Correcting for days between sample and event.

Question

[ccbb7aab4]

Hi All,

I was wondering if you could help me with a problem I have.

Our participants bleed dates and the days since they had their event are not standardised. This means when we work out the days since their event based on the bleed dates, we get a range of days since event.

I was wondering if there is a way to correct/normalise/Standardise for this variation. Originally, we used to categorise them based on time periods (Short, Medium, and Long). As we where interested if there was a time affect between the event and bleed date. However, with the analysis we are now doing we, need a way to remove the potential bias/variance introduced due to the time differences.

We have already collected our data, which mostly contains meso scale discovery data.

Some people we have samples less than 30 days since their event and others we have their sample 100+ days since their event.

Hope you can help.

Answer 1

[Matthias Zepper]

To consult the statistician after an experiment is finished is often merely to ask him to conduct a post mortem examination. He can perhaps say what the experiment died of.

Ronald Fisher

I am sorry to be that harsh, but I fear you can't correct for that bias in sampling in any reasonable way. Either the timing has absolutely no impact on the measured results (then there is no need to correct for it) or it does have an effect. In that case, one needs to consider the "Time since the event" as an additional explanatory variable / predictor in the model term. A Generalized Additive Model might be suitable for this, since they are often used to interpret time-series data.

[ccbb7aab4]

Hi Matthias,

Ronald is right of course, but unfortunatly we had no control on the way the samples where collected. Also when planning the study, this question wasnt part of the plan. Its a follow-on from a study we looked at before,which led us to raise this question.

Thanks for your suggestions, i will take a look at them and see if they are suitable. I agree the time is either an effect or not, and problably treating it as an additional predictor is problable the only resionable way to solve this.

Thanks, Dave

[Matthias Zepper]

Unfortunately, I know only too well the pressure to find and publish something and the immense effort and money it would take to gather new material tailored to a new research objective. I am well aware of the temptation to oversell findings based on actually poor data and that almost everyone else it doing it as well.

Nonetheless, I urge you to be extra careful when using data that was never collected for a specific purpose in another context. It is easy to mislead others but also yourself, so ideally seek to corroborate such findings by other means, too. Harley at al. is for example one paper that we use(d) to dismantle in the statistics class as a group exercise for its blatant errors and yet the authors made worldwide headlines with it. Also in this study, urine samples initially collected for assaying Bisphenol A were later used in a completely different study design, without proper controls and calibration, eventually giving rise to those questionable results.

Of course, raising new questions based on previous results is fine and looking at existing data to give you hints whether it is worth working in this direction is too, but always mind: There is already enough crappy science published out there, please don't add more ;-)

[ccbb7aab4]

Hi Matthias,

Don't worry if i am not confident about the outcomes or the processes it will never see the light of day ever again.

We also have a stataticians looking at this problem, as my background is not orignally statistics. I felt we needed somebody with more experiance to make sure we dont fall into the trap you discribed via Harley at el. I hope they will give us some straight answers, which could include this won't work.

Thanks again for your help. I wil try and not add to the crappy science that has been publised in the past.