Risk for infection with highly pathogenic influenza a virus in chickens, Hong Kong, 2002

Emerging Infectious Diseases, March, 2007 by Nina Y. Kung, Roger S. Morris, Nigel R. Perkins, Les D. Sims, Trevor M. Ellis, Lucy Bissett, Mary Chow, Ken F. Shortridge, Yi Guan, Malik J.S. Peirist

Risk Factors for Infection of Farms

Univariate Analysis

Statistical comparisons were not done for 9 of the variables from the questionnaire because of either uniformity of response across all farms or excessive missing data. Summary information for farm area, stock numbers, and shed numbers on each farm are presented in Table 2. We performed [chi square] tests of association on 60 variables in the univariate analysis. Table 3 shows the variables that were associated with a p value <0.25 in the univariate analysis. Affected farms were concentrated in a small number of districts compared with controls, which were more widely distributed across districts (OR 123.0, p<0.01).

Other factors positively associated with case farms: number of chickens on farm; stock density; death rate higher for birds >30 days of age than for younger birds (OR 7.40, p = 0.02); survival rate at 1-30 days of age (OR 1.54, p<0.01); medication use during January-February 2002 (OR 4.67, p = 0.02); whether chickens were sold directly to retail markets (OR 11.15, p<0.01); whether automatic manure scrapers were installed (OR 4.55, p = 0.02); whether persons from retail markets visited during January-February 2002 (OR 10.00, p = 0.01); and whether a visitor went inside the shed during this period (OR 3.94, p = 0.04). Factors that had ORs significantly <1.0 for case farms were reports of wild birds eating in the chicken feed trough (OR 0.20, p = 0.04), farm owner living on farm (OR 0.05, p<0.01), and visitors from another chicken farm during January-February 2002 (OR 0.23, p = 0.02).

Multivariate Analysis

Three alternative final models were identified from the model-building procedures, each containing variables that had significant p values (Table 4). Three variables appeared in all models: owner lives off farm, age group with highest death rate at >30 days old, and sale of chickens direct to retail markets. Each model had 1 additional variable, which was different for the 3 models; wild birds in feed trough (protective, model A), number of chickens on farm (model B), and relative working in poultry industry (model C). On the basis of the Hosmer-Lemeshow statistic, model A provided the best fit to the data, while by the adjusted [R.sup.2] statistic, model B had the highest explanatory value. In this model, farms with a nonresident owner were 12.8x more likely to be a case farm; farms that sold chicken directly to retail markets were 30.3x more likely to be a case farm; farms with highest death rate in birds >30 days old were 20.5x more likely to be a case farm; and farms with higher chicken numbers were 1.1x more likely to be a case farm.

Residual components from all 3 models showed 1 farm (Farm ID 19) with a large standardized residual. This was a farm in Pak Sha area where influenza A (H5N1) was isolated on February 20, 2002, but the model predicted it would be a control farm. Farm 19 imported day-old chicks from China during mid-February and sold some chicks 10 days later to another nearby farm (case farm, Farm ID 20). Infection may have entered this farm directly with imported birds; therefore, it did not share risk factors with the other case farms.

Discussion

We describe the use of a combination of descriptive and analytic epidemiologic techniques to investigate possible risk factors associated with the 2002 influenza A (H5N1) outbreak in Hong Kong. The small sample size limited the number of risk factors we could combine in a multivariable model and prevented consideration of interaction terms. Models containing [greater than or equal to]4 variables and models containing interaction terms either did not converge or showed evidence of multicollinearity. Inspection of counts for combinations of explanatory variables indicated that these occurrences were likely to be the result of zero counts. Because the study included only 16 case farms and related directly to transmission processes that exposed these specific farms, our inferences apply only to the specific circumstances of this outbreak, and caution should be used in applying these findings to other situations.