitis occurring among local populations may be impacted by high levels of immunity from widespread vaccination or previous infection, resulting in few reported cases despite ongoing circulation of viruses and limiting the applicability of these data in making vaccine recommendations for travelers.

Research approaches for risk estimation and risk characterization in travel medicine

It is important to recognize and understand the strengths and limitations of the major study designs when interpreting individual study results. These are discussed here, with additional information shown in Table 1.

Table 1.  Research approaches for risk estimation and risk characterization
Study approachAdvantagesLimitationsComments
Randomized controlled trialsCan be used to compare the efficacy of different interventions for illness prevention, such as for comparison of the efficacy of different malaria chemoprophylactic regimens or different vaccines for a specific pathogen.Limitations in the research questions that can be addressed, as it is unethical to intentionally leave individuals unprotected. For example, cannot randomly assign individuals to certain pretravel health interventions versus none.Considered gold standard study design for many research questions but cannot be used to monitor disease trends or inform absolute risk estimates.
Use of notification dataCommon demographic features and common destinations among people who have acquired notifiable diseases can be examined. Trends of illness over time can also be determined.Only possible for infections that are notifiable. Underreporting is common. National differences in case definitions and varying behaviors of travelers from different geographic regions may limit generalizability of results.Reported cases can be used as numerator data, and if surrogate denominator data are available, estimates of absolute risk can be made.
Cohort studiesCan determine the relative frequency of various illnesses after travel, can highlight risk characteristics associated with acquisition of specific infections, and can also provide incidence estimates for different diseases.Often focus on syndromes rather than specific etiologic diagnoses, as enormous numbers of prospectively enrolled travelers are required for sufficient incident cases of individual pathogens. Generalizability is limited if focus is on travelers from one clinic or region or those traveling in one geographical area. Recruitment limited to individuals who have had a pretravel visit will result in selection bias. If follow-up is brief, estimating the incidence of diseases with long incubation periods will not be possible.One of the first major research papers in travel medicine (performed by Steffen and colleagues6 in the 1980s) used this approach. Results have been updated and often quoted as best estimate of the burden of travel-associated illness.
Cross-sectional studiesCan capture both ill and well travelers and so can characterize likely risk factors for illness. May be able to estimate disease incidence for illnesses with short incubation periods.Generalizability of findings limited to the specific population and time period studied.This approach often involves airport surveys and has been used to determine people’s knowledge, attitudes, and practices regarding preventive measures.
Case-control studiesCase-control approach is useful in the setting of disease outbreaks among travelers.Logistic difficulties in finding true controls—other similar travelers who have remained completely well—mean true case-control studies are performed infrequently in travel medicine.Many outbreaks have been described in travelers involving such infections as hepatitis A,7,8 cutaneous larva migrans,9 histoplamosis,10–13 schistosomiasis,14–16 and Legionnaires’ disease.17–19 Either a case-control or a cohort approach will be used for outbreak investigations.
Case series and chart reviewsCan describe spectrum and relative frequency of particular health problems among unwell travelers. Can also identify common demographic or itinerary risk characteristics associated with acquisition of specific illnesses.Generalizability is limited if the study is performed by a single institution as travelers from different regions may have dissimilar behaviors. Often focus on travelers returning from one specific destination or alternatively on one specific disease outcome, so separate regions and illnesses need to be examined one at a time. Usually include only unwell patients who present for medical care, so predominantly include more severe illnesses and cannot compare characteristics to those who have remained well.Cumulatively can provide information on different infections acquired in diverse geographical destinations but cannot inform absolute risk assessment.
Case reportsCan document novel observations such as emergence of new strains of infection, new mechanisms or new areas of transmission, or rare adverse events after vaccination or chemoprophylaxis.Cannot be used to infer disease incidence.Particularly appropriate mechanism for communicating rare outcomes.

Case series and chart reviews

Many retrospective chart reviews and prospective collation of case series have been performed as they are relatively cheap and easy to do. These studies often collate information on unwell travelers who present for medical care to describe the spectrum of diseases seen and analyze the relative frequency of different illnesses. Sometimes, they are limited to individuals returning from a single destination, so that common adverse health outcomes following exposure in that region can be identified. Sometimes, data are collected by clinics that travelers frequent during travel, as these clinics may see a different spectrum of illnesses (often those with short incubation periods), and they can provide focused country-specific risk information.20,21 Alternatively, case series may focus on a specific disease outcome, collating patients who have developed that illness to identify factors that contribute to risk, such as common demographic or itinerary characteristics. A case-control study approach is also sometimes performed on accumulated data to better define and delineate observed risk factors.

However, there are a number of limitations to this approach. Chart reviews and case series are typically performed by a single institution, so results are influenced by the characteristics, behaviors, and travel destinations favored by its travelers, which limits the generalizability of results. To adequately examine one specific disease etiology, studies often need to be conducted in large referral centers that have accumulated many cases of the disease, and even if sufficient case numbers are available for analysis, such studies will be prone to referral bias. Also, separate destinations and disease outcomes need to be studied one at a time. Additionally, these studies frequently include only unwell—often hospitalized—patients, so they generally reflect only the more severe cases of travel-related health problems rather than the spectrum of morbidity resulting from travel infections, and comparisons cannot be made between unwell travelers who have presented for medical care and those who have remained well. Some of these issues may be able to be addressed via systematic reviews or meta-analyses that collate and examine data obtained by a number of institutions; however, some biases and limitations will remain. Finally, these studies lack accurate numerator or denominator data and so cannot inform absolute risk calculations.

As an example of this approach, a number of studies by individual centers have focused on travelers returning with fever and have reported on the contribution of malaria as the underlying cause. Results have varied from 27% to 75%22–25 depending on the makeup of the traveling population served by the center, whether all patients seen or only inpatients were included, and according to the common destinations visited by patients from each site. Nevertheless, collectively these studies have enabled examination of a large number of febrile travelers who required medical care and have highlighted the importance of malaria infection.

Another example that demonstrates the utility of this research approach is a large study of schistosomiasis performed by a single center in Britain.26 Common symptoms, investigation findings, treatment effects, and follow-up difficulties among more than 1,100 travelers and immigrants from Africa with schistosomiasis were described. Few study centers could accumulate such large numbers of cases, and it is likely that the major findings are applicable to other travelers from Africa returning to other sites around the world. However, comparative morbidity analyses and examination of other disease outcomes would require additional studies.

Cross-sectional studies

Cross-sectional studies involve administration of a questionnaire to a sample population to collect information regarding the prevalence of specific exposures or conditions at one time point. In travel medicine, cross-sectional studies have often been performed as airport surveys, enabling capture of both ill and well travelers for study. These studies provide estimates of the frequency of exposures among those with and without symptoms, therefore enabling characterization of likely risk factors for illness. They can also estimate the incidence of illnesses that have short incubation periods and therefore occur during travel and can be used to compare attack rates for these acute infections following travel to different destinations. Cross-sectional studies have also often been used to determine people’s knowledge and practices associated with certain preventive measures. However, this approach has limited utility for estimating risks of infections for which symptom onsets are delayed and the generalizability of findings is limited to the specific population, region, and time period being examined, and these studies are also subject to selection bias as people participating in the study may not be representative of the entire travel population.

An example of the contribution this study design can have on risk estimates of illness following travel in different countries is evident in a large, multicenter cross-sectional survey conducted to examine travelers’ diarrhea among 73,630 short-term visitors completing airport surveys just prior to flying home from Kenya, India, Jamaica, or Brazil.27,28 Diarrhea was reported by a high of 55% in Mombasa and a low of 14% in Fortaleza, and calculated 14-day incidence rates varied between 20 and 66% according to destination. This highlights the differences in results according to itinerary and suggests that similar studies in other airports would be required to obtain globally comparative data.

Another series of cross-sectional surveys were conducted among travelers departing from a number of different geographical locations (Australasia,29 Europe,30,31 the United States,32 and South Africa33) assessing the knowledge, attitudes, and practices of travelers from different regions in relation to travel-related diseases and preventive measures. The proportion of responders who had sought pretravel health advice ranged from 32% to 86% between the different sites, thereby highlighting the limited generalizability of single-site cross-sectional studies of pretravel practices.

Prospective cohort studies

Cohort studies in travel medicine generally involve recruitment of people undergoing a pretravel health assessment and then follow-up of these people on return to determine the magnitude and types of adverse health outcomes that occurred. As cohort studies are often questionnaire based, they are essentially the same as longitudinal surveys. This strategy enables determination of the comparative frequency of various illnesses after travel and can highlight risk factors associated with acquisition of different infections. It is also the most common methodological approach for providing incidence estimates of illness since both numerator and denominator data are captured. Cohort studies generally focus on syndromes (eg, travelers’ diarrhea or respiratory symptoms) rather than on specific etiological diagnoses (eg, confirmed influenza), as enormous studies are required to obtain sufficient numbers of each etiological diagnosis for risk calculations. As the source for recruitment of participants is often via a pretravel clinic, these studies are prone to self-selection bias as not all at-risk travelers present for pretravel advice. In fact, it is probable that those most at risk, namely, young or budget travelers, are the least likely of all to seek pretravel advice. Additionally, cohort studies are often performed at a single site or may focus on travelers visiting specific high-risk destinations, and these factors can limit generalizability of results. Most prospective cohort studies have followed patients for the first 3 to 6 months after return and therefore are unable to determine attack rates of diseases with longer incubation periods. Finally, these studies are expensive, time consuming, and labor intensive to perform, and the validity of results will be compromised if there are significant losses to follow-up.

The first definitive prospective cohort study to look at the burden of travel-associated illness was performed in the 1980s.6 Over 10,500 Swiss travelers visiting developing countries were prospectively enrolled and followed. Fifteen percent reported a health problem, 8% consulted a doctor, and 3% were unable to work for an average of 15 days. The monthly incidence of many travel-related illnesses such as severe diarrhea, acute respiratory tract infection, giardiasis, hepatitis, and malaria was able to be calculated; however, despite the large number of travelers enrolled and reasonable follow-up rate (74%), only 115 had a diagnosis confirmed, and the extensive resources delivered only 10 confirmed cases of hepatitis (A or B) and 8 cases of malaria on which to base risk estimates. These results highlight both the great potential for contribution to risk assessment of the prospective cohort approach and the logistic difficulties in obtaining specific and confirmed diagnostic information. The meticulous statistical analysis performed has allowed these data to stand the test of time and be cited repeatedly as the basis for many disease risk estimates in travel medicine. However, the results reflect the behavior of Swiss travelers and may not be representative of the entire traveling population.

A number of other questionnaire-based cohort studies have been performed since, with frequent citation of a 2-year study of 784 American travelers who traveled to 123 different developing countries for up to 90 days.34 This study found that illness was reported by 64% during travel and illness following return home by an additional 26%. Common morbidities among travelers were identified as diarrhea, respiratory illness, and skin disorders. However, because of the relatively limited sample size, risk of a given diagnosis was not analyzed by region or country.

Use of notification data

Another method for inferring travel-associated health risks is to examine surveillance data for notifiable illnesses. Descriptive information regarding the number of cases of illness reported can be obtained, and common demographic features and destinations among reported cases can be examined to identify risk factors for acquisition of the specific disease. However, this approach is obviously limited to travel-associated diseases that are reportable, such as malaria, typhoid, and hepatitis A. Additionally, diseases may be underdiagnosed and underreported, infections diagnosed and treated abroad will not be captured in the patient’s country of origin, and details regarding the travel history and the precise trip characteristics are often lacking or inaccurate.35–37 Furthermore, national differences in case definitions may affect generalizability of results, as some countries may include only definite cases, whereas others include probable cases as well. Varying behaviors of travelers from different geographic regions may also mean that results are not generalizable to travelers from other areas.

Although notification data cannot itself provide information on absolute risks of infections, some studies have used novel approaches to estimate total numbers of travelers to a region (denominator data) so that attack rates can be estimated. A variety of denominator estimates have been used, often based on national travel data collected by the country performing the analysis (if available) or by using data collected on a global scale such as World Tourism Organization (WTO) data. Despite some limitations and potential inaccuracies in the data used, this approach provides another method for inferring absolute risk estimates.

As an example, notified malaria cases have been collated from 28 industrialized countries (mainly European), with WTO data as denominator estimates, to assess malaria attack rates among travelers to Kenya. Average annual attack rates ranged from 50 to 135 of 100,000 travelers, depending on the reporting country, which highlights both the potential utility and the variability of this approach to risk estimation. Other similar studies have estimated malaria incidence at 9 of 100,000,38 149 of 100,000,39 or up to 240 of 100,00040 following travel to Kenya, perhaps reflecting differences in notification practices, different study periods, and/or different travel behaviors of Italian, UK, and Danish travelers, respectively, and again showing how studies performed by a single center may have limited generalizability.

Risk characterization based on data collected by sentinel surveillance networks for travelers

Over the past 10 years, collaborative sentinel surveillance networks involving clinics that provide specialized services in imported infectious diseases have been established to track travel-related diseases. These networks prospectively and systematically collect data on large samples of ill travelers. The GeoSentinel Surveillance Network, established in 1996, was the first major travel-related surveillance network developed. It is the largest existing database of travel-related illness, comprising 39 geographically dispersed travel/tropical medicine clinics on six continents, and is the only global travel-related surveillance system. Collected data include over 75,000 patient records from over 220 countries of acquisition. Another major coordinated surveillance activity is the European Network on Imported Infectious Disease Surveillance (TropNetEurop) founded in 1999.

Although surveillance networks have not been specifically established for research purposes, the ability to analyze large samples of diverse travelers enables better precision, reliability, and generalizability of findings. For GeoSentinel in particular, the patient groups included are diverse in origin, places visited, types of travel undertaken, and types of illness acquired, enabling a multinational and broad description of traveler morbidity.41,42 However, information is collected only on ill travelers who seek medical care from specialized sentinel sites, so travelers who have remained well and/or those who have not required medical follow-up will not be represented. Additionally, there are selection and reporting biases in the types of patients and types of diagnoses that present at these specialized sentinel clinics,41 and the data fields collected are relatively limited. Multicenter networks also lack uniformity in patient referral patterns, consistency in coding of diagnoses by clinicians, and central laboratory reference facilities. Also, sentinel surveillance cannot capture all incident cases of disease in travelers, and no external denominator data representing the total number of travelers to different regions are reported, so absolute risk calculations cannot be determined.

Despite these limitations, there are a number of analytical approaches for using these surveillance data to provide new methods for risk estimation and characterization. Patterns of both syndromic and specific etiologic diagnoses can be examined. In addition, trends in diseases over time can be studied.

Analytical approaches for using sentinel network data for risk estimates

Proportionate morbidity

Determination of proportionate morbidities—ascertainment of the number of patients with a given diagnosis divided by the total number of ill travelers to a destination—is one method for using sentinel surveillance network data to inform disease risk. In other words, comparative morbidity according to destination of travel is calculated for different diagnoses. The broader the range of travelers captured, the more generalizable are the results, making GeoSentinel the ideal database for such risk analyses. These calculations do not provide information on absolute risks of infection according to itinerary but do enable examination of the association between travel destinations and the probability of returning ill and being diagnosed with certain diseases. Calculations of the proportion per 1,000 ill-returned travelers from specific regions with particular diagnoses can also be performed. These data are valuable when considering the spectrum of differential diagnoses in unwell travelers and in guiding empiric therapy. Additionally, repeating analyses in the future can facilitate temporal comparisons of proportionate morbidities and examination of trends. Indeed, regular examination of GeoSentinel data for the proportionate morbidity of individual diagnoses plotted against month has been effective in recognizing irregular patterns of disease. However, a limitation of this approach is that if a certain disease is overwhelming and very commonly acquired at a particular destination, it may make other illnesses appear trivial by comparison even if there are in fact significant absolute risks.

The definitive paper describing this methodology and outcomes included data on 17,353 ill-returned travelers.41 The frequency of occurrence of different diagnoses was compared among travelers returning from six developing regions of the world, and significant regional differences in proportionate morbidity were detected for many syndromic categories of diagnoses. For example, among travelers presenting to GeoSentinel sites, systemic febrile illness occurred disproportionately among travelers from sub-Saharan Africa or Southeast Asia, whereas acute diarrhea was proportionately more common among those returning from South Central Asia. These results enable practitioners to assess likely disease outcomes according to travel destination.

Sentinel surveillance network reporting rates and reporting rate ratios

To the extent that patients included in surveillance network databases represent the larger population of travelers, the numbers of presentations to sentinel clinics with an illness of interest can be used as surrogate numerator data in risk estimate calculations. Ideal denominator data are all travelers to a destination, but instead, proxy data can be obtained from other sources, such as use of inbound and outbound tourism data collected by the WTO. Although somewhat similar to the approach if notification data are used (described above), using sentinel network data expands the diseases that can be examined beyond the limited few reportable infections. However, incomplete capture of disease numerators by selected sentinel clinics means that disease incidence cannot be estimated. Instead, it enables calculation of sentinel network reporting rates, which reflect the number of patients presenting to clinics in the surveillance network with a diagnosis (or group of diagnoses) following travel to a country or region divided by the WTO estimate of all travelers to the region. These reporting rates give an indication and a quantitative estimate of relative regional risks of specific disease acquisition among travelers. In addition, sentinel network reporting rate ratios, which are calculated by dividing the sentinel network reporting rate for a country or region of interest by the reporting rate for a low-risk reference region or country, can be determined. These latter calculations provide quantitative information regarding the incremental risks of visiting various regions.

Despite limitations, this approach has been used for estimation of relative regional risks of malaria43 and gastrointestinal infections44 according to region of travel, enabling development of a hierarchy of risk according to destinations visited, which is difficult to obtain via most other study designs.

Calculating odds ratios of associations between patient characteristics and diagnoses to inform risk

Another approach is to look for and quantify associations between patient demographic profiles, itinerary characteristics, and disease outcomes via calculation of odds ratios. Although somewhat similar to the approach used for risk factor identification in traditional case-control studies, the “controls” are in fact other unwell travelers in the network database who have presented with illnesses other than the one under study. Optimally, this approach should use a large global database that can facilitate a multicenter, multinational perspective such as GeoSentinel so that generalizability can be enhanced compared to single-region analyses.

As an example of this approach, a range of travel-related illnesses were compared among two subgroups of travelers visiting friends and relatives (VFRs) (“immigrant VFRs” and “traveler VFRs”) to those reported among tourist travelers.45 Findings suggested that immigrant VFRs were proportionately more likely to have serious and/or potentially preventable travel-related illnesses such as malaria, nondiarrheal intestinal parasitic infections, tuberculosis, and sexually transmitted infections than traveler VFRs. Such subgroup analyses involving multiple disease outcomes require examination of a very large number of patients to obtain statistically significant and globally generalizable results and would unlikely be achievable in the absence of network surveillance data.

Network surveillance for individual diseases in travelers

Surveillance networks for individual diseases also enable analyses to be performed regarding risk characteristics and trends in specific diagnoses among travelers. For example, TropNetEurop collects data specifically on malaria, dengue fever, and schistosomiasis occurring among returned international travelers, immigrants, and foreign visitors seen at sentinel sites in Europe.46,47 It has provided important and pioneering descriptive data on cumulative numbers of cases of each of these three common infections in European travelers.48–51

As an example, descriptions of almost 300 patients seen at sentinel sites were analyzed to facilitate understanding of the epidemiology and especially the clinical characteristics of imported dengue fever. However, denominator data were not available either for attack rate calculations or for calculation of proportionate morbidities. As with any regional network, travel destinations favored by travelers from that region may be overrepresented.

Future directions for travel health research

The ideal study for determination of risk estimates for specific diseases among travelers would involve a prospective cohort study similar to the study design used by Steffen6 but using a multicenter approach to recruit a diverse group of travelers originating from and traveling to a variety of destinations globally. Collaboration could be achieved using currently existing networks established for collection of surveillance data. Novel approaches for recruitment could be considered to limit selection bias, potentially using the media, travel agencies, and local doctors rather than specialized travel clinics only to advertise the study. Pre- and post-travel questionnaires regarding symptoms, behaviors, and itinerary details would be administered. In addition, efforts to maximize specific etiologic diagnoses should be considered such as encouraging presentation to health-care facilities for investigations if symptoms occur during travel and optimizing protocols for collection of appropriate clinical specimens including swabs, cultures, and pre- and post-travel serological tests for a variety of pathogens. Although accurate incidence data could still not be obtained for all pathogens, optimizing the use of questionnaire and laboratory data would enable improved risk estimates.

This approach to refining travel-associated risks clearly would involve considerable effort, resources, and expense. For uncommon infections, even very large multicenter prospective cohort studies may not result in sufficient numbers of infection for accurate etiologic risk determination. In addition, precise absolute risk data for all possible travel scenarios and all destinations will not realistically be achieved. For this reason, performing additional studies using conventional study approaches, additional analyses using sentinel network surveillance data, as well as combining and refining research methodologies will continue to be important and collectively will help to overcome some of the limitations posed by individual study approaches.

Conclusions

To date, both prospective and retrospective studies of various designs have provided important insights into the broad spectrum of travel-associated infections. Improved understanding of the risks of travel and of characteristics associated with heightened risk has relevance at a number of levels. Because of the defined period of risk, there are some unique methodological issues related to travel medicine research, which can both facilitate and hinder study performance. Limitations in available data have prompted a variety of approaches to be used to understand the health risks associated with travel.

Refinements in risk estimates are continuously evolving as new evidence emerges and as new approaches to risk quantification are used. Changing epidemiological patterns of disease and emerging health problems also need to be monitored. As research in travel medicine increases, it will become progressively more important to combine the information sourced from a number of different methodologies and consider novel approaches, as this will enhance the evidence base for risk assessment and optimize risk estimates.

Declaration of interests

D. O. F. is a director of the GeoSentinel Surveillance Network. The other authors state that they have no conflicts of interest.

Ancillary