Validating the Diagnosis of Delirium and Evaluating Its Association With Deterioration Over a One-Year Period

The findings reported here were derived from a longitudinal study of delirium and related symptoms in a residential care population. Descriptions of methods and other findings have been reported by the authors here (42) as well as elsewhere (14, 15). Consenting subjects who met the criteria for inclusion in the study were followed over 1 year, receiving assessments at baseline and at 4-, 8-, and 12-month follow-up sessions, with additional assessments during hospitalizations and after discharge. Baseline and follow-up assessments were identical, including tests of cognitive, physical, and affective functioning. In addition to assessments given at 4-month intervals, the Mini-Mental State Exam (MMSE) was repeated the second and third week after the baseline and follow-up assessments. Those who were hospitalized at some point during this study received clinical evaluations to establish the presence or absence of definite or possible DSM-III-R delirium within 48 hours of admission, at Day 6, and every 4–5 days thereafter, as previously described. Note that DSM-III-R criteria were utilized because they were the best validated criteria available at the time that this study was designed. Also, patients received research assessments with full evaluations (including the MMSE, Buschke Selective Reminding Tests, Stroop Test, Verbal Vigilance, and electroencephalogram [EEG]) or brief evaluations (including the MMSE and measures of accessibility) given on alternate days. They also received full evaluations 2, 4, and 8 weeks after discharge from the hospital. For those who were hospitalized toward the end of the planned 1-year period of observation, additional assessments were conducted, extending the period of observation until at least 2 months after hospital discharge.

The resident population consisted of those living in a nursing home and a high-rise congregate apartment complex. Medical records were reviewed to restrict the subjects considered for participation to a group of elderly persons who did not demonstrate severe cognitive impairment and were medically stable. Subsequently, they were evaluated for inclusion/exclusion criteria. Subjects were eligible if they spoke English and had no severe visual or hearing impairment. They had to have a score of less than 17 on the Blessed Memory-Information-Concentration Test; no acute medical illness, current psychosis, or major depression; no medication change in the 2 weeks before the initial assessment; and no hospitalizations in the last month. After accounting for these criteria, 102 subjects were included in this study; 51 were from a nursing home and 51 from a congregate housing facility. To be included in the analyses of change over time reported here, subjects needed to have been assessed at two or more time points at least 3 months apart; six subjects who did not meet these criteria were omitted from these analyses, leaving 96 subjects. During the course of the year, 47 of the 96 subjects (49%) were hospitalized.

Physical health was evaluated at admission to the study by the subjects’ physicians, using the Cumulative Illness Rating Scale (CIRS) (16) to assess the severity of physical health problems in 13 (non-psychiatric) systems; a comorbidity index was computed by counting the number with moderate-to-severe pathology. The CIRS has been validated as an indicator of health status among medical and psychiatric populations (17, 18) and, more specifically, for use with this population of frail, institutionalized older adults (19).

The MMSE (20) was used as a measure of global cognitive performance. Physical functioning was measured with the Physical Self Maintenance Scale (PSMS) (21), using data from the subjects’ self-report or staff reports. The eight items were scored on a three-point scale, totaling between 8 and 24, with higher scores indicative of greater impairment. Also, a series of specific neuropsychological tests was administered. The Stroop Test (22–27) is a measure of cognitive inhibition, flexibility, and mental control. Scores included the number of errors in the final condition with incongruent colors and words, and the interference score. The Buschke Selective Reminding Test (SRT) (28–33) assesses verbal list learning and memory. The number of words the subject recalled in six trials is added to obtain a measure of Total Recall. Verbal Vigilance is a measure of auditory vigilance and sustained concentration. The double-letter task requires the subject to respond to a target letter only when it follows a second specified letter. Scoring counted the total correct out of 10 trials.

EEG measures. A portable, microprocessor-based, two-channel EEG device was used to measure EEG (Neurotrac, Interspec, Inc.; Conshohocken, PA). This machine provides output in the form of fast-Fourier–analyzed EEG. The method used identified data contaminated by muscle movement and electromyogram activity, the alpha squeak phenomenon, and drowsiness. Recordings were obtained from parieto-occipital leads (P3–O1, P4–O2), because background slowing, primarily in posterior leads, is a classic hallmark of delirium (34). Subjects were asked to close and open their eyes 12 times. Data were taken from the first two 2-second epochs of artifact-free EEG after eyes were closed, totaling 20 epochs or 40 seconds. Data are reported in terms of percent power in standard delta (0–4 Hz), theta (4–8 Hz), and alpha (8–13 Hz) EEG bands. Measurements were conducted by research nurses who were trained in techniques by an EEG technician. The entire procedure took 15 to 20 minutes. This method of obtaining EEG data was previously shown to be reliable and appears to be a valid way of detecting changes in cerebral state in elderly subjects (14).

Repeated measures. In light of repeated testing, 12 different variations of the MMSE, Stroop Task, Verbal Vigilance, and Buschke SRT were given to limit potential learning effects. The 12 forms of the MMSE varied according to the list of three words used to test short-term memory, the word used to test attention, and the verbal three-step command. The 12 forms of the Stroop varied the order of stimulus presentation; the Buschke SRT provides 12 different word lists; and Verbal Vigilance varied in the target letters and order of presentation. Each subject entered in the study was given different forms initially and was rotated through the 12 forms in sequential order, starting over when given all 12 forms. In all, 1,197 MMSE measurements were obtained on 96 subjects, 486–560 other cognitive measures on 88–96 subjects; and 491 EEG measures on 90 subjects. For in-hospital data, there were 162 MMSE measures on 46 subjects; 72–76 other cognitive measures on 36–38 subjects; and 77 EEG measures on 38 subjects.

Other assessments. Other measures included self-ratings of depression conducted throughout the period of observation, and nurses’ ratings of behavioral disturbances conducted during hospitalizations. Findings will be reported elsewhere.

Several different types of analyses were performed to evaluate changes over time for each cognitive, EEG, clinical, and functional status measure over 1 year and to make comparisons of these changes with respect to hospitalization, delirium, and baseline cognitive status. First, we assessed changes across time outside of the hospital and compared these changes between hospitalized and non-hospitalized subjects and between subjects with and without delirium. Second, we evaluated changes across time during hospitalization and non-hospitalization periods, making comparisons between these two contexts and between delirium and non-delirium patients. Third, we evaluated how changes across time differed between cognitively impaired and intact patients as ascertained at baseline.

These three types of analyses were performed with mixed-effects linear regression models (35) consisting of random intercepts and slopes across time for each individual. This model is an appropriate method to analyze repeated measures over time in studies such as this, where there are multiple repeated measures for the same subjects and when there are unequal numbers of assessments across subjects depending on whether or not they were hospitalized, and the extent to which there are missing data (36, 37). Limitations of this model include effects of certain types of non-random attrition, which are assessed, and insensitivity to nonlinear change. The models used for these analyses are presented in Appendix 1.

We supplemented these mixed model-based analyses with analogous mixed models adjusting for baseline age and CIRS. These baseline variables did not differ significantly statistically or clinically between subjects with and without delirium at baseline. The respective means and standard deviations were 84.5±6.4 and 86.1±5.9 for age and 1.70±0.32 and 1.72±0.63 for CIRS. Consequently, their inclusion in the above models only marginally altered the presented estimates, standard errors, confidence intervals (CIs), and P values for delirium comparisons under the models without these covariates. Similar robustness to adjustments for age and CIRS were observed for comparisons between hospitalized and non-hospitalized status.

For each dependent variable (cognitive, clinical, EEG, and functional measures), we obtained an estimate of each of the above effects (e.g., difference in mean at baseline, difference in slopes, significance of a slope) accompanied with 95% confidence intervals, t-statistics, and corresponding P values. All t-tests reported in the article for slopes and intercepts are based on the mixed effects models; t-tests comparing delirium and non-delirium patients may lack power because of the small number of delirium patients (12). However, basing these t-tests on the mixed models that incorporate all data collected across time for each subject improves the power, as the effective sample size is greater than the number of subjects but less than the total number of data-points collected for an outcome.

To assess the sensitivity of our results to non-random attrition related to the missing data (i.e., informative drop-out), we reanalyzed the data with the above models augmented with additional effects for linear and quadratic time to drop-out. Such models represent pattern-mixture models that serve as approximations to selection models (38). The resulting estimates were very similar to those obtained under the previous models, suggesting that the inference in this article is not affected significantly by non-random attrition.

We also considered the extent of intraindividual variability in study measures by use of measures described in Appendix 1. In order to determine whether baseline global cognitive status predicted variability in the above measures, subjects who were considered cognitively intact (MMSE≥24) were compared, by means of t-tests with respect to the variability measures, with the cognitively impaired subjects (<24) at baseline. Also, the relationship between baseline medical status (as measured by CIRS) and each variability measure was estimated with a multiple-regression analysis, with the variability measure as the dependent variable. The significance of this relationship was determined after adjusting for baseline age, gender, and education. Subjects who were hospitalized during the study were compared with those who were not with respect to variability scores for all measures described above by including the covariate for hospital status in the multiple-regression models outlined above. Also, hospitalized subjects who were diagnosed with delirium and those who were not were compared with respect to variability of each of the measures, including EEG measures, by including a new covariate in each of the above models. Subjects were included in some comparison analyses but not others because they completed some measures, such as the MMSE, but were unable to complete other measures over the year for multiple reasons.

The changes exhibited by study subjects during the course of the year of follow-up can be characterized with the mixed-model slopes calculated from all available data obtained outside of the hospital (Table 2). For the sample as a whole, there were significant declines during the course of the year in functional status (PSMS scores) and in EEG measures (increases in % delta and % theta, decreases in % alpha power). There were increases over the year in correct responses on the Stroop test that can be attributed to practice or learning effects; these will not be discussed further. In subgroup analyses, the PSMS and EEG slopes were significantly different from zero in the patients who were hospitalized during the year, but not in those who were not. No other changes were significant in either the hospitalized or non-hospitalized subgroups. Among those who were hospitalized, changes in PSMS were significant both in those patients who were diagnosed with delirium and in those who were not. EEG slopes were significantly different from zero only in those patients who were found to show delirium during their hospitalization. In addition to those measures that exhibited changes over time in the entire study sample, MMSE variables exhibited significant changes over time, specifically in the subgroup of patients who were diagnosed with delirium when they were in the hospital.

However, these analyses reported on whether the mixed-model regression slopes were significantly different from zero in relevant subgroups. In interpreting the differences between the hospitalized and non-hospitalized patients or those between the delirium and nondelirium patients, it is also important to test for significant differences in slopes between subgroups. There were no significant differences between those subjects who were hospitalized and those who were not. Among those who were hospitalized, there was a significant difference in slopes on the MMSE between those with and without delirium (t=2.49; P=0.013), but no differences in the PSMS or in the EEG measures. There were significant differences between hospitalized and non-hospitalized subjects in their variability over time outside the hospital in the PSMS and in EEG measures (% theta, % alpha). Among those who were hospitalized, there were differences in variability between those with or without diagnoses of delirium in the PSMS but not in other measures (Table 3).

When subjects classified as cognitively intact on the basis of baseline MMSE scores were compared with those who were impaired, there were no significant differences in the proportion of subjects who were hospitalized or in the proportion of those hospitalized who were diagnosed with delirium. Those with cognitive impairment at baseline exhibited greater decline over the course of the year in the measure of verbal vigilance (t=3.76; P=0.0003) but not in other cognitive, clinical, or EEG measures. They exhibited greater variability over time in MMSE measures (t=4.20; P=0.001), interference on the Stroop test (t=2.64; P=0.01), and verbal vigilance (t=4.97; P=0.0001). There was a significant relationship between baseline CIRS scores and variability in MMSE scores (R²_change=0.0416; P=0.035) even after controlling for baseline MMSE, age, gender, and education; no other associations with measures of variability were significant.

Data acquired in the hospital were merged with the previously presented non-hospital data in a series of analyses that allowed comparison of changes in slopes and intercepts for data acquired during hospitalization with other data acquired while patients were in the long-term care setting. The slope over time for the hospitalization data did not differ significantly from that obtained outside of the hospital for any of the variables considered. The decline for hospitalization data for those diagnosed with delirium was steeper than that for all non-delirium patients for the MMSE (−4.261; 95% CI: −8.346 — −0.177 vs. −0.130; 95% CI: −0.781 — 0.520; P=0.048), but not for other variables.

The zero time intercept for hospital data differed from that for non-hospital data in measures of verbal memory, verbal vigilance, and in EEG measures (% delta, % theta, and % alpha; Table 4). The intercept for hospital data in patients diagnosed with delirium differed from that for others only for the EEG measures.

Another series of analyses compared hospital and non-hospital data after controlling for the differences that could be attributed to diagnosed delirium. Analyses of slopes demonstrated that differences for the MMSE scores remained significant (difference= −1.718; 95% CI: −3.357 — −0.079; P=0.040) after controlling for the effects of delirium. Evaluations of intercepts demonstrated that differences for verbal memory and verbal vigilance remained significant, but the EEG effects were no longer present.

In this sample, patients with a diagnosis of delirium differed from the remainder of the hospitalized patients in their pre-hospital baseline performance on the Stroop test, but not in their MMSE. This study did not replicate findings from previous research demonstrating that preexisting cognitive impairment was a risk factor for delirium. This result may have followed from our exclusion of patients with severe dementia from the study and from the diagnostic heterogeneity of those with cognitive impairment. Alternatively, it may reflect the low power of this study as a result of both the limited overall sample size and the small number of patients diagnosed with delirium. The findings on the Stroop test suggest that specific executive-function deficits may be a risk factor for delirium. Although this could be of conceptual importance, the relevant analyses must be considered exploratory, and the finding must be viewed with caution.

Research data acquired during hospitalizations on patients with delirium demonstrate significant differences in the MMSE slope and intercepts for EEG measures. Patients with delirium experienced greater decline in global cognition, on average, during their hospitalization. The differences in intercepts for the EEG variables indicate increased levels of delta and theta activity and decreased alpha activity in patients who were diagnosed with delirium during hospitalization. Given that our design ensured the independence of clinical and research assessments, these findings confirm current understanding of delirium as a disorder associated with global deficits in cognitive performance as well as its association with background slowing on the EEG.

To evaluate the extent to which clinical features of delirium may have occurred in patients who did not receive clinical diagnoses, we conducted a series of analyses that tested for differences between hospital and non-hospital data after controlling for the differences that were associated with the diagnosis of delirium. The EEG changes associated with hospitalization are accounted for by those patients who received clinical diagnoses. However, findings from the regression analyses demonstrate differences in MMSE slope and in intercepts for measures of verbal memory and verbal vigilance that remained significant after controlling for diagnoses of delirium. These findings suggest that although the clinical strategy used in the study for the identification of delirium was validated in terms of its association with increased mortality, it did not identify all of the individuals who experienced cognitive changes associated with hospitalization. The observation of MMSE, verbal memory, and attentional deficits that were not associated with clinical diagnoses could reflect either transient cases of delirium that appeared and resolved between clinical assessments or the existence of subclinical cases not captured by DSM-III-R diagnostic criteria. Although our clinical assessments did not capture all of the cognitive decrements associated with hospitalization, these findings demonstrate that frequent screening for the elements of delirium should identify such changes.

According to the research design, clinical evaluations were conducted to probe for the diagnosis of delirium only when patients were in the hospital. We recognized that delirium could also occur in the long-term care settings and that prevalence estimates of 6%–7% have been reported in the nursing home (39). It was our assumption that the extensive clinical services available in the facility that housed these studies would be adequate to identify these cases and that they would be captured in our design because they would, in general, lead to hospitalization. However, we have no tests for the validity of this assumption. In fact, findings on the variability of MMSE and EEG measures (see below) suggest that ephemeral, unrecognized episodes of delirium may have occurred in the long-term care facility as well as in the hospital.

The sample, as a whole, deteriorated over the course of the year in self-care ability as measured with the PSMS. Interestingly, these changes were significant only for those who were hospitalized during the year, whether or not they were diagnosed with delirium. This finding is consistent with a growing body of data suggesting that an identifiable subgroup of older patients are at risk for deterioration after an acute hospitalization (40, 41) and demonstrates that a major component of the functional decline that occurs in frail elderly patients is associated with significant medical events that require hospital admission. It suggests that increasing the intensity of medical care of older patients with chronic illness might serve a role in preventing deterioration.

No significant decrements in cognitive performance were found over the course of this study in this study sample as a whole. Although cognitive deterioration is a characteristic of Alzheimer’s disease, the analyses conducted identified greater decline in those patients who were classified as cognitively impaired at baseline only in tests of vigilance. In interpreting these findings, it is important to recall that each of the regression analyses controlled for baseline measures on the dependent variable. Therefore, the lack of effects may have reflected the nature of the analyses conducted as well as the heterogeneity of the subjects with baseline impairment, limitations in statistical power, or learning and practice effects related to the multiple repeated administrations of the test instruments. We did, however, observe significant within-subject decrements over the course of the year in MMSE score for those patients who were both hospitalized and diagnosed with delirium. Moreover, among those who were hospitalized, there were significant differences in the MMSE slopes for non-hospital data between those who experienced delirium and those who did not. These findings suggest that the combination of medical events and cognitive vulnerability that leads to delirium is a strong predictor of cognitive decline. Medical events and the vulnerability of the brain to the effects of medical illness, more than cognitive impairment itself, were associated with deterioration over time.

These conclusions are also consistent with the EEG findings reported here. The patient sample, as a whole, deteriorated over the course of the year, with increases in delta and theta and decreases in alpha power. This effect, however, could be attributed to significant declines only in those patients who were both hospitalized and diagnosed with delirium. Taken together, these data demonstrate that patients who experience delirium are at specific risk for declines in cognitive performance and cerebral activity during the course of a 1-year period. They cannot, however, distinguish between models in which the deterioration is a sequel or consequence of the delirium vs. those in which both delirium and deterioration occur because patients exhibit a combination of medical illness and a cerebral vulnerability to the physiological effects of illness. Even so, the EEG findings, like those on cognitive functioning, raise the possibility that medical care directed toward preventing acute illnesses might prevent deterioration in frail elderly patients.

The statistical models used in these analyses allow the quantification of within-subject variability as well as changes over time. Understanding the subjects’ variability in functional and cognitive measures could be clinically significant because fluctuations over time are characteristic of delirium and because such variability may imply that subjects are exhibiting excess disability some of the time. In fact, the observed within-subject variability in study measures may have a number of causes. Some of the associations may reflect limitations in the linear model used to characterize change over time. Others, such as the association of global cognitive impairment with increased variability in specific cognitive measures, could reflect limitations in the reliability of the instruments in specific populations. However, the findings on the association of hospitalization with greater variability in self-care during course of the year once more supports the importance of medical illness as a determinant of functional status in elderly persons. The associations of variability in MMSE scores with baseline measures of medical illness and those of variability in EEG measures with hospitalization suggest that medical vulnerability may be associated with variability in cerebral activity. These findings lead to the hypothesis that cerebral activity in frail older individuals may be sensitive to day-to-day variations in medical status.

To summarize, the findings from this prospective cohort study add to the existing body of evidence that confirms the validity of the clinical criteria used to diagnose delirium. They demonstrate that significant medical events leading to hospitalization are important markers for functional decline in elderly patients. Moreover, they demonstrate that either the occurrence of delirium or the combination of general medical and cerebral vulnerability that predispose individuals to it are important markers for global cognitive decline. Our findings demonstrate that a clinical diagnosis of delirium during a hospitalization has long-term significance as an index of frailty and a marker for a high risk for deterioration. Furthermore, several of our findings suggest that delirium can exist along a continuum in terms of its severity and persistence. One type of milder or more time-limited delirium may have been demonstrated through the finding that hospitalization was accompanied by cognitive decrements that could not be attributed to clinically diagnosed episodes of delirium. The existence of another type was suggested by the findings that subjects with greater medical illness burden at baseline and those who required hospitalization exhibited greater variability in (non-hospital) MMSE scores and EEG measures, respectively. Thus, for frail elderly patients, delirium appears to represent a spectrum of conditions ranging from those that are highly significant complications of acute illnesses to those that are part of the psychopathology of everyday life.