This is a prospective clinical study with 2 repeated measurements at a time interval of 7 days (T1 and T2). The rationale for the study design was to look at temporal variability in order to study predisposing and predicating risk factors for delirium, something which is the objective of a future article.

Subjects were recruited in three LTC facilities and one LTC unit of a large regional hospital, all located in the Quebec City area, Canada. A convenience sampling procedure was used. Residents with the diagnosis of dementia, aged 65 years and more, and with no history of psychiatric illness, were eligible for this study. Following all appropriate institutional ethics committee clearances, a research assistant met with the head nurse from each participating facility so as to identify eligible residents. Specifically, the unit head nurse was asked to use resident chart information to identify residents with a diagnosis of dementia.

Following identification of eligible residents, there was an informative letter about the study sent to the families of said residents, inviting them to contact the research assistant to obtain further information, if so desired. Once their questions had been satisfactorily answered, those who had contacted the research assistant were then invited to sign proxy consent forms. As well, the research assistant obtained assent for participation from those residents whose cognitive impairment was judged to be mild or moderate.

In total, there were 293 families of demented residents solicited. Of these, 122 did not respond (41.6%), 11 refused to take part (3.8%) and 160 agreed to give their consent (54.6%). Following enrolment, three residents died before the first evaluation and one was transferred, leaving a total of 156 participants at T1. One resident was hospitalized at T2 and was therefore excluded from the analyses.

The nurses of the residents were also invited to take part in the study and their consent also obtained. A total of 40 nurses participated in the study.

The research assistants were nurses who had completed 15 hours of instruction on delirium and dementia, given by a member of the research team. Instructions on the research procedures as well as direct supervision in the data collection for 15 participants were also provided. All observational measures were completed based on resident monitoring over a seven-hour period at T1 and T2. To avoid potential bias, there were two research assistants involved in this study. The first (RA-1) focused on completing the measures of delirium (forms and severity) and dementia (type, stage and severity). The second (RA-2), was blinded to each resident's delirium status, and conducted a standardized interview with the bedside nurses in order to gather the following information: the bedside nurse's rating of delirium (overall and symptoms-specific) and some baseline characteristics of the nurse (age, sex, level of education, years of experience as a nurse and years of experience in geriatrics). RA-2 also collected basic demographic data (age, sex, marital status, level of education, ethnicity) and health information on the resident (number of days since admission to LTC setting, behaviour problems, functional autonomy, comorbidity, pain, depression, sleep problems, visual and hearing impairment, dehydration, weight loss, brachial perimeter, geriatric fever, oxygen saturation, number of prescribed medications and the presence of physical restraints during the day). Some information was extracted from the medical chart and other information was observed or measured using specific instruments. It should be noted that the bedside nurses in our study received no specific instruction on dementia or delirium from the researchers.

At T1, RA-1 assessed the presence of delirium (using the Confusion Assessment Method [CAM]) and delirium symptoms (using 6 items of the Minimum Data Set-2 [MDS-2]. These measures were considered as the reference standard in analyses on sensitivity and specificity. During the seven-hour observation period, RA-1 had to perform at least three formal assessments of symptoms of delirium (e.g. attention, level of consciousness) by interviewing the residents; this before going on to complete the CAM at the end of observation period. At the end of the day, RA-2 then interviewed the bedside nurse and asked this latter to rate evidence of: 1) delirium (using one question) and 2) delirium symptoms (using the same 6 MDS-2 items), based on clinical judgment and observation made throughout the day. This same procedure was repeated at T2.

For the first two objectives of this study (detection rates of delirium and delirium symptoms by bedside nurses), sensitivities, specificities, positive and negative predictive values were computed. For the third objective, (factors associated with undetected delirium), delirium that was undetected by nurses at T1 was used as the outcome.

Our study was not of simple random design. In fact, every resident of each LTC facility who was eligible for the study was included in our sample. Nurses were not randomized on the residents either; they were the regular bedside nurses for those residents. In such a situation, it was important to verify that our design did not unduly inflate variance estimates in the planned analyses. Therefore the design effect was computed and was found to be very close to 1 (0.90–0.99), which means that the design of our study performs as if the resident-nurse pairs had been selected randomly.

First, simple descriptive analyses were used to describe the study population (residents and bedside nurses). Then, bedside nurse ratings of delirium (one question) were compared with RA-1 ratings of delirium, the reference standard. Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) with their 95% confidence intervals were computed using standard formulas for each time measurement (T1 and T2). This procedure was repeated to compare bedside nurse ratings of individual delirium symptoms (6 MDS items) with RA-1 ratings of the same delirium symptoms (6 MDS items). The decision to collect information on bedside nurse ratings of individual delirium symptoms using the MDS-2 was made after the beginning of the study. This meant that 31 residents were excluded from these analyses because of missing values.

To evaluate whether or not certain factors were associated with delirium that was undetected by bedside nurses, logistic regression analyses were performed. The outcome of these analyses was undetected delirium by bedside nurses at T1. Bivariate analyses were conducted to assess the crude association between each of the independent variables considered and undetected delirium. A significance level of 0.1 was used to select variables for subsequent multivariate analysis 49. All analyses were carried out using SAS for Windows, version 9.1.

It is worth noting that for two variables, forms of delirium and the age of the nurses, quasi-complete separation of the data was observed. This situation arises when, for example, nurses from one age group never detected delirium. It was important not to discard these effects because they may have some importance in the detection of delirium. As such it was decided to fit a logistic regression model using Firth's penalized likelihood procedure 50 as coded by Heinze and Schemper 51 in their FL SAS macro. Confidence intervals obtained are then approximate and their coverage probabilities may be larger than usually expected.

Baseline characteristics of the 156 residents are presented in Table 1. The mean age of the participants was 86.3 ± 6.9 (mean ± standard deviation) and the majority were female (73.7%). The mean time since their admission to their LTC facility was 2.6 years (SD = 2.3). The degree of cognitive impairment in this population was important with >95% of residents rated as having severe cognitive decline (Functional Assessment Staging) and Alzheimer disease was the most frequent type of dementia. About 41% of the participants had a high level of comorbidity (score ≥ 8 on the Charlson Comorbidity Index) and the mean number of prescribed medications was 9.1 (SD = 4.3). More than two third of the residents were showing important deterioration in their functional autonomy (score ≥ 29 on the SMAF).

Table 2 shows the baseline characteristics of the 40 bedside nurses who took part in the study. All were female with a mean age of 47.3 years (SD = 11.9). The majority of them (56.4%) had a college-degree educational level. The mean number of years of experience in nursing and in geriatrics was 21.2 (SD = 13.3) and 9.9 (SD = 8.3) respectively.

In all, 302 paired observations by bedside nurses and RA-1 were made in the 156 residents: 151 at T1 and 151 at T2. Five residents were excluded from the analyses because of missing values. Tables 3 and 4 present the comparison of bedside nurse and RA-1 delirium ratings for each time measurement (T1 and T2).

At T1, the prevalence of delirium was 71.5% (n = 108). Of the 108 residents with delirium, 14 were detected by the bedside nurses, corresponding to a sensitivity of 13%. Alternatively, bedside nurses identified 40 of the 43 cases without delirium, which represents a specificity of 93%. Of the 17 residents rated by the bedside nurses as having delirium, 83% were correctly classified (positive predictive value). However, of the 134 residents evaluated by the bedside nurses as not having delirium, only 29.9% were correctly classified (negative predictive value), suggesting that the absence of a delirium rating by the bedside nurse does not exclude delirium in this specific population.

At T2, RA-1 found evidence of delirium in 107 of the 151 residents (70.9%), thereby indicating practically no variation in the prevalence of delirium between T1 and T2. The comparison of bedside nurse and RA-1 delirium ratings at T2 yielded similar results to those observed at T1. Bedside nurses had difficulty detecting delirium when it was actually present (sensitivity of 18.7%). However, the majority (76.9%) of cases identified by the bedside nurses as having delirium were correctly classified (positive predictive value). Bedside nurses detected most of the cases (86.4%) without delirium but of the 125 cases they identified as not having delirium, only 30.4% of them were correctly classified (negative predictive value).

Table 5 presents a comparison of the bedside nurse and RA-1 ratings of 6 delirium symptoms (MDS-2 items): easily distracted; periods of altered perception; disorganized speech; periods of restlessness; periods of lethargy; and mental function varies over the course of the day. Sensitivity, specificity and confidence intervals are presented for each symptom. In total, 250 paired observations were made: 125 at T1 and 125 at T2. As previously mentioned, 31 residents were excluded from the analyses because of missing values.

At T1, sensitivities ranged from 32.4% to 58.1%, indicating an overall under-recognition of delirium symptoms by bedside nurses compared to the RA-1 ratings for these same symptoms. Nevertheless, 3 symptoms were detected more easily by the bedside nurses: disorganized speech (58.1%), easily distracted (53.4%) and mental function varies during the course of the day (45.8%). The following symptoms were more difficult for the bedside nurses to detect: periods of altered perception (39.1%), periods of lethargy (34.9%) and periods of restlessness (32.4%). Specificities ranged from 57.1% to 82.3%. Periods of restlessness, periods of altered perception and periods of lethargy were the symptoms with the higher specificities (82.3%, 75.8% and 73.5% respectively). These relatively high specificities indicate that these features were rarely identified by nurses when said features were indeed absent. Lower specificities were observed for disorganized speech (69.2%), easily distracted (66.7%) and mental function varies over the course of the day (57.1%) indicating that these features were more often identified as present (compared to the other features) when in reality, they were absent.

At T2, sensitivities and specificities ranged from 28.4% to 62.4% and 62.5% to 82.6% respectively. Despite some differences in the sensitivities and specificities recorded for T2 compared to those of T1, the interpretation of the results remains the same as that described for T1.

Tables 6, 7, and 8 present the results of univariate analyses linking several potential risk factors to the delirium that was undetected by bedside nurses at T1. From among the 18 variables considered, only age was found to be associated with undetected delirium: residents over 85 years old were at greater risk for being undetected (OR: 4.1; 90% CI: [1.5–11.0]).

The first objective of this study was to determine the rates of delirium detection by bedside nurses. The results have confirmed the importance of the under-recognition of delirium by bedside nurses among this very frail population. Despite the high prevalence of delirium in this cohort (delirium of 71.5% at T1 and 70.9% at T2), nurses were able to identify delirium in only a minority of cases (13% at T1 and 18.7% at T2). However, where nurses did identify delirium, in the majority of cases (93% at T1 and 86% at T2), they were correct. These results underscore how difficult it is for nurses to recognize delirium among this specific population, based solely on their own observations and clinical judgement.

Only one study looking at detection by nurses of delirium among elderly LTC residents was identified in the literature. Culp et al.19 studied the detection rate of delirium by nurses among LTC residents (n = 37). Contrary to that of the present study, only10% of their cohort was made up of demented residents. Nevertheless, nurses in their study were found to have detected only 26.7% of delirium cases. This relatively higher detection rate compared to that observed in the present study is not surprising since, due to overlapping of some symptoms, delirium superimposed on dementia can make delirium detection more difficult for untrained professionals 5.

Other studies on nurse-detection of delirium among older adults have been carried in hospital settings. Inouye et al. 20 studied nurses' ability to recognize delirium among 797 (>70 years old) hospitalized patients (medical and surgical units). They found an overall sensitivity of 19.3% and specificity of 95.8%. These results are very close to those of the present study. It should be noted that while their study population was not composed exclusively of older patients with dementia, the degree of cognitive impairment in their cohort was substantial (40% of patients having a score of less than 24 on the Mini Mental State Examination) 20. The reference standard used in Inouye's study was the CAM but the specific algorithm used was not specified (definite vs. probable).

A study by Lemiengre et al. 3 yielded higher sensitivities for detection of delirium by bedside nurses than the ones obtained in the current study. Depending on the method used to measure delirium, different sensitivities were reported: 23.8% for definite delirium and 66.7% for probable delirium. In our study, sensitivities of 13% and 18.7% were obtained with a definition that includes both definite and probable delirium. The higher sensitivities observed in the Lemiengre et al. study compared to those of the present study can probably be explained by differences in the population under study, as well as differences in methodology. For example, contrary to the present study where all participants were demented, only 12% of their study population had a diagnosis of dementia. The greater proportion of dementia in the current study may have complicated the detection of delirium by nurses. As well, the method used by the bedside nurses to measure the presence of delirium differed in the two studies. In the present study, bedside nurses had to respond to only one question seeking to determine if a resident had delirium or not. In Lemiengre's study, bedside nurses had to fill in a form using the CAM algorithm where each delirium feature was illustrated with practical examples. More importantly, the bedside nurses in Lemiengre's study were invited to an hour long information session on delirium and its detection and posters with information about the use of the CAM algorithm were placed in the nursing stations as well. It is quite likely that this intervention had a positive impact on the nurses' ability to detect delirium.

The second objective of the present study was to identify the delirium symptoms that presented the biggest challenge for nurses to detect. Interestingly, despite an overall under-recognition of individual symptoms of delirium, nurses were better at identifying individual features of delirium than delirium itself, as the higher sensitivities indicate. Disorganized speech (T1:58.1% and T2:57.1%), easily distracted (T1:53.4% and T2:62.4%) and mental function varies over the course of the day (T1:45.8% and T2:49.6%) were more readily recognized by nurses than periods of altered perception (T1:39.1% and T2:42.9%), periods of restlessness (T1:32.4% and T2:28.4%) and periods of lethargy (T1:34.9% T2:42.1%). These results are somewhat similar to those obtained in hospital settings. For example, Morency et al. 21 studied nurses' ability to recognize symptoms of delirium among older patients in acute care settings. They found that perceptual disturbances were detected by bedside nurses in only 41% of the cases, restlessness in 38% and lethargy in 39%. Another study conducted by Lemiengre and colleagues 3 in an acute geriatric ward revealed that it was difficult for nurses to detect fluctuation of symptoms (sensitivity of 45%), inattention (sensitivity of 59%) and disorganized thinking (sensitivity of 61%). Detection rates of delirium symptoms in the Inouye et al. study 20 were relatively lower than either those obtained in the current study or those reported in both Morency's and Lemiengre's studies 321. In fact, they found that nurses were able to recognize inattention in only 15% of the cases and disorganized thinking in only a quarter of the cases. However, the authors did mention that nursing turnover was high in their clinical setting and that the nurses' experience with older patients varied widely. It is likely that these factors explain the sensitivities they observed.

The results of the current study indicate that recognition of delirium and the symptoms of delirium among older LTC residents with dementia is a real challenge for nurses. Interestingly, the rate of detection among this very frail population was generally quite similar to that observed in hospital settings. Three of the key features of delirium, inattention, disorganized thinking and fluctuation of symptoms were more easily detected by nurses. The fact that these results were not reflected by a higher detection rate of delirium itself, suggests that nurses may not have the necessary knowledge about the diagnostic criteria required to define a resident as being delirious. Not only do nurses need to improve their ability to recognize the symptoms of delirium, they also need to develop an awareness of the link between symptoms and the presence of delirium.

The last objective of this study was to identify factors linked to undetected delirium by nurses. From among all the variables considered, only the age of residents was found to be associated; more precisely it was found that delirious residents 85 years of age and over, were 4 times more likely to be undetected than were younger residents.

One possible explanation may be related to the nurses' perception of ageing. In that sense, McCarthy et al. 52 have found that detection of delirium was influenced by the attitude of nurses toward health in ageing. Nurses who regard health in ageing as negative were found to be less likely to detect delirium. However, this assumption still needs to be tested in this specific population.

As far as the present authors are aware, only one study has identified the risk factors for undetected delirium by nurses. Inouye et al. 20 found that an older age (≥ 80 years old) (adjusted odds ratio-OR: 2.8; 95% confidence interval-CI: [1.7–4.7]), the presence of the hypoactive form of delirium (adjusted OR: 7.4; 95% CI: [4.2–12.9]), vision impairment (adjusted OR: 2.2; 95% CI: [1.2–4.0]) and the presence of dementia (adjusted OR: 2.1; 95% CI: [1.2–3.7]) were independently associated with undetected delirium. The current study corroborated the finding that age was an associated factor of undetected delirium.

The strengths of the present study include the quality of the reference standard ratings of delirium: 1) the RA-1 had received intensive training on delirium and its detection by the principal investigator from the study; 2) the presence of delirium was measured using an instrument recognised for its psychometric properties (CAM); 3) ratings of delirium were based on a seven-hour observation period; 4) information was collected in a consistent manner and the involvement of two RAs minimized the possibility of introducing an information bias in the measures of detection of delirium by bedside nurses. As well, 5) it is important to mention that detection rates of delirium were measured twice with a one week time lapse. Conclusions were the same for the two different time measurements. These findings support the internal validity of the study.

This study has several limitations that should be noted. Although the prevalence of delirium in the study population (n = 109) was high, the number of residents with delirium may still have been insufficient to detect a statistically significant association between the variables considered and undetected delirium. Therefore, the lack of significant association in the present study may be an indicator of a lack of power and should not exclude the possibility of these factors playing a role. Another limitation relates to the sampling method utilized. When compared to randomized sampling, convenience sampling, although very practical, may lead to biased results. On the other hand, the design effect was computed and our study found to perform as a simple random designed study. A final limitation concerns the decision to define delirium as meeting the CAM algorithms for definite and probable delirium based on the difficulty of obtaining a valid assessment of the criterion involving the acute onset of symptoms. By so doing, it is possible that some cases of delirium were in fact a type of dementia, e.g. dementia with Lewy bodies is known for its fluctuation feature which can convey a delirium-like picture. Although this information bias may have inflated the prevalence rate of delirium in our study population, it could not have influenced the overall results of this study, especially with regard to nurses' difficulty in recognising the features of delirium.

Given the high prevalence of delirium in LTC settings and the results observed in this study, it is urgent that nurses, who play a key role in the care of these residents, be knowledgeable about delirium. Nursing education, continuing education and new detection tools tailored to demented LTC residents are certainly ways to address the challenge.

Given the aging population, it is critical for newly registered nurses to have both the necessary knowledge and skills to care for elderly adults. However, recent surveys carried out in the U.S and Canada 5354 have provided evidence of the inadequate gerontological and geriatric content in undergraduate nursing programs for preparing nurses to adequately care for this population. Integrating more geriatric content into the nursing curriculum as well as increasing the number of clinical hours spent on the nursing care of older adults was identified as possible strategies 5355. The necessity for nursing schools to commit to the provision of greater gerontology education, as well as the need to increase the number of faculty staff members with expertise in geriatric care, were also identified as key factors in better preparing the next generation of nurses 54.

Continuing education is another important means for improving nurses' knowledge about delirium. Several studies using both formal and informal teaching sessions have yielded encouraging results 5657. Interventions exploiting the concept of mentoring have gained increasing importance in assisting nurses to develop best practice care for older patients and this approach has yielded promising results 58. Based on the findings obtained in the present study, educational programs on delirium should emphasize recognition of those primary features of delirium concerning which nurses showed significant shortcomings (i.e. periods of altered perception, periods of restlessness, and periods of lethargy). In addition, nurse awareness about the link between the presence of symptoms and the diagnosis of delirium also has to be increased.

A further key element in the improvement of delirium detection is the use of screening tools to detect delirium 59. In addition to being valid and reliable, the screening tool should be easy to administer, impose minimal discomfort in the residents, be integrated into clinical routine and tailored to a demented population. Such instruments now exist for palliative care 60 and the intensive care unit 61. Future research is needed to develop and validate an instrument with these characteristics among demented residents in LTC settings.