Developmental prosopagnosia (DP) is a neurodevelopmental condition characterised by a severe deficit in the recognition of facial identity, which is thought to result from a failure to develop the necessary visuo-cognitive mechanisms [1
]. Estimates suggest that DP affects approximately 2% of the adult population [2
] and 1–4% of children [3
]. While these figures are a product of the statistical techniques used to infer impairment [4
], they also result from the diagnostic protocols adopted in each study. The latter reflects the general state-of-the-art, whereby vast differences in inclusion criteria prohibit many cross-study comparisons. As a result, it is often unclear whether differences in cognitive presentation result from inconsistencies in neuropsychological screening or theoretically important heterogeneity between individuals.
The first case report of DP was presented by McConachie in 1976 [6
], and over 150 papers have subsequently been published. Increased interest in DP coincided with the development of the Internet and large-scale media coverage of the condition, allowing large numbers of people to contact researchers in the self-belief that they experience severe face recognition difficulties. A wide variety of techniques have been used to assess these individuals (for a recent review, see [7
]), with a minority of authors relying on self-reported anecdotal evidence alone (e.g., [8
]). While respectable associations have more recently been reported between multi-item subjective ratings of face recognition ability and objective measures (e.g., [9
]), it is nevertheless clear that some “typical” performers erroneously under-rate their face recognition ability and self-refer for DP assessment [10
]. Most authors therefore agree that self-rating or anecdotal evidence of DP should always be supplemented by objective measures of face recognition performance [4
], although the proportion of confirmed diagnoses versus false alarms has not yet been examined in a large group of individuals.
Using objective measures to identify cases of DP has followed the precedent set by the longstanding literature on acquired prosopagnosia. Traditionally, the acquired condition was confirmed using famous face tests (e.g., [13
]) or bespoke assessments composed of personally familiar faces (e.g., [15
]). These tests were later supplemented with standardized tests of unfamiliar face memory (e.g., using the Recognition Memory Test for Faces: [16
]) or perception (e.g., using the Benton Facial Recognition Test: [17
]). While famous face tests remain popular for the assessment of DP (e.g., [18
]), bespoke tests of personally familiar face recognition are labour-intensive to prepare for the typically larger samples of developmental compared to acquired cases, and it is also difficult to collect well-matched control data. Further, both the Warrington and Benton tests have been subject to some criticism, as successful recognition performance can be achieved using extra-facial cues alone (e.g., the hairline or clothing: [20
]). These tests have therefore been superseded by the Cambridge Face Memory Test (CFMT: [21
]) and Cambridge Face Perception Test (CFPT: [22
]). While the CFMT assesses participants’ ability to encode six previously unknown faces and to subsequently recognize these individuals from memory, the CFPT has no mnemonic demands, requiring participants to sort similar images of simultaneously-presented faces in relation to a target.
Together, tests of face perception, face memory and famous face recognition offer a theoretically-driven assessment battery, and acquired deficits have correspondingly been reported (and occasionally dissociated) at perceptual (e.g., [14
]), mnemonic (e.g., [14
]) and semantic (e.g., [25
]) stages of processing. Evidence that face perception can sometimes be preserved in acquired prosopagnosia has had particular theoretical impact, suggesting its independence from face memory. As such, this process is represented at an earlier, distinct stage in dominant theoretical models of face-processing (e.g., [26
]), and apperceptive and associative subtypes of acquired prosopagnosia have been proposed [28
Given that developmental disorders often present with heterogeneous profiles of cognitive and perceptual deficits (e.g., [29
]), and evidence supporting the independence of face perception and face memory has also been found in typical children [3
], it seems likely that cases of DP are similarly variable in their presentation. That is, we would not expect all individuals with DP to present with a significant perceptual impairment. However, evidence from developmental cases is sparse and contradictory, and it remains unclear whether the two processes unfold independently (for an overview, see [33
]). Although some adult DP case studies support the independence of face perception and memory (e.g., [34
]), larger case series disagree about the prevalence of dissociable deficits. For instance, Dalrymple and colleagues [35
] used conservative single-case statistics to report dissociations between intact CFPT and impaired CFMT performance in 6 out of 16 DPs, whereas a very recent study used group-based analyses on the same measures to suggest a much greater overlap in impairment in a larger sample of participants [36
]. Without additional data, it is difficult to reconcile the findings of these two studies given the differences in sample size and statistical approach.
Further, almost no work to date has considered whether isolated semantic-level impairments can present in DP, as has been reported for acquired cases. Bowles et al. [2
] raise this possibility in an individual who self-reported severe everyday face recognition difficulties and was impaired at famous face recognition, but achieved typical scores on the CFMT and CFPT. While it is possible that this person used compensatory strategies to obscure impairment on the latter two tests (see [21
]), evidence from another case study suggests that long-term memory for unfamiliar faces can be independently impaired in development. McKone et al. [37
] reported an individual who also achieved typical CFPT and CFMT-Aus (an alternate version of the standard CFMT) scores in the context of impaired famous face recognition, and additionally performed poorly on long-term recall (at 20-min and 24-h delays) of the CFMT-Aus faces. As this pattern of presentation has not yet been systematically examined, it remains unclear whether further, as yet unidentified, subtypes of DP may exist.
Understanding the different patterns of presentation that can exist within the DP umbrella has important implications for current screening protocols: many authors administer the CFPT, CFMT and a famous faces test at screening, and adhere to the recommendation that impairment should be noted on at least two tests for DP diagnosis [4
]. This recommendation for repeated testing is important to minimize false alarms, reducing “the chance that it happened by chance” ([14
], p. 945). Yet, the above review indicates that some individuals may genuinely only be selectively impaired on one test, and would not meet current criteria for DP—prohibiting both theoretical and clinical developments in our understanding and management of the condition.
The current study aimed to examine the substantive issues associated with the diagnosis of DP, as outlined above. We screened the largest sample of self-reported DP cases that has been reported to date, using the dominant battery of assessment tests that is currently used by most researchers in the field (i.e., the CFMT, CFPT and a famous faces test). First, we examined the proportion of cases that meet existing diagnostic criteria for DP (i.e., impairment on any two tests). Second, we examined whether specific patterns of dissociation that have been observed in acquired cases (i.e., preserved versus impaired face perception, and selectively-impaired famous face memory) occur in DP participants. Importantly, we also note the patterns of presentation that do not occur, and their implications for theoretical models of face recognition. As a supplement to these analyses, we also examine the utility of each diagnostic test across our target age range (18–78 years), and present UK age-appropriate norms that will be of widespread value to the field.
This investigation examined patterns of face recognition impairments in 165 adults who self-referred for DP. A total of 38.2% of the sample met standard diagnostic criteria of impairment on at least two out of three dominant tests. Conservative single-case analyses found dissociations between face perception and both unfamiliar and familiar face recognition tests in four individuals who met DP criteria (6.3% of the overall sample who met DP criteria). More surprisingly, a dissociation between impaired famous face recognition and both unfamiliar face memory and face perception was observed in a large proportion (25.5%) of the sample. While these individuals do not reach existing diagnostic criteria for DP, a number of possible interpretations of this finding have implications for developmental theories of face recognition, our understanding of DP, and existing diagnostic protocols.
First, we provide evidence to support the independence of face perception and face memory in DP, using conservative criteria that required face perception to be intact relative to impaired performance on both the CFMT and famous faces test. This finding supports parallel findings in acquired prosopagnosia (e.g., [24
]; see also [54
]) and previous DP case studies (e.g., [55
]), suggesting that face perception and face memory follow at least partially independent developmental trajectories that can be selectively impaired [32
]. Interestingly, the wider patterns of impairment reported here support linear accounts of face-processing (suggesting that early impairments in face perception likely result in downstream impairments in face memory): performance of the one individual who was only impaired at face perception did not dissociate from performance on either face memory test. Likewise, while there were two individuals who showed impairments on the CFMT and CFPT, but not the famous faces test, their performance with famous faces did not significantly dissociate from the CFMT or CFPT scores. In short, none of the 165 cases examined here show evidence of being able to compensate for abnormal perceptual processing or short-term face encoding, even with high levels of exposure to a face.
However, perhaps most interesting is the low prevalence of mnemonic cases of DP in our large sample (i.e., in contrast to the six out of 16 cases reported by Dalrymple et al.: [35
]), as traditionally defined by intact CFPT performance but impaired CFMT and/or famous faces scores. This may result from our conservative comparison of CFPT to both
CFMT and famous face performance, whereas Dalrymple and colleagues only compared to the CFMT. However, if we include the cases who showed a dissociation between intact CFPT performance and impaired CFMT scores (but not famous faces), only an additional four cases would have been included. On the one hand, the low prevalence of mnemonic cases reported here fits well with the recent work of Biotti and colleagues [36
], who used group analyses to suggest a widespread overlap between impaired face perception and unfamiliar face memory in DP—a finding that is also supported by our PCA. Together, these findings suggest that the CFPT and CFMT tap similar processes, and that unfamiliar face memory impairments in DP are mostly (but not always) underpinned by at least some weakness in face perception (affecting the encoding of faces; see also [19
]). However, it is likely that this widespread weakness in face perception may not exceed diagnostic cut-offs in many cases (31 of the 63 DPs who met existing diagnostic criteria in this study did not achieve impaired scores on the CFPT). While this finding may result from the poorer psychometric properties of the CFPT ([2
]; also see discussion below about age effects), it is equally likely that impairments in face perception are a matter of degree, rather than absolute. Such a suggestion is not new, and was offered by Farah [57
] in her account for apperceptive versus associative acquired visual agnosia. Indeed, Farah suggests that, when tested appropriately, most visual agnosia patients show perceptual deficits, but these reside on a continuum. It is therefore conceivable that the same pattern occurs in DP.
However, this explanation does not sit well with our finding of selectively impaired famous face recognition, compared to both CFPT and CFMT performance, in a large number of individuals. Indeed, if DP can be explained by a unitary hypothesis of suboptimal face perception skills, it seems unlikely that these deficits were not also detected by a test that is solely intended to tap face perception skills (the CFPT), nor a test of short-term unfamiliar face memory that contains an initial perceptual encoding phase (the CFMT). One could argue that successful performance can be achieved on both the CFPT and CFMT using suboptimal strategies (e.g., feature matching), and such compensatory mechanisms are less effective in a test of famous face recognition that uses single presentations of static images. However, this argument is countered by a lack of correlation between famous face and CFMT/CFPT scores in this group of individuals (see Figure 3
), and evidence of the reverse dissociation (i.e., intact famous face recognition that is significantly dissociated from impairments scores on both the CFPT and CFMT) in one DP. That is, an individual who performed within the typical range on a test that is more difficult to “cheat”, but not on two tests that are known to be susceptible to compensatory strategies. In addition, existing ERP evidence indicates an intact N250 component (thought to reflect activation of stored memory traces of individual faces) in six DPs viewing famous faces, but an impaired P600f component, thought to be linked to the semantic stages of identification [58
]. Likewise, previous work from our group has reported typical patterns of eye-movements in a DP with no known deficits in face perception when viewing famous faces [59
Alternatively, our findings can be seen to provide additional weight to previous indications from single-case studies that long-term face memory [37
], or memory for familiar faces [2
], may be selectively impaired. Strikingly, evidence of this dissociation in terms of impaired famous face recognition was not only observed in isolated cases in the current study, but in rather a large proportion of our sample: while 63 of the 165 individuals were only impaired at famous face recognition, this dissociation reached significance with regard to both CFMT and CFPT performance in 42 cases (reduced to 36 when cases that could be influenced by age-related decline in control norms were removed). However, it cannot be concluded from the available data whether at least some of these individuals represent a subtype (or even multiple subtypes) of DP that is not yet recognised, and further exploration is required to interpret the precise functional and cognitive underpinnings of the deficit. Indeed, the impairment may be in creating detailed and stable perceptual representations of familiar individuals, in retaining these representations in long-term memory, or in storing, linking or accessing semantic information about familiar people. Of course, all these possibilities may independently exist, and the potential diagnostic entities that reside between the processes tapped by the CFMT and famous faces test are currently unknown. One possibility that can be excluded here is that any famous face recognition deficits resulted from a developmental form of proper name anomia, as we accepted biographical information as a correct response in the absence of naming. While no developmental case of proper name anomia has been reported to date, it is also pertinent that we did not detect any potential candidates for the condition in the large sample reported here.
Importantly then, the above discussion casts the existing array of diagnostic tools and protocols into question. While many would agree that the field lacks a reliable test of face perception (e.g., [60
]), it is also generally well-accepted that intact scores can sometimes by achieved on the CFMT via compensatory mechanisms. Famous face tests are not without fault: they may be undermined by overly liberal cut-offs that result from ceiling effects and small SDs in controls, and can also be confounded by variations in lifetime exposure to target celebrities. In the current study, we attempted to address the latter by removing low familiarity trials from overall scores on an individual basis, and we used conservative case-by-case statistics to indicate both impairment and dissociation. Nevertheless, there were six false alarms in controls on the famous face test (compared to two on the CFMT and 11 on the CFPT), although this proportion of isolated famous face recognition impairments is clearly much lower than in the DP sample. One way of dealing with these issues is to allow more liberal cut-off scores for the CFMT and CFPT to capture cases of compensatory processing, while simultaneously making the famous face cut-off more conservative to ensure that “atypical” scores are truly impaired rather than suboptimal. It is likely that allowing more liberal cut-offs on the CFMT and CFPT would be controversial to the field, and may introduce a higher false-positive diagnosis rate. On the other hand, using more conservative cut-offs for the famous faces test may be more practical. Due to the format of the famous faces test, it is extremely unlikely that an individual can perform well by randomly guessing. Consequently, the effective “floor” of performance that can be meaningfully interpreted is much lower than in tasks which require forced responses, such as the CFMT and CFPT. Pertinently, Figure 3
indicates that even if we dropped the cut-off to seven SDs below the control mean, we would still find cases to support the impaired famous face dissociation compared to intact CFPT and CFMT performance. That is, some DPs scored −7 SDs below the control mean on the famous faces test, yet achieved positive z
scores on both the CFPT and CFMT). Not only do these individuals support the novel dissociation evidenced here, but also demonstrate the large range of impaired performance that can be tapped by a famous faces test, and perhaps used to grade the severity of face recognition impairments in a way that the CFMT cannot.
It is also pertinent that famous face tests have traditionally been the preferred means of diagnosing acquired cases of prosopagnosia, as impairments in familiar face recognition tend to be more striking and robust than those of unfamiliar face recognition. Indeed, the high mean scores and small standard deviations of control participants reported here (even for older participants) make very low scores particularly salient. It is therefore possible that famous face tests currently offer a more accurate assessment for prosopagnosia than tests of unfamiliar face recognition, perhaps due to (a) their greater ecological validity, (b) because familiar face recognition is less affected by vast individual differences in the typical population, (c) because familiar faces are less affected by extraneous cues and non-identity-related changes to images (e.g., viewpoint or lighting; [61
]) than unfamiliar face recognition, (d) they largely prohibit compensatory processing when non-iconic images are used, or (e) that we simply do not understand what tests of unfamiliar face processing are actually tapping. Thus, the findings reported here could be interpreted as evidence that famous face tests should carry relatively more weight in diagnosis, at least until we have a more fluent understanding of other diagnostic tools.
In the meantime, given general agreement that diagnosis should follow evidence of impairment on at least two objective measures [4
], the current findings suggest that some modification of existing protocols may be necessary. While face perception impairments may be more common than previously envisaged, they may not be severe enough to reach diagnostic cut-offs. Thus, at least two measures of face familiarity should be administered at the initial screening phase. However, given evidence of more highly prevalent selective impairments in famous face recognition, both short- and long-term face memory needs to be independently assessed, preferably each by two independent measures. Our finding that 17 control participants performed in the impaired range on any one test, and only one older individual performed below cut-off on multiple tests, supports this protocol. In terms of the testing battery itself, for now, an efficient solution would be to use the CFMT, CFMT-Aus [37
] and famous face tests at screening. The CFMT-Aus provides a second measure of unfamiliar face recognition (analogous to the standard CFMT), as well as a delayed recall measure that could be compared to famous face performance. Face perception may then be later assessed in those who meet criteria, if theoretically and/or practically relevant. Importantly, famous face tests could be modified from the format used here (and in most papers within the field), to include measures of face perception, familiarity, semantics, and naming.
However, one subgroup of individuals where famous faces tests are impractical for the diagnosis of DP is children, who are less likely to be exposed to a homogenous group of famous faces. Consequently, there is a pressing need to adapt existing diagnostic batteries for children to incorporate tests of long-term face memory (potentially through adapting current CFMT-based formats to include delayed recall measures, as in the CFMT-Aus). Incorporating these measures has significant practical benefits, as it would ensure that children and adults with DP are comparable in terms of their face recognition deficits, and facilitate cross-sectional and longitudinal studies in the area. Furthermore, effective assessment of long-term memory in children with face recognition impairments would allow researchers to develop a better theoretical understanding of the developmental trajectory of different aspects of face processing and identification.
Importantly, our findings underline the importance of objective testing in DP diagnosis. Even if we are very liberal and accept that the additional 42 individuals who were independently impaired at famous face recognition may represent an additional subtype of DP, approximately a third of the sample still failed to reach any inclusion criteria. It is likely that these people are either misjudging their face recognition ability, or that they experience less severe face recognition difficulties that are not captured by binary approaches to diagnosis. Indeed, it is possible that DP resides on the end of a single face recognition continuum, and this concept is also supported by the varied performance observed in our controls (see Figure 1
and Figure 2
). While further work needs to explore this issue and map the severity of real-world functional experiences onto diagnostic test scores, it is likely that a reliable mapping cannot currently be achieved. This may be because (a) existing assessment tasks are not capturing a critical aspect of real-world face recognition performance, and/or (b) some individuals (both those with and without impairment) have little insight into their face recognition skills. Until this issue is resolved, and diagnostic tasks are developed that prohibit compensatory strategies, it would be prudent for diagnostic approaches to heed both objective and subjective information (for a recent discussion, see [62
]). This may particularly be necessary for older adults, where self-report may need to carry more weight in diagnosis, alongside famous face and/or bespoke personally familiar face recognition tests. Indeed, the progressively lower means and larger standard deviations in older adults make it very difficult to infer impairment on the CFMT and CFPT (akin to the findings of Bowles et al.: [2
]), and this may have prohibited diagnosis of some of the older adults in our sample.