Introduction

The ability to recognize emotions is a key element in maintaining interpersonal relationships in aging and an important resource for mental health [1]. Newborns and infants younger than 5 months have the preference of smiling faces, which has been suggested to be automatic and universal and reflect their earliest experience of human face [2, 3]. Later in life, deficiency in emotion recognition has been associated with withdrawal from social interaction, poor communication, and mental health problems [4–6]. In clinical studies, poorer recognition of emotions in facial expressions has also been linked to Alzheimer’s disease [4, 7, 8] in which interpersonal problems have been identified as important targets for intervention [9].

Cognitive function, such as short-term memory, speed of reasoning and reaction time, tend to decrease at older ages even in the absence of neurodegenerative disorders, such as Alzheimer’s disease. With population ageing, ‘cognitive ageing’ has become a major public health issue affecting every day functioning of an increasing proportion of people [10, 11]. Tests of emotion recognition might represent a potential tool for detection of early-stage cognitive impairment. In addition, research on potential mechanisms, such as deficits in the recognition of other people’s emotions, which might link impaired cognitive function to interpersonal problems and behavioural disturbances, would increase our understanding of life with cognitive impairment in old age.

However, while severe states of neurodegenerative diseases have been shown to be associated with impaired recognition of emotion, knowledge of the degree to which ‘mild cognitive impairment’ is associated with impaired recognition of emotions, is limited. Studies that have included participants with mild cognitive impairment have been small in scale and have produced mixed findings [12–18]. Analyses of more than 4000 participants allowed us to more reliably examine the link between cognitive function and recognition of facial expressions across the full continuum of cognitive capacity, ranging from minor cognitive impairment to possible dementia.

Methods

Results

The mean age of participants was 69.1 years, 25.3% were women, and 25.4% had less than 13 years of education (Table 1). Depressive symptoms were reported by 7.3% and antidepressant use by 4.1%. Sixty-seven participants (1.7%) had the MMSE-score of <24 (range 7 to 23; 18 participants had less than 21 points) and 841 (20.8%) had the highest score (30 points). In descending order, the highest accuracy in recognition of emotions was found for happiness (mean of 69.3% correct), followed by ‘neutral’ (66.1%), fear (57.2%), disgust (54.3%), anger (50.8%), and sadness (45.8%).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics of the Whitehall II study participants (n = 4039).

https://doi.org/10.1371/journal.pone.0185513.t001

There were strong associations between the total MMSE cognitive function score and all accuracy task outcomes in the age, sex, education, depressive symptoms and antidepressant use adjusted models (Fig 1, panels A to F, S1 Table, Models 2). In general, study members with greater cognitive impairment had lower accuracy across all FERT accuracy domains and decreased performance in accuracy was evident already at minor (1 to 3-point) deficits in cognitive performance. The Bonferroni-corrected analyses did not remarkably change this picture; Compared to the highest MMSE group (score = 30), the first significant decline in emotion recognition was observed at the MMSE score of 28 in accuracy of anger, fear, and disgust, 27 for sadness, and 25 for happiness.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. (panels A to F) Multivariable adjusted association between the Mini-Mental State Examination (MMSE) cognitive function test score and accuracy of emotion recognition in the Facial Expression Recognition Task (FERT)^a.

^aPoint estimates are mean and error bars 95% confidence intervals. Models are adjusted for age, sex, educational level, depressive symptoms, and antidepressant use.

https://doi.org/10.1371/journal.pone.0185513.g001

The association was curvilinear in all accuracy tests (p≤ 0.023 for curvilinear trend), except for anger and happiness for which the association was linear (p< 0.0001 for linear term). In this case, curvilinearity suggests that facial expression recognition deteriorated with an accelerated rate along with lowering cognitive function scores. There was some variation in the difference between mean accuracy in the recognition of different emotions: the multivariable adjusted difference between the highest MMSE group and the group with the lowest score was 14.9 for anger, 15.5 for fear, 18.5 for disgust, 11.6 for sadness, 6.3 for happiness and 15.0 for neutral (S1 Table). Thus, recognition of happiness seemed to be less affected by a severe decline in cognitive function than other emotions.

The findings using an alternative indicator for emotion recognition, that is, emotion misclassification, largely replicated the findings on accuracy (Fig 1; S2 Table). The associations were curvilinear (all p≤ 0.016 for curvilinear trend), except for misclassification of neutral emotion where the trend was linear (p = 0.020). Multivariable adjusted mean difference in misclassification was 3.3 for anger, 4.0 for fear, 4.7 for disgust, 4.0 for sadness, 3.4 for happiness and -3.5 for neutral (S2 Table). Bonferroni-corrected analyses did not change the main conclusions.

S3 Table shows bivariate correlations between MMSE total score and subscores and accuracy of emotion recognition in the FERT, suggesting a slightly stronger association of the total MMSE score with recognition of emotions than any of the MMSE subscores. Thus, it seems that the findings are based on the global cognitive capacity rather than a deficit in any particular component of cognitive function.

Discussion

In this study among 4039 older adults, we examined the association between cognitive function and recognition of emotions in facial expressions across the full continuum of cognitive capacity. The main finding in our study was that the ability to recognize emotion in facial expressions might be affected at a relatively early phase of cognitive decline. Furthermore, a curvilinear association that we observed in several tests suggests that the capacity to recognize emotions may decline at an accelerated rate with the deterioration of cognitive function. The findings were robust to adjustment for age, sex, educational level, depressive symptoms and antidepressant use.

This evidence extends research on this field by showing that cognitive decline might not only relate to memory, executive function and reaction time, but also to processing of emotional information. Our findings add to the evidence obtained from earlier smaller-scale clinical studies showing that disturbance in various neurocognitive processes (e.g. vigilance, speed of processing, and cognitive flexibility) decreases a person’s ability to recognize emotion in facial expressions [31–35]. Impaired recognition of emotions has been found among patients with Alzheimer’s disease [4, 7, 8] while findings among participants with mild cognitive impairment have been mixed [12–18].

We demonstrated that depending on emotion, the performance in the accuracy test started to decline from 27 to 29 MMSE points out of the maximum 30 points. Scores 23 or less are used to define ‘possible dementia’ whereas the scores between 24 and 30 are within the normal range of cognitive function, although the low-bound scores might be associated with minor cognitive impairment [24, 25]. The few previous experimental studies available on this issue are small, used varying definitions of ‘mild cognitive impairment’ [14–16]. Although mild and moderate Alzheimer’s disease have been found to adversely affect components of attention [36], these deficits may be present already before the manifest disease and may further reduce the performance in facial expression recognition.

Not all emotions seem to be affected to the same extent by cognitive function. The adjusted mean difference in accuracy between the best and worst performed MMSE groups was 11.6 to 18.5 in other emotions but only 6.3 for happiness, suggesting that recognition of emotions with negative valence might be more strongly affected than recognition of happiness. Interestingly, infants younger than 5 months prefer happy faces over other emotions [2, 3]; we might hypothesise that happy emotion is also the last one to be affected by severe cognitive decline although this observations needs confirmation in other studies.

Clinical studies suggest better recognition of negative than positive stimuli among patients with Alzheimer’s disease [37]. However, the present cohort was from a general population and capable of performing the tests, thus, it is unlikely that many of the participants in our study were at an advanced stage of Alzheimer’s disease or dementia. A future line of research could apply neuroimaging studies to analyse the potential relationship of recognition of facial emotions with accelerated cognitive ageing (not associated with a neurodegenerative disease), Alzheimer’s disease, other dementias, and small vessel cerebral ischemia which has been found to be associated with mild cognitive impairment [38–40]. It would be important to examine whether different types of stimuli, e.g., static versus dynamic, produce different findings [41]. It is also important to note that poor performance on emotion recognition tasks might reflect impaired cognitive function rather than impaired emotion processing, per se, for example, with difficulties in following task instructions or technical performance of tasks [14]. However, neuroimaging data supports the hypothesis that impaired emotion recognition might indicate early signs of impaired cognitive processing [39].

Two leading theories to explain emotion recognition are Theory-Theory and Simulation Theory [42]. Theory-Theory emphasises rule-based cognitive processing in interpretation of other people’s mental states, involving the use of conceptual (semantic) emotion-related knowledge [43]. In Simulation Theory, the emphasis is on the other cognitive processes rather than appraisal, such as motivational and physiological changes and the subjective emotional experiences [44]. Simulation Theory postulates that humans recognize emotions by simulating the other people’s emotional state in themselves. Although within the present study, we did not directly test these theories, they highlight complex cognitive processes involved in emotion recognition, which helps us to understand the findings.

A strength of the present study is a detailed analysis of the large range of cognitive performance which allowed us to show that recognition of emotions in facial expressions starts to decline already at the an early phase of impaired cognitive performance. Also a novel finding in our study was a curvilinear trend between the MMSE scores and recognition of several emotions; that is, the association between poorer cognitive function and poorer recognition of facial expressions accelerated along with the deterioration of cognitive function.

Limitations of this study should be noted. The cross-sectional study design prevents us from making inferences about the direction of the associations. Furthermore, the MMSE has been developed to provide a brief screening test that quantitatively assesses the level of cognitive impairment and cognitive changes over time, for example, among patient groups [23]. Thus, the MMSE does not provide a diagnostic tool to detect dementia which, however was not the focus of our study. Moreover, as negative emotions elicit a higher arousal level than the positive ones, arousal level might act as a mediator or effect modifier which should be examined in future studies. We had only one emotion with positive valence (i.e., happiness); the rest of the emotions had negative valence [1]. In the future, it would be important to examine whether equal number of emotions with negative and positive valence in the study would lead to similar findings with ours. Although an advantage of our study population was its large size, it represents an elderly, originally occupational cohort and therefore the findings may not be generalizable to the total population of the United Kingdom.

Keeping in mind these limitations, this study provides new evidence on the association between cognitive function and recognition of emotion in facial expressions. Because recognition of emotions in other people’s faces is a central aspect of social interaction, our findings add to understanding of the determinants of interpersonal problems and mental health in older people.

Supporting information