Introduction
Cognitive impairment in critically ill survivors who have undergone invasive mechanical ventilation (MV) during the Intensive Care Unit (ICU) stay is a well-established health problem [
1‐
11]. ICU admission is associated with a greater cognitive decline than general ward hospitalization [
12], and the rate of dementia diagnosis after 3 years of follow-up has been reported to be higher in elderly ICU survivors than in the general population [
13]. This cognitive decline affects the functional and socioeconomic status of ICU survivors and their families, and reduces their quality of life after ICU discharge [
5,
14‐
16]. Cognitive impairment may be as high as 64% in ICU survivors [
17]. The domains most commonly affected are attention/concentration, memory and executive function [
18].
The pathophysiological mechanisms that lead to cognitive dysfunction after critical illness are not well understood. Many precipitating or modifiable factors and predisposing or non-modifiable factors have been related to the short- and long-term cognitive deficit observed in ICU survivors [
19]. Among precipitating factors such as MV [
2,
20], length of ICU stay [
9,
20], hypoxemia, hypoglycemia, hyperglycemia, fluctuations in serum glucose [
20‐
22] and perceived stress levels during the ICU stay [
23], the presence of delirium has shown the closest association with the cognitive impairment observed in ICU survivors [
2,
3,
24‐
26]. Predisposing or non-modifiable factors have been less explored and in most studies have been considered as confounding factors. Nonetheless, older age, previous cognitive impairment and higher illness severity seem to increase the risk of developing cognitive dysfunction after ICU stay [
18,
27]. The role of other individual predisposing factors has only rarely been studied in critical illness. One of these factors is cognitive reserve—that is, the ability of the brain to actively address brain dysfunction by using pre-existing cognitive processing approaches or by enlisting compensatory approaches [
28]. Cognitive reserve may confer a better resilience to pathological brain changes; that is, people with higher cognitive reserve may be less vulnerable to neurophysiological insults such as the impact of critical illness and its management. Cognitive reserve may also be a target for rehabilitation programs when the brain insult has already occurred.
Most of the studies evaluating cognitive impairment in ICU survivors after MV have focused on global cognitive impairment, and little is known about the characterization of different phenotypes of alteration [
12]. The early detection of distinct groups of patients regarding cognitive deficits might reveal the involvement of diverse pathophysiological mechanisms. Thus, the main objective of this study was to describe the cognitive phenotypes 1 month after ICU discharge in survivors of critical illness who had undergone MV during their ICU stay, using an unsupervised machine learning method. Two secondary objectives were also proposed. To warrant clinical interpretation of these results, we contrasted the cognitive phenotypes with the classical definition of cognitive impairment in critically ill patients established by Jackson et al. [
7]. An exploratory analysis of the predisposing and precipitating factors (e.g., gender, medications, severity of illness and days with MV) of the cognitive dysfunction after critical care was also carried out.
Discussion
The main finding of this study is the characterization of three different cognitive phenotypes in critically ill-ventilated patients 1 month after ICU discharge using the unsupervised machine learning K-means clustering algorithm. The descriptive and exploratory analysis of factors revealed female gender and older age as potential risk factors for specific cognitive impairment, while cognitive reserve emerged as a protective factor against cognitive deficit.
This is the first study to report patterns of cognitive impairment after ICU discharge in mechanically ventilated patients based on an unsupervised machine learning algorithm clustering method. This novel approach allowed us to detect three different phenotypes in the ICU survivors based on the exploration of six cognitive domains, instead of the two categories (impaired vs. non-impaired) differentiated in the classical method by Jackson et al. [
7]. Furthermore, in two of the three phenotypes, different levels and types of cognitive impairment were observed in the participants: while the K1 phenotype was characterized by severe alterations in speed of processing and executive function, the K2 phenotype was distinguished by moderate-to-severe deficits in learning and memory retrieval, and impaired speed of processing and executive function. The last phenotype (K3) was characterized by an almost normal performance in most participants on most of the cognitive domains assessed. These results are in line with other studies [
5,
8,
9], except for the low impairment in attention.
The presence of cognitive alterations in different domains in ICU survivors might suggest different patterns of brain dysfunction which probably involve various pathophysiological mechanisms or pathways. More specifically, while speed of processing, executive functions and memory retrieval impairments are related to dysfunctions in subcortical areas and the frontal lobe, the presence of learning and memory storage problems points to alterations in the hippocampus and the medial temporal lobe.
Using the classical approach for cognitive impairment, almost half of the participants (47%) showed moderate-to-severe cognitive impairment 1 month after ICU discharge. This low incidence in comparison with other studies [
2,
3,
7] may be related to the patients’ clinical characteristics, given that our sample presented lower severity of illness and shorter duration of delirium during ICU stay than samples in other reports. Moreover, while pre-existing cognitive impairment was ruled out in our patients, in other studies it may have contributed to the overall cognitive impairment observed. Interestingly, when the results of the unsupervised learning machine method were compared with the classical approach, 86% of the participants were accurately classified between the two categories of impaired and non-impaired cognition. Thus, the unsupervised learning machine method not only allowed detection of cognitive decline but also improved the classification of patients by identifying different patterns of cognitive impairment among ICU survivors.
The three cognitive phenotypes differed in terms of several demographic and clinical factors, a circumstance that may have had an impact on how the clusters were configured. Patients with the K1 phenotype had significantly fewer days of opioid treatment than patients in K2 and K3, and lower accumulated doses of opioids than K3. Participants in the K2 phenotype were mostly women, older and had more comorbidities than those in K1 and K3. Moreover, they presented lower accumulated doses of sedative than K3. Finally, participants with the K3 phenotype showed higher levels of cognitive reserve than K1 and K2.
When the most cognitively affected phenotypes were combined (K1 and K2), the exploratory analysis of the predisposing and precipitating factors suggested that certain factors may play a more important role than others for the development of cognitive decline after ICU. Specifically, female gender, older age and a lower level of cognitive reserve were significantly associated with cognitive impairment.
Looking at these factors individually, we found that women were more likely to present cognitive impairment 1 month after ICU discharge. The role of gender in the cognitive impairment after ICU has not been specifically addressed, and the occasional references in the literature are contradictory [
5,
34,
35]. Our results coincide with Habib et al. [
35] in suggesting that female gender may be a risk factor for developing post-ICU cognitive impairment, at least early after ICU discharge. However, these conclusions should be interpreted with caution: although in healthy populations older women usually perform better in verbal memory tests than older men [
36], normal aging itself entails cognitive deficits, especially memory and speed of processing [
37]. Previous results for the impact of age on the cognitive status of ICU survivors are controversial [
3,
4,
9,
25,
34,
35,
38]. Our results corroborate the notion that older critical care patients are more likely to present cognitive decline after critical illness. However, we cannot rule out a relation between age and gender in our sample, since patients in phenotype K2 (deficits in memory, speed of processing and executive dysfunction) tended to be female, older and presented more comorbidities. If female patients are commonly older, and if aging affects cognitive status in women differently than in men, it may be that the impact of ICU stay on cognition in older critically ill patients also differs between genders.
In this study, patients in the phenotype with the best cognitive performance (K3) presented the highest level of cognitive reserve. Although the analysis is only exploratory, cognitive reserve was found to be a protective factor against cognitive alterations 1 month after ICU survival. Only one previous study [
10] has included this concept as a predisposing factor for cognitive decline in ICU patients. Interestingly, cognitive reserve has also been identified as a protective factor for cognitive decline in healthy older adults [
39,
40] and in a wide range of medical populations [
41‐
43], including older patients with postsurgical delirium [
44].
One of the phenotypes with cognitive impairment (K1) presented significantly fewer days with opioid treatment than the others. Although the ratio between days with opioids and days of ICU stay only reached a trend towards significance in the exploratory analysis, opioid treatment was the only precipitating factor that could be related to cognitive decline in our sample. It should be borne in mind that opioid treatment improves the welfare and comfort of critically ill patients, enhancing their emotional status. This emotional well-being related to the management of the analgesia and sedation during ICU stay may impact the cognitive status of survivors.
Duration of delirium was not related to cognitive impairment. This may have been due to the notably short duration of delirium in our patients and the low inter-subject variability; furthermore, our patients were assessed 1 month after ICU discharge, while in the other studies the impact of delirium on cognition was assessed at 3 or 12 months.
Although the analysis was only exploratory, our results suggested a higher burden of predisposing factors (such as gender, age and cognitive reserve) than precipitating factors in the specific cognitive impairment detected early after ICU discharge.
The current results should be confirmed in future studies with a higher number of participants. Nevertheless, our preliminary findings may serve as a starting-point for further research. Of particular interest is the evolution of the cognitive sequelae in the two phenotypes with cognitive impairment. Establishing how patients in K1 and K2 resolve (or maintain) their cognitive deficits might help to clarify the burden of predisposing factors in long-term cognitive decline in ICU survivors. Determining how brain changes associated with aging in both genders may be impacted by critical illness, and how cognitive reserve may decrease this impact, also deserves further investigation.
The main limitation of this study is the size of the sample obtained in one of the clusters generated by the K-means clustering algorithm. This small size hampered the analysis of the role of the predisposing and precipitating factors in the three cognitive phenotypes in the ICU MV survivors. The two phenotypes that included most of their participants with cognitive impairment had to be combined in a single group in order to run the analysis. Nonetheless, the analysis was underpowered and it must be considered as exploratory. However, the strict selection of participants (with control of any previous cognitive impairment), and the careful statistical analysis vouch for the accuracy of our conclusions. The optimal interval for cognitive assessment may also be a limitation for performing comparisons with other studies, although it was appropriate for detecting the cognitive phenotype in the early stages of the recovery phase.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.