Megan Skiendel: David, quick question before we even start — do you know your cholesterol number off the top of your head?
David Sterling: I do. Why?
Megan Skiendel: Because statistically it may tell us almost nothing about whether you specifically are going to have a cardiac event. Up to 50% of heart attacks — half — happen in people whose cholesterol read as normal beforehand.
David Sterling: So the marker fails half the time on the exact outcome it exists to prevent.
Megan Skiendel: And yet it remains the primary clinical marker — not because the evidence says it's the best tool, but because nothing cheaper and better-validated has shown up to replace it. That's the honest version.
David Sterling: The Framingham Heart Study is where this starts — population-level correlations across thousands of subjects, genuinely solid science. The problem is, I mean — at some point those correlations stopped being described as group statistics and started getting applied as individual predictions.
Megan Skiendel: Which is a category error dressed up as clinical practice.
David Sterling: Frankly, yes. A number that describes a crowd being handed to one person as their future. That's the central tension here.
Megan Skiendel: And the category error has a very specific shape. It's not just 'population data applied to one person.' It's that SCORE2, the Pooled Cohort Equations — these tools were built to allocate resources across a population. They were never designed to tell you, this specific patient, what happens next.
David Sterling: Right — but it's structural. Think about a weather forecast: '30% chance of rain across the northeast.' That number is real. It's useful. But it tells you almost nothing about whether it rains on your specific street corner.
Megan Skiendel: That's the gap. That's exactly it.
David Sterling: SCORE2 and the Pooled Cohort Equations paper over that gap every day in clinical practice. A relative risk from a population study tells you how outcomes differ between groups — it cannot tell you the absolute probability for one person. Those are structurally different questions.
Megan Skiendel: And the reason it breaks down — I mean, put two patients in the same cardiologist's office, identical LDL readings. One smokes, has kidney inflammation, father died of a cardiac event at 52. Same number on the lab report. Completely different risk profile. The Pooled Cohort Equations don't — they can't — hold all of that.
David Sterling: Age, smoking, kidney function, inflammation, physical activity, family history — that's before you even get to medication use and prior disease. The number is the same. The risk is not.
Megan Skiendel: And the 2019 ACC/AHA guideline actually acknowledged this — quietly. They introduced 'risk-enhancing factors,' things like lipoprotein(a) and high-sensitivity CRP, specifically because the standard lipid panel was leaving clinical decisions unresolved. That's the guidelines admitting the single-number model is insufficient.
David Sterling: Which is actually a concession buried in procedural language. The Framingham correlations were the foundation. But the distance between 'this marker predicts outcomes in a group' and 'this marker predicts your outcome' — that's the load-bearing gap nobody named out loud until a guideline in 2019 started routing around it.
Megan Skiendel: And the 'routing around it' language is doing a lot of work there — because what the 2019 ACC/AHA guideline actually did by naming lipoprotein(a), apolipoprotein B, high-sensitivity CRP as 'risk-enhancing factors' is admit that the same LDL number produces genuinely different outcomes in different people. That's not a procedural footnote. That's the entire architecture of the problem.
David Sterling: Which is multifactorial disease causation, stated in guideline language.
Megan Skiendel: Right — but it goes deeper than 'there are multiple factors.' The reason the same HbA1c reading produces diabetes in one person and not in another isn't just lifestyle variation. It's that the genetic architecture underneath it is different. The environmental exposures are different. HbA1c predicts risk across a population beautifully — and then you try to apply it to one person carrying a specific gene variant and a specific stress profile and — I mean, the population signal just — it doesn't hold.
David Sterling: So the number is real. The prediction breaks on the individual.
Megan Skiendel: Every time. And this isn't only a cardiovascular problem — that's what gets me. There's an IBD study where the Harvey-Bradshaw Index couldn't distinguish active from inactive Crohn's disease. Standard biochemical biomarkers failed to predict endoscopic remission. Different disease entirely, same structural collapse.
David Sterling: Wait — the Harvey-Bradshaw Index failed to differentiate active from inactive? That's — what's the mechanism there, clinically?
Megan Skiendel: Same one. Crohn's is polygenic, environmentally triggered, individually variable. One patient's 'active' looks like another's 'remission' on the biochemical panel. The single marker is measuring the wrong layer of a multifactorial disease. Cardiology, metabolic disease, gastroenterology — the failure mode is identical across all three.
David Sterling: So the 4% validation rate from the EMA's 883 assessment reports isn't a cardiology number. It's a medicine number.
Megan Skiendel: That's the structural claim, yeah. And — honestly, the part that makes this even stranger comes when you look at individual patients whose biomarker readings appear healthy. Because 'healthy' on a panel turns out to not be a state so much as a timestamp. We'll get to exactly how dynamic that is — the data on how long 'healthy' actually lasts is — it reframes everything we just said.
David Sterling: That timestamp idea — let me make it concrete, because I think the data is harder than 'health is dynamic.' Imagine a 44-year-old woman, annual physical, late October. BMI 22, LDL 118, blood pressure unremarkable. Every number green. Doctor says see you next year. The longitudinal finding is — well, only about 15% of women classified as metabolically healthy at normal weight actually stayed that way after 30 years of follow-up. So that clean panel had an 85% failure rate as a long-term description of her.
Megan Skiendel: Eighty-five percent. That's not drift. That's the rule.
David Sterling: And here's what makes it structurally strange — the failure runs in both directions. You can have a perfect score and be in genuine danger. Or a 'bad' number and actually be protected.
Megan Skiendel: Which is exactly what MHO and MUNW prove. Metabolically healthy obesity — obese by BMI, but blood pressure, glucose, and lipids all normal. The number says danger. The metabolic picture says, actually, not yet. And then Ruderman, back in the 1980s, described the flip: metabolically unhealthy normal weight — BMI under 25, looks fine on paper, but carrying abdominal fat, elevated blood pressure, hyperinsulinemia. The weight number is clean. The risk is not.
David Sterling: So the biomarker is measuring — what, exactly? Not the underlying biology.
Megan Skiendel: It's measuring a proxy that correlates with the biology at the population level. But for the MUNW individual — I mean, their BMI is the 'right' number, they leave reassured, and the actual risk is invisible to every standard tool in the room.
David Sterling: That's the clinical consequence. Optimizing the reading and improving the outcome are two different operations. They can move in opposite directions.
Megan Skiendel: And no one's telling that 44-year-old woman her clean panel has an 85% chance of not describing her in 30 years. She leaves with a number. The number feels like an answer.
David Sterling: The number is a snapshot being sold as a forecast. That's the load-bearing problem — and it's not solvable by adding one more marker to the panel.
Megan Skiendel: And the escape route everyone reaches for is — more data. Multi-omics panels, genomics, metabolomics, proteomics all layered together. Precision medicine. The idea that if we just measure enough, the individual finally comes into focus. But the EMA ran 883 assessment reports on candidate predictive biomarkers — 883 — and 37 validated. Roughly 4%. So the question I can't shake is: does adding more unvalidated layers actually close the gap, or does it just compound the complexity?
David Sterling: Well, the math on that is uncomfortable. If individual markers fail at 96%, combining them doesn't automatically improve prediction — you're stacking unmeasured confounders, not canceling them.
Megan Skiendel: Although — HER2 in breast cancer. BRAF in melanoma. Those are single markers that do predict therapeutic response at the individual level. So it's achievable.
David Sterling: They're real. But they're the exception, not the template. And the incentive structure doesn't produce more HER2s — it produces more cheap, reimbursable, legally defensible single tests. The system rewards treating the number. That's a different problem than a measurement one.
Megan Skiendel: Which maybe means the actual question isn't whether to measure more. It's whether medicine is willing to name the uncertainty it's been hiding behind one clean number. I don't know that it is.
David Sterling: I think that's where we actually land. Uncertainty as a feature — not a bug. That's a hard sell inside a system built on defensible thresholds. I don't have a resolution. I don't think you do either.
