In the uplink of network MIMO systems, in order to cope with backhaul capacity limitations, base stations (BSs) must compress the received baseband data signal and the channel state information (CSI) for communication to the central unit (CU). Assuming ergodic fading, an Estimate-Compress-Forward (ECF) approach is investigated, whereby the BSs perform CSI estimation and forward a compressed version of the CSI to the CU. This approach contrasts with the previously studied Compress-Forward-Estimate (CFE) strategy, whereby CSI estimation is performed at the CU, and is motivated by the information-theoretic optimality of separate estimation and compression. Various ECF schemes are proposed that perform either separate or joint compression of estimated CSI and received baseband signal. Via numerical results, it is shown that a proper design of ECF strategies leads to substantial performance gains compared to the CFE approach.