The Minnesota trial began in November 1968 and included more than nine thousand men and women in six state mental hospitals and one nursing home. Half of the patients were served a typical American diet, and half a cholesterol-lowering diet that included egg substitutes, soft margarine, low-fat beef, and extra vegetables; it was low in saturated fat and dietary cholesterol and high in polyunsaturated fat. Because the patients were not confined to the various mental hospitals for the entire four and a half years of the study, the average subject ate the diet for only a little more than a year. Average cholesterol levels dropped by 15 percent. Men on the diet had a slightly lower rate of heart attacks, but the women had more. Overall, the cholesterol-lowering diet was associated with an increased rate of heart disease. Of the patients eating the diet, 269 died during the trial, compared with only 206 of those eating the normal hospital fare. When I asked Frantz in late 2003 why the study went unpublished for sixteen years, he said, “We were just disappointed in the way it came out.” Proponents of Keys’s hypothesis who considered the Helsinki Mental Hospital Study reason enough to propose a cholesterol-lowering diet for the entire nation, never cited the Minnesota Coronary Survey as a reason to do otherwise.

As I implied earlier, we can only know if a recommended intervention is a success in preventive medicine if it causes more good than harm, and that can be established only with randomized, controlled clinical trials. Moreover, it’s not sufficient to establish that the proposed intervention reduces the rate of only one disease—say, heart disease. We also have to establish that it doesn’t increase the incidence of other diseases, and that those prescribed the intervention stay healthier and live longer than those who go without it. And because the diseases in question can take years to develop, enormous numbers of people have to be included in the trials and then followed for years, or perhaps decades, before reliable conclusions can be drawn.