Studies on delirium and associated cognitive and functional decline in older surgical patients
Beishuizen, Sara
DOI:
10.33612/diss.135861414
IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document version below.
Document Version
Publisher's PDF, also known as Version of record
Publication date: 2020
Link to publication in University of Groningen/UMCG research database
Citation for published version (APA):
Beishuizen, S. (2020). Studies on delirium and associated cognitive and functional decline in older surgical patients: The time is now to improve perioperative care and outcomes. University of Groningen.
https://doi.org/10.33612/diss.135861414
Copyright
Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).
Take-down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum.
Chapter 6
Distinct cognitive trajectories in the first year after
hip fracture
Sara J. Beishuizen Barbara C. van Munster Annemarieke de Jonghe Ameen Abu-Hanna Bianca M. Buurman Sophia E. de Rooij
ABSTRACT
Objectives: Change in cognitive functioning is often observed after hip fracture. Different patterns, with both improvement and decline, are expected, depending on premorbid cognitive functioning and events that occur during hospitalization. These patterns are unknown and important for older hip fracture patients with different levels of premorbid cognitive functioning. Design, Setting, Participants, Measurements: We conducted a secondary analysis of a multi-center randomized controlled trial. 302 consecutive patients aged 65-102 years old, admitted for hip fracture surgery, were enrolled. The Mini Mental State Examination (MMSE) was obtained at hospital admission, at discharge, and at 3 and 12 months after discharge. Cognitive trajectories were identified with Group Based Trajectory Modelling, using the repeated MMSE measurements as outcome variable. To illustrate the specific characteristics of this relative novel methodological approach, it was contrasted with results obtained from linear mixed effects modeling.
Results: 146 (48.3%) patients had premorbid cognitive impairment and 85 patients (28.1%) experienced delirium during admission. Three distinct cognitive trajectories were identified and labeled based on different MMSE course over time: improvement (57.9%), stable (28.1%) and rapid decline (13.9%), with an annual MMSE change of 1.7, 0.8 and -3.5 points respectively. With mixed effects modeling an overall annual increase of 0.7 MMSE points was estimated for the group as a whole.
Conclusion: Three distinct cognitive trajectories were identified in a population of older hip fracture patients. These trajectory groups can be used as a starting point to inform patients and caregivers on the possible prognosis after hip fracture. Group based trajectory modelling is a useful technique when the purpose is to describe patterns of change within a population and a variety of trajectories is expected to exist.
BACKGROUND
Hip fracture is mostly encountered by the oldest, most vulnerable patients1 and as the population
ages, the absolute number of hip fractures is expected to increase to 6.3 million annually worldwide by 2050.2,3 Aside from the morbidity and mortality that arise in the perioperative
period, hip fracture is associated with increased risk of subsequent cognitive impairment, which can result in loss of independence, institutionalization and decreased quality of life.4,5
There is a range of factors that can contribute to cognitive impairment after hip fracture. Delirium has been studied extensively in this context, as it is encountered by 10-60% of older hip fracture patients.6 Several prospective cohort studies have found an increased incidence of
dementia in patients who suffered from perioperative delirium and were free of dementia prior to surgery.7-9 In patients with diagnosed dementia, it has been shown that delirium, encountered
during stay in a long term care facility or during acute hospitalization, can accelerate cognitive decline.10-13 Additionally, hospitalization in itself, irrespective of the occurrence of delirium, is
associated with cognitive decline and subsequent increased incidence of dementia.14,15 The nature
and direction of this relationship have not been firmly established. Previously unrecognized symptoms of preclinical dementia might be unmasked or exacerbated by acute illness and hospitalization. On the other hand, some patients are only temporarily, or not at all affected by these events.16
Previous research has identified different cognitive trajectories during and after hospitalization in populations of both medical and surgical older patients without substantial premorbid cognitive impairment. Surgery and delirium were main predictors of worse cognitive outcome.16,17
Cognitive trajectories after hip fracture in the mixed population of both cognitively intact and impaired older patients have not yet been studied, although cognitive impairment is an established outcome. Since dementia is a risk factor for both sustaining a hip fracture and delirium18, it is relevant to also include patients with premorbid dementia.
To better predict the prognosis of our normal clinical population, our primary aim was to investigate which cognitive trajectories occur after hip fracture and how the different trajectory groups are composed. There exists a number of different methods to assess trajectories of change over time in cohort studies.19 The general idea of all methods is to minimize the variance within
cluster analysis, different variations of latent class analysis and hidden Markov chains.20 We
have chosen to apply the method of Group Based Trajectory Modeling (GBTM)21, which is a
form of latent class analysis. The advantage of this method is that it makes no stand on the distribution of trajectories which fits in with the explorative nature of our research.21 In order to
better illustrate the specific characteristics of this relative novel methodological approach, it was contrasted with results obtained from the commonly applied method of linear mixed effects modelling.
METHODS
Study designWe performed a secondary analysis of a multi-center double blind randomized controlled trial that was conducted in the Netherlands between November 2008 and May 2013. The trial, of which the protocol and results have been published previously 22,23 investigated whether
prophylactic in hospital use of melatonin could prevent delirium after hip fracture surgery, which could not be demonstrated. In the present sub study, data on cognitive functioning were analyzed. The study was undertaken in compliance with the Helsinki Declaration and Good Clinical Practice Guidelines and approved by the Medical Ethics Committee of the Academic Medical Centre. From all patients, or a legal representative in case of cognitive impairment, written informed consent was obtained. The trial was registered with the Dutch Clinical Trial Registry (NTR1576).
Setting and subjects
The study locations were the surgical, orthopedic and trauma surgery ward of the Academic Medical Centre in Amsterdam, a 1000-bed university teaching hospital, and both locations of Tergooi in Hilversum and Blaricum, a 633-bed regional teaching hospital.
The study population of the original trial consisted of consecutive patients, aged 65 years or older, who were acutely admitted for hip fracture surgery. Exclusion criteria were transfer from another hospital to the trial location, anticipated postoperative stay on intensive care or coronary care unit, and inability to speak Dutch. For this sub-study we also excluded patients with less than two (out of possible four) MMSE measurements during the entire study period.
Clinical assessments
All patients aged 65 years or older with hip fracture were screened for eligibility for the original trial and asked to participate in the study within 24 hours after admission. At baseline, we recorded demographic data, medical history and medication use. Specifically use of benzodiazepines was noted, as this has recently been linked to an increased risk of Alzheimer’s disease (AD).24 Number and severity of comorbidities was scored with the Charlson comorbidity
index.25 We assessed daily functioning with the 15-item modified Katz Index of Activities of
Daily Living (Katz-ADL), based on the situation two weeks prior to admission.26 The
questionnaire was completed by the patients or, in case of cognitive impairment, by their primary caregiver. Functional impairment was calculated as the sum of all activities with impairment. Cognitive assessment
Our primary outcome, cognitive functioning, was assessed with the MMSE on four occasions: at admission, at discharge and at 3 and 12 months post-discharge. The MMSE is a validated 30-point questionnaire covering 11 cognitive domains. It assesses global cognitive impairment, with higher scores indicating better cognitive functioning.27 The Informant Questionnaire on
Cognitive Decline Short Form (IQCODE-sf) was completed by the primary caregiver, by comparing the situation two weeks before admission with 10 years earlier.28 Premorbid cognitive
impairment was defined as a score of 3.4 or higher on this questionnaire or a record of dementia in the medical history.28 The presence of delirium was assessed daily during hospitalization by
visiting the patient and using the criteria from the Diagnostic and Statistical Manual of Mental
Disorders, IV edition (DMS-IV-R),29 by an experienced team of geriatric nurses and
geriatricians. All patient information from sources such as medical and nursing records and a delirium observation screening (DOS) scale30 filled in three times a day, from the previous 24
hours were incorporated in this assessment. When delirium was present, severity was assessed using the13 severity items of the Delirium Rating Scale – Revised-98 (DRS-R-98, Dutch version).31,32 The scale has a maximum severity score of 39 points with higher scores indicating
higher severity. For the patients with delirium at any point during admission a maximum DRS-R-98 score was calculated as a single representation of delirium severity. Patients without delirium were not assessed with this scale and thus received a score of zero. In our analysis we used this DRS-R-98 maximum score as a single variable to indicate both the presence and severity of delirium. Total doses of haloperidol and benzodiazepines (in oxazepam equivalents) administered during delirium were also recorded.
Data analysis
Demographics and clinical variables were summarized using descriptive statistics in SPSS version 20.0. Baseline differences between patients with 0, 1 and 2 or more MMSE scores available, were assessed with One-way ANOVA, Chi square test or Kruskal-Wallis test, as appropriate. Differences were considered statistically significant at P<0.05 (2-tailed).
To identify distinct cognitive trajectories, we applied group-based trajectory modeling (GBTM) on the repeated measurements of MMSE over time. GBTM uses trajectory groups as a statistical device to approximate the unknown distribution of trajectories across the study participants. 21
This method was developed for the use in social sciences, and has recently been introduced in biomedical research.33,34 We used PROC TRAJ in SAS version 9.3.35 To evaluate model fit we
used the Bayesian Information Criterion, which balances model complexity (number of trajectory groups) and sample size on the one hand, with goodness of fit to the sample data (maximized value of the likelihood function) on the other hand.35 The adequacy of the final
model was evaluated by use of the average posterior probabilities of group membership (≥0.9 is excellent fit, ≤0.7 is poor fit).36 We added age, sex, premorbid cognitive impairment
(dichotomous), baseline functional status (KATZ-ADL), benzodiazepine use at home, delirium, and maximum delirium severity as covariates, to control for possible confounding. When assessing potential differences in patients’ characteristics between members of trajectory groups, the Holm-Bonferroni multiple-comparator adjustment method was used to control for type I errors.37 A multinomial logistic regression was performed to calculate the adjusted odds ratios
(OR’s) of the different patient characteristics for membership to the trajectory groups. Group membership was modelled as depending on age, sex, premorbid cognitive impairment, baseline functional status, benzodiazepine use at home, delirium and maximum delirium severity. We did not include baseline MMSE as a predictor of group membership because it is also used as an outcome that defined group membership. Model fit was assessed with Akaike’s Information Criterion (AIC).
Next, the average change in MMSE score was estimated with a linear mixed effects model, and in particular a random intercept model. MMSE score was the dependent variable, the patient constituted the random effect (the intercept), and time, premorbid cognitive impairment, delirium and maximum delirium severity were the fixed effects. Interaction terms for delirium, cognitive impairment and time were considered. In addition, age, sex, baseline functional status and
benzodiazepine use at home were added as covariates. We selected the best fitting model based on the AIC and by inspection of the distribution of the residuals.
In order to assess the impact of missing MMSE measurements on our outcomes of the GBTM analysis, we performed two sensitivity analyses. In the first sensitivity analysis we included all patients who had at least one MMSE measurement. We generated 10 datasets with multiple imputations using the approach of Multivariate Imputations by Chained Equations (MICE)38
using the MICE package in the R statistical environment39. We imputed any missing value of any
variable, not only the MMSE variables. For imputing a variable value any of the following variables were used as a predictor when the predictor showed a minimal correlation of at least 0.1 to the predicted variable: MMSE score at each time point, death at each time point, age, sex, number of functional impairments at baseline, cognitive impairment at baseline, prior benzodiazepine use at home, delirium during admission and delirium severity. Predictive mean matching was used for numeric outcomes, like the MMSEs, and for binary outcomes logistic regression was used. We set the number of iterations for each imputed dataset to 20 (default is 5). In the second sensitivity analysis we used the same approach, but now a value of zero was imputed on the MMSE in case of missing due to death. Outcomes of the two sensitivity analyses were summarized by calculating means and standard deviations across the 10 imputation sets.
RESULTS
Patient recruitment and baseline data
850 patients were assessed for eligibility, of which 315 patients declined to participate and another 115 did not meet inclusion criteria for the MAPLE trial. This left us with 420 eligible patients (Figure 1). 118 patients had less than two MMSE scores available (50 had one MMSE score and 68 patients had no MMSE score) and were thus excluded from further analysis. Reasons for having one or no MMSE scores were: death during the follow-up period (n=69), refusing follow-up visits (n=13), lost to follow-up (n=1) or being too frail to be interviewed (the remainder). The baseline characteristics of the 302 included and 118 excluded patients are summarized in Table 1. Excluded patients were significantly older, were less often living at home and more often had functional impairment. Premorbid cognitive impairment was present in 48.3% (146/302) of included patients and 88.1% (104/118) of excluded patients (p<0.001).
Delirium occurred in 28.1% (85/302) of included patients and 44.1% (52/118) of excluded patients (p=0.007) and was more severe in the latter group.
At the first follow-up assessment three months later, 18 included patients refused follow-up visits, five were lost to follow-up and 22 patients had died. At the second follow-up assessment 10 patients refused follow-up visits, six were lost to follow-up and 32 more patients had deceased. So 209 out of the 302 (69.2%) included patients completed all follow-up assessments (Figure 1).
Table 1. Characteristics of included and excluded patients at admission Variable Included: ≥ 2 MMSE scores (n=302) Excluded: 1 MMSE score (n=50) Excluded: no MMSE score (n=68) p-value Age, mean ± SD 83.3 ± 7.8 86.3 ± 7.2 86.2 ± 6.4 0.002 Sex, % male 88 (29.1) 15 (30.0) 12 (17.6) 0.14 Living at home, % 228 (75.5) 21 (42.0) 4 (5.9) <0.001 Katz-ADL preadmission, median (IQR) 4 (1-8) 8 (6-12) 13 (11-14) <0.001 missing data 4 1 5
Charlson, median (IQR) 1 (0-2) 2 (1-3) 2 (1-3) <0.001
Diagnosis of dementia 41 (13.6) 24 (48.0) 59 (86.8) <0.001
Admission MMSE, mean ± SD
22.5 (6.8) 15.5 (8.6) na <0.001
missing data 10 15
IQCODE-sf, median (IQR) 3.4 (3.0-4.2) 4.7 (4.0-5.0) 5.0 (4.8-5.0) <0.001
missing data 29 8 10
Premorbid cognitive impairment, %
146 (48.3) 36 (72.0) 68 (100.0) <0.001
Delirium during admission, %
85 (28.1) 23 (46.0) 29 (42.6) 0.007
Duration of delirium, median days (IQR) 2 (1-3.5 2 (2-4) 2 (1-3) 0.22 Maximum DRS-R-98, median (IQR) 16.8 (12.1-22) 22.0 (18.0-24.5) 22.5 (18.5-24.5) <0.001 missing data 5 4 2
SD= Standard Deviation, ADL= Activities of Daily Living, IQR= Inter Quartile Range, MMSE= Mini Mental State Examinatio (range 0-30), IQCODE-sf= Informant Questionnaire on Cognitive Decline Short Form (range 0-5), Premorbid cognitive impairment: IQCODE-sf≥3.4 and/or diagnosis of dementia, DRS-R-98= Delirium Rating Scale – Revised-98 (range 0-39)
Groups identified by group based trajectory modeling
The model with three cognitive trajectory groups proved to have the best fit. In all groups the average posterior probability of group membership was ≥0.9, which is considered an excellent
fit. Trajectories were labeled based on MMSE course over time: 13.9% of participants were assigned to the rapid decline group, 28.1% to the stable group and 57.9% to the improvement group (Figure 2). For all groups, the first MMSE assessment was most often present and the last was most often missing. All three trajectories showed an improvement in MMSE score between admission and discharge from the hospital. After discharge, the rapid decline group, who had a mean MMSE of 10.0 at admission (standard deviation (SD) 4.9), showed a deterioration with a linear slope of -3.5 points per year (95% confidence interval (CI) -5.9, -1.1). The stable group started with a mean MMSE of 18.9 (SD 3.9) and was characterized by steady course during the post-discharge period, resulting in a flat slope of 0.8 points improvement per year (95% CI -0.7, 2.3). The improvement group had a mean admission MMSE of 26.9 (SD 2.5) and a positive linear slope of 1.7 points per year (95% CI 0.9, 2.6).
In Table 2 the baseline characteristics of patients assigned to each cognitive trajectory group are described. There were clear differences between groups in terms of age, number of comorbidities, functional and cognitive status and delirium occurrence and severity during admission. Compared to the other groups, patients in the improvement group were significantly younger, had fewer comorbidities and functional impairments, less often had premorbid cognitive impairment and less often experienced delirium during admission. None of the patients in this group had a formal diagnosis of dementia. Additional differences between the rapid decline group and the stable group were noted. Patients in the former group had significantly more often functional and cognitive impairment, diagnosed with dementia, and more often experience delirium during hospitalization than patients in the latter group. Patients in the rapid decline group had the most severe delirium, although this did not result in higher doses of haloperidol or benzodiazepines administered during delirium. Death during follow up was more common in the rapid decline group (28.6%) and the stable group (29.4%) than in the improvement group (10.9%, chi-square 16.29, p<0.001). After correction for possible confounders, premorbid functional and cognitive impairment, and delirium presence and severity remained associated with increased odds of membership to both the stable and the rapid decline group as compared to the improvement group (Table 3).
Table 2. Characteristics of participants assigned to each trajectory group
SD= Standard Deviation, ADL= Activities of Daily Living, IQR= Inter Quartile Range, MMSE= Mini Mental State Examination (range 0-30), IQCODE-sf= Informant Questionnaire on Cognitive Decline Short Form (range 0-5), Premorbid cognitive impairment: IQCODE-sf≥3.4 and/or diagnosis of dementia, DRS-R-98= Delirium Rating Scale – Revised-98 (range 0-39), *α adjusted with Holm-Bonferroni method
Characteristic Rapid decline
(n=42) Stable (n=85) Improvement (n=175) Significant post hoc comparisons* Mean age ± SD 87.5 ± 6.5 86.1 ± 6,8 80.9 ± 7.7 1≠3, 2≠3 Sex, % male 13 (31.0) 21 (27.4) 54 (30.9)
Mean years of education missing data 8.8 ± 3.5 9 8.5 ± 2.7 7 10.1 ± 3.1 10 2≠3 Living at home, % 15 (35.7) 52 (61.2) 161 (92.0) 1≠2, 1≠3, 2≠3
Katz-ADL preadmission, median (IQR)
10.0 (6.9, 12.8) 7.0 (4.0, 9.1) 2.0 (0.0, 4.0) 1≠2, 1≠3, 2≠3 Charlson Comorbidity Index,
median (IQR)
1 (1-3) 1 (1-2,5) 1 (0-2) 1≠3, 2≠3
Benzodiazapine use at home, % 5 (11.9) 8 (9.4) 14 (8.0)
Diagnosis of dementia 25 (59.5) 16 (18.8) 0 (0) 1≠2, 1≠3, 2≠3
IQCODE-sf, median (IQR) 4.7 (4.4, 5.0) 4.1 (3.4, 4.6) 3.1 (3.0, 3.4) 1≠2, 1≠3, 2≠3
missing data 0 7 22
Premorbid cognitive impairment % 40 (95.2) 63 (74.1) 43 (24.6) 1≠2, 1≠3, 2≠3 Perioperative delirium % 28 (66.6) 39 (45.9) 18 (10.3) 1≠2, 1≠3, 2≠3 Duration of delirium, median days
(IQR) 2 (1-4) 2 (1-4) 2 (1-3) Maximum DRS-R-98, median (IQR) 21.0 (17.5, 24.0) 14.0 (12.3, 21.8) 12.0 (7.8, 17.0) 1≠2, 1≠3 missing data 0 2 1
Use of haloperidol during delirium, median mg (IQR)
5.0 (2.5, 7.8) 4.8 (1.8, 7.1) 4.0 (2.0, 6.7) missing data
Use of benzodiazepines during delirium, median mg oxazepam equivalents 0 100.1 (45.4, 195.8) 1 75.1 (33.4, 125.3) 0 50.0 (26.8, 79.2) missing data 6 14 9
Table 3. Patient characteristics associated with three cognitive trajectory groups
Characteristic Rapid decline vs
Improvement OR (95% CI) Rapid decline vs Stable OR (95% CI) Stable vs Improvement OR (95% CI) Age in years 1.04 (0.96, 1.12) 1.01 (0.95, 1.08) 1.02 (0.97, 1.08) Sex (male) 0.87 (0.27, 2.74) 0.69 (0.26, 1.84) 1.26 (0.57, 2.80) Katz-ADL preadmisson§ 1.62 (1.38, 1.89)** 1.22 (1.07. 1.39)* 1.33 (1.19, 1.49)** Charslon Comorbidity Index 1.15 (0.80, 1.66) 1.02 (0.75, 1.39) 1.13 (0.89, 1.44) Benzodiazepine use at home 0.98 (0.22, 4.49) 1.55 (0.44, 5.76) 0.64 (0.21, 1.93) Premorbid cognitive
impairment
24.08 (4.82-129.37)** 6.16 (1.23, 30.68)* 3.91 (1.87, 8.18)** Maximum DRS-R-98§ 1.22 (1.14, 1.30)** 1.08 (1.03, 1.13)* 1.13 (1.07, 1.18)** The final model was adjusted for age, sex, premorbid cognitive impairment, baseline functional status, benzodiazepine use at home, delirium and maximum delirium severity. OR= Odds Ratio, CI= Confidence Interval, ADL= Activities of Daily Living, DRS-R-98= Delirium Rating Scale – Revised-98 (range 0.30) § OR for every point increase on this scale
*p<0.05, **p<0.001
Mixed effects modeling
The model with a random intercept for each patient yielded the best fit, a random slope for each patient did not result in improvement of the model. The interaction terms for delirium and premorbid cognitive impairment with time were not significantly associated with the outcome, so they were discarded from the final model. As a result, it was not possible to estimate different slopes for subgroups, and one single slope of MMSE change was estimated for all patients. The estimated admission MMSE score was 23.3 (95% CI 17.1, 29.5) and the change over time was an increase of 0.7 MMSE points per year (95% CI 0.1, 1.3).
Patient characteristics associated with a lower MMSE were premorbid cognitive impairment (-2.6 points, 95% CI -3.8, -1.4, p<0.001), functional impairment (-0.7 point per additional ADL-activity with impairment, 95% CI -0.8, -0.5, p<0.001) and presence of perioperative delirium (-3.9 points, 95% CI -5.1, -2.9, p<0.001).
Sensitivity analysis
In the sensitivity analysis the 352 patients who had at least one MMSE score were included. In the first sensitivity analysis, in all 10 imputation sets a model with three groups had the best fit to the data, based on the criteria used in the original analysis. For easy comparison with the original analysis, we used the same names for the groups, but we add quotation marks to distinguish the groups based on imputations. The “rapid decline group” became relatively larger (mean group size 17.4%, SD 1.5) but less rapidly declining (-2.1 points per year, 95% CI -3.4, -0.7). The “stable trajectory group” became relatively smaller (mean group size 26.4%, SD 0.5) and actually showed more improvement over time (increasing 2.1 points per year, 95% CI 1.1, 3.2). The “moderate improvement group” did not change substantially in size (mean group size 56.3%, SD 1.3) nor shape (improving 1.9 points per year, 95% CI 1.2, 2.6).
In the second sensitivity analysis, were a zero was imputed for the MMSE after death, in six out of 10 imputation sets a model with three groups best fitted the data. In the remaining four sets, models with three and four groups both fitted the data well. We further evaluated the models with three groups in all 10 sets. It was noted that the group sizes did not substantially change (“rapid decline group” mean 13.0% (SD 0.4), the “stable group” mean 29.3% (SD 0.9) and the “improvement group” mean 57.7% (SD 0.9)). The latter group did not change is shape, but the two former ones did. A linear time trend was no longer observed, and instead a polynomial function of degree two, showing an accelerated cognitive decline over time, better fitted the data.
DISCUSSION
In this study of 302 older hip fracture patients with different levels of premorbid cognitive functioning, we identified three clinically distinct cognitive trajectory groups. Our results can be used as a starting point to counsel patients and caregivers on the possible cognitive prognosis after a hip fracture by assigning a patient to the most likely trajectory group based on easily available patient characteristics.
The trajectories that we have found with this rather novel methodological approach show some overlap with findings from previous studies using conventional methods. Cohort studies of non-demented hip surgery patients found an improvement in MMSE score during the months after
surgery for non-delirious patients and no improvement for delirious patients. 40,41 This pattern is
comparable with the trajectories of our improvement group, which was mainly composed of non-delirious patients without prior dementia, and our stable group, in which almost half of members experienced delirium but few were diagnosed with premorbid dementia. The rapid decline group is mostly composed of patients with delirium superimposed on dementia (DSD), a phenomenon that has not yet been studied in a hip fracture population. McCusker et al. used a mixed effects linear model to study MMSE trajectories of acutely admitted medical older patients with and without premorbid cognitive impairment and found that patients with DSD had the lowest admission MMSE and an annual decline in MMSE of -0.9 points per year.11 This is comparable
to our results of linear mixed effects modeling. However, with GBTM we were able to show that this group actually had a very different trajectory with a much faster rate of decline over time. Interestingly, this group also had the most severe delirium, as measured by the highest maximum DRS-R-98 scores. Several previous authors have found that delirium is more severe in patients with worse premorbid cognitive functioning,42-45 and that a more severe delirium is associated
with increased nursing home placement and mortality.46 We found that a more severe delirium
was associated with increased odds of membership to the rapid decline trajectory group, which implicates an association with worse cognitive outcomes. It has been shown before that a longer duration of delirium is associated with this outcome.47
Our study clearly illustrates the different theories that underlie the methods of trajectory modeling and mixed effects linear modeling. Trajectory modelling assumes that there is a variety of shapes of slopes (trajectories) within a population, were mixed effects linear modeling (with random slope) assumes there is a common slope and covariates can off-set an individual from this slope. If we had identified only trajectories that were parallel but started at a different intercept, a mixed effects linear model with a random intercept would have described the data well, although it is not able to distinguish different groups. If on the other hand we had found many trajectories with only few subjects in each group, a random slope and intercept linear model would probably have been better suited. In our case, we identified two trajectories with different intercepts and similar slopes (the improvement group and the stable group), and one trajectory that differed both in intercept and slope (the low and rapid decline group). An important utility of trajectory modeling is that it can identify this smaller subgroup that would have remained unnoticed when only mixed effects linear modeling had been applied.
Our study has several limitations. First of all, because participants with less than two MMSE scores were excluded from our analysis, a relatively younger and less functionally and cognitively impaired sample was created. This shows that even when a maximum effort is made to include the most vulnerable patients in a study it is still difficult to retain them, mostly due to death during follow up. To estimate the amount of bias that was introduced due to missing values we performed two sensitivity analyses were missing MMSE values were imputed with multiple imputation. This allowed us to include an additional subset of 50 patients who had only one MMSE score available. The first sensitivity analysis, were all missing MMSE scores were imputed, showed trajectories very similar in shape to our original analysis, but the declining group increased substantially in number. This means that we underestimated the number of patients who show a marked cognitive decline after hip fracture. Imputing a zero for patient who died (sensitivity analysis 2) resulted in an accelerated cognitive decline in the two less favorable trajectory groups because patients in these groups were most likely to die. Although a zero is probably an underestimation of the cognitive abilities prior to death, it is suggested from previous research that rapid deterioration of cognitive functions is often observed in the last months of life.48 The lesson that can be drawn from the sensitivity analyses is that in real life the
proportion of patients with cognitive decline after hip fracture is larger than our estimate in our original data, and that possibly the decline is more accelerated in patients who die during follow up.
A second limitation is that the pre-admission MMSE assessment was influenced by hospitalization and surgery. However, we tried to circumvent this problem by including the IQCODE-sf, which provides proxy-based information on premorbid cognitive functioning in the period before hospitalization, in our definition of cognitive impairment, instead of the first MMSE score. Lastly, no formal sample size calculation was conducted as it does not exist for GBTM. In theory we might have been able to find additional, smaller trajectory groups if we had had more participants, although the clinical relevance of these findings is questionable.
Our study also has several strengths. Firstly, our cohort is a close reflection of a genuine, real life hip fracture population, which makes it easy to integrate our findings in clinical practice. We have shown that patients that are prone to a specific cognitive trajectory can be identified based on simple baseline characteristics like age and level of premorbid functional and cognitive status. This will help clinicians in informing patients and caregivers on what outcome can be expected
after hip fracture, given the patients’ profile and the course of events following the hip-fracture. This information can be of use in arranging assisted care after hip fracture like rehabilitation. Secondly, by performing a detailed baseline assessment, a daily delirium assessment including delirium severity, and up to four MMSE measurements per patients during follow-up, we have provided powerful data to study changes in cognitive functioning and associated factors. Lastly, we have introduced a relatively novel statistical method of trajectory modeling in this paper that produces easy interpretable visual results that can provide insights in patterns of change over time. By contrasting this method with a well-known regression analysis, we have illustrated the useful characteristics of trajectory modeling for the analysis of longitudinal data to readers who are not yet familiar with this technique.
Future studies should verify to what extent the observed patterns of cognitive change are influenced by specific events that occur during hospitalization by modeling dual trajectories both pre- and post-hospitalization. Further characterization of the trajectory groups can potentially lead to the identification of new predictors of cognitive decline, which can help us in developing more tailored treatment strategies.
CONCLUSION
We have shown that there are distinct cognitive trajectories in the first year after hip fracture in older patients. We identified clinically different subgroups, shown to be robust by sensitivity analysis, that can aid clinicians in counseling patients and caregivers on possible cognitive outcomes. The novel method of group based trajectory modeling is useful in describing distinct patterns of change within a population when a variety of trajectories are expected to exist. By extending the observation period to include the months prior to hip fracture, this technique can provide us with even more knowledge on determinants of cognitive change in this vulnerable population.
REFERENCES
1. Kannus P, Parkkari J, Sievanen H, Heinonen A, Vuori I, Jarvinen M. Epidemiology of hip fractures. Bone. 1996;18(1 Suppl):57S-63S.
2. Stevens JA, Rudd RA. The impact of decreasing U.S. hip fracture rates on future hip fracture estimates. Osteoporosis international. 2013;24(10):2725-2728.
3. Cooper C, Campion G, Melton LJ, 3rd. Hip fractures in the elderly: a world-wide projection. Osteoporosis international. 1992;2(6):285-289.
4. Witlox J, Eurelings LS, de Jonghe JF, Kalisvaart KJ, Eikelenboom P, van Gool WA. Delirium in elderly patients and the risk of postdischarge mortality, institutionalization, and dementia: a meta-analysis.
Jama. 2010;304(4):443-451.
5. Boonen S, Autier P, Barette M, Vanderschueren D, Lips P, Haentjens P. Functional outcome and quality of life following hip fracture in elderly women: a prospective controlled study. Osteoporosis
international. 2004;15(2):87-94.
6. Siddiqi N, Stockdale R, Britton AM, Holmes J. Interventions for preventing delirium in hospitalised patients. The Cochrane database of systematic reviews. 2007(2):Cd005563.
7. Maria Lundström R, Agneta Edlund, RN, Gösta Bucht, MD, PhD, Stig Karlsson, RN, PhD and Yngve Gustafson, MD, PhD. Dementia after Delirium in Patients with Femoral Neck Fractures. Journal of
the American Geriatrics Society. 2003;51:1002-1006.
8. Bickel H, Gradinger R, Kochs E, Forstl H. High risk of cognitive and functional decline after postoperative delirium. A three-year prospective study. Dementia and geriatric cognitive disorders. 2008;26(1):26-31.
9. Krogseth M, Wyller TB, Engedal K, Juliebo V. Delirium is an important predictor of incident dementia among elderly hip fracture patients. Dementia and geriatric cognitive disorders. 2011;31(1):63-70. 10. McCusker J, Cole MG, Voyer P, et al. Six-month outcomes of co-occurring delirium, depression, and dementia in long-term care. Journal of the American Geriatrics Society. 2014;62(12):2296-2302. 11. McCusker J, Cole M, Dendukuri N, Belzile E, Primeau F. Delirium in older medical inpatients and subsequent cognitive and functional status: a prospective study. CMAJ. 2001;165(5):575-583. 12. Fong TG, Jones RN, Marcantonio ER, et al. Adverse outcomes after hospitalization and delirium in persons with Alzheimer disease. Annals of internal medicine. 2012;156(12):848-856, W296.
13. Gross AL, Jones RN, Habtemariam DA, et al. Delirium and Long-term Cognitive Trajectory Among Persons With Dementia. Archives of internal medicine. 2012;172(17):1324-1331.
14. Katz IR, Curyto KJ, TenHave T, Mossey J, Sands L, Kallan MJ. Validating the diagnosis of delirium and evaluating its association with deterioration over a one-year period. The American journal of
15. Ehlenbach WJ, Hough CL, Crane PK, et al. Association between acute care and critical illness hospitalization and cognitive function in older adults. Jama. 2010;303(8):763-770.
16. Chen CC, Chiu MJ, Chen SP, Cheng CM, Huang GH. Patterns of cognitive change in elderly patients during and 6 months after hospitalisation: a prospective cohort study. International journal of
nursing studies. 2011;48(3):338-346.
17. Saczynski JS, Marcantonio ER, Quach L, et al. Cognitive trajectories after postoperative delirium.
The New England journal of medicine. 2012;367(1):30-39.
18. Inouye SK. Delirium in older persons. The New England journal of medicine. 2006;354(11):1157-1165.
19. Twisk J, Hoekstra T. Classifying developmental trajectories over time should be done with great caution: a comparison between methods. J Clin Epidemiol. 2012;65(10):1078-1087.
20. Ciampi A, Dyachenko A, Cole M, McCusker J. Delirium superimposed on dementia: defining disease states and course from longitudinal measurements of a multivariate index using latent class analysis and hidden Markov chains. International psychogeriatrics / IPA. 2011;23(10):1659-1670.
21. Nagin DS, Odgers CL. Group-based trajectory modeling in clinical research. Annual review of
clinical psychology. 2010;6:109-138.
22. de Jonghe A, van Munster BC, Goslings JC, et al. Effect of melatonin on incidence of delirium among patients with hip fracture: a multicentre, double-blind randomized controlled trial. CMAJ. 2014;186(14):E547-556.
23. de Jonghe A, van Munster BC, van Oosten HE, et al. The effects of melatonin versus placebo on delirium in hip fracture patients: study protocol of a randomised, placebo-controlled, double blind trial.
BMC geriatrics. 2011;11:34.
24. Billioti de Gage S, Moride Y, Ducruet T, et al. Benzodiazepine use and risk of Alzheimer's disease: case-control study. BMJ (Clinical research ed). 2014;349:g5205.
25. Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis. 1987;40(5):373-383. 26. Weinberger M, Samsa GP, Schmader K, Greenberg SM, Carr DB, Wildman DS. Comparing proxy and patients' perceptions of patients' functional status: results from an outpatient geriatric clinic.
Journal of the American Geriatrics Society. 1992;40(6):585-588.
27. Folstein MF, Folstein SE, McHugh PR. "Mini-mental state". A practical method for grading the cognitive state of patients for the clinician. Journal of psychiatric research. 1975;12(3):189-198.
28. Jorm AF. The Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE): a review.
International Psychogeriatrics. 2004;16(3):275-293.
29. Diagnostic and statistical manual of mental disorders (DSM V). American Psychiatric
30. Schuurmans MJ, Shortridge-Baggett LM, Duursma SA. The Delirium Observation Screening Scale: a screening instrument for delirium. Research and theory for nursing practice. 2003;17(1):31-50. 31. Trzepacz PT, Mittal D, Torres R, Kanary K, Norton J, Jimerson N. Validation of the Delirium Rating Scale-revised-98: comparison with the delirium rating scale and the cognitive test for delirium. J
Neuropsychiatry Clin Neurosci. 2001;13(2):229-242.
32. de Rooij SE, van Munster BC, Korevaar JC, et al. Delirium subtype identification and the validation of the Delirium Rating Scale--Revised-98 (Dutch version) in hospitalized elderly patients.
International journal of geriatric psychiatry. 2006;21(9):876-882.
33. Wilkosz PA, Seltman HJ, Devlin B, et al. Trajectories of cognitive decline in Alzheimer's disease.
International psychogeriatrics / IPA. 2010;22(2):281-290.
34. Xie H, Mayo N, Koski L. Identifying and characterizing trajectories of cognitive change in older persons with mild cognitive impairment. Dementia and geriatric cognitive disorders. 2011;31(2):165-172. 35. Jones BL, Nagin DS. Advances in Group-Based Trajectory Modeling and an SAS Procedure for Estimating Them. Sociological Methods & Research. 2007;35(4):542-571.
36. Nagin DS. Group-Based Modeling of Development. Cambridge, Harvard University Press. 2005. 37. Holm S. A simple sequentially rejective multiple test procedure. Scand J Statist. 1979;6(2):65-70. 38. Van Buuren S, Groothuis-Oudshoorn K. mice: Multivariate Imputation by Chained Equations in R. Journal of Statistical Software. 2011;45(3):1-67.
39. Team RDC. R: A language and environment for statistical computing, first ed. R Foundation for
Statistical Computing. 2009;Vienna.
40. Witlox J, Slor CJ, Jansen RW, et al. The neuropsychological sequelae of delirium in elderly patients with hip fracture three months after hospital discharge. International psychogeriatrics / IPA. 2013;25(9):1521-1531.
41. Kat MG, Vreeswijk R, de Jonghe JF, et al. Long-term cognitive outcome of delirium in elderly hip surgery patients. A prospective matched controlled study over two and a half years. Dementia and
geriatric cognitive disorders. 2008;26(1):1-8.
42. Chong E, Tay L, Chong MS. Identifying phenomenological differences and recovery of cognitive and non-cognitive symptomatology among delirium superimposed upon dementia patients (DsD) versus those without dementia (DaD) in an acute geriatric care setting. International psychogeriatrics / IPA. 2015;27(10):1695-1705.
43. Voyer P, McCusker J, Cole MG, Khomenko L. Influence of prior cognitive impairment on the severity of delirium symptoms among older patients. J Neurosci Nurs. 2006;38(2):90-101.
44. Voyer P, Richard S, Doucet L, Carmichael PH. Factors associated with delirium severity among older persons with dementia. J Neurosci Nurs. 2011;43(2):62-69.
45. Boettger S, Passik S, Breitbart W. Delirium superimposed on dementia versus delirium in the absence of dementia: phenomenological differences. Palliat Support Care. 2009;7(4):495-500.
46. Marcantonio E, Ta T, Duthie E, Resnick NM. Delirium severity and psychomotor types: their relationship with outcomes after hip fracture repair. Journal of the American Geriatrics Society. 2002;50(5):850-857.
47. Pandharipande PP, Girard TD, Jackson JC, et al. Long-term cognitive impairment after critical illness. The New England journal of medicine. 2013;369(14):1306-1316.
48. Wilson RS, Beck TL, Bienias JL, Bennett DA. Terminal cognitive decline: accelerated loss of cognition in the last years of life. Psychosomatic medicine. 2007;69(2):131-137.
