So far release of “Bio Passport” data has gone something like here’s the data from the last year. Look it doesn’t cross the doping cutoffs therefore this rider is clean. There have been discussions of historical norms and Bayesian something ‘ r ‘nuther, but no follow through with regard to a specific data set. Hopefully some of you have noticed this discrepancy, and that what has been marketed as such, is not exactly Biopassport science.

Cycling news picks up the story and gets into it a bit. While the focus of debate seems to be headed for a he said she said about wheter or not Lance Armstrong had diarrhea, I would like to redirect you this particular quote,

“A logical follow-up question… that is, even if Armstrong has a unique physiology, how do his Tour results square with those from the Giro, which showed over a five point drop in hematocrit? “I can speculate a lot between the two races and the differences between the two,” Damsgaard answered. “[But] these kinds of blood profiles should be exposed to the scientific algorithms, to the Bayesian model and WADA and UCI rules of upper and lower limits.””

The key here is that the debate right now rests on a failed Super EyeBall Test or SEB test as I like to call it. SEB testing is fun, but unfortunately it isn’t quite BioPassport.

To discuss the difference lets first go back to the SEB testing we did in in part 1 of this article. We took a look at numbers from Lance Armstrong’s data set and attempted to take the analysis a step further than simply comparing them to populational cutoffs.

To make that next step we had to make some assumptions;

1. The human body will always attempt to maintain a tight physiologic equilibrium, therefore it will respond to equivalent stress in a consistent way.

2. The greatest physiologic stress that a cyclist will experience is an elite performance in a grand tour.

3. All other stresses are not on the order of magnitude of a grand tour, therefore their influence should be washed out when assessing the physiologic response to a grand tour.

These assumptions allowed us to hold all outside factors as constant to test the hypothesis that the two data sets where essentially the same. In the debut use the LC SEB test it was determined that the data sets where not the same. Since they where different we had to look for some other factor on the order of magnitude of a grand tour to explain why the human body reacted differently to an equivalent stress.

From the Cycling News article it looks like the search will come down to Diarhea vs Dope. Based on your perssonal whiff test you will likely end up in one camp or the other, either way nobody wins.

However, my congratulations still go out to Cyclingnews. They are the first of the mainstreamers to rise from populational cutoffs and join the elite ranks of Local Cyclist and NY Velocity at a level of debate that still isn’t exactly Biopassport territory either.

What? SEB test does not equal BioPassport?

Say that again. Entire Biopassport articles written about an over-simplified analysis that wasn’t even really a proper Biopassport analysis?

Why would you do that?

For Local Cyclist the reason was the same one as why you might choose to debate the blood values of a mega-celebrity; It grabs your attention and creates a teachable moment. Additionally, the effect of using an approachable problem with an engaging question is that it enables you to sneek in the tedious ground work needed to tackle more complex issues. It also lets this author get write two articles instead of one.

Now that you are primed, lets look at some more numbers. Note that the calculations are to illustrate what Jakob Moerkeberg is saying and to set up the point of this article. Keep in mind that the calculations, particularly involving Retic, may stretch the limmits of validity.

From the first tour data set we start with a Hgb of 14.8 that drop to 13 over the course of about three weeks. We will assume a hemodynamic steady state and just take the average of the Retic for the three weeks. We can then use the average Retic to estimate Red Blood Cell Production.

Simple right, Retic = Red Blood Cell production.

The justification for the geeks who might be interested:

During a steady state Red Blood Cell (RBC) turn over is about 100 days and Retic turnover is 1 day allowing you to say that Retic approximates RBC turnover i.e. it provides an estimate of RBC production. So if the average Retic is 1% then average daily RBC production is 1% of RBC total. An easy way to think of this is that if no RBCs are destroyed and production is at 1% your total number of RBCs would increase by 1% per day. Or if RBCs are undergoing normal destruction of 1% and you stopped making new ones your total number would drop by 1% per day.

So using the Retic from the first data set, which averages out to 0.97%, we can estimate RBC production to be 0.97%. Since 0.97% daily production is well within the 5% daily maximum capacity we can assume that the body is capable of meeting this demand. Therefore any drop in Hgb or Hct is not because your body couldn’t keep up causing a decrease in the number of Red Blood Cells, but because the numbers look lower because of dilution caused by your blood volume increasing.

In the first grand tour data set, to go from a Hgb of 14.8 to 13 you would need to increase blood volume by 14%.

Having made this calculation we can apply this information to the second grand tour data set and make some predictions based on the 14% increase in blood volume, and a RBC production of 0.97% from the first grand tour.

Starting with a Hgb of 14.3, a blood volume increase of 14% would give you a predicted drop in Hgb to 12.5 (z -2.65) by the end of the second grand tour.  Also, if an RBC production of 0.97% is necessary to match the rate of RBC destruction, as in the first grand tour, then you might expect the average RBC production of 0.6% in the second tour to cause a deficit of 0.37% per day, (0.97% -0.6%). Remember RBCs are always dieing, so if you don’t replace them fast enough your Hgb drops. Multiply that by 3 weeks and you should be down 8.5%. So the predicted final Hgb for the second grand tour might be as low as 11.28 (z -4.2).

For those who want to think about it another way, the final Hgb value of 14.5 may look like a nearly unchanged Hgb. However, staying at 14.5 is actually like the Hgb going up to 16.5 (z +2.54), given a 14% expansion in blood volume. Or if you want to account for the difference in Retic as well, you might calculate that staying at a Hgb of 14.5 is equivalent to jumping up to 17.7 (z +4.12). These numbers would certainly fail most people’s SEB test.

Of course that’s just speculation based on shaky science and bad math. But it does illustrates what your Super EyeBall test has been saying; The second set data set should look more like the first than it does.

So wouldn’t it be nice if your SEB test was on to something. Wouldn’t it be nice if you could just use the first data set to predict what the averages should be for the next data set. And what if you didn’t have to make tons of assumptions about all outside factors being the same, and didn’t have to choose between Dope and Diarhea. Instead, imagine plugging all the factors in exactly the way they were. What if you could do all this and end up with appropriately adjusted averages and cutoffs automatically tailor-fit to the athlete instead instead of the ill fitting populational norms. And wouldn’t it be nice if you could have these appropriate averages with their tight normal ranges to calculate a probability of doped vs clean?

Apparently you can. Its called the Biopassport.

Lets bring it all home with the example of Chuck running a quarter mile. You are Chuck’s time keeper but say you happen to be blind. I know it sucks but you still need to time Chuck at the track. You can’t see him but he swears up and down that he’ll run around the whole track and tell you when he crosses the line. So you say ok. You tell him to start and you listen closely as he runs off. At the far ends of the track you can’t quite here him but otherwise it sounds like he went all the way around. You do it again and again standing at different parts of the track trying to make sure he goes all the way around. Now Chuck gets his best time. You like Chuck but can’t be sure he didn’t cheat. You can compare his time to what an average person might do. But since the range is so wide (average can mean anything from toddler to Olympian) you only know if Chuck cheated if he’s way faster than the world record, or maybe you happen to be standing at the right part of the track at the right time to catch Chuck cutting across the field. At this point you’ve done the equivalent of pre-Biopassport testing.

Now say that you know Chuck is 13. Suddenly you can change the average to something more specific for 13 year olds and the range will narrow as well. Add in that Chuck is a great wide receiver and the average might go up a bit while the range narrows even closer. Then you plug in all the times from Chuck’s previous runs and readjust your average and range. Pretty soon anything that isn’t completely consistent is going to start to stick out. Congratulations you just constructed a Bayesian Network, a theoretical model based on probability which takes into account all available information. Finish the work off by taking the new time and calculating the probability that he cheated using the Bayesian Network (BN) that you just created.

In straightforward language, the idea is that you start with an average value from the population and your normal range is that of the population. Now start plugging in test results to adjust the average and narrow the normal range. Then you use Bayesian Networks to predict where the average should be adjusted to given the all of the available information. Now set cutoffs at a point that the probability of anything beyond the cutoff is practically proven cheating. Finally, see if your result fits.

What about Armstrong’s data sets.

I’m assuming that the z scores provided are calculated using an appropriate BN. If so, only the Hgb of 13 falls outside of 2 standard deviations. This value while outside of normal doesn’t cross the threshold of 3.09 standard deviations needed for conclusive proof of doping. Even after crossing this threshold the data would still require review by an expert panel including further analysis and calculations to rule out other explanations.

So how did we go wrong with our SEB test? Maybe we didn’t. At this point, the BNs are still in their early incarnations. To the best of my knowledge they don’t yet use the data the same way that we did in the eyeball test. For example the factors currently used in the BN are along the lines of age, gender, race, and altitude exposure. They are not yet taking the change in Hgb from first to last day of one grand tour to plug into the calculation of an appropriate Hgb for the last day of a second grand tour. At least that type of work is not yet published.  Also, while blood parameters are combined in the OFF score and Abnormal Blood Profile score they are used more in a pattern recognition manner. I am not aware of published literature showing them being used as explicit factors in a BN Hgb probability calculation.

I realize the conclusion is anti-climactic. Fortunately, more definitive answers will likely come. The beauty of the Biopassport is that the data isn’t going anywhere and BNs continue to learn as additional data is collected. BNs can also be taught new tricks by adding in factors that where previously not considered. It will be interesting to see just how far down the rabbit hole ultimately goes.

Even without the addition of new tricks, there is a lot more to discuss. Only 2 out of 7 hematologic markers are being considered here. The OFF score is sitting this one out and the Abnormal Blood Profile Score hasn’t even made an appearance. Nor did we mention the possibility of using multi-tiered sanctions for lowerer cutoffs, or bother to ask about calculating practically proven clean.

But for now, I hope that you can appreciate that the Biopassport is on the verge of unlocking an incredibly powerful new tool for doping prevention.

Finally, I would like to publicly thank Lance Armstrong. More than any other person he has the most to lose by releasing results. And it is the magnitude of what is at stake that has truly brought the BioPassport front page awareness it deserves.
Part 3 Part 4

P.S. If your eyes haven’t glazed over I encourage you to get to a library and start with this.