Dansgaard-Oeschger (DO) events are abrupt millennial-scale climate changes mainly detected in the North Atlantic region during the last glacial cycle. The frequency of the DO events varied in time, supposedly because of changes in background climate conditions. Here, we investigate the influences of external forcings on DO events with statistical modelling. We assume two types of generic stochastic dynamical systems models (double-well potential-type and oscillator-type), forced by the northern hemisphere summer insolation change and/or the global ice volume change. The models are calibrated by maximizing their likelihood and compared using the Bayesian Information Criterion (BIC). Among the models, the stochastic oscillator model forced by both insolation and ice volume changes is favored by the NGRIP calcium ion data. The BIC scores provide positive evidence for the ice volume forcing in the presence of the insolation forcing but weak evidence for the insolation forcing in the presence of the ice volume forcing. Consistently with their BIC scores, stochastic oscillator models with at least ice volume forcing reproduce well the frequency changes of warming transitions in the last glacial period across Marine Isotope Stages (MISs) 2, 3, and 4. Finally, using the selected model, we simulate the average number of warming transitions in the past four glacial periods, and compare the result with an Iberian margin sea-surface temperature (SST) record (Martrat et al., Science, vol.317, p.502, 2007). The simulation result supports the previous observation that abrupt millennial-scale climate changes in the penultimate glacial (MIS 6) are less frequent than in the last glacial (MISs 2, 3, and 4). On the other hand, it suggests that the number of abrupt millennial-scale climate changes in older glacial periods (MISs 6, 8, and 10) might be larger than inferred from the SST record.