Causation models validity

If several safety professionals investigate a given hazards-related incident, they should identify the same causal factors, with minimum variation. That is unlikely if the thought processes they use have greatly different foundations. At least 25 causation models have been published. Since many of them conflict, all of them cannot be valid. A review of some of them is followed by a discussion of principles that should be contained in a causation model. 

Dr. Roger L. Brauer, Executive Director of the Board of Certified Safety Professionals, made the following comment about proving the validity of causation models ... 

Judicial decisions in nonregulatory contexts such as toxic tort and product liability suits are likewise inconsistent in their consideration of the linear, no threshold model. As in the regulatory context, most cases find no problem with an expert s reliance on a risk assessment using the linear model. In a handful of cases, however, the court rejects reliance on a linear dose-response assumption. Eor example, one court in addressing the cancer risks from a low concentration of benzene in Perrier held that there is no scientific evidence that the linear no-safe threshold analysis is an acceptable scientific technique used by experts in determining causation in an individual instance (Sutera 1997). Another court decision concluded that [t]he linear non-threshold model cannot be falsified, nor can it be validated. To the extent that it has been subjected to peer review and publication, it has been rejected by the overwhelming majority of the scientific community. It has no known or potential rate of error. It is merely an hypothesis (Whiting 1995). The inconsistency and unpredictability of judicial review of risk assessments adds an additional element of uncertainty into the risk assessment process. 

The QSAR algorithm establishes correlation between the studied response and the molecular descriptors, but recently some concerns have been raised in the lit-erature, emphasizing that correlation between variables does not automatically imply that one causes the other and that chance correlation could occur, mainly if not understandable descriptors are used. However, correlation is a fundamental requirement for causation. The best way to exclude chance correlation is to carefully verify the statistical predictivity of the QSAR models by their validation, as requested by OECD principle 4, also externally on new chemicals, and by scrambling of the response, additionally, if possible, to mechanistically interpret the molecular descriptor (OECD Principle 5). If the correlation is confirmed after rigorous verification, it has a reason for existing and the problem is one of the human mind if the cause is not discovered or understood. 

The use of standardized terminology is required across all industries and professions so that people working in prevention will have better access to relevant and meaningful data on causation factors. Effective risk cormnunication expressed through clearly defined terms and simplified models is needed to assist in meaningful event analysis. The establishment of safe systems of work is not possible until our safety initiatives come to be based on valid and rehable data. 

---