Quasi-stationary Rossby waves (QSWs) modulate persistent (lasting days to weeks) atmospheric ridges and troughs, and can lead to extreme weather events, particularly in the midlatitudes. Due to their persistent nature, QSWs provide a unique opportunity to improve subseasonal forecasts of extreme events. Here, we evaluate the forecast skill of weather models in the Northern Hemisphere (NH) winter QSWs in the ECMWF dynamical subseasonal-to-seasonal (S2S) forecast model against ERA5 reanalysis data. The model shows varying prediction skill, as expected, with skill declining as the lead time increases. The North Pacific region shows the highest skill across all lead times studied (7 to 35 days). Further investigation shows an effect of a La-Nina like sea surface temperature (SST) pattern on North Pacific QSWs, which the model is able to reproduce. We find very large inter-annual variability in the subseasonal skill of the North Pacific region. This inter-annual variability of skill is not captured by variability in the ensemble spread, i.e. more skillful years do not have more certain forecasts. The annual time-series of aggregated subseasonal skill shows consistency between different S2S models, indicating that the aggregated annual skill may be partially driven by a physical forcing. We find strong correlations between aggregated subseasonal skill and SSTs, upper troposphere zonal winds, and waveguides. Overall, the results indicate that although the S2S skill of QSWs is currently low in forecast models, there is potential to utilize natural modes of variability to better capture uncertainty of model outputs.