A retrospective cohort explorative research design (Sedgwick, 2014) was employed to investigate data sets, case histories, self-rating assessment questionnaires regarding perceived LPR symptoms, voice quality and voice handicap as well as patient stroboscopy reports.

Data collection took place from April to July 2019 and was obtained from the existing database of an established clinical ENT, specialising in voice disorders, at a private hospital in Gauteng, South Africa.

Participants were adults aged 18 and older. Individuals with organic pathologies and underlying neurological causes were excluded to ascertain that changes in vocal characteristics were not because of any physical factors. In addition, research has shown that treatment can cause a ‘suggestive effect’ by causing an improvement in one tool (subjective RSI scores) while not in the other (objective reflux finding scores (RFS) [e.g. oedema, oro- and hypopharyngeal erythema and laryngeal keratosis]) (Lechien et al., 2018). Reassessment data could thus affect the correlations and associations between the RSI and other assessment tools. Consequently, to avoid confounding variables, reassessment data sets were excluded and only the data of initial assessments were collected. Only data sets containing a primary diagnosis, vocal fold edge data, a complete RSI, and at least one other complete assessment measure, were included. Some participant data sets included up to three diagnoses – classified as primary, secondary and tertiary – where secondary and tertiary diagnoses specify coexisting conditions other than the main condition affecting the participant’s voice. For example, a participant may have a primary diagnosis of hyperfunctional dysphonia with a secondary diagnosis of LPR. The main diagnoses that were given to participants, based on physical examination, acoustic and perceptual voice analysis as well as self-rating scales, were functional dysphonia, hyperfunctional dysphonia and muscle tension dysphonia (MTD). Hyperfunctional dysphonia and MTD both fall under functional dysphonia, but have progressed enough so as to warrant more specific diagnoses. Hyperfunctional dysphonia is specific to the misuse and abuse of the voice, whereas MTD is specific to a functional dysphonia that specifically involves muscle tension (Van Houtte et al., 2011).

From a database of more than 800 clients, 143 met the inclusion criteria, of which 68 were excluded because of missing or incomplete RSI data sets, and six because of the multi dimensional voice programme (MDVP) having been used with them instead of the Praat programme. Praat is a programme used to analyse and reconstruct recorded speech signals (Boersma & Weenink, 2022). The MDVP and Praat programmes are similar programmes used for the analysis of voice, but individual numerical values obtained by each programme have been shown to vary greatly, making the combined use of results from these programmes inadvisable (Amir et al., 2009). Thus, the data of 51 participants with non-organic voice disorders were analysed.

The dependent variable was the RSI which measured the subjective reflux symptoms of each patient. The independent variables were gender, age, caffeine intake, physical vocal characteristics, glottal function index (GFI), VHI, jitter, shimmer, fundamental frequency (F0), maximum phonation time (MPT) and perceptual characteristics (GRBASI).

The RSI is a Likert-scale questionnaire consisting of nine questions, as outlined in Table 1 (Belafsky et al., 2002; Lee et al., 2018). Each question is answered by selecting a number from zero to five, ranging from no problem to severe, respectively (Johnston et al., 2013). The total score can be any number between 0 and 45 (Johnston et al., 2013), with a score greater than 13 being considered as abnormal (Belafsky et al., 2002).

This study looked at the following physical vocal characteristics: vocal fold edge, glottic closure, mucosal wave, periodicity, phase closure, phase symmetry and vertical level of approximation. The physical vocal characteristics were observed by means of a fibreoptic flexible distal chip optic camera stroboscopy examination. A fibreoptic stroboscopy examination is considered to be the clinical ‘gold standard’ for assessing vocal fold function (Bonilha et al., 2015).

Acoustic characteristics, such as jitter, shimmer and F0, were also analysed. Jitter is the measure of the variation of periodicity in the fundamental frequency of the vibratory characteristics of the vocal folds. A jitter score higher than one is considered abnormal (Schwartz, 2004; Teixeira & Fernandes, 2014). Shimmer is the measure of the variation of periodicity in the amplitude of the vibratory characteristics of the vocal folds, with an abnormal score being anything higher than 5 (Schwartz, 2004; Teixeira & Fernandes, 2014). F0 is the lowest frequency of the voice which correlates with a physical measure of vocal fold vibration, with a score lower than 175 or higher than 230 being considered abnormal for females, and a score lower than 110 or higher than 160 being considered abnormal for males (Ferrand, 2012; Teixeira & Fernandes, 2014).

The GRBASI 4-point rating scale (Yamauchi et al., 2010) is a measurement for the perceptual evaluation of voice quality and is widely used because of its brevity and user-friendliness. It looks at six characteristics of voice quality, including grade of hoarseness (G), roughness (R), breathiness (B), asthenia (A), strain (S) and instability (I). The VHI is a Likert-scale questionnaire which measures the perceived psychosocial effect of voice disorders and consists of 30 items, each of which can be scored between zero and four, with a maximum total score of 120 (Johnston et al., 2013). VHI abnormal scores vary for each subsection. The symptom-focused vocal impairment of each participant was measured using the GFI. The GFI is a self-administered Likert-scale questionnaire which has four questions and is answered by selecting a number from zero to five, with a possible score ranging from 0 to 20 (Johnston et al., 2013), with a score higher than four being considered as abnormal (Bach et al., 2005).

Data collection took place at the voice clinic, where data dating from 2015 to 2019 were extracted from electronic assessment reports and then manually inserted into an MS Excel spreadsheet. The data were then manually coded. The RSI, VHI, GFI and case history were completed subjectively by each client. Flexible distal chip stroboscopy examinations were used by the ENT to conduct objective patient evaluations. In addition, the speech-language therapist (SLT) used the GRBASI and the Praat computer programme to evaluate the patients’ perceptual and acoustic vocal characteristics, respectively.

The ENT and the SLT who conducted the assessments both had more than 10 years of experience in treating voice disorders, ensuring reliable outcomes (Ajmi & Aase, 2021). The cross-check principle was used to ensure the accuracy of measurements. This principle is used to improve accurate diagnosis by cross-checking the results of a single test, in this study the RSI, against an independent test measure; in this case, the reported stroboscopy evaluation outcomes, thus confirming the results of the first test (Hall, 2016; Jerger & Hayes, 1976). This principle was used to account for the reported contradictory findings regarding the RSI’s validity and reliability (Hoon Park et al., 2006). The ENT and SLT reached consensus with the classification (mild, moderate or excessive) of caffeine intake. In addition, bias of evaluation may have influenced GRBASI scores, as the rater had prior knowledge of the patient’s symptoms and history and was, therefore, not blinded during the scoring of the GRBASI.

The Statistical Package for Social Sciences v24 was used by a statistician to analyse data. Descriptive statistics were calculated for the different variables. For continuous variables, the Shapiro–Wilk test showed a lack of normality; thus, nonparametric tests were used (Field, 2018). The Spearman correlation coefficient (r_s) was used for correlations between continuous variables; the Chi-square test was used to test for associations between nominal categorical variables. Because of a small cohort and sparse data, the p-value of the Fisher’s exact test was used to determine associations instead of the p-value of the Chi-square test. The ad hoc tests – Phi coefficient and Cramer’s V – were used following the Chi-square test to determine the strength of association. A statistically significant correlation (or association) exists if the p-value is less than 0.05 (Field, 2018). Correlations range from -1 to +1; the categorisations of their strength (i.e. poor, fair, moderate, very strong and perfect) were done as per the recommendations of a 2018 user’s guide to correlation coefficients (Akoglu, 2018). A positive correlation indicates that as one variable increases, so does the other, and vice versa, whereas a negative correlation indicates that as one variable increases, the other decreases and vice versa. Phi coefficients and Cramer’s V range from 0 to 1; categorisations of their strength were again done as per the guidelines in the 2018 user guide (Akoglu, 2018). Only statistically significant correlations and associations are reported on. The level of significance and the power of this study are 0.05 and 0.9685, respectively. It should be noted that a statistical power of 0.8 or higher indicates that the sample size was large enough to ensure statistical power. This being said that, prior to data collection, an a priori power analysis was conducted using G*Power version 3.1.9.4 (Faul et al., 2007) to compute the required sample size needed for a power of at least 0.8. For conciseness, details are omitted here, as the achieved power was greater than 0.8, indicating that the sample size was sufficient for the tests conducted.

An application for full ethical approval was made to the Institutional Review Board (University of Pretoria Faculty of Health Sciences Research Ethics committee), and ethics consent was received on 27 March 2019, reference number: 1/2019. All data sets used, had previously signed voluntary informed consent forms, giving permission for future research projects to use the data. A permission letter was also obtained from the ENT of the private voice clinic to analyse the existing database. The study was conducted in accordance with the Helsinki Declaration as revised in 2013 (World Medical Association, 2013).

This study explored the associations between LPR symptoms, vocal characteristics and patient-completed questionnaires, to better understand the characteristics and diagnosis of LPR. More than half of the participants (63%) had a primary, secondary or tertiary diagnosis of LPR. The analysis of the associations between RSI, MPT, VHI, GFI, acoustic-, perceptual-, and physical vocal characteristics produced various significant correlations.

Correlations between the GRBASI and other variables varied from fair to moderate. The strongest of these was between G and jitter (r_s = 0.755 [moderate]; p < 0.001). Other studies have found similar results, and the correlation between jitter and G is the most constant finding when comparing acoustic variables with GRBASI (Lechien et al., 2016; Ziwei et al., 2014). This might be explained by the close relationship between jitter and the stability of mucosal movement of the vocal cords, which is also affected by airflow and amount of mucous on the vocal cords, which in turn is linked to hoarseness of voice (Jin et al., 2008). Thus, an improvement in hoarseness of voice should reflect in the jitter.

Other correlations with jitter were all fair except between jitter and shimmer (r_s = 0.888 [very strong]; p = 0.005). These results confirm recent findings for correlations between jitter and shimmer (Karlsen et al., 2018; Ziwei et al., 2014), and between jitter and VHI_P (Dehqan et al., 2017; Karlsen et al., 2018; Ziwei et al., 2014). The correlations, between perturbation (jitter and shimmer) and VHI_P, indicate that there is a common underlying entity, that is, vocal fatigue associated with functional voice disorders, between them (Karlsen et al., 2018; Ziwei et al., 2014).

A positive correlation was found between RSI scores and caffeine intake VHI_P and F0-male (r_s = 0.874 [very strong]; p = 0.005), but all other significant correlations between the VHI and acoustic variables were fair. This may be because of the non-organic selection criteria of this study, as research has indicated that associations are not strong between VHI scores and acoustic parameters in participants without direct vocal cord disease (Dehqan et al., 2017; Karlsen et al., 2018; Schindler et al., 2009). Further research, exploring acoustic scores and subjective parameters across populations with specific vocal pathologies, is needed (Karlsen et al., 2018; Ziwei et al., 2014).

A positive correlation was found between high RSI scores and excessive caffeine intake (r_s = 0.322 [fair]; p = 0.043), indicating that participants with excessive caffeine intake had higher RSI scores. Although some studies have shown that caffeine intake may be related to reflux (Pehl et al., 1997), more recent literature has found this data weak (Katz et al., 2013; Kroch & Madanick, 2017). The relationship between caffeine and the RSI specifically has not been explored to the authors’ knowledge.

The analysis of the associations between reflux symptoms and acoustic-, perceptual-, and physical vocal characteristics, GFI and VHI produced three statistically significant fair correlations. Correlation between the RSI and physical vocal characteristic phase closure was positive, indicating that patients with abnormal phase closure were more likely to have reflux than those with normal phase closure (r_s = 0.424 [fair]; p = 0.035). No other studies have explored this relationship and further research is, therefore, necessary. The correlation between RSI and VHI_P was positive (r_s = 0.302 [fair]; p = 0.035), indicating a link between patients’ perception of their physical voice symptoms and of their reflux symptoms. Similar correlations were reported, showing a positive correlation between improved RSI and VHI scores after treatment (Belafsky et al., 2002). In addition, a positive correlation was found between the RSI and the GFI (r_s = 0.366 [fair]; p = 0.008), indicating a link between glottal dysfunction and increased reflux. This presents the possibility of the RSI being used in conjunction with the GFI when evaluating LPR. This relationship has also not been explored in the past, perhaps because of the GFI having been created to assess glottal dysfunction and not specifically LPR (Bach et al., 2005). Thus, if further research confirms these correlations as meaningful, the RSI may be used with the GFI and VHI for more reliable diagnosis of reflux-related voice disorders. In future studies, objective clinical testing should be included so as to ensure that unbiased results are obtained for comparison. Currently, however, empirical treatment is considered best practice for the diagnostic confirmation of reflux-related voice disorders (Falk & Vivian, 2016). Reported associations between the RSI and vocal characteristics may be an indicator of the individuals who would benefit from empirical diagnostic testing.

Correlations of moderate strength were found between the GFI and the VHI_P, VHI_T, VHI_E and the VHI_F. Little research is available on the relationship between the GFI and VHI other than that by Bach et al. (2005), who found a correlation of 0.61 (p < 0.001) between total GFI and VHI scores, which significantly reduced post-therapy for organic voice disorders. However, albeit not the aim of the current study, these significant correlations indicate that although these are two distinct clinical tools, they may both be used in conjunction with the RSI to increase reliability in monitoring vocal pathology. As significant correlations were found between the RSI and VHI_P and between the VHI and GFI, these relationships should be further explored to establish whether the RSI, GFI and VHI have meaningful significant relationships, which could make the diagnosis of LPR using subjective-based instruments more certain.

The results of this study indicate that associations exist between reflux symptoms and vocal characteristics, but further research is necessary to determine whether these results hold any true value for improving the diagnostic criteria for LPR.

Because of the strict exclusion criteria, confounding variables were limited and results were very specific. All data measures that were included are applicable to the typical voice practice. In addition, this study also has a high achieved power (0.9685) with a high power (typically 0.8 or greater), indicating that there is a large chance of a significance test detecting a true correlation or association. Limitations of the study include the strict exclusion criteria and retrospective nature of this study made it necessary to exclude participants who met the requirements, but did not have complete RSI forms, and may have led to selection bias. The cohort size (n = 51) of this study was smaller than expected. Only a minority of people seek private healthcare services for their voice disorders, even though its subjective impact on their lifestyles may be significant in terms of vocational demands and social interaction (Bhattacharyya, 2014). Furthermore, Fourie et al. (2017) found that the incidence of voice disorders amongst private practices in Gauteng specifically was only 5.2% (Fourie et al., 2017). It can be expected that the prevalence of voice disorders is much higher than the incidence rate but that very few people seek treatment (Fourie et al., 2017), often because of time and leave constraints, as well as a lack of awareness of when to seek help and whom to seek help from (Da Costa et al., 2012). An international study reported that people often seek help only when it severely affects their ability to perform their vocational duties (Roy et al., 2005). This, together with the strict exclusion criteria, may explain the small cohort, which limited the types of statistical tests that could be used for data analysis.