Re-evaluation of the traditional diet-heart hypothesis: analysis of recovered data from Minnesota Coronary Experiment (1968-73)
Christopher E Ramsden, Daisy Zamora, Sharon Majchrzak-Hong, Keturah R Faurot, Steven K Broste, Robert P Frantz, John M Davis, Amit Ringel, Chirayath M Suchindran, Joseph R Hibbeln - BMJ, 2016
Recovered data from the Minnesota Coronary Experiment show that replacing saturated fat with linoleic acid (corn oil) effectively lowered serum cholesterol but did not reduce mortality. Lower serum cholesterol was instead associated with a higher risk of death, and a meta-analysis of five RCTs found no mortality benefit from coronary heart disease or any cause. It is a key skeptic-cited result that lowering cholesterol via linoleic acid did not translate into longer life.
Key findings
- Each 30 mg/dL (0.78 mmol/L) reduction in serum cholesterol was associated with a 22% higher risk of death (HR 1.22; 95% CI 1.14-1.32; P < 0.001) in covariate-adjusted models - lower cholesterol tracked with higher mortality.
- The linoleic acid intervention lowered serum cholesterol as intended (mean change from baseline -13.8% intervention vs -1.0% control; P < 0.001).
- Meta-analysis of five RCTs showed no benefit on coronary heart disease mortality (RR 1.13; 95% CI 0.83-1.54).
- Meta-analysis of five RCTs showed no benefit on all-cause mortality (RR 1.07; 95% CI 0.90-1.27).
Effect measures
- Hazard Ratio: 1.22 (per 30 mg/dL / 0.78 mmol/L reduction in serum cholesterol)95% CI 1.14-1.32; P < 0.001
- Other: RR 1.1395% CI 0.83-1.54
- Other: RR 1.0795% CI 0.90-1.27
- Other: Mean serum cholesterol change -13.8% (intervention) vs -1.0% (control); P < 0.001
Why this evidence tier (Moderate)
- Risk of bias:
- Based on recovered, incomplete data from a 1968-73 randomized trial that was never fully published; the trial itself was randomized, but the re-analysis depends on data recovery and on covariate-adjusted observational modelling within the cohort for the cholesterol-mortality association.
- Precision:
- The within-cohort cholesterol-mortality hazard ratio is precise, but the pooled RCT mortality estimates have wide intervals that cross the null.
- Directness:
- Directly tests replacing saturated fat with linoleic acid on hard mortality endpoints, which is exactly the diet-heart question.
- Consistency:
- Concordant with other linoleic-acid replacement trials that failed to show a mortality benefit, but in tension with reviews showing cardiovascular-event benefit from polyunsaturated replacement.
- Funding / COI:
- Funded by the NIH Intramural Research Programs (NIAAA and NIA); no industry or commercial sponsorship, and authors declared no relevant financial conflicts. No funding-related concern.
Moderate certainty. A randomized intervention with recovered data showing cholesterol was lowered without a mortality benefit; strengthened by independent NIH funding, tempered by data-recovery limits and the observational nature of the cholesterol-mortality association.
- Population:
- Minnesota Coronary Experiment (1968-73): 9,423 men and women aged 20-97 in one nursing home and six state mental hospitals in Minnesota, randomized to a diet high in linoleic acid (corn oil) replacing saturated fat versus a control diet high in saturated fat. An accompanying systematic review and meta-analysis pooled five RCTs (n = 10,808).
- Conflicts of interest:
- Funded by the NIH Intramural Research Program with no declared industry funding; all authors declared no financial relationships that could appear to have influenced the work. None known.
- Funding:
- NIH Intramural Research Programs (NIAAA and NIA); no industry or commercial sponsorship.
Limitations
- Recovered, incomplete data from a trial conducted in 1968-73 and never fully published at the time.
- The 22% higher death risk per 30 mg/dL cholesterol reduction is an observational, covariate-adjusted association within the trial cohort, not proof that the intervention caused deaths.
- A 2024 technical correction exists (PMID 38942429) that only fixes an outdated link to the Broste thesis; it does not alter the hazard ratios or meta-analysis figures.
- Pooled RCT mortality estimates have wide confidence intervals that cross the null.