2.1. Ethical statement
The art therapy study was approved by the management of the privately run Marienheim, a social therapy facility for the mentally ill. The written project permission was received on 30 March 2016. The written, informed consent of all seven study participants was also available before the start of the study. The legal guardians of these seven patients were also informed about the study participation of their wards and made the consent forms legally binding by signing them. After completion of these formal requirements, the author drew up the final list of study participants, which were signed by the the two treating psychiatrists. They endorsed the art therapy envisaged in the study and described it as a 'medically indicated therapeutic measure' integrated in the overall treatment plan. Furthermore, the Ethics Committee of the University of Augsburg confirmed the ethical harmlessness of the art therapy research project in its statement of 11 May 2017. The study was registered at www.isrctn.com (identifier: ISRCTN12365070) https://doi.org/10.1186/ISRCTN12365070.
2.2 Sampling
Seven patients (n = 7) aged between 46 and 62 years were included in the explorative single case study (SSR in an isolated B design) over a period of six months (05/2016 to 10/2016). They were inpatients at the Marienheim facility in Peiting and suffered from long-term schizophrenia. The patients of the Marienheim are people of all genders who suffer from different, partly combined mental illnesses that are classified in the ICD code. Their stay is under a judicial accommodation order according to §1906 BGB and they have legal care. The construction of the sample from the total population of the home for this art therapy study resulted primarily from the specific objective. Secondarily, the complete lack of third-party funding for this research project also played a role, albeit a subordinate one.
The Marienheim was occupied by 69 patients in February 2016 (according to the resident list of 12 February 2016). The selection of the sample took place under inclusion and exclusion criteria in cooperation with both attending consultant physicians. One of them was a specialist in neurology and psychiatry from an independent practice and the other was a specialist from a psychiatric outpatient clinic (PIA); they were both commissioned by the Marienheim facility.
Inclusion criteria: Diagnosis of Paranoid Schizophrenia (F20.0) with chronic course of at least ten years or/and Schizophrenic Residual (F20.5), stable medication of psychotic symptoms for at least three months, preference towards artistic therapies (art therapy), and the presence of written informed consent. Exclusion criteria: Primary addictive disorder, acute accessory symptoms, suicidality, intelligence reduction, brain organic disorder, autistic disorder, personality disorder, visual impairment, language barrier, hemiparesis including speech disorder, moving out soon, or participation in another creative therapy (in group or/and individual therapy).
Based on the above 'inclusion and exclusion criteria', 15 patients were selected as suitable for the sample with the treating psychiatrists on 09 April 2016. Thereupon, the author informed the 15 patients about the details of the study in individual interviews, offering them the opportunity to participate in the study. The obligatory 'information and consent form' was also discussed in detail and handed out if required. Of these 15 patients, eight agreed to participate in the study. After a reflection period of several days, seven patients signed the 'information and consent form'. One study participant (P7) asked to participate in the study group but did not allow audio recording of the interviews.
The list of study participants signed by the psychiatrists–not anonymized in the original–is shown in Table 1 below with year of birth, gender, diagnosis(es), and date of entry into the Marienheim. In the further course of the study, the study patients were named exclusively as P1, P2, P3, P4, P5, P6, and P8.
Table 1
List of study participants (anonymized).
Patient
|
Birth year
|
Gender
|
Diagnosis
|
Move-in
|
P1
|
1967
|
M
|
paranoid schizophrenia (F 20.0)
mental behavioral changes due to alcohol (F10.3)
|
2015
|
P2
|
1958
|
M
|
paranoid schizophrenia (F 20.0)
schizoaffective disorder (F 25.1)
|
2015
|
P3
|
1970
|
W
|
paranoid-hallucinatory schizophrenia (F 20.0)
|
2014
|
P4
|
1963
|
M
|
schizophrenic residual (F20.5)
|
2002
|
P5
|
1956
|
W
|
paranoid-hallucinatory schizophrenia (F20.0)
|
2014
|
P6
|
1954
|
M
|
paranoid schizophrenia (F 20.0)
|
2015
|
P8
|
1965
|
M
|
paranoid schizophrenia (F 20.0)
schizophrenic residual (F 20.5)
|
2015
|
M: male; W: female.
2.3. Procedure
The course of the individual study phases with specifics of study interruption/dropout with reasons are shown in the CENT-2015 Flow Diagram in Figure 1.
2.5. Art therapy intervention
2.5.1. Picture creation
Group art therapy with image creation took place once a week, always on Fridays from 10.00 am to 11.30 am over a period of six months (05/2016 to 10/2016) and up to 20 possible sessions. The group was led exclusively by the author for the entire study and took place in the facility´s creative room. The art therapy appointments were fixed in the therapy plan as well as in the calendar of the residential group. All the materials needed for the artistic work were available for each task or topic; in detail, these were acrylic and watercolor paints, oil, and pastel crayons, colored, felt, lead and neon pencils, brushes of various thicknesses, palettes and water glasses and various adhesive materials (e.g., glue and tape). Pursuant to the diagnoses of the study participants, "solid materials" such as wax and colored pencils and collage materials were offered at the beginning [37]. Painting and drawing paper, cardboard, cardboard in various colors and thicknesses, primed canvases or painting cardboard and watercolor paper served as picture supports. Boxes with photographic material were available for the collages. In the first art therapy session, in addition to the timetable and schedule, the most important rules were explained and repeated as needed. These rules included that art therapy is not about creating works of art, that there is no 'right' or 'wrong', that the tasks and topics are for orientation and not binding, and that the materials offered can be used freely–according to individual expression needs. The group sessions of the study followed a fixed structure with regard to both the timing and the content.
The author moderated through the three phases, consisting of the opening round, picture design and closing round, according to Schrode [38] and von Spreti [37]. In the opening round, which lasted about 15 minutes, each participant was given the opportunity to talk about how they were feeling now. Especially at the beginning of the art therapy study–when the participants did not know each other that well–it was the author's task to create a trusting atmosphere within the art therapy group [37]. This was followed by the author setting the tasks and themes for the day to create the pictures. The tasks and themes were a central element in the art therapy study approach. The empirical findings of the art therapist Landgarten [40] formed the basis for the tasks and themes. In a group working on a long-term basis and thus 'experienced', the identification of themes also took place in consideration of current events, moods or wishes expressed by individual or several patients during the project [39]. The detailed description of the tasks and themes for the creation of the pictures is described in detail under A.1 (Appendix).
The pictures created by study participants in art therapy were placed in a personalized collection folder after drying. At the end of the study (beginning of November 2016), the images were also digitally photographed, anonymized, and archived on a hard drive. After the end of the study, study participants received their original pictures back.
2.5.2. Picture reflection
The art Therapist-Guided Picture Reflection (TGPR) took the form of a guided question interview. Although the guided interview belongs to the "semi-structured forms of data collection for obtaining verbal data" [41], it was used as a structured form of data collection in this research. This was because the questions in the guide enable the necessary structure in the sense of a standardized course of the survey. At the same time, this form of interview opened enough space for 'free speech', which was required here to obtain sufficiently large speech and voice samples. Furthermore, the standardization of the interview procedure established a certain comparability of the interviews during the study, specifically for intra-individual comparisons of each participant's speech and voice over the course of six months and for inter-individual comparisons across the study group. The structure of the TGPR (see Table A.2) was based on four phases according to Misoch [41]: the "information phase", the "warm-up and introduction phase", the "main phase", and the "fade-out and conclusion phase" [41].
It was necessary to create a catalogue of questions that, despite stereotypes, made the guide an event in which it should always be possible from image to create in the same way the essential emotional arc of tension, the actual therapeutic moment of the interview. When searching for suitable questions and their sequence, the disease-related limitations of the study participants had to be a primary consideration.
For the construction of such a catalogue of questions, it was possible to draw from a wealth of scientific foundations and our own practical experience. Method is a prerequisite for goal-oriented drawing, and here the choice fell on the structured CCSS method (Collecting, Checking, Sorting, Subsuming) by Helfferich [42]. This process was accompanied by the doctoral colloquium of the Research Association of Artistic Therapies (RAT) and by the upper seminar of art education research at the Chair of Art Education at the University of Augsburg in the period from September 2015 to May 2016. In a brainstorming session, all significant questions and topics for the art therapy guiding questions were 'collected'. Subsequently, which questions were suitable for the guideline with chronic schizophrenic patients, the setting, and the research question were checked. Subsequently, the questions were 'sorted' according to content as whether they were guiding questions or open-ended narrative prompts, maintenance questions, or specific follow-up questions. The content aspects and the prompt 'What is the question aiming at?' are shown in columns 2 and 5 of Table A.3. In a final step, the collected, reviewed, and sorted questions were 'subsumed' into a guideline of eight questions.
The thematic areas of the question catalogue were based on the art-therapeutic method of picture discussion from a phenomenological perspective according to Betensky [19] and the art-based approaches to picture viewing and discussion according to Stuhler-Bauer and Elbing [43], Bader [44], Dannecker [4] and Titze [45], and according to the specially elaborated conversational style for psychotherapy with schizophrenic patients according per Süllwold [10, 11]. In the interviews for data collection, the study participants talked about the images they had created. The conversation about the study patients' own pictures took place in the week after the pictures were created on Monday or Tuesday afternoon as individual therapy in the form of an interview with a maximum duration of 50 minutes. The author picked up the study participant from the agreed meeting point and went with them to the meeting room reserved for this purpose.
2.6. Research instruments
This art therapy study was directed at recording the vocal expression of emotion or emotionality and the speech and thinking style from the verbal and paraverbal speech signals of the study participants. To this end, the study used two computer-assisted research methods to analyze the digitized audio recordings, the computer-assisted quantitative text analysis Linguistic Inquiry and Word Count (LIWC2015) and the voice analysis software Vocal Emotion recognition by Appraisal Inference, the VocEmoApI technology (see Figure 2). Thus, a dual-automated, computerized analysis of the audio documents was achieved.
2.6.1. Range of target parameters of LIWC2015
The study interviews were conducted in German. To evaluate the recordings, the current software version of LIWC2015 according to Pennebaker et al. (46) was combined with a working version of the German LIWC2015 program by Meier et al. (46). The analysis took place in 2017/2018. The German DE-LIWC2015 dictionary consists of around 18000 words and word stems [47]. In the same way as in the English version of LIWC2015, the recognized words can be assigned to several word categories [47]. The quality of the German LIWC2015 was examined by Meier et al. [47] in two studies: first by comparing the German translation (DE-LIWC2015) with the English original (ENG-LIWC2015) and then with the previous German version of DE-LIWC2001. The findings in the studies showed a high level of equivalence between the two dictionaries [47].
The computer-assisted text analysis LIWC is a computer program that automatically examines written or transcribed verbal texts for features of the author in a formal, quantitative way. In doing so, the application focuses less on the content of human speech, but rather on individual words and word stems, which are counted, assigned to defined word categories and illustrated as a percentage in relation to the text length [48]. The recognized words can be assigned to several word categories–so-called upper and lower categories–at the same time and consequently be recorded and counted several times. The word cried can therefore be assigned to the categories sad (sadness), affect (all affect words as an upper category), the category negemo (negative emotion), verb (verbs), and focus past (past) [46].
For the recording and automatic evaluation of the verbal speech signals, a complete transcription of all individual interviews (full transcription) was necessary. The transcription rules were those from the LIWC2015 user manual according to Pennebaker [45] and the “German LIWC” according to Meier et al. [47].
The LIWC2015 text analysis consists of 90 output parameters. For the present work, however, only 61 of these parameters were included, since the remaining 29 may not be included in the statistics as additional share of an upper- or lower-level LIWC2015 category.
For the statistical analysis, it was first relevant how high the percentage of speech data is in an audio document. For this purpose, the LIWC output parameter Dictionary Words (Dic) was used, which determines the proportion of words that are recorded and evaluated by the LIWC2015 text analysis procedure. In addition, it was necessary to select the study-relevant parameter sets for the exploratory factor analysis. These included the three general descriptive variables with the words counted per interview (WC), the percentage of sentence length (WPS) and words longer than six characters (Sixltr). Of the output parameters of the categories of 'basic linguistic and grammatical parameters' (I) and 'psychological processes' (II), only those belonging to a subordinate category entered exploratory factor analysis. Therefore, of the 'basic linguistic and grammatical parameters' (I), all subcategories with 18 language features were included. This concerned the five linguistic features of personal pronouns (i, we, you, shehe and they) and impersonal pronouns (ipron) as well as the linguistic categories of article (article), prepositions (prep), auxiliary verbs (auxverb), adverbs (adverb), conjunction (conj) and negation (negate), verbs (verb), adjectives (adj), comparative words (compare), interrogative words (interrog), number words (number) and words that quantify (quant). Also, of the word categories for thematic content 'psychological processes' (II), only the subcategories with 30 parameters were included in the exploratory factor analysis. These included two parameters relating to affective (posemo, negemo), four relating to social (family, friend, female, and male), six relating to cognitive (insight, cause, discrep, tentat, certain, and differ), three relating to perception (see, hear, and feel), and four relating to biological processes (body, health, sexual and, ingest). In addition, five parameters describing drives (affiliation, achieve, power, reward, and risk), three describing temporal orientation (focuspast, focuspresent, and focusfuture), and three describing relativities (motion, space, and time) were analyzed. Finally, six parameters for 'personal concerns' (work, leisure, home, money, relig, and death) and four for 'formless language' (swear, assent, nonflu, and filler) were included. In summary, 61 parameters from the LIWC2015 text analysis were included in the exploratory factor analysis.
2.6.2. Range of target parameters of VocEmoApI
The voice analysis procedure used here is the further development of a software called sensAI (sensitive Audio Intelligence), a technology for the computer-aided identification of affective speaker states via human speech or voice. The following description of the software used is based on the working manual "sensAI WebAPI Documentation - Version 1.3 from 5.12.17" [49]. The sensAI technology analyses the input audio files and scales the automated evaluation of the output variables from speech or voice as well as emotional content. It recognizes the human voice with the help of "voice activity detection" (VAD) and can also distinguish it from loud background noises (e.g., street noise). The speech signals of the speaker are decoded about characteristic and personal features (personality, age, and gender), patient emotional speech states (emotional categories such as joy, anger, fear, etc.), the speech activity (duration, speech tempo, etc.), and the speech prosody (volume, pitch of voice, etc.).
The software, which works with 88 parameters, is based on the basic standard parameter set Geneva Minimalistic Acoustic Parameter Set (GeMAPS) for voice research and affective computing [50]. It was applied in the voice analysis of the present study under the name VocEmoApI technology with its category_v2_scores. This includes the intensity parameter, 52 emotion scores of the category_v2_scores, the three emotion dimensions pleasantness, urgency, and control, and the four prosodic features pitchAverage, pitchVariation, loudnessAverage and speakingSpeed, which will be explained in the following paragraph. The parameter intensity (line 4) is used to check the emotional intensity of the category_v2_scores. The range goes from 0.0 (neutral) to 1.0 (highly emotional). A score below 0.25 can be considered neutral language. Scores above this indicate emotionally colored speech [49].
The 52 emotion scores are found in the category_v2_scores (line 5). For the category_v2_scores, a reference range is given in which the scores move. This reference range was determined from voice samples of unspecified speakers. It is set between the values -1.0 and +1.0. Scores between -1.0 and 0 are called poor match, i.e., the recognized emotion 'does not fit at all' (-1.0) to 'insignificantly' (0) into the designated category [49]. Values between 0 and 1.0 are referred to as good match and quantify the gradual fit of the acoustic signal into the specific emotion category. A value of 1.0 cannot be exceeded in this definition of the reference range, here the recognized emotional signal fits completely into the designated category. All 52 emotion scores of the category_v2_ were included in the explorative factor analysis. The three emotion dimensions pleasantness, urgency and control are in row 6. The scores of pleasantness are between -1.0 (negative), 0.0 (`neutral'/in between) and 1.0 (positive). The scores of urgencies and control are between -1.0 (low), 0.0 (`normal'), and 1.0 (high) [49]. The four prosodic features pitchAverage (average fundamental frequency F0 in Hz), pitchVariation (variation of the fundamental frequency F0 in Hz), loudnessAverage (average perceived loudness of the speaker) and speakingSpeed (speaking tempo in syllables per second) are in line 8. Pitch is expressed in semitones relative to a base note (A0) of 27.5 Hz. The lowest value for pitch is 12 semitones (55 Hz) and the highest value is 62 semitones (~ 1000 Hz). For loudness, a psycho-acoustically-corrected loudness measure is employed, considering the human ear's selective frequency response and non-linear frequency and intensity perception. The speaking speed (speakingSpeed) also includes the pauses of a speaker. Thus, values below 3 - 4 Ss are called slow speaking and values above 4 Ss are called fast speaking. Values below 2 Ss result from hesitant speaking, short utterances or when many pauses are made during speaking [49]. From the total data set of 88 parameters, 28 were left out for further calculation. With the remaining 60 selected parameters, the affective features from the human voice are analyzed.
2.7. Statistical analysis
The analysis of all data sets and the graphical visualization of the results were carried out with the statistics programmed Statistics Standard from IBM Version 25.0 (SPSS®). The graphs were also created alternatively with the software GraphPad Prism version 8.1.0. (221) (GraphPad®) if this resulted in an optimized visualization.
The statistical evaluation of the LIWC2015 and VocEmoApI data sets was carried out using the methods of descriptive statistics, exploratory factor, and linear regression analysis.
2.7.1. Descriptive statistics
First, in the process of data analysis the calculation of mean (M), range (Range), minimum (Min), maximum (Max), and standard deviation (SD) was carried out as characteristics of the measured output parameters of both test methods. This was done in tabular form for the entire study group and for each individual study participant.
On the correlation of study outcomes from the dual study instruments used. The effect size of the bivariate correlation coefficient (r) was defined according to Cohen [50] as `small´ with r greater than/equal to 0.10, as `medium´ with r greater than/equal to 0.30 and as `large´ with r greater than/equal to 0.50 [51].
2.7.2. Exploratory factor analysis
Exploratory Factor Analysis (EFA) was used to evaluate the extensive output parameters of the LIWC2015 text analysis (with 61 parameters) and the voice analysis category_v2_scores (with 52 parameters). As a prerequisite for an exploratory factor analysis, the following statistical procedures or tests were applied to the output parameters of both research methods: A bivariate correlation matrix according to Pearson first gave an overview of the number and effect strength of calculated intercorrelations of the output parameters, which ultimately allow a statement on whether an exploratory factor analysis is suitable for uncovering "structures, trends and patterns" among the measured parameters (here the speech and voice analysis) [52]. The effect size of the correlation coefficient (r) was defined according to Cohen [51] as 'small' with r greater/equal 0.10, as 'medium' with r greater/equal 0.30 and as 'large' with r greater/equal 0.50 [52]. As a further prerequisite, the standard test procedure developed by Kaiser, Meyer, and Olkin (KMO) was applied with the so-called KMO value. A KMO value > 0.60 [53] is assumed an acceptable lower limit for suitability. Thirdly, Horn's parallel analysis [54] was used to determine the number of factors. For factorization, all items with a loading amount > 0.40 were considered. Factor loadings below this were removed according to the 'rule of thumb' of Wentura and Pospeschill [52] [53]. To check the internal consistency of the factors, a reliability test was carried out using Cronbach's alpha. Wentura and Pospeschill [53] state that values between 0.30 and 0.50 are to be interpreted as medium and above 0.50 as high [53]. Therefore, a value > 0.60 was assumed as a measure of reliability. With the exploratory factor analysis completed, further statistical analysis of the study was based on the factor scores alone.
2.7.3. Linear regression analysis
The following criteria are stated as prerequisites for linear regression analysis: First, an interval scaling of the independent and dependent variables, second, a linear relationship between the two variables, third, that the sample must be 'random', fourth, a normal distribution of the residuals and fifth, a variance of the residuals (homoscedasticity) [55]. The basic prerequisite of an interval scaling for the independent and the dependent variable was given by the intervention (TGPR) as measurement time points 1 to 20 over 25 weeks (six months) and by the metric output parameters of LIWC2015 and VocEmoApI. Normal distribution and homoscedasticity (homogeneity of variance) were visually checked. Since the method of non-probabilistic sampling was used in this study with seven study participants (n = 7), the requirement of "random sampling" was not met. However, this exploratory study considered all statistical findings exclusively in a chance-critical manner [52].
The strength of the linear relationship between the independent (IV) and Dependent Variable (DV) was indicated by the regression coefficient (b). In addition, a "sample significance test" was performed using a t-test to test the regression coefficient (b). The probability level of the t-value was explained with p < 0.05. The coefficient of determination (R2) provided information about the quality of the regression model [52]. For all seven individual cases, the extracted factors of the two study instruments were visualized as a scatter plot with a regression line, including a 95% confidence interval. The individual interpretation of the process courses is based both on the visual examination of the graphical representations for normal distribution (histogram) and on the interference statistical data for linear regression analysis.