Assuming the data are contaminated with broadband (rather than band-limited) noise, the different columns are observations and each row corresponds to a monotonically-increasing time value, use the sgolayfilt function to elimiinate as much noise as possible (this will require experimentation, I usually use an order 3 polynomial and then vary the framelen value until I get the desired result) and then use the gradient function to determine the direction of change, for example:
dPdt = gradient(Pressure) ./ gradient(time);
.
