METHODS

Sources of data

Data sources are shown in box 1.

Estimating all infectious intestinal disease (IID)

The IID study¹³ established that one in every 5.8 cases of IID in the population present to general practice. This ratio was applied to the annual rates of presentation to general practice for IID¹⁰ to produce estimates for the annual number of cases of all IID in England and Wales. The figure for 1995 was calculated using the point estimate for the rate of IID in the community from the IID study.

Aetiology

Only pathogens causing gastrointestinal symptoms,^13,18 and therefore likely to be diagnosed as food poisoning by clinicians in England and Wales, were included in these analyses (box 2). Foodborne botulism and Trichinella spiralis infection were excluded because of their extremely low incidence.

Pathogen specific rates for illness in the population were derived from the IID study.¹³ These were used to estimate all illness in England and Wales in 1995 due to each pathogen (box 3, worked example, salmonellas, step 1). Ascertainment ratios (AR), or multipliers, were then calculated by dividing the estimated number of cases due to each pathogen by the corresponding number of LabBase⁹ reports received in 1995 (box 3, step 2). For each of the other years, the total number of cases due to each pathogen was calculated by multiplying the annual total of laboratory reports by the appropriate AR (box 3, step 3). Annual figures for total IID of unknown aetiology were calculated by subtracting the total number of cases of disease due to known pathogens from the estimate for all IID.

Adjusting for travel associated infection

Data from LabBase and special studies^14–16 were used to determine the percentage of travel associated infection for each pathogen (including infection of unknown aetiology). For each year the number of travel associated cases was subtracted from the total number of cases to produce pathogen specific estimates for indigenous cases (box 3, step 4).

Estimating the number of cases of indigenous foodborne disease (IFD)

The pathogen specific percentage of foodborne transmission in outbreaks (includes foodborne plus person to person spread) from GSURV⁹ was applied to the corresponding number of indigenous cases to produce pathogen specific estimates for IFD (box 3, step 5).

Cases of IFD presenting to general practice

Using the approach described above, data from the general practitioner component of the IID study¹³ were used to produce pathogen specific estimates for new IFD consultations to general practice (box 3, step 6).

Hospital admissions

Data from GSURV, HES, and special studies^11,14–16 were used to estimate pathogen specific hospital admission rates. These were applied to cases of IFD presenting to general practice to produce annual pathogen specific estimates of hospital admissions due to IFD (box 3, step 7).

Hospital occupancy

Mean hospital stay (bed days) for each pathogen was derived from HES.¹¹ This was multiplied by the pathogen specific number of hospital admissions resulting from IFD to estimate the number of bed days (box 3, step 8).

Deaths

Pathogen specific case fatality rates from GSURV and special studies¹⁵ were applied to the corresponding number of cases of IFD presenting to general practice to derive annual estimates of deaths (box 3, step 9).

Quality of evidence

Each of the above steps was classified according to whether the pathogen specific data elements used were direct measures, extrapolations, or inferences in order to evaluate the effects of potential biases on the final estimates produced.

International comparison

To compare the CDC⁸ and PHLS estimates, ratios of the rates of foodborne illness, hospital admissions, and death for all aetiologies, known pathogens, and known bacteria were calculated.

RESULTS

Overall disease burden (table 1)

There were an estimated 10 464 004 cases of IID in England and Wales in 1995. Nearly 14% were acquired abroad, leaving 9 021 129 indigenous IID cases. Of these, 2 365 909 (26.2%) were estimated to be IFD and 989 928 (41.8%) were attributable to known pathogens. Of the known pathogens, six were responsible for 92.7% of IFD—yersinias, campylobacters, Cl perfringens, non-typhoidal salmonellas (salmonellas), Norwalk-like viruses (NLV), and non-VTEC. In 1995, IFD resulted in 511 941 presentations to general practice, 21 138 hospital admissions, 99 874 hospital bed days, and 718 deaths.

Trends in IFD (fig 1, table 2)

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Trends in indigenous foodborne disease in England and Wales, 1992–2000. NLV, Norwalk-like viruses.

Between 1992 and 2000, IFD fell by 53.3% from 2 869 735 to 1 338 772 cases. In 1992, IFD of unknown aetiology (1 644 515), yersinias (392 753), and Cl perfringens (276 266) formed the majority of cases. By 2000 these had all declined sharply. Since 1997 there has also been a fall in salmonellas. IFD due to NLV infection rose by 125.5% and campylobacter by 45.0%.

IFD presenting to general practitioners (table 2)

Campylobacters were the most common cause of IFD presenting to general practice in 2000.

Hospital admissions (table 2)

The contribution of campylobacters rose from 54.8% to 81.6%. Salmonellas remained the second most common cause of hospital admission despite a 55.6% fall between 1995 and 2000. In 2000 VTEC O157 infection ranked third among known pathogens.

Hospital bed occupancy (table 2)

Despite a decline of 19.6% overall, the contribution of campylobacters to bed occupancy rose. In 2000, salmonellas ranked second, Cl perfringens third, and L monocytogenes fourth for bed occupancy.

Estimated deaths (table 2)

Deaths fell by 48.1% (principally salmonella and Cl perfringens deaths). Listeria monocytogenes ranked highly in terms of estimated deaths for the whole period.

Potential effects of assumptions made on final estimates (table 3)

In general, the effects of extrapolation and inference on the final estimates would be relatively small except for campylobacters and unknown agents.

IFD in England and Wales in 2000 (table 4)

Despite accounting for just under half (47.4%) of all cases of IFD, pathogens under routine national laboratory surveillance represented the majority of cases presenting to general practice (74.2%), hospital admissions (95.9%), hospital occupancy (96.4%), and deaths (84.8%).

Comparison with the USA (table 5)

There were 76 million cases of food related illness in the USA per year⁸ compared with 1.3 million cases of IFD in England and Wales in 2000—that is, a ratio of 57:1. When rates were considered, the ratio for all illness fell to 11:1, and to 1.4:1 for bacterial illness. Taking disease severity into account, the two models also converged. The population adjusted ratio of estimates from the CDC and PHLS models for hospital admissions for food related illness as a whole was 3:1, and for deaths was 2:1. For hospital admissions and deaths due to all known pathogens and known bacterial pathogens, the US rates fall below those of England and Wales.

DISCUSSION

In 1992 there were an estimated 2 869 735 cases of IFD in England and Wales. By 2000 this had fallen by over half to 1 338 772. Measures of health service usage due to IFD fell less sharply owing to a rise in the incidence of campylobacters. However, there was a reduction of almost half in the number of estimated deaths. This was due in almost equal part to declines in illness caused by Cl perfringens, following a decline in the consumption of red meats in the UK,^19,20 and salmonellas which followed the introduction of a vaccination programme against Salmonella enterica serotype Enteritidis in chickens by the British poultry industry.²¹ Campylobacters, Cl perfringens, salmonellas, VTEC O157, and L monocytogenes accounted for the greatest disease burden.

Quality of data

Direct measurements were used wherever possible. However, extrapolation or inference was used when accurate measurements were not available because event frequencies were below the level of detection of epidemiological studies or surveillance. Therefore, the effects of these assumptions on the final model will be minimal.

Total cases of IID

For this study, it is important that incidence estimates for common pathogens are accurate. The IID study data were robust for those pathogens contributing most to IFD in England and Wales.¹³ IFD caused by yersinias, aeromonads, and non-VTEC may have been overestimated as not all strains are pathogenic.¹³ Conversely, the role of NLV might have been underestimated as a result of the use of electron microscopy, rather than molecular techniques, as the method of detection in the IID study.²²

Changes in patient presentation, diagnostic practice in primary care, or improved laboratory methods might affect laboratory trends. There has been no shift in the relative proportions of blood and faecal isolates in LabBase from patients with salmonella or campylobacter infections, suggesting no changes in general practitioner or patient behaviour. The widespread use of immunological assays and molecular techniques for NLV in the future means that the reliability of estimates based on IID study data will decay over time. Periodic incidence measurements for specific infections, such as NLV, will be needed to recalibrate the model.

Adjusting for travel associated infection

In general, LabBase data underestimate the extent of imported infection and therefore special study data were used where available.^14–16 However, the final estimates for IFD might not fully account for infection acquired abroad.

Estimating the number of cases of foodborne infection

Using outbreak surveillance data to estimate the proportion of foodborne disease requires care.²³ The validity depends on the extent to which disease transmission in general outbreaks represents all disease transmission. There was however no alternative. Recent national studies of sporadic gastrointestinal infection^13,16,24,25 were not designed to provide attributable fractions for foodborne transmission as a whole.

Clostridium difficile, Shigella spp, Cryptosporidium parvum, adenovirus 40/41, Sapporo-like viruses, and rotavirus are not usually transmitted through food.^8,13,26–29Clostridium difficile and adenovirus 40/41 were included in the PHLS model but, with no foodborne outbreaks reported, neither contributed to the overall burden of IFD. Foodborne transmission rates for Shigella spp, C parvum, and rotavirus from GSURV were low.

Estimates of foodborne NLV transmission vary from 68%³⁰ at one extreme to 7.6%³¹ at the other. Given its high incidence it is important to use an accurate figure for percentage foodborne transmission. The 10.75% figure in the PHLS model was considerably lower than most other published estimates but is derived from the largest and most contemporary dataset (1992–2000; n=1592 outbreaks).

In the CDC model, 85% of VTEC O157 was considered to be foodborne. In England and Wales 63% of VTEC O157 outbreaks were foodborne, which is consistent with recent studies of sporadic infection where person to person spread and contact with livestock were also important.^16,24,25,32

Outbreaks of yersiniosis are scarce in England and Wales. However, as yersinias appear to be one of the most common causes of IID,¹³ better data on pathogenicity and transmission pathways are needed.

Data on outbreaks of disease of unknown aetiology are held in GSURV. However, this might conceal a wide range of agents with differing modes of transmission and therefore represents an area of great uncertainty, requiring further research.

Estimating the number of cases presenting to general practice

Mostly, rates of presentation to general practice were taken directly from the IID study. Estimates for L monocytogenes, S paratyphi, S typhi, VTEC O157, and C cayatenensis were derived by extrapolation or inference. Given the relatively low incidence of each of these pathogens, the effect of inaccuracies on the final estimates would be trivial.

Estimating the number of cases admitted to hospital, hospital occupancy, and deaths

HES for hospital admissions for IID¹¹ as a whole were consistent with data from GSURV and PHLS enhanced surveillance schemes. However, a disproportionate number of patients were assigned to generic International Classification of Disease 10 codes such as “diarrhoea and gastroenteritis of presumed infectious origin”.³³ Therefore, special study^14–17 and GSURV data were used for acute admissions. Chronic disease or long term sequelae were not considered. Detailed NS mortality figures were poor and therefore enhanced surveillance and GSURV data were used.

International comparisons

First impressions are that foodborne illness is 11 times higher in the USA, with an additional 69 million cases after adjustment. However, the CDC baseline population estimate for IID⁸ was fourfold greater than that used in the PHLS model. The US acute gastroenteritis rate was mainly derived from a retrospective population survey.³⁴ However, the IID study team performed a comparison of retrospective and prospective methodologies for assessing rates of gastroenteritis.¹³ The retrospective method yielded a rate of IID in the population that was 2.8 times the rate derived through prospective follow up of the same cohort. It was concluded that recall bias led to the retrospective method overestimating the rate of IID in the community. Using the prospective method, the Sensor study²² in the Netherlands also yielded rates of IID much lower than those employed in the CDC model.

CDC used laboratory surveillance data for most pathogens except NLV.⁸ Instead, on the basis of a single population study,³⁵ the proportion of illness due to NLV was estimated to be 11% of total IID, approximately 40% of which was regarded as foodborne. This represented 67% of all foodborne illness caused by known pathogens. This is crucial as the percentage foodborne transmission in known agents was used as a proxy for unknown agents, and these accounted for 82% of all IID in the USA. Thus varying the percentage foodborne transmission of NLV changes foodborne illness due to unknown agents dramatically.⁸

Using a nearly fourfold greater foodborne transmission rate for NLV, alongside a much higher baseline level for IID results, created 59 million extra cases of illness due to unknown agents in the USA—that is, 85% of the difference between the two models. Similarly, an extra nine million cases of foodborne NLV in the USA accounted for a further 13% of the difference.

When illness due to known bacteria is considered, the CDC and PHLS estimates converge. In both, campylobacter was the most common bacterial cause of foodborne disease. Using population rates, the PHLS estimate for IFD due to campylobacter infection in 2000 was 95% of the CDC estimate. However, for salmonellas the PHLS estimate was only 16% of that from the CDC. Part of the explanation is a substantial decline in salmonellas in England and Wales since 1997.³⁶ Furthermore, CDC used an AR of 38 for both campylobacter and salmonellas. In the IID study the AR for campylobacter was higher than that for salmonellas.¹³

The estimates converged further when disease severity was considered. This is because these data are not directly influenced by the disparity in the baseline estimates for IID. Generally, there were more hospital admissions due to known pathogens and known bacteria in England and Wales than in the USA. In the PHLS model, hospital admissions rates were applied to cases presenting to general practice rather than to laboratory reports as in the CDC model.⁸ However, not all cases presenting to general practice are sampled let alone reported.³⁷ For hospital admissions due to viruses, the CDC model exceeded the PHLS model. However, in the CDC model, data were extrapolated from a single study.³⁸ Hospital admissions due to unknown agents were also higher in the USA.

The same arguments hold true for deaths. Both models highlighted the importance of salmonellas, L monocytogenes, campylobacter, and VTEC. In the PHLS model more deaths from Cl perfringens were due to high numbers reported through GSURV.²⁰

Other evidence from England and Wales

It might appear surprising that we describe a fall in IFD over a period when food poisoning notifications increased. However, our analyses have incorporated data from clinical¹⁰ and microbiological⁹ sources which independently show parallel declines over a nine year period. One study has demonstrated that food poisoning notifications are closely bound to the laboratory reporting of salmonellas and campylobacters.³⁹ This is borne out by an examination of recent trends. Notifications rose to reach a peak in 1998⁴⁰ reflecting an increase in the combined laboratory reporting of these pathogens. Since then notifications have fallen in line with declines in the reporting of both salmonellas and campylobacters.³⁶ The trends in IFD that we have described take into account a much wider range of pathogens and crucially measure the burden of infection due to each of these agents in a way that food poisoning notification data cannot.

In a retrospective survey it was estimated that over five million people per year in the UK suffered from acute gastroenteritis which they ascribed to contaminated food.⁴¹ Our analyses suggest that such a retrospective survey of this type would be expected to produce a figure in this range given that recall bias would result in at least a three fold overestimation in the rate of illness.¹³ This might have been compounded by misclassification bias because individuals made subjective judgements about illness causation. Based on symptoms alone, an individual cannot be certain if their illness was due to foodborne, person to person, or environmental transmission,^{13,16,24,25,42} with the possible exception of those involved in recognised and proven point source foodborne outbreaks. Some individuals with gastrointestinal illness will inappropriately blame contaminated food.

CONCLUSIONS

We have developed estimates using five separate criteria for IFD in England and Wales. A wide range of agents was included and, importantly, an adjustment for foreign travel. Between 1992 and 2000, overall illness fell by over half but hospital admissions declined by only 3%.

Illness due to known pathogens, particularly those under routine laboratory report surveillance, caused the most severe disease and greatest health service usage. In 2000, the majority of general practitioner consultations, hospital admissions, and hospital bed days were due to campylobacter infection. Salmonellas were the most common cause of death, also resulting in high levels of health service usage. Cl perfringens was second only to salmonellas as a cause of death.

If total IFD were taken as the sole measure of disease burden, the impact of VTEC O157 and L monocytogenes would be completely overlooked. Their importance only appeared when hospital occupancy and deaths were considered. By contrast, NLV infection caused far fewer deaths but caused large numbers of cases. We only considered the acute effects of foodborne disease because there is little routine information on chronic disease or long term sequelae.

The pattern of IFD is complex and evolving. Pathogens emerge, laboratory tests improve, and new data streams will require incorporation into the model. The most recent, robust, and reliable data currently available were used, but improvement requires continuous validation of those data sources.

Finally, reducing IFD in England and Wales means tackling campylobacter. Lowering mortality rates however also requires better control and prevention of salmonellas, Cl perfringens, L monocytogenes, and VTEC O157.