IH’s Dirty Little Secret

Blogging is a remarkable media; indeed, this blog has opened communication among colleagues in a manner that I did not anticipate.  This week’s blog represents a prime example.

You may remember that two of the last three blogs deal with the subject of sampling statistics in which I tried to make the point that exposure is naturally quite variable and that by just taking one or a few samples, we can be tricked as to whether a workplace is really in compliance relative to an OEL.  The blogs advocated using good statistical software such as IH STAT to help interpret the data. 

In response to these blogs I heard from a respected colleague who has a lot more field experience as an IH than I do.   Indeed, I would describe him as a seasoned and very thoughtful exposure assessor.   Because he does not want to jeopardize his future employment opportunities he has asked that I keep his comments anonymous which I am pasting below: 

“I think that the AIHA as an association has failed to a great degree to educate industry about this particular failing of our profession [the severe limitations of small sample size].  I'm not aware of a single piece that has ever been produced that attempts to educate the layperson (most business owners are) about the limitations of small sample sets.  

I also think that a pertinent question has been missed by the IH community.  That is - a reality-based question:  would a chemical-handling business be better off by having a professional IH visit and observe their processes, and conduct a single sample vs. no visit at all?  I have lost sleep over this.  Having worked with hundreds of small (and medium, large) businesses over the years, I recognized this issue soon after attending my first continuing education course on statistics back in the early '90s.

The reality is that a sizable proportion of chemical handling businesses will NEVER find their way to taking ANY samples.   Of the ones that do, VERY FEW would ever entertain the idea of repeat sampling.  This bothered me so much, that for a while I started giving proposals to clients with an explanation of my own regarding statistical limitations, and providing them two options in proposals - one with a single visit/sampling event, a second (obviously more expensive) option that would provide more confidence in conclusions reached - and leave the decision with them. Maybe I'm not a very concise scientific communicator, or just not a good salesman, but NONE of them ever opted for the multiple sampling option.

Another issue addresses a level playing field for consultants. This is really important because if you look at the demographics of the AIHA, consultants are the single largest group, and likely perform more than half of all exposure assessments. If I were to immediately insist that my new customers only go the repetitive sampling route, I would be out of business as a consultant tomorrow.  There will always be someone else to step in that gives the customer what they want. Luckily, my larger, more sophisticated clients will opt for building data validation over time through repeated sampling events, but they are still few and far between.

Lastly, and sadly – OSHA regulations simply don’t address the issue adequately.  Other than a few of the substance specific regulations for repeat sampling, there is no legal impetus for employers to provide a statistically sound approach to exposure assessment.

Hey - thanks for letting me spout off!  This is truly the 'dirty little secret' of the IH field, and in my opinion - no amount of shaming is going to solve this.” 

I appreciate this expression of opinion and truth and I know this person enough to believe he is correct.

From my perspective, I believe that taking one sample is MUCH better than taking none as long as you attempt to factor in the very large uncertainty associated with such action.   Assuming that a geometric standard deviation of about 2 is typical for workplace exposures, then, in my opinion, a single sample value that comes out to be less than 10% of an OEL is a pretty good indication that the average exposure will be below the OEL at least 90-95% of the time.  If true, OEL/10 could become the new action level for single samples.   Indeed, I would love to have some sharp statistical minds to look over this suggested approach and comment.

Another way of appropriately using but a single sample is to be sure you monitor reasonable worst case.  That is, if possible, be sure that your single sample is capturing reasonably foreseeable worst case.  Such factors as maximum product rate and/or minimum ventilation rate would be examples of this approach. 

My sense of all of this is that it seems to point directly to the value of modeling in determining actually what scenarios and conditions should be monitored using our precious resources.