The Adult Respiratory Distress Syndrome Cognitive Outcomes Study

Abstract

Rationale: Cognitive and psychiatric morbidity is common and potentially modifiable after acute lung injury (ALI). However, practical measures of neuropsychological function for use in multicenter trials are lacking.

Objectives: To determine whether a validated telephone-based neuropsychological test battery is feasible in a multicenter trial. To determine the frequency and risk factors for long-term neuropsychological impairment.

Methods: As an adjunct study to the Acute Respiratory Distress Syndrome Clinical Trials Network Fluid and Catheter Treatment Trial, we assessed neuropsychological function at 2 and 12 months post–hospital discharge.

Measurements and Main Results: Of 406 eligible survivors, we approached 261 to participate and 213 consented. We tested 122 subjects at least once, including 102 subjects at 12 months. Memory, verbal fluency, and executive function were impaired in 13% (12 of 92), 16% (15 of 96), and 49% (37 of 76) of long-term survivors. Long-term cognitive impairment was present in 41 of the 75 (55%) survivors who completed cognitive testing. Depression, post-traumatic stress disorder, or anxiety was present in 36% (37 of 102), 39% (40 of 102), and 62% (63 of 102) of long-term survivors. Enrollment in a conservative fluid-management strategy (P = 0.005) was associated with cognitive impairment and lower partial pressure of arterial oxygen during the trial was associated with cognitive (P = 0.02) and psychiatric impairment (P = 0.02).

Conclusions: Neuropsychological function can be assessed by telephone in a multicenter trial. Long-term neuropsychological impairment is common in survivors of ALI. Hypoxemia is a risk factor for long-term neuropsychological impairment. Fluid management strategy is a potential risk factor for long-term cognitive impairment; however, given the select population studied and an unclear mechanism, this finding requires confirmation.

At a Glance Commentary

Scientific Knowledge on the Subject

Cognitive and psychiatric morbidities are prevalent, long-lasting, and potentially modifiable in survivors of critical illness. Whether neuropsychological function can be assessed in a multicenter trial is unknown. Furthermore, the frequency and risk factors for long-term neuropsychological impairment in survivors of acute lung injury have not been studied in a multicenter trial.

What This Study Adds to the Field

We found that neuropsychological function in survivors from a multicenter trial could be assessed using a validated telephone battery. Our findings show that most survivors of acute lung injury experienced long-term cognitive and psychiatric morbidity, and we detected that hypoxemia is a risk factor for the development of long-term cognitive and psychiatric impairment.

In the United States, approximately 200,000 patients develop acute lung injury (ALI) each year (1). Advances in ALI management have decreased mortality to 25–40% (1–4), resulting in an expanding population of survivors who have been ravaged by their acute illness. Survivorship, it has been put forth, will be the defining challenge of modern day critical care (5). Beyond the physical ailments that survivors endure, neuropsychological impairment is increasingly recognized as a prevalent, important, and potentially modifiable outcome among survivors of ALI (6–21) and critical illness in general (22–29).

Traditionally, neuropsychological (cognitive and psychiatric) function is assessed in-person by an expert, a constraint that has limited the ability to study the frequency and determinants of neuropsychological impairment in critically ill populations and to measure the effects of an intervention on long-term neuropsychological function. Neuropsychological function has not been studied in survivors from a multicenter trial. Consequently, the generalizability of observations from single-center cohort studies is unknown. Furthermore, evidence regarding the etiology of neuropsychological impairment in survivors of ALI is extremely limited.

As a practical approach to assess the neuropsychological function of survivors from a multicenter trial, we developed a telephone battery of standardized neuropsychological tests which could be administered by a nonexpert (10). We then validated the telephone battery against standard in-person assessments (10, 30) and in survivors of ALI specifically (10, 11). Here, we administered the battery to a subset of survivors from the Acute Respiratory Distress Syndrome Clinical Trials Network (ARDSNet) Fluid and Catheter Treatment Trial (FACTT) (4, 31). FACTT tested the hypothesis that a conservative fluid-management strategy targeted to lower intravascular pressures as measured by either a pulmonary artery catheter (PAC) or central venous catheter improves outcomes in ALI. Our goals were to determine whether a telephone-based neuropsychological test battery was feasible in a multicenter trial, to determine the frequency of long-term neuropsychological morbidity in survivors from a multicenter trial, and to identify potential risk factors for the development of long-term cognitive and psychiatric impairment. Results of this study were previously reported in abstract form (32).

Methods

An expanded Methods section is available in the online supplement. The study was approved by the institutional review board of participating hospitals and as a substudy of FACTT by the ARDSNet.

Study Design

The ARDS Cognitive Outcomes Study (ACOS) is a prospective, multicenter cohort study of a subset of survivors from FACTT. FACTT enrolled patients between June 2000 and October 2005. Between July 2002 and July 2003, FACTT was halted and new regulatory approval related to the study was prohibited. The regulatory process for ACOS began before the halt and testing was conducted between March 2003 and September 2006 in concert with the Economic Analysis of the Pulmonary-Artery Catheter (EA-PAC) study, a long-term follow-up study of FACTT (33).

To be eligible for ACOS, subjects had to be enrolled in FACTT and EA-PAC (33) and ACOS-specific regulatory approval had to be in place. FACTT enrolled mechanically ventilated adult patients with ALI (34). Subjects not consented before discharge for ACOS were recruited by the telephone and through written communication when telephone contact failed.

Neuropsychological Function and Quality of Life

We used a validated telephone battery of standardized neuropsychological tests (10, 11, 30). The telephone battery was administered from the University of Pittsburgh to consenting, English-speaking subjects at 2 and 12 months post–hospital discharge. Subjects were not required to undergo 2 month testing to be tested at 12 months.

The instruments used to assess cognition, anxiety, depression, and post-traumatic stress disorder (PTSD) symptoms are presented in Table E1 in the online supplement (35–42). Quality of life (43) was assessed at 12 months as part of EA-PAC (33). We focused our study on survivors tested at 12 months to assess long-term neuropsychological function.

Data Analysis

The cognitive battery yielded scaled scores for each domain that were normalized to allow for comparisons across tests and subjects (44, 45). We report the median and interquartile range of results as percentiles. We defined impairment in a single domain as a score greater than 2 SD below the population normative data (6, 7, 9, 44–46). Cognitive impairment at the subject level was defined as impairment in memory, verbal fluency, or executive function in subjects who completed tests in each of these domains (6, 9–11, 27).

We a priori hypothesized that long-term cognitive impairment would be associated with duration of mechanical ventilation, either conservative or liberal fluid-management strategy (4), hypotension (9), hypoxemia (6), PAC use (19, 47, 48), sepsis as the cause of ALI (49, 50), and severity of illness (26). The prespecified potential confounding variables (6–12, 51–53) are detailed in the online supplement.

Psychiatric impairment was defined as impairment in any of the three psychiatric measures (anxiety, depression, and PTSD) in subjects completing each measure. We tested the following candidate risk factors for the development of long-term psychiatric morbidity: age (18, 27, 28); sex (18, 27–29); race and ethnicity; level of education (19, 27); Acute Physiologic and Chronic Health Evaluation III scores; trial interventions (fluid-management strategy, PAC); hypotension; hypoxemia (18, 27); any episode of hypoglycemia (glucose <60 mg/dl) during the hospitalization (20, 27); corticosteroid administration (54); ICU length of stay (17, 21, 27); and duration of mechanical ventilation (17, 18, 27).

Statistical Analysis

Multivariable logistic regression was used to investigate the relationship between candidate risk factors and long-term neuropsychological (cognitive or psychiatric) impairment. We adjusted for each candidate risk factor and potential covariates with an α level of significance less than 0.20 in univariate analyses (55). To avoid over-fitting, adjustment was performed one covariate at a time (56). A P value of less than or equal to 0.05 was used to signify statistical significance.

The cognitive impairment risk factor analysis was limited to 75 survivors who completed memory, verbal fluency, and executive function testing at 12 months. The Hayling Sentence Completion Test (HSCT) assesses executive function with an error score and response latency component to produce an overall score (35, 36). The HSCT response times were not measured before February 2005 because of an error in the administration of the timing of the test. The timing error resulted in 21 survivors tested at 12 months having incomplete testing. Secondary analyses were performed in 90 survivors who completed assessments in memory, verbal fluency, and the error score of the HSCT at 12 months.

We tested for associations between identified risk factors and memory, verbal fluency, and executive function scores as continuous variables using Spearman rank-correlation coefficients. Finally, we performed sensitivity analyses to determine the effects of missing data on the observed associations. Statistical analyses were performed using Stata 10.0 software (Stata Datacorp, College Station, TX).

Results

Enrollment

Of 1,001 patients randomized in FACTT, 406 survivors were eligible for ACOS (Figure 1) (see Figure E1 for enrollment by fluid-management strategy). Of 406 eligible survivors, we approached 261 to participate and 213 consented. Consent rates did not differ by age, sex, race and ethnicity, or geographic region after adjusting for mortality and excluding ineligible subjects in whom the time window to be tested had elapsed (see Table E2). Of 213 consenting subjects, 14 died before initial testing, and we were able to contact and test 122 of 199 alive and consenting subjects (61%) at least once. The 122 subjects were drawn from 28 hospitals and constitute the ACOS cohort.

Figure 1.

Enrollment and outcomes. ACOS = Adult Respiratory Distress Syndrome Cognitive Outcomes Study. *Of 1,001 Fluid and Catheter Treatment Trial (FACTT) subjects who underwent randomization, 227 were ineligible for long-term follow-up as part of Economic Analysis of the Pulmonary-Artery Catheter (EA-PAC) because no regulatory approval was in place and 119 declined consent (33). ^†Subjects were categorized as ineligible if the time window to be tested had elapsed because of the regulatory halt, as incapable if self-determined or determined by a surrogate to be physically or mentally incapable of telephone-based neuropsychological testing, and as lost to follow-up if consent was obtained but the subject was not tested. Of 53 subjects categorized as lost to follow-up, no contact information was available for 1 subject, no telephone service was available for 1 subject, no explanation was provided for 1 subject, and the remaining 50 subjects were recorded as “lost to follow-up.” ^‡Of 52 subjects tested at 2 months, 32 were retested at 12 months. The reasons for 20 subjects being tested at 2 months but not at 12 months were that 11 patients were lost to follow-up, 5 declined, 3 died, and 1 was incarcerated. Subjects were not required to undergo testing at 2 months to be tested at 12 months.

Compared with non-ACOS FACTT participants who survived 60 days (n = 609), the ACOS group was more likely to be female, more likely to be non-Hispanic whites, less likely to have HIV-AIDS, and less likely to have pneumonia as the cause of ALI (Table 1). The number of ACOS subjects enrolled from each hospital ranged from 1 to 17 (median = 3; interquartile range [IQR], 1–5).

TABLE 1.

COMPARISON OF BASELINE CHARACTERISTICS BETWEEN FACTT 60-DAY SURVIVORS NOT ENROLLED IN ACOS AND ACOS PARTICIPANTS

Variable	FACTT Survivors Not Enrolled in ACOS (n = 609)	ACOS Cohort (n = 122)	P Value
Age, yr	47 (37–57)	49 (40–58)	0.23
Male sex, %	54	43	0.04
Race or ethnic group, %			<0.001
White non-Hispanic	64	86
Black non-Hispanic	21	11
Hispanic	7	3
Other	8	0
Primary lung injury, %			0.005
Pneumonia	49	36
Sepsis	20	25
Aspiration	16	16
Trauma	9	8
Multiple transfusions	1	2
Other	5	13
Coexisting conditions, %
None	71	75	0.39
Diabetes	16	18	0.65
HIV infection or AIDS	7	0	0.001
Cirrhosis	3	2	0.78
Solid tumors	1	1	1.00
Leukemia	2	1	1.00
Lymphoma	0	2	0.13
Immunosuppression	6	7	0.84
APACHE III score, mean (SD)	85 (66–105)	85 (63–102)	0.32
Medical intensive care unit	65	56	0.06
Glasgow Coma Scale	8 (6–12)	8 (4–11)	0.38
Mean arterial pressure, mm Hg	76 (68–87)	77 (67–85)	0.49
Vasopressor use, %	34	33	0.81
PaO2:FiO2	122 (83–168)	122 (86–165)	0.73
Conservative strategy, %	51	55	0.38
Pulmonary artery catheter, %	51	52	0.70

Definition of abbreviations: ACOS = Adult Respiratory Distress Syndrome Cognitive Outcomes Study; APACHE III = Acute Physiologic and Chronic Health Evaluation III scores; FACTT = Fluid and Catheter Treatment Trial.

Values expressed as a percentage or median (interquartile range).

By 12 months, 22 consenting subjects were deceased, 2 were incarcerated, and 9 were deemed physically or mentally incapable of telephone-based neuropsychological testing. Of the remaining 180 subjects who were alive and able to be tested, 102 (57%) were tested at 12 months. These 102 long-term survivors completed individual cognitive domain testing to various degrees (see Table E1), and 74% (75 of 102) completed testing in all cognitive domains. Long-term survivors were tested on average 12 months post–hospital discharge (IQR, 11–13 mo). None of these long-term survivors had a history of preexisting dementia.

Neuropsychological Test Battery Administration

The test battery required 45–60 minutes to complete. During the first several tests, subject fatigue was noted to be significant when the battery was administered during the same telephone call as the EAPAC questionnaire, which required 15–20 minutes (33). Subsequently, the battery was administered during a separate call. None of the initial tests were included in the primary analyses of long-term cognitive impairment.

Using the separate testing session, survivors occasionally expressed frustration and requested to end testing. Despite encouragement, explanation of the reasons for the tests, and reassurance that the tests were meant to be challenging, eight long-term survivors declined to complete cognitive testing.

Baseline Characteristics and Effects of Fluid-Management Strategy

Acute Physiologic and Chronic Health Evaluation III score was higher (P = 0.04) in the liberal-fluid strategy group (see Table E3). No other baseline characteristic differed between fluid-management strategies. The ACOS patients in the conservative-strategy group received more furosemide (P < 0.001), resulting in a more negative 7-day cumulative fluid balance (P = 0.02), lower central venous pressures (P < 0.001), and a trend toward lower cardiac index (P = 0.06) (see Table E4). As in the larger FACTT study, in ACOS the conservative fluid-management strategy resulted in shorter duration of mechanical ventilation (median difference of 2.5 d; P = 0.04) and a trend to shorter ICU stays (median difference of 3 ICU d; P = 0.07).

Long-Term Cognitive Impairment and Psychiatric Morbidity

Cognitive function was below normal population mean scores in each domain tested (see Table E1). Cognitive impairment was observed in 41 of 75 survivors (55%) at 12 months. Using the error score of the HSCT to assess executive function rather than the overall score, cognitive impairment was observed in 54 of 90 survivors (60%).

Vocabulary and reasoning, domains hypothesized to be resilient to acquired brain injury, were impaired in 3 of 98 survivors (3%). Among the domains hypothesized to be susceptible to impairment after ALI, memory was impaired in 12 of 92 (13%), verbal fluency in 15 of 96 (16%), and executive function in 37 of 76 (49%) survivors at 12 months. Executive function was impaired in 57 of 100 (57%) survivors when assessed using the error score of the HSCT.

Symptoms of moderate or severe depression occurred in 37 of 102 (36%) long-term survivors. PTSD screening was positive in 40 of 102 (39%) survivors. Symptoms of moderate or severe anxiety occurred in 63 of 102 (62%) survivors. Psychiatric symptoms were present in depression, anxiety, or PTSD in 67 of 102 (66%) survivors. Symptoms in two or more psychiatric domains occurred in 43 of 102 (42%) survivors. Cognitive impairment was significantly associated with the presence of psychiatric symptoms (P = 0.04) at 12 months. Specifically, cognitive impairment was associated with anxiety (P = 0.04), but not depression (P = 0.20) or PTSD (P = 0.33).

Long-Term Cognitive Impairment, Psychiatric Morbidity, and Quality of Life

Quality of life was low in ACOS survivors (median utility = 0.67; IQR, 0.37, 0.90) at 12 months. There were no significant differences in quality of life between cognitively impaired and nonimpaired survivors (median utility in impaired = 0.66; IQR, 0.45, 0.9; nonimpaired = 0.69; IQR, 0.42, 0.82; P = 0.85). In survivors with psychiatric symptoms, quality of life was significantly worse (median utility in 59 survivors with psychiatric symptoms = 0.57; IQR, 0.26, 0.76; utility in 35 survivors without symptoms = 0.90; IQR, 0.53, 0.95; P < 0.001).

Risk Factors for Long-Term Cognitive Impairment

Lower Pa_O2 during FACTT was associated with cognitive impairment at 12 months (P = 0.02), as were enrollment in the conservative fluid-management strategy (P = 0.004) and lower central venous pressure (P = 0.04) (Table 2, Figure 2; see Table E5). As detailed in the online supplement, lower Pa_O2 values (P = 0.05), lower central venous pressures (P = 0.02), and enrollment in the conservative fluid-strategy group (P < 0.001) correlated with worse executive function. After adjustment for potential covariates, lower Pa_O2 and enrollment in the conservative fluid-management strategy were associated independently with cognitive impairment at 12 months (Table 2; see Table E6).

TABLE 2.

ASSOCIATIONS WITH LONG-TERM COGNITIVE IMPAIRMENT IN SURVIVORS OF ALI

	Long-Term Cognitive Impairment
	Not Impaired (n = 34)	Impaired (n = 41)	P Value	Unadjusted Odds Ratio (95% CI)	Adjusted Odds Ratio (95% CI)*
Candidate risk factors
Duration of mechanical ventilation	8 (4–11)	6 (4–9)	0.43
Conservative fluid-management strategy, %	32	66	0.004	4.03 (1.53–10.59)	3.35 (1.16–9.70)–5.46 (1.92–15.53)
Hypotension (hemodynamic data on-study)†
Systolic blood pressure (mm Hg)	108 (102–113)	104 (96–112)	0.32
Cardiac index (L/min/m2)‡	4.5 (3.8–5.3)	4.5 (3.7–5)	0.49
Shock, %†	32	29	0.77
Vasopressor use	26	24	0.84
Hypoxemia (respiratory variables on-study)
PaO2	86 (70–98)	71 (67–80)	0.02	1.56 (1.09–2.24)§	1.51 (1.01–2.26)–1.68 (1.14–2.49)§
PaO2:FiO2	152 (132–192)	157 (133–190)	0.63
Oxygenation index	7.38 (4.55–10.42)	7.67 (5.97–10.02)	0.57
Oxygen saturation (%)	95.1 (93.3–96.8)	94.2 (92.6–95.8)	0.10
Pulmonary artery catheter, %	53	61	0.48
Primary lung injury, %			0.10
Pneumonia	24	46
Sepsis	29	20
Aspiration	9	17
Trauma	12	10
Multiple transfusions	6	2
Other	21	5
Severity of illness
APACHE III	87 (60–106)	72 (60–94)	0.42
ICU length of stay	11 (8–18)	10 (7–16)	0.65
Potential covariates
Age, yr	50 (43–58)	51 (45–60)	0.51
Male sex, %	32	51	0.10
Race or ethnic group, %			0.53
White	91	90
Black	6	10
Hispanic	3	0
Level of education, yr	13 (12–16)	12 (12–15)	0.36
Coexisting conditions, %
Heavy alcohol use	3	13	0.21
Cerebrovascular disease	0	2	1.00
Psychiatric impairment, 1 yr, %	59	80	0.04
Hospital length of stay before enrollment, d	2 (1–3)	3 (1–4)	0.22
Time from discharge to testing, mo	12 (11–13)	12 (11–13)	0.51

Definition of abbreviations: ALI = acute lung injury; APACHE III = Acute Physiologic and Chronic Health Evaluation III scores; CI = confidence interval; CVP = central venous pressure; FACTT = Fluid and Catheter Treatment Trial; ICU = intensive care unit.

Cognitive impairment at the subject level was defined as impairment in memory, verbal fluency, or executive function in the 75 subjects who completed tests in each of these domains at the 12-month follow-up. Values expressed as a frequency (percent) or median (interquartile range).

On-study variables were summarized at the subject level as means to account for influential observations. With the exception of systolic blood pressure, in which the worst values over the preceding 24 hours were recorded to derive the cardiovascular component of the Brussels organ failure score, values were measured and recorded daily (measure closest to 8:00 a.m.) during the study.

^*Potential covariates were included in the multivariable logistic regression models one covariate at a time. The adjusted odds ratios presented represent the range of odds ratios observed in multivariable logistic regression models. For additional details, see Table E6.

^†Shock defined as mean arterial pressure less than 60 or vasoactive agent use (assignment to FACTT protocol cells 1 or 2) at any time during FACTT (4).

^‡Cardiac index was measured in 39 survivors (20 in the liberal-strategy group and 19 in the conservative-strategy group).

^§For each 10-unit decrease in Pa_O2, odds ratio for the development of cognitive impairment increased by 1.56.

Figure 2.

Long-term cognitive impairment according to fluid management strategy. Impairment in a domain was defined as a score greater than 2 SD below the population norm. Cognitive impairment was defined as impairment in memory, verbal fluency, or executive function in the 75 survivors completing testing in each of these cognitive domains. Memory, verbal fluency, or executive function was assessed in 92, 96, and 76 subjects, respectively. Proportions of impaired subjects are specified by domain, according to fluid-management strategy. Error bars reflect upper bound of 95% confidence interval. Cons = conservative fluid strategy; Lib = liberal fluid strategy.

In secondary analyses, using the untimed error score of the HSCT for the executive function assessment, the conservative fluid-management strategy remained associated with cognitive impairment at 12 months (71% [34 of 48] vs. 48% [20 of 42] in the liberal group; P = 0.02). The association between Pa_O2 values and cognitive impairment was no longer significant in these analyses (P = 0.32). As detailed in the supplement, we performed multiple sensitivity analyses to determine the effects of missing data and incomplete cognitive testing. These analyses may suggest that a relationship exists between hypoxemia and fluid-management strategy and long-term cognitive impairment.

Risk Factors for Long-Term Psychiatric Morbidity

Survivors with psychiatric morbidity at 12 months had lower Pa_O2 values (P = 0.02), lower systolic blood pressures (P = 0.05), and were more likely to have had an episode of hypoglycemia (P = 0.03) (Table 3). As detailed in the online supplement, lower Pa_O2 values (P = 0.05), lower systolic blood pressures (P = 0.04), and an episode of hypoglycemia (P = 0.01) were associated with anxiety. After adjustment for potential covariates, lower Pa_O2 was associated independently with long-term psychiatric morbidity.

TABLE 3.

ASSOCIATIONS WITH LONG-TERM PSYCHIATRIC MORBIDITY IN SURVIVORS OF ALI

	Psychiatric Morbidity
	Not Present (n = 35)	Present (n = 67)	P Value
Baseline characteristics
Age, yr	51 (38–64)	50 (44–57)	0.58
Male sex, %	40	45	0.64
Race or ethnic group, %			0.19
White	94	81
Black	6	15
Hispanic	0	4
Level of education, yr	14 (12–16)	12 (12–15)	0.15
Medical ICU, %	49	58	0.44
APACHE III	88 (63–108)	77 (60–99)	0.31
Conservative strategy, %	46	60	0.18
Pulmonary artery catheter, %	57	52	0.64
On-study variables*
Hemodynamic variables†
Systolic blood pressure (mm Hg)	111 (102–116)	104 (98–111)	0.05
Shock, %	37	30	0.46
Respiratory variables
PaO2‡	86 (72–98)	72 (68–90)	0.02
Metabolic variables
Glucose, mg/dl	123 (107–140)	119 (109–136)	0.99
Hypoglycemia‡, <60 mg/dl, %	0	13	0.03
Hyperglycemia‡, >180 mg/dl, %	34	30	0.65
Corticosteroid therapy§
Corticosteroids (mg of methylprednisolone, cumulative)	0 (0–60)	0 (0–140)	0.58
Corticosteroids received, %	29	33	0.66
ICU length of stay (d)	11 (7–20)	12 (8–16)	0.92
Duration of mechanical ventilation (d)	7 (4–15)	7 (4–10)	0.94

Definition of abbreviations: ALI = acute lung injury; APACHE III = Acute Physiologic and Chronic Health Evaluation III scores; FACTT = Fluid and Catheter Treatment Trial; ICU = intensive care unit.

Values expressed as a frequency (percent) or median (interquartile range).

Psychiatric morbidity was defined as present if symptoms of anxiety, depression, or post-traumatic stress disorder were identified at the 12-month follow-up.

^*On-study variables were summarized at the subject level as means to account for influential observations. With the exception of systolic blood pressure, in which the worst values over the preceding 24 hours were recorded to derive the cardiovascular component of the Brussels organ failure score, values were measured and recorded daily (measure closest to 8:00 a.m.) during the study.

^†Shock defined as mean arterial pressure less than 60 or vasoactive agent use (assignment to FACTT protocol cells 1 or 2) at any time during FACTT (4). The association between systolic blood pressure and psychiatric impairment was nonsignificant after adjustment for potential covariates (P = 0.14).

^‡For each 10-unit decrease in Pa_O2, odds ratio for the development of psychiatric morbidity increased by 1.35 (95% confidence interval, 1.02–1.78) and the association remained after adjustment for potential covariates. Hypoglycemia and hyperglycemia were defined as any episode of serum glucose less than 60 mg/dl or greater than 180 mg/dl, respectively, during the hospitalization. We were unable to adjust for potential covariates in the association between hypoglycemia and psychiatric morbidity because psychiatric symptoms were present in each subject who had a hypoglycemia episode.

^§Corticosteroid therapy was neither associated with PTSD (11 of 32 subjects [34%] who received corticosteroids experienced symptoms of PTSD compared with 29 of 70 subjects [41%] who did not receive corticosteroids; P = 0.50), nor associated with psychiatric morbidity in the subgroup of 27 sepsis-associated subjects with ALI (6 of 12 subjects [50%] who received corticosteroids experienced long-term psychiatric morbidity compared with 11 of 15 subjects [73%] who did not receive corticosteroids; P = 0.26).

Discussion

We found that neuropsychological function in survivors from a multicenter trial can be assessed using a validated telephone battery. Similar to previous investigations, we found most survivors of ALI experience long-term cognitive and psychiatric morbidity. Furthermore, our findings demonstrate a relationship between cognitive impairment, psychiatric symptoms, and adverse quality of life in long-term survivors. Finally, we confirmed prior associations that hypoxemia is a potential risk factor for the development of long-term cognitive and psychiatric impairment.

We found that neuropsychological function can be assessed over the telephone in a multicenter trial. This has the potential to be an important development for future critical care trials, because neuropsychological impairment is increasingly recognized as a common, long-lasting, and potentially modifiable outcome. However, our experience highlights potential limitations of measuring neuropsychological function in multicenter trials.

First, attrition was significant. Attrition is a recognized barrier to conducting long-term outcomes studies in critical care trials (24, 25, 57–59), caused in part by the mortality experienced by critically ill patients. After accounting for mortal losses, the requirement that subjects be enrolled in EA-PAC, and the delay in study initiation, only 406 of the 1,001 subjects randomized in FACTT were alive and eligible for ACOS.

However, because 12% of eligible survivors declined and 36% were unable to be contacted to obtain consent for ACOS after discharge, the opportunity to increase subject participation in future studies exists. Potential solutions to increase consent rate in future studies include enrolling subjects prospectively and prioritizing in-person consent before discharge, and using multiple strategies and repeated attempts to contact potential subjects (60, 61). Furthermore, because 12% of subjects who consented to long-term neuropsychological assessment subsequently declined and 25% were lost to follow-up, the opportunity to improve retention also exists. Although it is unclear why subjects declined to be tested, potential explanations include that subjects were asked to complete multiple follow-up interviews as part of EA-PAC and ACOS and neuropsychological assessments were lengthy. Finally, subjects completed cognitive testing by the telephone to varying degrees. Fatigue or frustration appeared to play a role, as did an error in test administration. Future studies need to balance the merits of administering a more comprehensive battery with the potential burden imposed on study participants.

Potential solutions to improve retention and test completion in future studies include increasing the flexibility of scheduling by centralizing test administration in different time-zones to minimize potential conflicts with work or healthcare visits; increasing the number of trained staff to administer the neuropsychological battery and identifying and using the most effective strategies to ensure test completion; selecting an intermediate testing time point (e.g., 3 mo), which minimizes mortal losses yet ensures adequate time for survivors to transition through the healthcare system to home (24, 25, 62, 63); using a shorter test battery (64) or limiting the domains to be assessed (62); and using multiple strategies in a systematic fashion to retain participants (60, 61). Through the effective use of many of these strategies, more recent studies are demonstrating the ability to recruit and retain survivors of ALI successfully (61, 65, 66). Finally, compared with nonparticipants, ACOS subjects were more likely to be female and white; as such, our findings may not generalize to other populations.

In conclusion, we found that we were able to assess long-term neuropsychological function in survivors from a multicenter trial using a telephone battery. However, we encountered significant challenges in regards to recruitment, retention, and test completion. We acknowledge that these challenges pose a potential threat to the internal and external validity of such an approach. To improve the performance of this practical approach in future studies, we have highlighted potential solutions to improve recruitment, retain study participants, and ensure complete neuropsychological assessments.

The subjects with ALI in our study had long-term cognitive and psychiatric morbidity consistent with prior studies where data were obtained in person (6–9, 24, 25). Coupled with our prior work (10, 11, 30), our findings validate that many survivors of ALI experience clinically important long-term cognitive impairment. Approximately half of our survivors of ALI had executive dysfunction. Although cognitive impairment was not associated with decreased quality of life in survivors of ALI enrolled in ACOS, the evidence supports that cognitive impairment has a considerable adverse impact on individuals and society (6–21, 67, 68).

Survivors of ALI enrolled in ACOS experienced anxiety, depression, and PTSD symptoms. The presence of long-term psychiatric symptoms, and symptoms of anxiety in particular, was found to be associated with lower Pa_O2 values and hypoglycemia during the hospitalization. Our findings validate the recent work of Hopkins and coworkers (18), wherein hypoxemia was associated with anxiety at 1 year in ARDS survivors; augment the recent work of Dowdy and coworkers (20), which found that hypoglycemia was associated with depression at 3 months; and support the notion that psychiatric symptoms may result from brain injury sustained during the hospitalization. We found that psychiatric morbidity was associated with cognitive impairment and significantly worse quality of life. Because cognitive impairment may lead to development of psychiatric symptoms (18), and psychiatric symptoms may lead to impaired physical function (66), our findings support the idea that cognitive impairment, psychiatric symptoms, and quality of life are interrelated and impact on each other in survivors of critical illness.

We found that hypoxemia is a potential risk factor for the development of long-term cognitive impairment. Consistent with the work of Hopkins and colleagues (6), we found lower Pa_O2 values were associated with cognitive impairment in general and executive dysfunction specifically. Enrollment in the conservative fluid-management strategy was identified as a potential risk factor for the development of long-term cognitive impairment and executive dysfunction. To potentially explain this finding, we demonstrated that within the ACOS population lower central venous pressures, the explicit target of the conservative fluid-management strategy, were associated with cognitive impairment and executive dysfunction. However, there was no indirect evidence for reduced cerebral perfusion (e.g., cardiac index, systolic blood pressure) as the mediator for the observed association, thus it is unclear how conservative fluid management may have caused cognitive impairment. Given the highly selected population that enrolled in and completed ACOS testing, it is unclear if this finding generalizes to the entire FACTT cohort. This finding needs to be validated.

Our study sample size limits the generalizability of our findings and is a major limitation. We also acknowledge the potential for uncontrolled confounding. Specifically, we were unable to adjust for potential covariates simultaneously due to our sample size, we were unable to adjust for sedation or delirium (25, 69), and we did not screen for preexisting dementia formally (70) or preexisting psychiatric disease. There exists the potential for informative censoring, and although some subjects were determined to be incapable of telephone-based neuropsychological testing, we did not assess formally for hearing loss before testing. Although furosemide has been associated with hearing loss, there was no association between furosemide dose administered and cognitive impairment (see Table E5). We used conservative criteria to define cognitive impairment and limited the probability of a Type I error (46); however, our multiple comparisons could result in falsely rejecting the null hypothesis. If we used a Bonferroni correction factor, only the conservative fluid-management strategy would remain associated with cognitive impairment. Regardless, our risk factor assessment requires confirmation. Finally, depression, anxiety, and PTSD were identified using self-report measures, which may overdiagnose these disorders. These psychiatric instruments, although reliable and valid, are not the gold standard for clinical diagnosis.

In conclusion, we found that neuropsychological function can be assessed in a multicenter trial, but it was challenging and opportunities to improve the performance of this strategy exist. Our findings validate that cognitive and psychiatric morbidity are common in survivors of ALI, findings that have significant implications for these patients’ long-term functioning. Executive dysfunction, specifically, was a common morbidity. We found evidence that hypoxemia is a risk factor for long-term cognitive and psychiatric impairment and a signal that fluid management strategy is a potential risk factor for long-term cognitive impairment. Further studies are necessary to confirm this latter observation given the select population studied and an unclear mechanism. Future clinical trials in critically ill populations should include assessment of long-term outcomes, including neuropsychological outcomes. We need to be cautious in our interpretation of such studies and acknowledge that confirmatory studies may be required. Nevertheless, to improve the lives of critically ill survivors, we should assess the short- and long-term effects of an intervention.