Prediction of Long-term Cognitive Decline Following Postoperative Delirium in Older Adults

Abstract

Background:

Increasing evidence suggests that postoperative delirium may result in long-term cognitive decline among older adults. Risk factors for such cognitive decline are unknown.

Methods:

We studied 126 older participants without delirium or dementia upon entering the Successful AGing After Elective Surgery (SAGES) study, who developed postoperative delirium and completed repeated cognitive assessments (up to 36 months of follow-up). Pre-surgical factors were assessed preoperatively and divided into nine groupings of related factors (“domains”). Delirium was evaluated at baseline and daily during hospitalization using the Confusion Assessment Method diagnostic algorithm, and cognitive function was assessed using a neuropsychological battery and the Informant Questionnaire for Cognitive Decline in the Elderly (IQCODE) at baseline and 6-month intervals over 3 years. Linear regression was used to examine associations between potential risk factors and rate of long-term cognitive decline over time. A domain-specific and then overall selection method based on adjusted R² values was used to identify explanatory factors for the outcome.

Results:

The General Cognitive Performance (GCP) score (combining all neuropsychological test scores), IQCODE score, and living alone were significantly associated with long-term cognitive decline. GCP score explained the most variation in rate of cognitive decline (13%), and six additional factors—IQCODE score, cognitive independent activities of daily living impairment, living alone, cerebrovascular disease, Charlson comorbidity index score, and exhaustion level—in combination explained 32% of variation in this outcome.

Conclusions:

Global cognitive performance was most strongly associated with long-term cognitive decline following delirium. Pre-surgical factors may substantially predict this outcome.

Delirium, an acute confusional state, is a common condition affecting up to 50% of older adults in the hospital, with serious potential consequences including cognitive and functional impairment, institutionalization, and mortality (1). Previous studies have indicated that individuals with delirium following surgery often experience cognitive impairment, although few studies have assessed cognitive function beyond 6 to 12 months postoperatively (2); still, emerging evidence suggests that some individuals never return to their previous cognitive level (2), and exhibit accelerated cognitive decline over the longer term (3). In the SAGES study, we previously reported that individuals had linear trajectories of cognitive decline over postoperative months 2 to 36, and these trajectories were significantly steeper among participants who developed delirium compared to those who did not (3).

The etiology of delirium associated with long-term cognitive sequelae is unclear. Multiple risk factors have been identified for delirium (4), and validated prediction models have been developed to target high-risk individuals with successful intervention (5–7); however, risk factors specific for delirium associated with cognitive decline have not yet been explored. In the absence of delirium, several risk factors have been established for cognitive decline in older adults (8), but this outcome generally has been difficult to predict, and there may be similar challenges in identifying predictors of cognitive decline following delirium—likely a complex outcome. Thus, the goal of this paper is to demonstrate as a proof-of-concept that pre-surgical factors can predict the rate of long-term cognitive decline among older persons who developed postoperative delirium.

Methods

Study Sample

The SAGES study is an ongoing prospective cohort study of older adults undergoing elective major noncardiac surgery. The study design and methods have been described previously (9). Briefly, eligible participants were age 70 years and older, English speaking, scheduled to undergo elective surgery at one of two Harvard-affiliated academic medical centers and with an anticipated length of stay of at least 3 days. Eligible surgical procedures were: total hip or knee replacement, lumbar, cervical, or sacral laminectomy, lower extremity arterial bypass surgery, open abdominal aortic aneurysm repair, and colectomy. Exclusion criteria included evidence of dementia, delirium, hospitalization within 3 months, terminal condition, legal blindness, severe deafness, history of schizophrenia or psychosis, and history of alcohol abuse. A total of 566 patients were enrolled between June 18, 2010 and August 8, 2013, and six individuals were subsequently excluded due to possible dementia based on a detailed adjudication process described previously (9), leaving 560 cohort participants. Written informed consent was obtained from all participants according to procedures approved by the institutional review boards of Beth Israel Deaconess Medical Center and Brigham and Women’s Hospital, the two study hospitals, and Hebrew SeniorLife, the study coordinating center, all located in Boston, Massachusetts.

Assessment of Pre-surgical Factors

An initial home interview was conducted to gather detailed information on health and functioning approximately 2 weeks prior to hospitalization for scheduled surgery. Basic demographic, medical, and lifestyle information was obtained, and well-validated assessments of cognitive, physical, and mental function were administered (see Supplement A for details). The five components of the Fried frailty index (unintentional weight loss, exhaustion, low physical activity, low grip strength, and slow timed walk) were assessed (10), and a blood sample was collected from which apolipoprotein 4 genotype and C-reactive protein levels were analyzed. A trained physician reviewed medical records following hospital discharge, which provided additional information on pre-surgical factors.

Assessment of Delirium

Beginning with the first postoperative day, delirium symptoms were assessed daily during hospitalization using brief cognitive tests (9,11), the Delirium Symptom Interview (12), and acute changes in mental status reported by family and nurses. Delirium diagnosis was based on the Confusion Assessment Method diagnostic algorithm (13), a widely used and well-validated assessment tool with very high sensitivity, specificity, and inter-rater reliability (14,15). Delirium symptoms were also obtained from medical records using a validated delirium assessment (16,17) with adjudication of cases by an expert delirium panel. For this study, a combined approach based on either Confusion Assessment Method or medical record assessment of delirium (and used in prior studies) was utilized to identify incident delirium (17).

Assessment of Cognitive Function

Cognitive function was assessed at the home interview, and repeated at 1, 2, and 6 months after hospital discharge, and at 6-month intervals thereafter, up to 36 months following discharge. The neuropsychological battery evaluated cognitive domains putatively most affected by delirium (9,18) (see Supplement B for details). We combined all test scores into a weighted composite summary measure, the General Cognitive Performance (GCP) score, using previously published methods (19); this score was calibrated to a nationally representative sample of adults aged ≥70 years (with an expected mean of 50 and standard deviation [SD] of 10) (18). The GCP score is sensitive to change with minimal floor and ceiling effects (18,19) and has been applied in previous studies (3,20). For longitudinal GCP values, we applied a previously described method of correction to these scores to account for learning effects over time (21,22).

As previously reported in this cohort, the pattern of GCP scores over time was the following: cognitive decline at postoperative month 1, recovery of cognitive function above baseline at month 2, and gradual cognitive decline beginning at month 2 and dropping below baseline over months 12 to 36; this pattern was more pronounced among participants who developed postoperative delirium compared to those who did not (3) (see Supplementary Table 1 for details). Loss to follow-up was minimal due to death (6.6%) and withdrawal from the study (4.8%), and non-differential with respect to delirium status; therefore, 88.6% of participants had complete data on the GCP score over time (3).

Statistical Analysis

We divided pre-surgical factors into nine related groupings or domains: demographics (age, sex, race, education, and living arrangement), lifestyle factors (smoking status, alcohol intake, and socioeconomic status), cognitive function (baseline GCP score and Informant Questionnaire for Cognitive Decline in the Elderly [IQCODE] score), physical function (ADL and independent activities of daily living [IADL] impairments), mental health/quality of life (GDS scale and SF-12 composite and subscale scores), sensory function (hearing and vision impairment), frailty (Fried frailty index components), medical factors (surgery type, CCI score, cardiovascular disease, peripheral vascular disease, diabetes, and cerebrovascular disease), and biomarkers (apolipoprotein 4 genotype and C-reactive protein level). Our outcome of interest was the rate of cognitive decline over 3 years of follow-up, which was estimated previously for each participant based on mixed effects regression models with random effects for baseline and change over time; this slope was estimated beginning at postoperative month 2 because, on average, participants started a linear trajectory of gradual cognitive decline at that time (3).

We used simple linear regression to estimate mean differences in rates of cognitive decline (and 95% confidence intervals [CIs]) for each unit increase in continuous predictors and for each level of a categorical predictor compared to a chosen reference level. We divided mean differences associated with continuous predictors by twice the SD of the predictor to make the scaling similar to that of binary predictors (23). Next, for each domain of pre-surgical factors, we successively added variables to linear regression models (with rate of cognitive decline as the outcome) in an order that maximized the adjusted R² value (ie, total proportion of variation explained in the outcome corrected for the number of variables in the model) with the addition of each successive variable; we also obtained total R² values (ie, total proportion of outcome variation explained by all variables in the model) for each model. Finally, we evaluated all variables retained from domain-specific models (with age and sex forced into the model) using the selection procedure described above and obtained adjusted and total R² values; the subset of variables that maximized the adjusted R² value of this model constituted our final model. We evaluated the normality of studentized residuals in our final linear regression model using the Shapiro-Wilk test, and identified potential outliers based on graphical methods and the absolute values of these residuals (threshold of two); we detected influential observations based on Cook’s Distance (threshold of four divided by the sample size) The selection process was repeated if outliers or influential points were identified, and these participants were excluded.

For each model, participants with complete information on contributing variables were included; as previously reported, there is little missing data in this sample (24). All statistical analyses were conducted in SAS version 9.3.

Results

Of 134 participants with postoperative delirium, we identified five individuals who were potential outliers and three individuals with influential data in our final model, and subsequently excluded these participants from our analyses; therefore, the modeling steps of our analyses include the remaining sample of 126 participants with postoperative delirium.

In Table 1, selected pre-surgical characteristics are described in the overall SAGES cohort and separately for participants who developed postoperative delirium. Overall, the cohort had an average age of 76.8 years at baseline; it was 60% female and 8% non-white. Participants with postoperative delirium were slightly older, had lower GCP scores, and more often had multiple comorbidities (as indicated by the Charlson comorbidity index score) compared to the overall cohort. Other differences were relatively small comparing participants with delirium to the whole cohort.

Table 1.

Selected Baseline Characteristics of Participants, Including Those With Post-Operative Delirium, in the Successful AGing After Elective Surgery (SAGES) Study^a

	All Participants (n = 552)	Participants With Delirium (n = 126)
Continuous Variables	Mean ± SD
Age, in years	76.8 ± 5.2	77.2 ± 4.8
Education, in years	14.8 ± 3.0	14.7 ± 3.0
Leisure-time activity level, in Metabolic Equivalent Times expended per week	742 ± 1230	739 ± 1495
Global Cognitive Performance score, in standard units	56.9 ± 7.4	54.7 ± 6.4
Informant Questionnaire for Cognitive Decline in the Elderly score	3.1 ± 0.2	3.2 ± 0.3
Geriatric Depression Scale score, in points	2.7 ± 2.6	3.0 ± 2.8
Categorical variables	Percentages
Female sex	60	60
Non-white race	8	10
Living alone	30	27
Current smoking	5	5
≥1 weekly alcohol intake	38	30
Any activities of daily living impairment	8	9
Any independent activities of daily living impairment	31	35
Any cognitive independent activities of daily living impairment	6	10
Hearing impairment	31	35
Surgery type
Orthopedic	79	78
Vascular	7	9
Gastrointestinal	14	13
Charlson comorbidity index score, in points
0	44	41
1	26	17
≥2	30	42
Cardiovascular disease	13	17
Diabetes	22	25
Cerebrovascular disease	6	8

Note: ^aData are derived from 552 participants, except there is missing information on alcohol intake (four missing), Informant Questionnaire for Cognitive Decline in the Elderly score (12 missing), and Geriatric Depression Scale score (two missing).

In univariable models, worse cognitive function prior to surgery (Domain 3) was significantly associated with faster cognitive decline over follow-up; specifically, participants with lower GCP scores and higher IQCODE scores at baseline had greater slopes of cognitive decline (mean differences: 0.47 points/year [95% CI: 0.24, 0.71] per half SD higher GCP score at pre-operative assessment, and -0.34 points/year [95% CI: −0.61, −0.08] per half SD higher IQCODE score) (Supplementary Table 2). Living alone (a component of Domain 1) was related to significantly faster cognitive decline (mean difference: −0.32 points/year, 95% CI: −0.59, −0.05), and other variables (ie, any cognitive IADL, hearing impairment, vision impairment, weight loss, and cerebrovascular disease) appeared to be related to faster cognitive decline, but these associations did not reach statistical significance.

When we examined the domain-specific contribution of pre-surgical factors (Supplementary Table 3), baseline cognitive performance (Domain 3) explained the most variation in rates of cognitive decline: GCP scores accounted for 11.1% and IQCODE scores contributed another 3.7%, and both variables improved the explanatory power of the model based on increased adjusted R² values upon entry of these variables to the model. Medical factors (Domain 8) explained 7.8% of variation in cognitive decline, with the Charlson comorbidity index score, diabetes, cerebrovascular disease, and cardiovascular disease explaining 3.1%, 1.8%, 1.3%, and 0.9% of variation in cognitive decline, respectively; these variables improved the model based on their adjusted R² values. Demographic variables (Domain 1) explained 5.8% of variation, with the majority of this variation explained by living alone (4.4%); this variable was the only one to increase the adjusted R² value of this domain-specific model. Frailty variables (Domain 7) explained 4.5% of variation in cognitive decline, with exhaustion explaining 1.8% of variation and slow timed walk explaining 1.5%; both of these variables improved the model based on their adjusted R² values. Other domains contributed less to explaining variation in the outcome, and only had one variable that improved the explanatory power of the respective models.

In our final step, baseline GCP score was found to explain the most variation in rate of cognitive decline (13.0%) when factors retained from each of the domain-specific models were considered for inclusion (in addition to age and sex, which were forced into the model and contributed 3.0% to explained variation) (Table 2). Living alone, exhaustion level, Charlson comorbidity index score, IQCODE score, cerebrovascular disease, and cognitive IADL impairment contributed another 6.4%, 3.2%, 2.1%, 2.0%, 1.3%, and 1.0% to explaining variation in the outcome, respectively. Beyond age and sex, these seven factors improved the amount of explained variation in the outcome based on an increase in the adjusted R² value with variable entry into the model. The total amount of variation in rates of cognitive decline explained by these variables (ie, total model R² value) was 31.8%. The residuals from this final model did not violate the assumption of normality for linear regression models (Shapiro-Wilk statistic, W = 0.99, p = .5).

Table 2.

Overall Contribution of Remaining Potential Risk Factors to Explained Variation in Cognitive Decline Among Participants With Post-Operative Delirium in the Successful AGing After Elective Surgery (SAGES) Study^a,b

	Total R2 Value	Change in Total R2 Value	Adjusted R2 Value, Adding Variables Successivelyc
Variables forced into the model
1. Age	—	—	—
2. Sex	0.0301	—	0.0088
Variables retained from Supplementary Table 3
3. Global Cognitive Performance score	0.1601	0.1300	0.1322
4. Living alone	0.2237	0.0636	0.1888
5. Frailty component 2: exhaustion	0.2554	0.0317	0.2131
6. Cerebrovascular disease	0.2680	0.0126	0.2175
7. Informant Questionnaire for Cognitive Decline in the Elderly score	0.2881	0.0201	0.2301
8. Charlson comorbidity index score	0.3087	0.0206	0.2346
9. Any cognitive independent activities of daily living impairment	0.3184	0.0097	0.2362
10. Diabetes	0.3256	0.0072	0.2352
11. C-reactive protein level	0.3325	0.0069	0.2336
12. Cardiovascular disease	0.3391	0.0066	0.2317
13. Frailty component 5: slow timed walk	0.3470	0.0079	0.2313
14. Geriatric Depression Scale score	0.3504	0.0034	0.2255
15. Hearing impairment	0.3530	0.0026	0.2185
16. Smoking status	0.3533	0.0003	0.1981

Notes: ^aData are derived from 126 participants, except there is missing information on Informant Questionnaire for Cognitive Decline in the Elderly score (two missing), frailty component 5 (27 missing), and C-reactive protein level (three missing).

^bVariables are numbered according to the order in which they were selected into the model.

^cAdjusted R² values represent the proportion of variation explained as variables are added successively to the final model, with a correction for the number of variables entered in the model. An increase in adjusted R² with addition of a variable to the model (shown in bold type) indicates the variable improved the explanatory power of the model, and a decrease in adjusted R² indicates the variable does not improve the explanatory power of the model.

Discussion

We found that worse cognitive performance (GCP and IQCODE scores) and living alone prior to surgery were significantly associated with long-term cognitive decline in older participants with postoperative delirium in SAGES. Baseline GCP score contributed most substantially to explained variation in rates of cognitive decline, and a total of seven factors—GCP score, IQCODE score, cognitive IADL impairment, living alone, cerebrovascular disease, Charlson comorbidity index, and exhaustion level—accounted for 32% of the variation in this outcome. These results suggest that pre-surgical factors may have important influences on long-term cognitive decline following postoperative delirium in older adults.

Our findings serve to confirm and extend prior work. Baseline cognitive function has been previously demonstrated to be a strong risk factor for cognitive decline in older adults (8), and was shown to be the dominant predictor of cognitive decline over time in a small study of community-dwelling older individuals (R² = 37%) (25). This result is consistent with our finding that baseline GCP score is the main factor predicting long-term cognitive decline among older individuals with postoperative delirium, and that impaired IQCODE score appears to contribute to this prediction as well. In addition, previous work in SAGES found that lower GCP score at baseline was linearly associated with greater risk of delirium (relative risk = 2.0 for each half SD decrease in GCP score) (26), and delirium was associated with steeper rates of cognitive decline over the 3-year follow-up (3). The present study provides additional insight by suggesting that lower baseline GCP scores predict greater rates of long-term cognitive decline among those who develop postoperative delirium—in a graded fashion that holds across the full range of baseline GCP scores in this sample.

Prior epidemiologic studies have demonstrated that cerebrovascular disease (27,28) and living alone (29,30) are important risk factors for cognitive decline in older individuals, and we identified these variables as significant predictors of cognitive decline in our sample of older participants with delirium. Cerebrovascular disease has been recognized as increasing the risk of delirium as well (1), and brain injury from both types of events might predispose to subsequent deterioration in cognitive function over time. In contrast, living alone prior to surgery may result in less social support or increased social isolation, leading to reduced cognitive reserve and greater vulnerability to decline following postoperative delirium.

Other predictors that were identified in our analyses included the Charlson comorbidity index, exhaustion level, and cognitive IADL impairment, although these variables appeared to be associated with cognitive decline in a counterintuitive direction in our sample of participants with delirium. These results are likely due to inherent limitations of our multivariable modeling approach in disentangling interrelated chronic disease factors when entered into our models simultaneously, as predictors commonly change magnitude or even direction of association between univariable and multivariable analyses due to such intercorrelation (31–33). Moreover, although our modeling approach can generally identify true predictors, it can also incorporate extraneous predictors into a final model (34). Thus, our findings should be interpreted with caution and need replication in future studies; without replication, our individual predictors cannot be considered suitable for clinical applications. Despite these few potentially counterintuitive findings, the demonstration that clinical-epidemiologic factors can explain a substantial degree of variance in long-term cognitive decline following delirium is an important step forward in this area of research.

In recent studies, common neuropathologic factors associated with Alzheimer’s disease, cerebrovascular disease, and Lewy-body disease predicted 41% of variation in rates of cognitive decline among older adults (35), and postmortem factors (eg, indicators of neuronal density and neural tissue integrity) explained some additional variation in this outcome (36,37). In our study, the total percentage of explained variation was 32% based on a variety of clinical characteristics, but similar pathophysiologic or neuropathologic factors could potentially account for the remaining variation in cognitive decline. Thus, our study provides an innovative contribution for prediction of cognitive decline following delirium, and approaches prior prediction of more general cognitive decline by baseline cognitive function (R² = 37%) (25) and by neuropathology (R² = 41%) (35), which might be considered the gold standard. However, more studies are needed to explore the combination of clinical, pathophysiologic (lab-based assessments), and neuropathologic factors that could explain more of the variance in long-term cognitive decline following delirium and lead to better prediction.

Major strengths of this study include: rigorous data collection, standardized delirium assessments, careful characterization of baseline cognition, repeated neuropsychological assessments, relatively long follow-up, and careful optimization of the GCP score for analyses of cognitive decline. We also had a relatively wide range of clinical-epidemiologic data available to explore potential risk factors at baseline of this study. However, our study has several limitations that should be considered. First, we have evaluated a subset of 126 participants who developed postoperative delirium in the SAGES cohort, and this modest sample size limits the power of our analyses. We may have missed important associations of interest, and the certainty of our estimates related to observed associations and model building is decreased; low power also increases the risk of identifying false positive results, which may have occurred in this study (38). Hence, these results will need to be replicated in future, larger studies. Second, we must stress that we cannot rule out the possibility that some participants had preclinical dementia at baseline and were already on a declining trajectory, despite our extensive efforts to exclude individuals with dementia from this cohort. For example, if some individuals had mild cognitive impairment at baseline, we may have overestimated the association between lower GCP score and faster cognitive decline in our sample; however, this is less likely to be driving the observed association because we found the association held across the full range of baseline cognitive scores—even among those with the highest scores. Finally, our sample included participants who were aged ≥70 years, highly educated, and mostly white race, and therefore these results may be limited in their generalizability to younger or more diverse samples.

In summary, we found that pre-surgical factors—predominantly global cognitive performance—predicted long-term cognitive decline among older participants who experienced postoperative delirium. This study lays the groundwork for developing and validating predictive models for long-term cognitive decline following delirium, with the ultimate goal of identifying high-risk patients for randomized clinical trials and targeted prevention efforts. Future studies should utilize larger samples and a broader range of pathophysiologic and etiologic variables toward these ends.

Supplementary Material

Supplementary data is available at The Journals of Gerontology, Series A: Biological Sciences and Medical Sciences online.

Funding

This work was supported by the National Institute on Aging (P01AG031720 and K07AG041835 to S.K.I.; R01AG044518 to S.K.I./R.N.J.; and R01AG030618, K24AG035075, and R01AG051658 to E.R.M.). S.K.I. holds the Milton and Shirley F. Levy Family Chair.