My first approach was to use the measured differences to correct each light frame that I took. But that would have meant to do this over and over again every time I have new data...
Instead I decided to adjust my dark frame (subtract the difference in the respective columns). Then I would have to do it only once. Of course, I would only correct the dark frames that are used for lights. The much shorter dark frames that I use for calibrating flat frames I would leave as-is.
Using Pixelmath in Pixinsight, I could easily correct each column:
Doing this for both columns resulted in this dark frame:
One can see that the columns are still there, but much more subdued.
With this dark, I calibrated and stacked my Ha frames again:
No dark columns - compare with the original stack:
I also checked other filters and they got corrected well too. So, from now on I'll use the corrected dark frames for calibrating and everything is fine!!!