We show a higher order discretization scheme for the Bismut-Elworthy-Li formula, the sensitivity of expectation of the solution to stochastic differential equation with respect to initial value. A weak approximation type algorithm is constructed through the integration by parts on Wiener space and is efficiently implemented by a Monte Carlo method. We give a sharp error estimate for the discretization based on Malliavin calculus. Numerical sensitivity analysis for the delta in stochastic volatility model shows the validity of the proposed scheme. This is joint work with Kenta Yamamoto.