Some people skip the test because of the unpleasant steps needed to get ready for it.
"People complain to me all the time, 'It's horrible,' " said Sidney Winawer, a gastroenterologist at Memorial Sloan-Kettering Cancer Center in New York, who helped lead the study. "But look at the alternative."
A second study in Europe found that colonoscopies did a better job of finding polyps than another common screening tool - tests that look for blood in stool. Both studies were in Thursday's New England Journal of Medicine.
Colorectal cancer is the second-leading cause of cancer death in the United States and the fourth worldwide. More than 143,000 new cases of cancers of the colon or rectum are expected in the United States this year, and nearly 52,000 people will die from it, the American Cancer Society predicts.
Deaths from colorectal cancer have been declining for more than two decades, mostly because of screening including colonoscopies and other tests, the group says. People of average risk of colon cancer ages 50 to 75 should get screened, but only about half in the United States do.
A government-appointed panel of experts recommends one of three methods: annual stool blood tests; a sigmoidoscopy (scope exam of the lower bowel) every five years, plus stool tests every three years; or a colonoscopy once a decade.
In a colonoscopy, a thin, flexible tube with a tiny camera is guided through the large intestine. Growths can be snipped off and checked for cancer. Patients are sedated, but many dread the test because it requires patients to eat a modified diet and drink solutions the day before to clear out the bowel. It usually costs more than $1,000, compared with a $20 stool test.
Researchers at Sloan-Kettering have shown that removing polyps during colonoscopy can prevent colon cancer from developing, but it was not clear that it saved lives.
The new study followed 2,602 patients who had precancerous growths removed during colonoscopies for an average of 15 years. Their risk of dying from colon cancer was 53 percent lower than that expected among a similar group in the general population - 12 patients in the study died vs. 25 estimated deaths in the general population.