The cognitive basis of dyslexia in school‐aged children: A multiple case study in a transparent orthography

Abstract This study focuses on the role of numerous cognitive skills such as phonological awareness (PA), rapid automatized naming (RAN), visual and selective attention, auditory skills, and implicit learning in developmental dyslexia. We examined the (co)existence of cognitive deficits in dyslexia and assessed cognitive skills’ predictive value for reading. First, we compared school‐aged children with severe reading impairment (n = 51) to typical readers (n = 71) to explore the individual patterns of deficits in dyslexia. Children with dyslexia, as a group, presented low PA and RAN scores, as well as limited implicit learning skills. However, we found no differences in the other domains. We found a phonological deficit in 51% and a RAN deficit in 26% of children with dyslexia. These deficits coexisted in 14% of the children. Deficits in other cognitive domains were uncommon and most often coexisted with phonological or RAN deficits. Despite having a severe reading impairment, 26% of children with dyslexia did not present any of the tested deficits. Second, in a group of children presenting a wide range of reading abilities (N = 211), we analysed the relationship between cognitive skills and reading level. PA and RAN were independently related to reading abilities. Other skills did not explain any additional variance. The impact of PA and RAN on reading skills differed. While RAN was a consistent predictor of reading, PA predicted reading abilities particularly well in average and good readers with a smaller impact in poorer readers.

While this research has provided evidence of the specific role of phonological processing in dyslexia, it also showed that a single cognitive deficit is not sufficient to explain the heterogeneity in the dyslexic population. It is now widely accepted that the etiology of dyslexia is multifactorial and best explained by various risk factors that increase the likelihood of fulfilling diagnostic criteria. According to the Multiple Deficit Model (McGrath et al., 2020;Pennginton et al., 2006), neurodevelopmental disorders are explained by various genetic and environmental risk factors that increase the possibility of developing the disorder. However, it is not clear how these factors sum-up or interact in the case of dyslexia. Although evidence for clearly separable subtypes of dyslexia is missing, dyslexia is a heterogeneous disorder and individuals differ in underlying aetiologies. The different symptoms might at least partly be associated with different cognitive profiles.
Selection of deficits tested by researchers as a potential source of reading impairment is based on the major theories of developmental dyslexia. The most acknowledged is the phonological deficit theory (Ramus & Szenkovits, 2008;Snowling, 1995Snowling, , 1998. Support for the phonological theory comes from evidence that dyslexic individuals perform particularly poorly on tasks requiring phonological awareness (PA), that is, conscious segmentation and manipulation of speech sounds such as finding rhymes or phoneme deletion. PA at preschool ages is a good predictor for later literacy skills for speakers of different languages (e.g., Lonigan et al., 2000) including Polish (Krasowicz-Kupis, 2009). Intervention studies (Hatcher et al., 1994;Hulme et al., 2012, reviewed by: Melby-Lervåg et al., 2012 show beneficial effects of early phonological training on further reading skills, which provides further evidence for a causal relation between the two skills (Hulme & Snowling, 2009). The double deficit theory (Bowers & Wolf, 1993;Wolf & Bowers, 1999 also posits that phonological deficits are central to dyslexia but adds deficient naming speed as an independent source of reading dysfunction. Their combined presence is then responsible for more profound reading dysfunction. Difficulties with procedural learning or task automatization associated with cerebellar dysfunction may be linked to dyslexia (Nicolson & Fawcett, 2007). The cerebellum also plays a role in motor control including speech articulation, which, when dysfunctional, might lead to deficient phonological representations. However, Raberger and Wimmer (2003) showed that cerebellar deficit might be linked to co-occurring disorders, such as ADHD, rather than dyslexia. Therefore, the causal relationship between cerebellar dysfunction and reading is much less clear than in case of phonological processing.
Dyslexia may be also related to sensory deficits and two theories have been forwarded to make this connection. One that is based on difficulties in auditory processing, like rhythm perception (rise time theory; Goswami, 2011) or tone perception (anchoring theory; Ahissar, 2007), which in turn may interrupt phonological processing. According to the rise time theory, the patterns of stressed and unstressed syllables in language may be processed by the same neural mechanisms that are used for processing patterns of strong and weak beats in music, at least in childhood. Hence individual differences in phonological processing in language should be related to individual differences in nonlinguistic musical tasks based on the patterns of the beat distribution.

RESEARCH HIGHLIGHTS
• This study tests the (co)existence of cognitive deficits in dyslexia in phonological awareness, rapid naming, visual and selective attention, auditory skills, and implicit learning.
• The most frequent deficits in Polish children with dyslexia included a phonological (51%) and a rapid naming deficit (26%), which coexisted in 14% of children.
• Despite severe reading impairment, 26% of children with dyslexia presented no deficits in measured cognitive abilities.
• RAN explains reading skills variability across the whole spectrum of reading ability; phonological skills explain variability best among average and good readers but not poor readers.
As for the anchoring theory, Ahissar et al. (2006) noticed that dyslexics' detection of regularities in sound sequences is inefficient, which may also impair phonological processing.
The VA span hypothesis postulates atypical development of reading skills because of non-optimal grain size parsing mechanisms of orthographic inputs, for example, reduced quantity of information that can be processed at a glance simultaneously. Alternatively, the impact of attention deficit on dyslexia was linked to impairment in binding letters to words or difficulty in shifting attention during reading (Lukov et al., 2015).
The possible coexistence of several deficits in people with dyslexia has been considered and experimentally addressed in multiple case studies (Ramus et al., 2003;Sprenger-Charolles et al., 2009;White et al., 2006). These studies showed that the most frequent deficit among both children and adults with dyslexia was a phonological deficit, in line with the phonological theory of dyslexia (Ramus & Szenkovits, 2008;Snowling, 1995;Snowling, 1998). Nevertheless, in several studies, the phonological deficit was present in only 50% of the sample, and there was a group of participants with dyslexia without any additional deficit (Reid et al., 2007;Sprenger-Charolles et al., 2009;White et al., 2006, but see also Ramus et al., 2006). The phonological deficit was often found to coexist with other deficits (auditory, visual, or motor). It is also worth noting that rapid automatized naming (RAN) tasks were often included in phonological skills when opaque languages were studied (Ramus et al., 2003;White et al., 2006). Therefore, it is impossible to distinguish between the phonological and RAN deficits in some of the previous studies. However, this is not always the case, and a more recent study on French-speaking children (Saksida et al., 2016) differentiated between phonological accuracy (measured with standard phonological tasks: phoneme deletion and spoonerism) and phonological speed (assessed with RAN tasks). This study showed that these two deficits affected a much higher number of children with dyslexia than visual deficits and that phonological and RAN deficits coexisted in the vast majority of participants (Saksida et al., 2016). Importantly, in a transparent (Polish) orthography, RAN deficit was at the same time the most frequent one among adults with dyslexia (Reid et al., 2007).
The previous multiple case studies involved relatively small groups (Ramus et al., 2003;Reid et al., 2007;White et al., 2006). The small samples not only limited the statistical power of the between-group (Sassenberg & Ditrich, 2019;Szucs & Ioannidis, 2017) but particularly affected the method of finding deficits. Namely, a small control group cannot be representative for typical readers, and therefore the thresholds of deficits established based on the range of scores in such groups may not indicate real deficits. Additionally, some studies applied statistical procedures that resulted in boosting scores of the control group. In particular, these studies removed the lowest scores in the control group before calculating the ranges of typical scores (Ramus et al., 2003;Reid et al., 2007;White et al., 2006). Therefore, the chances of finding a deficit in participants with dyslexia were higher (as mean scores of typical readers were overestimated and standard deviations were underestimated). In fact, some of the deficits found could be false positives. The existence of false positives in the described studies is also suggested by a much higher number of subjects with dyslexia without any particular deficit found in a study that employed a bigger (n = 86) control group (Sprenger-Charolles et al.,

2009).
As most multiple case studies of dyslexia have been performed in individuals reading an opaque orthography such as English or French, their findings cannot be easily generalized to transparent languages (Landerl et al., 2013;Sprenger-Charolles et al., 2011). Moreover, previous studies on the subtypes of dyslexia in transparent languages (e.g., Heim et al., 2008;Jednoróg et al., 2014) focused rather on comparison of distinguishable subgroups of children with dyslexia than on estimating the prevalence of the deficits.
We aimed to explore the presence and coexistence of cognitive deficits in children with severe developmental dyslexia (n = 51) in a transparent (Polish) orthography. We take into account previous methodological and statistical problems from earlier multiple case studies of dyslexia. We wanted to explore (1)  Secondly, treating dyslexia as the low end of a continuum including normal reading skills, we wanted to test which cognitive skills are the best predictors of reading outcome when we took into account a large group of children with various reading skills. To do that, we analysed the impact of cognitive skills on reading abilities in children with a wide range of reading skills (N = 211). We expected the reading abilities to be positively related to the PA skills and RAN, as well as to the other auditory (rhythm perception, tone comparison), visual (selective attention, VA span), and cerebellar (implicit learning) skills known from other studies to have an impact on reading ability.

Participants
Participants (total N = 211) were school-aged children (  where they lived (mostly children from the urban area of Warsaw were included in the study) and parental education (the parents were more educated than in the general population).

Dyslexia diagnosis
We applied a standardized battery of tests used to diagnose developmental dyslexia (Bogdanowicz et al., 2009) to distinguish children with and without reading impairment from grade 3 onwards. Children selected as having a severe reading disorder (n = 51) presented both low reading accuracy (they scored below 16th percentile in a singleword reading test) and low reading fluency (they scored below the 16th percentile in at least one out of two tests: pseudoword reading or reading with a lexical decision). We applied a conservative inclusion criterion of dyslexia, based on both reading accuracy and fluency, to increase the reliability of our group assessment and expected effect sizes (as recommended in Ramus et al., 2008). Spelling was not included as a diagnostic criterion, but 42 out of 51 children assigned to the dyslexia group also had a spelling impairment (i.e., scored below 16th percentile in a writing to dictation test). This is reasonable since Polish is a relatively transparent orthography with higher grapheme-tophoneme regularity in reading than in spelling (Schüppert et al., 2017).
Children who scored above the cut-off point (16th percentile) in all reading tasks were assigned to the typically reading group (n = 71).
None of the control children were impaired in spelling. Children who scored low either in reading accuracy or reading fluency (but not in both; n = 81) and children attending the second grade who were too young to identify dyslexia (n = 8) were excluded from the analyses of dyslexia subtypes.
Selected groups of typical readers and children with dyslexia did not differ in age and FHD but differed in sex, nonverbal IQ, and SES (see :   Table 1). Therefore, we used these demographic variables as covariates in all analyses. We decided to include FHD and age in the regression model because they could show significant effects even if their mean values do not differ between groups.

Procedure
All tests were performed in two sessions, each lasting around 45 min.
The sessions were run individually and took place either in a quiet room at school or in a testing room at the Nencki Institute of Experimental Biology. The standardized reading, phonology, selective attention, and rapid naming tasks were performed with a paperpencil method. The rhythm perception, tone comparison, VA span, and implicit learning tasks were designed as animated computer games for this study. All animated visual stimuli were presented on ASUS laptops (BU400A-W3097X) with a 14-in. screen. All auditory stimuli were presented binaurally through headphones. Cedrus keyboards RB-540 (24 cm × 16 cm) were used for all computerized tests. The keyboards contained four active buttons (1.9 cm × 3.8 cm each) arranged in a cross shape. Depending on the task, the action buttons were limited to right and left, up and down, or all four were used in the task. A detailed description of all tools used in the study is available in Supplementary Material S1 and the available data on reliability of the tools is presented in Supplementary Material S2.

Data analysis plan
We divided our analyses into two parts to answer how cognitive skills are related to dyslexia and reading skills. First, we aimed to examine the  (Table S2.1).

Differences between control and dyslexia groups
We tested differences between the control and dyslexia group with a logistic regression model predicting group membership based on the cognitive skills (with SES, FHD, nonverbal IQ, and age used as covariates). The final group size without missing data points was N = 118 (51 with dyslexia, 67 typical readers).

Deficits in dyslexia
We aimed to identify one or more cognitive deficits in participants with dyslexia. For each cognitive skill, we chose a threshold for the deficit at the level of 1.65 SD (corresponding to the bottom 5th percentile) below the mean of the typically reading control group performance (cut-off point used in similar studies, e.g., Ramus et al., 2003;Reid et al., 2007;White et al., 2006). In contrast with the previous studies, we did not exclude any participants from the control group who For factors other than READING and PHONOLOGY, we controlled for age using linear regression. For every factor, we fitted a linear regression model predicting the factor level based on age. Then, we used regression residuals as age-controlled scores when calculating cognitive skill deficits in the dyslexia and control groups (participants' age was partialled out from the raw score). In the regression analysis on the whole sample of participants, we used raw scores, since age was included independently as control in the full regression model.

Differences between control and dyslexia groups
A summary of logistic regression results is given in Table 2. Estimated coefficients may be interpreted as an increase of log-odds (natural logarithm of P/(1-P), P -probability) of belonging to the dyslexia group assuming an increase of predictor value of 1 SD. We identified differences in cognitive skills between children with dyslexia and typi-TA B L E 2 Coefficients of logistic regression predicting subject group (control or dyslexia) based on seven cognitive skills factors and four controls (N = 118)

(Co)existence of deficits
We grouped children with dyslexia by their cognitive deficits (see Figure 2). The two most common deficits within the dyslexia group were: PHONOLOGICAL (n = 26 out of 51; 51%) and RAN (n = 13; 26%). These deficits coexisted in seven subjects (14% of the group with dyslexia). However, 19 subjects showed neither of these two deficits, five of whom had a deficit in IMPLICIT LEARNING and one in VA SPAN.
As the criteria used for deficit identification (−1.65 SD in the control group) differed from the criteria used for dyslexia diagnosis, one child assigned to the group with dyslexia presented no deficit in READING (z-score in READING −1.61 SD relative to the control group). The     Perhaps we observed no relation between reading skills and phonological skills in the lower end of the spectrum because of the limited range of low phonological scores.

Correlation analysis
Our results are in agreement with a previous multiple case study on English-speaking children (White et al., 2006) where the majority of children with dyslexia (52%) had a phonological deficit. These results were interpreted in favour of the phonological deficit explanation of dyslexia. However, in that study the results from RAN tasks (object naming and digit naming) were included in the phonology factor. Another study on prevalence of cognitive deficits in French children with dyslexia showed that if RAN was separated from the phonological skills (it was labelled phonological speed), a phonological deficit was noted in 92% of children with dyslexia, a RAN deficit was observable in 85% of the children, and 79% of the children had a double phonological-RAN deficit (Saksida et al., 2016). In our study, also treating RAN as a separate cognitive dimension influencing reading, we found only a weak correlation between RAN and phonological skills (see Swanson et al., 2003 for meta-analysis) after correction for multiple comparisons (r(201) = 0.18, p < 0.05), which suggests that these two skills should not be treated as a joined factor. The association between these two skills might depend on the developmental stage or the transparency of the orthography (de Jong & van der Leij, 1999;Kirby et al., 2010;Landerl et al., 2013;Wimmer et al., 2000).
Our results emphasize the independent role of RAN and phonological skills in predicting the reading level. This is in line with the double-deficit theory of dyslexia (Wolf & Bowers, 1999), where RAN and phonological skills are treated as two separate sources of reading difficulties and their combination leads to more severe impairment (Bowers & Wolf, 1993;Wolf & Bowers, 1999. Empirical evidence in favour of the double-deficit was provided in studies on transparent (Boets et al., 2010;Papadopoulos et al., 2009;Torppa et al., 2012) and more opaque orthographies (Miller et al., 2006 In our study, 14% of children with dyslexia showed an implicit learning deficit. This deficit was not accompanied by any other deficit in the case of five out of seven children. According to the cerebellar deficit hypothesis (Nicolson & Fawcett, 2007), alterations in the cerebellum may lead to impaired articulatory, phonological, and/or implicit learning skills, which in turn may contribute to the reading disorder. In our sample, only one child had coexisting phonological and implicit learning deficits. The clear distinction between phonological and implicit learning deficits may suggest that implicit learning is a skill that may explain a certain number of dyslexia cases independent of RAN and phonology.
Around 24% of children with dyslexia in our sample showed no apparent cognitive deficit in phonology, RAN, or implicit learning. Only a few of them present a single or coexistent deficit in other domains (attention, motor, auditory). The lack of deficits in almost a quarter of children with dyslexia might be surprising, as in the previous studies almost all participants with dyslexia presented at least one cognitive deficit (Ramus et al., 2003;Reid et al., 2007;White et al., 2006). However, in the study of Pennington et al. (2012) individuals (Ramus et al., 2003;Reid et al., 2007;White et al., 2006). There are also other possible explanations for the lack of cognitive deficits in 24% of children with dyslexia. Perhaps the children were tested too late. Some deficits might be present early in development (Carroll et al., 2016) but may not be detectable at school-age. Also, other cognitive deficits or environmental causes might interrupt learning processes that are not yet included in the main theories of dyslexia and were not controlled in the present study.
Finally, the criterion of −1.65 SD might be too strict to reveal the influence of all relevant deficits on dyslexia. Reading impairment may occur in the case of multiple weaker (instead of one stronger) deficits.
To explore this possibility, we examined the patterns of deficits while applying a much more liberal criterion of deficit at −1SD (Tables S3.1-S3.4). We noted that weak phonological and RAN deficits did not lead to dyslexia when not accompanied by each other. Among children who scored below −1SD in PA alone there were 11 typical readers (39%) and 17 children with dyslexia (61%), and among children with a weak RAN deficit, there was an equal number of typical (n = 10) and impaired readers. However, among children who scored below −1SD in phonological and RAN skills at the same time, there were 95% of children with dyslexia (n = 18) and just one typical reader.
This suggests that even if phonological and RAN deficits are weak but coexist, they may lead to dyslexia. This is plausibly the reason why we cannot detect more cases like the one mentioned above in the control group. The deficit cases in the control group are present as a consequence of the chosen statistical threshold (up to 76% of typical readers will present at least one score lower than −1SD when such a threshold is chosen). However, even if a child has a weak or even a strong deficit in one domain, it will not necessarily lead to reading problems. Nevertheless, the combination of phonological and RAN deficits seems to be damaging for reading skills in a transparent orthography. As for clinical implications, we cannot use the cognitive profiles to simply diagnose dyslexia instead of the reading level itself.
A good practice would be to test phonological and RAN deficits. If they coexist, likely, a child will also have reading difficulties. Similarly, early dysfunction in RAN and phonology may lead to dyslexia. Although our regression model was based on a wide set of cognitive skills, it explained only 30% of the variance in reading. This may seem surprisingly low. However, this is no different from some previous models which take into account cognitive skills like phonology or RAN (e.g., Compton et al., 2001;Landerl & Wimmer, 2008). More variance might be explained if something other than cognitive or socioeconomic factors were included in the models. Reading skills may be accounted for by the quality of education or even personality traits (Agler, Noguchi & Alfsen, 2019). On the other hand, there are studies with a larger group of participants where PA and/or RAN explained a higher portion of the variance in reading (e.g., Pennington et al., 2012, around 50%).
So, including more participants in the study may lead to an increase in prediction value of our model.
In summary, our results showed that, in a transparent orthography, phonological and rapid naming factors are the most reliable predictors of the reading level in children with dyslexia and a group of children with various reading levels. Our results suggest additive influence of rapid naming and phonology on reading. Rapid naming appears to be the most stable predictor of reading, regardless of the reading level whereas phonology explains variability in average and good readers but does not explain variability in poor readers.

CONFLICT OF INTEREST
None of the authors has a conflict of interest.