Apples, Arsenic, and Risk - Part 6: Theory and Reality

My Mom likes to resend emails she receives.  This is the old-timers with computer's 21st century version of sending her kids and grand kids clippings from the newspaper or Reader's Digest.

Her latest email is titled "Here are some facts that of course you already knew...Of  Course..... :)." At the beginning of the email she writes:

If you find that any of these are wrong you don't have to email me back and tell me.

You see, my mom likes to live in a world where being told "that's not true" does not exist.  As you can probably tell from this blog, I and my two boys, tell her these "facts" are not true quite a bit.

However, there is a difference between believing as fact that "Chinese headbands are made from used condoms" and Water with 10 ppb of arsenic has an excess cancer risk of one in 500 as reported by Consumer Reports.

The problem with health related information is what someone will do with this information passed off as a fact.  Not buying hairbands does not impact health.  That's also the point the TCEQ was making in their response to the EPA over the use of this new cancer slope factor for arsenic.

[I]f erring on the side of conservatism significantly overestimates risk or hazard and is not  fully  justified, then harm to public health may result from diverting public, industry, and government attention and resources away from chemicals which may represent more of a public health risk at environmental levels.

A while back I wrote a number of posts on laundered shop towels and their risk of contamination to workers who use them.  I stated what I felt were a number of "rules" we public health professionals should abide by.  Rule number five was:

Always make sure the model and equations reflect reality.

For Consumer Reports to write "For water with 10 ppb of arsenic, the excess cancer risk is one in 500," and for the EPA to generate data showing a Risk for Male Bladder Cancer of three excess cancers in 1000 if exposed daily for a lifetime to 1 µg of arsenic per liter of drinking water, one must have confidence in the numbers used to calculate that risk value.

In other words, should we trust the value spat out by our model?

Models are what we use to predict an outcome.  So regarding the development of a model, here are some rules that I think we should consider:

1. Does the formula represent what is going on in the real world?
2. Can the parameters that make up the formula be derived from sound data?
3. Are the assumptions used to generate the values sound?
4. Does the model's prediction reflect what we see in the real world?

For example, lets say we want to estimate - or predict - how many days it will take to get from Los Angeles to New York City.  The more variables in the formula, the better the prediction.  So we could look at the number of rest stops, fuel stops, speed limits, road closures and work, etc.  In it's simplest form, the prediction of how long is dependent on how far you want to go divided by the top speed times the amount of time driving at that speed.

How long = Total Distance / (MPH x Hours Driving)

That formula is sound, it may not be the best we can do, but it does represent what is going on in the real world and the parameters can be derived from sound data.

We know how many miles it is from Los Angeles to New York City, we can estimate the MPH we will drive, and we can estimate the amount of hours we will drive each day.

How Long = 2790 miles / (70 x 8) = 5 days.

Does 5 days reflect reality?  Yeah, it does, but it is also based on the assumption that we can drive for 8 hours in a day at an average speed of 70 mph.  That's doable, so the model's prediction reflects a reality - a possibility of that outcome.

Now lets say I have never driven a car before, in fact no one has, but our data shows that our car could go as fast as 120 mph.  Since we have never driven before, we could also assume that a person could drive for 22 hours - allowing two hours for fuel and bathroom breaks.

How long = 2790 miles /(120 x 22) = 1.05 days.

Is the number 1.05 days correct?  Yes, based on the values used in the formula.  Would you accept it as representing what is possible in the real world?  No.  Is it possible to get from Los Angeles to New York in a car in about one day?  Yes...but only if you drive for 22 hours and average 120 MPH.  For 1 day to be a reasonable - or sound - prediction, all the variables used to calculate it must be sound. If the output of one day does not reflect the real world, then either the model is incorrect or the values used to calculate the prediction are not valid.

That's the fundamental problem with risk assessment models.  It uses complex formulas to estimate - predict - a risk.  And that prediction of risk is only possible if the data is correct and the formula is sound.  Look at the equation used to determine the chronic daily intake (CDI) of a chemical for a resident (person) eating fish from a contaminated water body:

Source

One or two incorrect values entered into the calculation will skew the results - too much - or - too little.  And with risk, we use values that are based on assumptions that are very conservative; such as 2 liters per day for 70 years.  Or in the case of shop towels, a worker touching their hand that handled a shop towel to their mouth 117,600 times over their lifetime.

In the equation above, they assume the person will eat fish from the water body 350 days a year for 30 years.  Is C-fish the concentration in the whole fish or the part of the fish that is consumed?  Is that CDI possible? Yes.  Probable?  No.  Unfortunately the public will be told of the risk from eating fish from that area that is dependent on a possibility that is not probable.

It's the best we got...the people demand a number...and this is how we give it to them. And Consumer Reports will print that risk as if it will take place.

When it's all said and done, you must look at the prediction - the outcome - the value spat out by the formula - and ask:

Does the model's prediction reflect what we see in the real world?

That's what the TCEQ did when they looked at the cancer slope factor proposed by the EPA that was used for bladder cancer:

Drinking water in the US generally contains an average of 2 μg/L of arsenic (ATSDR 2007). Based on final draft SFo estimates, USEPA indicates that drinking water concentrations corresponding to 1 in 10,000 combined cancer risks for males and females are 0.21 and 0.14 μg/L, respectively. The implication is that on average all across the US, people’s drinking water contains arsenic levels that exceed the upper end of the USEPA acceptable risk range (1 in 10,000) by approximately 10-14 times. In other words, on average, the level of arsenic in the nation’s drinking water supply is unsafe. 

Why would the TCEQ doubt the cancer slope factor being used by the EPA?  Because the number it generated for the risk of bladder cancer does not reflect the reality of the world we live in:

For bladder cancer alone, the incidence risk calculated by USEPA based on final draft values for males/females is 3.1E-04 per μg/L. Therefore, based on 2 μg/L as an average drinking water concentration, the estimated bladder cancer risk for the US population would be 6.2 per 10,000 or 62 per 100,000. However, the actual occurrence of bladder cancer in the US is about 23 cases per 100,000 (males/females combined). It would take 3 times the actual bladder cancer incidence for US males/females combined to even make possible the 62 cases per 100,000 estimated due to arsenic exposure from drinking water alone. Thus, the incidence risk calculated by USEPA final draft values for bladder cancer appears to be inaccurate and overly conservative.

So what's the big deal if the potency being used is more protective?  Isn't less arsenic better?  Always, but in this case no...not if it is based on this model.  And here is why the TCEQ and I believe the EPA needs to discard this SFo:

Proceeding with this SFo will unnecessarily alarm the public by giving a greater perception of harm and risk than is actually taking place.

Speaking of harm....


Next Post: Apples, Arsenic, and Risk - Part 7: Who the heck is Sharyn Duffy of Geneseo, N.Y.?

.