Evaluating the effects of smoking on the voice and subjective voice problems using a meta-analysis approach

The objectives of this study were to identify the effects of smoking on the voice of smokers and present the baseline data for establishing the basis for preventing voice disorders. This study was evaluated using a meta-analysis from studies published between Jan 1, 2000, and Nov 15, 2018. As a result, the final meta-analysis was conducted using nine papers. The standard mean difference was analyzed after dividing the effects of smoking on voice into the pitch (F0), sound quality (jitter, shimmer, and noise to harmonic ratio; NHR), Maximum Phonation Time (MPT), and subjective voice problem. The results showed that there was a significant difference in F0 and MPT. On the other hand, the jitter, shimmer, NHR, and Voice Handicap Index (VHI) had different mean effect size but they were not significantly different. The analysis by sub-function of VHI results showed that the mean effect size was significantly different only in VHI-P (Physical). This study evaluated the effects of smoking on voice using meta-analysis. It was confirmed that smoking had significant and moderate effects on the F0 of voice, MPT, VHI, and physical functions. It is necessary for future meta-analysis studies to conduct randomized controlled experiments or longitudinal studies to confirm the effect sizes of variables.

Including and excluding criterions. This study excluded studies which did not use a statistical analysis such as case studies and reviews. Secondly, only studies with a control group were included. Thirdly, studies dealt with respiratory diseases or neurological diseases (e.g., vocal cord paralysis due to stroke). Fourthly, intervention studies to evaluate the effects of treatment and animal experiments were excluded. This study only examined papers published in English and Korean. The exposed group was defined as the current smokers, while the unexposed group (control group) was defined as people who never smoked.
A total of 1,293 papers were initially selected for the meta-analysis. Afterward, duplicated papers (n = 61), studies meeting the exclusion criteria such as cases studies (n = 1,163), and studies using median instead of mean and those which we could not examine the full contents (n = 60) were excluded from the study. As a result, the final meta-analysis was conducted using nine papers (Fig. 1).
Qualitative assessment. The quality of selected literature was assessed using the Newcastle-Ottawa Quality Assessment Scale (NOS). The NOS, developed by the University of Newcastle and University of Ottawa, is a tool www.nature.com/scientificreports www.nature.com/scientificreports/ to evaluate the quality of control cohort studies 22 . It is composed of four items for evaluating selection and comparability possibility and three items for examining results. The items for selection and result can have one star (★) and the items of comparability can have up to two stars. The more stars means the higher quality of a study ( Table 1). The target publications used for the analyses were categorized according to authors, year of publication, number of study subjects, and statistical values (i.e., mean and standard deviation) and coded accordingly. The quality of the literature was assessed by two researchers independently through NOS. When there was disagreement, it was concluded through discussion. The quality evaluation of the target studies ranged from four to seven and only one publication 15 was excluded from the study due to low quality (Table 2).

Quality assessment of target papers.
According to the quality evaluation of each item showed that all studies did not secure the representativeness of the exposure groups and the comparison groups (Table 2). Most studies recruited study subjects from hospitals 12 or a specific institute such as a school 13,15,17,20,21,23,24 (88.9%). Hamdan et al. 25 did not mention study subject recruitment (11.1%). Five studies (55.6%) defined non-smokers as people who stopped smoking 1year or more. Simberg et al. 15 did not provide the definitions of smokers and non-smokers but the study was conducted by distinguishing between smokers and non-smokers. The only study 24 only recruited smokers and it did not define non-smokers. Among the nine studies, one study did not report a laryngeal disease. Seven studies (77.8%) evaluated a laryngeal disease and Hamdan et al. 24 directly confirmed the presence of a laryngeal disease using laryngeal stroboscopy.
Five studies have controlled sex 13,15,21,24,25 , while Awan 23 and Tafiadis et al. 20 only targeted female subjects and Pinar et al. 17 and Lee et al. 12 only examined male subjects. Gonzalez et al. 13 evaluated both male and female subjects. Three studies (33.3%; Hamdan et al. 24 , Pinar et al. 17 , and Gonzalez et al. 13 ) controlled age. It is considered to be adjusted if the scope of adult age is within five.  www.nature.com/scientificreports www.nature.com/scientificreports/ As a scale to evaluate voice problems in a self-reporting pattern, three studies (37.5%) used the voice handicap index (VHI) 17,20,21 . Simberg et al. 15 (12.5%) only reported the prevalence of a voice disease using a questionnaire that evaluates the presence of a voice disease based on symptoms.
Acoustic evaluation tools were conducted in six studies (66.7%). Among them, two studies (22.2%) used the Visi Pitch system 24,25 , the one study (11.1%) used the Praat software package 17 , one study (11.1%) used Dr. speech software 20 , one study (11.1%) used Multi-Dimensional Voice Program 13 , and Awan 23 (12.5%) used the Dysphonia Severity Index (DSI). All studies used the same interventions to compare groups, and no dropout rate was reported.
Meta-analysis. The effect size was analyzed using 'meta' package of R (ver. 3.4.4). The means and standard deviations of the exposed group and the unexposed group were used in the analysis. For the effect size, standard mean difference (SMD) was used by using Hedge's g. The weighted mean effect size considering the sample size was used for the mean effect size. The effect size and significance was examined based on the 95% confidence interval. The effect size was divided into 'small effect (≤0. Homogeneity test was conducted to examine the statistical heterogeneity of the effect size of each study. The results showed that Q-df was higher than 0 and F was confirmed to be equal to or greater than 75%, which implied that the variances of groups were significantly different and the effect size of them was heterogeneous. Therefore, the random effect model was used for this meta-analysis.
The publication bias was examined in order to verify the validity of the meta-analysis results. The funnel plot and adjusted funnel plot revealed that studies were scattered near the effect estimates and it was determined that there was no publication bias. The trim-and-fill method was applied for adjusting visual asymmetry, and it was found that the risk ratios before and after adjustment were similar. Egger's regression test and Kendall's tau test were performed to analyze the statistical symmetry. The results showed that it was significantly different in the continuous correlation condition (tau=0.03; p < 0.001). In summary, there was a publication bias error in the publications used for this study. The fail-safe N was calculated to evaluate the level of error and it was found that the fail-safe N was 5.0, indicating a marginal publication bias error. Subjective voice problems and the effect size of smoking by sub-functions. The effect size of subjective voice problems was analyzed by sub-function (Fig. 7). The analysis results showed that the mean effect size was significantly different only in VHI-P. VHI-P was 0.41 (95% CI = −0.

Discussion
Many previous studies have shown that smoking is a causal factor changing voice and increasing the prevalence of voice disorders 4,5,9,10 . However, there are studies indicating that smoking does not affect the voice 19 . There was heterogeneity in the results of these studies. Therefore, this study conducted a systematic literature review and www.nature.com/scientificreports www.nature.com/scientificreports/ meta-analysis in order to identify scientific basis regarding the harmful effects of smoking on voice health. This study analyzed the experimental studies published between 2000 and 2018 evaluating the effects of smoking on voice using control groups.
The results of quality evaluation analysis ranged between 4 and 7 points, showing that there was no critical issue to exclude publications except one study. The study with the lowest score evaluated the voice symptoms between smokers and non-smokers using questionnaires, and it did not satisfy the subject selection criteria, comparison possibility, and results criteria 15 . However, it was not excluded from this study because it showed the voice symptoms according to gender and smoking.
It was found that Dr. Speech, Visi-pitch, DSI, Praat, and CSL were the tools used to measure the acoustic characteristics. VHI was used to evaluate the subjective voice. Laryngeal stroboscopy was used to check the symptoms in the larynx. Among them, VHI is a frequently used tool for evaluating voice.
The effect sizes of the nine publications included in this study were analyzed and it was found that smoking only affected the pitch and MPT of voice. Smoking may cause voice disorders by affecting voice directly 23,26,27 . It is related to the organic voice disorders such as laryngeal cancer, rather than functional voice disorders due to the abuse and misuse of voice 4,26,28 . Excessive smoking can irritate the vocal cords and dry the vocal cord mucosa. It can results in inflammation on the vocal cords. It can cause coughing, sputum, and vocal cord feeling of irritation. It may lead to changes in voice 29 . Pinar et al. 17 argued that smoking would increase the weight of the vocal cords, which in turn decrease the fundamental frequency. Moreover, smoking weakens pulmonary function, which can reduce the mean MPT time. Awan 23 examined the relationship between DSI and smoking using 30 women aged between 18 and 24 years, and reported that smoking was strongly related to fundamental frequency decrease and shortened MPT. It is difficult to exclude the possibility that the difference in the algorithms of the evaluations tools could affect the results of this study, because the nine studies included in this study used different acoustic evaluation tools such as CSL, Dr. Speech, Visi pitch, and Praat. It means that future meta-studies should include more studies because the interpretation of the results from these nine studies could be limited.
The effect size was compared by dividing VHI into sub-indices, and the results showed that there were significant differences only in physical functions. VHI is a self-reporting assessment tool that evaluates the voice subjectively. Therefore, it has the potential to underestimate or overestimate the voice problem 25,30 . Also, perceptions about the impact of smoking on their voice varies 14 . Nevertheless, Tafiadis et al. 14 reported that acoustic parameters were correlated with VHI scores, which was similar with the results of this study. Particularly, they showed that NHR and jitter variables were significantly correlated with VHI function and VHI emotional.
The importance of this study was that this study established a scientific foundation to identify the relationship between voice and smoking by summarizing the results of studies evaluating the effects of smoking on voice. The limitations of this study are as follows. First, it is possible that this study did not include studies published  www.nature.com/scientificreports www.nature.com/scientificreports/ in languages other than English and Korean (e.g., Spanish and German), although this study included diverse studies through three international research databases and five domestic research databases. Second, the generalization of the results should be careful because there were less than ten studies in this meta-analysis. Future studies need to conduct more in-depth analyses including more studies. Third, the tools for speech analysis of the studies