Heavy impurities, such as tungsten (W), can exhibit strongly poloidally asymmetric density profiles in rotating or radio frequency heated plasmas. In the metallic environment of JET, the poloidal asymmetry of tungsten enhances its neoclassical transport up to an order of magnitude, so that neoclassical convection is expected to dominate over turbulent transport. The modeling of poloidal asymmetries is hence necessary in the integrated modeling framework. The neoclassical drift kinetic code, NEO [E. Belli and J. Candy, Plasma Phys. Control. Fusion P50, 095010 (2008)], takes into account the impact of poloidal asymmetries on W transport. However, the computational cost required to run NEO slows down significatively integrated modeling. An analytical formulation was proposed to describe heavy impurity neoclassical transport in the presence of poloidal asymmetries in specific collisional regimes [C. Angioni and P. Helander, Plasma Phys. Control. Fusion 56, 124001 (2014)]. The present work compares the analytical formula to the numerical results produced with NEO. It investigates if the formula can be used to accurately predict heavy impurity transport in experiment when combined with a neoclassical model that is less CPU intensive but does not include the effect of poloidal asymetries, such as NCLASS [W. A. Houlberg, K. C. Shaing, S. P. Hirshman and M. C. Zarnstor , Phys. Plasmas 4, 3230 (1997)]. The analytical formulation was derived in the limit where the main ions are in the banana regime, with high-Z trace impurities in the Pfirsch-Schlüter regime. The formula is found to remain valid outside its definition domain. Indeed, considering main ions in the banana regime, it well reproduces NEO results whatever the collisionality regime of impurities, provided the poloidal asymmetry is not too large. However, for very strong poloidal asymmetries, agreement is only obtained for impurities in the Pfirsch-Schlüter regime. Within the integrated transport platform JETTO, it is demonstrated that NEO and the neoclassical formula combined with NCLASS lead to the same tungsten profile predictions while gaining a factor 1100 of CPU time.