The longitudinal relevance of phonological speech for predicting cognition in first-episode psychosis

(1)

The longitudinal relevance of phonological speech for predicting cognition in first-episode psychosis

Isa van Nimwegen

Bachelor Thesis Psychobiology

University of Amsterdam

Student number: 11582057 Supervisor: A.E. Voppel

Date: 22-01-2021

(2)

Abstract

Impairments in cognitive functioning and phonological speech are related to first-episode psychosis (FEP) and impair quality of life. However, to date no research has examined the relationship of cognition and phonological speech in FEP patients over time. Therefore, this research focused on mapping cognition and phonological speech measures over time and examining what phonological aspects are longitudinally relevant for the prediction of different facets of cognition in FEP patients. The participants of this study included 50 FEP patients aged between 21 and 65 years. Using multiple regression analyses the predictive value of phonological speech measures and different cognitive facetswas

examined. Results show instability of most but not all measures of cognition and phonological speech. Additionally, phonological speech was predictive for global cognition. The present study offers some validation of the relation between cognition and speech disturbances. Future research could focus on the role that type and dosage of medication has on the relationship between phonological speech and cognition.

(3)

Introduction

First-episode psychosis (FEP) is the first occurrence of a psychotic episode and has a variety of characterizing symptoms, including impaired cognition and speech disturbances. FEP has been associated with a variety of incidences. Anderson, Fuhrer, Abrahamowicz and Malla (2012) reported higher incidences in impoverished environments and higher incidences for male FEP patients (113:100 000) than for female FEP patients (43:100 000). Additionally, much higher incidences have previously been reported for younger FEP patients (aged

between 15 and 29) than for older patients (aged between 30 and 59), while the older patients had a significantly higher risk for developing FEP (Simon et al., 2017). FEP typically has an onset in adolescence and can occur in multiple disorders among which are brief psychotic disorder, schizophreniform disorder, schizophrenia, schizoaffective disorder and other specified schizophrenia spectrum and other psychotic disorder (American Psychiatric Association, 2013; see also Kilciksiz, Keefe, Benoit, Öngür, & Torous, 2020).

The diagnosis of post-traumatic stress disorder (PTSD) is related to FEP as well. Rodrigues and Anderson (2017) reported the experience of a psychotic episode can evoke feelings of extreme anxiety and depression as well as trauma. So much so that one in two FEP patients showed signs of PTSD, while one in three FEP patients met the diagnosis PTSD. The high prevalence of PTSD in FEP patients shows the impact of FEP on the quality of life and functioning of the patients. Hence, the early recognition and treatment of FEP is important to decreasethis negative impact.

To help recognize FEP studies have shown phonetic features are linked to psychosis or the presence of a psychotic diagnosis and are potential features useful for identification. Tahir et al. (2019) for example showed phonological speech measures, such as percentage of

speaking time, were relevant for recognizing negative psychotic symptoms. In addition, Parola, Simonsen, Bliksted and Fusaroli (2020) investigated the relevance of a number of acoustic markers for psychotic symptoms of which pause duration could be clinically relevant. In line with these findings, De Boer, Brederoo, Voppel and Sommer (2020a)

reported a greater number of pauses and longer pause durations have been found in speech of people with schizophrenia-spectrum disorder compared to healthy controls. Additionally, speech disturbances seem to be stable over time and are related to impairments in social functioning. Furthermore, a decreased percentage of speaking time appears to be related to more severe psychotic symptoms (Cohen, Najolia, Kim & Dinzeo, 2012). In sum,

(4)

phonological speech (i.e. percentage of speaking time, pause duration and number of pauses) is related to the severity of psychotic symptoms and impairs social functioning, which impacts the quality of life.

Additionally, one of the main symptoms of FEP is impaired cognition (Schuepbach, Keshavan, Kmiec, & Sweeney, 2002). Former studies showed multiple cognitive difficulties related to FEP, namely processing speed, verbal memory, executive functioning and working memory (Corrión et al., 2011; see also Riley et al., 2000). Moreover, impairments in verbal fluency is well known in psychotic disorders, specifically in schizophrenia (Allen, Liddle & Firth, 1993). Bowie et al. (2004) showed that impairments in verbal fluency were related to severity of other cognitive, linguistic and functional problems.

In addition, a variety longitudinal studies have found that impairments in cognition in FEP patients appeared to be stable over time (Bozikas & Andreou, 2011; Hill, Schuepbach, Herbener, Keshavan & Sweeney, 2004; Kurtz, 2005). Contrarily to results of Zanelli et al. (2019) which showed verbal knowledge and memory declined over time. As a result of these cognitive deficits everyday functioning of FEP patients is decreased (Bowie et al., 2004).

Varying theories exist of the connection between language and cognition. Tillas (2015) adheres to the view that language is of significant importance to gain control over abstract ideas. However, others see language and cognition as different independent concepts where language is primarily used for communication. Furthermore, proponents of the supra-communicative view build on this view where language is hypothesized to not only be necessary for communication but also influence cognition, behavior and problem-solving. Another theory describing the relationship between language and cognition is the ‘label-feedback hypothesis’, which suggests that language modulates cognition (Lupyan, 2012). Whereas, Nagumo et al. (2020) propose that motor processes are required for language production. These motor processes are impaired when cognitive functioning is decreased, which in turn causes impairments in language production. Additionally, speech of people with schizophrenia-spectrum disorder, one of the psychotic disorders related to FEP, has been related to cognition. Cohen, Kim and Najolia (2013) reported pause duration scores were correlated with attention in schizophrenia patients. This evidence shows the relationship between cognition and language. This results in theories stating that impairments and variances in spoken language may be underlying to impairments in cognition. However, studies testing this hypotheses have been lacking (Cohen & Elvevåg, 2014).

Even though theories suggest a relationship exist between language and cognition, little research has been done to examine this. Additionally, the little research that has focused

(5)

on investigating the relationship between phonological language and cognition, like Cohen et al. (2013), focused on a single point in time. Therefore, this study will mainly focus on the relationship between phonological language and various cognitive domains as well as the stability of acoustic phonetics and cognition in FEP patients over time and aim to answer the following research question: What phonological language aspects are longitudinally relevant for the prediction of different facets of cognition in first-episode psychosis? The present study will help in validating the relation between cognition and speech disturbances, especially regarding the as of yet missing longitudinal aspect, a key measure important for its use as a (bio)marker of psychosis as a disorder and especially as a marker for the specific cognitive impairments which influence quality of life. Phonological speech related to FEP is

hypothesized to be relevant predictors for cognition. This hypothesis is tested by investigating spontaneous speech and cognitive tasks. Chiefly it is expected that the mean pause duration, number of pauses and percentage of speaking time to be relevant for the prediction of verbal memory, verbal fluency, attention and speed of information processing and global cognition.

Materials & Methods

Participants

Participants were FEP subjects enrolled in the HAMLETT study of which speech and cognition data was available. Dedicated includers were responsible for the recruitment of the participants following treatment at one of the twenty-four both inpatient and outpatient Dutch mental health clinics used for the study. For more details regarding the recruitment see

Begemann et al. (2020). In short, participants were Dutch-speaking FEP patients in remission at the baseline of the study (number: 50, age: 21-65 (mean ± standard deviation (29.76 ± 10.19)), both males (n=36) and females (n=14).

Inclusion criteria were, additionally to HAMLET inclusion criteria (see Begemann et al., 2020), availability of both BACS and PRAAT data on the second visit (baseline) and fourth visit (6 months after the second visit) of the HAMLETT study. Three participants with insufficient audio quality were excluded.

Measures

As measures for cognition and speech data of the BACS and audio collected of FEP patients were used. The BACS collected data on numerous specific facets of cognition

(6)

including: verbal memory – as measured using the list learning task, working memory – as measured using the digit sequencing task, motor speed – as measured using the token motor task, verbal fluency – as measured using the category instances and controlled oral word association tests, attention and speed of information processing – as measured using the symbol coding test and executive functions – as measured using the Tower of London test (for an in-depth description of the BACS, see Keefe et al., 2004). The composite score of the BACS was used as a measure for global cognition. Raw scores of the BACS were converted to z-scores correcting for age and gender.

Audios of spontaneous speech of participants were collected using semi-structured interviews. The digitally recorded audio was split to exclude interviewer audio. From the subject’s speech phonological aspects of language were extracted using the PRAAT syllable nuclei script (De Jong & Wempe., 2009), yielding features number of pauses, mean pause duration and percentage of speaking time for analysis (for an in-depth explanation of methodology, see De Boer et al., 2020b).

Statistical analysis

Statistical analysis was carried out using R (version 3.4.1). The three explanatory variables include mean pause duration, number of pauses and percentage of speaking time. The response variables consisted of the global cognition, verbal memory, verbal fluency, working memory, motor speed, attention and speed of information processing and executive functions z-scores tested using the BACS. A multiple regression analysis was carried out to determine the effect of the explanatory variables on the response variables on timepoint 1. The z-scores of global cognition, verbal memory, verbal fluency, working memory, motor speed, attention and speed of information processing and executive functions (Yz-scorep) were predicted by the intercept (i0), the mean pause duration (Xpausedurationp), number of pauses (Xnumberofpausesp) and percentage of speaking time (X%speakingtimep) and the corresponding coefficients (bpausdurationp, bnumberofpausesp and b%speakingtimep) for each participant (p) (Simms, Engebretson, Pilipenko, Reeves & Clilverd., 2016):

Yz-scorep= i0 + bpausdurationp* Xpausedurationp + bnumberofpausesp* Xnumberofpausesp + b%speakingtimep* X%speakingtimep

This predictive model on timepoint 1 was used to produce predictions of verbal memory, verbal fluency, working memory, motor speed, attention and speed of information

(7)

processing and executive functions z-scores on timepoint 2. These predictions were compared to the actual z-scores of corresponding cognitive facets using a paired samples t-test. For an overview of the design see Figure 1. Finally, a power analysis was conducted.

The four assumptions of multiple regression analysis described by Flatt and Jacobs (2019) were tested. Outliers were not removed from the data to maintain a realistic

representation of variability in the FEP subjects. A significance level of α= 0.05/7=0.007 after Bonferroni correction was used (Bland & Altman, 1995).

Figure 1

Overview of the design

A)

B)

Note. Figure of the overview of the design. (1): multiple regression models for global

cognition, verbal memory, verbal fluency, working memory, motor speed, attention and speed of information processing and executive functions, with explanatory variables mean pause duration, number of pauses and percentage of speaking time. (2): Predictions of the z-scores of global cognition, verbal memory, verbal fluency, working memory, motor speed, attention and speed of information processing and executive functions were created using the multiple regression equation. (3): predictions created in (2) were compared to the actual z-scores of global cognition, verbal memory, verbal fluency, working memory, motor speed, attention and speed of information processing and executive functions using a paired samples t-test.

(8)

Results

The global cognition z-scores, mean pause duration, number of pauses and percentage of speaking time were visualized on timepoint 1 and 2 (see Figure 2). The same was done for verbal memory, verbal fluency, working memory, motor speed, attention and speed of information processing and executive functions (see Figure A1).

Figure 2

Global cognition z-scores, mean pause duration, number of pauses and percentage of speaking time

Note. Figure showing boxplots of the global cognition z-scores (A), mean pause duration (B), number of pauses (C) and percentage of speaking time (D) over time. Timepoint 1 is

represented by the red boxplots and timepoint 2 by the blue boxplots. The red individual dots on the vertical red line in the boxplots of timepoint 1 represent the global cognition z-scores as well as the mean pause duration, number of pauses and percentage of speaking time per participant. These are connected to the blue dots on the vertical blue line in the boxplots of timepoint 2 by a light grey line which represent the global cognition z-scores as well as the mean pause duration, number of pauses and percentage of speaking time on timepoint 2 of the same participant. The black line connects the mean scores on timepoint 1 with the mean scores on timepoint 2.

(9)

The results of the multiple regression analysis, paired-samples t-test and assumptions are shown per subtest of the BACS, beginning with global cognition. Additionally, the results of the additional power analysis of the multiple regression models for global cognition, verbal memory, verbal fluency, working memory, motor speed, attention and speed of information processing and executive functions are shown in Appendix D.

Global cognition

All assumptions of the multiple regression analysis were met, as well as the assumption of the paired samples t-test (see Appendix B). Figure 3 shows the relationship between the global cognition z-scores and number of pauses, mean pause duration and percentage of speaking time.

The multiple regression analysis showed that the mean pause duration, number of pauses and percentage of speaking time taken together significantly predicted the global cognition z-scores on timepoint 1 (F(3,46)= 4.741, p= 0.006, R2_{= 0.236). Neither mean pause} duration (t= -1.220, p= 0.229) or number of pauses (t= -0.601, p= 0.551) were individually significant predictors. However, percentage of speaking time was (t= 3.173, p= 0.003). An overview of the results are shown in Table 1. The corresponding table with results of coefficients can be found in Table C1.

The regression equation of y= -6.628 -0.460* mean pause duration -0.008* number of pauses + 0.085*percentage of speaking time was used to calculate the predictions of the global cognition z-scores for timepoint 2. A paired samples t-test showed that the global cognition z-scores on timepoint 2 (M= -0.8584, SD= 1.380) did not significantly differ from the predicted global cognition z-scores on timepoint 2 (M= -1.34, SD= 0.735; t=1.415, p= 0.164).

(10)

Figure 3

The relationship between each independent variable mean pause duration, number of pauses and percentage of speaking time and the dependent variable global cognition

Note. The white colored dots are the predicted values, thee red colored dots are the true values on the upper side of predicted values and the blue colored dots are the values on the lower side of the predicted values. The light grey lined connecting the true values with the predicted values represent the residuals.

Table 1

Model summary of the multiple regression analysis for mean pause duration, number of pauses and percentage of speaking time predicting global cognition

F df1 df2 R2 Adjusted R2 p 4.741 3 46 0.236 0.186 0.006** Note. * p < .05. ** p < .007. Verbal memory

All assumptions of the multiple regression analysis were met, as well as the assumption of the paired samples t-test was met (see Appendix B).

The multiple regression analysis showed that the mean pause duration, number of pauses and percentage of speaking time taken together did not significantly predict the verbal memory z-scores on timepoint 1 (F(3,46)= 2.372, p= 0.083, R2=0.134). Neither mean pause

(11)

duration (t= -1.527, p=0.134), number of pauses (t= -0.613, p= 0.543) or percentage of speaking time (t= 1.978, p= 0.054) were individually significant predictors. An overview of the results are shown in Table 2. The corresponding results of coefficients can be found in Table C2.

The regression equation of y= -3.687 -0.671* mean pause duration -0.010* number of pauses + 0.061*percentage of speaking time was used to calculate the predictions of the verbal memory z-scores for timepoint 2. A paired samples t-test showed that the verbal memory z-scores on timepoint 2 (M= -0.219, SD= 1.350) did not significantly differ from the predicted verbal memory z-scores on timepoint 2 (M= -0.532, SD= 0.516; t=1.642, p= 0.107).

Table 2

Model summary of the multiple regression analysis for mean pause duration, number of pauses and percentage of speaking time predicting verbal memory

F df1 df2 R2 Adjusted R2

p

2.372 3 46 0.134 0.077 0.083

Verbal fluency

Three of the four assumptions of the multiple regression analysis were met. Additionally, the assumption of the paired samples t-test was met (see Appendix B).

The multiple regression analysis showed that the mean pause duration, number of pauses and percentage of speaking time taken together did not significantly predict the verbal fluency z-scores on timepoint 1 (F(3,46)= 0.466, p= 0.708, R2=0.029). Neither mean pause duration (t= -0.208, p= 0.836), number of pauses (t= -0.740, p= 0.463) or percentage of speaking time (t= 1.173, p= 0.247) were individually significant predictors. An overview of the results are shown in Table 3. The corresponding results of coefficients can be found in Table C3.

The regression equation of y= -2.982 -0.100* mean pause duration -0.013* number of pauses + 0.040*percentage of speaking time was used to calculate the predictions of the verbal fluency z-scores for timepoint 2. A paired samples t-test showed that the verbal fluency z-scores on timepoint 2 (M= -0.702, SD= 1.211) did not significantly differ from the

(12)

Table 3

Model summary of the multiple regression analysis for mean pause duration, number of pauses and percentage of speaking time predicting verbal fluency

p

0.466 3 46 0.029 -0.034 0.708

Working memory

Three of the four assumptions of the multiple regression analysis were met. Additionally, the assumption of the paired samples t-test was not met (see Appendix B).

The multiple regression analysis showed that the mean pause duration, number of pauses and percentage of speaking time taken together did not significantly predict the working memory z-scores on timepoint 1 (F(3,46)= 1.834, p= 0.154, R2_{= 0.107). Neither} mean pause duration (t= -0.878, p= 0.384), number of pauses (t= 0.537, p= 0.594) or

percentage of speaking time (t= 1.092, p= 0.281) were individually significant predictors. An overview of the results are shown in Table 4. The corresponding results of coefficients can be found in Table C4.

The regression equation of y= -3.587 -0.387* mean pause duration + 0.009* number of pauses + 0.034*percentage of speaking time was used to calculate the predictions of the working memory z-scores for timepoint 2. A Wilcoxon Matched-Pairs Test showed that the working memory z-scores on timepoint 2 (M= -0.499, SD= 1.196) did not significantly differ from the predicted working memory z-scores on timepoint 2 (M= -0.627, SD= 0.562; V=817, p= 0.084).

Table 4

Model summary of the multiple regression analysis for mean pause duration, number of pauses and percentage of speaking time predicting working memory

p

(13)

Motor speed

Three of the four assumptions of the multiple regression analysis were met.

Additionally, the assumption of the paired samples t-test was met (see Appendix B). Figure 4 shows the relationship between the motor speed z-scores and number of pauses, mean pause duration and percentage of speaking time.

The multiple regression analysis showed that the mean pause duration, number of pauses and percentage of speaking time taken together did not significantly predict the motor speed z-scores on timepoint 1 after Bonferroni correction (F(3,46)= 3.002, p= 0.040,

R2=0.164). Neither mean pause duration (t= -0.248, p= 0.805) or number of pauses (t= -0.097, p= 0.923) were individually significant predictors. However, percentage of speaking time was (t= 2.518, p= 0.015). An overview of the results are shown in Table 5. The corresponding results of coefficients can be found in Table C5.

The regression equation of y= -7.364 -0.121* mean pause duration -0.002* number of pauses + 0.087*percentage of speaking time was used to calculate the predictions of the motor speed z-scores for timepoint 2. A paired samples t-test showed that the motor speed z- scores on timepoint 2 (M= -0.606, SD= 1.452) did not significantly differ from the predicted motor speed z-scores on timepoint 2 (M= -0.779, SD= 0.834; t=0.784, p= 0.437).

(14)

Figure 4

Plot of the relationship between each independent variable mean pause duration, number of pauses and percentage of speaking time and the dependent variable motor speed

Table 5

Model summary of the multiple regression analysis for mean pause duration, number of pauses and percentage of speaking time predicting motor speed

p

3.002 3 46 0.164 0.110 0.040* Note. * p < .05. ** p < .007.

Attention and speed of information processing

The assumptions of the multiple regression analysis were met, as well as the assumption of the paired samples t-test (see Appendix B).

The multiple regression analysis showed that the mean pause duration, number of pauses and percentage of speaking time taken together did not significantly predict the attention and speed of information processing z-scores on timepoint 1 (F(3,46)= 0.733, p=

(15)

0.538, R2= 0.046). Neither mean pause duration (t= 1.099, p= 0.278), number of pauses (t= -0.249, p= 0.804) or percentage of speaking time (t= 0.993, p= 0.444) were individually significant predictors. An overview of the results are shown in Table 6. The corresponding results of coefficients can be found in Table C6.

The regression equation of y= -1.805 -0.348* mean pause duration -0.003* number of pauses + 0.017*percentage of speaking time was used to calculate the predictions of the attention and speed of information processing z-scores for timepoint 2. A paired samples t-test showed that the attention and speed of information processing zscores on timepoint 2 (M= -0.826, SD= 1.039) significantly differed from the predicted attention and speed of information processing z-scores on timepoint 2 (M= -1.148, SD= 0.182; t=2.257, p= 0.0289).

Table 6

Model summary of the multiple regression analysis for mean pause duration, number of pauses and percentage of speaking time predicting attention and speed of information processing F df1 df2 R2 _Adjusted R2 p 0.733 3 46 0.046 -0.017 0.537 Executive functions

Three of the four assumptions of the multiple regression analysis were met. Additionally, the assumption of the paired samples t-test was not met (see Appendix B).

The multiple regression analysis showed that the mean pause duration, number of pauses and percentage of speaking time taken together did not significantly predict the executive functions z-scores on timepoint 1 (F(3,46)= 1.460, p= 0.328, R2= 0.087). Neither mean pause duration (t= -0.273, p= 0.786), number of pauses (t= -0.574, p= 0.5691) or percentage of speaking time (t= 1.990, p= 0.053) were individually significant predictors. An overview of the results are shown in Table 7. The corresponding results of coefficients can be found in Table C7.

The regression equation of y= -4.249 -0.126* mean pause duration -0.010* number of pauses + 0.065* percentage of speaking time was used to calculate the predictions of the executive functions z-scores for timepoint 2. A Wilcoxon Matched-Pairs Test showed that the executive functions z-scores on timepoint 2 (M= -0.184, SD= 1.670) did not significantly differ from the predicted executive functions z-scores on timepoint 2 (M= -0.181, SD= 0.760; V=876, p= 0.157).

(16)

Table 7

Model summary of the multiple regression analysis for mean pause duration, number of pauses and percentage of speaking time predicting executive functions

p

1.460 3 46 0.087 0.027 0.328

Discussion

Here the longitudinal relationship between cognition and phonological aspects of speech, i.e. pause duration, number of pauses and percentage of speaking time, within FEP patients is examined. The results show that a predictive model could be made based on phonological measures for global cognition. Within this model percentage of speaking time was the only significant predictor. The same result would apply to motor speed, but the model lost its significance after Bonferroni correction. From these results can be concluded that percentage of speaking time is of great value when predicting overall cognition, but loses its importance when trying to predict specific facets of cognition. These findings correspond with a small part of the expectations. However, this study for the first time shows a longitudinal examination of the relation between phonological speech and cognition over time, assessing stability of phonological speech and cognition separately as well as the predictive value of phonological speech for cognition. Next to this, the study contributes to the following insights.

First, the finding that global cognition was predicted by percentage of speaking time implies that language and cognition are not independent processes ruling out this hypothesis described in Tillas (2015). Additionally, this study found that phonological speech was not predictive for verbal fluency, a cognitive facet related to categorization (Keefe et al. (2014). This indicates that phonological speech does not appear to modulate cognition and is

interesting information regarding the label-feedback hypothesis (Lupyan, 2012). This

hypothesis states that language modulates thinking processes and categorization, but does not specify what part of language modulates cognition. Here results indicate that another part of language than phonological speech could have this modulating role.

In addition, a possible explanation for the non-significant results regarding verbal fluency and mean pause duration, number of pauses and percentage of speaking time could be that verbal fluency problems are related to impairments in semantic processes opposed as to

(17)

speech estimates, a finding described by Allen et al. (1993). Furthermore, Bowie et al. (2004) implicated that impairments in other cognitive functions could induce impairments in verbal fluency. This could mean that cognitive measures influence each other and impairments in certain cognitive facets might predict outcomes on other cognitive measures.

Furthermore, the finding that phonological measures did not predict attention and speed of information processing is not in line with previous findings of Cohen et al. (2013). This could be because of lacking construct validity of the symbol coding test, here used as a measure for attention and speed of information processing (Keefe et al., 2004). This test possibly measures more cognitive facets opposed to specifically testing attention and speed of information processing (Andersen et al., 2013).

Second, the finding of longitudinal stability of mean pause duration and the

longitudinal instability of number of pauses and percentage of speaking time contribute to the current knowledge of the variation of phonological speech over time in FEP patients. Here, the instability of the number of pauses and percentage of speaking time over timepoint 1 and 2 cannot be noticed by comparing the means. However, substantial individual differences appear to be present when looking at the number of pauses and percentage of speaking time of participants. More so than in mean pause duration. This finding is not in line with former research (Cohen et al., 2012). However, Cohen et al. (2012) used epochs of one week, whereas this study uses a time lag of 6 months. This means this research contributes to new insights of phonological speech over a relatively long period of time in FEP patients.

Last, this study gives insights of cognitive functioning in FEP patients over time. The results of the descriptive statistics show that every cognitive facet of the BACS shows a light ascending line when comparing scores on timepoint 2 with timepoint 1, except for executive functions(see Figure A1). It appears that participants slightly improved cognitively on average. However, there are substantial individual differences and a balance of ascending and descending scores explains the average score. These results are contrary to the former

research showing cognitive impairments being stable over time (Bozikas & Andreou, 2011; Kurtz, 2005). This instability of cognitive functioning could be caused by differences in medication and dosage. Participants in this group differed greatly in type of medication used, but ever more so in dosage. While some were continuing medication others were reducing their dosages. The differences in ascending and descending scores could be explained by differences in continuation and dose-reduction, which could be caused by the positive impact antipsychotic medication has on some cognitive domains and negative impact on other cognitive domains (Hill et al., 2004). Due to the dataset being single-blind at the time of

(18)

analysis, we could not correct for medication types and dosage. However, this should be a focus of further research.

Next to medication, no corrections for diagnosis and comorbidity were carried out. Because subjects with FEP could later develop a specific diagnosis, this could result in differences in severity in impairments of cognitive functioning and stability and decline of these impairments. Zanelli et al. (2019) showed that a decrease in cognitive function is present in FEP patients but varies in severity and facets of cognition for different psychotic disorders. This suggests that different processes related to specific disorders are impaired and could also possibly be an explanation for the differences in ascending and descending scores.

Another limitation of this study is the use of two timepoints 6 months apart (Hill et al., 2004). Even though two timepoints should be enough to get an idea of the relationship of cognition scores and phonological measures over time, with more timepoints multiple timepoints could be used to make a more accurate model instead of basing a model on 1 timepoint. This results in a better representation of changes in cognition and phonology over a longer period of time, which could potentially lead to more detailed fine-grained results.

Additionally, another methodological limitation is that assumptions of multiple regression were violated for some of the facets of cognition. Namely, the residuals were not normally distributed for verbal fluency, motor speed and executive functions. Transforming the data did not resolve this problem. Correspondingly, the assumption of linearity was violated for verbal fluency and executive functions. Furthermore, the assumption of

homoscedasticity is violated for working memory. Even though this is a minor violation, it could have had an effect on the data in the form of false positives (Flatt & Jacobs, 2019). However, this seems unlikely since the results of working memory were non-significant. It is unclear to what extent these violations truly had an influence, but it must be noted. Because there is no non-parametrical test for multiple regression, the multiple regression analysis was used even though assumptions were violated.

A possible explanation for the non-significant results is insufficient power of the models (see Appendix D). The global cognition model is the only model that appeared to have sufficient power. This means that the chance of rejecting the alternative hypothesis while it is in fact true (type II error) is relatively low for this model. This did not apply for the models for verbal memory, verbal fluency, working memory, motor speed, attention and speed of information processing and executive functioning. These models all have insufficient power meaning a relatively high chance on a type II error (Wassertheil-Smoller & Kim, 2010).

(19)

Further research should focus on obtaining sufficient power, for example by using a larger sample size.

To conclude, here the predictive function of percentage of speaking time on overall cognition is shown. This is an important insight in gathering long-term effects and interplay between phonological speech and global cognitive functioning. As noted everyday

functioning is impacted by impairments in cognition; an assessment of global cognition from speech, a natural, low-effort task, would be valuable for clinical practice even if not all

cognitive domains could be individually assessed. In addition, future research should still take the longitudinal aspect into account, but should also focus on a multiple points in time, as well as the modulating effect of type and dosage of medication on cognitive function and using tests with sufficient construct validity. With further research in this field, more knowledge can be acquired for the clinical use of phonological speech as a marker for psychosis as well as cognitive functioning.

(20)

References

Allen, H.A., Liddle, P.F., & Firth, C.D. (1993). Negative features, retrieval processes and verbal fluency in schizophrenia. British Journal of Psychiatry, 163(6), 769-775.

https://doi.org/10.1192/bjp.163.6.769

Andersen, R., Fagerlund, B., Rasmussen, H., Ebdrup, B.H., Aggernaes, B., Gade, A., … Glenthoj, B. (2013). The influence of impaired processing speed on cognition in first-episode antipsychotic-naïve patients. European Psychiatry, 28, 332-339.

https://doi.org/10.1016/j.eurpsy.2012.06.003

Anderson, K.K., Fuhrer, R., Abrahamowicz, M., & Malla, A.K. (2012). The incidence of first-episode schizophrenia-spectrum psychosis in adolescence and young adults in

Montreal: an estimate from an administrative claims database. The Canadian journal of psychiatry, 57(10), 626-633. https://doi.org/10.1177/070674371205701007

Begemann, M.J.H., Thompson, I.A., Veling, W., Gangadin, S.S., Geraets, C.N.M., van t’ Hag, E., … Sommer, I.E. (2020). To continue or not to continue? Antipsychotic medication maintenance versus dose-reduction/discontinuation in first episode psychosis: HAMLETT, a pragmatic multicenter single-blind randomized controlled trial. Trials, 21(1), 147.

https://doi.org/10.1186/s13063-019-3822-5

Bland, J. M., & Altman, D. G. (1995). Multiple significance tests: the Bonferroni method. British Medical Journal (Clinical research ed.), 310(6973), 170.

https://doi.org/10.1136/bmj.310.6973.170

Bowie, C.R., Harvey, P.D., Moriarty, P.J., Parrella, M., White, L., & Davis, K.L. (2004). A comprehensive analysis of verbal fluency deficit in geriatric schizophrenia. Archives of Clinical Neuropsychology, 19, 289-303.

https://doi.org/10.1016/S0887-6177(03)00041-6

Bozikas, V.P., & Andreou, C. (2011). Longitudinal studies of cognition in first episode psychosis: A systematic review of the literature. Australian & New Zealand journal of psychiatry, 45(2), 93-108. https://doi.org/10.3109/00048674.2010.541418

(21)

De Boer, J.N., Brederoo, S.G., Voppel, A.E., & Sommer, I.E.C. (2020a). Anomalies in language as a biomarker for schizophrenia. Current Opinion in Psychiatry, 33(3), 212-218.

https://doi.org/10.1097/YCO.0000000000000595

De Boer, J.N., Hoogdalem, M., Mandl, R.C.W., Brummelman, J., Voppel, A.E., Begemann, M.J.H., … Sommer, I.E.C. (2020b). Language in schizophrenia: relation with diagnosis, symptomatology and white matter tracts. NPJ Schizophrenia, 6(1), 10.

https://doi.org/10.1038/s41537-020-0099-3

De Jong, N.H., & Wempe, T. (2009). Praat script to detect syllable nuclei and measure speech rate automatically. Behavior Research Methods, 41(2), 385-390.

https://doi.org/10.3758/BRM.41.2.385

Cohen, A.S., & Elvevåg, B. (2014). Automated computerized analysis of speech in psychiatric disorders. Current Opinion in Psychiatry, 27(3), 203-209.

https://doi.org/10.1097/YCO.0000000000000056

Cohen, A.S., Kim, Y., & Najolia, G.M. (2013). Psychiatric symptom versus cognitive correlates of diminished expressivity in schizophrenia and mood disorders. Schizophrenia research, 146, 249-253. https://doi.org/10.1016/j.schres.2013.02.002

Cohen, A.S., Najolia, G.M., Kim, Y., & Dinzeo, T.J. (2012). On the boundaries of blunt affect/alogia across severe mental illness: Implications for Research Domain Criteria. Schizophrenia Research, 140, 41-45. https://doi.org/10.1016/j.schres.2012.07.001

Corrión, R.E., Goldberg, T.E., McLaughlin, D., Auther, A.M., Correll, C.U., & Cornblatt, B.A. (2011). Impact of Neurocognition on Social and Role Functioning in Individuals at Clinical High Rusk for Psychosis. American Journal of Psychiatry, 168(8), 806-813. https://doi.org/10.1176/appi.ajp.2011.10081209

Diagnostic and Statistical Manual of Mental Disorders: DSM-5. Fifth edition. Arlington, VA: American Psychiatric Association, 2013. Print.

(22)

Flatt, C., & Jacobs, R.L. (2019). Principle assumptions of regression analysis: testing, techniques, and statistical reporting of imperfect data sets. Advances in developing human resources, 21(4), 484-502. https://doi.org/10.1177/1523422319869915

Hill, S.K., Schuepbach, D., Herbener, E.S., Keshavan, M.S., & Sweeney, J.A. (2004). Pretreatment and longitudinal studies of neuropsychological deficits in antipsychotic-naïve patients with schizophrenia. Schizophrenia research, 68, 49-63.

https://doi.org/10.1016/S0920-9964(03)00213-5

Keefe, R.S.E., Goldberg, T.E., Harvey, P.D., Gold, J.M., Poe, M.P., & Coughenour, L. (2004). The brief assessment of cognition in schizophrenia: reliability, sensitivity, and

comparison with a standard neurocognitive battery. Schizophrenia research, 68, 283-297.

https://doi.org/10.1016/j.schres.2003.09.011

Kilciksiz, C.M., Keefe, R., Benoit, J., Öngür, D., & Torous, J. (2020). Verbal memory measurement towards digital perspectives in first-episode psychosis: A review. Schizophrenia Research: Cognition, 21, 100177. https://doi.org/10.1016/j.scog.2020.100177

Kurtz, M.M. (2005). Neurocognitive impairments across the lifespan in schizophrenia: an update. Schizophrenia research, 74, 15-26. https://doi.org/10.1016/j.schres.2004.07.005

Lupyan, G. (2012). Linguistically modulated perception and cognition: the label-feedback hypothesis. Frontiers in psychology, 3, 54.

https://doi.org/10.3389/fpsyg.2012.00054

Nagumo, R., Zhang, Y., Ogawa, Y., Hosokawa, M., Abe, K., Ukeda, T., ... Shimada, H. (2020). Automatic Detection of Cognitive Impairments through Acoustic Analysis of Speech. Current Alzheimer research, 17(1), 60–68.

https://doi.org/10.2174/1567205017666200213094513

Parola, A., Simonsen, A., Bliksted, B., & Fusaroli, R. (2020). Voice patterns in

schizophrenia: a systematic review and Bayesian meta-analysis. Schizophrenia research, 216, 24-40. https://doi.org/10.1016/j.schres.2019.11.031

(23)

Riley, E.M., McGovern, D., Mockler, D., Doku, V.C.K., ÓCeallaigh, S., Fannon, D.G., … Sharma, T. (2000). Neuropsychological functioning in first-episode psychosis –- evidence of specific deficits. Schizophrenia research, 43, 47-55.

https://doi.org/10.1016/s0920-9964(99)00177-2

Rodrigues, R., & Anderson, K.K. (2017). The traumatic experience of first-episode psychosis: A systematic review and meta-analysis. Schizophrenia Research, 189, 27-36.

https://doi.org/10.1016/j.schres.2017.01.045

Schuepbach, D., Keshavan, M.S., Kmiec, J.A., & Sweeney, J.A. (2002). Negative symptom resolution and improvements in specific cognitive deficits after acute treatment in first-episode schizophrenia. Schizophrenia Research, 53(3), 249-261.

https://doi.org/10.1016/s0920-9964(01)00195-5

Simms, L.E., Engebretson, M.J., Pilipenko, V., Reeves, G.D., & Clilverd, M. (2016). Empirical predictive models of daily relativistic electron flux at geostationary orbit: Multiple regression analysis. Journal of Geophysical research: space physics, 121, 3181-3197.

https://doi.org/10.1002/2016JA022414

Simon, G.E., Coleman, K.J., Yarborough, B.J., Operskalski, B., Stewart, C., Hunkeler, E., … Beck, A. (2017). Incidence and presentation of first-episode psychosis in a population-based sample. Psychiatric Services, 68(5), 456-461.

https://doi.org/10.1176/appi.ps.201600257

Tahir, Y., Yang, Z., Chakraborty, D., Thalmann, N., Thalmann, D., Maniam, Y., … Dauwels, J. (2019). Non-verbal speech cues as objective measures for negative symptoms in patients with schizophrenia. PloS one, 14(4), e0214314.

https://doi.org/10.1371/journal.pone.0214314

Tillas, A. (2015). Language as grist to the mill of cognition. Cognitive processing, 16(3). 219-243. https://doi.org/10.1007/s10339-015-0656-2

(24)

Wassertheil-Sommer, S., & Kim, M.Y. (2010). Statistical analysis of clinical trials. Seminars in Nuclear Medicine, 40(5), 357-363.

https://doi.org/10.1053/j.semnuclmed.2010.04.001

Zanelli, J., Mollon, J., Sandin, S., Morgan, C., Dazzan, P., Pilecka, I., … Reichenberg, A. (2019). Cognitive change in schizophrenia and other psychoses in the decade following the first episode. American Journal of Psychiatry, 176, 811-819.

(25)

Appendix A – Results: descriptive statistics Figure A1

Verbal memory, verbal fluency, working memory, motor speed, attention and speed of information processing and executive functions z-scores per participant over time

Note. Plots of the verbal memory (A), verbal fluency (B), working memory (C), motor speed (D), attention and speed of information processing (E) and executive functions (F) z-scores per participant over time. Timepoint 1 is represented by the red boxplots and timepoint 2 by the blue boxplots. The red individual dots on the vertical red line in the boxplots of timepoint 1 represent the z-scores of verbal memory, verbal fluency, working memory, motor speed, attention and speed of processing and executive functioning. These are connected to the blue dots on the vertical blue line in the boxplots of timepoint 2 by a light grey line which

(26)

represent the z-scores of verbal memory, verbal fluency, working memory, motor speed, attention and speed of processing and executive functioning on timepoint 2 of the same participant. The black line connects the mean scores on timepoint 1 with the mean scores on timepoint 2.

(27)

Appendix B – Results: assumptions multiple regression and paired samples t-test

Global cognition

The results of the Durbin-Watson test revealed that there was no autocorrelation present (D-W= 2.144, p= 0.620). Additionally, the Breusch-Pagan test showed that the assumption of homoscedasticity was met (BP(3)= 2.424, p= 0.489). Furthermore, the Shapiro-Wilk test revealed that the residuals were normally distributed (W = 0.985, p= 0.769). Additionally, Figure 3 shows the linearity between the explanatory and response variables. Finally, the results of the Shapiro-Wilk test showed normally distributed difference scores (W= 0.958, p=0.074).

Verbal memory

The results of the Durbin-Watson test revealed that there was no autocorrelation present (D-W= 1.989, p= 0.962). Additionally, the Breusch-Pagan test showed that the assumption of homoscedasticity was met (BP(3)= 1.599, p= 0.660). Furthermore, the

Shapiro-Wilk test showed that the residuals were normally distributed (W = 0.981, p= 0.590). Additionally, Figure B1 shows the linearity between the explanatory and response variables. Finally, the results of the Shapiro-Wilk test showed that the assumption of normally

(28)

Figure B1

The relationship between each independent variable mean pause duration, number of pauses and percentage of speaking time and the dependent variable verbal memory

Verbal fluency

The results of the Durbin-Watson test revealed that there was no autocorrelation present (D-W= 1.663, p=0 .222). Additionally, the Breusch-Pagan test showed that the assumption of homoscedasticity was met (BP(3)= 2.351, p= 0.503). Furthermore, the Shapiro-Wilk test showed that the residuals were not normally distributed (W = 0.919, p=0.002 ). Additionally, Figure B2 shows the assumption of linearity between the explanatory and response variables is violated for number of pauses and pause duration, but was met for percentage of speaking time. Finally, the results of the Shapiro-Wilk test showed normally distributed difference scores (W= 0.976, p=0.411).

(29)

Figure B2

The relationship between each independent variable mean pause duration, number of pauses and percentage of speaking time and the dependent variable verbal fluency

Working memory

The results of the Durbin-Watson test revealed that there was no autocorrelation present (D-W= 1.923, p= 0.732). Additionally, the Breusch-Pagan test showed that the assumption of homoscedasticity was violated (BP(3)=9.157, p= 0.027). Furthermore, the Shapiro-Wilk test showed that the residuals were normally distributed (W = 0.971, p= 0.255). Additionally, Figure B3 shows the linearity between the explanatory and response variables. Finally, the results of the Shapiro-Wilk test showed that the assumption of normally

(30)

Figure B3

The relationship between each independent variable mean pause duration, number of pauses and percentage of speaking time and the dependent variable working memory

Motor speed

The results of the Durbin-Watson test revealed that there was no autocorrelation present (D-W= 2.234, p= 0.406). Additionally, the Breusch-Pagan test showed that the assumption of homoscedasticity was met (BP(3)= 3.865, p= 0.276). Furthermore, the Shapiro-Wilk test showed that the residuals were not normally distributed (W = 0.946, p= 0.024). Additionally, Figure 4 shows the linearity between the explanatory and response variables. Finally, the results of the Shapiro-Wilk test showed that the assumption of normally distributed difference scores was met (W= 0.966, p=0.158).

Attention and speed of information processing

The results of the Durbin-Watson test revealed that there was no autocorrelation present (D-W= 1.832, p= 0.548). Additionally, the Breusch-Pagan test showed that the assumption of homoscedasticity was met (BP(3)= 4.292, p= 0.232). Furthermore, the

(31)

Shapiro-Wilk test showed that the residuals were normally distributed (W = 0.995, p= 0.999).

Additionally, Figure B4 shows the linearity between the explanatory and response variables. Finally, the results of the Shapiro-Wilk test showed normally distributed difference scores (W= 0.967, p=0.168).

Figure B4

The relationship between each independent variable mean pause duration, number of pauses and percentage of speaking time and the dependent variable attention and speed of

information processing

Executive functions

The results of the Durbin-Watson test revealed that there was no autocorrelation present (D-W= 1.787, p= 0.410). Additionally, the Breusch-Pagan test showed

homoscedasticity (BP(3)= 0.435, p= 0.933). Furthermore, the Shapiro-Wilk test showed that the residuals were not normally distributed (W = 0.776, p= 2.513e-07). Additionally, Figure B5 shows the assumption of linearity between the explanatory and response variables was met

(32)

for percentage of speaking time, but was violated for number of pauses and pause duration. Finally, the results of the Shapiro-Wilk test showed that the assumption of normally

distributed difference scores was violated (W= 0.819, p=1.429e-06).

Figure B5

The relationship between each independent variable mean pause duration, number of pauses and percentage of speaking time and the dependent variable executive functions

(33)

Appendix C Results: Multiple regression analysis

Table C1

Coefficients of the multiple regression analysis of mean pause duration, number of pauses and percentage of speaking time predicting global cognition

Variable B SE t pr(>|t|) Intercept -6.628 1.930 -3.434 0.001** Mean pause duration -0.460 0.377 -1.220 0.229 Number of pauses -0.008 0.014 -0.601 0.551 Percentage of speaking time 0.085 0.027 3.173 0.003**

Note. SE = unstandardized standard error. * p < .05. ** p < .007. Table C2

Coefficients of the multiple regression analysis for mean pause duration, number of pauses and percentage of speaking time predicting verbal memory

Variable B SE t pr(>|t|) Intercept -3.687 2.246 -1.642 0.107 Mean pause duration -0.671 0.439 -1.527 0.134 Number of pauses -0.010 0.016 -0.613 0.543 Percentage of speaking time 0.061 0.031 1.978 0.054

Note. SE = unstandardized standard error. Table C3

Coefficients of the multiple regression analysis for mean pause duration, number of pauses and percentage of speaking time predicting verbal fluency

(34)

Table C4

Coefficients of the multiple regression analysis for mean pause duration, number of pauses and percentage of speaking time predicting working memory

Variable B SE t pr(>|t|) Intercept -3.587 2.251 -1.594 0.118 Mean pause duration -0.387 0.440 -0.878 0.384 Number of pauses 0.009 0.016 0.537 0.594 Percentage of speaking time 0.034 0.031 1.092 0.281

Note. SE = unstandardized standard error. Table C5

Coefficients of the multiple regression analysis for mean pause duration, number of pauses and percentage of speaking time predicting motor speed

Variable B SE t pr(>|t|) Intercept -7.364 2.492 -2.955 0.005** Mean pause duration -0.121 0.487 -0.248 0.805 Number of pauses -0.002 0.018 -0.097 0.923 Percentage of speaking time 0.087 0.034 2.518 0.015*

Note. SE = unstandardized standard error. * p < .05. ** p < .007. Table C6

Coefficients of the multiple regression analysis for mean pause duration, number of pauses and percentage of speaking time predicting attention and speed of information processing

(35)

Table C7

Coefficients of the multiple regression analysis for mean pause duration, number of pauses and percentage of speaking time predicting executive functions

(36)

Appendix D Results: power analysis Power analysis per multiple regression model

Model f2 Power (1-β) Global cognition 0.309 0.919 Verbal memory 0.155 0.627 Verbal fluency 0.030 0.152 Working memory 0.120 0.507 Motor speed 0.196 0.741

Attention and speed of information processing

0.048 0.220

Executive functions 0.095 0.413

Note. f2_{= effect size. Numerator degrees of freedom = 3, denominator degrees of freedom =} 49 and α= 0.05.