Delirium is an important problem in hospitalized cancer patients. The objective of this study was to determine whether exposure to corticosteroids, benzodiazepines, or opioids predicted delirium.
Delirium is an important problem in hospitalized cancer patients. The objective of this study was to determine whether exposure to corticosteroids, benzodiazepines, or opioids predicted delirium.
A prospective cohort study was conducted in an oncology/internal medicine population. Patients were assessed continuously for the presence of delirium until they were discharged by using the Nursing Delirium Screening Scale (Nu-DESC). Follow-up for outcome began after incident delirium. The primary outcome was the presence of a delirium event, which was defined as a Nu-DESC score >1. Strengths of associations of medications with delirium were expressed as odds ratios (ORs) in univariate and multivariate analyses.
In total, 114 patients (1823 patient-days) met the inclusion criteria for the study. The mean follow-up from incident delirium was 16 days. The mean number of delirium events by patient was 6 (total number, 667 delirium events). Analysis by day on several occasions revealed significant associations between opioids and delirium. Corticosteroids and benzodiazepines were not associated significantly with an increased risk of delirium on any given day. Analysis by patient using generalized estimating equation (GEE) models showed an increased risk of delirium on any day of follow-up associated with opioid exposure in univariate analysis (OR of 1.70; P < .0001). The association remained significant after adjustment for corticosteroid, benzodiazepine, and antipsychotic exposure using GEE regressions (OR of 1.37; P = .0033). Truncating follow-up at 30 days did not affect the results (OR of 1.38; P < .032).
Exposure to opioids during hospitalization was associated significantly with an increased longitudinal risk of delirium. Cancer 2007. © 2007 American Cancer Society.
Delirium is a frequent psychiatric condition in hospitalized patients, especially in oncology populations.1, 2 One possible explanation is the high exposure rate of such patients to drugs prone to induce delirium. Contrary to common clinical wisdom, however, very few psychoactive medications were identified repeatedly as delirium risk factors in large-scale studies.3 The quality of research in drug-induced delirium has been questionable methodologically in the past3; however, the results from a recent and rigorously conducted study suggest that psychoactive medications, such as benzodiazepines, corticosteroids, and opioids, increase the risk of developing delirium during hospitalization in oncology patients.2 Clearly, determining that such drugs are associated with delirium onset is an important step toward better patient management. The longitudinal analysis of factors associated with delirium is particularly valuable from the standpoint of evidence-based medicine. Indeed, these data may be used not only to lend further support to the causal relation between a given type of drug and delirium but also to help optimize the prevention and/or treatment of delirium on a day-to-day basis by controlling exposure to these medications over time.
In psychiatric research, taking depression studies as an example,4 the patient's status (affliction, reversal, exacerbation, etc) is difficult to determine with precision. This is mainly because of the lack of laboratory tests or histologic analyses that may be monitored over time. In delirium research, an effective way to monitor delirium status over time is using a continuously rated instrument, such as the Nursing Delirium Screening Scale (Nu-DESC).5 Patients do not necessarily face the same risk of delirium after a first occurrence, and those patients should not be included in the analyses as new patients. Two approaches can be used to take this into account. The first approach is to wait until the patient is free of clinical manifestations. However, this approach raises important issues, especially regarding the duration of remission. It is well recognized that delirium fluctuates over time, such that there is currently no evidence to support the notion of cure or reversal after a single day of symptomatic remission: Delirium still may be present, albeit in a subclinical form. Moreover, there are no guidelines that define duration of remission such that, after a given period, the patient can be considered at risk for a new incident delirium. The second approach involves taking into account all observations for all patients as of the first manifestations of delirium and taking into consideration that these observations are nonindependent. In our opinion, this is the most appropriate approach.
Few studies have examined the course of delirium, and most of those studies focused on the associated mortality risk during hospitalization.6, 7 However, 2 pioneer studies on the course of delirium in association with exposure to psychoactive medications have been conducted in recent years.8, 9 Using the first approach, 1 study in patients with advanced cancer examined the relation between delirium reversibility and various risk factors, including psychoactive medications.9 In multivariate analyses, psychoactive medications were associated significantly with reversibility. However, the strength of the evidence was impaired by a number of shortcomings, such as including large numbers of prevalent cases of delirium, retrospectively determining causes of delirium based on chart review by unblinded investigators, considering multiple criteria for the involvement of psychoactive medications in delirium reversibility, not specifying clearly which medications known to cause delirium were studied, whether the variable was defined a priori, and defining the variable psychoactive medications as a heteroclite set comprising numerous medications (eg, opioids, benzodiazepines, antidepressants) as well as antipsychotics. The other study, which used the second approach, suggested a longitudinal association between anticholinergic medications and delirium severity.8 However, that study also had methodological shortcomings. The authors used a rather empirical method, based on clinical experience and on a correlative clinical index, to evaluate anticholinergic effect. Delirium assessments were performed irregularly and infrequently. The majority of patients who were included were prevalent cases of delirium. Patients with missing laboratory data were assumed to have normal values without evaluation for randomness. Notwithstanding their respective limitations, for several years, these 2 studies have been among the most helpful available references on the course of delirium in relation to exposure to psychoactive medications. From a statistical perspective, delirium is typically a recurrent event, given its fluctuating nature. Naive statistical techniques (eg, basic logistic regression analysis) ignore the existence of recurrent events or the fact that the recurrent events within patients are correlated. Conversely, longitudinal techniques (repeated-measures designs) analyze the whole pattern of recurrent events over time. Longitudinal studies provide a large number of data points, thus improving the statistical power of the analyses and attenuating the effects of collinearity among explanatory variables. In delirium research, naive techniques are commonplace. Longitudinal data analysis largely has been underused, especially regarding the study of delirium risk factors, such as psychoactive medications, because, to our knowledge, no study before ours has attempted to perform this type of analysis. In a prospective cohort study that involved innovative methodological characteristics, we endeavored to evaluate the relation between corticosteroids, benzodiazepines, and opioids and the longitudinal risk of delirium in hospitalized cancer patients.
The study population consisted of patients who were admitted to the hemato-oncology/internal medicine unit at the Hotel-Dieu de Quebec Hospital, Canada, from January 21, 2002 to September 10, 2005. Study inclusion criteria were as follows: 1) patients with a histologic diagnosis of cancer and 2) incident cases of delirium.
Consecutive incident cases of delirium from our previous work, which examined the role of psychoactive medications in delirium occurrence2 (n = 43 patients), were included in the current study along with the enrollment of 71 new incident delirium cases. The 2 phases of enrollment were conducted in the same manner. In total, sample size of 114 incident delirium cases enabled this study to have a power of 80% to detect a 15% difference in the odds of delirium between groups. Written consent was not obtained, because there was no perceived risk of harm to patients. The hospital's Research Ethics Committee approved this study.
The Nu-DESC is an observational delirium diagnostic instrument that was developed by our group and is performed on a 24-hour basis (once per 8-hour shift). It rates the presence and severity of delirium manifestations in 5 dimensions: orientation, behavior, communication, perception, and psychomotor activity. Total scores range from 0 (no manifestations) to 10 (severe manifestations). A total score ≥2 at any time during the day defines delirium according to the Nu-DESC (a positive score on any of the 3 8-hour nursing shifts). This approach maximizes delirium detection (eg, instrument sensitivity) to levels never achieved before in the field and has even allowed reducing delays in delirium diagnosis by up to 5 days.10 However, for the purposes of the current study, this precluded us from demonstrating with confidence that patients truly were delirious over the whole 24-hour period when the presented a positive Nu-DESC score on a given day. For example, positive delirium evaluations on consecutive days of follow-up may have represented either a single episode or distinct episodes. From a longitudinal analyses standpoint, however, any delirium-positive day is considered a recurrence. Thus, any Nu-DESC score >1 on a given day of follow-up was defined as a delirium event. This term has been defined similarly and used in a recently published delirium study.11 These delirium events would represent the smallest analyzable delirium unit (eg, each Nu-DESC delirium score for each 24-hour period) and, conceptually, would be equivalent to the statistical term recurrences mentioned above. This approach allowed us to include all observations for all patients as of the first manifestations of delirium, like previous work in delirium course studies.8
The Nu-DESC can be used to assess delirium status rapidly on a continuous basis with a high level of sensitivity and diagnostic accuracy. Sensitivity, specificity, and efficiency of 85.7%, 86.8%, and 86.4%, respectively, were achieved based on the Nu-DESC validation study.5 The performance of the Nu-DESC was tested previously by using the Confusion Assessment Method (CAM)12 as the reference standard. The CAM was administered by study nurses who were blinded to the patients' Nu-DESC delirium status, and the results were confirmed by experienced clinical psychiatrists who also were blinded to the patients' delirium status. The Nu-DESC was determined to be congruent to the overall concept of delirium based on the fourth edition of the Diagnostic and Statistical Manual of Mental Disorders and was of comparable sensitivity and diagnostic accuracy.5 To our knowledge, no other delirium instrument presents such advantageous characteristics. Follow-up began after incident delirium (the first Nu-DESC score ≥2) and ended with the patient's discharge from the hospital. The presence of a delirium event (a subsequent Nu-DESC score ≥2) was the outcome variable of the study. In addition, the nurses who rated the Nu-DESC were blinded to patients' medication use and the study hypothesis.
In a previous study, we observed significant associations between incident delirium and cumulative daily equivalent doses of corticosteroids, benzodiazepines, and opioids >15 mg of dexamethasone, >2 mg of lorazepam, and >90 mg of morphine, respectively.2 The dichotomization of doses did not compromise the results, because the difference in the fit of models with medications as dichotomous versus continuous variables was not statistically significant2 based on the likelihood-ratio chi-square statistic.1. These thresholds of medication exposure also were used in the current study to examine the associated risk of delirium. We have accounted for antipsychotic exposure, because patients frequently were receiving such medication during follow-up for their delirious state. We have applied the required methodological precautions when conducting drug-induced delirium studies.3 Medication exposure for corticosteroids, benzodiazepines, opioids, and antipsychotics was computed as cumulative daily doses in oral lorazepam,14 oral dexamethasone,15 subcutaneous morphine,15, 16 and oral haloperidol17, respectively, based on the available literature. In calculating the equianalgesic dose of morphine, the potencies of the intravenous, subcutaneous, and intramuscular routes are considered equivalent.18 Exposure to antipsychotic medications was retained as a continuous variable.
Potentially confounding variables of an association between medications and delirium were chosen based on past work in the field and because of clinical relevance. The patients' clinical characteristics, sex, age, disease-related variables (primary cancer site, presence and sites of metastases, if applicable), diagnosis of delirium on a previous hospitalization, and dementia were determined at admission from medical records using standardized forms by a trained study nurse. Thus, the severity of illness as a confounding factor was taken into account indirectly, through the presence of metastases, because metastatic spread is an indicator of disease severity. In our setting, the type of pain experienced by the patients was recorded in nursing charts on a 24-hour basis. Uncontrolled pain was documented according to nursing or physician notes that the patient was complaining of pain or exhibiting pain spontaneously or upon movement. An experienced research nurse conducted a chart review of nursing and physician notes to document uncontrolled pain during the study period. Given the rehabilitative focus of our setting, few demented patients are admitted. Nevertheless, the patients' medical and nursing charts were inspected carefully for potential evidences of cognitive impairment and/or dementia. All available laboratory data were collected from medical charts and from computerized databases both at admission and later during follow-up. Along with laboratory data that were collected after admission, other time-dependent precipitants for delirium were documented based on chart review by a trained nurse. These precipitants were similar to those from previous studies and included episodes of infection, dehydration, or hypoxia.9 However, because of missing values and/or the limited quantity of these observations compared with medication data, we lacked the power to analyze these precipitants in a longitudinal fashion. Missing values were not substituted with normal values, because precipitants typically were not measured for those patients who seemed to be healthier. Because of the nonrandomness of missing data, values were not imputed, as this may have compromised the validity of the results.19 For admission variables, the categories used for the analyses were as follows. Ion levels (sodium, potassium, and calcium) were trichotomized (normal/low/high); whereas aspartate aminotransferase, alanine aminotransferase, urea, and creatinine levels were dichotomized (normal/high).
Preliminary analyses included descriptive statistics of medication exposure and baseline patient characteristics. The distribution of variables associated with occurrence of delirium was examined using the chi-square test. When discrete time points are considered, generalized estimating equations (GEE analysis)20, 21 is the recommended longitudinal approach.22 An analysis of the odds of daily occurrence of delirium during follow-up was conducted using a GEE model. The GENMOD procedure in the SAS software package was used.23 All statistical analyses were performed using SAS software (version 9.1; SAS Institute Inc., Cary, NC).
Baseline characteristics of the patients who met the eligibility criteria for study entry are presented in Table 1. Sixty patients had metastatic spread (52.6%). No cases of dementia were recorded. Fourteen percent of the study follow-ups have ended with the patient's death. These patients did not present more delirium events than the others nor were they followed for a longer period than the rest of the sample. Eight percent of the patients were transferred to another facility, whereas the remaining patients were discharged home (78%).
|Characteristic||All patients (%), N = 114|
|Mean age± SD, y||63± 13|
|Mean follow-up, d||16|
|Mean no. of delirium events||6|
|History of delirium||18 (16)|
|Primary cancer diagnosis|
|Gastrointestinal tract||11 (9.6)|
|Bones/soft tissue||12 (10.5)|
|Extent of disease|
|Locoregional only||54 (47.4)|
|Primary site of metastases|
|Low sodium||25 (22.9)|
|High sodium||3 (2.8)|
|Low potassium||18 (16.5)|
|High potassium||3 (2.8)|
|Low calcium||24 (38.7)|
|High calcium||24 (38.7)|
|High ALT||15 (18.3)|
|High AST||24 (28.9)|
|High urea||31 (27.9)|
|High creatinine||29 (25.7)|
The major categories of cancer pain based on etiology were as follows: 1) direct tumor involvement: somatic (73 patients), visceral (41 patients), neuropathic (27 patients); 2) therapeutic procedures (16 patients); and 3) side effects/toxicities (14 patients). Many patients had >1 category. To control pain symptomatology, a total of 3265 doses of opioids were received by the 114 patients. Most patients (n = 75) presented a regular pattern of use of opioids (1, 2, 3, or 4 times daily) supplemented by opioids as needed. Thirty patients received opioids only when needed. The most widely used administration route was subcutaneous (101 patients at some point during follow-up), followed by oral (101 patients), and transdermal (39 patients). Based on our chart review, none of the patients reported uncontrolled pain during the study period.
The risk of delirium was analyzed first for each day of follow-up. The results of these analyses are presented in Figure 1. There were 114 patients who were at risk of delirium on the first day of follow-up. The total number of patient-days at risk was 1823 days. The total number of delirium events was 667 events. After 30 days of follow-up, 79% of the 667 delirium events had occurred; and, after 60 days, 92% of the delirium events had occurred.
Odds ratios (ORs) representing the risk of delirium were computed for each day of follow-up. Twenty ORs were computed for corticosteroids, and 39 ORs were computed for benzodiazepines. Days with absence of medication exposure in cases or controls (ORs equal to 0 or ∞) were not considered further. There did not appear to be a significant relation between exposure to medication above the thresholds and delirium for corticosteroids and benzodiazepines: ORs were scattered above and below the null value (Fig. 1), and none of the ORs appeared to indicate an increased risk of delirium on any day of observation.
There appeared to be a graphic trend toward an association between exposure to doses >90 mg of morphine equivalents and the presence of a delirium event (Fig. 1). Statistically significant associations were noted on 3 occasions during follow-up: on Day 8 (OR of 4.8; 95% confidence interval [95% CI], 1.3–17), on Day 15 (OR of 8.1; 95%CI, 1.5–43.2), and on Day 28 (OR of 10; 95%CI, 1–97.5). Because the power of an analysis by day is somewhat limited, next, we used analyses by patient by combining the total number of daily observations (precipitants and delirium events) for the entire cohort.
The results of the analyses based on the 1823 patient-days at risk for delirium during follow-up are presented in Table 2. Of the 3 classes of medications studied, a significant chi-square test result was noted for opioids (P < .0001). Further examination of the nature of the association between this class of medications and delirium was performed using logistic regression models.
|Medication||Delirium event (n = 667)||No delirium (n = 1156)||P|
ORs that represented the risk of delirium associated with exposure to opioids were calculated. For the follow-up period, on any given day, the crude odds of experiencing a delirium event were 1.70 times greater for patients who were exposed to daily opioid doses >90 mg compared with patients who were exposed to lower doses (P < .0001) (Table 3). To control for a potential bias toward positive associations caused by repeated measures, we used GEE models to evaluate the daily risk of delirium over time for patients who were exposed to opioids. The risk of delirium remained significant (OR of 1.39; 95%CI, 1.04–1.85; P = .027).
|With repeated measures|
|First month of follow-up|
|With repeated measures|
We also examined the associations between baseline patient characteristics and delirium in univariate analyses (data not shown). These variables included sex, age, primary cancer site, presence and sites of metastases, diagnosis of delirium on a previous hospitalization, and laboratory data at admission. Consistent with our previous work,2 a history of delirium and the presence of liver metastases were variables associated with an increased risk of delirium. Next, several multivariate models were evaluated for goodness of fit to data. None of the baseline variables remained a significant predictor of delirium when they were fitted simultaneously into models that included opioid exposure or other time-dependent variables. Thus, in the best-fitting multivariate model, the risk of delirium associated with opioid exposure was adjusted only for the other 3 classes of medications studied (benzodiazepines, corticosteroids, and antipsychotics) and still was significant (OR of 1.37; 95%CI, 1.03–1.83; P = .033). Exposure to opioids was a significant predictor of delirium in all of the models in which this variable was studied.
Finally, to control further for repeated measures, this time for the potential bias of patients who had longer follow-up, analyses similar to those described above were performed for a shorter follow-up but in which 80% of all patient-days of observation still would be available for analysis (30 days). These results were comparable to those obtained by using the complete length of follow-up for all patients. The adjusted risk of delirium associated with opioids was 1.38 (95%CI, 1.03–1.85; P = .032) (Table 3).
Delirium frequently is seen in consultation-liaison psychiatry, particularly in oncology settings. Psychoactive medications, especially opioids, are among the most consistent risk factors for incident delirium.3, 24 We showed previously that opioids were involved in the development of delirium in hospitalized patients.2 The current study provides convincing evidence that exposure to the latter drug class also strongly influences the course of delirium. In fact, to our knowledge, this is the first study to suggest that the risk of presenting a delirium event is almost 40% greater for hospitalized patients who are exposed to relatively high opioids doses (>90 mg of daily morphine equivalents) compared with patients who are exposed to lower doses. After having presented an incident delirium, study patients were more likely to present a delirium event later if they still were exposed to high opioid doses. A typical study patient presented approximately 7 delirium events, suggesting that, of 16 days of follow-up, a typical patient was delirious at some point each day for roughly 7 days. This draws attention to the importance of controlling risk factors, especially medication, even after delirium appears to be reversed.
Our study combined several important methodological characteristics, including but not limited to 1) a large sample of delirious patients, 2) a longitudinal and dynamic design that was suited particularly well to the fluctuating nature of delirium, 3) continuous assessment of delirium status performed by bedside nurses who were blinded both to patients' medication use and to the study hypothesis, and 4) taking into account antipsychotic exposure when establishing the role of other psychoactive medications on the risk of delirium. Foremost, our statistical approach using GEE models allowed us to take into consideration the temporal order between delirium and the time-varying variables, the changes in these variables that occurred between each reassessment and the preceding one, and within-person correlations. Comparison with current data is difficult, because few studies have examined the longitudinal risk of delirium and its associated factors. In 1996, Inouye and Charpentier presented a complex model that featured 5 precipitating delirium risk factors, including psychoactive medications.25 Although the variable was defined somewhat imprecisely in that study,3 the results nevertheless lend support to the important role of psychoactive medications in precipitating delirium. The data presented herein are in agreement with results from a recent study, which suggested that controlling exposure to psychoactive medications can reverse delirium.9 Our data also suggest that delirium was preventable/reversable if exposure to opioid medications was kept as low as possible. Although cognitive function seems to be preserved in patients taking moderate doses of opioids for chronic pain,26, 27 the documentation of the effect of high doses of opioids on cognitive function remains incomplete. Our current and past work2 in the field of drug-induced delirium has helped to identify a critical threshold of daily exposure to opioid medications (90 mg of morphine) above which the risk of delirium significantly increases. According to a recent review, daily doses >180 mg of morphine or a morphine equivalent may be considered excessive.28 To prevent/reverse delirium, escalation of the dose to such levels preferably should be avoided. More specifically, opioid doses should be reduced by 25% if analgesia is satisfactory; nonessential, centrally active medications should be discontinued; and sepsis, metabolic derangement, and/or tumor in the central nervous system should be excluded. If delirium persists, then a change to an alternative opioid, a change in opioid route to the intraspinal route (with or without local anesthetic), a trial of other anesthetic or neurosurgical options, or a trial of antipsychotic medication should be considered.18 Antipsychotic prophylaxis is one of the most promising research directions in pharmacologic delirium prevention.29 The longitudinal association between opioids and delirium noted in this study clearly lends support to the causality of the correlation but also minimizes the possibility of interference by a fixed confounding factor, such as age. This strongly suggests that delirium indeed is associated with a time-varying factor. Our results also lend support to a previous study showing that a change in exposure to anticholinergic medications was associated independently with a change in the severity of delirium symptoms.8
In contrast, we did not note significant associations for the other groups of medications studied. However, these results should be interpreted with caution because of study limitations. First, we could not analyze longitudinally the impact of other precipitating factors on delirium course during hospitalization (eg, episodes of infection). Second, we could not account for a potential dose-accumulation effect or possible drug interactions or for the possibility that the effect of a drug exposure may last more than 1 day. Third, our primary measurements were collected by using a newly validated tool that shows promise but that has not been proven sufficiently to date. For instance, its discriminatory capacity to separate out dementia from delirium could be studied in other populations, especially because the rate of dementia in our population is low. Fourth, this study included a number of patients from a previous study.2 Although both studies favor the involvement of opioids in causing delirium, the generalizability of the results presented here remains to be documented. Finally, including a specific measure of performance status or disease severity in the study potentially may have provided more extensive information on the demographic characteristics of the sample examined.
No associations were noted between corticosteroids or benzodiazepines and delirium. According to the available literature, the evidence linking these 2 classes of medications to delirium seems to be weaker than that for opioids,3 although all 3 groups of medications are plausible risk factors from a biologic standpoint.30 Benzodiazepines and corticosteroids have been associated with delirium in case series31–34 (a rather weak study design).35 In cohort and case-control studies, no significant associations were identified between corticosteroids and delirium, whereas benzodiazepines were associated significantly with delirium in 1 study. However, most of those studies had methodological flaws.3 Conversely, in our previous work,2 we observed significant associations between benzodiazepines, corticosteroids, and an increased risk of delirium by using a more rigorous methodology than that employed in previous studies. It also is conceivable that statistical power was lacking in the current study and, thus, that small differences were not identifiable. Nevertheless, based on our results, corticosteroids and benzodiazepines do not seem to increase the longitudinal risk of delirium. Obviously, the data regarding the associations between these drug classes and delirium still are contradictory, and further investigation into this matter is warranted.
In conclusion, based on our previous work2 and on the current report, we seem to have isolated a single modifiable delirium risk factor, exposure to opioid medications, that would influence both the development of the condition2 and its subsequent course. Prevention of delirium is a major aspect of its psychiatric management, and great efforts have been made in developing interventions designed to reduce its frequency, with undeniable success.36, 37 However, to date, these interventions have been somewhat complex, time- and money-consuming, and are not necessarily suited to all settings. Bearing in mind that our findings may relate only to specific populations and that the implications of these findings also may depend on other contributing factors in those specific populations, the results presented in this article constitute readily applicable clinical data. It appears that patients who are exposed to daily opioid doses <90 mg are at lower risk of delirium. Furthermore, these results provide a significant evidence basis for upcoming randomized controlled trials targeting these medications. Some examples of simple experimental interventions would include reducing the doses of opioids and/or adding antipsychotics.
Supported by a grant from the Sociobehavioural Cancer Research Network and funded through the National Cancer Institute of Canada (NCIC) Centre for Behavioural Research and Program Evaluation with funds from the Canadian Cancer Society.