Bias, Part Three

Last time we met to talk about bias, we talked about measurement bias and how choosing the wrong instrument, the wrong measurement, or measuring the wrong thing can make all the difference in the outcome of your study. We also talked about recall and attention bias, where the study subjects' expectations can have an effect on your outcomes as well. Today we're going to talk about Exposure/Intervention Bias, where the experimental group goes awry. All after the jump...

He would eventually get treated for exposure.

CONTAMINATION BIAS
This type of bias has nothing to do with getting stuff on you when you fail to heed the warnings of the biosafety level 1 lab personnel. No. This type of bias occurs when a member or members of the control group (the group that is getting placebo, or the old drug, or nothing at all) are exposed to the intervention without the knowledge of the researcher. (Or maybe the researcher knows but fails to control for it.) A good example is the comparison of the effects on teeth of the fluoridated versus un-fluoridated water. Suppose you have two towns, and you enroll the residents of both towns into a study to see if those in one town (whose water has added fluoride) have better or worse dental hygiene compared to the other town (whose water has no added fluoride). What would happen if a few in your control (no fluoride) town spent most of the day working at the experimental (fluoride) town?

Not this, that's for sure.

CO-INTERVENTION BIAS
Let's say that you're studying two groups of people. One group (control group) is taking the old medication for blood clots. Another group (experimental group) is taking the new medication for blood clots. For some quirk in the design, you managed to get older adults in the experimental group who also take a baby aspirin. Can you see how your experimental group may end up having less instances of blood clots? (Yes, it's both a selection bias and a co-intervention bias.) In the example of the fluoride towns, what would happen if the people in the fluoridated water group don't brush their teeth and the group in the regular water did?

TIMING BIAS
One of the very common statements made by people who try homeopathy is that they were sick "for years", and it wasn't until they tried homeopathy that they got better. We know that homeopathic remedies are nothing but water with one or two molecules here and there of the "active ingredient" (which itself is some unproven herb or mineral or something). What would explain these folks' miraculous recovery? Timing bias is also known in some circles as "return to normal". See, if you catch the flu, it is expected that you'll have the full-blown symptoms for 7 to 10 days. You'll feel miserable. You'll wish you were dead. But what if you started taking a homeopathic remedy at day 1? If you apply timing bias to your observation, you would proclaim the benefits of the remedy at day 12, once the disease had run its natural course.

He tried homeopathy for swine flu and, well, you draw your own conclusions.

On the other hand, if you give a person an antiviral at day 1 and they get worse because of some co-infection or underlying health condition, and they unfortunately die at day 3, timing bias will make it look like the antiviral didn't work. Some will even go as far as thinking that the intervention is what killed the person. In a large study, timing bias of this kind will be seen when the study is too short to really let the intervention kick in, or the endpoint it too far in the future so as to allow the "return to normal" to occur.

COMPLIANCE/WITHDRAWAL BIAS
In large studies, participants are always given a set of instructions along with the information about the study itself. When the participants fail to stick to the plan, the study may be biased. This also happens when participants drop out of the study, changing the statistical dimensions of your study when your sample size or number of interventions change. For example, what would happen if you end up with no one in the control group? Study over. How would your results look if the participants don't take the new drug you're researching, opting instead for some other drug without telling you (also a co-intervention bias)?

PROFICIENCY BIAS
Finally, imagine that you are testing the safety and effectiveness of a new vaccine that is given up the nose. (There's a flu vaccine like that available already.) What would happen if one of your treatment centers has staff that is not well-trained on how to give the vaccine? Or what if the vaccine needs to be reconstituted with 5ml of deionized water but one of the techs preparing it doesn't know the difference of to-contain and to-deliver pipettes? Proficiency bias has more to do with the people giving a drug (or preparing it), or the ability of those giving out a survey, etc.

THAT'S IT FOR BIAS
So this about wraps it up for bias. Next time, we'll talk about confounding, which is and isn't bias, per se. It's a special kind of animal that requires it's own "lesson"... At one point, not recognizing confounding made coffee drinkers incredibly nervous. But we'll talk about that next time.

Thank you for your time.

Powered by ScribeFire.