In this paper, we consider automatic repeat request based multihop communication between a source and a destination. We present the design of near-optimal power control policies which minimize the packet drop probability (PDP), when the energy cost to receive and decode a packet is non-negligible. The design problem is a nonconvex mixed integer nonlinear program, which, in general, is NP-hard to solve. We transform the problem into a complementary geometric program (CGP), and solve the CGP using a series of GP approximations. Simulations show that for slow (fast) fading channels, the obtained policy offers approximately one (two) orders of magnitude better performance compared to a conventional equal power policy. In addition, the results quantify the relative impact of the available power at the nodes and the effect of decoding cost on the PDP.