This is such a morbid topic that I've never acted on it, but I suspect that with historical news data you could figure out the growth rate laws which relate time since disaster occurred with number of reported casualties to build a predictive model which - given the inaccessibility and population density of region and the scale of a disaster - would fairly accurately predict the final number of dead.

The logic being that irrespective of final number of dead the 'effective death rate' (i.e. number of new reported casualties per day) is likely to be similar (conjecture).

Anyway - just writing this post makes me feel uncomfortable, hence why I've never done more than hypothesize.