3 ms·

> Analyses adjusted for age, sex, duration of light intensity physical activity, duration of moderate intensity physical activity, smoking history, alcohol cons

by dynm 4y ago

> Analyses adjusted for age, sex, duration of light intensity physical activity, duration of moderate intensity physical activity, smoking history, alcohol consumption, accelerometer estimated sleep duration, fruit and vegetable consumption, education, self-reported parental history of cardiovascular disease and cancer, and self-reported medication use (cholesterol, blood pressure, and diabetes), and prevalent cancer and CVD. We should be very nervous about any results that depend on this many adjustments. Remember: If their measure of smoking history is crude or inaccurate, this basically means they only "half control" for it. What I would love to see in this kind of a study is what the strength of the association is WITHOUT adjusting for anything and then show how it changes as more adjustments are added. If the association drops a lot with each adjustment, then we should be suspicious that enough controls haven't been included or the controls are too noisy. Of course, at the end of the day, nothing substitutes for an RCT. These kinds of observational studies are worth thinking about even though they are much less informative, given that they are orders of magnitude cheaper. It's just worrisome that most popular reporting doesn't seem to recognize how much less they should be trusted.

nextos 4y ago
Nature Medicine is known in the field for sometimes publishing studies with dubious statistical methods. It's a great journal in terms of metrics, but more often than not they publish things that look interesting but not very robust. For example, most studies that do multivariate analyses (not applicable here) do not do partial pooling. Hence, they end up with inflated effect sizes. Nature Genetics is also letting this happen all the time, which is annoying and leads to non-reproducible findings.
psychphysic 4y ago
That would be really nice. And reproducible science needs to be more prevalent. But you must realise that effects would be entirely obscured without corrections. That's just the central limit theorem. But I think this thread misses the main thrust of the research. Next time you read a paper about someone living longer for say. Combing their hair think, should they have corrected for VILPA?
mohammad_ali85 4y ago
I'm not sure I agree with the level of adjustment being a negative here. The point they are making is that have adjusted for confounders ie. considered other covariates (all those listed) and the results still show their exposure (VILP) being associated with their outcome of interest (mortality). It's usually better to account/adjust for as many confounders as possible (happy to be corrected by someone on this!), especially those that are clinically relevant. Usually, more covariates in the model takes it towards the null so if there is still an effect estimate, it's likely to be a better association than using crude values. My issue with these studies from the UK Biobank though is that the wearable devices are only worn for up to 2 weeks or so, although the follow period is significantly longer in this study. For this study the "participants were required to have at least three valid monitoring days, with at least one of those days being a weekend day". Would be better to have a prospective cohort but good to see associations so further studies can investigate.

by dynm 4y ago

nextos 4y ago

Nature Medicine is known in the field for sometimes publishing studies with dubious statistical methods. It's a great journal in terms of metrics, but more often than not they publish things that look interesting but not very robust. For example, most studies that do multivariate analyses (not applicable here) do not do partial pooling. Hence, they end up with inflated effect sizes. Nature Genetics is also letting this happen all the time, which is annoying and leads to non-reproducible findings.

psychphysic 4y ago

That would be really nice. And reproducible science needs to be more prevalent. But you must realise that effects would be entirely obscured without corrections. That's just the central limit theorem. But I think this thread misses the main thrust of the research. Next time you read a paper about someone living longer for say. Combing their hair think, should they have corrected for VILPA?

mohammad_ali85 4y ago

I'm not sure I agree with the level of adjustment being a negative here. The point they are making is that have adjusted for confounders ie. considered other covariates (all those listed) and the results still show their exposure (VILP) being associated with their outcome of interest (mortality). It's usually better to account/adjust for as many confounders as possible (happy to be corrected by someone on this!), especially those that are clinically relevant. Usually, more covariates in the model takes it towards the null so if there is still an effect estimate, it's likely to be a better association than using crude values. My issue with these studies from the UK Biobank though is that the wearable devices are only worn for up to 2 weeks or so, although the follow period is significantly longer in this study. For this study the "participants were required to have at least three valid monitoring days, with at least one of those days being a weekend day". Would be better to have a prospective cohort but good to see associations so further studies can investigate.