A follow up after discharge from intensive care unit (ICU) [1] revealed that vivid 'dreams' occur in 32% of patients. Although it is questionable whether these are dreams in the true sense, there is no doubting their impact on the sufferer, as has been described by anecdotal and published accounts [2]. Aetiological factors are unknown. We undertook a study of a relatively homogeneous population who had undergone cardiac surgery in order to identify the incidence of 'dreams', to identify possibly causative factors and to inquire into the ongoing effects the 'dreams' have had.

Following Institutional Ethical Committee approval, and consent from each patient's general practitioner, we performed a retrospective, questionnaire-based study. Included in the study were all patients still living who had stayed 4 days or longer on the ICU after cardiac surgery. Patients were excluded if their general practitioner objected.

The questionnaire was a modification of that used by Jones et al [1] and inquired into dreams, memories for normal events on ICU (eg endotracheal suction), and subsequent flashbacks and panic attacks. Demographic and clinical details were obtained from the medical records of all those who replied. The population was split into dreamers (D) - if they could describe one or more of dreams, nightmares, travelling and persecution - and nondreamers (ND). The groups were analyzed by stepwise logistic regression and ?² testing where appropriate.

Of 423 eligible patients, 240 were still alive and contactable. Replies were received from 161 (response rate 67%). The duration of stay on the ICU ranged from 4 to 64 days. Dreams were recounted in 77 patients (48%). These varied from vivid stories (29%) to severe persecutory beliefs (19%), 60% of which involved members of staff. The mean number of 'normal' events remembered in the ICU was significantly higher in group D (P < 0.001). Sepsis and midazolam usage were more common in group D (odds ratios 3.6, 95% confidence interval 1.7-7.3 for sepsis; odds ratio 2.9, 95% confidence interval 1.3-6.2 for midazolam). The development of new neurological signs while in the ICU protected against dreams (odds ratio 0.3, 95% confidence interval 0.1-0.7). Flashbacks and episodes of panic after discharge from ICU were more common in group D (P < 0.001).

ICU patients have been shown not to enter rapid eye movement (dream) sleep, but 'dreams' are common. Unlike true dreams, the patients studied here could recall their experiences from years previously in minute detail. Although many of the 'dreams' could be interpreted as being a distortion of reality (an illusion), some bore no bearing on real events (a hallucination). It appears that, with time and adequate explanation, all patients had insight into their experiences, unlike in psychoses.

Illusions and hallucinations are common after prolonged recovery in an ICU after cardiac surgery, are more frequent among patients receiving midazolam and in those who are septic, and are associated with flashbacks and episodes of panic. In those with less memory for their stay in ICU, illusions are rare.