Karl Lauterbach tweeted on 15 June 2026: “New study in JAMA Internal Medicine shows clearly: the corona vaccine protects the heart.” Kay Klapproth — immunologist, freelance journalist and board member of the ida association in Heidelberg — replied: the control group was already 88.7 percent primary-immunized — what is being compared is not vaccinated versus unvaccinated, but boost-willing versus boost-weary. That, he said, is no proof of heart protection.
Both saw something correct. Both overreach — but in different ways: Lauterbach’s error is communicative in nature, Klapproth’s methodological, and it weighs more heavily. This is not a verdict in favor of the middle — it follows from the numbers.
What the study measures#
The study: Cai M, Xie Y, Al-Aly Z: “2024–2025 COVID-19 Vaccine and Major Adverse Cardiovascular Events Among US Veterans,” JAMA Internal Medicine, 15 June 2026, DOI 10.1001/jamainternmed.2026.1929. Design: propensity-weighted observational study with a target-trial-emulation approach, data basis the US veterans system, 349,085 in the COVID group, 690,574 in the control group, 8 months of follow-up.
The decisive methodological component is the active comparator: both groups received an influenza vaccine in the 2024/25 season. One group simultaneously took the current COVID-19 booster, the other did not. The comparison is therefore not vaccinated versus unvaccinated — but boost-takers versus boost-decliners, with both groups being active vaccination-deciders in the same season.
This design controls for the so-called healthy-vaccinee bias (HVB): the systematic finding that people who take up vaccination offers are healthier and live healthier than the population average anyway. A comparison with the completely unvaccinated would not be able to separate this effect out.
88.7 percent — the number is correct#
Table 1 of the study states it in black and white: in the COVID vaccine group, 97.3 percent had already received the COVID-19 primary immunization, in the control group 88.7 percent. Klapproth quoted the number correctly.
What it means: the difference between the groups is not vaccination history as such, but primarily the decision for or against the current 2024/25 booster. The study compares an already largely primary-vaccinated veteran population — and measures whether the additional booster makes a measurable difference in cardiac events within that population.
Klapproth’s thesis: if the groups differ mainly in their boost-readiness, the boost group might simply be more health-motivated — even beyond the influenza comparison point. This so-called adherer bias is known in the methods literature and is not entirely excluded a priori in active-comparator designs.
The question is whether the study tested this objection.
What the negative controls show#
It did test it. In eTable 5 the authors report the results of negative controls: clinical events that cannot be biologically influenced by a COVID vaccine. If a systematic selection bias exists — if the boost group is simply healthier — that bias would also produce an apparent “protective effect” at negative-control endpoints. This is the classic test mechanism.
The authors write: “All negative control analyses produced null results, consistent with a priori expectations.” Not a single negative-control result shows a significant effect in any direction.
This is the empirical test for exactly the confounder Klapproth is pointing at. It was carried out. It found nothing.
Added to this is eTable 4: twelve sensitivity analyses with alternative weighting models, different propensity-score estimations and multiple imputation — “all consistent in both direction and magnitude with the primary analysis.”
This does not mean that every theoretically conceivable residual bias is excluded. Negative controls do not test every possible confounder. But the specific health-related selection effect Klapproth points to was searched for — and not found.
What the numbers say#
The measured effect (primary endpoint: COVID-related major adverse cardiovascular events, MACE, 8 months of follow-up):
- Overall population: vaccine effectiveness 37.7% (95% CI 18.2–54.9), risk difference 2.0 in 10,000 (CI 0.9–3.7). Absolute risk: 3.37 versus 5.43 in 10,000.
- Under 65 years: not significant.
- 65–75 years: not significant.
- Over 75 years: VE 50.7% (CI 31.8–65.6), RD 5.5 in 10,000 — statistically significant.
- Single endpoints (statistically significant, some with wide confidence intervals): cardiovascular death VE 57.9% (95% CI 25.2–78.2), heart attack VE 38.5% (CI 4.3–62.3), heart-failure hospitalization VE 41.9% (CI 4.1–67.5). Heart attack and heart-failure hospitalization are technically significant — the lower CI bound lies just above zero — but the wide intervals show these subgroup analyses are underpowered.
- Stroke: VE 30.6% (CI −24.7 to 64.3) — not significant.
- All-cause MACE (all major cardiac events, not just COVID-attributed ones): VE 6.2% (CI 3.8–8.9), RD 23.7 in 10,000 — twelve times the COVID-MACE RD. The study does not fully resolve this tension. Mechanistically explicable: because the absolute cardiac risk of this older population is high and the COVID-specifically attributed share is a small fraction of it, even a low VE produces a large absolute RD. Alternatively readable: the sharp drop from VE 37.7% to 6.2% could be a signal that part of the primary effect stems from residual bias that has a stronger impact on narrowly defined COVID-attributed endpoints than on all of them. Both explanations fit the data. Neither side in this debate confronts this tension.
For a geriatric high-risk population — over-75s with cardiovascular pre-burden — an RD of 5.5 in 10,000 over eight months is not an irrelevant value. For the general population under 75, the study shows no significant effect on COVID-related cardiac events.
Where both overreach#
Lauterbach reads a general cardiovascular vaccine protection out of an RD of 2 in 10,000, a significance almost exclusively among the over-75s, and an all-cause MACE effect of 6.2%. The headline “the corona vaccine protects the heart” is broader than what the study measures: an age-dependent booster effect on COVID-attributed cardiac events in a US veteran population.
Klapproth concludes that the design renders the finding methodologically invalid. That overlooks what the authors set against it: negative controls with consistently null results and twelve consistent sensitivity runs. Whoever claims the described selection bias is present must explain why it left no mark in a single negative-control endpoint.
The interest structure is open on both sides: Lauterbach defends years of vaccination policy with this study and has an obvious interest in a broad heart-protection headline. Klapproth — immunologist and freelance journalist — has built a profile in the COVID debate with methods criticism of vaccine studies; the technical background makes his objection more worth examining than a politician’s, but it does not replace the answer to the negative controls. Neither does the interest situation explain the argument, nor does it refute it — but it does explain the intensity.
What the study does not say#
The study was conducted on US veterans: predominantly male, older, a specific health-care context. Transferability to the European general population is not established and is not claimed by the authors either.
This is missing in both camps: Lauterbach communicates the finding as general heart protection without naming the population specificity and age-dependence. Klapproth criticizes the design without addressing the negative controls. The methods debate being conducted publicly is incomplete in both camps.
Conclusion#
Cai et al. have presented a methodologically careful study — more careful than the public fight over it suggests. The active-comparator design is an established tool against HVB. The negative controls found no systematic bias. The sensitivity analyses are consistent.
The measured effect is real, small and age-dependent: 2 in 10,000 avoided COVID-related major cardiac events over eight months in a US veteran population, statistically significant almost exclusively for the over-75s.
Whoever turns this into “the vaccine protects the heart” reads beyond the numbers. Whoever declares the design invalid because of it does not explain why the described bias left no mark in any of the negative-control endpoints.
