Browse the corpus

Effective spoken communication is central to military work. Military leadership doctrine emphasizes communication, command and coordination as core elements of effective military functioning (). Commands, training instructions and operational coordination often need to be delivered in noisy, physically demanding and time-critical environments. In such settings, the voice is not only a biological function but also an occupational tool. Impaired voice production may therefore affect communication reliability, training performance, leadership, safety and operational readiness (–).

Military personnel may be exposed to several work-related risk factors for voice disorders. High vocal load is common in roles requiring sustained or forceful voice use, such as drill instruction, command, physical training and communication in high-noise environments (–). Ambient noise from weapons, vehicles, aircraft and machinery can increase vocal effort and may contribute to vocal fatigue, phonotrauma and persistent dysphonia (–). In addition, military training and deployment may involve exposure to airborne irritants, dust, smoke and chemical agents, which can affect the larynx and upper airway (, –). Neurological injury, including traumatic brain injury, may further impair communication and vocal control (). Combat-related respiratory and laryngeal disorders may also affect upper-airway function and voice production ().

Although occupational voice disorders have been studied extensively in civilian voice-demanding professions, particularly teaching, their burden in military populations remains poorly quantified (, , ). Military personnel differ from civilian workers because of selection effects, fitness requirements, deployment-related exposures and the potential underreporting of symptoms that may be perceived as minor or non-operational. The “healthy warrior effect” may also lead to lower apparent morbidity in active-duty cohorts, while veteran and exposure-specific cohorts may capture longer-term consequences of service-related hazards (, ). From a public-health and workforce-health perspective, the relevance of these disorders extends beyond individual symptoms. In military populations, communication capacity may be essential to instruction, coordination, safety, and operational readiness. However, the extent to which voice and related laryngeal outcomes are captured by existing occupational-health surveillance remains unclear, and the military roles or exposure contexts that may warrant targeted prevention and early referral have not been systematically identified. Synthesizing the available evidence by occupational and environmental risk group is therefore important for informing surveillance priorities and identifying gaps in preventive military health practice.

Reported prevalence estimates vary widely across military populations, ranging from low rates in broad military cohorts to substantially higher rates in personnel with intensive vocal demands or documented chemical exposure (–, –, , ). Differences in study populations, case definitions, and assessment methods have limited a clear understanding of the occupational burden of voice disorders in military settings. Because the military literature uses overlapping terms and different assessment methods, “voice and related laryngeal disorders” was used in this review as an umbrella term for the synthesis of functionally relevant outcomes rather than as a single clinical diagnosis.

To our knowledge, no systematic review and meta-analysis has synthesized the prevalence of voice and related laryngeal disorders in military personnel and veterans. The primary aim of this review was to estimate prevalence separately in general military populations, high-vocal-demand personnel and chemically exposed cohorts. Secondary aims were to explore differences between these subgroups, examine the influence of outcome definition and case ascertainment, and synthesize findings from clinically referred or symptom-selected military populations. The review was intended to characterize the available evidence and identify priorities for future occupational-health research rather than to establish causal effects or the effectiveness of screening or preventive interventions.

This systematic review and meta-analysis was conducted in accordance with the PRISMA 2020 statement (). The review was registered in PROSPERO on 19 March 2025 (CRD420250640753; https://www.crd.york.ac.uk/PROSPERO/view/CRD420250640753 ). Registration occurred after completion of the formal literature searches and study screening but before data extraction, risk-of-bias assessment and quantitative synthesis. The primary review question, population, outcome domain and core eligibility criteria were established before the literature search. Nevertheless, because registration occurred after study identification and screening, it did not provide fully prospective documentation of these stages. The timing of registration and its implications for methodological transparency are acknowledged as a limitation of the review. No separate full protocol document was prepared.

The final synthesis differed from the registered analysis plan in several respects because of the limited number and substantial clinical diversity of the available studies. The registered plan anticipated separate analyses according to diagnostic and risk categories, including vocal cord dysfunction, dysphonia, vocal strain and traumatic brain injury. However, the available data were insufficient for reliable diagnosis-specific pooling, and several studies reported overlapping voice and laryngeal outcomes. The quantitative synthesis was therefore organized according to occupational and environmental risk context: general military populations, high-vocal-demand personnel and chemically exposed groups. Planned trim-and-fill and Bayesian analyses were not performed because the small number of studies within each subgroup would not have supported reliable interpretation. These changes did not alter the primary review question or the core population and outcome eligibility criteria. Two reviewers independently conducted study screening, data extraction and risk-of-bias assessment. Disagreements were resolved by discussion; if necessary, a third reviewer was consulted.

We searched PubMed and Embase on 6 September 2024, Scopus on 11 September 2024, Web of Science on 12 September 2024, ClinicalTrials.gov on 13 September 2024, and Google Scholar on 19 September 2024. Reference lists of eligible articles and relevant reviews were also hand-searched. Randomized trials were eligible if they reported relevant baseline prevalence data, but no eligible randomized trials were identified. Complete database-specific search strategies, exact search dates, applied filters, and numbers of records retrieved are provided in . Records were managed in EndNote. After deduplication, two reviewers independently screened titles and abstracts, followed by independent full-text assessment of potentially eligible articles. Reasons for full-text exclusion were recorded. No formal statistic of inter-reviewer agreement, such as Cohen's kappa, was calculated. Disagreements were resolved through discussion and, when necessary, consultation with a third reviewer. The study-selection process is shown in the PRISMA flow diagram ().

Two reviewers independently extracted study characteristics, including author, year, design, country or military setting, population type, active-duty or veteran status, sample size, number of voice-disorder cases, prevalence estimate, diagnostic or screening method, occupational or environmental risk factors, and eligibility for quantitative synthesis. Extracted exposure categories included high vocal demand, chemical or inhalational exposure and traumatic brain injury where reported. Disagreements were resolved by discussion; if necessary, a third reviewer was consulted.

Risk of bias was assessed independently by two reviewers using tools selected according to study design. Cohort studies were evaluated using the Newcastle–Ottawa Scale (), which assesses selection, comparability and outcome ascertainment; scores of 7–9, 5–6 and 4 or fewer were interpreted as low, moderate and high risk of bias, respectively. Cross-sectional and prevalence studies were evaluated using the Joanna Briggs Institute critical appraisal checklist for studies reporting prevalence data (), and retrospective case series were evaluated using the JBI checklist for case series (). JBI items were rated as Yes, No, Unclear or Not applicable, and overall judgements were based on the number and seriousness of identified limitations rather than on an unvalidated numerical summary score. The certainty of evidence for the three subgroup-specific pooled prevalence estimates—general military populations, high-vocal-demand personnel and chemically exposed cohorts—was assessed using the GRADE framework (). Two reviewers independently evaluated risk of bias, inconsistency, indirectness, imprecision and publication bias. Imprecision was judged serious when the width of the 95% confidence interval exceeded 10 percentage points and very serious when it exceeded 20 percentage points or encompassed substantially different prevalence levels. Because the evidence consisted of observational prevalence studies, certainty began at Low and was downgraded for serious or very serious concerns in the relevant domains. Publication bias could not be assessed reliably because each subgroup contained fewer than 10 studies. Disagreements were resolved through discussion or, when necessary, consultation with a third reviewer. Study-level risk-of-bias assessments and GRADE evidence profiles are provided in –.

The primary outcome was the prevalence proportion of voice or related laryngeal disorders, calculated as the number of affected individuals divided by the corresponding study population. The outcome definition reported by each study was retained. Accordingly, administrative or ICD-based diagnoses, self-reported symptoms, screening-questionnaire results and clinically assessed disorders were not treated as clinically identical outcomes. Differences according to the method of outcome identification were examined in sensitivity analyses where sufficient data were available. Because the included military populations differed substantially in occupational demands, environmental exposures and expected baseline risk, no overall pooled prevalence was calculated across all datasets. Quantitative analyses were instead conducted within three occupationally and environmentally defined subgroups: General military populations: broad active-duty, recruit or veteran cohorts that were not selected on the basis of high occupational vocal demand, documented chemical exposure or a pre-existing voice or laryngeal disorder. High-vocal-demand personnel: military personnel in roles involving sustained, intensive or forceful voice use, including drill instructors, trainers and comparable instructional personnel.

Chemically exposed populations: military personnel or veterans with documented exposure to chemical or inhalational irritants potentially affecting the larynx or upper airway. Active-duty personnel, recruits and veterans were not combined in a single analysis across all exposure contexts. All of these service-status groups were eligible because military service may involve occupational and environmental hazards relevant to upper-airway health (). They were assigned to subgroups according to the occupational or environmental context represented by each dataset. Service status was not used as a separate primary subgroup because too few datasets were available to support reliable stratified pooling. Nevertheless, because veterans may differ from active-duty personnel in exposure history, healthcare access and baseline risk, an additional sensitivity analysis was conducted after excluding the veteran dataset from the general military subgroup. When a study reported results for two non-overlapping military populations with different occupational exposure profiles, each population was treated as an independent dataset and assigned to the appropriate subgroup. This applied to the general and high-vocal-demand populations reported by Sapir et al. and Perry et al.

Sixteen studies met the eligibility criteria and were included in the systematic review. The quantitative synthesis included 11 independent datasets from nine studies, comprising 1,617,774 military-affiliated participants. Seven studies were retained for narrative synthesis and were not included in the prevalence meta-analysis. Six examined clinically referred or symptom-selected populations whose denominators did not represent unselected military source populations. The remaining study was a reanalysis of a cohort already represented in the quantitative synthesis and was not pooled separately to avoid double-counting participants. The study-selection process is presented in . The 11 datasets included in the quantitative synthesis were classified according to occupational or environmental context. The general military subgroup comprised five datasets from broad military, recruit or veteran populations that were not selected for high occupational vocal demand or chemical exposure (, , , , ). The high-vocal-demand subgroup comprised three datasets involving drill instructors, trainers or other personnel in roles requiring sustained or forceful voice use (, , ). The chemically exposed subgroup comprised three datasets from personnel or veterans with documented exposure to chemical or inhalational irritants, including sulfur mustard, tear gas or pepper spray (–).

Two studies contributed more than one non-overlapping dataset. Perry et al. reported separate results for general military personnel and personnel classified as having high occupational vocal demand (), whereas Sapir et al. reported separate results for female army recruits and instructors (). These populations were treated as independent analytical units and assigned to the appropriate subgroups. A subsequent analysis of the Sapir cohort was not included in the quantitative synthesis to avoid double-counting participants (). Definitions of voice and related laryngeal outcomes varied considerably across the included studies. Studies of broad military populations included administrative or ICD-based diagnoses, self-reported symptoms and clinical assessments. High-vocal-demand studies primarily used self-reported symptoms or questionnaire-based assessments, whereas chemically exposed studies used clinical evaluation or screening for persistent laryngeal symptoms. These outcomes were retained according to the definitions used in the original studies and were not interpreted as representing a single clinically homogeneous disorder.

Traumatic brain injury was identified in the registered plan as a potential risk-based subgroup. However, no included study provided extractable prevalence data specifically for voice disorders among personnel with traumatic brain injury. TBI-related findings were therefore considered narratively and were not subjected to separate quantitative pooling. Study characteristics and eligibility for quantitative synthesis are summarized in , with additional details provided in .

Risk of bias was assessed for all 16 included studies using tools selected according to study design. Four cohort studies were evaluated using the Newcastle–Ottawa Scale and received scores of 5–6, corresponding to moderate risk of bias. Of the nine cross-sectional or prevalence studies assessed using the JBI checklist for prevalence studies, eight were rated at moderate risk and one at high risk. The principal concerns were non-random or selected sampling, inadequate sample size, limited control for potential confounding and differences in the methods used to identify or measure voice and laryngeal outcomes. Three retrospective case series were assessed using the JBI checklist for case series; one was rated at low risk and two at moderate risk. Overall, one study was rated at low risk of bias, 14 at moderate risk and one at high risk. No study was excluded solely because of its risk-of-bias assessment.

The certainty of evidence was rated as Very Low for all three subgroup-specific pooled prevalence estimates. Observational prevalence evidence initially started at Low certainty and was downgraded for serious or very serious concerns relating to risk of bias, inconsistency and imprecision. The very wide confidence and prediction intervals, substantial unexplained heterogeneity and small number of studies in the exposure-defined subgroups were the principal reasons for downgrading. Publication bias could not be assessed reliably because each subgroup contained fewer than 10 datasets. Study-level risk-of-bias assessments and the complete GRADE evidence profiles are provided in –.

Subgroup-specific prevalence estimates were calculated using random-effects models with restricted maximum likelihood estimation and Hartung–Knapp adjustment (; ). No overall prevalence was calculated across the three occupational and environmental subgroups. The general military subgroup comprised five datasets with 6,139 cases among 1,613,921 participants. The pooled prevalence was 3.3% (95% CI 0.2–37.5%), with substantial heterogeneity (I 2 = 99.8%; τ 2 = 5.3114). The 95% prediction interval ranged from 0.001% to 99.1%. Study-specific prevalence estimates ranged from 0.35% in Norman et al. () to 31.1% among the recruits reported by Sapir et al. (). The high-vocal-demand subgroup comprised three datasets with 483 cases among 3,708 participants. The pooled prevalence was 40.0% (95% CI 0.8–98.2%), with substantial heterogeneity (I 2 = 99.5%; τ 2 = 3.1315). The 95% prediction interval ranged from approximately 0% to 100%. Study-specific prevalence estimates ranged from 9.1% in the high-vocal-demand population reported by Perry et al. () to 76.9% among the army instructors reported by Sapir et al. ().

The chemically exposed subgroup comprised three datasets with 105 cases among 145 participants. The pooled prevalence was 72.8% (95% CI 43.3–90.4%), with moderate heterogeneity (I 2 = 57.5%; τ 2 = 0.1448). The 95% prediction interval ranged from 0.6% to 99.9%. Study-specific prevalence estimates ranged from 61.5% in Shah et al. () to 79.1% in Balali-Mood et al. (). Although the subgroup point estimates differed markedly, their confidence intervals and prediction intervals were wide and overlapped substantially. The pooled values therefore summarize the center of highly variable study-specific estimates and should not be interpreted as prevalence values expected to apply uniformly across military settings.

An exploratory mixed-effects meta-regression was conducted across the 11 datasets using subgroup membership as a categorical moderator. The omnibus test of the subgroup moderator was statistically significant [F(, ) = 5.61, p = 0.0300], indicating an overall association between subgroup membership and logit-transformed prevalence. Subgroup membership accounted for an estimated 48.1% of the between-study heterogeneity (R 2 = 48.1%); however, residual heterogeneity remained extremely high [residual I 2 = 99.9%; Q?() = 2,686.94, p < 0.0001]. In pairwise comparisons, the estimated logit prevalence was higher in the high-vocal-demand subgroup than in the general military subgroup, but the difference did not reach statistical significance after the small-sample adjustment (logit difference 2.9751, 95% CI −0.1923 to 6.1426; p = 0.0622). The chemically exposed subgroup had a higher estimated logit prevalence than the general military subgroup (logit difference 4.3987, 95% CI 1.2066 to 7.5908; p = 0.0130). The difference between the chemically exposed and high-vocal-demand subgroups was not statistically significant (logit difference 1.4236, 95% CI −2.1399 to 4.9870; p = 0.3839). Pairwise comparisons are presented in .

These results should be considered exploratory because the meta-regression included only 11 datasets, with three datasets in each of the two exposure-defined subgroups. Statistical power was therefore limited, and the study-level comparisons may also have been affected by differences in population selection and outcome definition.

The generalized linear mixed model sensitivity analyses produced pooled point estimates similar to those obtained using the primary REML–Hartung–Knapp models. The GLMM estimate was 3.3% (95% CI 0.6–17.0%) for the general military subgroup, 40.0% (95% CI 11.4–77.6%) for the high-vocal-demand subgroup and 72.9% (95% CI 62.6–81.2%) for the chemically exposed subgroup. The similarity of the point estimates indicated that the central estimates were not materially affected by the choice between the two statistical models, although the widths of their confidence intervals differed. Sensitivity analyses according to outcome definition demonstrated substantial differences in prevalence between methods of outcome identification. Within the general military subgroup, the two datasets using ICD or administrative diagnoses produced a pooled prevalence of 0.36% (95% CI 0.22–0.60%; I 2 = 81.6%). The two datasets based on self-reported symptoms produced a pooled prevalence of 10.2% (95% CI approximately 0–100%; I 2 = 99.1%). The remaining clinically assessed dataset, Forbes et al., reported a prevalence of 22.2% (95% binomial CI 14.8–31.2%). All three high-vocal-demand datasets used self-reported symptoms or questionnaire-based assessments and therefore corresponded to the primary estimate of 40.0% (95% CI 0.8–98.2%; I 2 = 99.5%).

Within the chemically exposed subgroup, the two studies using clinical dysphonia assessment produced a pooled prevalence of 78.5% (95% CI 12.9–98.9%; I 2 = 0%). The remaining study, which used the Laryngeal Hypersensitivity Questionnaire, reported a prevalence of 61.5% (95% binomial CI 47.0–74.7%). Because the outcome-definition categories contained only one or two studies, these analyses were descriptive and should not be interpreted as providing stable alternative pooled estimates. Exclusion of Forbes et al., whose participants had been recruited because of post-deployment dyspnoea, reduced the pooled prevalence in the general military subgroup from 3.3% to 2.0% (95% CI 0.05–43.5%; 95% prediction interval approximately 0–99.9%; I 2 = 99.8%). Exclusion of the veteran dataset reported by Norman et al. increased the pooled prevalence to 5.7% (95% CI 0.2–67.6%; 95% prediction interval approximately 0–100%; I 2 = 99.9%). Neither exclusion resolved the substantial heterogeneity or imprecision.

In leave-one-out analyses, pooled prevalence estimates ranged from 1.75% to 5.71% in the general military subgroup, from 23.1% to 63.4% in the high-vocal-demand subgroup and from 70.0% to 78.5% in the chemically exposed subgroup. The high-vocal-demand results were particularly sensitive to the exclusion of individual datasets, whereas the chemically exposed point estimates were comparatively stable. Confidence intervals remained wide, especially in analyses containing only two studies. Complete leave-one-out results are provided in and . The model-comparison, outcome-definition and study-exclusion sensitivity analyses are presented in . Formal tests of funnel-plot asymmetry and small-study effects were not performed because each subgroup contained fewer than 10 datasets. Publication bias could therefore not be assessed reliably.

Seven studies were synthesized narratively and were not included in the prevalence meta-analysis. Six studies examined clinically referred or symptom-selected military populations (, , –). Because inclusion in these studies was conditional on the presence of voice, respiratory or upper-airway symptoms, their denominators did not represent unselected military source populations and could not provide population-based prevalence estimates. The remaining study, Sapir et al. (), was a reanalysis of a cohort already represented by Sapir et al. () and was not pooled separately to avoid double-counting participants.