Our work is preliminary and has a number of limitations. Our WARM interface objects are early prototypes and underwent substantially fewer design iterations than we would have liked. The test environment itself is a VR simulation of AR, and though we presume that our conclusions would transfer from the one setting to the other, this has not been demonstrated. The study itself was only a pilot study involving a limited number of mostly "expert" subjects. The study also involved only two, quite artificial simultaneous clinical tasks and thus provided only a limited reproduction of the target clinical environment.
Despite these limitations, our results show that WARM interfaces, by allowing concurrently performed tasks to be perceptually aligned, reduces the amount of time needed to perform the tasks correctly. This conclusion is not as foregone as it might at first appear, since there is an important trade-off between reducing visual scanning and adding visual clutter [36].
The clear failure of the 3D ECG object in all performance measures was not a great surprise; Merwin et al. in a different context found 3D representations to be less successful than more conventional variants [38]. Many additional design refinements such as further color and animation encoding were envisioned but couldn't be implemented because of resource constraints; we think the 3D object would have fared better with them. On the other hand, a real-time object of this sort only accomplishes half of the two-phase decision task involved in interpreting an ECG rhythm. The first part (event classification) involves determining to which cardiac electrophysiological event each component of the ECG waveform pertains, and our 3D ECG model actually provides a higher-level interpretation of this phase. However, the second part of the interpretive task (context recognition) is not supported at all in a strictly real-time animation model. In contrast, the conventional ECG waveform shows at least a few seconds so the user can scan back in time. Further, the real-time model scales time 1:1, whereas the conventional waveform trace allows a user to scan the entire period virtually instantaneously, providing a huge magnification of time scale. Nevertheless, we believe real-time models probably have an appropriate use, and we're encouraged by similar work in "rapid serial visual presentation" interfaces such as that reported by Konrad et al. [37].
We believe that future work on WARM interfaces should focus on improved 3D information metaphors and visualization implementations, development of "control surfaces" for information interactivity, and continued formalization of performance testing methodologies.