Syndromic surveillance of Norovirus using over-the-counter sales of medications related to gastrointestinal illness

community using over-the-counter (OTC) sales of relevant medications (ie, antinauseants and antidiarrheals) is a form of syndromic surveillance. The usefulness of this type of surveillance has been reviewed in a number of investigations in which OTC medication sales have been shown to provide an earlier signal of outbreaks of diarrheal (1) and respiratory (2) disease than do hospital diagnoses. A recent Canadian study (3) of temporal distributions of GI-related OTC medication sales and emergency room (ER) visits for vomiting, diarrhea and bloody diarrhea found that seasonal patterns for ER visits and OTC medication sales were similar, but different than those of reportable GI cases based on laboratory isolates of bacteria and parasites. The study took place in a province where Norovirus and Rotavirus are not reportable. As part of the Canadian National Enteric Surveillance Program (4), weekly aggregate counts of cases of infectious GI due to reportable bacteria, parasites and viruses are collected for each province. This laboratory-based surveillance system has shown that although the organism-specific infection patterns vary somewhat depending on regional climate, the incidence of bacterial and parasitic infections tends to be higher in summer and early fall, whereas that of viral infections (particularly Norovirus and Rotavirus) appears to peak in winter and spring (5). Under normal circumstances, this pattern is also seen in OTC medication sales (3), suggesting that the pattern reflects underlying community viral infections rather than bacterial or parasitic infections. If OTC medication sales are to be considered for use in syndromic surveillance of community GI, the nature of this relationship needs to be clarified.

OBJECTIVE: To assess whether over-the-counter (OTC) sales of gastrointestinal illness (GI)-related medications are associated with temporal trends of reportable community viral, bacterial and parasitic infections. METHODS: The temporal patterns in weekly and seasonal sales of nonprescription products related to GI were compared with those of reportable viral, bacterial and parasitic infections in a Canadian province. RESULTS: Temporal patterns of OTC product sales and Norovirus activity were similar, both having highest activity in the winter months. In contrast, GI cases from both bacterial and parasitic agents were highest from late spring through to early fall. CONCLUSIONS: Nonprescription sales of antidiarrheal and antinauseant products are a good predictor of community Norovirus activity. Syndromic surveillance through monitoring of OTC product sales could be useful as an early indicator of the Norovirus season, allowing for appropriate interventions to reduce the number of infections.
The objective of the present study was to compare temporal distributions of GI-related OTC medication sales with laboratory-isolate patterns of bacterial, parasitic and viral cases of human GI infections in a province where viral infections from Norovirus and Rotavirus are reportable. A specific objective was to determine the similarity between patterns in OTC medication sales and those linked with Norovirus and Rotavirus activity.

Pharmacy sales data
One major retailer was able to provide electronic data from their 19 pharmacies in the study area. These were dispersed in a region representing approximately 53% of the provincial population, and represented approximately 12% of all pharmacies in that region. The database provided included daily aggregate counts of in-store 'point of sale' purchases of antinauseant and antidiarrheal products between January 2001 and April 2004.

Statistical analyses
SAS Version 9.1 (SAS Institute Inc, USA) was used for all statistical analyses. Temporal patterns were plotted for each organism, as were totals for bacteria, parasites and viruses. Using PROC ARIMA (SAS), weekly OTC medication sales were cross-correlated with the number of reported bacterial, parasitic, Norovirus and Rotavirus infections in the same week, and then with weekly counts lagged one to 24 weeks before and after. Twelve season-byyear indicators ('SEAS_YR') were created: 'Spring 2001' to 'Winter 2003/04'. 'Winter' was defined as December 1 to mid-April, 'Spring' was mid-April to the third week of June, 'Summer' was the last week of June to mid-September and 'Fall' was mid-September to the end of November. Descriptive statistics for weekly and seasonal infections and sales were done using PROC UNIVARIATE (SAS). Seasonal patterns of OTC medication sales and GI cases for combined bacteria and parasites, Norovirus and Rotavirus were presented as the difference between the weekly total and the overall mean, and were plotted together to highlight positive and negative changes from a common mean value of zero on the y-axis. Tukey's test (with Bonferonni adjustment) and LSMEANS within PROC GLM (SAS) were used to determine any significant differences between the SEAS_YR periods (

RESULTS
Of all reported cases from April 2001 to April 2004, approximately 46% were due to bacteria, 18% to parasites and 36% to viruses (29% and 7% due to Norovirus and Rotavirus, respectively). Figure 1 shows the temporal patterns in weekly counts of reportable cases of Salmonella, Campylobacter and E coli infections, which accounted for 96% of bacterial cases. Giardia was responsible for 60% of all parasitic infections in the study, followed by Cryptosporidium (26%) and Entamoeba (14%). Approximately 80% of viral infections were due to Norovirus. Similar seasonal patterns were seen for all reportable bacterial and parasitic infections, with the highest number of cases in summer and also in late spring and early fall; this pattern was not seen for Norovirus or Rotavirus cases ( Figure 2). Norovirus infections peaked in late fall and winter, particularly during the winter of 2002 to 2003. Rotavirus infections occurred somewhat later than those of Norovirus, being highest in late winter and spring. Seasonal patterns, presented as differences from the mean (Figure 3), showed very little concordance between the combined bacteria and parasite patterns and OTC medication sales. Similarly, no temporal relationship between Rotavirus cases and OTC medication sales was evident. Temporal patterns within the OTC medication sales patterns were more closely synchronized with those of Norovirus infection, with peaks and troughs in the differenced values occurring during the same time periods.
Cross-correlation results showed that the highest correlation between weekly counts of OTC medication sales and Norovirus cases was in the same week ('lag 0') with an r 2 of 0.44. Correlation decreased with increasing weekly lags: r 2 =0.32 and r 2 =0.2 for ±1 and ±2 weeks, respectively. OTC medication sales were negatively correlated with counts of both bacteria and parasites for lags 0±10 weeks (r 2 =-0.32 and less); the highest correlation with counts of Rotavirus (r 2 =0.21) occurred at lags of ±4 and ±6 weeks.
Statistical comparisons of weekly averages by year and season for both OTC medication sales and cases of reported Norovirus and Rotavirus infections are shown in Table 1.

DISCUSSION
Previous retrospective studies of large-scale waterborne outbreaks of GI (involving Cryptosporidium, E coli and Campylobacter) have shown that OTC medication sales increased dramatically in synchrony with the underlying epidemic curve (6)(7)(8). This suggests that automated access to electronic OTC medication sales data for use in a syndromic surveillance system may be useful as a public health tool for providing near real-time (within 24 h to 48 h) indicators of community GI activity. Our results show, however, that GI trends under nonoutbreak conditions are predominantly driven by Norovirus infections. This finding has important ramifications when deciding on the usefulness of an OTC medication sales-based syndromic surveillance tool for routine surveillance or as an early warning system. If this tool is biased toward Norovirus infections and the early detection of extremely large outbreaks, it is yet to be determined whether it can make a significant contribution to surveillance above and beyond current laboratory-based methods. However, our findings do suggest that OTC medication sales data could contribute a unique aspect to surveillance with established laboratory approaches.
Under-reporting of GI and associated etiology is a recognized problem worldwide (9)(10)(11)(12). In England, an estimated 1500 more cases of NLV occur in the community for each case recorded (13). A study based in Ontario (14), where Norovirus is not reportable, indicated that approximately one in 300 community GI cases are actually reported. Knowledge of the proportion of GI cases due to viruses, bacteria and parasites is still formative and incomplete in Canada, and would be particularly difficult to determine in provinces where Norovirus is not reportable. However, the etiologies of reported outbreaks alone indicate a preponderance of viral infections. In Ontario, as much as 62% of all GI outbreaks (2000 to 2002) of known etiology were reported to be viral (15). Internationally, the same picture is emerging. In the United States, a study by Fankhauser et al (16) indicates that 93% of nonbacteriological outbreaks are of NLV origin; research by Miller and Mikol (17) indicates that viral agents (NLV and Rotavirus) may account for most GI cases from January to April.
A national laboratory survey (18) in Canada shows that a relatively small percentage of laboratories tested for viruses, 67% of which conducted on-site (stool) enteric testing for bacteria, 31% for parasites and 10% for viruses. Of those laboratories that tested for viruses, 95% tested for Rotavirus and 31% tested for Norovirus. Additionally, the motivation for physicians to request tests for viruses is low (19). This relates in part to the cost and availability of suitable tests and to symptom expression of the various agents. Test results for viral identification using polymerase chain reaction assay would generally only be available after the illness episode, because in reasonably healthy individuals, Norovirus infections are characterized by the rapid onset of severe and relatively short-lived (24 h to 60 h) symptoms of nausea, vomiting and diarrhea. Self-medication may provide adequate symptom relief for the one or two days of illness, likely precluding a physician visit. In comparison, symptoms from infections due to Salmonella, Campylobacter and E coli can develop more slowly and last longer, possibly increasing the likelihood of an individual seeking professional medical care, which, in turn, may also discourage the use of antidiarrheals in particular. Consequently, proportionally more people with Norovirus may be purchasing OTC products than those who have sporadic bacterial or parasitic infections, or those who are involved in outbreaks, increasing the probability of detecting Norovirus cases through sales of medications. The observed lack of correlation between

R o ta v irus O TC
Week number Week number Week OTC medication sales trends and Rotavirus cases may reflect that most of these cases are pediatric and that parents may tend to take their children to a physician rather than medicate with OTC products. Because we were unable to differentiate between adult and pediatric OTC products, we could not investigate this issue further. The use of OTC medication sale trends as a supplemental surveillance tool for monitoring Norovirus levels could, theoretically, be very useful for public health officials. Indications are that viral GI likely has, through sheer volume, a very large impact on community burden of illness, particularly on the more frail or susceptible members of the community. This is especially evident where person-to-person contact is high (nursing homes, day care facilities and hospitals) and where spread is exacerbated by increased confinement in cold weather. An early warning of a rise in community infections (based on OTC medication sales) would allow for extra vigilance and preventive actions (as simple as handwashing) in these facilities.
A limitation of our study was that we did not know whether people were buying products in response to illness (which could be chronic, infectious or noninfectious), for prevention (holidays or travel) or because of a store promotion. Thus, a pharmacy-level consumer study would be valuable for ascertaining more specifically the type of medications being purchased and the reason for the purchase. As well, associated demographic, economic, social and cultural information would further enhance our understanding of OTC medication practices.
Targeted investigations to determine the percentage of GI cases due to bacterial, parasitic or viral infections would address our current lack of information on the relative magnitude of viral illnesses. Key locations for such studies would be hospital emergency rooms, because they would provide community representation and because sample-taking would be easier. An important complement to this would be a populationbased study on GI-related behaviours, including the investigation of factors that prompt self-medication, media and marketing influences, and changes in the availability of health care in Canada.
Improved monitoring and understanding of population GI due to Norovirus, either through syndromic surveillance using OTC medication sales or changes in laboratory testing techniques and protocols, would be expected to have a direct and significant impact on reducing the burden of illness due to this virus.