Social cognition broadly refers to the complex set of cognitive processes that allow us to perceive, process, and interpret social information (Adams et al., Reference Adams, Schweitzer, Molenberghs and Henry2019; Henry et al., Reference Henry, Von Hippel, Molenberghs, Lee and Sachdev2016; Poletti, Enrici, & Adenzato, Reference Poletti, Enrici and Adenzato2012). In many neurodegenerative disorders, deficits in core social cognitive abilities, such as theory of mind (ToM), empathy, social perception, and social behavior, are now recognized to be common presenting symptoms, and to be as, if not more important, than deficits in other neurocognitive domains (Christidi et al., Reference Christidi, Migliaccio, Santamaría-García, Santangelo and Trojsi2018; Henry et al., Reference Henry, Von Hippel, Molenberghs, Lee and Sachdev2016). This is because social cognitive deficits fundamentally disrupt the ability to build and maintain supportive social relationships, thereby eliminating the benefits that social interactions have for people living with neurocognitive impairments (see, Henry et al., Reference Henry, Von Hippel, Molenberghs, Lee and Sachdev2016). It is therefore unsurprising that social cognitive impairment has been argued to be a key predictor of many important prognostic outcomes including mental health, well-being, social integration, and quality of life more broadly (Christidi et al., Reference Christidi, Migliaccio, Santamaría-García, Santangelo and Trojsi2018; Henry et al., Reference Henry, Von Hippel, Molenberghs, Lee and Sachdev2016).

Alzheimer’s disease (AD) is the most common neurodegenerative disorder, with amnestic disturbances the most frequent initial presenting feature. However, in the mild to moderate stages of the illness, abnormalities in a number of distinct social cognitive domains are also evident. In particular, meta-analytic reviews show that there are consistent AD-related deficits in both social perception (Bora, Velakoulis, & Walterfang, Reference Bora, Velakoulis and Walterfang2016; Klein-Koerkamp et al., Reference Klein-Koerkamp, Beaudoin, Baciu and Hot2012) and ToM (Bora, Walterfang, & Velakoulis, Reference Bora, Walterfang and Velakoulis2015). Failures of social perception typically present as difficulties recognizing and responding to basic social and emotional cues, such as interpreting facial expressions, body language, or voices. Failures of ToM refer to difficulties understanding the mental states of others and appreciating that these mental states might differ from our own (Poletti et al., Reference Poletti, Enrici and Adenzato2012). In AD, these social cognitive deficits have been linked to both volumetric loss and white matter pathology (Dermody et al., Reference Dermody, Wong, Ahmed, Piguet, Hodges and Irish2016; Guntekin et al., Reference Guntekin, Hanoglu, Akturk, Fide, Emek-Savas, Rusen, Yıldırım and Yener2019; Kanske et al., Reference Kanske, Böckler, Trautwein and Singer2015; Lee et al., Reference Lee, Sun, Leung, Chu and Keysers2013; Park et al., Reference Park, Kim, Shin, Kim, Sohn, Park, Youn and Lee2017; Poletti et al., Reference Poletti, Enrici and Adenzato2012; Sturm et al., Reference Sturm, Yokoyama, Seeley, Kramer, Miller and Rankin2013). Consequently, it is unsurprising that both types of impairment grow more severe with disease progression (Fliss et al., Reference Fliss, Le Gall, Etcharry-Bouyx, Chauvire, Desgranges and Allain2016; Kumfor et al., Reference Kumfor, Irish, Leyton, Miller, Lah, Devenney, Hodges and Piguet2014; Sturm et al., Reference Sturm, Yokoyama, Seeley, Kramer, Miller and Rankin2013; Torres et al., Reference Torres, Santos, Sousa, Simoesneto, Nogueira, Belfort, Dias and Dourado2015).

However, while it is now well established that deficits in social perception and ToM are common presenting problems, less clear is how empathy – another core aspect of social cognition – is affected. Empathy is a multifaceted construct, which consists of both cognitive and affective components. Whereas cognitive empathy refers to the ability to understand another’s emotional state, affective empathy refers to one’s emotional response to the perceived situation of another (Bartochowski et al., Reference Bartochowski, Gatla, Khoury, Al-Dahhak and Grossberg2018). A deficit in cognitive empathy may manifest as an inability to comprehend what another individual is feeling, with direct implications for how one then behaves toward that individual. Alternatively, affective empathic responses reflect one’s emotional response to the perceived emotional states of others. An example here would be when an individual observes another person crying in distress, they may feel sad that they are unhappy. Importantly, these personal emotional reactions can be distinct from the other individual’s experienced emotions (i.e., when one feels embarrassed for someone who is overconfident; Henry et al., Reference Henry, Von Hippel, Molenberghs, Lee and Sachdev2016). In day-to-day living, an exaggerated or deficient affective empathic response has implications for interpersonal conduct and social relations with others. Since behavioral problems and breakdowns in patient–carer relationships are key predictors of institutionalization (Banerjee et al., Reference Banerjee, Murray, Foley, Atkins, Schneider and Mann2003; Spitznagel et al., Reference Spitznagel, Tremont, Davis and Foster2006), a clearer and more nuanced understanding of how empathic processing is affected by AD is required.

Consequently, it is of particular interest that in a recent qualitative review, Fischer, Landeira-Fernandez, Sollero de Campos, and Mograbi (Reference Fischer, Landeira-Fernandez, Sollero de Campos and Mograbi2019) argued that AD is associated with impairment in cognitive but not affective empathic processing. Indeed, while prior work has identified a core neural network that appears to underlie both components of empathy (Fan et al., Reference Fan, Duncan, De Greck and Northoff2011), studies have also identified distinct neural regions for cognitive and affective empathy. For instance, Fan et al. (Reference Fan, Duncan, De Greck and Northoff2011) found that overall empathy was associated with activation in the dorsal anterior cingulate cortex, anterior mid-cingulate cortex, supplementary motor area, and the bilateral anterior insula. The right anterior insula was more likely to be activated in affective empathy (the left anterior insula was activated for overall empathy), while the dorsal anterior cingulate cortex was more frequently activated during cognitive empathic tasks than for overall empathy. These findings align with Eres et al. (Reference Eres, Decety, Louis and Molenberghs2015) who found that affective empathy was associated with greater gray matter density in the insula cortex, whereas cognitive empathy was associated with greater gray matter density in the midcingulate cortex/dorsomedial prefrontal cortex. Thus, it is possible that these neural regions may be differentially affected by AD (Dermody et al., Reference Dermody, Wong, Ahmed, Piguet, Hodges and Irish2016; Eres et al., Reference Eres, Decety, Louis and Molenberghs2015; Uribe et al., Reference Uribe, Puig-Davi, Abos, Baggio, Junque and Segura2019). To date, there has been no meta-analytic study to quantitatively test the magnitude and significance of any AD-related effects in each of these two empathic components. This is therefore the next important step in this literature and the first aim of the present study.

For ToM, a distinction between the affective and cognitive subcomponents has also been identified. Whereas cognitive ToM requires an understanding of another’s mental states, affective ToM is concerned with understanding another’s emotional state (Heitz et al., Reference Heitz, Noblet, Phillipps, Cretin, Vogt, Philippi, Kemp, de Petigny, Bilger, Demuynck, Martin-Hunyadi, Armspach and Blanc2016). Although conceptually there are differences between affective ToM and cognitive empathy (see, Singer, Reference Singer2006) at the behavioral level of assessment, these two constructs are difficult to distinguish and their overlap has frequently been noted (e.g., Bensalah, Caillies, & Anduze, Reference Bensalah, Caillies and Anduze2016; Dodich et al., Reference Dodich, Cerami, Crespi, Canessa, Lettieri, Lannaccone, Marcone, Cappa and Cacioppo2016; Dvash & Shamay-Tsoory, Reference Dvash and Shamay-Tsoory2014; Moreau et al., Reference Moreau, Rauzy, Viallet and Champagne-Lavau2016; Preckel, Kanske, & Singer, Reference Preckel, Kanske and Singer2018; Shamay-Tsoory, Aharon-Peretz, & Perry, Reference Shamay-Tsoory, Aharon-Peretz and Perry2009; Shamay-Tsoory et al., Reference Shamay-Tsoory, Tomer, Berger, Goldsher and Aharon-Peretz2005). In particular, both affective ToM and cognitive empathy involve understanding another’s emotional state. Thus, these constructs will be considered interchangeable for the purposes of this paper. A deficit in cognitive ToM affects one’s ability to take the perspective of another (i.e., as shown by failures of false-belief understanding; Takenoshita et al., Reference Takenoshita, Terada, Yokota, Kutoku, Wakutani, Nakashima, Maki, Hattori and Yamada2018). This in turn affects the ability to relate to others as individuals may demonstrate a failure to reciprocate socially, even when obvious social cues are given. Cognitive ToM is different from cognitive empathy as the former represents one’s capability to infer another’s mental state (thoughts, beliefs, perspectives), while the latter is the ability to infer another’s emotional state (which may manifest as happiness, sadness, etc.).

In the only meta-analysis of the ToM-AD literature to date, Bora et al. (Reference Bora, Walterfang and Velakoulis2015) found evidence for an AD-related deficit in ToM function that was large in magnitude (d = 1.15). However, it was unclear whether this was attributable to problems with both or only one of the subcomponents, as no distinction was made between affective and cognitive ToM. Importantly, it has been found that in the mild to moderate stages of AD, key brain regions involved in overall ToM processing – the precuneus, temporal poles, and posterior superior temporal sulcus – are affected (Bailly et al., Reference Bailly, Destrieux, Hommet, Mondon, Cottier, Beaufils, Vierron, Vercouillie, Ibazizene, Voisin, Payoux, Barré, Camus, Guilloteau and Ribeiro2013; Ciaramidaro et al., Reference Ciaramidaro, Adenzato, Enrici, Erk, Pia, Bara and Walter2007; Enrici et al., Reference Enrici, Adenzato, Cappa, Bara and Tettamanti2011; Gomez-Isla et al., Reference Gomez-Isla, Hollister, West, Mui, Growdon, Petersen, Parisi and Hyman1997; Ramos et al., Reference Ramos Bernardes Da Silva Fiho, Oliveira Barbosa, Rondinoni, Dos Santos, Garrido Salmon, Da Costa Lima, Ferriolli and Moriguti2017; Ryu et al., Reference Ryu, Kwon, Lee, Yang, Kim, Song, Yang, Kim and Lee2005; Schroeter et al., Reference Schroeter, Stein, Maslowski and Neumann2009; Sebastian et al., Reference Sebastian, Fontaine, Bird, Blakemore, De Brito, McCroy and Viding2011; Stone, Baron-Cohen, & Knight, Reference Stone, Baron-Cohen and Knight1998). While some studies have provided evidence for AD-related impairment on one or both ToM components (Dodich et al., Reference Dodich, Cerami, Crespi, Canessa, Lettieri, Lannaccone, Marcone, Cappa and Cacioppo2016; Moreau et al., Reference Moreau, Rauzy, Viallet and Champagne-Lavau2016; Yamaguchi et al., Reference Yamaguchi, Maki, Takatama and Yamaguchi2019), others have identified no significant AD-related impairment for either (Heitz et al., Reference Heitz, Noblet, Phillipps, Cretin, Vogt, Philippi, Kemp, de Petigny, Bilger, Demuynck, Martin-Hunyadi, Armspach and Blanc2016; Kumfor et al., Reference Kumfor, Honan, McDonald, Hazelton, Hodges and Piguet2017) or have reported evidence of a disconnect between the two (Bertoux et al., Reference Bertoux, De Souza, O’Callaghan, Greve, Sarazin, Dubois and Hornberger2016; Fliss et al., Reference Fliss, Le Gall, Etcharry-Bouyx, Chauvire, Desgranges and Allain2016). For instance, one recent study reported that only the affective but not the cognitive component of ToM was impaired (Bertoux et al., Reference Bertoux, De Souza, O’Callaghan, Greve, Sarazin, Dubois and Hornberger2016), while a different study reported the reverse pattern of impairment (Fliss et al., Reference Fliss, Le Gall, Etcharry-Bouyx, Chauvire, Desgranges and Allain2016). The second aim of the present study was therefore to use meta-analytic methodology to quantitatively test the magnitude and significance of AD-related effects for these two components of ToM.

In sum, the present study was designed to provide the first meta-analytic integration of the AD literature focused on empathy and ToM, in which the cognitive and affective subcomponents of both of these abilities are differentiated. Analyzing empathy and ToM, both broadly and in terms of their subcomponents, will provide a clearer and more nuanced understanding of the types of social cognitive difficulties persons with AD exhibit. Based on the prior empirical studies in this literature (Bora et al., Reference Bora, Walterfang and Velakoulis2015; Dermody et al., Reference Dermody, Wong, Ahmed, Piguet, Hodges and Irish2016; Fischer et al., Reference Fischer, Landeira-Fernandez, Sollero de Campos and Mograbi2019), it was predicted that significant deficits would be identified for the cognitive, but not the affective component of empathy, and that deficits would also be evident for both components of ToM. Prior literature has shown that older adults often exhibit greater social cognitive difficulties than younger adults ( Grainger et al., Reference Grainger, Steinvik, Henry and Phillips2019, Reference Grainger, Rakunathan, Adams, Canty and Henry2020; Henry et al., Reference Henry, Phillips, Ruffman and Bailey2013; Sun et al., Reference Sun, Luo, Zhang, Li and Li2017). Because of this, age was included as a predictor variable in the meta-regression analyses. Additionally, broader cognitive decline and disease duration have been linked to greater social cognitive difficulties in AD and other clinical groups (Castelli et al., Reference Castelli, Pini, Alberoni, Liverta-Sempio, Baglio, Massaro, Marchetti and Nemni2011; Cuerva et al., Reference Cuerva, Sabe, Kuzis, Tiberti, Dorrego and Starkstein2001; Fliss et al., Reference Fliss, Le Gall, Etcharry-Bouyx, Chauvire, Desgranges and Allain2016; Henry et al., Reference Henry, Von Hippel, Molenberghs, Lee and Sachdev2016; Laisney et al., Reference Laisney, Bon, Guiziou, Daluzeau, Eustache and Desgranges2013). As such, these clinical variables were also included as predictors of ToM and empathy impairment.

Methods

Literature Search Strategy

The current meta-analysis was conducted in accordance with PRISMA guidelines (Moher et al., Reference Moher, Liberati, Tetzlaff, Altman and Group2009). A systematic literature search was conducted across PubMed, Web of Science, MEDLINE, and PsycINFO databases in August 2019. The search terms used were as follows: AD or dementia; in combination with, emotion perception, social cognition, social perception, emotion recognition, facial expression, prosody, ToM, pragmatic impairment, lie*, joke*, mentalising, non-literal language, sarcas*, empath*, perspective taking, Peer-Report Social Function Scale. This was accompanied by a backward citation search of studies identified in relevant meta-analyses and systematic reviews (Bartochowski et al., Reference Bartochowski, Gatla, Khoury, Al-Dahhak and Grossberg2018; Bora et al., Reference Bora, Velakoulis and Walterfang2016, Reference Bora, Walterfang and Velakoulis2015; Christidi et al., Reference Christidi, Migliaccio, Santamaría-García, Santangelo and Trojsi2018; Henry et al., Reference Henry, Von Hippel, Molenberghs, Lee and Sachdev2016; Poletti et al., Reference Poletti, Enrici and Adenzato2012).

Search Eligibility Criteria

Titles then abstracts were first screened to remove ineligible studies (i.e., studies with no mention of AD, systematic reviews or meta-analyses, animal studies, or studies with no healthy age-matched control group; see Figure 1). Full-text screening was then conducted by the first author. In instances where study inclusion was ambiguous, the second author was consulted.

Fig. 1.

PRISMA flowchart displaying study screening and selection process.

Studies were considered eligible for inclusion if (1) they compared AD participants to healthy age-matched controls, (2) the AD group consisted of more than one participant – (i.e., it was not a single-case study), (3) the required statistics to derive effect sizes were published or provided upon request, (4) the studies and their data were unique, (5) the study was published in a peer-reviewed journal in English, (6) the tasks assessed measured the participant’s cognitive/affective abilities of ToM and/or empathy, and (7) the studies were empirical. There was no restriction placed on the year in which a study had to be published.

Theory of Mind Eligibility Criteria

For cognitive ToM, eligible studies had to include a behavioral, informant-rated, or self-report measure that assessed mental state attribution. Eligible affective ToM/cognitive empathy measures had to assess the participant’s understanding of another’s emotional state. In this instance, an emotional state referred to a state of arousal which externally manifested as emotions (happiness, sadness, anger, etc.). Overall ToM was measured through combining scores from the cognitive and affective ToM measures.

Empathy Eligibility Criteria

Tasks were considered eligible for affective empathy if they included a self-reported, informant-rated, or behavioral assessment of one’s affective empathic capacity. Overall empathy was measured by combining scores from the cognitive empathy/affective ToM and affective empathy measures. For further detail on the categorization of these constructs, see Coundouris et al. (Reference Coundouris, Adams and Henry2020).

Data Extraction

Where available, the data extracted from the eligible studies included:

1. 1) Sample characteristics from the control and AD groups, such as sample size, means and standard deviations for age, cognitive status as indexed by the Mini Mental State Examination (MMSE), and, for the AD participants, disease duration.

2. 2) Means and standard deviations for cognitive and affective measures of empathy and ToM, or other data which allowed the precise effect sizes for these measures to be calculated.

When required data were unavailable, the corresponding author on the manuscript was contacted to request the data. The following studies were excluded because the data were not available (Martinez et al., Reference Martinez, Multani, Anor, Misquitta, Tang-Wai, Keren, Fox, Lang, Marras and Tartaglia2018; Poveda et al., Reference Poveda, Osborne-Crowley, Laidlaw, Macleod and Power2017; Rowse et al., Reference Rowse, McCarthy-Jones, Knowles, Corcoran and Bentall2013; Zaitchik et al., Reference Zaitchik, Koff, Brownell, Winner and Albert2006). In certain instances, data were unavailable upon request due to specific participant groups not being assessed on the measures of interest (Cuerva et al., Reference Cuerva, Sabe, Kuzis, Tiberti, Dorrego and Starkstein2001; Dodich et al., Reference Dodich, Cerami, Crespi, Canessa, Lettieri, Lannaccone, Marcone, Cappa and Cacioppo2016; Duclos et al., Reference Duclos, Bejanin, Eustache, Desgranges and Laisney2018; Fernandez-Duque, Baird, & Black, Reference Fernandez-Duque, Baird and Black2009; Youmans & Bourgeois, Reference Youmans and Bourgeois2010). If participants scored at ceiling or floor on any measure, data for both control and AD participants from the measure in question were not permitted to contribute to the analyses. This decision was made because these types of data have the potential to overestimate effect sizes, and the software used to conduct the analyses does not analyze data points with a standard deviation of zero (Castelli et al., Reference Castelli, Pini, Alberoni, Liverta-Sempio, Baglio, Massaro, Marchetti and Nemni2011; Gregory et al., Reference Gregory, Lough, Stone, Erzinclioglu, Martin, Baron-Cohen and Hodges2002; Le Bouc et al., Reference Le Bouc, Lenfant, Delbeuck, Ravasi, Lebert, Semah and Pasquier2012; Yamaguchi et al., Reference Yamaguchi, Maki and Yamaguchi2012; Zaitchik et al., Reference Zaitchik, Koff, Brownell, Winner and Albert2006). However, the overall study was not excluded if it reported appropriate data for other measures. When studies tested a single control group and more than one AD group (for instance, a mild AD group and a moderate AD group), data from the two AD groups were averaged together to derive one overall score for AD performance for each dependent measure of interest in the study (Cuerva et al., Reference Cuerva, Sabe, Kuzis, Tiberti, Dorrego and Starkstein2001). Some measures did not clearly differentiate cognitive and affective ToM (Fliss et al., Reference Fliss, Le Gall, Etcharry-Bouyx, Chauvire, Desgranges and Allain2016; Heitz et al., Reference Heitz, Noblet, Phillipps, Cretin, Vogt, Philippi, Kemp, de Petigny, Bilger, Demuynck, Martin-Hunyadi, Armspach and Blanc2016; Narme et al., Reference Narme, Mouras, Roussel, Devendeville and Godefroy2013; Scheidemann et al., Reference Scheidemann, Petermann and Schipper2016), and consequently contributed to analyses of overall ToM but not to either of the specific ToM subcomponents (see Table 1). Finally, all of the data extracted were independently cross-checked by a second coder. In instances where measure inclusion was ambiguous, the fourth author was consulted.

Table 1

Individual study characteristics

Note. AD = Alzheimer’s disease. AffTom/CogEmp = Affective ToM/Cognitive Empathy. FB = False Belief. RMET = Reading the Mind in the Eyes Test. IRI = Interpersonal Reactivity Index. TASIT = The Awareness of Social Inference Test. ToM = theory of mind.

^a Data from this measure were excluded from analyses due to ceiling effects.

^b Study had a second AD participant group who were excluded from analyses due to floor effects.

^c Measure was excluded from analysis due to insufficient data.

^d Other ToM or empathy tasks used in this study were excluded as they measured social perception.

^e Measure could not be separated into cognitive or affective ToM, so was used for overall ToM analyses but not subcomponent analyses.

^f Other measures of ToM that were excluded due to having no control data.

^g Study had data for two separate AD groups which were averaged together for analyses.

^h Authors separated RMET task into a cognitive ToM component and an affective ToM/cognitive empathy component.

ⁱ A measure of ToM was excluded because data were deemed inconclusive by authors.

Key background characteristics of the contributing studies are reported in Table 1. A number of studies assessed a single-dependent measure using multiple tasks (i.e., a study may have reported data for multiple separate ToM tasks). In these instances, overall mean effect sizes were calculated by averaging the data across measures to create a single value to ensure that each participant contributed only once to the estimation of each mean effect. The same participants were, however, allowed to contribute to different overall means (i.e., the same participants were permitted to contribute both to the mean effect for affective empathy as well as the mean effect for cognitive empathy). As this resulted in mean effects not being independent, these subcategories were not compared statistically. Instead, to interpret differences between mean effect sizes, differences in effect size magnitude, as well as the percentage of variance accounted for (PVAF) by having a diagnosis of AD versus being a control participant was referred to (see, Adams et al., Reference Adams, Schweitzer, Molenberghs and Henry2019; Coundouris et al., Reference Coundouris, Adams, Grainger and Henry2019).

Statistical Analyses

Meta-analyses were conducted using the Comprehensive Meta-Analysis Version 3 software. A Hedges g index of effect size was used (Borenstein et al., Reference Borenstein, Hedges, Higgins and Rothstein2013), and effect sizes were deemed small, medium, or large when their values were equal to or larger than 0.2, 0.5, or 0.8, respectively (Cohen, Reference Cohen1988). The effect sizes for each study were calculated so that positive values indicated a deficit in ToM or empathy for the AD group (relative to the control group). The random effects model was used as it better accommodates heterogeneous effect distributions (Lipsey & Wilson, Reference Lipsey and Wilson2001).

A series of meta-regressions were then conducted to test whether clinical variables (i.e., MMSE scores, disease duration) or age explained variance in any of the effects identified. For each of these analyses, a minimum of 10 data points were required for each relevant predictor variable and the social cognitive ability under assessment (see Higgins, Reference Higgins2008). For this reason, no meta-regressions were conducted with affective empathy as the dependent measure, as fewer than 10 studies contributed data to this subcomponent. Where sufficient data were available, meta-regressions were conducted assessing the influence of MMSE scores, disease duration, and age on overall ToM and empathy, cognitive ToM, affective ToM/cognitive empathy.

Testing for publication bias

“The file drawer problem” refers to the higher probability of significant results being published, relative to nonsignificant results (Easterbrook et al., Reference Easterbrook, Berlin, Gopalan and Mathews1991; Rosenthal, Reference Rosenthal1979). To assess whether or not publication bias influenced the results of the current study, four analyses were conducted. First, the Begg and Mazumdar rank correlation test was used to calculate the correlation between study size and effect size. Egger’s test of the intercept was then used to provide a test of the same bias using precision (the inverse of the standard error) to predict the standardized mean effect size. For both these tests, a nonsignificant correlation/intercept would suggest no presence of bias.

Additionally, the robustness of each mean effect size to any potential bias was assessed. Orwin’s “Fail safe N” was calculated to quantify the number of unpublished studies averaging null results that must exist to reduce the mean correlation past a specific threshold (0.10 in the present study). The trim-and-fill method was then used to establish how each respective mean effect size would change if any identified bias were removed.

Testing for within study bias

To assess any potential biases at the level of the individual study, the appraisal tool for cross-sectional studies (AXIS; Downes et al., Reference Downes, Brennan, Williams and Dean2016) was used. The AXIS was designed to evaluate the quality of cross-sectional studies on both methodological and more general issues. Although studies are assessed against 20 criteria that are classified as either yes (recorded as one) or no/unclear (recorded as zero), no numerical scale is provided, and thus interpretations involve degrees of subjectivity. Downes et al. (Reference Downes, Brennan, Williams and Dean2016) describe this subjectivity as a positive aspect of the tool as it allows for more flexibility during decision making, especially as the user can assess individual aspects of a study’s design.

Twelve of the AXIS criteria were met by the included studies, five criteria were met by the majority of studies, and three criteria were met by less than half of the studies appraised (see Table S1). Specifically, of the 31 studies, only one justified their sample size, nine ascertained their samples from an appropriate population base, and four undertook measures to categorize and address nonresponders. It is important to note, however, that issues of nonresponders were likely not an issue that needed to be addressed in the remaining 27 studies. Overall, while most of the studies satisfied the majority of the AXIS criteria, there were some limitations that should be taken into consideration when evaluating this literature.

Results

Literature Search

As can be seen in Figure 1, the initial literature search yielded 6117 results. After the removal of duplicates, 3643 articles remained, which were then subjected to title and abstract screening. Following this screening, 104 articles were subjected to a full-text review. Ultimately, 31 studies published between 2001 and 2019 met the inclusion criteria. This included 30 studies examining overall ToM, 14 studies examining overall empathy, 24 studies assessing cognitive ToM, 13 studies assessing cognitive empathy/affective ToM, and four studies assessing affective empathy. Overall, the studies included data from a total of 748 AD participants and 750 controls.

Participant Characteristics

An independent groups t-test indicated no significant difference between AD participants (M = 74.79, SD = 6.60) and controls (M = 71.73, SD = 7.08) on age (p >.05). The mean MMSE scores were 22.28 (SD = 2.98) and 28.79 (SD = 0.47) for AD and control participants, respectively. On average, AD participants had been formally diagnosed with the disorder for 3.31 (SD = 0.82) years.

Meta-analytic results

Table 2 reports the key results from the meta-analysis. Compared to healthy controls, people living with AD were significantly impaired in overall ToM, with this deficit large in magnitude (g = 1.03, K = 30; see Figure 2). People with AD were also significantly and moderately impaired in their overall empathic ability (g = 0.67, K = 14; see Figure 3). Examining the overlapping and distinct subcomponents of these constructs revealed that AD was associated with significant, and large-sized deficits in cognitive ToM (g = 1.09, K = 24) and affective ToM/cognitive empathy (g = 0.76, K = 13). However, no group differences were evident for affective empathy (g = 0.36, K = 4).

Table 2.

Mean effects for ToM and empathy subcomponents comparing participants with Alzheimer’s disease against healthy controls and tests for publication bias

Note. *p < .05, **p < .01, ***p < .001. CogEmp/AffToM = cognitive empathy/affective theory of mind. A positive effect size indicates that the control group performed better than the AD group. PVAF = percentage of variance accounted for by group membership. K = the number of studies. N = the number of participants that contributed to the effect. AD = Alzheimer’s disease. The Orwin fail-safe N threshold used was 0.10. Trim and fill: look for missing studies to left of mean; using random effects model. Imputed mean is random effects.

Fig. 2.

Forest plot for overall ToM, displaying effect size (Hedges g) calculated using a random effects model. Positive effect sizes indicate better performance by controls.

Fig. 3.

Forest plot for overall empathy, displaying effect size (Hedges g) calculated using a random effects model. Positive effect sizes indicate better performance by controls.

Meta-Regressions

The next step in the analyses was to test whether level of cognitive impairment (as indexed by the MMSE) was a significant predictor of AD-related impairment in overall ToM, both ToM subcomponents, and overall empathy. MMSE scores accounted for significant variance in overall ToM, cognitive ToM, affective ToM/cognitive empathy, and overall empathy (R ² = 0.44, between-group Q = 4.06, p = .04, K = 25, R ² = 0.65, between-group Q = 7.33, p = .007, K = 20, R ² = 0.26, between-group Q = 4.63, p = .031, K = 12, and R ² = 0.20, between-group Q = 4.44, p = .035, K = 13, respectively). Using the same meta-regression approach, it was then tested whether disease duration predicted impairment in overall ToM and cognitive ToM ability. Disease duration did not account for significant variance in either overall ToM or cognitive ToM (R ² = 0.15, between-group Q = 1.76, p = .184, K = 11, R ² = 0.05, between-group Q = 1.46, p = .226, K = 10, respectively).

A meta-regression was then conducted to assess whether the age of the AD participants accounted for variance in their overall ToM and empathy, as well as their subcomponents (excluding affective empathy, for which K < 10). No effect of age was identified for overall ToM, cognitive ToM, affective ToM/cognitive empathy, or overall empathy (R ² = 0.00, between-group Q = 0.23, p = .632, K = 29, R ² = 0.00, between-group Q = 0.17, p = .681, K = 23, R ² = 0.00, between-group Q = 0.13, p = .717, K = 13, and R ² = 00, between-group Q = 0.00, p = .965, K = 14, respectively). These data therefore suggest that at least some of the heterogeneity between the contributing studies is attributable to variation in levels of cognitive impairment linked to AD, but that this is not the case for either disease duration or normative age.

Heterogeneity of Effect Sizes

Cochran’s Q was used to measure the extent to which the studies contributing to each subcomponent’s mean could be regarded as homogenous. As reported in Table 2, Cochran’s Q statistics for all mean effects were significant, indicating that there is meaningful variance between the studies that are contributing to each mean (Borenstein et al., Reference Borenstein, Hedges, Higgins and Rothstein2013).

Testing for Publication Bias

The tests used to quantify publication bias (Begg’s and Egger’s methods) suggested a potential presence of publication bias for overall ToM, but not for either of the two ToM subcomponents, overall empathy, or affective empathy. Orwin’s “Fail safe N” indicated that for overall ToM and empathy, 256 and 76 unpublished studies averaging null results would be needed to reduce the mean effect size value past 0.1 (respectively). Additionally, the trim-and-fill method indicated that the mean effect size for all constructs did not change (see Table 2 for further information). Therefore, these data suggest a potential publication bias for overall ToM, but not for overall empathy, affective empathy, cognitive ToM, or affective ToM/cognitive empathy.