HoldsUp

Eric Topol

Eric Topol is a medical researcher whose work focuses heavily on the application and evaluation of artificial intelligence and digital technologies across healthcare. His published research explores clinical AI foundation models, dynamic AI evaluation, robotic surgery, and biological aging clocks. Additionally, he has published studies on multi-cancer early detection, COVID-19 treatments, metabolic mechanisms in neurodegeneration, and appraisals of the wellness industry.

59 claims checked on air: 1 context 1 contradicted 11 overstated 41 supported 5 unverified

What they said on air - overstated

0:27:20overstatedmoderateIt's Not Cholesterol. Inflammation Is What's Actually Causin

Artificial intelligence analysis of retinal photographs can predict the onset of Alzheimer's disease 5 to 7 years in advance.

"You can do a retina AI exam. So I have a picture of the retina and you do AI on it, and it tells you when you're going to have Alzheimer's, if you're going to have Alzheimer's 5 to 7 years in advance." (said at 0:27:20)

Large prospective cohort studies (including data from the UK Biobank) demonstrate that artificial intelligence and deep-learning models applied to retinal photographs or optical coherence tomography can identify retinal structural features and biological aging markers associated with an increased relative risk of developing Alzheimer's disease or dementia years (e.g., 5 to 12 years) before clinical diagnosis. However, framing this as a ready clinical exam that deterministically "tells you when you're going to have Alzheimer's, if you're going to have Alzheimer's" significantly overstates the evidence. Current AI retinal models quantify statistical risk and biological vulnerability in research cohorts; they are not clinically diagnostic or deterministic prognostic tests for individual patients.

0:32:24overstatedhighIt's Not Cholesterol. Inflammation Is What's Actually Causin

Statins cause severe leg cramps and muscle-related symptoms in patients.

"Some of the data that comes out of these big meta-analyses, which say, "Oh, people don't get any leg cramps." That's not true. You and I know that's not true. People do get severe leg cramps where they can't even sleep at night, you know, uh, and and all sorts of other, you know, leg, uh, and muscle-related symptoms." (said at 0:32:24)

While statin therapy is causally associated with a small excess rate of muscle pain or weakness, the severity and frequency attributed to statins are significantly overstated. A comprehensive individual participant data meta-analysis of 19 large double-blind, placebo-controlled randomized trials (n=123,940) by the Cholesterol Treatment Trialists' Collaboration found that statins caused a small excess of mostly mild muscle symptoms during the first year of treatment (absolute excess of 11 events per 1,000 person-years; rate ratio 1.07). However, after the first year, there was no significant excess risk, and more than 90% of all muscle symptom reports among statin-treated participants were not attributable to the drug, occurring at near-identical rates in the placebo groups (nocebo/drucebo effect and background prevalence). Severe muscle damage (such as rhabdomyolysis or severe myopathy) is very rare.

0:33:00overstatedmoderateIt's Not Cholesterol. Inflammation Is What's Actually Causin

Failing to lower LDL cholesterol below 100 mg/dL or 70 mg/dL places individuals at higher risk for dementia.

"if you don't have the LDL lowered to let's say less than 100, less than 70, you're going to be at higher risk for dementia." (said at 0:33:00)

While some observational cohort analyses find that individuals with LDL cholesterol levels below 70 mg/dL have lower rates of incident dementia compared to those with higher levels, randomized controlled trial evidence does not support the claim that actively lowering LDL cholesterol to specific targets (e.g., <100 or <70 mg/dL) reduces dementia risk. Cochrane systematic reviews of randomized trials evaluating statin-induced LDL lowering found no significant reduction in the incidence of dementia or Alzheimer's disease, and trial data from PCSK9 inhibitor studies similarly show no significant effect on dementia outcomes.

0:00:06overstatedmoderateAlzheimer's Starts 20 Years Before You Notice It — Here's Wh

Plasma p-tau217 levels begin to rise 20 years before the onset of mild cognitive impairment.

"p-tau217 is the very first one that goes up and it starts 20 years before mild cognitive impairment. 20 years." (said at 0:00:06)

The speaker bundles two claims: that plasma p-tau217 is the 'very first' biomarker to change, and that it rises approximately 20 years before the onset of mild cognitive impairment (MCI). Longitudinal cohort data, such as a 25-year follow-up from the Women's Health Initiative Memory Study (PMID: 41805953), confirm that elevated baseline plasma p-tau217 levels are significantly associated with incident MCI and dementia up to 20 to 25 years before clinical manifestation. However, p-tau217 is not the 'very first' biomarker to become abnormal along the Alzheimer's disease continuum; trajectory modeling demonstrates that reductions in the plasma Aβ42/Aβ40 ratio precede increases in plasma phosphorylated tau species (PMID: 40539416).

0:22:45overstatedmoderateAlzheimer's Starts 20 Years Before You Notice It — Here's Wh

Blood-based p-tau217 testing performs as accurately as cerebrospinal fluid analysis and PET tau neuroimaging for Alzheimer's biomarker detection.

"because it's as good as a cerebrospinal fluid, it's as good as a PET tau scan, you know, which is a lot of radiation and hard to get" (said at 0:22:45)

While blood-based phosphorylated tau 217 (p-tau217) is an accurate biomarker for identifying Alzheimer's disease pathology (amyloid-beta positivity) and approaches the diagnostic accuracy of cerebrospinal fluid (CSF) testing in clinical screening, asserting that it is universally 'as good as a PET tau scan' is overstated. Cohort studies demonstrate that plasma p-tau217 assays primarily reflect early soluble tau phosphorylation in response to amyloid rather than the anatomical distribution or severity of insoluble neurofibrillary tangle pathology. For differentiating late-stage tau accumulation (A+T+ from A+T-), the accuracy of plasma p-tau217 assays decreases significantly (AUC 0.69–0.77), where tau-PET imaging remains the standard for staging.

0:27:10overstatedhighAlzheimer's Starts 20 Years Before You Notice It — Here's Wh

In clinical trials, diabetic patients on GLP-1 drugs only lose 3 to 4 pounds on average, whereas non-diabetic individuals with obesity lose 40 to 80 pounds.

"And they kept saying to her, 'Well, Lotte, it's not going to work because the diabet- the diabetics only lose three or four pounds.' Well, now we see they we can get people to lose, you know, 40, 50, 60, 80 pounds. These drugs are so potent, and the reason we we it was blown was because the diabetics don't lose that weight" (said at 0:27:10)

While clinical trials consistently show that patients with type 2 diabetes experience somewhat less weight loss on GLP-1 receptor agonists than individuals without diabetes, the claimed disparity (3–4 pounds vs. 40–80 pounds) is a substantial exaggeration. In Phase 3 randomized trials of semaglutide 2.4 mg, adults with type 2 diabetes achieved an average weight loss of approximately 9.6% to 10% (around 20–22 pounds, as in the STEP 2 trial), far exceeding 3 to 4 pounds. Meanwhile, non-diabetic adults with overweight or obesity average approximately 15% to 17% weight loss (about 33–37 pounds across STEP 1, 3, and 5), rather than an average loss of 40 to 80 pounds.

0:30:44overstatedmoderateAlzheimer's Starts 20 Years Before You Notice It — Here's Wh

The United States has the highest consumption of ultra-processed food in the world, with average consumption exceeding 70% of total diet and many individuals consuming 80% or more.

"and the US has the highest consumption in the world, 70% plus. And of course, a lot of people are 80% or more." (said at 0:30:44)

While systematic reviews confirm that the United States and the United Kingdom have the highest ultra-processed food (UPF) consumption rates internationally (generally exceeding 50% of total caloric intake), the speaker substantially overstates average US intake levels. Nationally representative data from NHANES show that the average contribution of UPFs to total energy intake is approximately 53% among adults and 62% among youth (overall average near 57–60%), well below the claimed average of '70% plus'.

0:53:25overstatedmoderateAlzheimer's Starts 20 Years Before You Notice It — Here's Wh

GLP-1 receptor agonists reduce cardiovascular inflammation prior to weight loss and prevent heart failure with preserved ejection fraction (HFpEF).

"We've already seen how GLP-1 drugs do that before any weight loss. So that should work well in people who aren't even obese, and we've seen how that can prevent heart—preserve ejection fraction heart failure, which is half of all heart failure, right? GLP-1s prevent that." (said at 0:53:25)

Glucagon-like peptide-1 receptor agonists (GLP-1 RAs), such as semaglutide, demonstrate significant anti-inflammatory effects that occur partly independent of the magnitude of weight loss. In the STEP-HFpEF program (STEP-HFpEF and STEP-HFpEF DM randomized controlled trials), semaglutide significantly reduced C-reactive protein (CRP) levels, improved heart failure symptoms and physical limitations, and in pooled analyses reduced worsening heart failure events in patients with established heart failure with preserved ejection fraction (HFpEF). However, stating that GLP-1 drugs 'prevent HFpEF' and 'should work well in people who aren't even obese' overstates the evidence. The pivotal clinical trials specifically enrolled individuals with overweight or obesity (e.g., BMI ≥30 kg/m²), and evaluated the management of established obesity-related HFpEF and secondary worsening events rather than primary prevention in non-obese populations.

0:54:07overstatedlowAlzheimer's Starts 20 Years Before You Notice It — Here's Wh

AI analysis of retinal photographs can predict the onset of Alzheimer's disease 5 to 7 years in advance.

"So I have a picture of the retina and you do AI on it, and it tells you when you're going to have Alzheimer's, if you're going to have Alzheimer's 5 to 7 years in advance." (said at 0:54:07)

While research using deep learning on retinal imaging (such as fundus photography and optical coherence tomography) shows that retinal microvascular and structural features are associated with neurodegenerative risk factors and preclinical changes years before diagnosis, artificial intelligence cannot deterministically predict whether or precisely when an individual will develop Alzheimer's disease 5 to 7 years in advance. Existing deep learning models are investigational tools evaluated in large research cohorts (such as the UK Biobank) that demonstrate statistical risk associations and classification capabilities, but they are not validated clinical diagnostic tools that pinpoint future Alzheimer's onset.

1:00:15overstatedmoderateAlzheimer's Starts 20 Years Before You Notice It — Here's Wh

Failure to lower LDL cholesterol below 100 mg/dL or 70 mg/dL is associated with an increased risk of developing dementia.

"and Alzheimer's, as you know, accounts for 70% of dementia—that if you don't have the LDL lowered to, let's say, less than 100, less than 70, you're going to be at higher risk for dementia." (said at 1:00:15)

While observational studies and meta-analyses indicate that elevated midlife LDL cholesterol (e.g., >121 mg/dL) is associated with an increased risk of Alzheimer's disease, there is no evidence establishing that failing to achieve specific cardiovascular targets (<100 mg/dL or <70 mg/dL) increases dementia risk. Furthermore, an individual participant data meta-analysis of over 21,000 older adults found no significant association between LDL cholesterol levels and incident dementia or cognitive decline, and meta-analyses of randomized trials of intensive LDL-lowering therapies (such as PCSK9 inhibitors) have not shown a statistically significant reduction in the risk of dementia or Alzheimer's disease.

1:10:54overstatedlowAlzheimer's Starts 20 Years Before You Notice It — Here's Wh

Personalized neoantigen vaccines have achieved cures in patients with treatment-refractory pancreatic cancer and renal cell carcinoma.

"personalized neoantigen vaccines to cure pancreatic cancer, to cure renal cell carcinoma, intractable, that is people that failed everything else." (said at 1:10:54)

Personalized neoantigen vaccines have shown promising immunogenicity and improved recurrence-free survival in early phase I clinical trials, but they have not been shown to 'cure' pancreatic cancer or renal cell carcinoma (RCC), nor were the landmark trials conducted in patients with intractable disease who had failed all other therapies. In a phase I trial for pancreatic ductal adenocarcinoma (PDAC), an individualized mRNA neoantigen vaccine (autogene cevumeran) was administered as adjuvant therapy following complete surgical resection alongside chemotherapy and immunotherapy; 8 of 16 patients developed neoantigen-specific T-cell responses and experienced delayed disease recurrence, not confirmed cures. Similarly, a phase I trial of a neoantigen vaccine in RCC enrolled 9 patients with fully resected, high-risk disease in the adjuvant setting rather than treatment-refractory disease. These phase I trials demonstrate biological activity and preliminary feasibility in post-surgical settings, but neither establishes curative efficacy nor applies to intractable, end-stage disease.

Fact-checked episodes

Publications