The hypothesis was so obvious it barely needed testing. Good teams make fewer mistakes.
Amy Edmondson set out to show it in the place where mistakes matter most. On eight nursing units in two teaching hospitals, trained investigators spent six months tracking medication errors. Edmondson, working separately, surveyed the nurses, doctors and pharmacists on each unit about their leaders and their teamwork.
When she ran the numbers, the result was statistically significant. “The sign was in the wrong direction,” she said later. The better a unit’s leadership and teamwork, the more errors it had. Detected error rates ran from 2.3 to 23.7 per thousand patient days, roughly a tenfold spread.
“I sat there, I think I froze,” she recalled. “I was upset and I was scared.” Her first reaction: I must have made a mistake.
Then she noticed something. Better teams might not be making more mistakes. They might just be more willing to report them.
To check, she sent a second researcher onto the units. She told him nothing about the error rates or the surveys. He watched each unit for several days and came back with a verdict: “They really are very different places to work.”
His ranking of how open each unit felt matched the error ranking almost unit for unit. On the most open unit, the nurse manager worked in scrubs and called a non-punitive atmosphere essential. A nurse there had phoned a physician at home to question a dose she thought was too high. On the most closed unit, the manager wore a business suit and held conversations behind a closed door. A nurse who had hurt a patient while drawing blood said she was treated like a two-year-old, and expected trouble for saying so.
One survey question told the story on its own: if you make a mistake on this unit, is it held against you? Edmondson said the answers were “profoundly, significantly correlated” with the error rates.
There was one more clue. Some adverse drug events can’t be prevented, like an allergy no one could have known about. For those nonpreventable events, the study found no significant link to how a unit was run.
The units that looked the most dangerous may have been more willing to let mistakes enter the record.