Full Terms & Conditions of access and use can be found athttps://www.tandfonline.com/action/journalInformation?journalCode=plcp21

Language, Cognition and Neuroscience

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/plcp21

Determinants of aphasia recovery: exploratorydecision tree analysis

Durjoy Lahiri , Souvik Dubey , Alfredo Ardila , Debasish Sanyal & Biman KantiRay

To cite this article: Durjoy Lahiri , Souvik Dubey , Alfredo Ardila , Debasish Sanyal & Biman KantiRay (2020): Determinants of aphasia recovery: exploratory decision tree analysis, Language,Cognition and Neuroscience

To link to this article: https://doi.org/10.1080/23273798.2020.1777314

Published online: 09 Jun 2020.

Submit your article to this journal

View related articles

View Crossmark data

REGULAR ARTICLE

Determinants of aphasia recovery: exploratory decision tree analysisDurjoy Lahiria, Souvik Dubeya, Alfredo Ardilab,c, Debasish Sanyald and Biman Kanti Raya

aBangur Institute of Neurosciences, IPGMER and SSKM Hospital Kolkata, Kolkata, India; bInstitute of Linguistics and Intercultural Communication,I.M. Sechenov First Moscow State Medical University, Moscow, Russia; cPsychology Doctoral Program, Albizu University, Miami, FL, USA;dDepartment of Psychiatry, K.P.C. Medical College and Hospital, Kolkata, India

ABSTRACTOne hundred and sixty-three aphasia patients underwent initial language examination during thefirst week following stroke and 90–100 days post-stroke. Demographic factors (age, gender, andnumber of years of formal education), lesion-related factors (type of stroke, lesion volume,cortical versus sub-cortical location, and site of lesion), as well as initial severity and type ofaphasia were taken as independent variables while aphasia recovery (in terms of no changeversus change to a milder type or complete recovery) was the dependent variable. Chi squareautomatic interaction detection (CHAID) was performed to assess predictor importance andformulate a predictive model for aphasia recovery. Initial severity of aphasia followed by initialaphasia symptomatology was found to be the most important factor determining aphasiarecovery. Age and gender had some importance. Lesion-related factors did not reach statisticalsignificance as independent determinant of aphasia recovery. The predictive value of the modelwas 66.87%

ARTICLE HISTORYReceived 11 December 2019Accepted 27 May 2020

KEYWORDSAphasia recovery; aphasiaseverity; aphasiasymptomatology; decisiontree analysis; recoveryprediction

Introduction

Contemporary literature has pointed out that a diversityof factors are involved in aphasia recovery (Doogan et al.,2018). These include but are not limited to lesion relatedas well as aphasia related factors (Watila & Balarabe,2015). The literature, however, is not convergent whenit comes to determining the relative significance ofthese predictors in aphasia outcome and recovery.

Different predictors have been analysed. In one earlystudy in this field, Lazar et al. (2010) demonstratedinitial aphasia severity as the most important predictorfor recovery in post-stroke language dysfunction. Amore recent study that explored factors predictingrecovery in chronic aphasia observed- “chronic aphasiarecovery is positively influenced by higher education,smaller lesions, younger age, and current antidepressantuse (independently of one another)” (Suneja et al., 2014).In the same year, another study pointed out that recov-ery predictors after stroke are different in cases of rightor left hemisphere damage; the right hemisphere playsa crucial role in the recovery of aphasia following lefthemispheric stroke (Forkel et al., 2014). In their reviewpaper, Watila and Balarabe (2015) observed that ana-graphic factors like age, gender, handedness, and edu-cation were not robust predictors of aphasia recovery;rather lesion location and size, aphasia type and severity,and the nature of early hemodynamic response and

treatment received were the most important determi-nants of aphasia recovery after stroke. As far as thelong-term prognosis of aphasia after stroke is concerned,“SPEAK” model proposed by El Hachioui et al. (2013) hascontributed substantially to our knowledge in this area.According to this model, the outcome of aphasia atone year after stroke can be predicted in the first weekby phonology score, Barthel index (a scale used tomeasure performance in activities of daily living,Mahoney & Barthel, 1965), age, educational level, andstroke subtype, among which phonology score was themost robust predictor. The model has recently under-gone successful external validation (Nouwens et al.,2018).

Thus, it is understood that there is considerable diver-gence in the existing literature regarding the relativeimportance of different predictors in aphasia recovery.The aim of the present study was to explore the determi-nants of aphasia recovery in the first 3 months followingfirst ever acute stroke using a decision tree analysis toevaluate the importance of different predictors.

Materials and methods

With prior approval from the institutional ethics commit-tee, a study on post-stroke aphasia in Bengali-speakingpatients was conducted from 2016 to 2018, at a tertiary

© 2020 Informa UK Limited, trading as Taylor & Francis Group

CONTACT Alfredo Ardila ardilaalfredo@gmail.com

LANGUAGE, COGNITION AND NEUROSCIENCEhttps://doi.org/10.1080/23273798.2020.1777314

care stroke centre situated in eastern India. Basic findingsof “Kolkata aphasia study” have been included in pre-vious papers by the authors (Lahiri et al., 2019, 2020).

Participants

Consecutive patients with first ever acute stroke present-ing to our stroke unit were recruited in our study. Theinclusion criteria were (1) conscious and alert at thetime of language examination; (2) adult (more than 18years) literate subjects with aphasia after first everstroke; (3) speakers of the Bengali language (thatmeans any person capable of communicating in theBengali language during pre-morbid state). Exclusion cri-teria were as following: (1) aphasia as a consequence ofvascular catastrophe inside intra-cranial space-occupyinglesion; (2) non-linguistic cognitive disturbance affectinglanguage assessment;(3) documented or, suspecteddementia; (4) pre-morbid psychiatric illness hinderingcommunication (e.g. personality disorder);(5) abuse ofalcohol or drug; (Lahiri et al., 2019).

During the study period, 607 patients of first everacute stroke were collected, of whom 92 got excludeddue to reasons as following- death before first languageassessment (33); severe systemic complications duringan early post-stroke period (25); severe dysarthria oranarthria hindering language assessment (15); completeresolution of language deficit before initial assessment(12); extensive bilateral micro-vascular ischaemicchanges (7). In consequence, only 515 patients remainedfor further analyses.

In the index study, the definition of bilingualism wasas follows – the ability to communicate in two or morelanguages during interaction with other speakers ofthese same languages (Mohanty, 1994). This definitiontakes into account the ability for active language usagerather than grammatical proficiency and therefore is inalignment with the current concepts of bilingualismas life-long activity in contrast to passive learning(Bak, 2016).

Language examination

The Bengali version of Western Aphasia Battery (BWAB), avalidated scale for assessing the language function inadults (Keshree et al., 2013), was used to determine thepresence, severity, and type of aphasia. The parametersof BWAB are as following- Spontaneous Speech, Compre-hension, Repetition, Naming, and Aphasia Quotient (AQ).The AQ is an estimate of severity of aphasia and is a com-bined score based on the oral-auditory language sub-scales of Fluency, Comprehension, Repetition, andNaming. AQ below 93.8 is taken as the quantitative

cut-off for aphasia diagnosis. For the purpose of theindex study, initial language examination was performedbetween 3rd and 7th day following stroke in each of therecruited patients.

Severity of aphasia

Severity of aphasia in the present study was determinedby AQ scores which is a composite score based onfluency, comprehension, repetition, and naming accord-ing to BWAB. Magnitude of severity was semi-quantifiedas follows- mild (AQ = 76–93.8); moderate (AQ = 51–75);severe (AQ = 26–50); and very severe (AQ = 25 orless). Then the participants were further classified intonon-severe (AQ 50 or more) and severe group (AQ lessthan 50).

Aphasia recovery

Recruited patients were followed up for 3 months andrepeat language assessment was done between 90 and100 days post-stroke. During follow-up 45 patientswere lost; in consequence, only 163 participants wereanalysed after the repeated assessment. All the recruitedpatients underwent speech and language therapy admi-nistered by 2 speech-language pathologists for 1–2 h/week over 8–10 weeks, in addition to adequatemedical management. Aphasia recovery followingstroke was categorised as no change, change to amilder type/ lower severity score across severity scalesor recovery. Recovery from aphasia was defined as AQof 90.0 or more in the follow-up visit. The cut-off scoreassigned to recovery was data-driven. The lower andupper bounds of AQ (in the first visit) were 2.0 and90.0 respectively in our study population. Technicallyspeaking, it would mean that recovery is not equal tocomplete freedom from aphasia given the cut-off foraphasia diagnosis. Individuals who do not reach theaphasia cut-off score of 93.8, but are in the “recovery”group achieve significant recovery but may still presentclinically with aphasia. However, BWAB does not giveany cut-off scores for denoting recovery as the score of93.8 is used for aphasia diagnosis only.

Brain imaging

Brain imaging for ischaemic strokes in our study was per-formed by using Siemens 3T MRI machine (Magnetom-Verio DOT, 16 channels) with a standard quadraturehead coil. Computed tomography (CT) scan (Philips, 16slice) was utilised for initial brain imaging of patientssuffering from haemorrhagic stroke. Three patient ofischaemic stroke, who had contraindications (artificial

2 D. LAHIRI ET AL.

metallic heart valve or permanent pace-maker) for MRI,was evaluated by CT scan.

Visual analysis was performed in SiemensRSyngoviewer for lesion assessment and the images were readby neuro-radiologists to pinpoint the location of lesion.The definition of infarcted tissue was as following-tissue having abnormal high signal onT2 weightedand/or FLAIR (Fluid attenuated inversion recoverysequences) images. Diffusion weighted imaging wasemployed for detection and localisation of acute infarctin cases where MRI was performed within 48 h post-stroke. Location and extent of the lesion were deter-mined in reference to the sulcal anatomy. Cortical andmixed lesions were localised to the extent of lobes andgyri according to standard sulcal anatomy by studyingaxial, coronal and sagittal sections of brain in T2-weighted and diffusion weighted sequence. Similarly,for haemorrhagic stroke, CT brain was studied in threeorthogonal planes to determine the location andextent of lesions.

Lesion volume was calculated by using standard ABC/2 method in CT scan images for intra-cerebral haemor-rhage (Kothari et al., 1996) and in MR images (DWI orFLAIR) for ischaemic stroke (Sims et al., 2009). For sub-jects (3) with ischaemic stroke and contraindication forMRI, infarct volume was estimated by ASPECT scoringand CT perfusion score (Gaudinski et al., 2008).

Statistical analysis

Statistical analyses were conducted using IBM SPSS Stat-istics (version 25.0) and SPSS modeller (latest version).Demographic factors (age, gender, bilingual status, andthe number of years of formal education), lesion-related factors (type of stroke, lesion volume, and corticalversus sub-cortical location), initial aphasia severity aswell as type of aphasia were taken as independent vari-ables whereas aphasia recovery in terms of no change,change or complete recovery was the dependent vari-able. Chi square and Fisher’s exact tests were done tostudy the factors associated with and without aphasiatype change or recovery (as applicable). The means ormedians of continuous variables were compared usingthe Student t test or Mann Whitney U test (as applicable).

CHAIDChi square automatic interaction detection (CHAID),which is a decision tree analysis, was performed toassess predictor importance and formulate a predictivemodel for aphasia recovery.

Decision tree algorithm belongs to supervised datamining technique and is considered an effective toolfor multivariate data exploration or prediction (Chen

et al., 2003; Piramuthu, 2008). As the name implies,decision tree has the appearance of a tree containingroot node, branches, and leaf nodes. Initially, the rootnode is established in a decision tree followed by theaddition of nodes based on problem and classificationattributes. Each node is representative of an attributewhile branches represent the values of the tested attri-bute followed by the leaf nodes that represent classifi-cation categories. Branches are employed to decide ifdata entry is to be applied to the child node in the follow-ing layer. The process continues until all data is enteredup to the level of leaf nodes. Rule of so-called if–then canbe generated from the results available at leaf nodes.Decision tree makes no assumptions on relationshipsbetween features. It just constructs splits on single fea-tures that improve classification, based on an impuritymeasure like Gini or entropy. On the basis of theextracted classification rules, a pruned decision treemodel is valid for data prediction. Pruning addressesthe problem of skewed interpretation that can occurfor variables with more splits. Three kinds of decisiontree algorithms are widely used for data prediction pur-poses in various settings. These are – classification andregression tree, CHAID and quick unbiased efficient stat-istical tree algorithms. A recent study analyzing the capa-bility of these three models in terms of their accuracy ofprediction found, CHAID generated the maximumclassification rules and displayed the highest accuracyof prediction in comparison to the other two (Lin &Fan, 2019). The application of CHAID model has beengradually increasing in the medical field to address pro-blems that relate to data prediction, although it has notbeen employed so far in predicting aphasia outcome/recovery which in itself is a complicated subject.

Decision tree in this paper was built in SPSS modeller,the method of growth being exhaustive CHAID. Inputvariables were all the independent variables mentionedabove that included demographic factors, lesion-related factors and aphasia related factors.

Results

The initial aphasia sample included 208 patients, ofwhom 163 turned up for repeat language assessment.The mean age of patients with aphasia was 52.19 years(SD = 10.96) with male/female ratio of 2.25:1. Among208 participants with aphasia, 53 (25.48%) qualified asbilingual and rest 155 were monolinguals. The ratio ofischaemic/haemorrhagic stroke in our sample of partici-pants with aphasia was 1.97:1. Proportion of participantspresenting aphasia between ischaemic and haemorrha-gic stroke did not differ significantly (0.40 versus 0.39).

LANGUAGE, COGNITION AND NEUROSCIENCE 3

On univariate analysis the following factors wereassociated with a change to milder type or recovery –haemorrhagic stroke (p <.001); pure sub-cortical stroke(p < .001), and initial non-severe classification (p < .001).From the perspective of initial aphasia type, mostchange (p < .001) was observed with global aphasia(usually becoming a Broca’s aphasia) while most recov-ery (p < .001) was observed with milder aphasia types(transcortical motor, transcortical sensory, and anomicaphasia). Sex and age were weaker predictors (p < .291and 0.512, respectively) with men and younger patientsrevealing better recovery. In CHAID analysis, the mostimportant predictor favouring type change or recoverywas initial non-severe classification (Figures 1 and 2).The overall predictive value of the CHAID model, as com-puted in the SPSS modeller by cross-validation of thedecision tree (the algorithm was trained by 70% of thedata), was found to be 66.87%. Confusion matrix wasdeveloped and the recall percentage for each group ofaphasia outcome alongside the overall accuracy of pre-diction has been shown in Table 1. It is observed thatthe recall percentage of the CHAID model is highest for“no change” group (71.27%). However, there is notmuch difference in terms of recall percentage for theremaining two groups as well, namely, “change” and“recovery” (59.09% and 64.00%, respectively).

Discussion

The issue of predicting aphasia recovery following strokehas gained particular importance in recent times giventhe development of rehabilitation measures. Althoughseveral contemporary studies have explored the determi-nants of aphasia recovery, the relative importance ofthese factors remains controversial. To the best of our

knowledge, this is the first time exploratory decisiontree analysis was performed to determine the relativeimportance of predictors for aphasia recovery. Thestrengths of CHAID decision tree analysis are as follows:the technique is based on adjusted significance testing(Bonferroni testing); it does not require the data to be nor-mally distributed; continuous variables can be utilised ascriterion variable and finally output is highly visual aswell as easy to interpret. These features of CHAID analysishave been particularly instrumental in our work becausethe study setting being a tertiary care centre majority ofthe patients presented severe aphasia. We did not use amultiple regression model in the present study becauseof the following reasons. Firstly, the objective was toexplore the data to build a decision tree which is knownto give a visual output that can be better interpreted byclinicians. Algorithms rather than regression statisticsare more appealing in clinical situations. Secondly, thedata in our study was not normally distributed. Reasonsare related to the study setting which is a tertiary carecentre and we encounter mostly severe cases. For adataset which does not satisfy the test of normality,regression model assumptions will be violated. Thirdly,CHAID model is known to take care of collinearity issue,had it been present. We did run the collinearity diagnos-tics in SPSS. The maximum variance inflation factor (VIF)was for AQ in first visit (aphasia severity) which is 2.312.As far as collinearity is concerned, as long as the VIF isbetween 0.2 and 10, it should not concern us.

According to our observation, the initial severity ofaphasia remains the most important determinant foraphasia recovery following stroke given comparablerehabilitation measures to all the participants. Initialnon-severe classification of aphasia was a predictor infavour of type change or recovery. This information is

Figure 1. Predictor importance for type change or recovery.

4 D. LAHIRI ET AL.

Figure 2. Predictive model for aphasia type change or recovery (CHAID analysis) showing aphasia type change or recovery as thedependant variable in the root node. The parent node shows initial aphasia severity as the first branching point followed by ageand sex as the subsequent branching points generating the child nodes for non-severe aphasia and severe aphasia, respectively.

Table 1. Predictive value of the developed decision tree model (exhaustive CHAID) for each group; bold numbers denote the positivepredictive value for each group.Actual (n = 163) Predicted Total Recall (%) Classification accurate (%)

No change Change RecoveryNo change 67 22 5 94 71.27Change 11 26 7 44 59.09Recovery 2 7 16 25 64.00Overall 66.87

LANGUAGE, COGNITION AND NEUROSCIENCE 5

crucial in the prognosis as well as in the appropriateapplication of rehabilitation measures particularly inresource limited settings such as in developing countries.The second important factor, as was observed in thedecision tree, is the type of aphasia. Patients with initialglobal aphasia were more prone to undergo change intype as well as severity. People with global aphasia hada tendency to evolve into Broca’s aphasia; which is inagreement with the current literature (e.g. Pedersenet al., 2004). It is understandable that global aphasia fol-lowing acute stroke sometimes would represent anetwork dysfunction, particularly if caused by a small cor-tical lesion or sub-cortical lesion. In that case, it is not sur-prising to find global aphasia cases showing type andseverity change when assessed in chronic phase. Onthe contrary, Broca’s aphasia and Wernicke’s aphasiacases were less likely to undergo type or severitychanges. Probably these cases represent actual damageto the inferior frontal gyrus and superior temporalgyrus rather than network dysfunction. It is interestingthat transcortical motor aphasias with low AQ scoresbehaved similar to Broca’s and Wernicke’s aphasia.

Demographic factors, such as age and gender, andmore surprisingly, lesion-related factors such as lesionvolume did not reach statistical significance on univariateanalysis. This absence of statistical significance may bedue to the fact that the majority of the patients presentedsevere aphasia and consequently dispersion was low.However, in CHAID analysis demographic factors turnedout to be important. Older patients with global aphasiawere less likely to undergo type or severity change prob-ably related with a decreased brain plasticity. When itcomes to Broca’s and Wernicke’s aphasia, females wereobserved to lag behind in recovery compared to theirmale counterparts. However, in our sample gender hadquite uneven distribution: the male/female ratio was2.25:1. This uneven distribution may suggest that malesand females, because of some uncontrolled socioculturalfactors, may have a different probability to recur to aneurological hospital in cases of stroke, and hence theaphasia severity could be different, probably being as awhole less severe in males. Formal education, the otherdemographic factor that was taken into account in thisstudy was not found to have influenced recovery to a sig-nificant extent. Interestingly, the bilingual status of theparticipant appeared to be important in post-strokeaphasia recovery, even though it did not reach statisticalsignificance on univariate analysis. Patients with bilingualstatus in their pre-morbid state presented better recoverycompared to monolingual participants (p = .124). Thisobservation is in keeping with that of recent literature(Paplikar et al., 2019). A very recent paper by theauthors, as part of the “Kolkata aphasia study”,

demonstrated that bilingualism is associated with theless severe magnitude of aphasia (Lahiri et al., 2020).Although the finding did not pass the regression test, ithints towards a possible protective role of bilingualismin post-stroke aphasia. Going a step further in this direc-tion, when the specific effect of bilingualism on aphasiaseverity was analysed with lesion volume and educationas co-variates, it was concluded by the authors that“aphasia is less severe and consequently, recovery isexpected to be better in bilingual patients” (Ardila &Lahiri, 2020). The influence of bilingual status on aphasiarecovery however can be better understood if these vari-ables are explored in a more specific and direct mannerand not as part of analyzing other determinants as wasdone in the index paper. In addition, taking into accountthe nuances of bilingualism, for instance, age of acqui-sition and level of proficiency in both languages, mayreveal more valuable information in this context. Havingsaid that, firm conclusions about bilingualism andaphasia cannot be drawn at this point, for evidences oneither side of the assumption are available.

As per our univariate analysis, ischaemic strokepatients were found to have poor recovery; which canbe explained by the nature of damage imparted to thebrain tissue by ischaemia in comparison to haemorrhage.Cortico-subcortical mixed lesions also presented poorrecovery particularly in comparison to pure sub-corticalstrokes. Recent literature has recognised the crucialrole of sub-cortical structures in language function (e.g.Negwer et al., 2017). Sub-cortical damage, in additionto the cortical lesion, therefore is a hindrance toaphasia recovery following stroke.

There are several limitations of the present studywhich we would like to improve upon in our subsequenteffort. Firstly, newer stroke therapy was not included inthe analysis as because the number was too small (n =5). Secondly, most patients belonged to very severeaphasia group possibly caused by the study setting (ter-tiary care stroke centre). Thirdly, modern techniques oflesion volume assessment could not be employed dueto resource constraint. The potential limitations ofdecision tree analysis are also worth mentioning. Thetree is significantly susceptible to variations in theinput variables. In addition, the findings may not beextrapolated to out-of-sample population and theresults are to be interpreted in light of clinical meaning-fulness. Therefore further validation of the findings in alarger sample is required.

Conclusion

In the present study, which is part of a larger epidemio-logical data on aphasia after stroke, we attempted to

6 D. LAHIRI ET AL.

explore the determinants of post-stroke aphasia recoveryin Bengali-speaking subjects. To our knowledge, it is thefirst time that decision tree analysis has been performedto explore the question- which factors are most impor-tant in determining aphasia recovery after stroke.According to the CHAID analysis conducted for thisstudy, the initial severity of aphasia followed by initialaphasia symptomatology was found to be the mostimportant factor determining aphasia recovery instroke patients receiving comparable rehabilitationmeasures. Among the demographic factors, age andgender were also observed to be of some importance.Lesion-related factors, particularly lesion volume, didnot reach statistical significance as independent determi-nant of aphasia recovery in stroke, probably due to thecharacteristics of our aphasia sample. Despite few afore-mentioned limitations of the present endeavour, westrongly believe that our present work will be a foun-dation for stronger predictive results in the future.

Acknowledgements

We are sincerely thankful to Professor Stefano F. Cappa for hisvaluable guidance in preparing a former version of the manu-script, Professor David Copland and Professor Loraine K. Oblerfor sharing their thoughts during the initial presentation ofthis work and Professor Marsel M. Mesulam for his suggestionsat different points while conducting the “Kolkata aphasiastudy”. We extend our sincere gratitude to Adriana Ardila forher editorial support. Preliminary findings of this paper werepresented at the 57th annual meeting of the Academy ofAphasia in Macau, China (October 27–29, 2019). This paperwas accepted for presentation and received an internationalscholarship award (2020) at the 72nd Annual Meeting of theAmerican Academy of Neurology (this meeting was called offdue to COVID-19 pandemic).

Disclosure statement

No potential conflict of interest was reported by the author(s).

References

Ardila, A., & Lahiri, D. (2020). Aphasia in bilinguals. medRxiv2020.04.22.20075432. https://doi.org/10.1101/2020.04.22.20075432

Bak, T. H. (2016). Cooking pasta in La Paz. Linguistic Approachesto Bilingualism, 6(5), 699–717. https://doi.org/10.1075/lab.16002.bak

Chen, Y. L., Hsu, C. L., & Chou, S. C. (2003). Constructing a multi-valued and multi-labeled decision tree. Expert Systems withApplications, 25(2), 199–209. https://doi.org/10.1016/S0957-4174(03)00047-2

Doogan, C., Dignam, J., Copland, D., & Leff, A. (2018). Aphasiarecovery: When, how and who to treat? Current Neurologyand Neuroscience Reports, 18(12), 90. https://doi.org/10.1007/s11910-018-0891-x

El Hachioui, H., Lingsma, H. F., van de Sandt-Koenderman, M. E.,Dippel, D. W., Koudstaal, P. J., & & Visch-Brink, E. G. (2013).Recovery of aphasia after stroke: A 1-year follow-up study.Journal of Neurology, 260(1), 166–171. https://doi.org/10.1007/s00415-012-6607-2

Forkel, S. J., de Schotten, M. T., Dell’Acqua, F., & Kalra, C. M.(2014). Anatomical predictors of aphasia recovery: A tracto-graphy study of bilateral perisylvian language networks.Brain, 137(7), 2027–2039. https://doi.org/10.1093/brain/awu113

Gaudinski, M. R., Henning, E. C., Miracle, A., Luby, M., Warach, S.,& Latour, L. L. (2008). Establishing final infarct volume:Stroke lesion evolution past 30 days is insignificant. Stroke,39(10), 2765–2768. https://doi.org/10.1161/STROKEAHA.107.512269

Keshree, N. K., Kumar, S., Basu, S., Chakrabarty, M., & Kishore, T.(2013). Adaptation of the Western Aphasia Battery in Bangla.Psychology of Language and Communication, 17(2), 189–201.https://doi.org/10.2478/plc-2013-0012

Kothari, R. U., Brott, T., Broderick, J. P., Barsan, W. G., Sauerbeck,L. R., Zuccarello, M., & Khoury, J. (1996). The ABCs of measur-ing intracerebral hemorrhage volumes. Stroke, 27(8), 1304–1305. https://doi.org/10.1161/01.STR.27.8.1304

Lahiri, D., Dubey, S., Ardila, A., & Ray, B. K. (2020). Factorsaffecting vascular aphasia severity. Aphasiology, 1–9.https://doi.org/10.1080/02687038.2020.1712587

Lahiri, D., Dubey, S., Ardila, A., Sawale, V. M., Roy, B. K., Sen, S., &Gangopadhyay, G. (2019). Incidence and types ofaphasia after first-ever acute stroke in Bengali speakers: Age,gender, and educational effect on the type of aphasia.Aphasiology, 1–14. https://doi.org/10.1080/02687038.2019.1630597

Lazar, R. M., Minzer, B., Antoniello, D., Festa, J. R., Krakauer, J.R., & Marshall, R. S. (2010). Improvement in aphasia scoresafter stroke is well predicted by initial severity. Stroke, 41(7), 1485–1488. https://doi.org/10.1161/STROKEAHA.109.577338

Lin, C., & Fan, C. (2019). Evaluation of CART, CHAID, and QUESTalgorithms: A case study of construction defects inTaiwan. Journal of Asian Architecture and BuildingEngineering, 18(6), 539–553. https://doi.org/10.1080/13467581.2019.1696203

Mahoney, F. I., & Barthel, D. (1965). Functional evaluation: TheBarthelIndex. Maryland State Medical Journal, 14, 56–61.

Mohanty, A. K. (1994). Bilingualism in multilingual society:Psychosocial and pedagogical implications. Central Instituteof Indian Languages.

Negwer, C., Ille, S., Hauck, T., Sollmann, N., Maurer, S., Kirschke, J.S., & Krieg, S. M. (2017). Visualization of subcorticallanguage pathways by diffusion tensor imaging fiber track-ing based on rTMS language mapping. Brain Imaging andBehavior, 11(3), 899–914. https://doi.org/10.1007/s11682-016-9563-0

Nouwens, F., Visch-Brink, E. G., El Hachioui, H., Lingsma, H. F.,van de Sandt-Koenderman, M. W., Dippel, D. W., & de Lau,L. M. (2018). Validation of a prediction model for long-termoutcome of aphasia after stroke. BMC Neurology, 18(1), 170.https://doi.org/10.1186/s12883-018-1174-5

Paplikar, A., Mekala, S., Bak, T. H., Dharamkar, S., Alladi, S., & Kaul,S. (2019). Bilingualism and the severity of poststroke aphasia.Aphasiology, 33(1), 58–72. https://doi.org/10.1080/02687038.2017.1423272

LANGUAGE, COGNITION AND NEUROSCIENCE 7

Pedersen, P.M., Vinter, K., &Olsen, T. S. (2004). Aphasia after stroke:Type, severity and prognosis. Cerebrovascular Diseases, 17(1),35–43. https://doi.org/10.1159/000073896

Piramuthu, S. (2008). Input data for decision trees. ExpertSystems with Applications, 34(2), 1220–1226. https://doi.org/10.1016/j.eswa.2006.12.030

Sims, J. R., RezaiGharai, L., Schaefer, P. W., Vangel, M., Rosenthal,E. S., Lev, M. H., & Schwamm, L. H. (2009). ABC/2 for rapid

clinical estimate of infarct,perfusion, and mismatchvolumes. Neurology, 72(24), 2104–2110. https://doi.org/10.1212/WNL.0b013e3181aa5329

Suneja, A., Gonzalez-Fernandez,M., & Hillis, A. (2014). Predictors ofrecovery of chronic aphasia. Neurology, 82(10), 6.228.

Watila, M. M., & Balarabe, S. A. (2015). Factors predicting post-stroke aphasia recovery. Journal of the Neurological Sciences,352(1-2), 12–18. https://doi.org/10.1016/j.jns.2015.03.020

8 D. LAHIRI ET AL.