Consider converting the data to berkleydb or SQLite formats as databases lend themselves to "asking questions of the data." When your module launches it could test the date-stamp on the source textfile and create/reload the data as updates are obtained. Written generically enough, an import utility could cope with the addition of or removal of columns. Text::CSV and Text::CSV::DetectSeparator would be a good starting point for a making flexible reader/loader even if you didn't convert to a database.
Another idea would be to create a loader for each year's data and presumably call the right loader based on some header information in the source files. You could extend this to load a specific year's data on demand provided the end user had copies of previous and current year's data.
Just food for thought.