toucan

you should basically ignore ramez's tweet. i am not going to bother with a writeup because it is quite obvious if you skim, but the main issue is the study is basically measuring "are people who do not move around like at all going to die from all-cause mortality" and the answer is yes, obviously. it is not attempting to measure, nor does it measure, how much exercise you should obtain for longevity benefits. the crux is that the threshold for activity is extremely, extremely low and failing to meet this threshold, any under common sense reasoning, means that you are probably extremely unhealthy, even when controlling for covariates of obesity and age. the threshold for activity is VPA=vigorous physical activity, defined by "physical activity at an energy expenditure rate of at least six metabolic equivalents". 6 METs is roughly, playing basketball, jogging at a 15min / mile, or climbing hills without load. the reason this is a very low threshold is because, as measured, 92% of all VPAs are less than a minute. when the study measures 15min of VPA over a week, that is cumulative, and that means roughly fifteen independent incidents of VPA less than a minute each. ask yourself how many VPA-minute slices somebody who is actually doing exercise would expect to incur. a single workout of a fifteen minute-mile-run should be fifteen minutes of 6 MET, definitionally! basically nobody in this dataset is working out at all! it is all being accumulated from random life moments of literally less than a minute each! the study is not measuring exercise! it is measuring VPA parameterized by a 6 MET threshold! if you are not hitting 10 min of VPA in a week you are probably dying soon because your lifestyle is incredibly unhealthy and / or your body is incapable of literally walking up a hill! and that's if you take the study data at face value. there are some questionable things going on with the accelerometer data, such as the fact that the accelerometer is on the dominant hand, rather than on the non-dominant hand, or the thigh, which is the case for many of the data collection validation studies that are cited, and also the fact that the random forest activity classifier classification scheme is seemingly unable to capture many kinds of exercise, such as weightlifting. again, as a simple thought experiment, imagine the kind of forces on the dominant wrist during a squat, compare it to dishwashing, consider the distinct metabolic load of these two activities, then realize that they get squished into the same classification. also the classifier calibration stuff is super suss. all that is to say, that the buckets of VPA total time over 10min are probably including a lot of very active people who are incorrectly classified and people who are not that active. so i highly suspect that if you actually had ground truth you would get a much tighter correlation between activity level and health. but again, a VPA total time of 30minutes over a week is VERY LOW because it is measured in 10 SECOND SLICES over ALL DAILY ACTIVITIES. If you run to catch the bus that counts as a VPA!

Small amounts of vigorous exercise provide the bulk of the benefits of all exercise. Exercise volume is really unimportant for fitness. This is true for cardio fitness, for mortality and longevity, and for strength. This particular paper shows that <10 minutes of vigorous exercise* per week (yellow) gets you more than half of all the longevity benefits of much longer periods. It's probably an even lower amount at higher intensity. * - Vigorous exercise here is defined as roughly 6 METs or above, which is jogging, moderate cycling, hiking, etc.. Sprinting is ~4x that intensity.