If you can be sure the regex won't match any other field, you can try:

perl -p -i -e 's/\b0(\d{10})\b/\+91$1/g;' filename

If there are other fields that it could match and munge, then it could all go horribly wrong and you're better off with something that accurately replaces only that field, like Tux already posted.

Update: I just took a look over at your thread at SO. You're trying a dangerous thing by counting commas in a CSV. Any element of a CSV can have commas inside of quotes to protect them, so you could have any number of commas before the 35th element in a row. CSV is a deceptive format-- it seems all simple and benign, but can be complicated in really annoying ways. If the quick and dirty regex won't work, use a module like Tux shows that is designed to deal with the unexpected but legal cases in the CSV.