The main objective of modeling the degradation path of a device is to predict its eventual time-to-failure. When a system is operated on-field, abrupt and unexpected changes in ambient conditions could potentially cause deviations from the expected degradation path, such as an acceleration to a state of failure. Previous estimates of lifetime distributions become inaccurate because the fitted model may no longer be a satisfactory representation of the degradation path. This paper shows the application of a Cuscore statistic to detect a downward shift in the gradient of a deterministic trend buried in autocorrelated noise. The proposed diagnostic methodology finds an application in monitoring the natural degradation of solar photovoltaic modules installed on-field.