The Tool That Makes You Better Without It
Clinical decision support should build expertise, not replace it. There's a difference
The Future of Clinical Decision Support is a new series from MDCalc exploring how AI, evidence, and physician judgment are reshaping medicine. These essays are intended to spark discussion about where technology is taking clinical decision support – and where physicians feel it should go.
There's a metric no clinical AI vendor publishes on their website: How good are the clinicians who used their tool for a year - when the system is down?
That question sounds contrarian, but I think it's one of the more honest measures of whether a clinical decision support tool is doing what it claims. Does it augment expertise, or does it create reliance? Is a clinician who's used it for a year a sharper diagnostician – more calibrated, better pattern-matched, more confident in their reasoning – or have they outsourced a part of their cognition to the tool in a way that would show up if the tool disappeared?
The answer matters clinically, and it's been largely absent from how we evaluate these tools.
The distinction I'm drawing is between two different models of what decision support actually does.
The first model: the tool as a colleague. It surfaces information you hadn't considered, presents evidence you knew existed but hadn't recalled, flags a calculation you were about to do in your head anyway but now you can do precisely. You consult it the way you'd consult a second opinion – it adds to your reasoning rather than substituting for it. Use the tool for a year and your reasoning gets better because you've been exposed to more evidence, more calibration, more clinical variation than you'd have seen on your own.
The second model: the tool as a crutch. It makes decisions easier by making them shorter. You provide the inputs; it provides the output; you act. The reasoning stays inside the tool. Use it for a year and you've done a thousand clinical consultations without building the pattern recognition those consultations should have produced.
The first model is what clinical decision support is supposed to be. The second is what you get when the design optimizes for engagement over expertise.
Consumer technology has a different relationship with dependence. If Netflix's recommendation algorithm becomes so good that you can't choose a movie without it, that's a product success metric. The dependence is the value.
Medicine doesn't work that way — or shouldn't. The goal of clinical education has always been to make trainees progressively less dependent on supervision as their expertise grows. A clinical AI tool should have the same ambition: to make the clinician better over time, not more reliant. If the tool is the only thing standing between a competent clinician and a bad outcome, the tool has created a problem it didn't solve.
I want to be precise about what this critique does and doesn't apply to.
There are tasks where tool dependence is entirely appropriate — drug dose calculations where an error is dangerous, allergy checks where comprehensiveness matters, documentation that doesn't require clinical judgment. For those tasks, reliance is fine, because the alternative (manual calculation, memory-dependent cross-referencing) is worse.
The critique applies where clinical judgment is the crux. Diagnosis. Risk stratification. Differential generation. The question of whether to pursue a workup at all. These are the domains where a tool that does the thinking for you is not augmenting your practice — it's atrophying it.
The best clinical AI tools I've used feel more like working with a brilliant medical librarian than a substitute attending. They surface evidence I should know. They remind me of considerations I've thought about before and can evaluate. They speed up the research I would have done anyway. They promote critical appraisal and prompt unique considerations.
And when they're off, I'm better than I was before I started using them.
That's the test.
MDCalc has always aimed to be the second opinion that makes clinicians sharper, not the assistant that makes the decision. Our Quality Rating System isn't a decision — it's a structured way of capturing your clinical reasoning in a form the evidence can speak to. The reasoning is still yours.