Ethics statement.

The current study was approved by the local ethics committee of the Graduate School of Engineering, Chiba University, and was conducted in accordance with the Declaration of Helsinki. Prior to the survey, participants were informed that participation was voluntary and that they could quit at any time. Written and electronic informed consent was provided by each participant in Survey 1 and 2, respectively.

Translation of the Visual Discomfort Scale.

Before the commencement of the current study, the first author acquired permission to translate the VDS into Japanese from the first author of the original version [22]. The present first author and two professional translators individually translated the original English VDS into Japanese. The present first author then compared these translated versions and developed the Japanese VDS. The back-translation was conducted by two other professional bilingual translators, who did not know about the study’s purpose and the original VDS. After two other non-professional bilinguals and the present authors individually confirmed consistency between the original and back-translated VDS versions, the present first author added some minor modifications to the format and completed developing a tentative version of the Japanese VDS. Two Japanese female undergraduates completed the tentative version and confirmed its comprehensibility. Finally, we decided to use this as the final version of the Japanese VDS in the surveys that followed.

Participants and procedures.

A total of 453 Japanese students from two colleges located in Tokyo and Chiba (both in Eastern Japan) voluntarily participated in this survey and received no compensation. Printed questionnaires, which included an ethics statement, consent form, and demographic questions (sex, age, and visual acuity), followed by psychometric measures (see Measures), were distributed to the participants by visiting some classrooms and laboratories. A total of 25 of the participants, who provided more than one missing response, were excluded from the analysis. We finally analyzed data collected from 428 participants (247 females; mean age = 23.21 years, standard deviation [SD] = 6.77 years, range = 18 to 60 years). All participants reported that they had normal or corrected-to-normal visual acuity. Our sample did not greatly differ from that used in the study in which the original VDS was developed [22], which comprised 515 students in Australia, whose age ranged from 16 to 48 years (mean age and sex ratio unavailable).

Measures.

The Japanese VDS and a migraine screening questionnaire were completed by the participants.

The VDS is a Rasch-based unidimensional scale consisting of 23 items [22]. Each VDS item (Table 1; see S1 File for the Japanese version) assesses the frequency of everyday experiences regarding the perceptual and somatic aspects of visual discomfort induced when reading texts and when observing striped patterns and lights. The items are answered using a four-point scale with 0 (“Event never occurs”), 1 (“Occasionally. A couple of times a year”), 2 (“Often. Every few weeks”), and 3 (“Almost always”). The sum of the item scores serves as the VDS scale score that ranges from 0 to 69, according to the original version and those reported in previous studies [26, 32, 43, 45–47]. There are no reverse-scored items.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Rasch item measures, fit statistics, and latent principal component structure of the Japanese version of the Visual Discomfort Scale in Survey 1.

https://doi.org/10.1371/journal.pone.0191094.t001

The migraine screening questionnaire included 18 items on the occurrence of primary headaches and their duration, characteristics, and accompanying symptoms (see S2 File for details), which were designed to conform to the third edition (beta) of the International Classification of Headache Disorders [48, 49]. Responses to the migraine screening questionnaire were evaluated according to the criteria of the International Classification of Headache Disorders. The criteria for migraine without aura were as follows (aura refers to transient visual, sensory, or motor disturbance before a migraine attack). The experience of five or more migraine attacks during one’s lifetime was required. Each migraine attack was required to last for 4–72 hours and be accompanied by at least two of the following characteristics: unilateral location, pulsating quality, moderate or severe intensity, and aggravation by physical routine activity (e.g., walking). In addition, the attacks were required to be associated with at least one of the following symptoms: nausea and/or vomiting, and photophobia and/or phonophobia. Participants who met all the above criteria were classified into the Migraine category, while those who satisfied all but one of the criteria were classified into the Probable Migraine category, by reference to the International Classification of Headache Disorders and previous studies (e.g., [50, 51]). Based on this questionnaire, participants were classified into four categories: Headache-free (i.e., never experienced a primary headache; number [N] = 130), Migraine (N = 74), Probable Migraine (N = 144), and Others (i.e., experienced primary headache, but neither classified as migraine nor probable migraine; N = 80). Characteristics of the participants’ (probable) migraine headaches have been summarized in S3 File.

As migraine with visual aura (i.e., scintillating scotoma) can be associated with altered functional and anatomical alterations in visual cortices [52, 53], one might want to know about the potential effect of visual aura on the participants in the categories “Migraine” and “Probable Migraine,” all of whom were classified as migraine without aura. At the same time, we had to check whether there was a confounding effect of potential visual aura experience on the variables of interest. The migraine screening questionnaire for exploratory purposes includes an item about the occurrence of visual aura using a dichotomous response. Therefore, we could compare the participants with (probable) migraine without aura, who did and did not have an experience of visual aura, although this item did not ask about aura in detail and was not intended to formally classify as migraine with aura. Consequently, 12 participants each from the Migraine and Probable Migraine categories reported that they experienced visual aura at least once in their lifetime. Thus, our preliminary analysis using two-tailed t-tests examined differences in age and VDS score between participants with and without experience of visual aura, separately for the Migraine and Probable Migraine categories. We found no effects of visual aura for both Migraine (age: t(72) = 0.14, p = 0.89, Cohen’s d = 0.04; VDS: t(13.14) = -0.31, p = 0.76, d = 0.13) and Probable Migraine categories (age: t(142) = 1.16, p = 0.25, d = 0.35; VDS: t(142) = 0.38, p = 0.71, d = 0.11). Moreover, the chi-square test revealed no relationship between visual aura and sex for both headache categories (Migraine: χ²(1) = 1.94, p = 0.16, φ = 0.16; Probable Migraine: χ²(1) = 1.64, p = 0.20, φ = 0.11). These were consistent with previous studies that have investigated visual characteristics in migraine by comparing whether visual aura accompanies them or not, but found no differences in VDS scores [26, 46, 47, 54] and perceptual performances such as contrast sensitivity [21, 55]. Therefore, we decided not to discriminate participants with aura from those without aura in the following analyses.

Although the psychometric properties of the migraine screening questionnaire have not been evaluated and questionnaire-based screening may be quasi-valid in terms of a clinical diagnosis, our previous studies have partially validated this questionnaire. For instance, our epidemiological study in Japan using this questionnaire has revealed the prevalence of migraine and its triggering factors [56]. Importantly, our experimental studies in Japan have successfully differentiated the visual [20, 57, 58] and visuo-vestibular [59] perceptual characteristics in people with migraine from those in people without migraine classified by this questionnaire. The current study also aimed to investigate the perceptual characteristics in terms of visual discomfort, and thus adopted the migraine screening questionnaire.

Data analysis.

According to the development study of the original version [22], the Rasch model analysis was performed using Winsteps 3.92.1 [60], to examine how well individual items of the Japanese VDS fit a unidimensional model for measuring visual discomfort. Andrich rating scale model [61] was adopted because all items shared the same polytomous response structure [62, 63]. The Rasch analysis procedure followed the recent guidelines [62–65]. First, we assessed whether the four response categories of the VDS were evenly ordered by checking a continuous ordering of the thresholds, which are points at which two adjacent category probability curves cross. Disordering of the thresholds occurs when categories are underused, ambiguously defined, or hard for respondents to distinguish [66]. Measures indicating location on the unidimensional latent variable, expressed in logits, were calculated for each person and item. Infit and outfit statistics for each item were used as indices of the fit to the Rasch model. Infit (i.e., inlier-pattern-sensitive fit statistic) is based on the chi-square statistic weighted using model variance, while outfit is based on the conventional chi-square statistic and more sensitive to outliers. Since infit and outfit mean-squares indicate the amount of distortion of the measurement scale and their expected values are close to one, values less than one indicate an overfit to the Rasch model, while those more than one indicate an underfit to the model. Mean-squares within the range from 0.70 to 1.30 reflect that the item fits well to the unidimensional model for measurement [63, 65]. We adopted this criterion while there has been a more lenient criterion between 0.50 and 1.50 has also been acceptable for considering unidimensional measurement [62, 67]. Infit and outfit z-standardized statistics (Zstds, standardized t-statistics being reported with infinite degrees of freedom) indicate statistical significance of the mean-squares occurring by chance, based on the assumption that the data fit the Rasch model. A Zstd less than zero indicates overfit, while that more than zero indicates underfit. When we obtained the acceptable infit and/or outfit mean-squares, their Zstds could be ignored because mean-squares close to one indicate little distortion of the measurement scale, regardless of the Zstd [62]. The unidimensionality of the scale was also assessed by a principal component analysis (PCA) of the residuals based on the amount of raw variance explained by the measure and the eigenvalue of the unexplained variance in the first contrast.

Furthermore, we analyzed the person or item reliability based on the Rasch modeling. The Rasch reliability indicates reproducibility; high person- or item-reliability means the high probability that persons or items with high estimated measures show high actual measures than persons or items with low estimated measures do. More specifically, person reliability can serve as an indicator of reproducibility of person ordering that can be expected if this sample of persons were given another set of items measuring the same construct [68], and item reliability can serve as an indicator of reproducibility of the items’ hierarchy and/or given scores to each item if the same items were given to another sample with comparable characteristics [64]. We also reported the Cronbach’s alpha, a traditional measure of internal consistency, in order to compare the findings with those of previous studies and with those of our Survey 2. Targeting, referring to how well the item difficulty matches with the abilities of the study sample, was assessed by the differential between person and item mean location measures [65]. Differential item functioning (DIF), which indicates that one subgroup of a sample is scoring different from the other subgroup on an item [65], was checked across three subgroups: sex (male or female), age (younger than the median 21 years or not), and headache (Headache-free, Migraine, Probable Migraine, or Others). DIFs for sex and age were assessed by subgroup differentials of item measure in logits, while DIF for headache was assessed by the chi-square test where the null hypothesis was that the item had no overall DIF across subgroups [62].

We tested known-groups validity by analyzing the effect of migraine headache on the Japanese VDS. A one-way analysis of variance on the VDS score was performed to compare between four participant categories: Headache-free, Migraine, Probable Migraine, and Others. If the Japanese VDS has sufficient known-groups validity, Migraine participants would score higher on the VDS than the Headache-free, Probable Migraine, and Others participants would. However, we could also expect that the difference in the VDS score between Migraine and Probable Migraine might be small or null due to the similarity of their symptoms. To test for any effect of sex and age on the VDS and primary headache properties, unpaired two-tailed t-test on the VDS score comparing sexes and Pearson’s zero-order correlation analysis between age and VDS were performed. In addition, chi-square tests were performed to analyze the relation between sex and headache categories, and a one-way analysis of variance was performed on age with the headache categories as a factor. If there were any sex and age effects, we performed an analysis of covariance (ANCOVA) on the VDS score with the headache categories as a factor and with the sex and/or age as covariates only when the data satisfied the assumption of the parallelism of regressions (i.e., absence of interaction between independent variable and covariate) and the significance of regression by covariate. In the current study, statistical analyses, except for Rasch modeling, were performed using SPSS 24.0 (IBM Corporation, Armonk, New York). The significance level was set at p < 0.05.

Rasch analysis.

There was no disordering of thresholds of the four-response categories of the Japanese VDS (Fig 1), suggesting that all response categories were understood, distinguished, and evenly endorsed by the participants. The Rasch statistics for the items and persons have been reported in Tables 2 and 3 (see S4 File for the person statistics without minimum extreme). Descriptive statistics have been presented in Table 4. Location measures for persons and items have been summarized in the Wright’s person-item map (Fig 2). The Rasch modeling analysis provided the fit statistics indicating how well each item fit to the unidimensional model of the Japanese VDS (Table 1). As a whole, items were well fitted to the model; averaged infit mean-square was 1.03 and averaged outfit mean-square was 0.96 (Table 2). Specifically, infit and outfit mean-squares were in between 0.70 and 1.30, which is the criterion range for considering that the item fits well to the model [63, 65], except for the item 7, whose outfit mean-square was 1.42, and the item 19, whose infit mean-square was 1.31 (Table 1). As another index of the unidimensionality, the PCA revealed that 38.1% of the raw variance was explained by the measures. This was below the criterion of 50.0% for unidimensionality of the scale [65]. Moreover, the eigenvalues of the unexplained variance in the first and second contrast were 2.31 and 2.03 respectively, which exceeded the cutoff of 2.00 [65]. These suggest that the Japanese VDS has a low degree of unidimensionality and that there were other underlying dimensions in the residuals. We thus examined the presence of multidimensionality in terms of the correlation between item clusters within each PCA contrast (i.e., latent dimension) [62, 64]. The items were divided into three item clusters according to loadings on each of the first and second contrast. For the first contrast, cluster 1 included items 6 and 8; cluster 2 included items 23, 7, 2, 4, 22, 3, 5, 1, 17, 10, and 14; and cluster 3 included items 13, 18, 11, 19, 15, 9, 21, 20, 16, and 12. For the second contrast, cluster 1 included item 2, 3, 5, 4, and 1; cluster 2 included items 9, 19, 20, 11, 18, 16, 7, 12, and 10; and cluster 3 included items 21, 22, 8, 6, 23, 17, 14, 13, and 15. Clusters and items were numbered in the order of contrast loading (contrast loadings of items are shown in Table 1). Each person was measured on each item cluster for each contrast. These measures for each item cluster were correlated for each pair of clusters. The correlation was reported as the disattenuated Pearson correlation coefficient, which removed a standard error of measurement for each item cluster. If the disattenuated correlation coefficient approaches 1.00, we can consider the pair of item clusters to measure the same construct [62]. The disattenuated correlation coefficients were 0.94 to 1.00 for the first contrast and 0.88 to 1.00 for the second contrast. This suggests that the item clusters defined by two latent dimensions (i.e., PCA contrast) indeed measure the same construct. Given the low degree of raw variance explained by the measures, it can be said that the Japanese VDS has unidimensionality, but there is a high amount of unexplained variance, which is considered as random noise.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Category probability curves for the Japanese version of the Visual Discomfort Scale.

The four curves show ordered response categories. Arrows indicate three thresholds for the four categories.

https://doi.org/10.1371/journal.pone.0191094.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Person-item map of the Japanese version of the Visual Discomfort Scale.

Person locations are displayed on the left column, while item locations are displayed on the right. Each hash represents three participants, and each period represents one or two participants. Numbers on the right column correspond to the item numbers of the Visual Discomfort Scale. M: mean; S: one standard deviation from the mean; T: two standard deviations from the mean.

https://doi.org/10.1371/journal.pone.0191094.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Rasch summary statistics for 23 items of the Japanese version of the Visual Discomfort Scale.

https://doi.org/10.1371/journal.pone.0191094.t002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Rasch summary statistics of the Japanese version of the Visual Discomfort Scale for 428 persons.

https://doi.org/10.1371/journal.pone.0191094.t003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Descriptive statistics of the measures in Survey 1 and 2.

https://doi.org/10.1371/journal.pone.0191094.t004

We obtained a person reliability of 0.82 (Table 3) and an item reliability of 0.98 (Table 2). These measures suggest that the Japanese VDS is sufficient for classifying respondents and verifying item hierarchy, according to the previously explained criteria in which a person reliability larger than 0.80 and item reliability larger than 0.90 are desirable [62]. Moreover, the Cronbach’s alpha was sufficiently high (0.90, Table 3), again suggesting good internal consistency. The targeting of 1.99 was between 1.00 and 2.00, and suggested that the Japanese VDS has a medium level of matching between item difficulty and respondents’ ability [65]. Several items showed “mild” DIF for sex (i.e., differential of item measures between 0.50 and 1.00 [65]); 0.67 for the item 4 and 0.54 for the item 5, while the other items showed ignorable DIFs, lower than 0.34. There were no substantial DIFs for age, suggested by all items showed the differential lower than 0.45. Finally, some items showed substantial DIFs for the four headache categories; item 1: χ²(3) = 7.95, p = 0.047; item 4: χ²(3) = 33.64, p < 0.001; item 5: χ²(3) = 12.10, p = 0.007; item 6: χ²(3) = 10.44, p = 0.02; item 8: χ²(3) = 14.39, p = 0.002. The other items did not show DIFs (χ²s < 6.18, ps > 0.10). In sum, since several items showed mild or substantial DIF for sex and/or headache, the Japanese VDS exhibited medium quality in terms of DIF [65].

Items 7 and 19 might be considered as candidates to be removed from the VDS because they showed the fit statistics that indicated a misfit to the model. We analyzed the 21-item version of the Japanese VDS without items 7 and 19 using the same strategy explained above (see S5 File for detailed results). Even after the removal of two items, item 1 still showed an outfit mean-square of 1.34, thus exceeding the criterion range [65]. We further analyzed the 20-item version without items 1, 7, and 19. The infit and outfit mean-squares for the 20-item version were within the criterion range [65], suggesting a good fit with the model. PCA revealed that the 20-item version showed dimensionality comparable to the 23-item version. The 20-item version also showed reliability and DIF comparable to the 23-item version; however, person reliability of 0.78 did not reach the criterion of 0.80 [62, 63]. Low person reliability indicates that the scale may need more items [62]. Moreover, the 20-item version showed a low level of targeting of 2.21, larger than 2.00 [65]. In sum, even when we removed three items that seemed to misfit the model, some aspects of psychometric property (e.g., dimensionality and DIF) were comparable with its full version. However, importantly, the 20-item version would be inferior in person reliability and targeting to the full version. Therefore, we decided to utilize the 23-item full version of the Japanese VDS in the subsequent analyses.

Known-groups validity with migraine.

The VDS score did not correlate with age (r(426) = -0.06, p = 0.20), but was higher in female than in male participants (Mean_male = 10.73, SD_male = 8.92; Mean_female = 13.95, SD_female = 9.91; t(426) = 3.46, p = 0.001, d = 0.34), which is in line with the results of a previous study [23]. Moreover, age did not differ among the four headache categories (F(3,424) = 1.48, p = 0.22, η²_p = 0.01), while the distribution of the four headache categories differed between the sexes (χ²(3) = 40.00, p < 0.001, Cramer’s V = 0.31; see also S3 File). These findings suggested that female participants were more likely to have migraine or probable migraine (24.29 and 38.87%, respectively) than male participants were (7.73 and 26.52%), which is consistent with previous epidemiological evidence [69, 70]. Thus, we performed an ANCOVA on the VDS score with the headache categories as a between factor and sex as a covariate. At first, we confirmed that the ANCOVA satisfied the assumption of the parallelism of regression lines (F(3,420) = 0.85, p = 0.47, η²_p = 0.006) and the significance of regression by the covariate (t = 2.09, p = 0.04). As shown in Fig 3, the ANCOVA revealed a significant effect of headache categories on the VDS score (F(3,423) = 5.64, p = 0.001, η²_p = 0.04). A post-hoc test revealed that the Migraine participants (Mean = 15.86, SD = 9.99) scored significantly higher than the Headache-free (Mean = 10.06, SD = 8.22; p = 0.003, Bonferroni-corrected) and Others participants did (Mean = 10.94, SD = 8.32; p = 0.04). Moreover, the Probable Migraine participants (Mean = 14.11, SD = 10.56) also scored higher than the Headache-free participants did (p = 0.02), but did not differ from the Migraine (p = 0.99) and Others (p = 0.21) participants. These results suggest the known-groups validity of the Japanese VDS, which can discriminate the individuals with migraine, who are more likely to experience visual discomfort than headache-free individuals are [3, 15–21], from the individuals with other primary headaches and without any primary headaches who also experience visual discomfort from various visual stimuli [9–12]. Consequently, the higher VDS score in the Migraine participants than that in the Probable Migraine participants might also be expected. However, this was not the case for our results, perhaps because probable migraine was similar to full-blown migraine in terms of headache characteristics and accompanying symptoms [48, 49], and might be accompanied by reports of visual discomfort comparable to those accompanying full-blown migraine, or simply because there may be a limitation in the known-groups validity of the Japanese VDS.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Visual Discomfort Scale scores in participants with and without migraine, probable migraine, or other headaches.

Error bars denote 95% confidence interval. Asterisks denote significant differences (*p < 0.05, **p < 0.01, Bonferroni-corrected). VDS: Visual Discomfort Scale.

https://doi.org/10.1371/journal.pone.0191094.g003

Participants and procedures.

A total of 227 Japanese adults were recruited from Yahoo Crowdsourcing, which is an online labor market similar to Amazon Mechanical Turk [71]. They participated via SurveyMonkey (http://www.surveymonkey.com), using their own computer. At this Time 1, participants were first given the ethical statement and consent form. Participants, who provided informed consent, provided their sex and age, completed the Japanese VDS, and reported their email address. Finally, participants were thanked and given a cash voucher equivalent to 100 Japanese yen (approximately 0.9 US dollars). Due to the repeated participation, 24 participants, who showed an identical IP address, sex, and age, were excluded in order to ensure a level of data quality by screening invalid respondents [72], although previous studies have suggested that data collected from online samples can be psychometrically reliable and valid, and comparable with data collected from students and community samples [73–77] even when using Japanese online labor markets [78].

Two weeks after Time 1, we sent an email asking the remaining 203 participants to participate in the retest of the VDS (Time 2) via SurveyMonkey. A total of 127 of the participants consented to participate, and again reported their sex and age, completed the Japanese VDS, and were asked to report the email address, similar to that done at Time 1. Participants were given a cash voucher equivalent to 100 Japanese yen. In order to screen invalid respondents, we excluded nine participants who reported inconsistent sex, age, and email address at Time 1 and 2 (note that age increment of one year was acceptable). Finally, we analyzed the data from 118 participants (38 females; mean age = 40.20 years, SD = 8.42 years, range = 22 to 73 years).

Measure.

Participants completed the Japanese VDS at Time 1 and 2.

Data analysis.

Descriptive statistics including internal consistency (Cronbach’s alpha) were reported. To examine the test-retest reliability of the VDS, correlation analysis between the VDS scores at Time 1 and 2, and a paired two-tailed t-test comparing Time 1 and 2, were performed. The Pearson’s correlation coefficient (r) was used if the data normally distributed based on the Shapiro-Wilk’s test, while the Spearman’s correlation coefficient (ρ) was used if the data was not normally distributed. The Welch’s t-test with corrected degrees of freedom was used when the data violated the assumption of homogeneous variance in accordance with the Levene’s test. We further used the Bland-Altman method [79] to quantify the agreement between Time 1 and 2 by calculating the limits of agreement (LoA) that was estimated by mean ± 1.96 SD of differentials between the two measurements from the same individuals. We also reported the 95% confidence interval for the mean differential and the upper and lower limits according to a recent guideline [80]. Our sample size of 118 was above 100, which has been recommended for the Bland-Altman method [80]. The Bland-Altman plot (Fig 4) was generated using MedCalc 17.6 (MedCalc Software, Ostend, Belgium). Moreover, we examined any effect of sex and age on the VDS score at Time 1 and 2, similar to that in Survey 1. If an effect was observed, we performed the post-hoc correlation analysis partialling out the effect of sex and/or age.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Bland-Altman plot for the Visual Discomfort Scale scores in Survey 2.

Each dot represents data from a single participant. Solid lines indicate the averaged difference and the upper and lower limits of agreement. Broken lines indicate 95% confidence intervals.

https://doi.org/10.1371/journal.pone.0191094.g004