Abstract
Rationale: Delirium severity and duration are independently associated with higher mortality and morbidity. No studies to date have described a delirium trajectory by integrating both severity and duration.
Objectives: The primary aim was to develop delirium trajectories by integrating symptom severity and duration. The secondary aim was to investigate the association among trajectory membership, clinical characteristics, and 30-day mortality.
Methods: A secondary analysis of the PMD (Pharmacologic Management of Delirium) randomized control trial (ClinicalTrials.gov Identifier: NCT00842608; N = 531) was conducted. The presence of delirium and symptom severity were measured at least daily for 7 days using the Confusion Assessment Method for the intensive care unit (CAM-ICU) and CAM-ICU-7 (on a scale of 0–7, with 7 being the most severe). Delirium trajectories were defined using an innovative, data-driven statistical method (group-based trajectory modeling [GBTM]) and SAS v9.4.
Results: A total of 531 delirious participants (mean age 60 yr [standard deviation = 16], 55% female, and 46% African American) were analyzed. Five distinct delirium trajectories were described (CAM-ICU-7: mean [standard deviation]); mild-brief (CAM-ICU-7: 0.5 [0.5]), severe-rapid recovers (CAM-ICU-7: 2.1 [1.0]), mild-accelerating (CAM-ICU-7: 2.2 [0.9]), severe-slow recovers (CAM-ICU-7: 3.9 [0.9]), and severe-nonrecovers (CAM-ICU-7: 5.9 [1.0]). Baseline cognition and race were associated with trajectory membership. Trajectory membership independently predicted 30-day mortality while controlling for age, sex, race, cognition, illness severity, and comorbidities.
Conclusions: This secondary analysis described five distinct delirium trajectories based on delirium symptom severity and duration using group-based trajectory modeling. Trajectory membership predicted 30-day mortality.
Keywords: delirium, trajectory of illness, delirium severity, critical care, prediction
Approximately 80% of critically ill patients receiving mechanical ventilation experience delirium, a type of acute brain failure (1–4). Delirium is independently associated with intensive care unit (ICU) mortality, a prolonged length of stay, and in-hospital complications costing an additional $600 per patient ICU day (4–11). Delirium is traditionally described in terms of psychomotor behaviors and categorized as hypoactive (drowsy, sedated), hyperactive (restless, agitated), and/or mixed (fluctuation between the two states) (12–16). These subtypes follow the “delirium spectrum,” are linked to clinically meaningful outcomes, and provide insight into underlying motoric neural mechanisms (7, 12, 13, 17). Nonetheless, the outward display of agitation and restlessness in hyperactive delirium attracts disproportionate attention, leading to an underrecognition of hypoactive delirium even though it is associated with worse outcomes. Furthermore, the multidimensional nature of delirium is not captured (14). Thus, current delirium subtypes are limited by their basis on motoric behavior, do not include the severity or duration of the core diagnostic features of delirium (inattention, fluctuation of arousal level, and disorganized thinking), and do not illustrate how the course or trajectory of delirium severity relates to clinically relevant outcomes such as mortality. Understanding the trajectory of delirium severity may provide opportunities to alter the course of illness to prevent adverse outcomes, and provide supportive therapies to improve patients’ quality of life.
To advance research and clinical practice, we need to define dynamic subtypes or phenotypes of delirium that incorporate its complex, heterogeneous features using clinically meaningful diagnostic criteria (14). The term “dynamic” is used to represent one of the hallmark features of delirium, the fluctuation of symptoms over time. Current delirium research and clinical management are limited by a lack of precise interventions that target specific symptom profiles and differentiate among individuals. Dynamic delirium trajectories are the first step toward implementing such an approach. These trajectories would allow differentiation among the various complex symptom profiles of delirium, facilitating the clinical use of targeted therapies to mitigate symptoms and improve outcomes associated with each trajectory. Furthermore, each trajectory could then be examined to elucidate the (possibly differential) biological underpinnings of delirium and be further developed into a phenotype (18–20).
The purpose of this study was to characterize clinically meaningful, dynamic delirium trajectories. The main goals were to 1) define delirium trajectories based on the presence and duration of core diagnostic symptom criteria using a data-driven advanced statistical technique, 2) identify admission and clinical correlates, and 3) examine the associations between trajectory membership and 30-day mortality.
Methods
This was a secondary data analysis of deidentified, prospective, longitudinal data collected in two federally funded, randomized controlled clinical trials: PMD (Pharmacologic Management of Delirium; ClinicalTrials.gov Identifier: NCT00842608) and de-PMD (deprescribe-PMD) (21, 22). The institutional review board approved the original trials, and details were previously described (23). The study protocol is provided in the online supplement and outlines both studies. The patient’s legally authorized representatives provided written informed consent. This secondary analysis did not meet the definition of human subjects research and therefore did not require institutional review board review or approval.
Dataset
The dataset consisted of 551 patients who were admitted to the ICU of three Indianapolis hospitals and were enrolled in the PMD trials. Patients who were ≥18 years of age and had delirium based on the Confusion Assessment Method for the ICU (CAM-ICU) were eligible for study enrollment. Patients were excluded if they 1) were non–English speaking, 2) were hearing impaired, 3) were legally blind, 4) had been admitted with alcohol intoxication, 5) were prisoners, 6) had a history of severe mental illness, or 7) were pregnant/nursing. PMD participants completed baseline assessments before they were randomized to the intervention or usual care (21–23). Results of the main clinical trial were previously published (21, 22). For this analysis, we used all available patient data from study enrollment through Day 7 of the hospital stay. Twenty patients did not have delirium severity data after randomization and were excluded from this analysis (N = 531 patients, with a total of 4,438 observations).
Delirium Measurement
The dataset included trained raters’ daily assessment of PMD participants’ level of consciousness, delirium, and delirium severity from enrollment until death (CAM-ICU-7 score: mean = 3.8, standard deviation [SD] = 2.8) or hospital discharge (length of hospital stay: mean = 25.6 SD = 29.8 d). The participants’ level of consciousness was assessed with the Richmond Agitation-Sedation Scale (RASS) (24, 25). Participants with a RASS score of −4 or −5 were recorded as “coma,” a type of acute brain dysfunction. The presence of delirium was assessed by the CAM-ICU (26–29). The CAM-ICU-7 measures delirium severity using core diagnostic features and is scored directly from the CAM-ICU (5). The CAM-ICU-7 is an objective 7-point scale (0–7) with high internal consistency (Cronbach α = 0.85) (5). A score of 0 indicates no delirium, 1–2 indicates subsyndromal delirium, 3–5 indicates mild-to-moderate delirium, and 6–7 indicates severe delirium. The scoring system is outlined in Table E1 in the online supplement. Of note, for this analysis, individuals recorded as “coma” were assigned a CAM-ICU-7 score of 7, indicating acute brain dysfunction (11, 30–33).
The original study (January 1, 2009 to July 1, 2010; n = 113) protocol required once-daily assessments of RASS, CAM-ICU, and CAM-ICU-7. The study protocol was adjusted on July 1, 2010 to assess twice daily for RASS, CAM-ICU, and CAM-ICU-7. For this reason, we have included all available CAM-ICU-7 measurement scores from the time of study enrollment to Day 7 of study participation.
Baseline Demographic Data, Clinical Data, and Outcomes
Baseline demographics included age, sex, years of education, and race. Clinical data included chronic comorbidities and illness severity, which were assessed using the Charlson comorbidity index (CCI) and the Acute Physiology and Chronic Health Evaluation II (APACHE II) scale, respectively. Data for these assessments were obtained from electronic health records (EHRs). Information regarding the participant’s prehospital cognitive and functional status was provided by the his or her legally authorized representative via the Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE), Instrumental Activities of Daily Living (iADL), and the Barthel Index for Activities of Daily Living (ADL). Further information about these scales is provided in Table E2. Trained research assistants who were unblinded to the subjects’ delirium status reviewed the EHRs and conducted in-person, once or twice daily observations for the presence of the following inpatient clinical characteristics: bladder catherization, intravenous devices, restraints, and mechanical ventilation. Unsafe behavior was defined as the documentation or observation of one of the following behaviors during the 7-day analysis period: pulling out tubes/lines, trying to get out of bed, and documented need for restraints. These defined variables were preselected for inclusion in this analysis based on published studies of delirium that showed evidence of associations between these variables and delirium, as well as availability in the secondary dataset.
In-Hospital Outcomes and Mortality
Trained research assistants collected data daily on documentation in the EHR and/or the observed presence of the following inpatient clinical adverse events: nosocomial infection, pressure ulcer, delayed procedures, and falls (23). Adverse events were categorized as the presence of one of these events during the 7-day analysis period. Mortality, length of ICU and hospital stays, and discharge disposition were collected from EHRs.
Statistical Analysis
Characteristics of the study cohort were described using means/SDs and frequency counts depending on the type of data. Delirium severity trajectories were identified using trajectory modeling. This approach permitted us to estimate probabilities for several trajectories instead of a single population mean, as in growth-curve models or traditional regression. The maximum likelihood method was used to fit a semiparametric or discrete-mixture model to the longitudinal data with the SAS procedure Proc Traj (34–38). All available CAM-ICU-7 scores in the 7-day analysis period were modeled as a zero-inflated Poisson distribution. Participants who experienced study attrition (defined as either death or discharge in that 7-day period) were included in the initial analysis and excluded from the sensitivity analysis.
The optimal trajectory number (one to six) and shape (constant, linear, quadratic, cubic, or quartic terms) were selected using the Bayesian information criterion. For each number of trajectories, the shapes were varied until the best-fitting model was identified using the maximum Bayesian information criterion. The model fit was confirmed by evaluating the average posterior probability of trajectory assignment (with <0.70 considered a poor fit) and odds of correct classification (≥5.0). Further details on the trajectory modeling are included in the online supplement.
The associations among clinical characteristics, 30-day mortality, and delirium severity trajectories were assessed using analysis of variance (ANOVA) followed by multinomial regression to control for confounding. Inpatient clinical outcomes, including length of stay, occurrence of adverse events, and inpatient mortality, are described per trajectory. Finally, a prediction model for 30-day mortality was built using Least Absolute Shrinkage and Selection Operation (LASSO) (39–41). Multiple-comparison correction was completed for each model using the Bonferroni method (42). Effect sizes were evaluated for categorical variables using Cramer’s V (43). The influence of randomization in the original studies on trajectory modeling was examined using multinomial regression. SAS v9.4 and NCSS v12.0 were used for data analysis. The SAS Proc Traj package (38) was used to perform group-based trajectory modeling (GBTM). Significance was defined as P ≤ 0.05.
A total of 4,438 CAM-ICU-7 observations over 7 days were included in the analysis. Given the design parameters described above, there were 6,643 possible data points. This implies that we were missing 34% of potentially observable CAM-ICU-7 assessments. There were two types of missing assessment data on longitudinal CAM-ICU-7. The first type was intermittent missing (i.e., patients may have missed a particular assessment but had CAM-ICU-7 data after the missed assessment). Reasons for intermittent missing assessment data included incomplete CAM-ICU-7 scores (66), unable to assess owing to clinical care (666 data points, 10% of possible data points), and not collected (497 data points, 7% of possible data points). These reasons suggest that the data were “missing completely at random,” and the trajectory modeling approach is robust to this scenario (35). Missing assessment data could also result from study attrition due to death (n = 20, 110 data points, 2% of possible data points) or discharge (n = 134, 866 data points, 13% of possible data points) before Day 7. We included CAM-ICU-7 data collected before death or discharge in our trajectory modeling because delirium severity scores before the attrition events were highly predictive of these events. These missing assessment data are called “missing at random” (MAR) (i.e., the probability of missing assessment data depends on the observed data) and are appropriately handled in the trajectory modeling, which produces unbiased estimates under the MAR (35, 44).
We conducted a number of sensitivity analyses to probe the consistency of our inferences given the high proportion of missing data. These analyses included the following scenarios: 1) “coma” excluded, 2) participants who were missing seven or more CAM-ICU-7 assessments, and 3) participants who experienced study attrition due to death or discharge on or before Day 7.
Results
A total of 531 delirious participants were included in this analysis. At study enrollment, overall, the participants had a mean age of 60 (SD = 16) years, 55% were female, 46% were African American, and 73% were on mechanical ventilation. Additional clinical characteristics are shown in Table 1.
Table 1.
Descriptive characteristics of overall participants and by trajectory
| Demographics and Baseline Clinical Data | Total Cohort (N = 531) | Mild-Brief (n = 90) | Severe-Rapid Recovers (n = 79) | Mild-Accelerating (n = 35) | Severe-Slow Recovers (n = 104) | Severe-Nonrecovers (n = 223) |
|---|---|---|---|---|---|---|
| Mean ± SD | ||||||
| Age, yr | 60 ± 16 | 58 ± 16 | 59 ± 16 | 64 ± 15 | 59 ± 15 | 61 ± 16 |
| Years of education (n = 76 missing) | 11.4 ± 2.4 | 11.3 ± 2.2 | 11 ± 2.2 | 11.4 ± 2.8 | 11.4 ± 2.3 | 11.6 ± 2.6 |
| APACHE II | 20 ± 8.1 | 19.7 ± 8.7 | 19.1 ± 7.9 | 20.4 ± 9.0 | 19.6 ± 8.2 | 20.5 ± 7.8 |
| CCI | 3.2 ± 2.8 | 2.7 ± 2.5 | 3.2 ± 2.7 | 2.9 ± 3.7 | 3.1 ± 2.8 | 3.4 ± 2.9 |
| IQCODE | 3.2 ± 0.5 | 3.1 ± 0.4 | 3.2 ± 0.4* | 3.4 ± 0.6* | 3.2 ± 0.3* | 3.3 ± 0.5* |
| bADL | 5.4 ± 1.4 | 5.7 ± 0.9 | 5.4 ± 1.3 | 5.3 ± 1.3 | 5.4 ± 1.3 | 5.2 ± 1.5 |
| iADL | 6.1 ± 2.6 | 6.9 ± 2.0 | 6.0 ± 2.5* | 5.6 ± 2.7* | 6.5 ± 2.4* | 5.7 ± 2.8* |
| Days to delirium | 7.5 ± 7.0 | 6.2 ± 5.5 | 6.9 ± 5.8 | 11.3 ± 12.3* | 7.7 ± 7.2 | 7.3 ± 6.5 |
| Percentage (n) | ||||||
| Female | 55 (292) | 60 (54) | 48 (38) | 63 (22) | 51 (53) | 56 (125) |
| White | 53 (280) | 70 (62) | 57 (45) | 63 (22) | 50 (51) | 48 (107) |
| African American | 46 (241) | 30 (27) | 43 (34)* | 37 (13)* | 50 (52)* | 52 (115)* |
| ICU location | ||||||
| MICU | 68 (353) | 68 (76) | 71 (56) | 69 (24) | 63 (65) | 63 (140) |
| PICU | 10 (55) | 7 (8) | 10 (8) | 14 (5) | 10 (10) | 11 (25) |
| SICU | 23 (121) | 14 (16) | 19 (15) | 17 (6) | 27 (29) | 26 (57) |
| Admission diagnosis | ||||||
| Respiratory | 28 (148) | 30 (27) | 29 (23) | 26 (9) | 26 (27) | 28 (62) |
| AMS | 5 (27) | 6 (5) | 10 (8) | 3 (1) | 3 (3) | 4 (10) |
| Neurologic | 3 (17) | 6 (5) | 4 (3) | 0 (0) | 3 (3) | 3 (6) |
| Gastrointestinal | 9 (50) | 10 (9) | 6 (5) | 9 (3) | 13 (14) | 9 (19) |
| Resp + sepsis | 13 (70) | 13 (12) | 4 (3) | 17 (6) | 13 (13) | 16 (36) |
| Sepsis | 16 (84) | 13 (12) | 19 (15) | 23 (8) | 17 (18) | 15 (33) |
| Surgical | 5 (27) | 11 (10) | 6 (2) | 0 (0) | 8 (8) | 6 (136) |
| Trauma | 10 (53) | 4 (4) | 13 (10) | 9 (3) | 9 (9) | 13 (28) |
| Other | 10 (54) | 16 (14) | 13 (10) | 14 (5) | 9 (9) | 7 (16) |
| Ventilated at enrollment | 73 (390) | 67 (60) | 75 (59) | 54 (19)* | 79 (82) | 76 (170) |
Definition of abbreviations: AMS = altered mental status; APACHE II = Acute Physiology and Chronic Health Evaluation II; bADL = Barthel Index for Activities of Daily Living; CCI = Charlson comorbidity index; iADL = Instrumental Activities of Daily Living questionnaire; ICU = intensive care unit; IQCODE = Informant Questionnaire on Cognitive Decline in the Elderly; MICU = medical ICU; PICU = progressive ICU; Resp = respiratory; SD = standard deviation; SICU = surgical ICU.
Analysis of variance: iADL (P = 0.001, F = 4.69), IQCODE (P < 0.001, F = 5.57), African American (P = 0.008, F = 3.45), days to study enrollment from ICU admission (P = 0.006, F = 3.67). Chi-square for binary. Statistical significance is noted as P < 0.05.
Indicates statistical significance between trajectories (the reference trajectory is mild-brief).
Delirium Trajectories
Five distinct trajectories were identified (CAM-ICU-7 mean [SD]): mild-brief delirium (17% of the cohort, initially had a positive CAM-ICU-7 score of ≥3 followed by a decrease in symptoms over a 12- to 24-h period to a CAM-ICU-7 score of ≤2, indicating subsyndromal delirium across all time points, CAM-ICU-7: 0.5 [0.5]); severe-rapid recovers delirium (16% of the cohort, initially had severe delirium [CAM-ICU-7 score of 6–7] and symptoms quickly resolved within 60 h, CAM-ICU-7: 2.1 [1.0]); mild-accelerating delirium (7% of the cohort, began with mild-to-moderate delirium symptoms [CAM-ICU-7 scores 3–5], which quickly worsened, demonstrating a nonlinear, exponentially increasing trajectory to severe delirium [CAM-ICU-7 score 6–7], CAM-ICU-7: 2.2 [0.9]); severe-slow recovers delirium (22% of the cohort, severe delirium symptoms [CAM-ICU-7 score 6–7] that gradually declined over the 7-day period, CAM-ICU-7: 3.9 [0.9]); and severe-nonrecovers delirium (40% of the cohort, severe delirium symptoms [CAM-ICU-7 score of 6–7] that persisted over the 7-day assessment period, CAM-ICU-7: 5.9 [1.0]). Trajectories are illustrated in Figures 1 and E1, and described in Table 2. Model selection statistics are displayed in Tables E4 and E5. Multinomial regression indicated that randomization into the intervention cohorts did not significantly influence the trajectory modeling (P = 0.76, F = 0.47).
Figure 1.
The five delirium severity trajectories over the 7-day assessment period (x-axis). The mean Confusion Assessment Method for the Intensive Care Unit 7 (CAM-ICU-7) scores are shown on the y-axis. To the right of the graph, each trajectory is named (combining a severity term with a time element). The percentage represents the distribution of participants in each trajectory.
Table 2.
Descriptions of delirium trajectories
| Trajectories | Description |
|---|---|
| Mild-brief | Low CAM-ICU-7 (CAM-ICU-7 score: 1–2) across all time points |
| Severe-rapid recovers | Initial severe CAM-ICU-7 score with a sharp decline/resolution of delirium symptoms within 60 h (CAM-ICU-7 score of 6–7 indicating severe, declining to a mild-moderate score of 3–5, followed by a subsyndromal symptom score of 1–2) |
| Mild-accelerating | Mild CAM-ICU-7 scores (CAM-ICU-7 score: 1–2) that exponentially increased to moderate to severe delirium (CAM-ICU-7 score: 5–7) over the 7-d assessment period |
| Severe-slow recovers | Severe CAM-ICU-7 scores (CAM-ICU-7: 6–7) that gradually declined to mild delirium (CAM-ICU-7 score: 3–5) over the 7-d assessment period |
| Severe-nonrecovers | Severe CAM-ICU-7 scores (CAM-ICU-7: 6–7) that persisted over the 7-d assessment period |
Definition of abbreviation: CAM-ICU-7 = Confusion Assessment Method for the Intensive Care Unit 7
The table describes the delirium trajectories that are illustrated in Figure 1. CAM-ICU-7 scoring system: 0 = no delirium symptoms; 1–2 = subsyndromal delirium; 3–5 = mild-to-moderate delirium; 6–7 = severe delirium.
Clinical Correlates of the Trajectories
In ANOVA (Table 2), trajectories varied according to baseline cognition (mild-accelerating, IQCODE score of 3.4 [0.6] vs. mild-brief IQCODE of 3.1 [0.4]; P ≤ 0.001), identifying as African American (severe-nonrecovers 52% vs. mild-brief 30%; P = 0.008), baseline functional ability (mild-accelerating iADL of 5.6 [2.7] vs. mild-brief of 6.9 [2.0]; P = 0.001), and mechanically ventilated at enrollment (severe-slow recovers 79% vs. mild-accelerating 54%; P = 0.02).
In multinomial regression models (Table E6), trajectories varied by baseline IQCODE and identifying as African American while controlling for age, sex, and illness severity (APACHE II); comorbidities (CCI); and functional ability (iADL). After multiple-comparison correction, African Americans were almost three times more likely to belong to the severe-nonrecovers delirium trajectory (adjusted odds ratio, 2.8; 95% confidence interval [CI], 1.6–5.0).
Trajectories varied by number of days with a bladder catheter (severe-nonrecovers 4.5 [1.9] d vs. mild-brief 1.7 [1.7] d; P = 0.001), restraints (severe-nonrecovers 3.0 [2.1] d vs. mild-brief 0.3 [1.6] d; P = 0.001), and mechanical ventilation (severe-nonrecovers 2.3 [1.9] d vs. mild-brief 0.2 [0.8] d; P = 0.001) while controlling for age, sex, race, APACHE II, IQCODE, and iADL, CCI, discharge disposition, hospital mortality, and hospital length of stay. The frequency of unsafe behavior was also associated with trajectory membership while controlling for age, race, sex, iADL, IQCODE, APACHE II, CCI, discharge disposition, hospital mortality, and length of stay (Table 3).
Table 3.
Clinical characteristics of the study participants
| Total Cohort (N = 531) | Mild-Brief (n = 90) | Severe-Rapid Recovers (n = 79) | Mild-Accelerating (n = 35) | Severe-Slow Recovers (n = 104) | Severe-Nonrecovers (n = 223) | |
|---|---|---|---|---|---|---|
| Clinical assessments | ||||||
| Mean ± SD | ||||||
| CAM-ICU-7 | 3.8 ± 2.8 | 0.5 ± 0.5 | 2.1 ± 1.0* | 2.2 ± 0.9* | 3.9 ± 0.9* | 5.9 ± 1.0* |
| Variation in CAM-ICU-7 | 4.0 ± 3.6 | 1.0 ± 1.4 | 6.3 ± 3.5* | 4.6 ± 3.3* | 7.5 ± 2.8* | 2.5 ± 2.6* |
| RASS | −1.5 ± 1.3 | −0.2 ± 0.3 | −0.8 ± 0.6* | −1.0 ± 0.8* | −1.6 ± 0.7* | −2.4 ± 1.2* |
| Daily clinical profile | ||||||
| Ventilation days | 1.5 ± 1.8 | 0.2 ± 0.8 | 0.6 ± 0.9* | 0.7 ± 1.3 | 1.5 ± 1.7* | 2.3 ± 1.9* |
| Bladder catheter days | 3.7 ± 2.0 | 1.7 ± 1.7 | 2.9 ± 1.5* | 3.1 ± 1.9* | 4.5 ± 1.4* | 4.5 ± 1.9* |
| Restraint days | 2.0 ± 2.0 | 0.3 ± 1.6 | 1.1 ± 1.1* | 0.7 ± 1.2* | 2.5 ± 1.7* | 3.0 ± 2.1* |
| Additional symptoms, % (n) | 17 (91) | 15 (14) | 12 (11) | 7 (6) | 21 (19) | 45 (41) |
| Unsafe behavior, % (n) | 15 (77) | 5 (4) | 13 (10) | 5 (4) | 18 (14) | 58 (45)* |
Definition of abbreviations: CAM-ICU-7 = Confusion Assessment Method for the Intensive Care Unit 7; RASS = Richmond Agitation-Sedation Scale; SD = standard deviation.
Shown are the descriptive statistics for the twice-daily clinical assessments, which included delirium severity assessments (CAM-ICU-7), variation in CAM-ICU-7 score (calculated as the average of the squared differences from the mean, illustrating the spread of the delirium severity scores), sedation (RASS), additional symptoms (observation or patient report of hallucinations, delusions, excessive physical agitation, verbose language, or uncooperativeness), unsafe behavior (observation or nurse report of patient pulling at tubes/lines, trying to get out of bed, or need for restraints), the presence of a bladder catheter, restraint, or mechanical ventilation. Multinomial regression examined the relationships between trajectory membership and twice-daily clinical assessments and clinical profile. Age, sex, race, illness severity, Charlson comorbidity index, and baseline cognition and function (Instrumental Activities of Daily Living) were added as covariates to the models, and the adjusted P value is displayed.
In row indicates statistical significance (adjusted) between trajectories, reference trajectory is mild-brief, P < 0.05.
The inpatient and 30-day outcomes of the trajectories are described in Table 4. Because these trajectories are approximations of unobserved patterns in the data and may be influenced by study attrition due to death or discharge by Day 7, statistical differences are not reported for outcomes that occurred during that 7-day period.
Table 4.
Descriptive statistics for patient outcomes per trajectory
| Full Cohort (N = 531) | Mild-Brief (n = 90) | Severe-Rapid Recovers (n = 79) | Mild-Accelerating (n = 35) | Severe-Slow Recovers (n = 104) | Severe-Nonrecovers (n = 223) | |
|---|---|---|---|---|---|---|
| Hospital complications | ||||||
| Adverse events | ||||||
| Yes, % (n) | 43 (228) | 26 (23) | 39 (31) | 51 (18) | 48 (50) | 43 (106) |
| Falls, n | 3 | 0 | 0 | 0 | 0 | 3 |
| Sitter patient days, total | 24 | 2.5 | 7 | 3 | 4.5 | 8 |
| Mortality % (n) | ||||||
| 7-d | 4 (20) | 2 (2) | 0 (0) | 0 (0) | 0 (0) | 8 (18) |
| ICU | 7 (39) | 2 (2) | 1 (1) | 17 (6) | 5 (5) | 11 (25) |
| Hospital | 9 (50) | 3 (3) | 1 (1) | 20 (7) | 8 (8) | 14 (31) |
| 30-d* | 12 (65) | 3 (3) | 4 (3) | 31 (11)* | 10 (10) | 17 (38)* |
| Length of stay | ||||||
| ICU | 21.5 ± 28.4 | 12.5 ± 10.5 | 15.5 ± 15.8 | 18.7 ± 11.6 | 27.2 ± 46.8 | 25.1 ± 25.1 |
| Hospital | 25.6 ± 29.8 | 17.1 ± 16.4 | 20.1 ± 18.3 | 25.4 ± 21.5 | 30.9 ± 47.6 | 28.5 ± 26.7 |
| Discharge, % (n) | ||||||
| Home | 38 (189) | 62 (55) | 46 (36) | 20 (7) | 36 (37) | 24 (54) |
Definition of abbreviation: ICU = intensive care unit.
The table displays the descriptive statistics for patient outcomes per trajectory. P value was adjusted by age, sex, race, illness severity, comorbidity index, and baseline cognition and function for 30-day mortality. Adverse events are defined as the presence of nosocomial infection, new pressure ulcer, delayed procedure, or fall during the 7-day analysis period.
P < 0.001.
Trajectories and 30-Day Mortality
In ANOVA (Table 4), 30-day mortality outcomes significantly varied between mild-accelerating 31% vs. severe-nonrecovers 17% vs. mild-brief 3%; P < 0.001). The mild-accelerating trajectory was associated with 30-day mortality (31% vs. 3% in the mild-brief trajectory; adjusted P < 0.001) after controlling for age, sex, race, and illness severity (APACHE II); comorbidity burden (CCI); and baseline cognition and function (IQCODE and iADL). These associations remained after multiple-comparison correction and are displayed in Table 5.
Table 5.
Logistic regression results for the association between trajectory membership and 30-day mortality in the original analysis and two sensitivity analyses
| Outcome: 30-d Mortality | χ2 | P Value | Mild-Brief | Severe-Rapid Recovers | Mild-Accelerating | Severe-Slow Recovers | Severe-Nonrecovers |
|---|---|---|---|---|---|---|---|
| Odds Ratio (95% Confidence Interval) |
|||||||
| Original analysis, (N = 531) | 50.70 df = 11 | <0.001 | Ref | 0.58 (0.09–3.61) | 9.39* (2.28–38.60) | 2.0 (0.50–8.01) | 4.24* (1.22–14.66) |
| Sensitivity analysis #1 (n = 491) | 47.77 df = 11 | <0.001 | Ref | 0.47 (0.07–3.00) | 7.86* (1.89–32.78) | 1.55 (0.38–6.27) | 3.44 (0.98–1.02) |
| Sensitivity analysis #2 (n = 376) | 44.70 df = 11 | <0.001 | Ref | 0.39 (0.06–2.63) | 8.80* (1.82–42.47) | 0.81 (0.18–3.54) | 2.15 (0.58–7.98) |
Definition of abbreviations: CAM-ICU-7 = Confusion Assessment Method for the Intensive Care Unit 7; df = degrees of freedom; Ref = reference.
The mild-brief trajectory was used as a reference category. Age, sex, race, and illness severity (Acute Physiology and Chronic Health Evaluation II); Charlson comorbidity index; baseline cognition (Informant Questionnaire on Cognitive Decline in the Elderly); and function (Instrumental Activities of Daily Living) were covariates within the models. These significant relationships remained after multiple-comparison correction (P < 0.001). Sensitivity analysis #1 removed participants missing more than seven CAM-ICU-7 assessments. Sensitivity analysis #2 removed all participants who experienced study attrition before Day 7 (death or discharge). In logistic regression procedures using Least Absolute Shrinkage and Selection Operation regression, age, illness severity (Acute Physiology and Chronic Health Evaluation II), comorbidities (Charlson comorbidity index), Instrumental Activities of Daily Living scores, and trajectory membership significantly predicted 30-day mortality (receiver operating characteristic, 0.78 [95% confidence interval, 0.73–0.84]; P < 0.001). Baseline cognition, sex, and race were not significant predictors of 30-day mortality and dropped out of the Least Absolute Shrinkage and Selection Operation modeling procedures.
Denotes significance after multiple-comparison correction.
LASSO regression identified age, comorbidities (CCI), illness severity (APACHE II), functional status (iADL), and trajectory to predict 30-day mortality (receiver operating characteristic (ROC), 0.78; 95% CI, 0.73–0.84). The addition of trajectory membership to the prediction model significantly improved performance (P < 0.0001) from an ROC of 0.70 (95% CI, 064–0.77) to an ROC of 0.78 (95% CI, 0.73–0.84). Baseline cognition, sex, and race were not identified as predictors of 30-day mortality and dropped out of the LASSO modeling procedures.
Three different sensitivity analyses were completed. In the first sensitivity analysis, modeling procedures were run without coma coded as a “7” indicating severe delirium. The same five-trajectory model was identified as the best fit for the data. The membership distribution per trajectory and mean CAM-ICU-7 scores were not significantly changed. In the second sensitivity analysis, participants who were missing more than seven CAM-ICU-7 assessment points were removed (n removed = 39, n analyzed = 492). Again, the same five-trajectory model was identified as the best fit for the data. The membership distributions per trajectory and mean CAM-ICU-7 scores were not significantly changed. Further details are provided in the online supplement. The final sensitivity analysis was based on those participants who survived or remained in the hospital past Day 7 (n analyzed = 376). The modeling procedures identified the five-trajectory model as the best fit for the data, and the membership distribution and mean CAM-ICU-7 scores were not significantly changed. Baseline and clinical characteristics that described trajectories largely remained the same; however, IQCODE and identifying as African American were no longer significantly different between trajectories. Nonetheless, medium effect sizes were maintained (n = 531, Cramer’s V for African American 0.16; n = 376, Cramer’s V for African American 0.11) indicating that the loss of significance may be due to the sample size. In logistic regression, trajectory membership (mild-accelerating) remained associated with 30-day mortality while controlling for age, CCI, APACHE II, race, sex, IQCODE, and iADL (P = 0.005). Tables E7 and E8 display the attrition due to death or discharge by trajectory and per data analysis day.
Discussion
In this analysis, we defined five distinct dynamic delirium trajectories based on the duration and severity of diagnostic symptom criteria using a data-driven advanced statistical method, GBTM. These defined trajectories were significantly associated with admission and clinical characteristics as well as clinically relevant patient outcomes, including 30-day mortality. The inclusion of diagnostic symptom criteria, as outlined in the CAM-ICU-7 assessment tool, over a 7-day period captures the complex and dynamic nature of delirium and has implications for both clinical assessment and research. Clinically, therapies to mitigate residual delirium symptoms and prevent future adverse health outcomes after the ICU stay can be customized and tailored to each trajectory to maximize recovery. Further research to examine the biological underpinnings of each trajectory may elucidate unique genetic and biomarker profiles, leading to further refinement of a dynamic delirium symptom phenotype.
This study builds on previous research by defining delirium trajectories based on the duration of the four central diagnostic features of delirium. Previous subtypes have been defined by psychomotor behavior (15, 16, 45). Although this approach may be useful for clinical settings, it focuses on only one diagnostic feature of delirium and is likely influenced by the presence of ICU sedation. Other trajectories have used presumed etiology, location, or demographic characteristics to build subtypes (11, 46, 47). Although these are valuable for research and categorization, they do not capture the complexity of delirium symptoms over time as well as the core features of delirium. The trajectories we identified are in line with a previous study in which subgroups of delirium symptoms were based on the delirium index symptom scale over a 15-day period in geriatric medical patients (n = 230) (48). That study included five clusters, or subgroups, of delirium symptoms over time with similar patterns. In the current analysis, we built on those findings by using a delirium severity tool twice daily that was based solely on the core diagnostic features of delirium in an ICU population.
Variations in admission and clinical characteristics by trajectory membership can be used to identify (target) those at greatest risk and/or personalize (tailor) prevention measures for each individual patient. Impaired baseline cognitive status, measured using the IQCODE, was significantly higher in the mild-accelerating trajectory. This finding is supported by previous studies, as preexisting cognitive impairment is the highest, or most robust, predictor of delirium (6, 49–51). Individuals with higher IQCODE scores likely have ongoing neurodegeneration, are more vulnerable to delirium, and lack compensatory mechanisms to manage delirium symptoms (52). Although preexisting cognitive impairment increases the overall risk of delirium, in this particular analysis, preexisting cognitive impairment significantly increased the risk of belonging to the mild-accelerating trajectory, but not other trajectories. Identifying as African American was predictive of the severe-nonrecovers delirium trajectory while controlling for age, comorbidities, and illness severity. Few studies to date have examined associations between race and delirium. Additional studies to replicate our findings are recommended.
In this study we used a novel method, GBTM, to identify trajectories of delirium severity in a critically ill population. Although this is the first study to use this method in delirium research, GBTM and trajectory modeling have been applied to identify varying courses of illness in other disease states, such as multiple sclerosis and congestive heart failure (53, 54). In both diseases, trajectories are used to identify therapies that may alter or slow the disease course, and to understand what supportive services are needed to enhance the patients’ quality of life (55–57).
Modeling the trajectories of a dynamic measure such as delirium severity has implications for clinical care. First, the addition of a dynamic measure to a static prediction model can not only improve predictive ability, it can also enable clinicians to assess the impact of interventions in real time and adjust care accordingly. Researchers and designers of clinical care models should consider incorporating dynamic or “real-time” measurements into prediction models to continually evaluate the effectiveness of the care and intervention provided (30).
Delirium severity trajectories can be used to personalize (target or tailor) delirium management therapies in an effort to change the trajectory and predicted outcome. For example, the mild-accelerating trajectory demonstrated a nonlinear trajectory, with an exponential increase in delirium severity, starting at ICU Day 4. This trajectory had the highest percentage of adverse hospital events, suggesting that this trajectory may have been altered by the occurrence of such events, resulting in moderate-to-severe delirium and a high mortality rate. The baseline characteristics of this trajectory reveal two known risk factors for delirium: increasing age and preexisting cognitive impairment. Although neither of these conditions is modifiable, prevention measures to mitigate patients’ vulnerability and bolster resilience could be instituted upon admission to prevent the acceleration of delirium severity and occurrence of poor clinical outcomes.
Our findings are the first step toward precision delirium prevention and management. In line with the collaborative framework outlined by the National Institutes of Nursing Research Symptom Science Model, this study identified a trajectory of delirium severity (symptoms) in critically ill patients over the course of 7 days. The next steps are to characterize the phenotype associated with these symptom trajectories and investigate possible biomarkers using an -omic approach (19, 20). Future studies may also begin to examine which therapeutic treatments may prevent or alter these trajectories. Because this is one of the first studies of its kind, investigators should continue to explore and validate delirium severity trajectories and characterize phenotypes in similar samples, additional patient populations, and more diverse populations through secondary data analyses and/or prospective, longitudinal research. Furthermore, in this study we focused on delirium severity based on all symptoms and did not investigate the prevalence of specific features of delirium, such as inattention and disorganized thinking. Future research could expand to include investigations of a specific symptom burden within trajectories.
Strengths and Limitations
This study has several strengths. We collected carefully annotated data for the trajectories from at least daily delirium severity assessments using a validated clinical tool for the ICU population. We used a robust statistical technique to identify delirium trajectories based on severity over time, allowing for inclusion of individual and heterogeneous fluctuating trends in delirium. We confirmed our findings using three different sensitivity analyses, each of which identified a five-trajectory model. Lastly, this large sample of delirious patients was demographically diverse.
The study’s limitations include the fact that the study population was limited to an ICU cohort and the results may not generalize to other medical and surgical populations. Participants who had a RASS score of −4 or −5 were categorized as “coma” and included in the analysis with a CAM-ICU-7 score of 7, indicating severe delirium, or acute brain failure. This method has been used in previous studies; however, this might introduce bias, as the literature has shown differences between coma and delirium with regard to outcomes (9). To account for this potential bias, we conducted a sensitivity analysis that excluded coma as severe delirium. Missing assessments due to study attrition within the 7-day analysis period may introduce bias. We attempted to account for this potential bias by selecting an advanced modeling procedure that is able to accommodate MAR data. The MAR assumption requires that the variables we do have that predict missingness are also sufficient to predict the values that we would have observed, but did not (44). We also probed the adequacy of the MAR assumption by completing three different sensitivity analyses to identify how missing data may have influenced our findings. Therefore, these results should be considered “hypothesis-generating” and need to be validated in a subsequent study. All three sensitivity analyses identified the same five-trajectory model as the best fit for the data and mean CAM-ICU-7 scores. However, we did note that differences in baseline characteristics were identified in the multinomial model that accounted for study attrition due to discharge or death before ICU Day 7.
Conclusions
In this secondary analysis of carefully annotated existing data, we described five distinct delirium trajectories based on diagnostic delirium symptom criteria over time. These five trajectories were associated with demographic and clinical correlates as well as clinically relevant patient outcomes, including 30-day mortality. Our findings have important implications for the development of a precision healthcare approach to delirium prevention and management, including the characterization of phenotypes and underlying biological markers.
Acknowledgments
Acknowledgment
The authors thank all of the individuals who provided guidance and support throughout the conduct of this study, and the participants of the original trial, without whose participation this study would not have been possible. They also thank Patrick Monahan, M.D., Ph.D. for reviewing the statistical methods and findings.
Footnotes
Supported by a grant from the U.S. National Heart, Lung, and Blood Institute (T32: 5T32HL091816-07) (H.L.). S.W. is supported by grants from the National Institute on Aging (NIA; 2P30AG010133 and K23AG062555-01). B.A.K. and S.G. are supported by grants from the National Heart, Lung, and Blood Institute (R01HL131730) and the NIA (R01AG055391). M.A.B. was supported by a grant from the NIA (R01AG034205).
Author Contributions: Execution of original randomized control trial: B.A.K. and M.A.B. Study conception and protocol for secondary analysis: H.L., B.A.K., J.S.C., S.G., and M.A.B. Guidance on trajectory/phenotype development: J.S.C. and M.A.B. Data analysis and interpretation: H.L., S.G., A.J.P., and R.N.J. Manuscript writing: H.L., B.A.K., and J.S.C. Manuscript revision: H.L., B.A.K., J.S.C., S.G., A.J.P., S.H.K., S.W., R.N.J., and M.A.B.
This article has an online supplement, which is accessible from this issue’s table of contents at www.atsjournals.org.
Author disclosures are available with the text of this article at www.atsjournals.org.
References
- 1.Hayhurst CJ, Pandharipande PP, Hughes CG. Intensive care unit delirium: a review of diagnosis, prevention, and treatment. Anesthesiology. 2016;125:1229–1241. doi: 10.1097/ALN.0000000000001378. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Mehta S, Cook D, Devlin JW, Skrobik Y, Meade M, Fergusson D, et al. SLEAP Investigators; Canadian Critical Care Trials Group. Prevalence, risk factors, and outcomes of delirium in mechanically ventilated adults. Crit Care Med. 2015;43:557–566. doi: 10.1097/CCM.0000000000000727. [DOI] [PubMed] [Google Scholar]
- 3.Inouye SK, Westendorp RG, Saczynski JS. Delirium in elderly people. Lancet. 2014;383:911–922. doi: 10.1016/S0140-6736(13)60688-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Girard TD, Thompson JL, Pandharipande PP, Brummel NE, Jackson JC, Patel MB, et al. Clinical phenotypes of delirium during critical illness and severity of subsequent long-term cognitive impairment: a prospective cohort study. Lancet Respir Med. 2018;6:213–222. doi: 10.1016/S2213-2600(18)30062-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Khan BA, Perkins AJ, Gao S, Hui SL, Campbell NL, Farber MO, et al. The confusion assessment method for the ICU-7 delirium severity scale: a novel delirium severity instrument for use in the ICU. Crit Care Med. 2017;45:851–857. doi: 10.1097/CCM.0000000000002368. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Singh TD, O’Horo JC, Gajic O, Sakusic A, Day CN, Mandrekar J, et al. Risk factors and outcomes of critically ill patients with acute brain failure: a novel end point. J Crit Care. 2018;43:42–47. doi: 10.1016/j.jcrc.2017.08.028. [DOI] [PubMed] [Google Scholar]
- 7.Han JH, Brummel NE, Chandrasekhar R, Wilson JE, Liu X, Vasilevskis EE, et al. Exploring delirium’s heterogeneity: association between arousal subtypes at initial presentation and 6-month mortality in older emergency department patients. Am J Geriatr Psychiatry. 2017;25:233–242. doi: 10.1016/j.jagp.2016.05.016. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Brummel NE, Jackson JC, Pandharipande PP, Thompson JL, Shintani AK, Dittus RS, et al. Delirium in the ICU and subsequent long-term disability among survivors of mechanical ventilation. Crit Care Med. 2014;42:369–377. doi: 10.1097/CCM.0b013e3182a645bd. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Pandharipande PP, Girard TD, Jackson JC, Morandi A, Thompson JL, Pun BT, et al. BRAIN-ICU Study Investigators. Long-term cognitive impairment after critical illness. N Engl J Med. 2013;369:1306–1316. doi: 10.1056/NEJMoa1301372. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Devlin JW, Fong JJ, Fraser GL, Riker RR. Delirium assessment in the critically ill. Intensive Care Med. 2007;33:929–940. doi: 10.1007/s00134-007-0603-5. [DOI] [PubMed] [Google Scholar]
- 11.Vasilevskis EE, Chandrasekhar R, Holtze CH, Graves J, Speroff T, Girard TD, et al. The cost of ICU delirium and coma in the intensive care unit patient. Med Care. 2018;56:890–897. doi: 10.1097/MLR.0000000000000975. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Krewulak KD, Stelfox HT, Ely EW, Fiest KM. Risk factors and outcomes among delirium subtypes in adult ICUs: a systematic review. J Crit Care. 2020;56:257–264. doi: 10.1016/j.jcrc.2020.01.017. [DOI] [PubMed] [Google Scholar]
- 13.Han JH, Hayhurst CJ, Chandrasekhar R, Hughes CG, Vasilevskis EE, Wilson JE, et al. Delirium’s arousal subtypes and their relationship with 6-month functional status and cognition. Psychosomatics. 2019;60:27–36. doi: 10.1016/j.psym.2018.05.003. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Stevens RD, Zink EK. Subtypes of delirium: a step toward precision medicine. Crit Care Med. 2018;46:2058–2059. doi: 10.1097/CCM.0000000000003462. [DOI] [PubMed] [Google Scholar]
- 15.Grover S, Sharma A, Aggarwal M, Mattoo SK, Chakrabarti S, Malhotra S, et al. Comparison of symptoms of delirium across various motoric subtypes. Psychiatry Clin Neurosci. 2014;68:283–291. doi: 10.1111/pcn.12131. [DOI] [PubMed] [Google Scholar]
- 16.Pandharipande P, Cotton BA, Shintani A, Thompson J, Costabile S, Truman Pun B, et al. Motoric subtypes of delirium in mechanically ventilated surgical and trauma intensive care unit patients. Intensive Care Med. 2007;33:1726–1731. doi: 10.1007/s00134-007-0687-y. [DOI] [PubMed] [Google Scholar]
- 17.Han JH, Zimmerman EE, Cutler N, Schnelle J, Morandi A, Dittus RS, et al. Delirium in older emergency department patients: recognition, risk factors, and psychomotor subtypes. Acad Emerg Med. 2009;16:193–200. doi: 10.1111/j.1553-2712.2008.00339.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Agache I, Akdis CA. Precision medicine and phenotypes, endotypes, genotypes, regiotypes, and theratypes of allergic diseases. J Clin Invest. 2019;130:1493–1503. doi: 10.1172/JCI124611. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Saligan LN. Collaborative framework to advance symptom science: an intramural perspective. J Nurs Scholarsh. 2019;51:17–25. doi: 10.1111/jnu.12445. [DOI] [PubMed] [Google Scholar]
- 20.MacRae CA, Vasan RS. The future of genetics and genomics: closing the phenotype gap in precision medicine. Circulation. 2016;133:2634–2639. doi: 10.1161/CIRCULATIONAHA.116.022547. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Khan BA, Perkins AJ, Campbell NL, Gao S, Farber MO, Wang S, et al. Pharmacological management of delirium in the intensive care unit: a randomized pragmatic clinical trial. J Am Geriatr Soc. 2019;67:1057–1065. doi: 10.1111/jgs.15781. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Campbell NL, Perkins AJ, Khan BA, Gao S, Farber MO, Khan S, et al. Deprescribing in the pharmacologic management of delirium: a randomized trial in the intensive care unit. J Am Geriatr Soc. 2019;67:695–702. doi: 10.1111/jgs.15751. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Campbell NL, Khan BA, Farber M, Campbell T, Perkins AJ, Hui SL, et al. Improving delirium care in the intensive care unit: the design of a pragmatic study. Trials. 2011;12:139. doi: 10.1186/1745-6215-12-139. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Sessler CN, Gosnell MS, Grap MJ, Brophy GM, O’Neal PV, Keane KA, et al. The Richmond Agitation-Sedation Scale: validity and reliability in adult intensive care unit patients. Am J Respir Crit Care Med. 2002;166:1338–1344. doi: 10.1164/rccm.2107138. [DOI] [PubMed] [Google Scholar]
- 25.Khan BA, Guzman O, Campbell NL, Walroth T, Tricker JL, Hui SL, et al. Comparison and agreement between the Richmond Agitation-Sedation Scale and the Riker Sedation-Agitation Scale in evaluating patients’ eligibility for delirium assessment in the ICU. Chest. 2012;142:48–54. doi: 10.1378/chest.11-2100. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Tomasi CD, Grandi C, Salluh J, Soares M, Giombelli VR, Cascaes S, et al. Comparison of CAM-ICU and ICDSC for the detection of delirium in critically ill patients focusing on relevant clinical outcomes. J Crit Care. 2012;27:212–217. doi: 10.1016/j.jcrc.2011.05.015. [DOI] [PubMed] [Google Scholar]
- 27.Guenther U, Popp J, Koecher L, Muders T, Wrigge H, Ely EW, et al. Validity and reliability of the CAM-ICU Flowsheet to diagnose delirium in surgical ICU patients. J Crit Care. 2010;25:144–151. doi: 10.1016/j.jcrc.2009.08.005. [DOI] [PubMed] [Google Scholar]
- 28.Ely EW, Inouye SK, Bernard GR, Gordon S, Francis J, May L, et al. Delirium in mechanically ventilated patients: validity and reliability of the confusion assessment method for the intensive care unit (CAM-ICU) JAMA. 2001;286:2703–2710. doi: 10.1001/jama.286.21.2703. [DOI] [PubMed] [Google Scholar]
- 29.Ely EW, Margolin R, Francis J, May L, Truman B, Dittus R, et al. Evaluation of delirium in critically ill patients: validation of the Confusion Assessment Method for the intensive care unit (CAM-ICU) Crit Care Med. 2001;29:1370–1379. doi: 10.1097/00003246-200107000-00012. [DOI] [PubMed] [Google Scholar]
- 30.Marra A, Pandharipande PP, Shotwell MS, Chandrasekhar R, Girard TD, Shintani AK, et al. Acute brain dysfunction: development and validation of a daily prediction model. Chest. 2018;154:293–301. doi: 10.1016/j.chest.2018.03.013. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Reddy DR, Singh TD, Guru PK, Sakusic A, Gajic O, O’Horo JC, et al. Identification of acute brain failure using electronic medical records. J Crit Care. 2016;34:12–16. doi: 10.1016/j.jcrc.2016.03.008. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Wang S, Sigua NL, Manchanda S, Gradney S, Khan SH, Perkins A, et al. Preoperative STOP-BANG scores and postoperative delirium and coma in thoracic surgery patients. Ann Thorac Surg. 2018;106:966–972. doi: 10.1016/j.athoracsur.2018.05.089. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Matsuishi Y, Shimojo N, Unoki T, Sakuramoto H, Tokunaga C, Yoshino Y, et al. Type D personality is a predictor of prolonged acute brain dysfunction (delirium/coma) after cardiovascular surgery. BMC Psychol. 2019;7:27. doi: 10.1186/s40359-019-0303-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Gill TM, Gahbauer EA, Han L, Allore HG. Trajectories of disability in the last year of life. N Engl J Med. 2010;362:1173–1180. doi: 10.1056/NEJMoa0909087. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Nagin D. Group-based modeling of development. Cambridge, MA: Harvard University Press; 2005. [Google Scholar]
- 36.Nagin DS, Odgers CL. Group-based trajectory modeling (nearly) two decades later. J Quant Criminol. 2010;26:445–453. doi: 10.1007/s10940-010-9113-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Jones BL, Nagin DS. Advances in group-based trajectory modeling and an SAS procedure for estimating them. Sociol Methods Res. 2007;35:542–571. [Google Scholar]
- 38.Jones BL, Nagin DS, Roeder K. A SAS procedure based on mixture models for estimating developmental trajectories. Sociol Methods Res. 2001;29:374–393. [Google Scholar]
- 39.Lindroth H, Bratzke L, Twadell S, Rowley P, Kildow J, Danner M, et al. Predicting postoperative delirium severity in older adults: the role of surgical risk and executive function. Int J Geriatr Psychiatry. 2019;34:1018–1028. doi: 10.1002/gps.5104. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Tibshirani R. The lasso method for variable selection in the Cox model. Stat Med. 1997;16:385–395. doi: 10.1002/(sici)1097-0258(19970228)16:4<385::aid-sim380>3.0.co;2-3. [DOI] [PubMed] [Google Scholar]
- 41.Tibshirani R, Bien J, Friedman J, Hastie T, Simon N, Taylor J, et al. Strong rules for discarding predictors in lasso-type problems. J R Stat Soc Series B Stat Methodol. 2012;74:245–266. doi: 10.1111/j.1467-9868.2011.01004.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42.Armstrong RA. When to use the Bonferroni correction. Ophthalmic Physiol Opt. 2014;34:502–508. doi: 10.1111/opo.12131. [DOI] [PubMed] [Google Scholar]
- 43.McHugh ML. The chi-square test of independence. Biochem Med (Zagreb) 2013;23:143–149. doi: 10.11613/BM.2013.018. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44.Enders CK. Applied missing data analysis. New York, NY: Guilford Press; 2010. [Google Scholar]
- 45.Peterson JF, Pun BT, Dittus RS, Thomason JWW, Jackson JC, Shintani AK, et al. Delirium and its motoric subtypes: a study of 614 critically ill patients. J Am Geriatr Soc. 2006;54:479–484. doi: 10.1111/j.1532-5415.2005.00621.x. [DOI] [PubMed] [Google Scholar]
- 46.Flaherty JH, Yue J, Rudolph JL. Dissecting delirium: phenotypes, consequences, screening, diagnosis, prevention, treatment, and program implementation. Clin Geriatr Med. 2017;33:393–413. doi: 10.1016/j.cger.2017.03.004. [DOI] [PubMed] [Google Scholar]
- 47.Kim SY, Kim JM, Kim SW, Kim ES, Kang HJ, Lee JY, et al. Do the phenotypes of symptom fluctuation differ among motor subtypes in patients with delirium? J Pain Symptom Manage. 2018;56:667–677. doi: 10.1016/j.jpainsymman.2018.07.022. [DOI] [PubMed] [Google Scholar]
- 48.Sylvestre MP, McCusker J, Cole M, Regeasse A, Belzile E, Abrahamowicz M. Classification of patterns of delirium severity scores over time in an elderly population. Int Psychogeriatr. 2006;18:667–680. doi: 10.1017/S1041610206003334. [DOI] [PubMed] [Google Scholar]
- 49.Jones RN, Marcantonio ER, Saczynski JS, Tommet D, Gross AL, Travison TG, et al. Preoperative cognitive performance dominates risk for delirium among older adults. J Geriatr Psychiatry Neurol. 2016;29:320–327. doi: 10.1177/0891988716666380. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 50.Van Rompaey B, Schuurmans MJ, Shortridge-Baggett LM, Truijen S, Bossaert L. Risk factors for intensive care delirium: a systematic review. Intensive Crit Care Nurs. 2008;24:98–107. doi: 10.1016/j.iccn.2007.08.005. [DOI] [PubMed] [Google Scholar]
- 51.Lindroth H, Bratzke L, Purvis S, Brown R, Coburn M, Mrkobrada M, et al. Systematic review of prediction models for delirium in the older adult inpatient. BMJ Open. 2018;8:e019223. doi: 10.1136/bmjopen-2017-019223. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 52.Sanders RD. Hypothesis for the pathophysiology of delirium: role of baseline brain network connectivity and changes in inhibitory tone. Med Hypotheses. 2011;77:140–143. doi: 10.1016/j.mehy.2011.03.048. [DOI] [PubMed] [Google Scholar]
- 53.Hum S, Lapierre Y, Scott SC, Duquette P, Mayo NE. Trajectory of MS disease course for men and women over three eras. Mult Scler. 2017;23:534–545. doi: 10.1177/1352458516655478. [DOI] [PubMed] [Google Scholar]
- 54.Kheirbek RE, Alemi F, Citron BA, Afaq MA, Wu H, Fletcher RD. Trajectory of illness for patients with congestive heart failure. J Palliat Med. 2013;16:478–484. doi: 10.1089/jpm.2012.0510. [DOI] [PubMed] [Google Scholar]
- 55.Lundgren JG, Dahlström Ö, Andersson G, Jaarsma T, Kärner Köhler A, Johansson P. The effect of guided web-based cognitive behavioral therapy on patients with depressive symptoms and heart failure: a pilot randomized controlled trial. J Med Internet Res. 2016;18:e194. doi: 10.2196/jmir.5556. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 56.Schwehr NA, Kuntz KM, Enns EA, Shippee ND, Kingwell E, Tremlett H, et al. BeAMS Study group. Informing medication discontinuation decisions among older adults with relapsing-onset multiple sclerosis. Drugs Aging. 2020;37:225–235. doi: 10.1007/s40266-019-00741-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 57.Bogosian A, Morgan M, Bishop FL, Day F, Moss-Morris R. Adjustment modes in the trajectory of progressive multiple sclerosis: a qualitative study and conceptual model. Psychol Health. 2017;32:343–360. doi: 10.1080/08870446.2016.1268691. [DOI] [PubMed] [Google Scholar]