The prediction of the thermal energy confinement time, tau(th), for ITER size experiments is based on power law scalings obtained using data sets of present tokamak results in specific regimes, the most relevant being the ELMy H mode regime. A thorough statistical approach has provided a bust fit to these data with an estimation of the error bars which forms the basis for the ITER EDA design parameters. In this article the range of variation of the main parameters in the database are studied individually and it is observed in particular that tau(th) depends linearly on the combined variable a kappa B. Taking advantage of this linear dependence and of the dependence on the plasma current, it is shown that the four variables a kappa B, n,P* = P-L/V and q(eng), or equivalently I, epsilon, n and P*, are good parameters which provide a simple lit to the data, namely: tau(th) 0.0307 (a kappa B)n(1/2)/(P*(2/3) q(eng)) = 2 pi x 10(-3) In-1/2/(epsilon p*(2/3)), which satisfies the high-beta or Kadomtsev constraint. This fit is as good as the best log-linear fit, using the eight variables I, B, R, epsilon, P, kappa, M and n, over the full set of devices used in the databank and it yields the same prediction for the ITER EDA confinement time. It turns out that the simple best fit used is a gyro-Bohm scaling. It is also shown that small changes in the density, n and P* exponents call give a Bohm-like scaling, which is less accurate and pessimistic, but can be used as a lower bound prediction. The use of this simple scaling law is illustrated by a proposal to slightly reduce the size and magnetic field of the ITER EDA design, while taking advantage of the favourable dependence of tau(th) on kappa.