Fellowship gets decided on interest, on lifestyle, on whether you can stand another few years of being the most junior person in the room. Those are real. But there is an arithmetic question underneath all of it that almost nobody actually works out, and it takes about a minute.
It has two inputs.
Forgone earnings is what the training costs you in income you never see. A fellow makes about $82,000, per the AAMC's 2025 stipend survey. A general internist makes $267,200, per the Bureau of Labor Statistics, May 2025. So each fellowship year costs roughly $185,000 in earnings you would have had. Three years is about $556,000, gone, before anyone has said anything about whether the fellowship was worth it.
The delta is what you gain, every year, forever after. Cardiology pays $454,940 against that same general internal medicine baseline. The delta is $187,740 a year.
Now divide. About $556,000 of forgone earnings, against $187,740 a year of gain, is a payback period of roughly three years. Finish fellowship, practice three years, and you have caught up with the version of yourself who never went. Every year after that you are ahead. On money alone, that fellowship is not a close call.
What makes it close, or wrong, is a smaller delta. A three-year fellowship that adds $40,000 a year takes almost fourteen years to break even. One that adds $100,000 takes about six. The delta can also be zero, or negative. These are pre-tax and ignore investment growth, so the real paybacks are longer, especially the small ones.
Within your own residency, the yearly hole barely changes. Multiply by your own fellowship's length. The delta changes enormously, and the delta is the whole decision.
One caution about where you get that second number. The Bureau of Labor Statistics publishes attending pay by specialty, and the spread is wide: $212,110 for general pediatrics, and that $454,940 for cardiology. But it does not print a separate line for every subspecialty, and the commercial surveys that do can disagree by tens of thousands. Use a source you would be willing to defend out loud.
None of this tells you whether to do a fellowship. Plenty of people should do one at a payback period that looks terrible on paper, because the work is the work they want. The point is narrower. The number exists whether or not you run it, and the people who never run it are not choosing to ignore it. They just never knew it was a number.
Run the division before you rank.