It is well documented that wastewater disposal by underground injection can increase seismic activity. However, it is challenging to predict whether and when this effect occurs, as it can manifest itself even many years after the beginning of disposal operations. Continuous instrumental monitoring of the seismic activity is a first step enabling early detection of symptoms of change, updating of site and regional risk assessment, and prompt response. To avoid misdetection, however, seismic records have to be rationally analyzed, properly modeling the statistical features of the earthquake occurrence process. In this study, we develop a statistical approach to detect increments in seismic rate, accounting for model uncertainty (which is particularly acute when the monitoring period is short) and interdependence among events. The approach is composed of two steps: (1) stochastic earthquake declustering identifies mainshocks, and (2) the hypothesis of a constant rate of mainshocks is statistically tested. The method is applied to the analysis of the Oklahoma injection region, demonstrating that it is able to detect an increment in seismic rate before the change is large enough to produce severe consequences. We also investigate the statistical power of the method using synthetic data simulated for a wide range of scenarios. We believe that the method can be employed as a preliminary decision-support tool to identify areas with trends in seismicity that require further investigation and to guide adaptive management and operation of disposal wells.