Introduction

Because of an increase in long-term survival in children with cerebellar tumors [1], interest has risen in identifying long-term sequelae of the tumor, the tumor treatment and the need for cognitive rehabilitation [2]. Apart from a broad range of motor and cognitive impairments, cerebellar tumor survivors may present with speech (e.g., dysarthria) and language impairments (e.g., agrammatism) [3,4,5]. Although 25% to 35% of children present with postoperative cerebellar mutism syndrome (pCMS), characterized by severe language deficits, this is not a prerequisite for language difficulties [4, 6].

Previous studies describe language disorders in pediatric cerebellar tumor survivors based on formal language measures (e.g., picture naming task; assessing a specific language function) [3, 5], but also in spontaneous language analyses (e.g., storytelling; assessing multiple language functions simultaneously) [5, 7, 8]. Nonetheless, comprehensive reports of strengths and weaknesses at specific levels of language processing are lacking in the literature though these are necessary for targeted language rehabilitation [2, 6]. Recently, Svaldi et al. [7] provided a detailed spontaneous language analysis across four language processing levels (i.e., semantic, lexical, phonological, morphosyntactic) using standard measures and psycholinguistic properties. Impairments were identified across all levels of language processing, irrespective of pCMS diagnosis. Importantly, several patients presented with reduced lexical and/or grammatical accuracy but an error analysis was lacking. Therefore, the present study will build on this work by 1) characterizing the types of (lexico-)phonological, lexical-semantic and morphosyntactic errors in the spontaneous language of pediatric cerebellar tumor survivors and 2) examining the potential association between pCMS and the nature of the language impairments. The participant group is the same as reported by Svaldi et al. [7].

Methods

Participants

The patient group consisted of 12 (five females, seven males) long-term survivors of pediatric cerebellar tumors with a mean age at assessment of 11;3 years (SD = 6;3 years, range = 3–24;2 years). Patients were recruited via the Erasmus Medical Centre/Sophia Children´s Hospital in Rotterdam as part of an earlier study [9] and time since surgery ranged between 11 months and 12;3 years (M(SD) = 4;8(3;8) years). Five patients suffered from pCMS. Five cerebellar tumor survivors were diagnosed with a medulloblastoma, five with a pilocytic astrocytoma, and two with an ependymoma.

Control data were gathered until each patient could be matched in age and gender with five controls resulting in a total of 39 participants with no history of language or learning impairment (M(SD) = 11;1(5;11) years; range = 3;0–24;3 years), including 24 male and 15 female participants. The absence of language impairment or delay was confirmed through formal language tests and questionnaires (see [7]). A detailed description of the inclusion and exclusion criteria of participants can be found in Svaldi et al. [7] and patient characteristics are reported in the Supplementary materials (reprint from Svaldi et al. [7]). See Appendix A for the cognitive status of patients within one year before language assessment.

Procedure

Spontaneous language samples (minimum 300 words) were gathered based on structured conversations and picture descriptions and were transcribed following the methods outlined in Svaldi et al. [7]. Patient data were collected in 2007 in one testing session. Control data were collected in 2020 and 2021 and consisted of two testing sessions since additional language tests were administered in the first session to rule out language impairment.

Data Coding and Analysis

The samples analyzed by Svaldi et al. [7] were coded for the types of (lexico-) phonological, lexical-semantic and morphosyntactic errors produced (see Appendix B for descriptions of the error types). The proportion of each error type was calculated by dividing the number of errors in the whole sample (i.e., conversation + picture descriptions) by the total number of utterances. The error proportions were compared between each patient and their individually matched set of five controls using modified t-tests for single-case statistics [10].

Results

Each patient produced a greater proportion of errors than controls on at least one of the error types. Significant differences occurred across all language processing levels and are indicative of impaired spontaneous language (see Table 1). Findings of increased error occurrence in patients appeared to be more frequent in the pCMS-group (19 significant findings in five patients) than the non-pCMS-group (12 significant findings in seven patients). Descriptive statistics of the proportions of errors produced and results of the individual case statistics can be found in the Supplementary materials. The level of significance was set at α < 0.05.

Table 1. Results of individual comparisons of error proportions between cerebellar tumor survivors and controls. Arrows indicate a significantly higher (↑) or lower (↓) proportion of an error type compared to controls. Grey shading indicates language processing levels for which there was evidence of impairment in Svaldi et al. [7]
Full size table

Three out of 12 patients, of whom two were diagnosed with pCMS, produced (lexico-) phonological errors (i.e., phonemic paraphasias). Lexical-semantic errors were observed in eight patients. The production of general-all-purpose (GAP) verbs was the most common lexical-semantic error for individuals with and without pCMS. Seven out of 12 patients produced morphosyntactic errors and, while keeping in mind small numbers, tended to occur more in the pCMS-group (4/5 in pCMS-group relative to 3/7 in non-pCMS-group). Here, morphological errors in verb inflection were most often observed, particularly in the pCMS-group.

For eight patients, error patterns aligned with the observations of Svaldi et al. [7] based on standard spontaneous language measures and psycholinguistic properties (see Table 1). For three patients (P23, P17, P16) the psycholinguistic analysis detected impairments in lexical-semantics or morphosyntax that were not confirmed by the error analysis. A greater mismatch was observed for (lexico-)phonological processing, where the error analyses did not reveal significant findings in five out of eight patients in whom the psycholinguistic analysis indicated a phonological impairment.

In contrast, four other patients (P2, P6, P8, P24) showed increased lexical-semantic errors, despite no difference to controls in spontaneous language measures included in Svaldi et al. [7]. This was possibly due to limited measurable spontaneous language for two of these patients (P2 and P6), which led to exclusion from some comparisons. For two patients (P6, P23), error analyses revealed additional morphosyntactic difficulties.

Discussion and Conclusion

The findings of this error analysis demonstrated that long-term (lexico-)phonological, lexical-semantic and morphosyntactic impairments may be present in cerebellar tumor survivors, in the presence of normal scores in other cognitive domains. This aligns with earlier studies and supplements previous reports with a more detailed evaluation of language processing abilities [5, 7, 11]. Such comprehensive evaluation seems thus warranted and necessary for targeted language therapy [2].

Lexical-semantic and morphosyntactic errors were most common in our patient group, aligning with earlier reports on difficulties in these language levels [5, 11, 12]. An increased number of GAP verbs was the most common lexical-semantic error and may reflect immature language or word-finding difficulties [13]. Verb inflection mistakes were the most common morphosyntactic errors and have previously been reported in this population [5]. Further research is necessary, however, to confirm these findings which should include more phonological error types (e.g., phonemic distortions, differentiating between programming and selection impairment).

More differences between patients and controls were found in the pCMS-group compared to the non-pCMS-group. Morphosyntactic disorders may be particularly frequent in the pCMS-group [6]. Notably, preoperative morphosyntactic disorders were suggested as risk factors for pCMS [11]. This could mean that the higher proportion of morphosyntactic errors in the pCMS-group reflects pre-surgical impairments. However, this can only be established in future research including a preoperative assessment and equal groups of patients with and without pCMS, as well as a larger sample size so that more generalizable patterns may emerge. Similarly, prevalence data is notoriously difficult to establish in small-N studies.

It should also be considered that the occurrence of pCMS may result in less reliable spontaneous language assessment, given the reduced output. However, where standard and psycholinguistic measures detected no impairments for some patients [7], the error analysis did. The opposite was also observed, indicating that different analyses can provide complementary information, potentially tapping into different underlying impairments. A combination of standard measures, psycholinguistic analyses, and error analyses may increase the sensitivity to identify impairments and to characterize their language processing nature [13].

The error analysis conducted in the present study suggested long-term language impairment in cerebellar tumor survivors. A detailed and individualized language follow-up seems necessary to investigate the processing nature of these impairments.