In this study, we were able to provide CrIs of annual community incidence of 3 important infections from surveillance data. We used a method for determining pathogen-specific underreporting factors in Australia that has been deduced without the need to collect costly new data and includes an estimation of precision. This method is applicable to diseases other than infectious diarrhea, provided data on the components of the notification fraction are available: the proportion of case-patients who visit a doctor, the proportion who have a laboratory test, the sensitivity of the test, and the completeness of reporting of illness to surveillance. Even if collecting some additional data is necessary to estimate certain components of the notification fraction, this collecting may be worthwhile to obtain the added insight into the effects of particular diseases. Although knowing the incidence may not be necessary for detecting outbreaks and monitoring increased or decreased disease over time, this information is vitally important to policy makers. We consider it most important to furnish estimates with a measure of their precision and have used a simple simulation technique that is easily applied. If information is to be used in public health policy making, the responsible interpretation of results involves a realistic appreciation of potential error. Simple point estimates may give a misleading picture of certainty; the estimates of the community incidence of these foodborne diseases show a high degree of uncertainty that should be acknowledged when comparing estimates from other countries or times.
If this level of uncertainty is also found in other countries, then our confidence in apparent differences may be compromised. However, some differences appear to be so large that they are of interest nonetheless. When compared with multipliers for enteric diseases in other countries, Australia’s estimates were most similar to the estimates in the United Kingdom that were derived from a cohort study. It was estimated that for every case reported to surveillance, 3.2 cases of salmonellosis and 7.6 cases of campylobacteriosis existed in the community (5
). However, the multiplier for salmonellosis in the United States has been estimated in the past at 39 (3
), and the same factor was estimated in a recent US study (6
). This recent study estimated the component probabilities of the notification fraction by using data from a population survey of diarrhea from 1996 in which the proportion of case-patients who visited a doctor was 12%. More recent surveys in the United States have put this estimate at ≈20% (20
). If 20% is now more appropriate, then the US multiplier would reduce to ≈25. In the US study, blood in stool was found to be highly influential on stool test ordering by doctors (100% vs. 18% requested stool tests, depending on blood in stool) and not so influential on the probability of a case-patient visiting a doctor (15% vs. 12%), which was similar to our Australian study findings. The US study did not report duration as a predictor of visiting a doctor and of having a stool test. Because salmonellosis frequently lasts >5 days, adjusting for duration had a marked impact on our multiplier, reducing it considerably. If a similar influence of duration on visiting a doctor and ordering stool tests exists in the United States, and the symptom profile of salmonellosis is similar in the 2 countries, then the US multiplier would likely be further reduced.
The choice of the case definition of diarrheal illness used in population studies may also have affected calculations for the multipliers when the method of component probabilities was used. If the case definition itself includes features of severity that are predictors of visiting a doctor or of having a stool test, this may affect the proportion of cases that undergo these steps, thereby affecting the component probabilities used to calculate the multiplier. The laboratory sensitivity testing that provided another component probability may also affect the calculation of the multipliers. The quality assurance testing mimics some of the transport issues that occur in real life, but it probably represent a “best cases scenario” in which a patient sheds microorganisms at the time of collection, and good transport methods are available.
In addition to unavoidable uncertainty due to paucity of data, methodologic differences in each study, combined with differences in surveillance systems, can make international comparisons of disease incidence difficult. However, applying the multiplier for each country leads to an estimate of community incidence that is likely to make comparisons more meaningful than simply comparing notification rates. The notification rates of salmonellosis in each country are currently ≈70/100,000 population for the United Kingdom (5
), 38/100,000 for Australia, and 12/100,000 (21
) for the United States. Applying the underreporting factors of 3, 39, and 7 for the United Kingdom, United States, and Australia gives estimates of annual community incidence of ≈220/100,000, ≈470/100,000, and ≈262/100,000 (95% CrI 150–624), respectively. If more recent estimates of the proportion of case-patients who visit a doctor are used to calculate the underreporting factor in the United States, and a factor of 25 were applied to the number of notifications, the US rate becomes 300/100,000.
To validate results or assess the potential degree of error, another useful approach is to use differing methods and compare results in the same country. In Australia, 1 other possible method is to use results from an Australian cohort study of diarrheal disease (22
). The study was a randomized controlled trial, conducted for 18 months, that assessed the health impact of water quality and treatment in Melbourne, Victoria, in 1999. Of 795 stool samples tested, 9 cases of salmonellosis (0.003 per person-year) and 24 cases of campylobacteriosis (0.007 per person-year) were identified. When these data are extrapolated to the notifications in Victoria, the community-to-notification ratios are 12.6 for salmonellosis and 9.3 for campylobacteriosis, which are comparable to our estimated multipliers of 7 (95% CrI 4–16) for salmonellosis and 10 (95% CrI 7–22) for campylobacteriosis.
Assessment of the functionality of surveillance is likely to lead to more effective control of disease in the community, and multipliers are 1 measure of the quality of surveillance systems. The relatively low ratio between reported enteric cases and the number of community cases in Australia suggests that the surveillance system is working reasonably effectively and therefore is likely to detect outbreaks.
This study provides an estimate of the community incidence of 3 important foodborne diseases in Australia. Such estimates are important in public health to assess the economic and human cost of these diseases and to help set priorities. The estimates in this study show that salmonellosis, campylobacteriosis, and STEC have considerable effects in the community, most of which go unreported. Calculation of multipliers for other diseases would also be worthwhile to inform public health practice.