The nation's sickest patients are being treated in wards that are frequently as loud as pneumatic drills, research suggests.
While most people's vision of intensive care units (ICUs) is one of dimmed lights, hushed voices and softly beeping machines, the reality is that noise levels are akin to a busy restaurant, researchers from Oxford University said.
On average, patients in ICUs can expected to be disturbed by noise once every seven to 16 minutes overnight, they found.
Professor Duncan Young, from Oxford's Kadoorie Centre for Critical Care Research and Education, said: "High levels of noise make it harder to sleep, sleep deprivation leads to confusion and confusion is thought to complicate the healing process and slow recovery.
"Yet our research found that during the day, noise levels in an ICU are equivalent to those of a busy restaurant.
"While things are quieter at night, we still found that sounds louder than 85 decibels - around the level of a road drill - were happening up to 16 times an hour."
The team's study, published in the Critical Care journal, saw them monitor five ICUs to see whether they complied with recommended noise levels.
The World Health Organisation (WHO) recommends the average hospital sound levels should not exceed 35 decibels (dB), with a maximum of 40 dB overnight.
The researchers found average sound levels always exceeded a perceived loudness (A-weighted decibels, dBA) of 45 dBA.
They wrote: "On average, there were approximately 25 minutes of every hour during the day when peak levels above 85 dBA occurred.
"Peak levels above 85 dBA occurred less frequently overnight but a patient can still expect to be disturbed at least once every 7 to 16 minutes of every hour between 10pm and 7am."
The team have worked to come up with a series of noise-reducing methods including replacing metal bins with plastic ones that close quietly and creating guidelines about volume settings on equipment.
They are also looking at developing a "noise display" so staff can actually see noise levels on the wards.