#### Descriptive summary

Data were available for 792 flocks on 33 farms, and of these, 217 (27.4%) tested positive for *Campylobacter*.
The median number of flocks per farm was 14 (mean 24, range 1 to 146),
and the median number of positive flocks per farm was 3 (mean 7, range
0 to 55). The proportion of positive flocks per farm ranged from 0 to
75%, with a median and mean of 25%. Three farms did not have any
positive flocks; these were primarily smaller farms that each raised a
total of 1 to 9 flocks during the four summer seasons of the study.
Other domestic livestock on farms included cattle only (1 farm), pigs
only (1 farm), sheep only (2 farms), and sheep plus cattle and/or
horses (5 farms).

Of the 792 flocks raised during the four summer seasons, the total
production was 5,828,772 broilers. This figure was slightly less
(5,659,534 broilers) for the 28 farms (758 flocks) included in the
multivariable analyses. The median age at slaughter of flocks raised
during the summer seasons was 37 days (mean 37, range 31 to 100). The
age distribution for flocks included in the multivariable analyses was
similar although the maximum age was 63 days. The number of houses per
farm ranged from 1 to 15 (median 2, mean 2.5). Individual flocks ranged
in size from 604 to 21,772 broilers (median 6,275, mean 7,366). A large
proportion of flocks (72%) were slaughtered in one catch lot. For
flocks with more than one catch lot, the mean catch lot size was 5,065
broilers (range 330 to 14,867). Each catch lot was sampled at
slaughter. Of the 217 positive flocks, 14 flocks were slaughtered in
three catch lots with four samples per catch lot for a total of 12
samples per flock, 46 flocks were slaughtered in two catch lots with
four samples per catch lot for a total of eight samples per flock, and
the remaining 157 flocks were slaughtered in one catch lot with four
samples per flock. On the basis of catch lot sampling, out of 291 catch
lots, 266 were positive in all samples (91.4%), 2 were positive in
three samples (0.7%), 6 were positive in two samples (2.1%), and 17
were positive in one sample (5.8%). On a flock basis, 14 of the 217
positive flocks were positive in only one pooled sample, likely
indicating early stages of flock colonization.

The characteristics of farms excluded from the analyses due to missing data for one or more variables are shown in Table 3.
There were no obvious patterns among the excluded farms other than none
of the farms raised other livestock and all had either one or two
houses. The proportion of positive flocks on these farms ranged from
25% to 75%.

**Table 3.** Characteristics of farms excluded from the farm-level analyses due to missing data

#### Multivariable analysis

The variables "manure spread on fields in summer season" and "manure
spread on fields in winter season" were strongly and positively
correlated with each other (τ_{b }=
0.86). Of the 28 farms included in the multivariable analysis, there
were 9 farms that spread manure in both summer and winter, 17 farms
that did not spread manure in either season, and 2 farms that spread
manure in the summer but not in the winter.

Coefficients and p-values for the variables in each model are presented in Table 4.
For categorical variables, exponentiation of the coefficient represents
the increase (positive coefficient) or decrease (negative coefficient)
in the risk of *Campylobacter *when the factor was present on
the farm compared to when it was not present on the farm. For example,
using the coefficient of 0.92 from the manual backward selection model
using "manure spread in summer", the risk of a flock being colonized
with *Campylobacter *was 2.5 times higher (e.g. e^{0.92 }=
2.5) on farms that spread manure on fields in the summer season
compared to farms that did not spread manure in the summer.
Exponentiation of the coefficient for the continuous variables
represents the increase in the risk of *Campylobacter *as the
median flock size increased by 1,000 birds, and the increase in risk
for each additional house on the farm (see discussion). The p-values in
Table 4
represent the probability that the increase or decrease in risk was due
to chance alone. For example, the p-value of 0.025 for manure spreading
in the summer indicates that there was a 2.5% probability that the
observed increased risk of *Campylobacter *colonization was due to chance.

**Table 4.** Six logistic models for farm-level factors associated with *Campylobacter *in broilers in Iceland (n = 28^{a})

For each variable in Table 4,
a range of coefficients and p-values are presented. The values differ
depending on the model selection method. The presence of data in the
table is an indication that the variable was associated with *Campylobacter *colonization
in the respective model, whereas the absence of data indicates that the
variable was not associated with flock colonization (i.e. the variable
was either removed (backward-type models) or it was not eligible for
addition (forward-type methods)). The variables are listed in
descending order, such that factors identified as being associated with
*Campylobacter *in all models are the top of the table. For
example, increasing median flock size was identified as a strong risk
factor in all six models, whereas an all-in-all-out policy was not a
significant predictor in any of the models. Factors that were
significantly associated with colonization regardless of modelling
approach could be considered to have a greater relative importance in
the epidemiology of *Campylobacter *on broiler farms in Iceland.

In general, the factors associated with an increased risk of *Campylobacter *were
increasing median flock size, spreading manure on the farm in the
winter, and increasing the number of broiler houses on the farm.
Protective factors included the use of official or official treated
water on the farm compared to the use of non-official untreated water,
storing manure on the farm at any time of year, and the presence of
other domestic livestock on the farm.

In the automated forward selection and forward stepwise models, one
farm had a large residual (standardized Pearson's residual = 3.2)
relative to the residuals of the other farms. The characteristics of
this farm were: non-official water, one house, an all-in-all-out
system, manure was spread and stored at all times of the year, absence
of other livestock and poultry, and a mean flock size of 4,579 birds.
Although this farm had a much higher proportion (7/15) of positive
flocks than predicted (2.5/15), it did not have undue influence on the
models.