Association Between Component Costs,Study Methodologies,and Foodborne Illness–Related Factors with the Cost of Nontyphoidal Salmonella Illness

Abstract

Nontyphoidal Salmonella spp. are one of the most common causes of bacterial foodborne illness. Variability in cost inventories and study methodologies limits the possibility of meaningfully interpreting and comparing cost-of-illness (COI) estimates, reducing their usefulness. However, little is known about the relative effect these factors have on a cost-of-illness estimate. This is important for comparing existing estimates and when designing new cost-of-illness studies. Cost-of-illness estimates, identified through a scoping review, were used to investigate the association between descriptive, component cost, methodological, and foodborne illness–related factors such as chronic sequelae and under-reporting with the cost of nontyphoidal Salmonella spp. illness. The standardized cost of nontyphoidal Salmonella spp. illness from 30 estimates reported in 29 studies ranged from $0.01568 to $41.22 United States dollars (USD)/person/year (2012). The mean cost of nontyphoidal Salmonella spp. illness was $10.37 USD/person/year (2012). The following factors were found to be significant in multiple linear regression (p≤0.05): the number of direct component cost categories included in an estimate (0–4, particularly long-term care costs) and chronic sequelae costs (inclusion/exclusion), which had positive associations with the cost of nontyphoidal Salmonella spp. illness. Factors related to study methodology were not significant. Our findings indicated that study methodology may not be as influential as other factors, such as the number of direct component cost categories included in an estimate and costs incurred due to chronic sequelae. Therefore, these may be the most important factors to consider when designing, interpreting, and comparing cost of foodborne illness studies.

Introduction

Nontyphoidal Salmonella spp. (NTS) are a common cause of foodborne illness (Rabsch et al., 2001; Scallan et al., 2011; Chen et al., 2012). In addition to acute gastrointestinal symptoms, sequelae such as reactive arthritis and irritable bowel syndrome can occur following NTS infection (Lindsay, 1997). Numerous studies have calculated the cost-of-illness (COI) associated with NTS illness, and the economic burden is substantial (Gadiel, 2010; Lake et al., 2010; Chen et al., 2012; Hoffmann et al., 2012; Scharff, 2012). COI analysis is the most common economic burden of illness approach and aims to identify and quantify costs incurred due to disease (Byford et al., 2000; Tarricone, 2006). Despite following a generalized methodology, COI studies can be conducted in various ways (Larg et al., 2011). COI studies employ analytical frameworks that differ in the study cost inventory (i.e., the type of component costs included in an estimate), the epidemiological approach, the estimation methods for direct and indirect costs, and the temporal relationship of study initiation and cost-incurring events (Tarricone, 2006; Larg et al., 2011). Inconsistencies in the way COI studies are conducted limit their usefulness, as different methodologies and cost inventories may influence the magnitude of estimates (Shiell et al., 1987; Oderda, 2003; Akobundu et al., 2006; Segel, 2006; Larg et al., 2011).

Costs inventories consist of individual and societal level costs (Hesmat, 2001; Centers for Disease Control and Prevention, 2005). Direct individual level costs represent the value of goods, services, and other resources consumed in providing care due to illness (Health Canada, 1998). Indirect individual level costs represent productivity losses due to illness or death, intangible costs such as pain and suffering, overhead healthcare-related activities that are shared amongst individuals, and expenditures incurred in the process of seeking care (McPherson et al., 2011). Costs incurred at the population level, such as industry, public health, legal, and government costs are societal costs (Byford et al., 2000; Segel, 2006; Larg et al., 2011).

Direct costs can be estimated using a top-down, bottom-up, or econometric approach. The top-down approach uses known total cost expenditures and portions these costs to broad disease or severity categories (Segel, 2006; Larg et al., 2011). The bottom-up approach requires the estimation of costs associated with a treatment or service and utilization. The costs are then estimated by multiplying unit costs and the number of units used (Segel, 2006; Tarricone, 2006). The econometric approach estimates the difference in costs in a cohort of individuals with and without a disease, but it is seldom used (Segel, 2006). The epidemiological approach of a study is dictated by whether an estimate is based on prevalence (existing cases) or incidence (new cases) data (Byford et al., 2000; Larg et al., 2011). COI studies can be conducted prospectively or retrospectively. In retrospective studies, all of the relevant events where costs can be incurred have already taken place when the study is initiated (Tarricone, 2006).

The two most common methods of estimating indirect costs are the human capital approach (HCA) (Rice, 1966; Scitovsky et al., 1987) and the willingness-to-pay (WTP) approach (Schelling, 1968). The HCA views an individual as producing a stream of output over a period of time (Oderda, 2003; Segel, 2006), and measures costs indirectly associated with illness (i.e., lost output or earnings due to morbidity and premature mortality) (Hodgson et al., 1982; Shiell et al., 1987). Intangible costs are not estimated in this approach (Hodgson et al., 1982; Rice, 1994). The WTP approach proposes that the value of health can be deduced from the amount of money an individual would be willing to pay to reduce the probability of an illness (Mishan, 1971). A “linked” human capital and WTP approach embodies characteristics of both methods (Landefeld et al., 1982). An additional approach to estimating indirect costs is the friction cost method, which is similar to the HCA, but measures only the production losses incurred during the time it takes to replace a sick worker (Oderda, 2003; Segel, 2006).

Under-reporting is an issue in determining the burden of foodborne illnesses, and results because of an unknown portion of ill individuals not seeking care. Even when care is sought, the causative pathogen may not be identified. This leads to an underestimation of cases in the databases used to monitor foodborne illnesses in a population (Majowicz et al., 2005; Thomas et al., 2006).

Variability in cost inventories and study methodologies limits meaningful interpretation and comparison of COI estimates (Akobundu et al., 2006; Segel, 2006; Larg et al., 2011). However, little is known about the relative effect these various factors have on a COI estimate. This is important not only for comparing existing estimates, but also when interpreting and designing new COI studies. Our objective was to explore the association between descriptive, component costs, methodological, and foodborne illness–related factors such as chronic sequelae and under-reporting with the estimated COI of NTS illness.

Methods

Study identification and selection

Studies that provided an overall COI estimate for NTS illness were selected from a scoping review (McLinden et al., 2014). Complete details of the search keywords, selection criteria, and data-charting forms are available elsewhere (McLinden et al., 2014). Briefly, following database searches (MEDLINE via PubMed and AGRICOLA between 1972 and 2012) and relevance screening, 74 references pertaining to the cost of foodborne illnesses were identified. Of these, 44 studies providing COI estimates for NTS illness were selected for the present study.

Data extraction and management

From each study, descriptive, component cost, study methodology, and foodborne illness–related data were extracted (Table 1). Studies were categorized broadly as including individual level costs or individual and societal level costs (Table 2). Studies were also categorized by the group or combination of costs that were included (direct, direct and indirect, direct, indirect and societal, or indirect and societal costs) and according to the total number of individual (direct and indirect) and societal level component cost categories, and the totals in each category. The detail in the extracted component cost data was as reported in the studies and thus, some categories represented more detailed subcategories. For example, a study that reported the broad category of “medical costs” would not have explicitly described more detailed direct medical cost–related categories.

Table 1.

Descriptive, Study Methodology, Component Cost, and Foodborne Illness–Related Variables Evaluated for an Association with the Estimated Cost of Nontyphoidal Salmonella spp. Illness Using Studies Identified from a Scoping Review

Descriptive information of studies (no. of estimates)	Component cost information (no. of estimates)
Country of estimation United States (16) England and Wales (4) Canada (2) New Zealand (2) United Kingdom (1) Spain (1) Netherlands (1) Denmark (1) Croatia (1) Taiwan (1) Region of estimation North America (18) Europe (9) Other (3) Year of estimation 1978–1982 (2) 1983–1987 (1) 1988–1992 (9) 1993–1997 (4) 1998–2002 (5) 2003–2007 (3) 2008–2012 (6) Number of foodborne pathogens included Single (11) Multiple (19)	Specific cost categories included Direct (4) Direct and indirect (9) Direct, indirect, and societal (16) Indirect and societal (1) Broad cost categories included Individual level costs (13) Individual and societal level costs (17) Overall number of component cost categories included (median: 6, range: 1–11) Number of direct component cost categories included (median: 3, range: 0–4) Number of indirect component cost categories included (median: 2, range: 0–3) Number of societal component cost categories included (median: 0.5, range: 0–4)

Descriptive information of studies (no. of estimates)

Component cost information (no. of estimates)

Country of estimation
United States (16)
England and Wales (4)
Canada (2)
New Zealand (2)
United Kingdom (1)
Spain (1)
Netherlands (1)
Denmark (1)
Croatia (1)
Taiwan (1)
Region of estimation
North America (18)
Europe (9)
Other (3)
Year of estimation
1978–1982 (2)
1983–1987 (1)
1988–1992 (9)
1993–1997 (4)
1998–2002 (5)
2003–2007 (3)
2008–2012 (6)
Number of foodborne pathogens included
Single (11)
Multiple (19)

Specific cost categories included
Direct (4)
Direct and indirect (9)
Direct, indirect, and societal (16)
Indirect and societal (1)
Broad cost categories included
Individual level costs (13)
Individual and societal level costs (17)
Overall number of component cost categories included (median: 6, range: 1–11)
Number of direct component cost categories included (median: 3, range: 0–4)
Number of indirect component cost categories included (median: 2, range: 0–3)
Number of societal component cost categories included (median: 0.5, range: 0–4)

Methodological information (no. of estimates)	Foodborne illness-related information (no. of estimates)
Epidemiological data used Prevalence-based (2) Incidence-based (28) Direct cost estimation method Top-down (aggregate approach) (19) Bottom-up (individual approach) (10) Did not estimate direct costs (1) Indirect cost estimation method Human capital approach (10) Willingness-to-pay approach (2) Linked approach (HCA-WTP) (13) Friction cost method (1) Did not estimate indirect costs (4)	Chronic sequelae costs included: Yes/No (12/18) Under-reported accounted for: Yes/No (15/15) Under-reporting accounted for: Explicitly not included (6) No information provided (9) Yes with multiplier value provided (7) Yes with no further detail provided (8)

HCA, human capital approach; WTP, willingness-to-pay.

Table 2.

Component Cost Categories Included in the 30 Cost of Nontyphoidal Salmonella spp. Illness Estimates Using Studies Identified from a Scoping Review ^a

Component cost category ^b	Number of times included in an estimate
Individual direct cost categories
Medical costs (described at this broad level only)	4
Treatment costs	18
Laboratory costs	9
Personnel costs	22
Hospital services costs	22
Community services costs	4
Long-term care services costs	4
Individual indirect cost categories
Productivity losses	24
Patient transportation costs (nonambulance)	11
Additional costs (intangible psychosocial costs, overhead costs)	21
Societal cost categories
Industry costs	12
Public health costs	12
Legal costs	6
Government (regulatory) costs	9

A study that included a more specific component in these broad categories was counted as including the broader category, with the exclusion of “medical costs.”

Refer to McLinden et al., (2014) for a detailed description of each component cost category.

The methodological information included (1) temporal relationship of the study (retrospective or prospective); (2) type of epidemiological data (prevalence or incidence); (3) direct cost–estimation method (top-down, bottom-up, and not applicable; i.e., studies that did not estimate direct costs); and (4) indirect cost–estimation method (HCA, WTP, linked human capital and WTP, friction cost method, and not applicable; i.e., studies that did not estimate indirect costs). It was also recorded whether a study included costs associated with chronic sequelae, and whether under-reporting of illness was considered. Sequelae were included if the researcher mentioned their inclusion anywhere in the study, or if a dollar value was provided for their burden. Under-reporting was categorized as (1) the study explicitly stated under-reporting was not accounted for (“no”); (2) the study provided no information on under-reporting (“no information provided”); (3) the study accounted for under-reporting but provided no additional information (“yes”); or (4) the study accounted for under-reporting and provided information on the level of under-reporting accounted for (“yes with detail on the magnitude,” i.e., provided detail on the multiplier value). The first two categories were grouped as “no” while the latter two were grouped as “yes.”

The estimated COI associated with NTS illness was extracted from each study and converted to a standardized outcome of COI per person per year in 2012 USD. This was done using country-specific Consumer Price Index (CPI) values to inflate COI estimates to the base year of January 2012 (Diewert, 2001; Bank of Canada, 2013; European Area Consumer Price Index, 2013; Reserve Bank of New Zealand, 2013; United States Department of Labor, 2013). For years prior to 2012, the average CPI for a given calendar year was used. Each estimate outside of the United States was then converted to January 2012 USD using a country-specific exchange rate (United Nations, 2013). The annual costs were converted to a per-capita figure by dividing the annual estimated cost of NTS illness by the location-specific population in January 2012 (or the available population estimate that was closest to this month and year) (United States Department of Commerce, 2013).

Statistical analysis

Linear regression was used to identify associations with descriptive, component cost, methodological, and foodborne illness–related variables with the annual cost of NTS illness per capita. The dependent variable was measured at the study level, and corresponded to the annual cost of NTS illness per capita in 2012 USD as calculated for each study estimate. The independent variables were coded categorically (e.g., country of estimation, year of estimation, cost categories included) and continuously (e.g., number of cost categories). A Shapiro-Wilk test of normality was performed on the outcome. Linearity of continuous independent variables and the outcome were assessed graphically. Then, each independent variable was tested for statistical associations in univariable analyses. For the direct- and indirect cost–estimation methodology variables, studies that did not estimate costs were included as a category. To provide an indication of collinearity between continuous variables, correlation coefficients were calculated (coefficient>|0.8| indicated collinearity). Variables significant at p≤0.2 in univariable analyses were entered simultaneously into a multivariable linear regression model. Factors were removed by order of the largest p-value (i.e., backward selection) until all factors remaining in the model were significant (p≤0.05). Following the removal of a categorical variable, a partial F-test was used to evaluate significance (p≤0.05). No interaction terms were included in the models. The mean cost of NTS illness for each independent categorical variable was calculated from the model coefficients. The normality of the residuals was assessed through examination of a histogram and a normal quantile plot. Homoscedasticity was qualitatively examined by plotting the standardized residuals against the linear predicted values. The amount of variation explained was evaluated using an adjusted R ²-value.

Following the creation of the final multivariable model, direct component cost category variables (binary-coded as included or excluded in a study) were explored in greater detail in a separate analysis, using the same model-building process as described above. All regression modeling was performed in Stata (Version 12: Intercooled; College Station, TX).

Results

Of the 44 studies identified, 15 studies were excluded from the analysis. Seven excluded studies estimated the cost of NTS illness attributable to a certain source only (e.g., the cost of poultry-borne Salmonella illnesses). Three studies did not provide the information required for standardizing the outcome (i.e., country or year of estimation), and a single study examined the costs associated with antibiotic-resistant NTS illnesses only. Four studies estimated the costs of individual outbreaks and were excluded from the analysis. One study provided two estimates as separate outcomes: one for the cost of NTS illness in Canada and the other in the United States. Therefore, 29 studies were included, providing 30 cost-of-NTS-illness estimates.

Table 1 displays descriptive, component cost category data, and methodological features for the 30 cost-of-NTS-illness estimates. Most estimates were from North America (n=18, 53%) and over half included both individual and societal level costs (n=17, 57%). The majority used incidence data (n=28, 93%), the top-down approach for estimating direct costs (n=19, 63%), and the linked human capital and WTP approach for estimating indirect costs (n=13, 43%). All estimates were based on retrospective study designs. The majority did not include the cost of chronic sequelae (n=18, 60%), while half included under-reporting (n=15, 50%). Table 2 indicates the number of studies that estimated a cost in each of the component cost categories. The most frequently included component cost category was indirect costs associated with productivity losses (n=24), followed by direct costs associated with hospital services (n=22) and personnel costs (n=22). Costs associated with community and long-term care services were seldom included (n=4).

The cost of NTS illness ranged from $0.01568 to $41.22 USD/person/year (2012). The mean cost was $10.37 USD/person/year (2012). A Shapiro-Wilk test of normality was not significant (p>0.05), indicating that the outcome was normally distributed. Correlation coefficients indicated collinearity between component sum variables that used the same data but were categorized differently (i.e., sums of the overall number of component categories and number of specific component categories). In univariable analyses using a liberal p-value (p≤0.2), the region (p=0.02), the number of pathogens included in the study (single or multiple; p=0.19), the overall number of component cost categories included in an estimate (p=0.07), and the number of direct component cost categories (p=0.05) were significant (Table 3). Whether under-reporting was accounted for (yes or no) was significant in univariable analyses (p=0.14), but the level of detail included for under-reporting was not. The inclusion of costs incurred due to chronic sequelae was significant (p<0.001). Following backward selection of variables significant at p≤0.2 in univariable analyses, the final multivariable model demonstrated significant positive associations with the number of direct component cost categories and chronic sequelae costs (inclusion/exclusion) with the cost of NTS illness (Table 4). Forty-five percent (adjusted R ²=0.4531) of the variation in the estimated cost of NTS illness could be explained by these two factors. The residuals of the multivariable model were normally distributed, and a plot of the standardized residuals against the linear predicted values displayed a constant variance (homoscedasticity). An additional analysis was undertaken to explore the associations between the specific direct component costs categories included in a study (binary-coded as included or excluded; Table 2) with the estimated cost of NTS illness (Table 5). Treatment costs, personnel costs, hospital services costs, and long-term care costs were significant at p≤0.2 in univariable analyses. Following backward selection of the direct component cost category variables to yield a multivariable model, the only component cost category that remained was long-term care costs (not shown).

Table 3.

Univariable Linear Regression Analyses for Associations of Descriptive, Study Methodology, Component Cost, and Foodborne Illness–Related Variables with the Estimated Cost of Nontyphoidal Salmonella spp. Illness Using Studies Identified from a Scoping Review

Variable	N	Coefficient (SE) ^a	95% CI	Mean cost (2012 USD)
Descriptive variables
Country of estimation		Overall p: 0.51	—	—
United States	16	Referent	7.84, 22.10	14.97
England and Wales	4	−11.79 (7.65)	−27.74, 4.16	3.18
Canada	2	8.43 (10.26)	−12.97, 29.83	23.40
New Zealand	2	−14.44 (10.26)	−35.84, 6.96	0.53
United Kingdom	1	−14.86 (14.10)	−44.27, 14.56	0.11
Spain	1	−14.49 (14.10)	−43.90, 14.93	0.48
Netherlands	1	−14.57 (14.10)	−43.98, 14.84	0.40
Denmark	1	−5.35 (14.10)	−34.77, 24.06	9.62
Croatia	1	−14.57 (14.10)	−43.99, 14.84	0.40
Taiwan	1	−14.82 (14.10)	−44.24, 14.59	0.15
Region of estimation		Overall p: 0.02	—	—
North America	18	Referent	10.06, 21.75	15.91
Europe	9	−13.27 (4.93)	−23.39, −3.15	2.64
Other	3	−15.50 (7.53)	−30.96, −0.04	0.41
Year of estimation		Overall p: 0.40	—	—
1978–1982	2	Referent	−10.16, 29.15	9.49
1983–1987	1	−4.52 (16.46)	−38.57, 29.53	4.97
1988–1992	9	−0.98 (10.51)	−22.71, 20.75	8.51
1993–1997	4	16.65 (11.64)	−7.42, 40.73	26.14
1998–2002	5	−2.93 (11.24)	−26.19, 20.33	6.56
2003–2007	3	−1.69 (12.27)	−27.07, 23.69	7.80
2008–2012	6	−1.18 (10.97)	−23.88, 21.52	8.31
Number of pathogens included		Overall p: 0.19	—	—
Single	11	Referent	−2.22, 14.28	6.03
Multiple	19	6.86 (5.06)	−3.51, 17.23	12.89

Methodological variables
Epidemiological data used		Overall p: 0.30	—	—
Incidence-based	28	Referent	5.84, 16.31	11.08
Prevalence-based	2	−10.54 (9.90)	−30.81, 9.74	0.54
Direct cost estimation method		Overall p: 0.74	—	—
Top-down	19	Referent	4.57, 17.65	11.11
Bottom-up	10	−1.10 (5.43)	−12.24, 10.03	10.01
Did not estimate direct costs	1	−11.04 (14.25)	−40.29, 18.20	0.07
Indirect cost estimation method		Overall p: 0.40	—	—
Human capital approach	10	Referent	0.89, 18.48	9.69
Willingness-to-pay approach	2	9.45 (10.46)	−12.09, 31.00	19.14
Linked HC–WTP approach	13	3.69 (5.68)	−8.01, 15.39	13.38
Friction cost method	1	−9.29 (14.16)	−38.46, 19.89	0.40
Did not estimate indirect costs	4	−9.24 (7.99)	−25.69, 7.22	0.45

Component cost variables
Specific cost categories included		Overall p: 0.32	—	—
Direct	4	Referent	−13.32, 14.23	0.45
Direct and indirect	9	9.91 (8.05)	−6.65, 26.47	10.36
Direct, indirect, and societal	16	13.06 (7.49)	−2.35, 28.46	13.51
Indirect and societal	1	−0.38 (14.99)	−31.19, 30.42	0.07
Broad cost categories included		Overall p: 0.29	—	—
Individual level costs	13	Referent	−0.36, 14.99	7.31
Individual and societal costs	17	5.40 (4.98)	−4.79, 15.60	12.71
Overall number of component cost categories included	—	Overall p: 0.07	—	—
(median: 6, range: 1–11)		1.46 (0.78)	−0.144, 3.06	—
Number of direct component cost categories included	—	Overall p: 0.05	—	—
(median: 3, range: 0–4)		3.80 (1.88)	−0.05, 7.66	—
Number of indirect component cost categories included	—	Overall p: 0.32	—	—
(median: 2, range: 0–3)		2.47 (2.44)	−2.54, 7.48	—
Number of societal component cost categories included	—	Overall p: 0.21	—	—
(median: 0.5, range: 0–4)		1.98 (1.53)	−1.17, 5.12	—

Foodborne illness–related variables
Chronic sequelae costs included		Overall p:<0.001	—	—
No	18	Referent	−1.54, 9.00	3.73
Yes	12	16.61 (4.07)	8.27, 24.95	20.34
Under-reporting accounted for		Overall p: 0.14	—	—
No	15	Referent	−0.35, 13.66	6.65
Yes	15	7.44 (4.84)	−2.47, 17.35	14.09
Under-reporting accounted for		Overall p: 0.50	—	—
No	6	Referent	−6.90, 16.07	4.59
No information provided	9	3.44 (7.21)	−11.38, 18.27	8.03
Yes with magnitude	7	9.29 (7.61)	−6.36, 24.95	13.88
accounted for provided
Yes with no further detail provided	8	9.69 (7.39)	−5.50, 24.88	14.28

Significance testing at p-value≤0.20 for selection into multivariable model.

SE, standard error of the estimate; CI, confidence interval; N, number of estimates; USD, United States dollar; HC, human capital; WTP, willingness-to-pay.

Table 4.

Multivariable Linear Regression Model of Significant Descriptive, Study Methodology, Component Cost, and Foodborne Illness–Related Variables (Identified Through Univariable Analyses) Associated with the Estimated Cost of Nontyphoidal Salmonella spp. Illness Using Studies Identified from a Scoping Review

Variable	N	Coefficient (SE) ^a	95% CI
Final regression model		Overall p:<0.001 Adjusted R² : 0.4531
Intercept		−6.45 (4.71)	−16.11, 3.21
Number of direct component cost categories included	—	p: 0.019	—
(median: 3, range: 0–4)		3.66 (1.47)	0.657, 6.67
Chronic sequelae costs included		p:<0.001	—
No	18	Referent	—
Yes	12	16.41 (3.74)	8.74, 24.08

Significance testing at p-value≤0.05.

SE, standard error of the estimate; CI, confidence interval; R ², coefficient of determination; N, number of estimates; USD, United States dollar.

Table 5.

Univariable Linear Regression Analyses for Associations of Direct Component Cost Category Variables with the Estimated Cost of Nontyphoidal Salmonella spp. Illness Using Studies Identified from a Scoping Review

Variable	N	Coefficient (SE) ^a	95% CI	Mean cost (2012 USD)
Direct cost category variables
Medical costs		Overall p: 0.53	—	—
(described at this broad level)		—	—	—
No	26	Referent	5.5, 16.50	10.99
Yes	4	–4.67 (7.36)	–19.74, 10.39	6.32
Treatment costs		Overall p: 0.11	—	—
No	12	Referent	–2.31, 13.25	5.47
Yes	18	8.17 (4.90)	–1.88, 18.21	13.64
Laboratory costs		Overall p: 0.92	—	—
No	21	Referent	4.03, 16.36	10.20
Yes	9	0.58 (5.49)	–10.67, 11.84	10.78
Personnel costs		Overall p: 0.09	—	—
No	8	Referent	–6.04, 12.94	3.45
Yes	22	9.44 (5.41)	–1.64, 20.52	12.89
Hospital services costs		Overall p: 0.18	—	—
No	8	Referent	–4.87, 14.47	4.80
Yes	22	7.60 (5.51)	–3.69, 18.89	12.40
Community services costs		Overall p: 0.46	—	—
No	26	Referent	5.63, 16.60	11.12
Yes	4	–5.55 (7.33)	–20.58, 9.47	5.57
Long-term care costs		Overall p: 0.06	—	—
No	26	Referent	3.34, 13.71	8.53
Yes	4	13.85 (6.93)	–0.35, 28.05	22.38

Significance testing at p-value≤0.20 for selection into multivariable model. Final multivariable model not shown as only one variable remained following backward elimination by p-value (i.e., long-term care costs).

SE, standard error of the estimate; CI, confidence interval; N, number of estimates; USD, United States dollar.

Discussion

The comparability of COI estimates is an important consideration in the economic burden of illness literature, and varying cost inventories and study methodologies are often cited as creating insular estimates in which meaningful comparisons cannot be made (Hodgson et al., 1982; Shiell, 1987; Tarricone, 2006; Larg et al., 2011). The present study explored the associations between descriptive, component cost, methodological, and foodborne illness–related factors with the estimated COI of NTS illness. The results indicate that although varying methodologies are being employed, the number of direct component cost categories included in an estimate, particularly long-term care costs, and chronic sequelae costs may be more important factors to consider when interpreting or comparing cost of foodborne illness studies.

The region of COI estimation was significant in an unconditional model, but not the country of estimation. However, although our data represented a census of the published cost of NTS illness literature, there was only a single COI estimate from most countries, making it difficult to evaluate the association of individual countries with the cost of NTS illness. Broad cost categories (individual vs. individual and societal) were not significantly associated with the estimated cost of NTS illness, nor were more specific cost categories of direct, direct and indirect, direct, indirect and societal, or indirect and societal costs. This finding was counterintuitive and contradictory to findings in the literature, suggesting that studies including different cost inventories would have significantly different overall estimated costs (Akobundu et al., 2006; Segel, 2006; Larg et al., 2011). However, a study that included a single direct and a single indirect component cost was categorized as including direct and indirect costs, as was a study that included multiple direct and indirect component costs (McLinden et al., 2014). Therefore, our groupings do not capture the relative number of component costs included in each category, much less the specific component costs included. For that reason, each study also was classified by the number of overall component costs categories and by the number of direct, indirect, and societal component cost categories. This method captured component cost–related information in greater detail, but had the disadvantage of weighting every category as equal.

The finding that the overall number of component cost categories was not significant was unexpected. However, the number of direct component cost categories showed a significant positive association with the cost of NTS illness in the final multivariable model. This factor was explored further (in a separate analysis) by examining the association between specific direct component cost categories with the cost of NTS illness. Following backward selection in a multivariable model (of the direct component cost categories significant at p≤0.2 in univariable analyses), long-term care costs was the only remaining category. However, the number of component cost categories included was a sum and still did not capture the specific component costs in each study. Additionally, because the detail in the extracted component cost data was at the level reported in the studies, a study that included “medical costs” may have included other direct medical cost-related categories as well. Therefore, there is a need for greater detail of reporting for component cost information, which will provide a greater understanding of what types of costs are being included in an estimate. Exploring whether the inclusion or exclusion of specific component costs are associated with the estimated cost of foodborne illnesses could be conducted to further evaluate these factors.

There is concern in the COI literature regarding the inability to meaningfully interpret and compare COI estimates due to varying methodologies (Akobundu et al., 2006; Segel, 2006; Larg et al., 2011). However, our finding that no methodological variables were significant indicates that methodologies may not be as important as other factors, such as direct component cost inventories, and more specifically, long-term care costs. However, statistical power was low, with only two estimates using prevalence data and few that used WTP (n=2) and friction cost methods (n=1) for estimating indirect costs.

Under-reporting was not significantly associated with the cost of NTS illness, which was unexpected as under-reporting is a major consideration when estimating the burden of foodborne illnesses (Majowicz et al., 2005; Thomas et al., 2006). However, the variable categories used explored this factor superficially. We did not attempt to determine associations between the magnitudes of under-reporting with the cost of NTS illness, as most studies did not provide this information. An important finding is that the inclusion or exclusion of chronic sequelae costs was significantly associated with the cost of NTS in both the univariable and multivariable models. This corresponded with previous research that illustrated chronic sequelae associated with foodborne illness added substantially to the burden and economic losses associated with these illnesses (Buzby et al., 1997; Lindsay, 1997; Lindqvist et al., 2001; Mangen et al., 2005).

Estimates of the cost of NTS illness varied widely, and only 45% of the variation in the cost of NTS illness was explained by the number of direct costs and whether the estimate included chronic sequelae. It is possible that the factors evaluated, particularly the component cost-related factors, were not refined enough. Therefore, the variation in the estimates may be due to the inclusion or exclusion of specific component costs within the broader categories (which was not explored in the present study), or due to additional unmeasured factors.

A limitation of these analyses was the small number of studies that provided the estimated annual national costs of NTS illness. This meant that several categories of variables contained very few observations, resulting in low statistical power to detect associations and large standard errors (i.e., wide 95% confidence intervals). However, this analysis used a census of the available published COI estimates (McLinden et al., 2014). Therefore, use of the models to predict the cost of NTS illness is not recommended, as the goal of this study was to identify and describe the potential association of factors with a COI estimate. Lastly, a large number of variables were also considered, some using the same data but categorized in different ways, increasing the likelihood that the associations identified were type-I errors.

Conclusions

Our objective was to investigate associations between descriptive, component cost, methodological, and foodborne illness–related factors with the estimated cost of NTS illness. Our findings indicated that study methodology may not be as influential as other factors, such as the number of direct component cost categories included in an estimate, particularly long-term care costs, and costs incurred due to chronic sequelae. Therefore, these may be the most important factors to consider when designing, interpreting, and comparing cost of foodborne illness studies.

Footnotes

Acknowledgments

Funding for this project was provided by the Canadian Institutes of Health Research/Public Health Agency of Canada (CIHR/PHAC) Applied Public Health Research Chair (awarded to JMS) and the Public Health Agency of Canada. Stipend funding for TM was provided by the Ontario Graduate Scholarship (OGS).

Disclosure Statement

No competing financial interests exist.

References

Akobundu

, Ju

, Blatt

, Mullins

. Cost-of-illness studies: A review of current methods. Pharmacoeconomics, 2006; 24:869–890.

Bank of Canada. Rates and Statistics—Inflation Calculator, 2013. Available at: www.bankofcanada.ca/rates/related/inflation-calculator, accessed June 2013 .

Buzby

, Roberts

. Economic costs and trade impacts of microbial foodborne illness. World Health Stat Q, 1997; 50:57–66.

Byford

, Torgerson

, Raftery

. Economic note: Cost of illness studies. Br Med J, 2000; 320:1335.

Centers for Disease Control and Prevention. Cost Analysis—Cost of Illness, 2005. Available at: www.cdc.gov/owcd/eet/Cost/fixed/3.html, accessed June 2013 .

Chen

, Li

, Hsieh

, Chang

, Lee

, Wu

, Lee

, Shih

, Ko

. Epidemiology, disease spectrum and economic burden of non-typhoidal Salmonella infections in Taiwan, 2006–2008. Epidemiol Infect, 2012; 140:2256–2263.

Diewert

. The Consumer Price Index and index number purpose. J Econ Social Meas, 2001; 27:167–248.

European Area Consumer Price Index. Historical Consumer Price Index (CPI) for Euro Area. 2013. Available at: www.rateinflation.com/consumer-price-index/euro-area-historical-cpi, accessed June 2013 .

Gadiel

. The economic cost of foodborne disease in New Zealand, 2010. Available at: www.foodsafety.govt.nz/elibrary/industry/economic-cost-foodborne-disease/foodborne-disease.pdf, accessed June 2013 .

10.

Health Canada. Economic Burden of Illness in Canada (EBIC), 1998. Ottawa, Canada: Policy Research Division, Strategic Policy Directorate, Population and Public Health Branch, 1998.

11.

Hesmat

. An Overview of Managerial Economics in the Health Care System. New York: Delmar–Thomson Learning Publishing, 2001.

12.

Hodgson

, Meiners

. Cost-of-illness methodology: A guide to current practices and procedures. Health Society, 1982; 60:429–462.

13.

Hoffmann

, Batz

, Morris

. Annual cost of illness and quality-adjusted life year losses in the United States due to 14 foodborne pathogens. J Food Prot, 2012; 75:1292–1302.

14.

Lake

, Cressey

, Campbell

, Oakley

. Risk ranking for foodborne microbial hazards in New Zealand: Burden of disease estimates. Risk Analysis, 2010; 30:743–752.

15.

Landefeld

, Seskin

. The economic value of life: Linking theory to practice. Am J Public Health, 1982; 72:555–566.

16.

Larg

, Moss

. Cost-of-illness studies: A guide to critical evaluation. Pharmacoeconomics, 2011; 29:653–671.

17.

Lindqvist

, Andersson

, Lindback

, Wegscheider

, Eriksson

, Tidestrom

, Lagerqvist-Widh

, Hedlund

, Lofdahl

, Svensson

, Norinder

. A one-year study of foodborne illnesses in the municipality of Uppsala, Sweden. Emerging Infect Dis, 2001; 7:588–592.

18.

Lindsay

. Chronic sequelae of foodborne disease. Emerging Infect Dis, 1997; 3:443–452.

19.

Majowicz

, Edge

, Fazil

, McNab

, Dore

, Sockett

, Flint

, Middleton

, McEwen

, Wilson

. Estimating the under–reporting rate for infectious gastrointestinal illness in Ontario. Can J Public Health, 2005; 96:178–181.

20.

Mangen

MJJ

, Havelaar

, Bersen

RAJAM

, Van Koningsveld

, De Wit

. The costs of human Campylobacter infections and sequelae in the Netherlands: A DALY and cost-of-illness approach. Acta Agric Scand Section, 2005; 2:35–51.

21.

McLinden

, Sargeant

, Thomas

, Papadopoulos

, Fazil

. Component costs of foodborne illness: A scoping review. BMC Public Health, 2014; 14:509.

22.

McPherson

, Kirk

, Raupach

, Combs

, Butler

. Economic costs of Shiga toxin–producing Escherichia coli infection in Australia. Foodborne Pathog Dis, 2011; 8:55–62.

23.

Mishan

. Evaluation of life and limb; a theoretical approach. Journal of Political Economics, 1971; 79:687–705.

24.

Oderda

. Cost-of-illness studies. J Pain Palliative Care Pharmacol, 2003; 17:59–62.

25.

Rabsch

, Tschape

, Baumler

. Non-typhoidal salmonellosis: Emerging problems. Microbes Infect, 2001; 3:237–247.

26.

Reserve Bank of New Zealand. Monetary policy—Inflation calculator, 2013. New Zealand. Available at: www.rbnz.govt.nz/monetary_policy/inflation_calculator, accessed June 2013 .

27.

Rice

. Estimating the cost of illness. Am J Public Health, 1966; 57:424–440.

28.

Rice

. Cost-of-illness studies: Fact or fiction?. Lancet, 1994; 344:1519–1520.

29.

Scallan

, Hoekstra

, Angulo

, Tauxe

, Widdowson

, Roy

, Jones

, Griffin

. Foodborne illness acquired in the United States—Major pathogens. Emerging Infect Dis, 2011; 17:7–15.

30.

Scharff

. Economic burden from health losses due to foodborne illness in the United States. J Food Prot, 2012; 75:123–131.

31.

Schelling

. The Life You Save May Be Your Own: Problems in Public Expenditure Analysis. Chase SB edition. Washington, DC: Brookings Institution, 1968.

32.

Scitovsky

, Rice

. Estimates of the direct and indirect costs of acquired immunodeficiency syndrome in the United States: 1985, 1986, and 1991. Public Health Rep, 1987; 102:5–17.

33.

Segel

. Cost-of-Illness Studies—A Primer. RTI International. Durham, NC: RTI-UNC Center of Excellence in Health Promotion Economics, 2006, pp. 1–39.

34.

Shiell

, Gerard

, Donaldson

. Cost of illness studies: An aid to decision making?. Health Policy, 1987; 8:317–323.

35.

Tarricone

. Cost-of-illness analysis: What room in health economics?. Health Policy, 2006; 77:51–63.

36.

Thomas

, Majowicz

, Sockett

, Fazil

, Pollari

, Doré

, Flint

, Edge

. Estimated numbers of community cases of illness due to Salmonella, Campylobacter and verotoxigenic Escherichia coli: Pathogen-specific community rates. Can J Infect Dis Med Microbiol, 2006; 17:229–234.

37.

United Nations. Treasury—UN operational rates of exchange, 2013. Available at:www.treasury.un.org/operationalrates/OperationalRates.aspx, accessed June 2013 .

38.

United States Department of Commerce. United States Census Bureau—International Data Base, 2013. Available at:www.census.gov/population/international/data/idb/informationGateway.php, accessed June 2013 .

39.

United States Department of Labor. Bureau of Labor Statistics—Consumer Price Index, 2013. Available at: www.bls.gov/cpi/cpifaq.htm, accessed June 2013 .