We searched for studies using standardized, normatively defined neuropsychological tools in the screening, diagnosis or assessment of KS, ARD, and ARBD. Studies assessing alcohol-dependent persons without one of the above ARCI diagnoses were omitted as the focus of the review was on the ability of neuropsychological tests to identify and assess those with clinically significant cognitive impairment (i.e., meeting diagnostic criteria) and to differentiate these individuals from those with little or no impairment. We made two deviations from the initial aim of including only standardized, normatively defined neuropsychological tests. The first to include studies employing standardized observational measures of confabulation which are not normatively defined, enabling a more comprehensive review of ARCI assessment tools. The second to include a study evaluating the utility of the Brown-Peterson paradigm (a non-standardized test with normative data available) to allow a more detailed discussion of working memory assessment. We also deviated from our protocol by using the term ARCI over ARBD to frame this review as the former is more widely used outside of the UK.

The following databases were searched on January 19th 2016 and again in August 2018: MEDLINE, EMBASE, Psych-INFO, ProQuest Psychology, and Science Direct. To achieve both aims of the review the process of study selection and subsequent data extraction was divided into two consecutive phases. In the first phase, we identified 105 studies meeting our original eligibility criteria (i.e., standardized test used to assess ARCI). From these, we extracted only the type of ARCI sample assessed and the tool used, addressing aim 1 (i.e., identify the tools used). In phase two, 43 studies were identified from within this selection which provided some evaluation of the psychometric, diagnostic or practical merits of the tools used. These studies were extracted in full using a table designed by the authors and presented in our protocol (Heirene et al., 2016). Extracted data from phase two was used to produce a narrative synthesis of findings focused on addressing aim 2 (i.e., evaluation of the measures) (see Figure 1 for an overview of the study selection process).

Title and abstract and full-text screening were conducted independently by two authors (RH & GR-D) via Covidence online systematic review software. A pilot full-text screen of 20 texts was conducted to check inter-reviewer consistency. A Kappa statistic of .839 (P <.001) was achieved from this process, indicating excellent consistency between reviewers (Viera and Garrett, 2005). Phase two of extraction and study appraisal was also completed independently by two authors (RH & BJ), though phase one extraction was carried out by one author (RH).

All studies extracted in full were appraised using the quality assessment checklist designed by the authors and presented in our protocol (Heirene et al., 2016). We made minor updates to this checklist to provide a less ambiguous and more comprehensive assessment of study rigor (see Supplemental Document 1). We discuss the outcomes of the study appraisal at the end of the findings section, though we also present multiple indices of study quality in the tables of findings (e.g., ensuring a sufficient period of abstinence achieved prior to testing) and highlight any notable methodological weaknesses or strengths of studies when discussing their contribution to the review.

As anticipated in our protocol, meta-analysis was not appropriate for the synthesis of findings due to heterogeneity in the tests used, the samples of focus, comparator groups, and the outcomes reported. Instead, findings are synthesized in the form of a narrative review, with tables used throughout to present key outcomes. Findings are divided into five sections covering cognitive screening instruments, memory, EF, intelligence and test batteries, and premorbid intelligence. Where relevant and possible, effect sizes were calculated for between-group comparisons (e.g., ARCI vs. controls) to supplement review findings (Sullivan and Feinn, 2012). Cohen's d was selected as a commonly used effect size which calculates the standardized difference in means between groups and can facilitate the relative comparisons of each test's sensitivity (Larner, 2014). Based on Cohen's (1992) original suggestion, effects are classified as small (d = 0.2), medium (d = 0.5), or large (d ≥ 0.8). However, these cut-off points are somewhat arbitrary. All effect sizes should be considered in conjunction with the methodological quality of the study (e.g., sample size, abstinent participants) before reaching conclusions regarding the size of the effect or the implications of this for the test used. Sensitivity and specificity values for tests were also calculated where relevant; though few studies provided the requisite information (all reviewer calculated outcomes are presented inside squared brackets [] in tables).

The 105 studies included in phase one of data extraction ranged in year of publication from 1968 to 2017. Ninety studies assessed KS participants, 15 ARD, six ARBD, and four included samples described as having mild alcohol-related impairment not fitting KS or ARD criteria (ALC), one of which was diagnosed according to the DSM-5's alcohol-related Mild Neurocognitive Disorder. The most commonly used tests are presented in Table 1 according to each domain of cognitive assessment discussed in the narrative synthesis (for a complete list of all neuropsychological tests used within the reviewed studies see Supplemental Document 2).

The final 43 studies included in phase two of data extraction spanned a range of 40-years from 1977 to 2017. Consonant with the studies included in the first phase of extraction, the majority focused on KS participants (n = 39), with few investigating ARD (n = 4), ALC (n = 4) or those with a diagnosis of ARBD (n = 2). Adding to the confusion surrounding the use of ARBD as a specific diagnosis that was highlighted in the introduction, neither of the two studies which assessed participants with this diagnosis (i.e., Welch et al., 1997; Horton et al., 2015b) presented criteria for the condition or clearly described the definition of ARBD which they adopted. However, it appears that Welch et al. (1997) used the term to refer to milder neurocognitive impairment not meeting criteria for WKS, while Horton et al. (2015b) used the term more broadly; though it is unclear whether this included individuals with WKS diagnoses.

Most studies employed a between groups design (n = 36), though four assessed only a single group, two assessed multiple clinical groups individually, and one was a case-study. Various outcomes of relevance to the review were reported, with the most common relating to the diagnostic value of tests, including sensitivity and specificity values (n = 23); the use of tests for discriminating between groups (n = 13); positive and negative predictive values (n = 1); and likelihood ratios (n = 1). Twelve studies commented on the practical values of tests, including ease or time of administration or the availability of parallel versions. Outcomes relating to validity were common, including convergent (n = 11), construct (n = 11), predictive (n = 2), concurrent (n = 1), and content validity (n = 1); while only four studies reported reliability values (inter-rater = 2; test-retest = 1; internal = 1).

Five reviewed studies evaluated Cognitive Screening Instruments (CSIs) in those with ARCI (Table 2 presents a summary of test and study details along with key outcomes). The Montreal Cognitive Assessment (MoCA; Nasreddine et al., 2005) was assessed by two studies and has received the comparatively greatest psychometric evaluation. In Wester et al. (2013c), the MoCA effectively differentiated between KS and ALC groups, between KS and controls, and between ALC and controls, mostly all with good to excellent sensitivity and specificity levels provided the cut-off score was adjusted accordingly. Only the KS vs. ALC MoCA comparison did not reach optimal 80% sensitivity and 60% specificity criteria cited by the authors (Blake et al., 2002).

Further support for the MoCA has come from Oudman et al. (2014), who directly compared the psychometric properties of the MoCA and the Mini-Mental State Examination (MMSE; Folstein et al., 1975) in a KS and control sample. Overall, both the MoCA and MMSE demonstrated good to excellent sensitivity and specificity, with the MoCA the superior of the two. Reviewer calculated effect sizes support this finding, indicating that the MoCA produced substantially larger relative differences when comparing controls and KS than the MMSE. Additionally, a MoCA cut-off score of <23 produced nearly perfect positive and negative predictive values and likelihood ratios in Oudman et al. (2014). However, while these values support the screening capabilities of the tool, they should be interpreted with caution as the equal proportion of KS to controls investigated by Oudman and colleagues is unreflective of typical clinical environments, and thus the values would likely decrease in such settings. As in Wester et al. (2013c), adjustments to the cut-off score of both CSIs were required by Oudman et al. to result in the best possible discriminatory abilities. Poor sensitivity of the MMSE to KS warranted a rise in the standard cut-off score of <24 to between 25 and 27 to produce optimal sensitivity and specificity levels, with <26/27 being most able to discriminate between KS and controls (Wester et al., 2013c). For the MoCA, the combined findings of Oudman et al. and Wester et al. suggest a cut-off score of ≤ 23 is needed to distinguish between unimpaired individuals and KS, ≤ 24 for unimpaired vs. ALC, and ≤ 20 for ALC vs. KS. Nonetheless, caution should be exercised when reducing or increasing cut-off scores to be conscious of increased false negatives and false positives, respectively.

Although Oudman et al.'s (2014) findings suggest the MMSE may have value in screening for ARCI, the test's utility is restricted by the absence of items specifically indexing executive abilities. Additionally, the memory component of the MMSE has been criticized for its simplicity and lack of sensitivity to alcohol-related memory disorders (Squire and Shimamura, 1986; Kopelman et al., 2009). The cognitive assessment component of the CAMDEX (Roth et al., 1986), the CAMCOG, includes all items from the MMSE in addition to more difficult tests of memory and measures of EF, and thus may be more suitable for ARCI screening. However, Deary et al. (1991) found several of the CAMCOG subtests (7 of 11) correlated strongly with the Digit Symbol Substitution test of visuospatial processing, which they stated indicates high redundancy within much of the CAMCOG. Still, research evaluating the CAMCOG in this population is scant–and nonexistent for the updated CAMCOG-R (Roth et al., 1999).

In total, 22 studies evaluated tests of memory, which are further divided into tests of episodic, autobiographical, procedural, and working memory and confabulation.

Six studies evaluated tests of EF, all in KS participants (see Table 8).

Two studies found evidence for the sensitivity of the ecologically-based Behavioral Assessment of the Dysexecutive Syndrome (BADS; Wilson et al., 1996) to KS-related impairments. When comparing KS and ALs, the Six Elements and Rule Shift sub-tests have produced particularly large effect sizes (η²), approaching those observed on memory testing (Doors & People; Maharasingam et al., 2013). Indeed, when converted to Cohen's d for comparison with the other effect sizes calculated here, the Six Elements effect size (KS vs. AL) was the largest of any observed in the review (d = 6.0; Maharasingam et al., 2013). When employing the BADS it is important to observe individual sub-test scores as only 40% of KS were classed as impaired on the overall age-corrected score in van Oort and Kessels (2009), although this figure rose to 87% in Maharasingam et al. (2013).

The Wisconsin Card Sorting Test (WCST; Heaton et al., 1993) appears highly sensitive to KS-related EF deficits (Shoqeirat et al., 1990). A shortened and simplified version of the WCST, the Modified Card Sorting Test (MCST; Nelson, 1976), also appears sensitive, although to a lesser extent. Comparing KS performance on both the full WCST and Nelson's version, Shoqeirat et al. (1990) found the WCST to be more sensitive to severe executive dysfunction and therefore better able to discriminate between KS and other amnesic groups. The authors stated that this cannot simply because the full WCST is harder (suggesting differences in test procedures are responsible) and advocate investing the additional time required to employ the full version over the MCST. A third rule derivation test which has been used to assess ARCI is the Brixton Spatial Anticipation Test (BSPAT). In a relatively large sample of KS (n = 41) participants, Van Den Berg et al. (2009) found the BSPAT demonstrated good sensitivity and specificity when discriminating between KS and controls. The measure may be particularly sensitive to the executive deficits associated with ARCI as the KS group performed significantly worse than multiple other cognitively impaired groups.

Tests of cognitive estimations are also impaired in those with KS (Shoqeirat et al., 1990; Kopelman, 1991), although whether this reflects executive dysfunction remains uncertain. Impaired Cognitive Estimation Test (CET) performance in those with KS may also result from semantic memory impairments and a failure to “error check” the estimation (Taylor and O'Carroll, 1995). Moreover, despite impairments on the CET, the cognitive estimation component of the BADS (i.e., Temporal Judgement) appears the least sensitive subtest to KS (Van Den Berg et al., 2009; Maharasingam et al., 2013); though, this may be due to the use of only four questions, compared with 10-15 in the CET.

Two reviewed studies examined the inter-correlations of tests of single EFs in those with ARCI in attempts to see where scores converged or diverged. The CET does not appear to correlate with the WCST, MCST, or FAS verbal fluency test (Shoqeirat et al., 1990; Kopelman, 1991), and moderate correlations have been observed between the MCST and FAS (Kopelman, 1991), but not consistently and not between the WCST and FAS (Shoqeirat et al., 1990). The absence of correlations likely reflects the distinct executive skills used to complete these diverse tasks.

In total, ten of the studies investigated intelligence or battery tests (see Table 9). The Wechsler Adult Intelligence Scale (WAIS; Wechsler, 1958) was investigated in its varying editions by nine studies, often in conjunction with its counterpart, the WMS. The WAIS-WMS combination brings multiple benefits in terms of its extensive assessment of cognition, shared norms, and comparisons both within and between the tests. Though, Taylor and Heaton (2001) found 47% of the normative sample scored within the impaired range on at least one factor score (e.g., Working Memory, Verbal Comprehension etc.) from the combined Wechsler battery and 14% were impaired on at least two, indicating poor specificity. This highlights the need to identify a disease-specific profile for ARCI using this test combination. Historically, this has come in the form of a substantially impaired WMS score compared with WAIS Full-Scale IQ (FSIQ) for KS. The discrepancy between these scores has been found to be greater for KS than ALs, those with various neuropsychiatric disorders, and controls (Oscar-Berman et al., 1993; Taylor and Heaton, 2001). However, impaired FSIQ in those with ARCI has also been found, primarily due to poor performance on tests within the Performance IQ (PIQ) component (Taylor and Heaton, 2001). As a result, the discrepancy between the relatively preserved Verbal IQ (VIQ) component of the WAIS and the MQ may be a more effective indicator of KS (Alekoumbides et al., 1987; Charter and Alekoumbides, 1988). However, caution should be exercised in using these discrepancies for the diagnosis of ARCI for reasons outlined in the discussion.

The specific WAIS tests on which KS have demonstrated impaired performance include the Digit Symbol Substitution Test (DSST), Block Design, Picture Completion and Object Assembly (Malerstein and Belden, 1968; Glosser et al., 1977; Harbinson, 1984; Oscar-Berman et al., 2004). Of these, the DSST and its replacements have shown the greatest sensitivity to ARCI (Glosser et al., 1977; Harbinson, 1984) and have consistently differentiated KS from ALs, who are (to a lesser extent) also impaired on the test (Oscar-Berman et al., 2004). DSST performance appears to depend on a combination of new learning (Kapur and Butters, 1977), psychomotor speed, and visuospatial processing (Glosser et al., 1977), explaining its sensitivity to ARCI.

The CANTAB computerized battery assessment was investigated by one of the included studies. Horton et al. (2015b) found those with ARBD were impaired relative to norm scores on various CANTAB tests, particularly the Rapid Visual Processing (RVP) test of sustained visual attention; though the authors suggest this may reflect a processing speed deficit rather than one of attention. Floor effects were also observed on the RVP, suggesting this test may be limited in assessing the spectrum of ability in those with ARCI. However, the application of Horton and colleagues' findings is restricted as they failed to clearly outline the definition of ARBD they adopted and the criteria by which this diagnosis was made. They also omitted whether their sample comprised individuals with differing diagnostic labels (e.g., WE, KS, ARD, and/or ALC), as has been reported by other studies of participants with “ARBD” (e.g., Wilson et al., 2012).

Three studies investigated methods of estimating premorbid ability (all attempted to predict IQ) in those with ARCI (see Table 10), providing some insight into the value of the three approaches typically used: [1] predictions based on demographics (e.g., education, gender, ethnicity), [2] current performance on tasks believed to be relatively impervious to neurological damage, and [3] a combination of these factors. Demographically based predictions may be useful when assessing individuals with ARCI as no testing is required and thus estimates are not influenced by participant effort or the effects of alcoholism and associated factors (e.g., head injuries). O'Carroll et al. (1992) found demographically based estimates strongly correlated with those derived by the National Adult Reading Test (NART) of premorbid IQ. However, Bright et al. (2002) found NART and NART-R based estimates accounted for substantially more of the variance in current WAIS FSIQ than demographic variables, suggesting the NART may be the more accurate predictor.

Crawford et al. (1988) have recommended the use of the NART in those with ARD after finding little difference between ARD NART scores and those of demographically matched (age, gender, education) controls. The authors also, however, suggested the NART may be unsuitable for use with KS as it produced a significantly lower IQ than was estimated for a matched control group. Nonetheless, Crawford et al. (1988) advocated the NART for use with KS over the WAIS Vocabulary test, which produced lower estimates again. O'Carroll et al. (1992) have also questioned the validity of using the NART with KS based on finding its estimates were significantly below those of demographically-derived predictions. Drawing on their findings and those of Crawford and colleagues, O'Carroll et al. (1992) speculated the impaired NART performance in KS may result from executive dysfunction, suggesting that individuals with KS will quickly read aloud the words according to standard phonetical rules of pronunciation, without the necessary error checking process required for irregular word pronunciation. However, the processes underlying impaired NART performance in KS are yet to be fully elucidated, and those with frontal lesions appear to perform comparably with matched controls (Bright et al., 2002; MacPherson et al., 2017), questioning the role of executive dysfunction.

In contrast to both Crawford et al. (1988) and O'Carroll et al. (1992), Bright et al. (2002) found no significant difference between a larger sample of KS participants and matched controls on both the NART and NART-R. Moreover, the NART-R produced a smaller discrepancy between its scores and those derived from demographic variables than the original NART, indicating an improvement in predictive accuracy. Finally, Bright et al. found no additional benefit of combining NART scores with demographic variables to better estimate premorbid function in KS, other various neuropsychiatric disorders or controls. However, some value in combining these approaches has been documented in the wider literature (e.g., Vanderploeg and Schinka, 1995).

The summarized results of the study quality appraisal are presented in Table 11 (see Supplemental Document 1 for a tabulated overview of the entire quality assessment outcome). The quality assessment of studies revealed that the description of ARCI samples was frequently poor, with many authors failing to state whether they were abstinent, if they were diagnosed according to nosological criteria (e.g., DSM or ICD), how they were diagnosed, and whether those with complicating comorbid disorders (e.g., dementia) were excluded. Additionally, one reviewed study (Weintraub et al., 2000) and several more identified during the study selection process failed to state whether their sample's KS was alcohol-related or caused by some other etiology (e.g., bariatric surgery), requiring reviewers to contact the authors directly. The distinction between etiologies is important as the two may result in different cognitive profiles (Nikolakaros et al., 2016). Similarly, the only two studies which assessed participants with a specific diagnosis of ARBD provided little participant description, did not refer to diagnostic criteria, and one failed to state whether the sample was inclusive of those with KS or ARD diagnoses (i.e., Horton et al., 2015b).

Methodological and statistical errors were also frequent. Firstly, there was a risk of circularity in studies exploring the sensitivity of tests as many authors reported using neuropsychological assessment to inform diagnostic decisions, thereby inflating the apparent diagnostic value of the test observed on repeat testing. This was noted by Wester et al. (2013c) as a particular issue in their evaluation of the RBMT-3's sensitivity to KS, though was avoided by Wester et al. (2013b) in their appraisal of the MoCA in KS and ALC participants. Secondly, while reviewer calculated effect sizes were typically large between ARCI and control samples, many studies included small sample sizes (72% investigated <25 ARCI participants) suggesting they may have lacked the statistical power to detect more subtle between-group differences (type-II error) or produced misleading statistically significant results (type-I error; Button et al., 2013). Thirdly, most studies failed to report effect sizes when conducting appropriate analyses. While this can often be easily calculated by readers (as done here), multiple authors in the included studies commended the discriminatory value of a test based solely on statistical significance between groups (e.g., Oscar-Berman et al., 2004), which does not provide information regarding the relative size of the discrepancy. Given the additional information provided by their inclusion and the limitations of P values (see Sullivan and Feinn, 2012), we recommend all future studies in this field report effect sizes where relevant.

CSIs are useful in providing a preliminary indication of the severity of cognitive impairment. Both the MoCA and MMSE have demonstrated some value in distinguishing those with ARCI from those without and between gradations of alcohol-related impairment, though the MoCA appears the superior of the two (Wester et al., 2013c; Oudman et al., 2014). Although, clinicians using CSIs with this population should be cognizant of the considerable risk of false-positive and false-negative outcomes and avoid making definitive diagnostic decisions purely on the basis of their use.

In the wider literature, additional support for the MoCA comes from Pelletier et al. (2016) who found it proved useful in monitoring improvements in cognitive function in ALs undergoing rehabilitation. However, it is not clear whether the authors' findings simply reflect a learning effect, given the short duration (M = 33.5 days, SD = 7) between the two administration points. CSIs have also been criticized in the wider literature for their poor sensitivity to alcohol-related memory disorders (Benedict and Brandt, 1992; Munro et al., 2001). Consequently, additional memory testing may be useful during initial assessments. Additionally, CSIs may be of little use in discriminating between ARCI and other neurocognitive disorders such as Alzheimer's, with little observable difference in MMSE scores between groups (Osuntokun et al., 1994; Saxton et al., 2000). Overall, while CSIs provide quick, easily administered assessments of cognitive function, there is a need for more comprehensive testing following screening to ensure correct diagnosis.

The testing of EF represents an important element of ARCI assessment, particularly as impairments may have deleterious consequences for rehabilitation and treatment efforts (Bates et al., 2002, 2005). Executive dysfunction is now an accepted feature of KS, with impaired performance in tasks requiring planning (Joyce and Robbins, 1991), rule detection and following (Kopelman, 1991), verbal fluency (Fujiwara et al., 2008), and cognitive estimation (Taylor and O'Carroll, 1995). Compared with ALs, KS demonstrate considerably poorer performance on tests of EF (Maharasingam et al., 2013), leading some to conclude that such impairments represent unique sequelae of organic KS and are not the result of alcohol neurotoxicity (Krabbendam et al., 2000; Brokate et al., 2003). Nonetheless, the literature is replete with evidence of impaired performance on various measures of EF in ALs compared with controls (Moriyama et al., 2002; Oscar-Berman et al., 2004; Pitel et al., 2007), which does not appear to be accounted for by differences in age or education. Thus, it seems the severe executive deficits of KS are superimposed on top of existing, albeit milder, impairments resulting from chronic alcoholism and related factors; although this could include sub-clinical episodes of thiamine deficiency (Pitel et al., 2011). Findings relating to samples with diagnoses other than KS are scant, though one study found 65% of a small sample (n = 16) with ARBD performed within impaired ranges on a test of EF (Horton et al., 2015b), but it is not clear from the sample description whether this included individuals with WKS diagnoses.

Despite substantially more impaired executive abilities observed in those with KS compared with ALs, less sensitive measures of EF such as the MoCA's EF subtest may result in little difference between these groups (Wester et al., 2013c), emphasizing the need for more sensitive measures. The ecologically-based battery assessment of EF, the BADS, appears sensitive to KS-related deficits across multiple subtests and can differentiate these from AL impairments (Maharasingam et al., 2013). KS performance on the BADS is not consistent across subtests, highlighting the need to examine subtest scores in addition to the overall profile score (Van Den Berg et al., 2009). In terms of ecological validity, high scores on the Profile score and Temporal Judgment and Zoo Map sub-tests have been associated with subsequent occupational success in ALs, though no link between BADS scores and drinking outcomes was observed (Moriyama et al., 2002).

Individuals with KS perform particularly poorly on tests of rule derivation such as the WCST, MCST, and BPSAT compared with controls and those with psychiatric disorders and other neurological conditions (Shoqeirat et al., 1990; Van Den Berg et al., 2009). Multiple studies have also found evidence of impaired cognitive estimation in KS (Kopelman, 1991; Taylor and O'Carroll, 1995), although an earlier study found preserved performance (Leng and Parkin, 1988). The commonly used test of cognitive estimation, the CET, has previously been criticized for being culturally bound, but this issue has recently been resolved with the production of a revised CET (MacPherson et al., 2014). Accurate cognitive estimations, however, may rely on the retrieval of knowledge from semantic memory and the subsequent error checking of this process as well as executive abilities (Taylor and O'Carroll, 1995). This view has since been corroborated in KS participants using an alternative test of cognitive estimation (Brand et al., 2003) and supported in a review of findings from various clinical groups (Wagner et al., 2011).

There appears to be little inter-correlation between tests of single EFs, including the CET, FAS, WCST, and MCST (Shoqeirat et al., 1990; Kopelman, 1991). Dissattenuated correlations (i.e., correlations corrected for the unreliability of the test scores) between the EF test scores were, however, not considered here and may be stronger than those originally reported (Schmidt and Hunter, 1996). Nonetheless, the lack of correlations is perhaps unsurprising as, although estimations of quantities, verbal fluency, and rule detection may depend on some common skills (e.g., attention), they likely recruit different skills for their completion. This view is consistent with studies using latent variable analyses to study the diversity of executive abilities which have concluded that EFs are separable but related functions that share some level of underlying commonality (e.g., Miyake et al., 2000; Friedman et al., 2008). The resulting models of EF from these investigations distinguish between key functions such as shifting, inhibition and updating, with some also including working memory and planning (Snyder et al., 2015). Such models could provide a useful theoretical basis for testing EFs in those with ARCI; although the EF tests identified here mostly assess shifting (e.g., WCST, BSPAT) and planning (e.g., BADS Zoo Map Test), with tests of fluency and cognitive estimations not neatly fitting within existing models of EF (see Snyder et al., 2015). Inhibition can be tested using the BADS or CANTAB response inhibition tests reviewed here, while updating is typically tested by paradigms such as the Spatial 2-back or Keep Track tests used by Friedman et al. (2008), indicating a need to incorporate standardized updating tasks into existing batteries.

Impairments on intelligence tests have also been observed in those with ARCI, particularly on subtests that involve processing speed, working and episodic memory, and executive abilities. Only two test batteries were evaluated by included studies, the WAIS and CANTAB. The WAIS is often viewed as the gold-standard assessment of intelligence and general ability (Hayes et al., 2016). Those with ARCI, particularly KS, demonstrate better performance on the Verbal IQ (VIQ) components of the WAIS compared with the Performance IQ (PIQ) components; though the reverse has also been documented (Oscar-Berman et al., 1993), highlighting heterogeneity within this population. A depressed WMS relative to WAIS IQ or VIQ may be indicative of KS, though we do not recommend using this discrepancy to diagnose the condition. The process has been repeatedly criticized based on several inherent flaws (Dennis et al., 2009; Kent, 2013). First, inferring memory dysfunction from the discrepancy is fallible, as an accurate measure of premorbid function from which any deterioration in IQ can also be established is rarely available. Second, those at extreme ends of the IQ spectrum may produce misleading discrepancy scores. Third, WMS and WAIS performance are not exclusive as there is overlap between the skills used in both, particularly working memory processes (Shipstead et al., 2016). Further, the historical view of KS as a circumscribed memory disorder is clearly challenged by the extant evidence, with wide-ranging cognitive deficits observed in the studies reviewed here. Indeed, Bowden (1990) has argued that the strict selection criteria used for KS by some researchers may have resulted in samples which were unrepresentative of the more varied presentation typical of this population. Accordingly, Scalzo et al. (2015) report a case-series investigation of those with WKS whose IQ minus MQ scores ranged widely from−14 to 36, demonstrating the poor diagnostic validity of the discrepancy approach.

Tests from the computerized CANTAB battery appear sensitive to the deficits associated with ARCI, namely episodic memory and EF (Horton et al., 2015b). The battery is psychometrically and practically appealing as its on-screen nature allows little room for administration or interpretation errors, parallel versions of tests are available, and normed scores are calculated using a large and diverse normative database. The tests are also almost entirely visual in style and are therefore unaffected by low verbal capabilities or cultural background.

Overall, intelligence and battery assessments appear useful for evaluating multiple cognitive abilities while using the same normative sample for comparison. In addition, composite scores derived from battery tests are typically more reliable and better predictors of functional outcomes than single test scores (Harvey, 2012). However, large batteries such as the WAIS can be time-consuming and demanding for some, potentially explaining why many studies have used only a selection of relevant subtests as opposed to the entire battery. Additionally, most batteries will require the inclusion of auxiliary tests to cover all functions known to be impaired in those with ARCI, such as orientation or aspects of memory and learning not assessed.

The assessment of premorbid intellectual function in those with ARCI is essential for correctly interpreting the extent of the suspected cognitive deficit. The NART may provide some value in predicting premorbid function in those with ARCI and the revised version appears more effective than the original. Still, it remains unknown whether NART performance is entirely impervious to the deficits associated with KS. Nonetheless, the measure appears to be the common choice with this population and the NART-R has recently been re-standardized in a non-clinical British sample for the new WAIS-IV (Bright et al., 2016), increasing its application.

In addition to test selection, several additional choices face clinicians using neuropsychological measures to assess ARCI. One of the foremost of these is the decision as to when tests should be administered. A systematic review by Walvoort et al. (2013) found it typically requires 6-weeks of abstinence before reliable neuropsychological performance can be achieved. However, the use of cognitive screening tests during or immediately following detoxification is common practice and may provide some indication of a person's cognitive ability (Royal College of Psychiatrists, 2014); although results should be interpreted in the light of Walvoort and colleagues' findings. The use of repeat testing may also be important to track the changes in cognition that have been observed during the first 2 years post-detoxification (Stavro et al., 2013). The Royal College of Psychiatrists (2014) in the UK have suggested that following initial assessments, more detailed cognitive assessments should be undertaken after 3-months, then repeated at 6-month intervals for 3 years. This emphasizes the need to use tests that have high test-retest reliability values and/or parallel versions to avoid practice effects.

Several person-specific factors should also be considered when interpreting the outcomes of cognitive testing in those suspected of ARCI to ensure accurate interpretation. First among these is the consideration of any comorbid psychopathology, as individuals with conditions such as depression (Bosaipo et al., 2017) and schizophrenia (Bora and Pantelis, 2015) have been found to display decreased performance on various cognitive tests compared with matched controls. Relatedly, psychiatric medications can have varying effects on cognitive performance, both during and immediately following use (e.g., Bouchard et al., 2015; Helmes and Ostbye, 2015). Additional non-medical substance use including cocaine, heroin, cannabis, and even (acutely) nicotine can also have deleterious effects on cognitive testing lasting for varying periods following abstinence, particularly in those with co-morbid psychiatric disorders (see Gould, 2010); though one study has found no additional impact of cannabis use on MoCA scores in ALs (Alarcon et al., 2015), suggesting the effects may vary. Finally, the presence of cerebrovascular disease and traumatic brain injuries should also be considered due to their high prevalence (25%) within this population (Wilson et al., 2012). Together, these findings highlight the need to combine neuropsychological testing with a clinical diagnostic process and other assessment measures—primarily neuroimaging and nutritional assessment (Scalzo et al., 2015)—to ensure accurate diagnosis.

As mentioned, few studies included in the review investigated non-KS forms of ARCI, suggesting further investigations of those with other diagnostic labels (e.g., ARD, ARBD) may be warranted. However, the concept of a prolonged alcohol-related neurocognitive disorder with a distinct neuropathological basis from WKS or other disorders of vitamin depletion (e.g., Pellagra) remains tenuous. Given the similarities in cognitive profiles between groups with diagnoses of ARD, ARBD and KS seen here, the present review findings appear consistent with suggestions that common neuropathological processes, namely thiamine depletion, underpin most forms of ARCI (Joyce, 1994; Zahr et al., 2011). This theory is supported by research with an AL population demonstrating a significant correlation between thiamine levels and memory performance, as well as a graded decrease in cognitive ability with each symptom of WE (Caine et al., 1997 criteria) present (Pitel et al., 2011).

Similar cognitive profiles are also evident across gradations of alcohol-related impairment, suggesting less impaired groups sit earlier in the same underlying pathological process as those with KS. Indeed, AL & ALC groups appear to be impaired, to a lesser extent, on many of the same tests of memory (e.g., Pitel et al., 2008; Wester et al., 2013b) and general cognitive ability (e.g., Wester et al., 2013c) as those with KS. However, those with KS typically display impaired performance on tests of recognition memory and orientation, yet ALC and ARD populations perform at control-level (Munro et al., 2001; Wester et al., 2013b), indicating that some neuropsychological distinctions (other than those of degree) are evident when thiamine depletion progresses to the point of a KS diagnosis. Nonetheless, normal recognition memory and an absence of disorientation have been observed in a case of KS meeting standard diagnostic criteria for the disorder (Noël et al., 2001), further highlighting the heterogeneity within this population.

The present findings suggest it may be valuable from a clinical perspective to adopt a broad diagnostic conceptualization of ARCI, assuming a degree of individual variability in the particular functions impaired and where the extent of impairment is the deciding factor in diagnostic decision making; as opposed to trying to decide into which of the highly overlapping diagnostic categories (e.g., ARD, WKS) an individual best fits. Those adopting this view could acknowledge the central role of thiamine deficiency in producing ARCI (Arts et al., 2017), and still specify when a person's symptomology closely resembles KS, whilst also recognizing that various other etiological factors (e.g., other nutritional deficiencies, impaired liver function etc.) can, like thiamine depletion, contribute to varying degrees and may explain heterogeneity in degree and types of impairment observed (see Bowden, 1990). The more transient, but still significant (e.g., Stavro et al., 2013), effects of alcohol neurotoxicity (Arts et al., 2017) should also be recognized within this broad approach, highlighting to clinicians the potential for recovery of certain functions over time. This perspective closely resembles the DSM-5's alcohol-related neurocognitive disorder, which bifurcates twice into mild and severe forms and from the latter into amnestic and non-amnestic types. The ARBD conceptualization postulated by Wilson et al. (2012)—which encompasses WKS and all clinically significant forms of ARCI not meeting the criteria for WKS—may also provide a highly pragmatic diagnostic conceptualization for clinical use consistent with the above approach. That said, if as the literature appears to suggest, that most cases of ARCI are variants of WKS caused by thiamine deficiency to a greater or lesser extent, then the term WKS may be more apt from an etiological perspective and has recently been advocated for use in place of terms like ARD or ARBD (Arts et al., 2017). Though, as mentioned, traditional views of WKS may fail to fully capture the syndrome as experienced by alcohol users in regards to its heterogeneity and the contributing role of etiological factors other than thiamine depletion, particularly alcohol neurotoxicity. Nonetheless, the value of neuropsychological assessment lies in characterizing the type and degree of cognitive dysfunction suffered by alcohol-users and monitoring it over time, regardless of diagnostic categories and nomenclature.

Overall, we recommend further appraisal of all tests discussed here in those with ARCI, but a selection of more specific recommendations for future studies are worthy of mention. First, the generalizability of findings and methodological rigor of future research in this area can be improved by addressing some of the limitations of the existing literature identified here, including clear reporting of abstinence duration, diagnostic criteria (particularly if assessing samples with “ARBD”), and the methods used for diagnosis (including whether the tests evaluated contributed to diagnostic decisions). Second, further evaluations of the test re-test reliability of tests used with this population are necessary to ensure any variability in scores on repeat testing represent true differences, not practice effects. Third, to support clinicians in identifying those with or at risk of ARCI, it would be beneficial for studies to continue to identify test cut-off scores which delineate gradations of impairment (e.g., from AL to ALC and more severe forms of ARCI), as has recently been done for the MoCA and MMSE (Wester et al., 2013c; Oudman et al., 2014). Fourth, it would also be useful to see how well performance on tests reflects the ability to carry out activities of daily living and can predict functional outcomes such as treatment adherence. While excluding treatment studies from the present review decreased the chance of identifying these outcomes, a brief review of the ARCI treatment and intervention literature did not find any such investigations.

Fifth, many of the most commonly employed neuro-psychological tools identified in Table 1 were not evaluated by any studies in the review and require investigation to support their continued use with this population. Two additional tests requiring validation in this population are the Addenbrooke's Cognitive Examination-III (ACE; Hsieh et al., 2013) and Repeatable Battery for the Assessment of Neuropsychological Status (R-BANS). In an on-going investigation of ARCI prevalence by the present authors, these are the most commonly reported neuropsychological tools used by UK clinicians in the ARCI diagnostic process. The ACE has demonstrated screening abilities commensurate with the MoCA and superior to the MMSE for AL impairments (Ridley et al., 2017), but neither the ACE nor R-BANS have been evaluated for use in ARCI diagnosis.

Finally, two CSIs have recently been developed specifically for assessing alcohol-related cognitive impairments, the Brief Examination of Alcohol-Related Neuropsychological Impairments (BEARNI; Ritz et al., 2015) and the Test of Detection of Cognitive Impairment in Alcoholism (TEDCA; Jurado-Barba et al., 2017), though both have only been validated in AL populations not meeting ARCI diagnostic criteria (both CSIs can be accessed in the corresponding publications [the BEARNI is in French]). The BEARNI was designed as an easily administered CSI for assessing the deficits associated with alcoholism: working and episodic memory, visuospatial abilities, EF, and ataxia. In its validation, Ritz et al. (2015) found the BEARNI was considerably more sensitive when differentiating between ALs and controls (sensitivity = 98.4%) compared with the MMSE (9.5%) and Dementia Rating Scale (7.9%; Coblentz et al., 1973). However, the tool displayed relatively poor specificity (50%). The TEDCA appears highly similar to the BEARNI in its assessment of working and episodic memory and visuospatial skills. The TEDCA's assessment of EF, however, appears more comprehensive, including the Trail Making Test-B, conceptual similarity tests (e.g., “What do food & gasoline have in common?”), and a go-nogo task; compared with only a category fluency test in the BEARNI. The TEDCA demonstrated moderate sensitivity (67%) and good specificity (76.7%) when differentiating between ALs and controls (Jurado-Barba et al., 2017), although these findings are obfuscated by the failure to state whether ALs were abstinent during testing. Overall, both tests are valuable additions to the assessment of this population and possess several strengths: short administration time, an ability to be administered by non-specialists, and a focus on the key areas of alcohol-related cognitive impairment. A priority for future research in this area should be to evaluate the screening capabilities of both tests for those with ARCI diagnoses such as KS.

The present review may be limited by the exclusion of studies evaluating non-standardized tests without normative data, as there may be tests not meeting these criteria which are particularly useful in the assessment of ARCI. However, expanding the review to include such tests would be beyond the scope of a single review and the evaluation of standardized normatively defined tests is likely of most value to those involved in the assessment of ARCI. Second, this review considered only one test of affect or social cognition (WMS-III Face Recognition Test; Holdnack and Delis, 2004) indicating the lack of studies evaluating standardized tests of these domains. If considering ARCI from a dual-processing perspective (e.g., Helfrich et al., 2018), as has recently been advocated (Brion et al., 2015), then the tests reviewed here focus mostly on the “reflective system” system (which draws upon memory and EF to initiate considered, deliberate responses), with few assessing the affective component of the “affective-automatic” system responsible for the impulsive processing of stimuli, including decoding the affective quality of a stimulus. This is notable as tests of affect and social cognition may prove informative for those involved in the treatment of this population (Brion et al., 2015). For example, ALs and those with KS display impaired processing of emotional facial expressions and, in the former group, this has been associated with interpersonal dysfunction (Kornreich et al., 2001; Montagne et al., 2006; Donadon and Osório, 2014). Thus, tests of affect should also be considered when assessing ARCI; to date, however, studies in this domain are scant (Arts et al., 2017).

Third, we did not include studies evaluating neuropsychological tools in AL participants, which could have provided additional information on the utility of tests for the assessment of alcohol-related deficits. However, we have introduced some of the more pertinent and up-to-date evidence relating to ALs (e.g., Alarcon et al., 2015; Ritz et al., 2015) into our discussion to supplement the review findings. Fourth, several studies in the review evaluated older versions of tests used today (e.g., WMS/WAIS editions), thus caution should be taken in applying these findings to newer versions. Although older versions are likely conceptually and structurally similar, they should not be viewed as the same test (Strauss et al., 2006). Finally, the findings from older studies in the review should be interpreted carefully as they were considerably more likely than newer studies to contain methodological weakness (e.g., lack of diagnostic criteria and short abstinence duration).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.