Application and validation of a statistically derived risk-based sampling plan to improve efficiency of inspection and enforcement

- Binomial probability statistics were applied to design risk-based plan of work.
- Predictability of statistical sampling models was validated using an actual data.
- Statistical sampling protocol provides unbiased estimates of population attributes.
- Sampling protocol would be optimized through the validation process for improvement.

The statistically derived risk-based sampling plan for surveillance sample assignments of chemical and biological hazards was designed using binomial probability distribution. The binomial statistics was applied to the past 3-year data to estimate a confidence interval and a sample size aiming to improve efficiency and effectiveness of the agency's sampling and inspectional activities. The accuracy of the statistical models and computed estimates were validated in the following years. The ranges of confidence interval and sample size appeared to be significantly influenced by a level of the violation rate of feed product samples, an acceptable error, a number of the analyzed samples, and a statistical significance level. The violation rates of feed products for target analytes (aflatoxins, fumonisins, Salmonella, and dioxin) in the validation data were lower than those of the average 3-year data in most feed products. Besides, the actual violation rates of the validation samples did not exactly fall within the anticipated range of the confidence interval estimates. Such a discrepancy is considered introduced by several factors such as sample size adequacy, skewed distribution of a target analyte in feed products, and unique analyte/product combination. The overall study results indicate that the risk-based plan of work would provide a more effective and efficient risk management tool to help improve the oversight of the feed industry and the compliance to feed safety standards.