The study was conducted at St. Mary's Hospital, Montreal, a 400-bed university-affiliated primary acute care facility. All patients aged 65 or more admitted to the 5 general medical units between Mar. 15, 1996, and Jan. 31, 1999, were eligible. Excluded were patients who met 1 or more of the following exclusion criteria: primary diagnosis of stroke, duration of stay on the intensive care unit or cardiac monitoring unit of more than 48 hours, admission to geriatric or oncology service, inability to speak English or French, or residence other than on the island of Montreal.
To detect prevalent cases of delirium, eligible patients were screened within 24 hours after admission by the study nurse using the Short Portable Mental Status Questionnaire. Those who scored 3 to 9 errors on this instrument or had symptoms of delirium recorded in the nursing notes were assessed by means of the Confusion Assessment Method.22
This is a structured instrument that allows standardized recording of the 9 symptom domains of delirium specified in the Diagnostic and Statistical Manual of Mental Disorders
, third revised edition2
(DSM-III-R). To detect incident cases of delirium, all patients without prevalent delirium were rescreened during the week following admission. Those who scored 1 point higher on the Short Portable Mental Status Questionnaire than on admission or had symptoms of delirium recorded in the nursing notes were assessed with the Confusion Assessment Method. Patients with prevalent or incident delirium (criteria of the DSM-III-R2
) were enrolled in the study.
Following enrolment, informed consent was obtained from the patient or substitute decision-maker. Patients were randomly allocated by means of computer-generated random numbers to receive the intervention or usual care on the 5 medical units. Stratified randomization was used; that is, independent randomization was done within groups of prevalent and incident cases respectively. Furthermore, we performed blocking using blocks of different sizes to guarantee similar numbers of patients in the control and intervention groups at any point and to ensure that the intervention team was not overloaded by a large number of patients during any period. Unequal block size also helped to maintain blinding as to treatment allocation. The study statistician prepared two series of sealed envelopes, one for each stratum, containing the treatment allocation. Following enrolment, the study nurse wrote the name of the patient on the appropriate envelope, then opened it to determine group allocation. If the patient was allocated to the intervention group, the study nurse started the intervention.
It was not feasible to randomly assign patients to intervention or control units in our hospital; however, in the previous trial, there was no evidence of contamination when patients in the 2 groups were managed on the same unit.20
Patients were asked for consent to participate in the study, and a close family member was asked for informed consent. The trial was approved by the hospital's Research Ethics Committee.
The intervention consisted of 2 parts: consultation and follow-up by a geriatric internist or psychiatrist, and follow-up in hospital by the study nurse. The consultation (within 24 hours after enrolment) determined the probable predisposing, precipitating and perpetuating factors of delirium (focusing on crucial factors associated with delirium, such as medication, infection and sensory deficits) and resulted in management recommendations (e.g., changes in medications and investigations to be carried out), which were recorded on a regular hospital consultation form and signalled in the progress notes. The follow-up by the study nurse involved daily visits (mean duration 35.7 minutes [standard deviation 28.8]) to conduct a brief structured mental status exam, monitor the completeness of the consultant's reports, ensure that previous recommendations had been implemented, ensure implementation of the nursing intervention protocol () by liaising with the primary care nurses, and meet with the patient's family to involve them in patient care.
The intervention in the current study was more intensive than that in the previous trial20
in 4 ways. First, the consultant not only assessed but also followed the patients as required. Second, the study nurse visited the patient 5 days per week. Third, the intervention team (comprising 2 geriatric psychiatrists, 2 geriatric internists and the study nurse) met after every 8–10 patients were enrolled in the intervention group to discuss delirium management problems. Finally, the primary investigator met weekly with the study nurse to discuss problems of diagnosis, enrolment and interventions.
Usual care consisted of standard hospital services. Referrals (by attending physicians) for geriatric or psychiatric consultation were honoured consistent with usual practice, but patients in the usual care group did not receive systematic consultation by the geriatric specialists, follow-up by the study nurse or the nursing intervention protocol. Hospital staff were not informed of the diagnosis of delirium. We felt this approach was justified, for 2 reasons. First, patients were receiving care equivalent to the standard of care in most Canadian hospitals and, second, there is no evidence that detection of delirium and disclosure of the diagnosis to other caregivers is beneficial.
Within 24 hours after enrolment a research assistant, blinded as to treatment allocation, administered a baseline Mini-Mental Status Exam (MMSE),23
the Delirium Index24
and the Barthel Index.25
The MMSE measures cognitive function on a scale of 0 (poor) to 30 (excellent), a score of 23 or less indicating cognitive impairment. The Delirium Index measures the severity of delirium; the total score ranges from 0 (no symptoms) to 21 (severe symptoms). The Barthel Index measures independence in personal care activities; we used a modified scoring system that ranged from 0 (dependent) to 100 (independent).26
Subsequently, patients were evaluated 3 times (every other day) during the first week and weekly thereafter until discharge or 8 weeks in hospital.
Information related to length of hospital stay, survival within 8 weeks after enrolment and living arrangements after discharge was obtained from the hospital database and from follow-up interviews with informants. Living arrangements were ordered hierarchically from least dependent (e.g., home alone) to most dependent (e.g., nursing home); living arrangements at discharge were compared with those at admission and were rated as more dependent, same or less dependent.
Processes of care involved documentation of delirium by attending physicians, medication changes, consultations to occupational therapy, recreational therapy or social work, personal possessions at the bedside and nursing documentation of emotional support or orientation. To reflect processes of care, a present state checklist was completed within 24 hours after enrolment by the research assistant. Details of the methods and results of this process-of-care assessment are found in Appendix 1 (available online at www.cmaj.ca
On enrolment the study nurse rated the overall clinical severity of illness27
and collected demographic data (including living arrangements before admission). Clinical severity of illness was rated on a scale from 1 (mild) to 9 (moribund). A family member was interviewed by the research assistant to complete the short form of the Informant Questionnaire on Cognitive Decline in the Elderly.28
This instrument assesses the possible presence of dementia before admission based on the responses of an informant who had known the patient for at least 5 years and could assess any decline in memory or cognition; the score is an average of the 16 item scores, each rated from 1 (much improved) to 5 (much worse). Patients with a mean score of 3.51 or more were considered to have dementia. After discharge (or 8 weeks in hospital) medical records were reviewed by a nurse-abstractor to complete the Charlson Comorbidity Index29
and living arrangements after discharge. The Charlson Comorbidity Index measures the number and severity of coexisting medical conditions, higher scores indicating more comorbidity.
The interrater agreement for delirium versus no delirium using the Confusion Assessment Method was excellent (kappa value 1.0) (n
= 14); the study nurse's diagnosis of delirium had a sensitivity of 0.89 (n
= 87) and a specificity of 1.0 compared with a consensus diagnosis.30
The interrater concordance correlation coefficient for the MMSE was 0.99 (n
To check on the extent of blinding of the research assistant during the study, the assistant was asked to guess which study group a convenience sample of 18 patients (whom the research assistant had recently assessed) belonged to. The research assistant's responses were correct for 2 of 4 patients in the intervention group and 3 of 5 patients in the usual care group; the assistant responded “don't know” for the 9 remaining patients (4 in the intervention group and 5 in the usual care group).
Characteristics of the patients at admission were compared between the study groups. Analyses of intervention effectiveness followed the intention-to-treat principle. The primary analysis was performed on a constructed binary outcome: “improved” or “not improved” during the hospital stay. Improvement in cognitive status was defined a priori as an increase in the MMSE score of 2 or more points (judged to be the minimum clinically significant change) compared to baseline, and no decrease below baseline plus 2 points in subsequent assessments. If the baseline MMSE score was 27 or more, improvement was defined as no decrease in the MMSE score below 27 in subsequent assessments. A patient who died was classified in the “not improved” category. The time to improvement was defined as the number of days between enrolment and the first assessment satisfying the definition of improvement. We used the Cox proportional hazards model31
to compare the intervention and usual care groups. The data were considered censored in the analysis for patients who had not improved at the time of discharge or withdrawal or 8 weeks after enrolment. In this context, a hazards ratio (HR) greater than 1 means a shorter time to improvement in the intervention group. In the secondary analyses, the interaction of the intervention with dementia, with comorbidity (Charlson Comorbidity Index < 3 v. ≥ 3) and with prevalent versus incident delirium were evaluated in separate Cox proportional hazards models because of the high level of multicolinearity between the covariates. Because the intervention was not expected to affect death, in secondary analyses we excluded deaths from the “not improved” category and considered deaths as censored.
To examine possible contamination effects, we divided the enrolment period into 3 equal time periods; the main analysis of time to improvement for the MMSE was repeated with the effects of time period and interaction of time period by intervention added to the Cox proportional hazards models used. Finally, we conducted an efficacy analysis in which we excluded patients in the intervention group who did not receive a consultation and those in the usual care group who did receive a consultation.32
We applied the same analyses for the secondary outcomes, the Delirium Index and the Barthel Index. The level of significance was set at 0.05. Confidence intervals (CIs) for the HRs were computed at the 95% level.