search

ITHIM / ITHIM-R

Development of the ITHIM-R, also known as ITHIM version 3.0. Started in January 2018.
https://ithim.github.io/ITHIM-R/
GNU General Public License v3.0
17 stars 11 forks source link

Calculation of RR for physical activity #1

Open markotainio opened 6 years ago

markotainio commented 6 years ago

This issue is about discussing how we transfer the physical activity (PA) levels to RRs.

The general methodology will be following:

@usr110

syounkin commented 6 years ago

Will we include all physical activities, or only recreational activities?

markotainio commented 6 years ago

Good clarification. Our recent meta-analysis paper was defined to include non-occupational physical activity volume, so that is the PA we are interested in. So both leisure and transport related PA, but not occupation.

syounkin commented 6 years ago

What definition was used for the spreadsheet version of ITHIM?

syounkin commented 6 years ago

Can we discuss the meta-analysis paper at our next meeting? I would like to incorporate these new risk estimates as soon as possible. Thanks.

markotainio commented 6 years ago

I don't remember details from spreadsheet version. I think it was similarly non-occupation PA.

We will talk on meta-analysis paper in next meeting (Thursday 1st).

JDWoodcock commented 6 years ago

the spreadsheet was a confusing mix depending on the studies available. The new MA is (at least for the core diseaseswe cover) much more consistent. It should be noted we have potential to distinguish incidence from mortality in some cases now. For dementia this is the paper we plan to use http://bmjopen.bmj.com/content/7/10/e014706

syounkin commented 6 years ago

@usr110 @markotainio Could we place a copy of the table(s) used in the table look-up step for the physical activity relative risks in ITHIM-R/ on Google Drive? I am interested in creating a function to replace the look-up procedure. Thanks.

JDWoodcock commented 6 years ago

when we had instruments calibrated to objective data we use that - marginalisng energy expenditure above that of a sedentary office worker. In theory one could therefore include occupational activity above that of a sedentary risk in your exposure estimate. However, would need to think about applying a minimum intensity to avoid counting hours in very light activity. Even for MVPA thre is a a risk of overestimating how much time manual workers actually spend in MVPA. Ideally we would have access to objective data.

usr110 commented 6 years ago

Why don't you start with all-cause mortality data? I've put it in the G-drive (at Google Drive\ITHIM-R\data\DR_all-cause mortality.csv), please have a look at it @syounkin Although the data is available till 77 MMETh, but I am capping it at 35. This means any value greater than 35, would have the same RR. From our last discussion, I think a better solution in the long run would have to be a generic function for all diseases. I will work towards it.

markotainio commented 6 years ago

@JDWoodcock wrote that "For dementia this is the paper we plan to use http://bmjopen.bmj.com/content/7/10/e014706". I checked the paper and the exposure for physical activity is METh/week, not marginal METh/week like we have in our meta-analysis. Should we scale the new study to marginal METs, or use METs for this endpoint? @syounkin @usr110

markotainio commented 6 years ago

Someone should go through the depression literature to see if there is better dose-response than what we have in old ITHIM (Paffenbarger et al. 1994 https://www.ncbi.nlm.nih.gov/pubmed/8053361). Note the old RR is based on study made only in men! Would @walkabilly be interested to check this one?

markotainio commented 6 years ago

The study where we can obtain dose-response for type-2-diabetes is this one: https://www.ncbi.nlm.nih.gov/pubmed/27747395

Is uses marginal METs and is based on same systematic review than the new meta-analysis study (however, based on older version of analysis and synthesis).

usr110 commented 6 years ago

The study where we can obtain dose-response for type-2-diabetes is this one: https://www.ncbi.nlm.nih.gov/pubmed/27747395 Is uses marginal METs and is based on same systematic review than the new meta-analysis study (however, based on older version of analysis and synthesis).

I created a Shiny App for this study. Please see it here: http://npct0.vs.mythic-beasts.com/meta-analyses/pa/diabetes/

JDWoodcock commented 6 years ago

@leandromtg can you look at the dementia paper and see how easy it would be marginalise it. Probably eaiser to just mets as @markotainio suggests but nice to see if we can be consistent

JDWoodcock commented 6 years ago

@usr110 I suggest we use 35 mmeth for all diseases as point becomes linear (even those not in our meta-analysis)

syounkin commented 6 years ago

@JDWoodcock Can you clarify the previous comment? Thanks.

usr110 commented 6 years ago

@JDWoodcock Can you clarify the previous comment? Thanks.

@JDWoodcock sure, that's what I meant. Let me clarify what James is saying there @syounkin This means that for all diseases and causes, we assume a straight line (with constant RR value) for MMETh greater than equal to 35. This is because we don't have sufficient confidence in the dose-response relationship for the higher doses.

JDWoodcock commented 6 years ago

@syounkin @usr110 @robj411 @leandromtg as sensitivity analysis assuming flat beyond 17.5 would be sensible. I would also like a way of checking the user provided PA mmeth distribution to compare against the ones we have from the studies used in the meta-analysis. I expect some users will propose distributions that are very different (too high) and we want to alert them to this.

leandromtg commented 6 years ago

Replying @JDWoodcock

@leandromtg can you look at the dementia paper and see how easy it would be marginalise it.

Sure. Let me deal with the ventilation rate x MET issue first and will come to this soon.

Replying @markotainio

@JDWoodcock wrote that "For dementia this is the paper we plan to use http://bmjopen.bmj.com/content/7/10/e014706". I checked the paper and the exposure for physical activity is METh/week, not marginal METh/week like we have in our meta-analysis. Should we scale the new study to marginal METs, or use METs for this endpoint?

In the meta-analysis we used an prediction equation to translate MET.h/wk to marginal MET.h/wk. I guess we could use it here too, but let me check more carefully. If so, perhaps we could have a function to go from one scale to the other.

leandromtg commented 6 years ago

On marginalisation in general: In our latest meta-analysis, we marginalised MET.h/wk using this equation obtained from EPIC cohort participants: MMET.h/wk = 0.0011055 MET.h/wk^2 + 0.739704 MET.h/wk

On the dementia meta-analysis: Authors obtained a dose-response function (DRF) between all-cause dementia and leisure-time physical activity (LTPA) with only 3 papers (1 potential study excluded because upper limit surpassed 10,000 kcal/week or 200 MET.h/week). Observed range was 0 to 45 MET.h/week, with no evidence of linearity (p = 0.86). Other 5 LTPA papers included in "high vs. low" meta-analysis, but were not harmonised to METs. Extraction matrix with data from original papers was made available, so potentially we could use our imputation and harmonisation (and marginalisation) processes on them and obtain a more robust DRF. If we decide to stick with the reported DRF, we can ask for a lookup table and apply the aforementioned rescaling equation.

markotainio commented 6 years ago

Sounds me that it makes sense to harmonise using the meta-analysis equation. This way all the calculations would be with MMETs. Calculating both METS and MMETS wont be big tasks, but it's a bit more cleaner to use only one.