Since the mid-1990s and the raise of concerns about bioterrorism and emerging diseases, non-diagnosis-based data have increasingly been used for routine disease surveillance and outbreak detection [1]. The CDC defined “syndromic surveillance” as an investigational approach where health department staff, assisted by automated data acquisition and generation of statistical alerts, monitor disease indicators in real-time or near real-time to detect outbreaks of disease earlier than would otherwise be possible with traditional public health methods [2].

In such efforts, different registries have served as data sources for public health surveillance [1],[3], including data on absenteeism at work or school [4], calls to health helplines [5],[6], emergency department consultations [7],[8], ambulance dispatching [9], or drug sales. Although unspecific, such data sources can have the advantage over diagnosis-based surveillance of providing information within short delays since the event and in readily available electronic form for relatively low-cost, while capturing large parts of the population.

Drug sales data analysis may overcome the limitation of poor specificity when groups of drugs are exclusively used for the disease or disease syndrome of interest. Furthermore, drug sales data may earlier capture changing population health status, as over-the-counter (OTC) sales and a dense network of pharmacies in most developed countries make drugs easily accessible to patients at the earliest appearance of their symptoms.

Despite this potential interest, no state of the art of drug sales based surveillance is available to date. The present systematic literature review therefore summarized the evidence for an added value of drug sales data for infectious disease surveillance. We limited the scope of the review to infectious diseases, as they represent a public health problem for which early and valid signal detection is of particular concern, in light of potentially rapid emergence and opportunity for control interventions.

We conducted a literature search from 1975 up to June 2012 to identify relevant peer-reviewed articles regarding surveillance of infectious diseases based on drug sales data. PRISMA guidelines were followed in the reporting of the review [10]. Published articles were searched for on electronic databases (Pubmed, Embase, Scopus, LILACS, African Index Medicus, Cochrane Library), using combinations of the following key words: (“surveillance” OR outbreak detection OR warning system) AND (over-the-counter OR “prescription drugs” OR pharmacy OR (pharmaceutical OR drug OR medication) sales). The search was limited to articles in English or French. There were no limitations on study settings.

To be included in the review, articles had to describe, test, or review an infectious disease surveillance based on drug sales data; and be original research that presented new data and results. We excluded studies that monitored chronic diseases, as well as prevalence studies whose purpose was not epidemic detection.

One reviewer screened and evaluated the titles and abstracts. Articles were widely included in a first stage. The full-text review and the final selection of the articles were made by two reviewers.

We reviewed and described the articles in terms of objectives, diseases studied, data sources, methodologies, and performance for real time surveillance. To describe methods and results, we separated the articles into three groups based on their main objective: descriptive retrospective studies, drug selection studies, and prediction studies. Outcomes selected to compare drug sales data to reference surveillance data of the corresponding disease were correlation measurements (strength and timeliness of the correlation) and indicators of outbreak detection performance (sensitivity, i.e. ability to identify true outbreaks; specificity, i.e. ability to identify true negative and timeliness of the detection).

The evidence gathered in this systematic literature review suggests that drug sales data analysis can be a useful tool for surveillance of acute respiratory and gastrointestinal infections.

As could be expected, prescribed drug sales data were strongly correlated with clinical case reporting. No lead time was observed, which is consistent with the fact that patients purchase drugs after seeing a healthcare professional. Analysis of prescribed drug sales data may nevertheless have an additional utility for epidemic detection, as these data might be available with a shorter delay than clinical surveillance data [27].

A high correlation between OTC drug sales data and reference surveillance data were found in almost all the retrospective studies. Several studies also showed that OTC drug sales can serve as an early indicator of disease epidemics. Patients may buy nonprescription drugs during the early phase of illness when they become symptomatic, before consulting a health practitioner [38]. A surveillance system based on drug data should ideally detect all the outbreaks, rapidly, with a low false alert rate. However, few studies in the review determined the sensitivity and specificity of the outbreak detection and those aspects should be analyzed in more details in future studies.

Surveillance based on OTC drug sales could be particularly relevant for diseases whose prodromal phase persists for several days before the onset of more severe symptoms. For example, the early stages of dengue fever symptoms are nonspecific (fever, headache, myalgia, arthralgia, etc.) [39]. The occurrence of grouped cases could trigger an excess of nonspecific drug sales over baseline levels, which in turn could provide an early warning of outbreak in an endemic area.

Results from drug selection studies showed that it is possible to identify groups of products strongly associated with incidence data, which can then be used to predict future trends in clinical data and help public health authorities to prepare response planning. Such product selection procedures, however, depend on the existence of large clinical surveillance databases of the diseases concerned.

Similarly, the validity of drug sales data analysis has been evaluated mainly for two disease groups, respiratory and gastrointestinal illness, for which clinical reference data, used as the gold standard, are readily available. Pertussis and syphilis have been evaluated in only one study each, and still require further confirmation. The concept of drug-based surveillance therefore needs to be validated for other infectious diseases.

All the studies were conducted in developed countries or area. Surveillance based on drug sales data requires electronic information systems for routine data analysis. Besides, its implementation requires that the population has access to the health care system and mainly buy drugs in pharmacies. This could limit the use of drug based surveillance systems in developing countries.

By improving the timeliness of epidemic detection compared to clinical data and giving information from a larger part of the population, drug sales data can be an additional source of information for already monitored diseases. Besides, drug sales data analysis could have its greatest value in the surveillance of diseases for which clinical surveillance is cumbersome and costly, or where substantial under-reporting is suspected. To confirm the selected drug group as a valid proxy of disease, clinical surveillance may be conducted for a defined period in a representative population. Examples of diseases for which this would be useful are typically varicella, urinary infections, allergies/asthma, and parasitic diseases.

Ideally, the drugs to be monitored should be specific to the disease and widely used to treat it in order to maximize the sensitivity of the signal. For example, benzylpenicillin benzathine 2.4 MUI is the quasi exclusive treatment for syphilis infection [13] and is a good candidate. In contrast, the treatment of measles is mostly symptomatic without a specific drug, which makes this disease unattractive for this approach. Another limitation applies to diseases that are usually treated in hospitals or specialized centers, such as tuberculosis.

Surveillance based on drug sales, may not be appropriate to accurately estimate incidence of diseases, as the source population size is not precisely known. Moreover, it may be difficult to link the number of drug packages sold to the number of patients with disease. However, the method is very efficient to determine temporal dynamics of a situation and to detect abnormal phenomena. Surveillance based on drug sales is therefore well adapted to diseases with seasonal variations such as norovirus gastroenteritis, influenza and other infectious respiratory agents, or community outbreaks (foodborne illnesses, waterborne illnesses, hepatitis A, etc.).

Drug sales can be influenced by store promotions, sales period (holidays, weekends), and the media. Also, we do not know whether people buy medications to treat a disease they currently have or a disease they fear they may have in the near future. For example, during the media coverage of avian influenza A (H5N1) in the US, an increase in antiviral medications sales was observed [40], which corresponded to stockpiling behavior of the population.

Health-seeking behaviors also vary by demographic, social, cultural, and economic factors. A survey [41] in Canada analyzed the healthcare-seeking behaviors of 351 patients with acute gastroenteritis. They found significant differences (patient age and sex) between the patients who used OTC drugs and those who did not. Consequently, factors that prompt self-medication should also be taken into account. The usefulness of drug sales based surveillance is also dependent on the available resources and the organization of the health care system. OTC drug sales surveillance is for example less relevant in countries where reimbursement rate are high and patients mainly get prescribed drugs.

Population mobility, particularly in tourist areas, may lead to an increase in remedy sales, which could wrongly be interpreted as a disease outbreak. Inversely, patients with high geographical mobility may not be included in the region of study and lead to an underestimation of the magnitude of an epidemic.

Despite some limitations, routine collection and analysis of drug sales data are likely to be developed in the coming years. Many automated surveillance systems that collect drug data the day after the sales have been implemented in the last decade [14],[19],[20],[27]. They allow a rapid assessment of the public health situation. Early detection of outbreaks allows public health authorities to set up epidemic investigations and control measures sooner. Most studies included in this review were published after the year 2000, with their number increasing recently. They illustrate the need for improved surveillance systems, evidenced by recent public health crises (e.g., anthrax in 2001, the SARS outbreak in 2003, the A/H1N1 influenza pandemic in 2009, etc.). Drug sales data present indeed many advantages in terms of public health surveillance. Data can be obtained in a real-time manner and usually cover a large portion of the population. Data collection may be exhaustive, without selection of specific sales, and allows the simultaneous monitoring of a large number of diseases, especially new or emerging diseases.

Although non-specific, drug sales data are directly linked to patients’ health conditions. Drug sales data are therefore more specific than other syndromic surveillance data, such as tracking search patterns on the web and are likely to reflect more accurately disease activity in the population. Moreover, it should be noted that alternative sources of data for disease surveillance are currently under development. Healthcare management databases that can provide exhaustive information on drug consumption and diagnosis, as the Dossier Médical Personnel [42] in France, are promising tools for disease surveillance.

Our review may be affected by a publication bias since studies unable to show correlations between drug sales and reference data may have been less published. In addition, selections bias may have occurred in the studies. Indeed, some studies in the review were based on a limited number of pharmacies and/or a limited study period (e.g. less than one year). Language bias may exist as we were not able to identify studies published in languages other than English and French. The review focused on the temporal dynamics of infectious disease; consequently, further analyses are required to determine the capacity of these systems to efficiently monitor other aspects of infectious diseases such as spatial spreading.

This review suggests that the analysis of drug sales data is a promising method for surveillance and outbreak detection of infectious diseases. It has the potential to trigger an outbreak alert earlier than most surveillance systems. However, the main challenges consist in the appropriate selection of indicator drug groups and the validation of this approach for diseases for which no or poor quality clinical surveillance data exists. The usefulness of the approach also depends on the available resources and the organization of the health care system. Drug sales databases with real-time or near real-time data transmission are available in several countries; future studies should be encouraged to expand their use on other infectious diseases.

All authors contributed to the study’s design. MP and JM carried out the literature search and reviewed articles. MP drafted the manuscript. All authors interpreted the results, revised and approved the final manuscript.

Mathilde Pivette has a Doctor of Pharmacy degree and a Master of Public Health. She is a PhD candidate in epidemiology at the French School of Public Health. Her research interest lies in the analysis of drug sales data for disease surveillance.