dence serving as the basis for the proposed system and praises FSIS for its resilience as it improves the proposal with public comments. In general, the committee found it a challenge to evaluate the adequacy of indicators of process control to rank establishments and allocate agency inspection resources without a clear understanding of the rationale for the general approach. The committee’s deliberations, based on its review of the report PHRBIS, open meetings, and personal communications with FSIS, resulted in the following findings:
The proposed inspection system consists of two components: one based on process control indicators and a second based on public health impact. The committee was tasked to review only the first component, but found it difficult to completely exclude deliberations on indicators of public health impact.
The report PHRBIS lacks details that are crucial to its evaluation. For example, the description of the algorithm, the scientific basis for the algorithm, the scientific basis for the use of the process indicators, the description and analysis of data, and the use of the process control indicator algorithm as it is integrated into the overall inspection system are not clearly articulated in the FSIS technical report.
The specific activities assigned to the three levels of inspection are not explicated. Likewise, the process of decision making to transfer a plant into a different level of inspection (LOI) (e.g., from LOI 2 to LOI 1) is not well defined. Further, it is not clear for how long or how frequently a plant in category LOI 2 or LOI 3 will be subject to an in-depth inspection or how these LOI designations relate to current regulatory requirements.
Key terms of the algorithm, such as “process control indicators,” are not well defined. In addition, the proposed algorithm assigns the same weight to all process indicators, even though they vary in their ability to predict loss of process control. For example, some indicators may predict future loss of control (e.g., the rate of health-related noncompliance records [NRs]), but others might only reflect past loss of control (e.g., recalls). For some foods, no adequate process control indicator is proposed.
The statistical analysis that was conducted to find associations between proposed process control indicators—lift analysis—is a data-mining tool appropriate for use in finding initial associations among events that occur infrequently. However, the identi-