The Case Against Finding Out

We tend to treat a test as free information. Worst case, it comes back clear; best case, it catches something early. But a test result is not information sitting in the world waiting to be collected. It is a probability, filtered through who you are and why you took it, and acting on it has costs of its own. Today's question: when does looking for a problem create one?

The intuitive case for screening is almost airtight. Find disease earlier, treat it earlier, get better outcomes. In a new episode of Sigma Nutrition Radio, physician Austin Baraki argues that this reasoning is incomplete rather than wrong, and the gap is where a lot of avoidable harm lives [1].

Start with the distinction Baraki draws between two very different acts that look identical from the outside: screening a person who has no relevant symptoms, and investigating an actual clinical problem [1]. The same scan, the same blood panel, means different things depending on which situation you are in. When you investigate a symptom, you already have reason to think something is there, so a positive result is more likely to be real. When you screen an asymptomatic person, most of them do not have the disease, and that changes how a positive result should be read [1].

This is the probabilistic interpretation Baraki emphasises [1]. A test is not a verdict; it is evidence that shifts an estimate. If a condition is rare in the population being screened, even a fairly accurate test will produce a large share of false positives simply because there are so many healthy people for it to be wrong about. The test hasn't gotten worse. The base rate has done the damage. This is why the answer to "should I get tested?" depends less on the test's advertised accuracy and more on your prior likelihood of actually having the thing.

From there the episode traces the downstream costs that the simple story leaves out: false positives, incidental findings, overdiagnosis, overtreatment, and what Baraki calls diagnostic cascades [1]. An incidental finding is something the scan picks up that you were not looking for and that may never have troubled you. Overdiagnosis is the detection of something that technically meets a definition of disease but would never have caused symptoms in your lifetime. Overtreatment is what follows: procedures, medications, and their side effects applied to a problem that was never going to hurt you. A diagnostic cascade is the sequence where one ambiguous result triggers another test, which triggers another, each carrying its own risk of a misleading answer [1].

The part most worth sitting with is why screening can look more effective than it is. Baraki points to biases that inflate the apparent benefit [1]. Consider what happens mechanically when you find a slow, harmless condition earlier: the person now carries a diagnosis for more years, and their outcomes look excellent, not because anything was prevented but because you started the clock sooner and selected for the mildest cases. Survival statistics can improve even if not a single life is extended. This is the trap in reading screening data: the numbers move in the reassuring direction whether or not the intervention actually helped.

Nnone of this is an argument against testing. It is an argument against treating testing as consequence-free. The honest version of the question is not "could this find something?" but "if it finds something, will acting on it leave me better off than not knowing?" For someone with symptoms or elevated risk, the math often favours investigating. For a healthy person chasing reassurance through more panels and more scans, the same act carries a different and quieter set of risks [1].

The evergreen point underneath: in a culture that equates vigilance with responsibility, more data feels like more control. But a measurement you cannot act on wisely is not neutral. It is a decision waiting to be made under uncertainty, and the sustainable, no-extremes version of health includes knowing which measurements are worth taking in the first place.

Research Radar

• Processing may not be the right lens for plant-based diets. A large prospective cohort stratified plant-based foods by whether they were ultra-processed, and the lead author argues this classification doesn't meaningfully change how we interpret the link between plant-based patterns and long-term health [2]. The healthfulness of the overall pattern seems to carry more weight than the processing label.
• Not all training supplements do the same job. A network meta-analysis of trained athletes, restricted to those in structured programmes for at least six months, compared protein, creatine, and omega-3 across strength, endurance, and recovery outcomes [3]. Treating them as interchangeable misreads the evidence; each maps onto different outcomes.
• Strength training is a bone-health tool women are talked out of. A recent review notes that lower peak bone mass and estrogen-related loss leave women at higher fracture risk after menopause, and that resistance training helps, yet misconceptions about muscle gain deter participation [4].

One Thing To Try

Before your next test or scan, ask one question out loud: "If this comes back abnormal, what would I actually do differently?" If you don't have an answer, that's worth noticing before you book it.

Worth Your Attention

• Sigma Nutrition Radio #616, with Austin Baraki, MD — the clearest recent explanation of why more screening isn't automatically better health [1].
• Sigma Nutrition Radio #615, with Alysha Thompson, PhD — for anyone unsure whether "ultra-processed" should change how they eat plants [2].
• Comparative Effects of Protein, Creatine, and Omega-3 (Nutrients) — a rigorous sort-out of which supplement matches which goal in trained athletes [3].
• Strength unseen (Annals of Medicine) — a case for putting resistance training at the centre of women's long-term bone health [4].

The open asked when looking for a problem creates one. Baraki's answer is not to stop looking, but to look with a reason [1]. Health that fits a full life isn't the one with the most data points. It's the one where each thing you measure earns its place by changing something you'd actually do.