Generalising from a hospital-based study to the big wide world of general practice and beyond is fraught with problems. The reverse is also true, as this study shows. Automated defibrillators save lives out in the community but that does not guarantee they will also be a good idea in hospitals.
All credit to Chan and colleagues for studying this question. They were not able to use a randomised design; hence much hangs on whether poorer survival in the AED group was due to something else - i.e confounding. Perhaps you don't bother with an AED for a simple arrest, only getting it out if things are complicated - if so then the results are confounded by a poorer expectation of survival in the AED group. The analysis adjusted for some confounders and other possibilities are mentioned in the discussion.
Finally, note that it took 10 years to gather and analyse this data.