Article Text

Download PDFPDF

Review
The cognitive profile of ALS: a systematic review and meta-analysis update
  1. Emma Beeldman1,
  2. Joost Raaphorst1,2,
  3. Michelle Klein Twennaar1,
  4. Marianne de Visser1,
  5. Ben A Schmand1,3,
  6. Rob J de Haan4
  1. 1Department of Neurology, Academic Medical Centre, University of Amsterdam, Amsterdam, The Netherlands
  2. 2Department of Neurology, Radboud University Medical Centre, Nijmegen, The Netherlands
  3. 3Department of Psychology, University of Amsterdam, Amsterdam, The Netherlands
  4. 4Clinical Research Unit, Academic Medical Centre, University of Amsterdam, Amsterdam, The Netherlands
  1. Correspondence to Emma Beeldman, Department of Neurology, Academic Medical Centre, P.O. Box 22700 DE, Amsterdam 1105 AZ, The Netherlands; e.beeldman{at}amc.uva.nl

Abstract

Cognitive impairment is present in approximately 30% of patients with amyotrophic lateral sclerosis (ALS) and, especially when severe, has a negative impact on survival and caregiver burden. Our 2010 meta-analysis of the cognitive profile of ALS showed impairment of fluency, executive function, language and memory. However, the limited number of studies resulted in large confidence intervals. To obtain a more valid assessment, we updated the meta-analysis and included methodological improvements (controlled data extraction, risk of bias analysis and effect size calculation of individual neuropsychological tests). Embase, Medline and PsycInfo were searched for neuropsychological studies of non-demented patients with ALS and age-matched and education-matched healthy controls. Neuropsychological tests were categorised in 13 cognitive domains and effect sizes (Hedges’ g) were calculated for each domain and for individual tests administered in ≥5 studies. Subgroup analyses were performed to assess the influence of clinical and demographic variables. Forty-four studies were included comprising 1287 patients and 1130 healthy controls. All cognitive domains, except visuoperceptive functions, showed significant effect sizes compared to controls. Cognitive domains without bias due to motor impairment showed medium effect sizes (95% CI): fluency (0.56 (0.43 to 0.70)), language (0.56 (0.40 to 0.72)), social cognition (0.55 (0.34 to 0.76)), or small effect sizes: delayed verbal memory 0.47 (0.27 to 0.68)) and executive functions (0.41 (0.27 to 0.55)). Individual neuropsychological tests showed diverging effect sizes, which could be explained by bias due to motor impairment. Subgroup analyses showed no influence of bulbar disease onset and depression and anxiety on the cognitive outcomes. The cognitive profile of ALS consists of deficits in fluency, language, social cognition, executive functions and verbal memory. Social cognition is a new cognitive domain with a relatively large effect size, highlighting the overlap between ALS and frontotemporal dementia. The diverging effect sizes for individual neuropsychological tests show the importance of correction for motor impairment in patients with ALS. These findings have implications for bedside testing, the design of cognitive screening measures and full neuropsychological examinations.

  • ALS
  • COGNITION
  • META-ANALYSIS
View Full Text

Statistics from Altmetric.com

Introduction

Cognitive impairment occurs in approximately 30% of patients with amyotrophic lateral sclerosis (ALS) and a small proportion of patients (±10%) have frontotemporal dementia, mostly the behavioural variant.1 ,2 Cognitive impairment, especially when severe, bears significant implications for patients with ALS and their families: it increases caregiver burden and reduces survival which may be related to restricted use of non-invasive ventilation.3 ,4 More insight into the cognitive profile of ALS could lead to better bedside testing and more adequate information to the patient and their caregiver.

Across studies, deficits have been shown in different cognitive domains. Most consistently, executive dysfunction and fluency deficits are found.5 There is increasing evidence of language impairment and memory dysfunction in patients with ALS, and recently, impairment of social cognition has been shown.6 ,7 In 2008, we performed a meta-analysis of the cognitive profile of patients with ALS without dementia (n=16 studies, 554 patients, published in 2010) showing impairment of fluency, executive functions, language and memory.8 For other cognitive domains, we were unable to draw firm conclusions due to limited data. Since 2008, many neuropsychological ALS studies have been published. With an update of the meta-analysis, we aimed to generate a more precise assessment of the cognitive profile of ALS.

Methods

Search strategy

We searched Embase (1970–2014), Medline (1966–2014) and PsycInfo (1970–2014) up to 28 November 2014 for articles written in English, German, French or Dutch. References of articles were also considered for inclusion. Key words included ALS and its synonyms, cognition and frontotemporal dementia (see online supplementary table S1). Two authors (EB, MKT) performed title/abstract screening and subsequently full text evaluation. Consensus meetings led to the inclusion or exclusion of articles.

Study selection

Eligible articles were controlled studies. In case of longitudinal studies, only data from the first visit were used for analysis. All studies had to meet the following criteria:

  • Patients had to be diagnosed with ALS according to the validated El Escorial criteria,9 that is, presence of a combination of upper and lower motor neuron signs. When patients with progressive muscular atrophy or primary lateral sclerosis were also studied, data of patients with ALS had to be reported separately.

  • Patients and controls had to be matched for age and education, or age-corrected and education-corrected standard scores had to be used.

  • Patients and controls also had to be free of frontotemporal dementia and Alzheimer's disease, according to standard clinical criteria10 ,11 or according to the authors’ statement. If demented participants were included in the study, the results of non-demented participants had to be reported separately.

  • Data of at least one validated neuropsychological test had to be reported for both patients and controls as the mean and SD of raw or standardised test scores.

  • Studies had to report unique cohorts. If they reported data of the same cohort, the study with the largest sample was included.

  • Studies had to correct for the presence of dysarthria or motor disabilities, or both, for example, use of adjusted neuropsychological tests or excluding patients with severe motor disabilities.

There were no exclusion criteria.

Data extraction

We extracted demographic and clinical variables from the articles: age (years), educational level (years of formal education), disease duration (months), site of onset (% bulbar onset), disease severity (ALS functional rating scale revised, (ALSFRS-R)),12 respiratory function (forced vital capacity (FVC, percentage predicted (%pred.) or pCO2, capillary/arterial blood gas), use of psychoactive medication and depression/anxiety. In order to facilitate the interpretation of data, we categorised neuropsychological tests in 13 cognitive domains, as described before, that is, language, executive functions, fluency, immediate verbal memory, delayed verbal memory, visual memory, visuoperceptive functions, visuoconstructive functions, verbal IQ, psychomotor speed and attention. Owing to the widespread use of the mini mental state examination (MMSE) in ALS studies, we included this test as a global measure of cognitive impairment. Social cognition was a new cognitive domain and was not included in the 2010 meta-analysis. Two of the authors (EB, MKT) performed the data extraction in order to reduce extraction errors.14

Heterogeneity

Subgroup analyses were performed to find explanations for moderate or substantial heterogeneity in the cognitive domains. The following demographic and clinical variables were assessed in the subgroup analyses: age (years), educational level (years of formal education), disease duration (months), site of onset (% bulbar onset), disease severity (ALSFRS-R), respiratory function, psychoactive medication use and depression/anxiety. For each variable, the studies were divided into two groups, based on a median split, except for respiratory function, use of psychoactive medication and depression/anxiety. Respiratory dysfunction and psychoactive medication use were not part of the exclusion criteria in all studies. For respiratory dysfunction, studies were divided into a group of “studies which excluded patients with severe respiratory dysfunction” and a second group of “studies which did not exclude patients with severe respiratory dysfunction or did not present data on respiratory function.” For psychoactive medication use, studies were divided into a group of “studies which excluded patients with psychoactive medication use” and a second group of “studies which did not exclude patients with psychoactive medication use or did not present data on psychoactive medication use.” For depression/anxiety, the studies were divided into a group with “depression/anxiety level comparable to healthy controls” and a group with “depression/anxiety level higher than healthy controls.”

Statistical analysis

We calculated effect sizes, expressed as Hedges’ g, per cognitive domain for each study, using Review Manager.13 Hedges’ g represents the mean difference between patients with ALS and healthy controls, divided by the pooled SD. When studies used multiple neuropsychological tests in one cognitive domain, an averaged effect size was computed in order to ascertain that each study only added one effect size to the final analysis.

We chose a random effects model to obtain an average weighted effect size across the studies. Effect sizes are considered small, medium and large when they are 0.2, 0.5 and 0.8, respectively.15 A positive effect size, that is, an effect size of more than 0, indicates impaired performance of patients with ALS compared with healthy controls. An effect size was considered ‘significant’ when the CI did not contain 0. Effect sizes of individual cognitive tests were also calculated when administered in ≥5 studies.

Statistical heterogeneity among studies was assessed with Cochran's Q test (significance set at p<0.10) and the I2 statistic. I2 describes the percentage of total variation across studies that is due to heterogeneity rather than chance. Cut-off points for low, moderate and substantial heterogeneity are 25%, 50% and 75%, respectively.16 Outliers within the domains with substantial heterogeneity were identified and effect sizes were recalculated after exclusion of the outlier.

Risk of bias

One of the authors (EB) assessed the quality of the included articles with the Newcastle—Ottawa quality assessment scale (NOS, maximum score 9 points).17 The NOS evaluated the following aspects: selection of participants, comparability of the participants and data collection. Owing to the design of the included studies, that is, not blinded to the status of the participant, studies could not be awarded a point on the item ‘ascertainment of exposure’, and the maximum score was therefore eight points. The risk of selection bias was evaluated with a subgroup analysis comparing studies with consecutively recruited patients to studies with a potential selection bias.

The risk of bias due to motor impairment and/or speech difficulties was assessed. Effect sizes were recalculated after exclusion of cognitive tests that require fine motor skills or normal speech; these recalculated (non-motor dependent) effect sizes were compared with the original effect sizes.

The presence of publication bias can result in a (more) significant mean effect size, because non-significant studies are less likely to be published. To investigate the probability of a relevant publication bias, we calculated the fail-safe N for all cognitive domains, that is, the number of studies without an effect which should be added to the meta-analysis to obtain a non-significant effect size, that is, p=0.05.18 A more conservative estimate of existing unpublished or unretrieved studies is the tolerance level ((5×number of studies examining the domain)+10). If the fail-safe N is large in comparison to the tolerance level, the observed result is considered to be a reliable estimate of the real effect.

We used IBM Statistics SPSS, V.20 and Review Manager, V.5.2 for statistical analysis.

Results

The literature search yielded 7380 hits and 44 studies were included in this meta-analysis. Thirty studies were new compared to our previous meta-analysis (table 1 and figure 1). For an overview of the excluded articles and reasons for exclusion, see online supplementary table S2.

Table 1

Study characteristics of included articles

Figure 1

Flow chart of literature search and study selection. Embase, Medline and PsycInfo were searched simultaneously with the same key words. *The flow chart shows the results after deduplication.

Study characteristics

A total of 1287 patients (62.8% males) and 1130 healthy controls (55.5% males) were included in the meta-analysis (table 1). In three studies, only patients with definite ALS were analysed, and in 26 studies patients had either probable or definite ALS. The mean age (SD) of the patients was 59.2 years (3.9) and the disease duration ranged from 8.1 to 80.0 months (median 23.5). The site of onset was reported in 28 studies (63.6%, bulbar onset in 33% of patients), of which five included a higher proportion of bulbar onset than patients with limb onset. Disease severity was measured with different scales across studies, that is, the Amyotrophic Lateral Sclerosis Functional Rating Scale—Revised (ALSFRS-R, maximum score 48=no disability), the ALS severity scale (ALSSS, maximum score 40=no disability), the Appel score (scores between 30 and 164, lower scores=less disability) and the Norris score (maximum score 100=no disability).12 ,19–21 The ALSFRS-R was used most frequently (n=23 studies, 52.3%), with a median score (range) of 33.2 (11.9–40.4). The median scores (range) of the ALSSS, Appel score and Norris score were 31.8 (28.4–32.2), 68.8 (30–123) and 75.5 (50–92), respectively. In most studies, the disability scores indicated a mild to moderate disease severity. One study examined patients with an advanced disease state, with a mean (SD) ALSFRS-R of 11.9 (8.2).

Studies corrected for the presence of dysarthria and/or motor disabilities by using an adjusted neuropsychological test battery (time-dependent tests were excluded or tests were corrected for slowing of speech), by excluding patients with severe motor impairment or dysarthria, or some patients did not perform all tests due to motor impairment or dysarthria (see online supplementary table S3).

Twenty-one studies (47.7%) reported data on the presence of respiratory failure. Eight studies excluded patients with an FVC (%pred) below 70%, while two studies performed blood gas analysis to investigate the presence of hypercapnia. In 11 studies, patients had no severe respiratory dysfunction, as stated by the authors. Depression and anxiety were measured in 26 studies (59.1%), most frequently with the Hospital Anxiety and Depression Scale (HADS, 53.8%). Other depression and anxiety scales were the Beck Depression Inventory (BDI, n=7, max score 63),22 the Hamilton Rating Scale for Depression (n=2, max score 52),23 the ALS depression inventory (n=1, max score 48), the Geriatric Depression scale (n=1, max score 30)24 and the Von Zerssen depression scale (n=1, max score 48).25 In the majority of studies (72.7%), the scores did not differ between patients with ALS and healthy volunteers. Six studies showed significantly more depressive and/or anxiety symptoms in patients with ALS compared with controls.

Twenty-one studies excluded patients if they used psychoactive medication, one study did not exclude patients, and the remaining 22 studies did not report medication use. References related to the aforementioned study characteristics can be found in online supplementary table S4.

Effect sizes of cognitive domains

All cognitive domains, except visuoperceptive functions, showed significant effect sizes, that is, impairment in patients with ALS, compared with controls (figure 2). In the domains with little or no bias due to motor impairment, medium effect sizes were found, that is, between 0.5 and 0.8 for the domains fluency, language and social cognition (table 2). Small effect sizes, that is, below 0.5, were found for the domains delayed verbal memory, immediate verbal memory, attention, executive functions, verbal IQ and visual memory. The domains in which bias due to motor impairment could not be ruled out, that is, visuoconstructive functions, psychomotor speed and MMSE, showed large or medium effect sizes. These domains were exclusively examined with time-dependent and motor-dependent tests and an adjusted effect size without motor bias could not be calculated.

Table 2

Pooled weighted effect sizes and heterogeneity statistics of all cognitive domains and frequently used neuropsychological tests

Figure 2

Effect size per cognitive domain. Effect size is expressed as Hedges’ g, larger values indicate more impairment in patients with ALS compared to healthy volunteers. Black line: meta-analysis update; grey line: meta-analysis 2010. For a description of the neuropsychological tests included in each domain, see online supplementary table S5. ALS, amyotrophic lateral sclerosis.

The main differences with the results of our 2010 meta-analysis are the inclusion of social cognition as a new cognitive domain, the significant effect size of delayed verbal memory and the narrow CIs (figure 2), due to the inclusion of more studies.

To reduce bias due to motor impairment, the effect sizes of the domains fluency, language, executive functions and visual memory were recalculated after exclusion of the time-dependent and motor-dependent tests, which did not considerably change the effect sizes of fluency, language and visual memory. The effect size of executive functions decreased after exclusion of the time-dependent and motor-dependent tests from 0.49 to 0.41 (table 2).

Effect sizes of neuropsychological tests

The neuropsychological tests used in each cognitive domain are listed in online supplementary table S5. The largest effect sizes were found for the trail making test part B, colour word interference of the Stroop test, symbol digit modalities test, letter fluency, word-reading and colour-naming of the Stroop test and naming tests (table 2).

Heterogeneity

Heterogeneity was low in the domains MMSE, fluency, immediate verbal memory, attention, verbal IQ and visual memory (range I2 values: 0–20%). Moderate heterogeneity (I2: 43–50%) was found in the domains language, social cognition and visuoperceptive functions. The domains visuoconstructive functions, psychomotor speed, executive functions and delayed verbal memory showed substantial heterogeneity (I2: 56–74%). In the domains visuoconstructive functions, psychomotor speed and delayed verbal memory, a main outlier could be identified, and effect sizes were recalculated after exclusion of the outlier (table 2).

Subgroup analyses

There was no influence of site of onset on any of the cognitive domains (table 3). Disease severity or the presence of depressive and anxiety symptoms did not significantly contribute to the effect sizes on any of the cognitive domains (data not shown, but available on request). Studies with younger patients had more impairment of psychomotor speed. Studies with a lower educational level had more impairment of executive functions. A shorter disease duration resulted in more impairment of visual memory and immediate verbal memory. Studies that excluded patients with respiratory dysfunction showed less impairment of psychomotor speed, delayed verbal memory and MMSE. Studies that excluded patients with psychoactive medication showed less impairment of attention and visual memory.

Table 3

Subgroup analysis of the percentage of bulbar disease onset in the included studies

Risk of bias

The representativeness of the patient cohort was infrequently stated (n=18). The subgroup analysis for selection bias showed no differences in effect sizes for any of the cognitive domains. The recruitment method of healthy volunteers was described in 26 studies. The non-response rate was also infrequently recorded (n=5). The fail-safe N exceeded the tolerance level at least five times in all cognitive domains, indicating that the mean effect sizes are a reliable estimate of the true value when publication bias is considered.

Discussion

We updated our meta-analysis of cognitive impairment of non-demented patients with ALS compared with healthy volunteers. The cognitive profile of ALS consists of deficits in fluency, language, social cognition, executive functions and verbal memory.

Social cognition and the overlap with FTD

Social cognition is a new cognitive domain compared to our previous meta-analysis and its effect size is comparable to the domains of language and fluency. It includes recognition of the emotional states of others and insight into social situations and social protocol.26 ,27 This finding corroborates the overlap between ALS and FTD, as studies in FTD have shown extensive impairment of social cognition which is correlated to cortical atrophy of the right orbitofrontal, superior temporal, occipital and posterior cingulate regions.26–28 However, in order to investigate the existence of a cognitive continuum, neuropsychological studies are needed that directly compare patients with ALS, ALS-FTD and FTD. To the best of our knowledge, there are only three such studies. Two studies only administered Addenbrooke's cognitive examination and therefore do not allow for a comparison of the cognitive profiles.29 ,30 Another study administered tests of social cognition, which showed similar scores of patients with ALS and healthy volunteers and comparable deficits in the ALS-FTD and FTD groups.7 In the current meta-analysis, we excluded studies which examined patients with ALS-FTD, in order to assess the cognitive profile of patients with ALS without dementia and reduce the heterogeneity in the data. Subtle effects of mild cognitive impairment in a large ALS cohort might be overshadowed by large effects in a small ALS-FTD subgroup in the cohort. Also, the inclusion of patients with severe behavioural changes may lead to seemingly impaired performance on cognitive tests, due to a lack of attention and interest. Despite the challenges of examining patients with ALS-FTD, neuropsychological studies are needed to elucidate the similarities and differences between ALS and FTD.

Associations between behavioural changes and cognitive impairment on the one hand and reduced survival and caregiver burden on the other have been found previously.4 ,31 Considering the prominent role of social cognition in this meta-analysis, social cognition deficits may also be associated with shorter survival and increased caregiver burden. This warrants a more thorough investigation in future studies.

Memory

Our previous meta-analysis showed significant effect sizes for immediate verbal memory and visual memory, and a large CI for delayed verbal memory which did not reach significance. The current update shows significant effect sizes for all memory domains, with the largest effect size for delayed verbal memory. Although both the medial temporal lobe (MTL) and the prefrontal cortex are involved in delayed and immediate memory, delayed memory relies primarily on intact functioning of the MTL, whereas immediate memory is thought to be a function of the mid-dorsolateral frontal cortex.32 ,33 The finding of memory changes in our meta-analysis ,therefore, could reflect changes in the MTL, which have been described as TDP-43 positive neuronal inclusions in the dentate gyrus, hippocampus and transentorhinal cortex, found in postmortem studies of patients with ALS, in particular when concomitant dementia was present.34 Recent MRI studies underlined the involvement of the MTL in non-demented patients with ALS by correlating hippocampal volume with impairment of verbal memory.35 ,36

Executive functions

Although executive dysfunction is often found in patients with ALS, the effect size for this domain was smaller than for the domains language and social cognition.6 This is probably related to our decision to examine fluency as a separate cognitive domain, for which we found a larger effect size, that is, 0.58 compared to 0.49 for executive functions. Furthermore, the neuropsychological tests of the executive domain showed diverging effect sizes, that is, the trail making test and the Stroop test showed large effect sizes, whereas the Brixton spatial anticipation test had a non-significant effect size. This could be explained by deficits in motor function and speech in patients with ALS. The tests with large effect sizes are time-dependent and require fine motor skills or normal speech and, when excluded from the analysis, executive functions had one of the smallest effect sizes. When examining patients with ALS, it is therefore key to administer tests that correct for physical impairment, in order to avoid overestimation of cognitive deficits.37

Respiratory function

Impairment of executive functions and memory (list and object learning) is known to be partially related to respiratory dysfunction in ALS and is thus partially reversible after initiation of night-time non-invasive ventilation.38 In the current meta-analysis, only 10 of 44 studies (22.7%) reported measurements of respiratory function which may have led to an overestimation of the effect sizes of the executive and memory domains. An exploratory subgroup analysis showed no effect of respiratory function on the majority of the cognitive domains, except for psychomotor speed, delayed verbal memory and MMSE. For a valid statement on the influence of respiratory dysfunction on cognitive function in ALS, more studies need to report detailed data about the presence or absence of respiratory dysfunction.

Bulbar disease onset

The relation between bulbar disease onset and cognitive impairment in ALS has been debated. The proportion of bulbar disease onset in incident ALS cohorts is approximately 33%, but ranges from 33% to 60% in cognitively impaired cohorts.1 ,39–41 With a subgroup analysis, we found no evidence of more severe cognitive deficits in patients with bulbar disease onset. The varying rates of bulbar disease onset in cognitively impaired cohorts may be related to the presence of dysarthria in combination with the use of time-dependent tests. A correction for slowing of speech is therefore essential.37

Limitations

In addition to its strengths, that is, a large number of patients, careful selection of articles, extraction of data by two individuals and a systematic analysis, this study has some limitations.

Bias due to motor impairment could not be ruled out in the domains visuoconstructive functions, MMSE and psychomotor speed, because these were exclusively examined with tests that require fine motor skills and/or normal speech. All studies corrected for severe motor impairment, but adjustments for mild motor impairment were often not performed. This might explain why the effect sizes of these domains were among the largest effect sizes, indicating that patients with ALS have more difficulty performing the motor part of these tests than healthy volunteers. However, the effect size of language, fluency and visual memory, which were partly examined with time-dependent or motor-dependent tests, did not change after exclusion of these tests. This implies that bias due to motor impairment was small in these three domains.

We included MMSE in our analyses as a global measure of cognitive impairment. However, there are some limitations in the use of MMSE. First of all, there is a large ceiling effect in the normal population, implying that healthy controls are likely to obtain (near) maximum scores.42 Second, three items of MMSE require normal hand function, and could lead to an underestimation of the abilities of patients with ALS. The combination of the ceiling effect and the motor-dependent items in MMSE could have resulted in an artificially large difference between patients with ALS and healthy controls.

There was substantial heterogeneity in the domains visuoconstructive functions, psychomotor speed, executive functions and delayed verbal memory, hindering the interpretation of the results. The most likely explanation for the heterogeneity is the presence of outliers within these domains. For the domains visuoconstructive functions, psychomotor speed and delayed verbal memory, a single outlier could be identified, and after exclusion, heterogeneity was reduced to 10%, 46% and 44%, respectively. For the domain executive functions, a single outlier could not be identified, but all individual studies, except one, showed impairment of patients with ALS, compared to healthy volunteers. Our results therefore indicate impairment of patients with ALS on executive function tasks, but the results of our study emphasise the importance of tests that correct for physical impairment and slowing of speech to avoid false positive results.

In conclusion, the cognitive profile of patients with ALS consists of deficits in fluency, language, social cognition, executive functions and verbal memory with sparing of visuoperception. These findings provide further support for the observation that the cognitive deficits of ALS are more than ‘just frontal’, which has implications for bedside testing, the design of cognitive screening measures and full neuropsychological examinations.

Acknowledgments

The authors thank René Spijker for his help with the systematic literature search.

References

View Abstract

Supplementary materials

Footnotes

  • Contributors EB was involved in the study concept, literature search, analysis and interpretation of the data, statistical analysis, drafting/revising the manuscript for content and study coordination. JR was involved in the study concept, analysis and interpretation of the data and drafting/revising the manuscript for content. MKT was involved in the literature search, analysis and interpretation of the data and drafting/revising the manuscript for content. MdV was involved in the analysis and interpretation of the data, drafting/revising the manuscript for content and study coordination. BAS was involved in the analysis and interpretation of the data, statistical analysis and drafting/revising the manuscript for content. RJdeH was involved in the analysis and interpretation of the data, statistical analysis and drafting/revising the manuscript for content.

  • Competing interests None declared.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.