Checking for voice disorders without clinical intervention: The Greek and global VHI thresholds for voice disordered patients

Voice disorders often remain undiagnosed. Many self-perceived questionnaires exist for various medical conditions. Here, we used the Greek Voice Handicap Index (VHI) to address the aforementioned problem. Everyone can fill in the VHI questionnaire and rate their symptoms easily. The innovative feature of this research is the global cut-off score calculated for the VHI. Therefore, the VHI is now capable of helping clinicians establish a more customizable treatment plan with the cut-off point identifying patients without normal phonation. For the purpose of finding the global cut-off point, a group of 180 participants was recruited in Greece (90 non-dysphonic participants and 90 with different types of dysphonia). The voice disordered group had higher VHI scores than those of the control group. In contrast to previous studies, we provided and validated for the first time the cut-off points for all VHI domains and, finally, a global cut-off point through ROC and precision-recall analysis in a voice disordered population. In practice, a score higher than the well-estimated global score indicates (without intervention) a possible voice disorder. Nevertheless, if the score is near the threshold, then the patient should definitely follow preventive measures.

www.nature.com/scientificreports www.nature.com/scientificreports/ to obtain the opinions of patients regarding the psychosocial effects of their voice disorders on their daily lives 1 , recent studies have also reported its discriminatory ability [40][41][42][43][44][45][46] . The latter is a very important finding, as it demonstrates that this questionnaire is a very important tool in the hands of specialists and every person who wishes to preliminarily check his/her vocal-voice status for potential abnormalities before the use of interventional actions (e.g., endoscopy).
The aforementioned discriminatory capability has also been reported by studies that included the cut-off points for diverse populations [40][41][42][43][44][45][46] . Those cut-off points were computed with the use of the Receiver Operating Characteristic (ROC) analysis. The ROC analysis was developed by physicists during the Second World War and was used as an accurate discriminatory method for radar signal analysis. Similarly, this well-established method is currently used by health sciences to distinguish between healthy populations and populations with different pathologies. Specifically, the ROC curve is used to determine the appropriate threshold. Thus, the first purpose of our study was to determine the cut-off point for the VHI and its three domains for a voice disordered population from Greece (limitations apply, please see section 4.4). The second purpose of this study was to determine the correlations between various populations, while the third purpose was to estimate a global cut-off point (for voice disordered patients). The latter could be a true innovation in global voice screening procedures.

Results
The sample consisted of 180 participants (90 controls and 90 voice disordered People -VDP). Non-dysphonic participants exhibited a VHI-T mean of 14.73 (SD = 2.27), a VHI-F mean of 4.56 (SD = 1.57), a VHI-P mean of 5.28 (SD = 1.78) and a VHI-E mean of 4.90 (SD = 1.63). The VHI-T mean score for the VDP group was 37.82 (SD = 20.89), while for the VHI-F group it was 10.72 (SD = 7.62), for the VHI-P group it was 16.98 (SD = 8.08), and for the VHI-E group it was 10.12 (SD = 8.10). The aforementioned are presented in detail in Table 1. VHI can detect dysphonic patients or people with potential voice problems with the use of statistical, ROC, and precision-recall analysis. More details are given below.
A Kruskal-Wallis test was conducted to compare the subgroup median VHI-T scores and the scores on the different domains. In particular, significant differences were observed between the non-dysphonic and dysphonic groups for the VHI-T score, with mean scores of 58.88 for non-dysphonic participants, 119. 14 (Table 3).
A ROC analysis was conducted to determine the cut-off points of the VHI-T and its three domains (functional, physical and emotional). The threshold (cutoff point) was derived from the maximum "Sensitivity + Specificity" coordinates (1-Specificity, Sensitivity) (please see Krzanowski and Hand 47 ). Another way of calculating the same value for the cutoff point similar is to find the maximum score obtained when subtracting "1-Specificity" from the"Sensitivity", i.e., the cutoff point coordinates' condition = the maximum {Sensitivity (i) -(1 -Specificity (i)} where i ϵ [n], [n] = {1, …, n}, for n ROC analysis points. The last can be justified as "Sensitivity" referring to true positives (i.e., voice disordered people are correctly identified as voice disordered) and "1-specificity" referring to false positives (i.e., people without disorders are incorrectly identified as voice disordered). Consequently, the best cutoff point should be the coordinates for which we have the maximum {Sensitivity (i) -(1 -Specificity (i)}, because that is the point at which the most people are identified correctly as voice disordered.
The cutoff point of the VHI-T score was found to be 19.50, with a sensitivity of 0.882 and a 1-specificity of 0.022 (Fig. 1). The VHI-F cutoff point was 7.50, with a sensitivity of 0.735 and a 1-specificity of 0.022. Additionally, the VHI-P cutoff point was 8.50, with a sensitivity of 0.868 and a 1-specificity of 0.000, and the VHI-E cutoff point was 8.50, with a sensitivity of 0.618 and a 1-specificity of 0.011. Specifically, with regard to the usage of the 3 domains' cutoff values, these were considered in the past of lower importance compared to the total score. This could be partially correct when the domains are not equal. In contrast, when they are equal, which we observed for the first time in 45 , then this identifies with more certainty a potential voice problem, in addition to the fact that the self-perceived score is at least equal to the total cutoff score.  www.nature.com/scientificreports www.nature.com/scientificreports/ To validate the proposed total cut-off point (threshold) of 19.50, we decided to produce a precision-recall curve (Fig. 2). The main idea behind this dual analysis is that rare events (e.g., positive samples are very few in number compared to the negative ones), then a precision-recall analysis is appropriate 48 . This case does not apply to our sample but takes into consideration that the ROC curve remains unchanged relevant to "rare" positive samples, making it wise to conduct an additional validation of the already calculated ROC cutoff point. Consequently, after creating the precision-recall curve and finding the AUC of 94.1% with a cut-off point at 19.50, the threshold proposed in this study was validated for the detection of potential voice disorders in the Greek population.

Discussion
The results of the present study (limitations apply, please see section 4.4) revealed that VDP exhibit significantly higher means and medians on all three domains and on the total score of the VHI compared to the control group. This finding indicates that dysphonic patients perceive their voice differently than non-symptomatic groups [15][16][17] . Consequently, the discriminatory ability of the VHI has been confirmed. Moreover, the following similarities and differences were found: • For VDP, the VHI-T median and mean scores were estimated to be 35.50 and 37.82, respectively, which is similar to many results from studies of the cross-cultural adaptation of VHI to dysphonic populations 1,21,23-33 .   www.nature.com/scientificreports www.nature.com/scientificreports/ and the Italian version of the VHI (mean = 38.38) 31 . The mean and median values of the VHI-F were 10.72 and 10.00, respectively, for the VDP subgroup. These scores were also in good agreement with those reported in other studies 1,[21][22][23][24][25][26][27][28][29][30][31][32][33] and were close to those on the original version of VHI (mean = 10.07) 1 and the Greek VHI (mean = 10.00 and median = 9.00) 25 .
The previously mentioned studies did not consider the cutoff values of the VHI domains, only the total VHI score. The use of cutoff points for the individual VHI domains enhances the discriminatory ability and the diagnostic capability of this tool, which has also been reported before 41,42 . Karlsen et al. 26 highlighted that the VHI-F score tends to accurately diagnostically discriminate between dysphonic and non-dysphonic individuals, which was also reported in the current study. Moreover, Helidoni et al. 25 emphasized the fact that the VHI-P domain had a good discriminatory capability because VD patients are more familiar with vocal symptoms, and physical signs are more detectable through a self-perceived diagnosis.
The use of patient-centred evaluations, both objective and subjective, is of great importance, as concluded by other studies 4,25 . Specifically, the use of VHI cutoff points will help clinicians better understand the condition of people with vocal symptoms to better address their problems 25 and to accurately monitor their health condition. To summarize the significant findings, we conclude the following (limitations apply, please see section 4.4), which are very important for accomplishing the globalization of the VHI and thus are related to the purposes of this study: • The Greek VHI cutoff point (COP) was 19.50. Consequently, every score higher than this in the Greek population should immediately be reported to a specialist so that the potential patient can be examined. • Various populations investigated in previous studies were also examined to determine commonalities with the Greek population. Most studies that met the following criteria exhibited similar cutoff points: the same type of sample (almost the same severity of disorders), nearly equal numbers of controls and patients, all the patients were in the same phase of therapy, and the study defined only a cutoff point and not a range. • Taking into consideration the previous criteria, we found the following cutoff points: the Greek version reported by Tafiadis 43 . The mean cut-off point for these populations was found to be higher than 19 (i.e., 19.10) with a standard deviation of ≈0.7. In turn, this means that globally, when someone exhibits a higher score than 18.50, (in practice this score should be at least equal to 18 as the minimum scale increase is 1) then he/she should visit a specialist. This value is lower than the Greek cutoff point (in practice equal to 19), but it provides a more stringent preventive screening method. • Τhis study demonstrated that the three domains of the VHI exhibit similar cut-off points for the Greek population and smokers 41,42 . This near equality between the subdomains (with cut-off points from 7.50 to 8.50) may be useful for regular monitoring in cases where there is a strong perception that a voice symptom is persistent. At the same time, the equality of the VHI thresholds may be beneficial for patients with voice problems who want to check their therapeutic progress, as has been suggested in the case of smoking cessation 41,42 . www.nature.com/scientificreports www.nature.com/scientificreports/ • When people want to check themselves for potential voice disorders, the procedure is simple. They must fill out the VHI 1 . Then, they calculate their score. If their total score is above 18, then there is a strong possibility of a clinical voice problem. If so, they should seek the help of a professional. Especially for the Greek population, subdomains should also be checked. If their scores are above the cutoff points, then there is an even stronger possibility of a clinical voice disorder. • Finally, a new factorial analysis could be performed, as in the research by Hongyan Li et al. 49 , who took into account various economic and cultural differences when they compared eastern and western populations and who designed a completely new method with new tools. However, these differences do not exist to a large extent between different western cultures. Consequently, there is no imminent need to redefine the VHI tool. Furthermore, our co-author Thomas Murry was the first to produce the VHI-10 50 and he, along with our other co-author, Meropi E. Helidoni, introduced the formal Greek VHI edition 25 . Finally, members of the team published in 2018 the formal Greek VHI-10 edition 51 . Consequently, based on a deep knowledge of the VHI, we declare that there is no need to re-evaluate the VHI tool. Nevertheless, factorial analysis could be useful in the future to determine how much our cultures have changed over time being due to economic issues and population migration. The last two facts could even constitute new balances and behaviours inside newly emerged societies.
Conclusions. The aim of this study was to estimate the VHI cutoff points (for all domains) for populations with voice disorders (VD) in Greece (limitations apply, please see section 4.4). The VHI can significantly distinguish VD and non-VD populations based on their perception of their voice status. The Greek VHI can be used as a monitoring tool for clinicians. Furthermore, it can be used as a screening procedure for voice professionals. The proper use of the VHI cut-off points can assist a voice clinician in better predicting the course of a patient's voice disorder. Finally, these cutoff points, in combination with a voice assessment, can better characterize an individual's voice disorder. In turn, this can lead to a more customized treatment.

Materials and Methods
Participants. This study was approved by the Ethical Committee of the Medical School-University of Crete. One All subjects with voice disorders were classified by ENT doctors and speech language pathologists (SLPs). The patients' subgroup was determined via video laryngeal endoscopy and stroboscopy. All VDP had not undergone a previous laryngeal surgery nor did they have other recent surgeries during recruitment. Sixty-seven (67) patients attended the ENT clinic of the University Hospital of Crete, while the remaining twenty-three (23) attended a private medical office in Athens-Greece and an SLP office in Ioannina-Greece. The geographic distribution of normative subjects (90 non-dysphonic) was the same compared to dysphonic subjects.
The control group consisted of subjects who accompanied the patients, those attending the ENT department for reasons other than voice disorders, clinical staff members and subjects from the School of Health and Welfare Professions at the TEI of Epirus.
Non-dysphonic subjects who were not included in the study presented any upper or lower respiratory system disorder or any laryngeal/vocal complaints in the last two weeks or had symptoms of gastroesophageal reflux (GERD) or laryngopharyngeal reflux (LRP) disease. In addition, excluded subjects were those who had attended voice therapy and/or presented voice disorders in the past, who had a history of alcohol and/or drug abuse, who lived or worked in environments that had factors influencing the voice (smoke, dust, exposure to chemicals, external noise, and/or allergens, etc.).
It must also be clarified that the number of participants was defined based only on the strict criteria selection and not on statistical estimations of the sample size. Consequently, many participants were found, but a large number was excluded.
The population sample was partially representative of the general population of Greece (limitations apply, please see section 4.4), and the distributions of pathology, gender and age were also appropriate. Particularly while the sample size seems small, many people were excluded based on the strict criteria mentioned in this section, which balances the drawback of the small total number. Moreover, (1) because the sample originated from different regions of Greece, it is partially representative (locations were only from Athens, Ioannina, Crete); (2) the sample was selected with strict exclusion criteria, which indicates additional sample representativeness; and (3) as the sample's collection was conducted by different independent professional researchers who collected this sample of each region, the sampling procedure could be defined as adequate with limitations (please see section 4.4).
www.nature.com/scientificreports www.nature.com/scientificreports/ Data collection. The Hellenic VHI index 25 and the translation of the Greek Voice Evaluation Template (VET) 52 were filled in by all participants. The VET is a consensus template about the voice that the American Speech Hearing Association (ASHA) developed that can be used in daily clinical practice. All subjects received information before enrolment about the study's purposes and the confidentiality of the obtained data. They were also asked to sign a consent form.

Statistical analysis. The distribution of variables was tested with the Kolmogorov-Smirnov and
Shapiro-Wilk tests. All skewed variables (VHI scores) are expressed as the medians and the interquartile range, and all normally distributed variables are expressed as the means and standard deviations (SD). A Mann-Whitney U test was used for the comparison of the two study groups (dysphonic and non-dysphonic participants). Furthermore, the Kruskal-Wallis test was used to compare the five subgroups (non-dysphonic participants, LML patients, LID patients, NVD patients and HVD patients). ROC and precision-recall analyses were conducted to estimate the best cut-off values for the VHI-Total (T) and its three domains. All reported p-values were two-tailed, and the statistical significance was set at the value of p < 0.05. The analysis was performed with SPSS statistical software (version 19.0, Armonk, NY, USA).
Limitations and future solutions. The sample collection was conducted only in three different regions of Greece (Athens, Ioannina, Crete) so this constitutes a limitation of this study. Future works should include more regions in order the sample to be characterized as Panhellenic and not as partially representative like in our study.
Moreover, most of the dysphonic samples (67) were collected from the University Hospital of Crete while the remaining twenty-three (23) attended a private medical office in Athens-Greece and an SLP office in Ioannina-Greece. At first sight the sample seems strongly representative of the Crete's population and not generally all of Greece, but this is not the case as we conducted various random checks to people all around Greece compared to the sample of this candidate paper. Unfortunately the gathered sample from each area was too small, so this does not constitute a strict validity in order to give further details in the manuscript. A future solution to this limitation should include a vast and well-balanced sample collection from many regions of Greece. The sample should be gathered in such a way that should depend on Eurostat data (the number of participants from every region should correspond to the real percentage of the Greek living population in these areas). Then this future study should be definitely treated as an epidemiological study.
Ethical approval. All procedures performed in studies involving human participants were conducted in accordance with the ethical standards of the institutional and/or the national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards.

Informed consent. Informed consent was obtained from all individual participants included in the study.
This study was approved by the Ethical Committee of the Medical School-University of Crete.

Data Availability
All the included data are fully available without restriction.