We present a useful research algorithm for detecting delirium in an ICU setting. This method utilizing both the CAM-ICU and a validated chart review demonstrates a more comprehensive approach to detection of delirium for research purposes.
Compared with research nurse ratings using the CAM-ICU, the chart-based method has a sensitivity of 64% and a specificity of 85%. The positive predictive accuracy of the chart-based method was 87%, which is much higher than the 39% reported in a non-ICU population and is probably related to the greater prevalence of delirium in the ICU [20
Numerous studies have verified the under-recognition and under-documentation of delirium by both physician and nursing staff [21
]. Under-documentation of delirium in the medical record is supported by our findings, in that there was no chart documentation for 36% (105/292) of delirium cases identified by the CAM-ICU.
Our false-positive rate for chart-based detection was 15%. Because of the fluctuating nature of delirium, the CAM-ICU may miss cases of delirium if it is performed only once a day. In this study research nurses performed the CAM-ICU during the day shift. When we examined shift of chart documentation for our 28 false-positive patients, we found that only five had chart documentation on the day shift whereas the rest were documented on nights only or evenings only or some combination of shifts. This reflects the fluctuating nature of delirium. In addition, our findings on the prevalence of delirium (80%) are similar to those other studies that reported on ICU delirium (40–87%) [1
Delirium is a fluctuating disorder, and reliance on a single daily observation can substantially underestimate the prevalence of delirium. Given the false-negative rate of 36% for chart review, we recommend that the CAM-ICU be used as the primary tool for detecting delirium in the ICU. For clinical care, ICU nurses should be trained to administer the instrument on each shift concurrent with assessments of sedation and acuity. Ely and coworkers [15
] have demonstrated that large-scale implementation of the CAM-ICU by nursing staff is feasible. However, when frequent CAM-ICU assessment is not feasible or when research staff members are unavailable, such as during weekends or holidays, using a validated chart review will markedly improve detection of delirium in the ICU, from 20% to 64% patient-days in our study. Chart review for detection of delirium has been used in multiple studies [20
]. Use of both the CAM-ICU and chart review represents a comprehensive delirium detection method in the ICU.
Depending on the nature of the study, coma and stupor may or may not have been included in delirium rates in previous studies [1
]. Prior research [30
] suggested that there is a spectrum of abnormal mental state and that patients may move between delirium, stupor, and coma. The CAM-ICU cannot be performed when a patient is in a state of stupor or coma, and these patients are often excluded from analysis when delirium and its impact on outcomes are evaluated. Previous research suggests that a large number of patients who have coma or stupor transition to delirium in the ICU. McNicoll and coworkers [1
] reported that 85% of patients who had coma or stupor transitioned to delirium, whereas 12% remained in coma/stupor and 3% transitioned to no delirium. In our study, of the 278 patient-days on which the CAM-ICU could not be performed because of stupor or coma, 205 (74%) had chart evidence for delirium. Only 5% (73/1,457) of our patient-days were stupor/coma with no chart documentation of delirium, and these were not counted as delirium but rather handled as a separate categorization.
Not surprisingly, the nursing staff documented delirium in their notes more frequently than did the physicians. The nurse-to-patient ratio in our ICU is usually 1:2, and thus the nurses spend much more time with the patients over the course of the day, allowing them to note changes in mental status as well as fluctuation in mental status. Our ICU nurses also give detailed sign out information when changing shifts so that the next nurses on duty are aware of each patient's baseline mental status, allowing them to assess better any changes that subsequently occur.
The strengths of the present study include the sizeable nature of the patient group with detailed daily clinical observations on 1,457 patient-days by a skilled and highly reliable research team. We applied two well validated instruments for delirium detection. In addition, this prospective ICU cohort was representative of the medical ICU population at our hospital. However, several caveats about the study deserve comment. No 'gold standard' method was used to validate our delirium diagnoses; however, both delirium measures used have been externally validated and are widely employed. One research nurse performed the chart-based abstraction, and this may be a potential source of bias and limit the generalizability of the findings. As with any single-site study, the generalizability of the results may be called into question. Although the external validity could be challenged, this does not compromise the internal validity of our findings, which require replication in other settings and populations. Finally, the proposed algorithm is intended for use in research studies and not for general clinical purposes, where more frequent application of the CAM-ICU is recommended because of its superior performance compared with the chart review method.