Participants
Patients with PPPD
Twenty-nine patients were recruited from the vestibular clinic at University Hospital Wales (UHW). All patients had received a diagnosis of PPPD from a Clinical Scientist in Audiology or a Consultant Audiovestibular Physician, following common tests to examine vestibular functioning, including Halmagyi bedside head thrust testing, Video Head Impulse testing (VHIT using Synapsys system), Videonystagmography (typically saccades, pursuit, gaze using GN system) and (sometimes) caloric testing if deemed necessary. The average age of participants was 44 (sd = 14.3, range 11–67), 60% were female.
General population cohort
Two methods were used to recruit participants from the general population: (1) A community health participant list in Wales, (2) Prolific academic, a website where members of the general public can sign up to take part in studies in return for payment.
Participants in a community health list were emailed an advert and link to take part in a survey. The survey was advertised as being about ‘Health and the Senses’ and contained the following text: ‘The School of Psychology at Cardiff University are investigating health and the senses through an online survey. Dizziness is common in the general population and can have serious consequences for daily functioning and health. The research team are interested in a particular type of dizziness that is triggered by being in certain environments. These tend to be environments where there is a lot of clutter, for example, a supermarket or a crowded street. They are interested in how common this dizziness is in the general population and how it might relate to other conditions (e.g. migraines). In the future, they hope this research will help them to develop more effective rehabilitation tools for dizziness. The online survey will include questions and pictures about sensory sensitivity, dizziness and migraines, and is open to everyone. They would like to hear from a range of people, whether or not you suffer from dizziness and migraines.’ We emphasised the inclusivity of the survey so that individuals with an interest in dizziness and migraines would not selectively participate.
From the 18,683 email addresses sent the invite, we received ~ 2500 responses, 972 of which had complete data for each of our measures (necessary to build the full SEM model, and we use the same sample of participants at each stage for direct comparability). The average age of participants was 57 (sd = 13.8, range 19–86), 72% were female. The median level of education attainment was 3, IQR 2–4 (where 0 = no education, 1 = GCSE/O Level, 2 = A-level/BTEC, 3 = Undergraduate, 4 = postgraduate).
To augment this sample, we used Prolific Academic to recruit 211 participants online in return for £5 compensation. Of these, we received 135 valid responses with data on each measure. The average age of these participants was 27 (sd = 7.1, range 18–55), 27% were female. The median level of education attainment was 3, IQR = 2–3.
Combining the two cohorts provided a sample size of 1107, with a mean age of 56 (sd = 16.47, range 18–86), 67% female, and median education attainment of 3 (IQR 2–4). We asked participants to report if they had a current diagnosis of any common vestibular related conditions (N = 111), the summary of which is shown in Table 1. For the comparison with patients, we removed from the ‘non-clinical’ control cohort participants who had reported vestibular conditions, resulting in a sample size of 996. For the SEM analysis (where we are interested in symptom variance in the population rather than having a ‘non-clinical’ control group) we included all the participants from the general population cohort (N = 1107), but did not include the PPPD patients we had recruited from the clinic.
Table 1 Self-reported vestibular related conditions in the general population cohort (i.e. separate to the 29 patients recruited through clinic) All procedures were approved by the Cardiff and the Vale University Health Board and the School of Psychology, Cardiff University, ethics committees, and have therefore been performed in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki and its later amendments.
Measures
All aspects of the survey were delivered via Qualtrics, an online survey tool.
Demographic information: Basic demographic information was collected including age, gender, educational attainment and if they had a current diagnosis of any common vestibular conditions.
Visual Vertigo Analogue Scale [VVAS, 5] was used to evaluate symptoms of PPPD. Participants rated on a scale of 0–10 the amount of dizziness they experience in 9 situations that are known triggers for visually-induced dizziness. These include walking down a supermarket aisle, walking across a patterned floor, and going to the cinema. Scores on each item are typically averaged and then multiplied by ten so that the total score an individual could achieve by rating all situations a 10 (maximum dizziness) is 100. However, in our SEM model, we included the nine items as separate indicators for a ‘PPPD symptoms’ latent factor. The overall internal consistency of these indicators was good (See Fig. 2 for full item loadings and supplementary materials Sections A-C for construct reliability and validity checks).
Adolescent/Adult Sensory Profile [AASP, 41]: The adolescent/adult sensory profile (AASP) is a questionnaire that measures individual differences in four sensory subtypes related to Dunn’s model [41]. We were primarily interested in the subscales that assess ‘low sensory threshold’ [47], which are called ‘sensory sensitivity’ and ‘sensory avoidance’ (i.e. behaviours attempting to counteract strong sensations, such as avoiding strong tastes). The AASP is a 60 item questionnaire that asks participants to rate on a Likert scale the frequency that they perform certain behaviours (almost never, seldom, occasionally, frequently and almost always). The questions are split into six sensory domains: taste/smell processing, movement processing, visual processing, touch processing, activity level and auditory processing. Items from the movement processing and visual processing domains were removed as these overlap with known PPPD symptoms. The sensory sensitivity and avoidance subscales showed acceptable internal consistency in our sample (sensory sensitivity Cronbach’s a = 0.73; sensory avoidance a = 0.77). Due to the high collinearity between these subscales (R = 0.72), the scores were included as indicators to one ‘multi-sensory processing’ latent factor in our SEM model, which again showed good construct reliability and validity (see Fig. 2 for loadings and supplementary materials for inner model checks; note we use the label ‘multi-sensory processing’ to avoid confusion with objectively measured sensory thresholds).
Hospital Anxiety and Depression Scale [HADS, 48]: The HADS is a 14 item scale containing 7 questions that contribute to an anxiety subscale and seven questions related to a depression subscale. Due to the previous focus on anxiety and dizziness, we only used the anxiety subscale. Example anxiety questions include: ‘I feel tense or wound up’ and ‘I get a sort of frightened feeling as if something awful is about to happen’. Participants are given four response options per question (.e.g. most of the time, a lot of the time, from time to time, not at all) and are asked to select the option that is closest to how they have been feeling in the past week. Questions are both positively and negatively worded. Options are scored from 0–4, where 4 indicates more anxiety. Items can then be summed to provide an overall anxiety subscale score, however, in our SEM model, we included the seven items as separate indicators to an ‘anxiety’ latent factor. The overall internal consistency of these indicators was good (see Fig. 2 for full item loadings and supplementary materials for construct validity checks).
Migraine Screening Questionnaire [MS-Q, 49]: The MS-Q is a five-item screening tool that identifies probable migraine. Participants answer yes/no questions about headache episodes they experience, which include ‘Do you usually suffer from nausea when you have a headache?’ and ‘Does light or noise bother you when you have a headache?’ Participants must respond ‘yes’ to four or more of the five questions to obtain a result of probable migraine.
Visual Activities Questionnaire [VAQ, 50, 51]: The VAQ is a measure of low vision difficulties that can affect every-day life across a number of domains, such as acuity, visual search, peripheral vision and colour vision. The original VAQ contains 33 items and 8 subscales [51], however, due to time constraints on participants, we used a reduced 13 item, unidimensional version [50]. Each question asks how often the problem occurs on a 5-category rating scale from 1 (never) to 5 (always). Items are averaged to produce a total score, where higher values indicate more low vision difficulties. The total VAQ score showed good internal consistency in our sample (a = 0.86).
Visual discomfort: Participants were asked to rate on a scale of 0–10 the amount of discomfort they experienced when viewing a selection of 20 images. These images were taken from Penacchio and Wilkins [52] and we used them previously in our paper on visual discomfort and symptoms of PPPD [28]. Half of these images were ‘high discomfort’ in terms of both previous participant ratings and spectral content analysis. The remaining half were categorised as low discomfort. The images spanned three categories: photographs of buildings, abstract art, and geometric shapes. The images were embedded in the Qualtrics questionnaire and were viewed on participants’ personal devices, so they were rendered at slightly different sizes and resolutions across participants. However, we asked participants at the beginning of the questionnaire to use the device with the biggest screen (e.g. tablet preferable over a phone). Most participants used a computer monitor or a tablet to view the images (computer = 62%, tablet = 21%, phone = 17%). The average resolution (width × height) for the devices used by participants was 1,059,422 pixels. On a standard 22in monitor with a viewing distance of 60 cm, the images subtended 25° × 15° of visual angle. We subtracted average ratings of discomfort on high discomfort images from low discomfort images, to yield a ‘visual discomfort score’. This score represented a participant’s particular aversion images that deviate from the statistical properties of natural scenes.
Situational characteristics questionnaire [SCQ, 53]: The SCQ is a 20 item questionnaire that asks participants to rate discomfort in different situations that include intense visual salience or visual-vestibular conflict. It was originally developed to measure space and motion discomfort, but this is now considered to be closely associated with the new diagnosis of PPPD [1]. Situations are rated between 0 and 3 and scores are normalised by subtracting responses to paired situations that are not commonly associated with visually-induced dizziness. The final score is obtained by dividing the summed ratings across all items by the total number of items and multiplying by 10, therefore the maximum score that can be given is 30. Item 15 from the Prolific academic responses was removed due to a question transcription error. We have previously shown that the VVAS has higher internal consistency than the SCQ in a general population sample [27], and so we used the SCQ as a secondary outcome measure of PPPD symptoms and report the results in the Supplementary materials (Section D). Due to a large number of items in the SCQ (N = 20), items were combined following standard procedures to produce one SCQ score, and this was included as the outcome factor in our SEM model.
Structural equation modelling (SEM) procedure
Data were analysed using Smart-PLS, a software that supports partial least squares, or component-based, SEM [54, 55]. In the PLS routine, indicator variables (e.g. questionnaire items or scores) are first standardised to a mean of 0 and a standard deviation of 1. Next, the standardised indicator variables are combined and equally weighted to produce latent variables scores (using a ‘reflective’ model, which assumes indicators, such as questionnaire components, reflect rather than cause the underlying construct, such as anxiety). Initial weights are then applied to the hypothesised paths between the latent variables in such a way that maximises the R-squared of each latent variable. After this initial estimation, the PLS algorithm iteratively adjusts the weighting of the indicators and the latent variable path connections to maximise the explained variance across the model. These iterations stop when there is no significant change in the weights of the indicator variables.
Before interpreting the results from the SEM, we carried out a number of checks to ensure the validity of the outer (indicator to latent variables) and inner models (latent variable paths) [54, 55]. We first checked the construct validity of the indicator relationships to each latent variable by examining indicator reliability (average loading > 0.7), internal consistency (Cronbach a > 0.7), convergent validity (average variance explained, AVE > 0.5) and discriminant validity (Heterotrait-Monotrait Ratio, HTMT < 0.85). If these were within the limits of acceptability, we then checked that none of the latent variables was collinear (VIP < 5). We then used bias-corrected accelerated bootstrapping (5000 iterations) to test the significance of each path coefficient and report the F square as a measure of effect size (where f2 > 0.3 = large effect, f2 > 0.15 moderate effect, f2 < 0.15 = small effect). Due to the moderately large sample size, even small coefficients tended to be significant, so it was important to examine the f2 values. Finally, we used a blindfolding technique to ensure that estimated model was robust by iteratively removing a subset of the cases and calculating the predictive accuracy of the reduced data model to estimate the omitted data points (Q2 > 0).
The main predicted variable in all models was PPPD symptoms. In the main analysis we used the VVAS as our measure of PPPD. We repeated all path models using the SCQ as our measure of PPPD and report these results in the supplementary materials.