Multiple-Community-Based Epidemiological Study of Stuttering among 3-Year-Old Children in Japan Open Access

Developmental stuttering, a speech disorder that typically emerges in early childhood, involves sound or syllable repetitions, sound elongations, fixed postures of the speech mechanism, as well as associated body movements. One research focus of developmental stuttering has been the basic epidemiological information regarding its various aspects such as onset, incidence, prevalence, and prognosis [1]. However, some of the reported results were inconsistent, likely due to differences in subjects’ age, observed time window, sample size, survey method, and criteria for identifying stuttering events [2‒16].

The reported prevalence of stuttering varied depending on the age-group sampled or survey method. In Denmark, Månsson [5] reported that 51 children were identified as exhibiting stuttering out of 1,021 children aged 3 years (prevalence: 4.99%). The stuttering cases were identified by the speech and hearing consultants who were trained in stuttering diagnosis, using an interaction between the speech and hearing consultants and the child, as well as a parent interview. Okalidou and Kampanaros [6] surveyed 1,113 children aged 4–5 years in Greece using teachers’ responses to a checklist and reported a prevalence of 2.20%. Proctor et al. [11] reported that among 3,164 children (2,223 African-American and 941 European-American preschoolers) aged 2–5 years, the total prevalence was 2.52%, with no statistically significant group differences, using a 3-pronged approach that included individual screening by clinicians, teacher identification, and parent identification. On the other hand, McLeod and Harrison [12] reported a higher prevalence (5.60%) in a larger study of 4,983 children in Australia aged 4–5 years who were surveyed based on parental interviews and questionnaires.

Other studies have reported lower prevalences [7, 9, 10]. In Australia, Craig et al. [7] reported prevalence according to age-group, covering subjects aged 2–95, using interviews and confirmation of speech over the telephone with strict criteria for defining stuttering. The prevalence was highest in preschoolers and early school grades, at 1.40% for children aged 2–5 years and 1.44% for those aged 6–10 years. For the age-groups older than 10 years, the prevalence was less than 1.0%. Also, in Australia, McKinnon et al. [10] surveyed more than 10,000 children aged 5–13 years using teacher identification and school speech-language pathologists’ records and found a prevalence of only 0.33%. Using questionnaires for teachers in Belgium, van Borsel et al. [9] surveyed 21,027 pupils aged 6–20 years. Among all those surveyed, 121 were found to have stuttering, with an overall prevalence of 0.58%. Prevalence decreased systematically with age, with reported values from 0.78% for ages 6–10 years, 0.53% for ages 11–15 years, and 0.27% for ages 16–20 years. More recently, a study by the USA Center for Disease Control and Prevention covered a wider age range of young people, with a much larger sample [13]. They analyzed responses obtained from parents or legal guardians of 119,367 children, aged 3–17 years, who were asked if their child had any of several developmental conditions, including stuttering during the past 12 months. The overall stuttering prevalence was 1.60%; it was 1.99% for those between 3 and 10 years of age and 1.15% for those between 11 and 17 years.

There have only been a few epidemiological studies on stuttering in Japan. In the last century, Ozawa [2] reported that 0.98% of students showed stuttering symptoms, based on a survey of 7,654 elementary and junior high school students. This survey involved two stages of screening by school teachers and direct evaluation by a dentist. The prevalence is comparable to that reported in Europe and the USA (around 1.0%) in the 19th and 20th centuries [17, 18]. In this century, Ooki [8] surveyed 1,896 Japanese twin pairs aged between 3 and 15 years, using a questionnaire asking their mothers to identify the frequency of stuttering from four categories: “often,” “sometimes,” “never,” or “do not know.” This prevalence (percentage of people who answered “often” and “sometimes”) was reported to be 6.7% and 3.6% among males and females, respectively. Shimada et al. [16] reported a stuttering prevalence of 1.41% by directly assessing the speech of 2,274 3-year-old children for any stuttering in speech in an approximately 10–20 min session. However, the identification of the presence or absence of stuttering by only a time-limited direct observation alone might have led to underdiagnosis as the reported prevalence is almost at the lowest end of the range (1.40–5.6%) among recent epidemiological studies with young children up to 5 years [5‒7, 11, 12]. Stuttering prevalence has also been reported as 1–2% in West Asia (e.g., Egypt and Iran) or Africa [4, 14, 15, 19] other than Western countries and Japan, however, with higher ages of the participants. The main studies published after 1960 that reported the prevalence and/or incidence of childhood stuttering are summarized in Table 1.

The differences in prevalence in these studies can be primarily explained by the natural course of early childhood stuttering. Stuttering most commonly occurs between the ages of 2–4 and naturally recovers within a few years after onset in the majority of the cases [20]. As a result, its prevalence is highest between the ages of 3 and 5 and declines thereafter. Furthermore, young children who stutter may experience fluctuations between typical disfluency, and mild stuttering, or non-stuttering over several weeks or months [21]. Stuttering symptoms can also vary depending on speaking situations and emotional states. Therefore, it is expected that differences will arise between assessments of symptoms made on a single occasion based on strict criteria by the specialists such as SLPs and observations made by parents and teachers who spend their daily lives together over a certain period before the survey time. A higher prevalence (4% or more) was reported in the previous studies involving participants aged 3–5 years and incorporating the assessments conducted within a certain period of time by guardians in daily contact with the child [5, 12]. Similar values of more than 4% were reported in the study covering a wider age range (3–15 years) when a child was determined as having “stuttering” if the guardian gave a response other than “never” (i.e., “often” or “sometimes”) on the questionnaire [8]. On the other hand, a lower prevalence (2–3%) was reported for children aged 3–5 years when the assessments were conducted by teachers [6, 11]. Even lower prevalence values (less than 2%) were reported in studies involving subjects in late childhood, adolescence, and beyond [2, 4, 9, 10, 13‒15, 19], as well as in those where the assessment was performed only at a single point in time by the specialists [7, 16]. Based on these previous studies, a recent review described that it is clear that the prevalence under age 6 is considerably higher than that in later periods in life [1]. Attention should be given to the proper interpretation of the age of the subjects, the screening and assessment methods (e.g., interviews, questionnaires, and direct assessments), the situations in which the surveys were conducted, and the assessors involved (such as speech-language pathologists, school teachers, and/or parents).

The incidence of stuttering can also vary greatly depending on the age of the children sampled. As children typically begin to stutter around the age of 2–4 years and the majority of them recover within a few years [20], it is crucial that surveys of the cumulative incidence of stuttering included children between the ages of 2–7. Prospective, longitudinal tracking of a large population over several years allows for the direct identification and verification of all new cases and has been considered the most accurate method for estimating the incidence of stuttering [1]. In the 20th century, Andrews and Harris [3] tracked approximately 1,000 children from birth to 16 years of age and reported a cumulative incidence of 4.9% and a recovery rate of 79%. Since the turn of the century, several longitudinal studies on early childhood stuttering have demonstrated higher rates of both onset and recovery of stuttering than those in the previous century, with the reported cumulative incidence ranging from 5.19 to 17.7% [5, 22‒25]. On the other hand, a recent study estimated the incidence or risk of stuttering as 2.8% (95% confidence interval [CI] 1.6–4.0%) in young children (2–5 years) based on retrospective data [7], obtained through a telephone survey in which households were asked if any person living in their household stuttered or had ever stuttered. This estimated incidence is considerably smaller than that reported in the aforementioned prospective cohort studies [5, 22‒25].

Although there is strong evidence that stuttering has a genetic basis [8, 26, 27], identifying the factors associated with the onset of stuttering, whether of genetic origin or otherwise, helps us understand the nature of stuttering. In the 20th century, two prospective cohort studies reported that more boys stuttered than girls at a ratio of 1.65:1 [5] and 1.58:1 [24]. Additionally, those who stuttered were more likely to come from a lower socioeconomic status and had delayed acquisition of words and phrases compared to children who did not stutter [28]. In a 21st-century cohort study, male sex, twin birth, mother’s educational level, and having higher communication skills and a greater vocabulary at the age of 2 years were identified as risk factors associated with the onset of stuttering [24, 25]. Delayed acquisition of words and sentences [28] and having a large vocabulary [24, 25] have both been reported as risk factors of stuttering, although they appear somewhat contradictory. Another study examining children’s development of receptive and expressive language concluded that those who stutter scored lower than non-stuttering peers on tests of expressive grammatical abilities. However, those who stutter did not differ from non-stuttering peers on tests of receptive vocabulary [29]. As the association between a child’s language skills and stuttering is inconsistent among studies, it remains a topic of ongoing debate.

The association between stuttering and various comorbid conditions in childhood has also been reported. It has been noted that a significant percentage of children who stutter exhibit concomitant speech and/or language disorders, in addition to their stuttering [30, 31]. Arndt and Healy [30] revealed that out of 467 children (aged 3–20 years), 205 (44%) had phonological disorders (32%), language disorders (35%), and phonological and language disorders (33%) as reported by 241 SLPs. Blood et al. [31] investigated 1,184 SLPs and found that out of 2,628 children who stuttered, 62.8% had other co-occurring speech disorders, language disorders, or non-speech-language disorders. Regarding non-speech-language disorders, many studies have shown that certain aspects of attention differ in children who stutter compared to children who do not stutter [32], and many children who stutter demonstrate an increased incidence of attention-deficit hyperactivity disorder (ADHD) characteristics [33, 34]. A recent population-based study that investigated the relationship between executive function, stuttering, and comorbidity reported that the majority (60.32%) of children who stuttered had at least one other comorbid condition in addition to stuttering. Among the various comorbid conditions, ADHD, asthma, and autism were the most prevalent [35]. Heart conditions, cerebral palsy, and Down’s syndrome have also been reported as comorbid conditions. Children with low mental abilities, including Down’s syndrome, have a higher incidence of stuttering than the general population [36]. These findings may indicate that certain common intrinsic factors may underlie these disorders or diseases and may serve as predictive or at least warning signs of the onset or persistence of stuttering.

In Japan, the establishment of an effective public health service policy for childhood stuttering is a matter of increasing demand. To make informed healthcare policy decisions, it is necessary to obtain solid epidemiological data without underdiagnosis. The present study aims to investigate the prevalence, cumulative incidence, and risk factors of early childhood stuttering in Japan. The investigation is based on a comprehensive diagnostic procedure involving several assessment methods and procedures (screening and confirming interviews, questionnaires, and direct assessment) and multiple assessors (SLHTs and parents) within a certain period of time. To collect a large amount of unbiased data from children during the time when stuttering is most likely to occur, the survey was planned to be conducted during the national health checkups in young children, which have been conducted as a national policy since 1965 under an initiative of the Ministry of Health and Welfare (currently Ministry of Health, Labour and Welfare). These health checkups are conducted at 1.5 and 3 (or 3.5) years of age. The uptake rate for these checkups is approximately 95% [37], making it a rare opportunity that could potentially allow for the most unbiased sampling. Since the onset of stuttering is most prevalent between the ages of 2 and 4 years [20], the health checkups of 3-year-old children were chosen for the present survey. To avoid excluding children who may have developed stuttering and subsequently recovered before reaching the age of 3 years, we calculated the cumulative incidence (risk of stuttering) using the combined method reported by Craig et al. [7], which integrates retrospective data from before the age of 3 years with the data of current stuttering status at the age of 3 years.

Once a child reaches the age of 3 or 3.5 years in Japan, they undergo a national health checkup at public healthcare centers within the local community in which they reside. According to a 2021 report by the Ministry of Health, Labour and Welfare [37], the checkup participation rates at 3 and 3.5 years of age (hereinafter collectively referred to as “3-year-old health checkup”) ranged from 94.5 to 95.9% during 2017–2021. The municipal healthcare centers in five distant prefectures (Fukuoka, Ishikawa, Ibaraki, Kanagawa, and Tokushima), where the affiliated institutions of five of the authors were located, were requested to cooperate with our survey. In these prefectures, 15 regional (branch) healthcare centers ultimately agreed to cooperate. The survey was conducted between July 2016 and February 2018. It consisted of a two-step procedure: (a) screening and (b) confirmation. The municipal healthcare centers were requested to include our package of screening questionnaires (details below), consent forms, and documents containing instructions regarding our study in their standard health checkup packages. In centers that did not agree to send our research packages in advance, the packages were distributed and explained to the participants by research staff on the day of the checkup. The instruction document asked children’s guardians who were willing to participate in the study to submit a signed consent form and a completed screening questionnaire to the research staff at the respective health checkup site. Since this study was planned to conduct a regression analysis with 10 explanatory variables (risk factors) for stuttering, and since the prevalence of stuttering was estimated to be approximately 5–8% from previous studies, the overall target sample size was set at 2,000 children so that the number of participants in the small sample group (stuttering group) could be at least 10 times larger than the number of explanatory variables (i.e., at least 100 participants). In total, 2,448 research packages were distributed and 2,055 guardians submitted the consent form and the completed screening questionnaire to our staff. Out of them, participants were excluded from this study if (1) the child had not yet begun to speak or (2) the child was under 3 years old or 4 years old or older.

The screening questionnaire consisted of (1) items regarding the child’s disfluent speech characteristics and (2) items regarding the child’s development and family background information. The items about the child’s speech disfluencies included questions that assessed the type and frequency of stuttering-like disfluencies (SLDs), focusing specifically on the presence or absence of sound or part-word repetition, prolongation, and block. There were three response choices: “yes,” “no,” and “previously yes.” Regarding to repetition, the frequency was assessed by selecting “rarely,” “sometimes,” “often,” and “very often.” Additionally, three questions regarding the features related to aggravation were included: (1) unnatural tension in voice and/or increases in pitch or loudness, (2) duration of SLDs, and (3) the degree of guardian’s concern. Items about the child’s development and family background yielded data regarding previously reported risk factors for stuttering onset and/or prognosis [17, 20, 24, 25, 28, 38‒40]. These items concerned (1) sex, (2) age, (3) birth order, (4) the age at which two-word sentences were first produced (selected from 5 choices of time), (5) family history of stuttering, (6) parents’ concern about their child’s development (if “yes,” selected from 9 choices of concrete concern), (7) the child’s currently diagnosed diseases or disabilities (if “yes,” described the diagnosis name in the free-text field), (8) family’s socioeconomic status (three choices for household annual income range), and (9) the parents’ highest educational attainment (four choices on the educational background). Item (4) was set as a language development indicator because the age at which two-word sentences appeared is used as a language development indicator during children’s health checkups in Japan, and this question with semi-annual choices was considered easy for the guardians to answer. If the guardian answered “yes” to the question about family history in item (5), they were required to answer who had stuttering (choosing between “father, mother, or other(s)”). If “other(s)” was selected, they were also required to provide a specific relationship. They were also required to answer whether the family member’s stuttering was temporary or persistent. The item did not specify the range of family (immediate or extended) included in “others.” The last two questions were specifically marked as “optional” to reduce non-responders due to a possible perception that these questions were too revealing.

When the screening questionnaires were submitted at the checkup site, research staff immediately evaluated the responses. If any one of the questions regarding characteristics of stuttering in Japanese, such as stuttering-like disfluencies (sound or part-word repetition, prolongation, and block) [41, 42], was marked as positive, the guardians were asked to cooperate with a confirmatory examination to determine whether their children’s disfluencies constituted stuttering, normal disfluencies, or something else.

The confirmatory questionnaire contained a section for free-text descriptions of concrete examples of stuttering (SLDs) and eight questions about the child’s stuttering (e.g., the child’s condition when first exhibiting stuttering, the progress after stuttering onset, the child’s awareness of stuttering, and the tension in the body and/or any concomitant symptoms). The free-text description section was particularly emphasized for the purpose of determining whether the child exhibited stuttering.

A clinical evaluation was then performed by (a) direct interviews with the guardians, (b) direct assessment of the child’s speech, (c) a questionnaire, and/or (d) telephone interviews. First, to the greatest extent, the direct interviews and direct assessments were conducted at the health checkup site on the same day or in a playroom at the SLHT’s office on a different day. The guardians were asked to provide concrete examples of the child’s disfluencies, and/or the SLHT directly assessed the child’s speech through facilitating interactions, using picture books and toys, for approximately 10–30 min. If the direct interviews and assessments could not be arranged due to the child or guardian’s lack of cooperation, a confirmatory questionnaire was sent to their home to confirm concrete examples of SLDs in the child’s speech. If responses to this detailed questionnaire were obtained but it was difficult to determine from the responses due to ambiguity or lack of information, or if responses to the questionnaire were not obtained, telephone interviews with the guardians were conducted. Confirmation about past stuttering was obtained only through interviews with the guardian or questionnaire responses. Figure 1 shows the flowchart of the survey and participants.

If none of the questions regarding SLDs in the screening questionnaire received positive answers, the child was classified as “non-stuttering.” If one or more positive responses were obtained, we proceeded to the confirmatory stage to determine whether the child was currently stuttering (or had stuttered in the past) using two criteria: (1) whether the type of disfluency was SLD and (2) whether repetition occurred more than once per disfluency event if stutter-like repetition was identified. The confirmatory survey placed the utmost importance on concrete examples of the child’s disfluent speech, which were obtained through direct observation or from their guardians’ responses, presented with transcribed or orally demonstrated child’s utterances. When ambiguities arose (e.g., whether sound repetition or word repetition), we confirmed the guardians’ response through the demonstration of the SLDs by researchers and verified whether the disfluency was of a normal type (e.g., whole-word repetition) or an SLD. Notably, while single-syllable word repetition is considered stuttering in English [20], it is not in Japanese because repetition of single-syllable words (one-mora words, to be exact, such as “ki” and “te” ) is also observed in children who do not stutter and is considered one type of the normal disfluencies [42]. Regarding the “blocking” of speech, the guardians were asked to determine whether it was caused by a problem in recalling or choosing a word to say, or by involuntary stopping of articulation. If the questions about tension in the body and/or concomitant symptoms (e.g., noisy breathing) received positive answers, they were used as additional information in determining the block. Finally, the child was classified as either “stuttering,” “non-stuttering,” or “stuttering in the past” based on the comprehensive discussion by two Speech-Language-Hearing Therapists (1st and 3rd authors) who are members of Japan’s national license holders (JSLHT) and have more than 20 years of clinical experience in stuttering. A child was designated as “undetermined” if their stuttering could not be verified in either way due to a lack of sufficient decisive answers, or if the guardian was unavailable for the confirmatory step.

First, the absence of significant regional differences concerning the rate of confirmed stuttering was verified as these differences could be caused by the different socioeconomic and urbanization backgrounds of the five regions. Then the prevalence among the 3-year-old children across all regions was calculated as the percentage of participants classified as exhibiting “stuttering.” The risk for stuttering incidence was calculated as the percentage of participants with either a current or past history of stuttering relative to the total number of participants.

Logistic regression analysis was applied to calculate the odds ratios (ORs) for the relationship between predictor variables and the status of stuttering (“stuttering” and “non-stuttering”) at the time of the 3-year-old health checkup. Crude ORs from bivariable models, in which each risk factor was used as an independent predictor, as well as adjusted ORs from multivariable models, in which all risk factors were simultaneously used as predictors, were obtained. Our selected predictors were (1) sex, (2) age in months, (3) presence of sibling(s), (4) time at which two-word sentences were first produced, (5) family history of stuttering, (6) concern about the child’s development, (7) child’s diseases or disabilities, (8) household income, and (9) highest educational attainments of the mother, and (10) highest educational attainments of the father. The continuous variable “age” was classified into two categories. Since data for both 3- and 3.5-year-old children were collected at the health checkup sites, a bimodal age distribution was expected. Based on the two peaks at 37 and 42 months, as well as the average age (39.68 months), the participants were classified into two age categories: (1) ≤40 months and (2) >40 months. In the event that the logistic analysis revealed that predictor #6 or #7 showed higher odds for stuttering onset, an analysis of which concern of guardians and/or which disease or disabilities of the child was associated with the stuttering status would be conducted using Fisher’s test. Statistical analyses were performed using IBM SPSS Statistics ver. 24 software (IBM Corp., Armonk, NY, USA).

Of the 3,336 children who underwent the national health checkup at the regional health centers in 5 prefectures, 2,448 research packages were distributed, and 2,055 participants (61.6%) responded to the screening questionnaire survey. Among these, 10 (0.5%) met the two following exclusion criteria: (1) the child had no speech yet (n = 6), (2) age <3 or ≥4 years (n = 4). An additional 57 (2.8%) were excluded from the analyses for the following reasons: (1) missing answers to the screening questionnaire (n = 6), (2) no consent form (n = 5), (3) consent withdrawal after the survey (n = 9), (4) due to several surveys, the child stuttering state was finally classified as “undetermined” (n = 37). Consequently, data available for analysis pertained to 1,988 children (average age = 39.68 months, range = 36–47 months, SD = 2.73 ), with 991 (49.8%) girls, 989 (49.7%) boys, and 8 of unspecified gender (0.4%). Initially, of the 2025 valid respondents, 334 parents (16.5%) indicated that their child was exhibiting stuttering characteristics (SLDs) at the survey time, while 131 (6.5%) parents indicated that their children had previously exhibited stuttering. Of the total 465 participant pairs (children and their guardians) (23.4%), 205 were assessed through face-to-face observation of the children and direct interviews with the guardians, 104 were assessed using the confirmatory questionnaire, and 121 were assessed through a telephone interview. The remaining 35 had not responded to the confirmation survey. The positive stuttering rate, calculated excluding those who were unable to determine, for each method was as follows: 44.1% (90/204) for the face-to-face observation and interview, 31.7% (33/104) for the confirmatory questionnaire, and 45.0% (54/120) for the telephone interview. The questionnaire survey indicated the significantly lowest stuttering rate (χ² test, p < 0.01). Our procedures confirmed only 177 (99 boys, 77 girls, and 1 of unspecified gender; 38.1%) as presenting legitimate stuttering histories out of the 465 screening-positive participants. Among the remaining 288 (61.9%), 251 participants (54.0%) were judged to show or have shown normal dysfluencies, and 37 participants (7.9%) were unable to be identified. Figure 1 shows the flowchart of the survey and participants. Of the 177 children with confirmed stuttering, the age of stuttering onset was 18–23 months in 15 (8.5%) children, 24–29 months in 37 (20.9%), 30–35 months in 76 (42.9%), 36–41 months in 45 (25.4%), and 42–47 months in 4 (2.3%) children. Of the 177 children, 129 children (72.9%) were classified as currently stuttering and 48 (27.1%) as stuttering in the past. Table 2 shows the numbers of children who were non-stuttering, stuttering, and stuttering in the past in each of the five regional areas. Fisher’s exact test revealed no significant differences among prefectures in the proportion of children classified as “stuttering,” “past stuttering,” and “non-stuttering” (p = 0.67). Given the absence of regional bias in the data, the calculated prevalence and incidence of stuttering among the 3 years old children were derived for the entire group of respondents. Hence, the prevalence at the survey point was 6.5% (129/1988*100; 95% CI 5.4–7.7%) and the overall risk for stuttering (incidence) for up to, and including, the age of 3 years was estimated to be 8.9% (177/1988*100; 95% CI: 7.7–10.2%). If the prevalence and overall risk of stuttering were to be estimated using the positivity rate for each survey method, the prevalence would be 7.4% (334*0.441/1988*100; 95% CI: 6.3–8.6%) by face-to-face observation and interview, 5.3% (334*0.317/1988*100; 95% CI: 4.4–6.4%) by the confirmatory questionnaire, and 7.6% (334*0.450/1988*100; 95% CI: 6.4–8.8%) by telephone interview, while the risk of stuttering (incidences) would be 10.3% (465*0.441/1988*100; 95% CI: 9.0–11.7%) by face-to-face observation and interview, 7.4% (465*0.317/1988*100; 95% CI: 6.3–8.6%) by the confirmatory questionnaire, and 10.5% (465*0.450/1988*100; 95% CI: 9.2–11.9%) by telephone interview. In terms of the risk of stuttering, the questionnaire survey indicated the significantly lowest stuttering rate.

Table 3 displays the characteristics of the classified groups. Table 4 summarizes the results of logistic regression analyses of risk factors for stuttering, presenting both crude (bivariable) and adjusted (multivariable) ORs. In both models, three variables: (1) family history of stuttering, (2) guardians’ concern about the child’s development, and (3) child’s diseases or disabilities predicted higher chances of stuttering onset. However, the multivariable model had low predictive power, accounting for only 5.3% of the total variation in stuttering onset.

In the stuttering group, 39.4% of the guardians reported concerns about their child’s development, whereas in the non-stuttering group, only 25.5% of the guardians reported such concerns. In both groups, unbalanced eating, obsessions, and articulation/phonology problems were the three most common concerns. A significant difference in the percentage of guardians in each group who answered “yes” occurred only for the item “the child not playing with friends” (p = 0.0009; Bonferroni correction following Fisher’s exact test). The percentage for the stuttering group was 7.1% compared with only 1.2% in the non-stuttering group. Due to the moderate size of the stuttering group (127 children), the findings should be viewed with caution.

In the stuttering group, 11.7% of the guardians reported a child’s disease or disorder, as compared to 5.1% in the non-stuttering group. Such conditions included asthma, allergies, heart diseases, Kawasaki disease (KD), and autism/ADHD, among others. A significant difference in the prevalence of each disease or disability between the groups was found only for the KD. Specifically, significantly more children with a medical history of KD exhibited stuttering at the age of 3 years than those without a history of KD (p = 0.0022; Bonferroni correction following Fisher’s exact test). In the case of the typically rare KD, 3.1% in the stuttering group reported positively versus only 0.1% in the non-stuttering group.

Historically, research on early childhood stuttering heavily relied on parents’ reports of past events. Often, information was obtained long after the events under investigation had taken place. Therefore, the data exhibited reduced accuracy due to ambiguous memory and/or other factors influencing informants, mostly parents, to provide the information. For example, Darley [43], in his study of the onset of stuttering, obtained information from parents more than 4 years post-onset. Furthermore, Sheehan and Martyn [44] studied several thousand much older university students to gather incidence and prevalence information related to events that had taken place many years in the past. Compared to these past studies, the present study, like several recent ones, has the advantage of being conducted much closer to the events under study. It should be also noted that some of our participants were observed firsthand by clinicians, and the accuracy of the information provided by their guardians was thus verified. Because stuttering symptoms are not always apparent, relying solely on time-limited direct observation might have led to underdiagnosis. To prevent this underestimation, we carefully interviewed guardians for specific examples of their child’s utterance, even if we could not directly observe symptoms, and determined the presence or absence of stuttering by referring to the SLD in Japanese. As mentioned earlier, this study was conducted with significant improvements compared to the previous designs in Japan at that time. However, compared to the proposed most accurate method for collecting the epidemiological data on stuttering [20], the present study has the limitation that some of the data were collected retrospectively. In the future, more reliable epidemiological data should be obtained by conducting a longitudinal prospective cohort study that enables the direct identification of all new cases at the time of onset.

In this study, the prevalence, cumulative incidence, and risk factors for early childhood stuttering were investigated based on a comprehensive diagnostic procedure that employed combined assessment methods. The aim was to obtain solid epidemiological data in Japan. The present surveys, employing three methods, have had an impact on the results. The risk of stuttering estimated from the results of the confirmatory questionnaire survey showed significantly lower values than the other two methods. The main difference between the questionnaire survey and the other two methods was the extent to which the question-response interaction took place. Since the average guardian’s knowledge of stuttering is limited, it is expected that the intent of the questionnaire items may not be clearly understood, leading to misaligned or inadequate responses. In such cases, a direct or telephone interview could continue by asking more questions until the essential information was obtained. However, in the questionnaire survey, we had to determine whether the child exhibited stuttering based solely on misaligned or inadequate responses in some cases, resulting in a large number of participants who were unable to be determined. Thus, if researchers are conservative in making judgments in an epidemiological survey using only a questionnaire, the prevalence and cumulative incidence may be underestimated, If the values of this study were used, the risk of stuttering would be reduced to 7.4% (CI: 6.3–8.6%). In this study, because these participants who were unable to be determined were subsequently surveyed by telephone interview, without missing them, we were finally able to calculate the overall prevalence and the risk of stuttering. These values do not differ when compared to previous studies [24].

The present study, which integrates data obtained by the three methods, revealed an incidence of 8.9% and a prevalence of 6.5% among 3-year-old children in Japan. While these values are higher than those reported in several previous studies [5‒8, 11, 16], they are similar to those reported in others [12, 23‒25]. In particular, our incidence data fall between the 8.4% reported by Dworzynski et al. [23] in the UK and the 11.2% found by Reilly et al. [25] in Australia. Support for an incidence higher than 5% was provided by several studies mentioned above, including the longitudinal investigation of more than 1,600 Australian children. In that study, which began when the children were 8 months old, the incidence had reached 11.2% by age 4, based on confirmation by speech clinicians [25]. Similarly, the prevalence observed in the present study (6.5%) is close to the 5.6% reported by McLeod and Harrison [12], who determined the presence of stuttering based on interviews with and questionnaires administered to parents.

We also note that the children’s age in the current study falls within the prime period for the onset of the stuttering. Available data show that approximately 85% of childhood stuttering cases have their onset prior to 48 months of age [5, 20, 45, 46]. This also corresponded to the upper age limit of our participant children. Considering the fact that, in most cases, natural recovery is a gradual process that may last from 1 to 4 years or longer, both high incidence and high prevalence are expected, especially among 3-year-olds. Hence, surveys that cover this age bracket generally provide more valid estimates of the disorder’s incidence [1]. Additionally, the proximity of these events (onset and/or early recovery) to the time when the present study was conducted increases confidence in the validity of parents’ reports. Thus, recent studies, including the present one, covering the age range during which stuttering onset is most frequent and conducting detailed surveys, including reports from guardians near the onset to avoid memory bias, have reported high values for both prevalence and incidence [3, 5, 12, 22, 24, 25]. On the other hand, values are lower when the subjects are in late childhood [2, 4, 9, 10, 13‒15, 19], when only a single point in time is used for assessment [7, 16], or when retrospective interviews are conducted [7]. Based on these previous studies, two reviews of the literature [1, 18] concluded that the incidence of stuttering is 8–10 percent, if not higher.

Finally, the finding that 8.9% of children had begun showing symptoms of stuttering by the time of their 3-year-old health checkup, but that only 6.5% of all participants were actually stuttering at that time, suggests that not all cases required immediate treatment post-onset. This, along with similar findings by other investigators [16, 28], reinforces the necessity of exploring means for early differentiation between those children who will recover from stuttering on their own and those at risk for a prolonged course of the disorder who need early clinical intervention.

Our logistic regression analysis revealed that three variables (family history of stuttering, guardians’ concern about development, and diagnosed diseases or disabilities) predicted higher chances for children to exhibit stuttering at the age of 3 years. However, these predictions were rather weak, explaining only 5.3% of the variations in stuttering onset. This is somewhat unexpected, considering that 27.4% of the stuttering group indicated a family history of stuttering, even though this figure falls at the low end of the range (20−74%) reported by many surveys of the familial backgrounds of people who stutter, most of which fell between 30 and 60 percent [26]. It is possible that the strong stigma against stuttering in Japanese culture [47, 48] may have restrained the level of candid reports in this respect. It is reported that the Japanese tend to view stuttering as a more serious disability than wearing hearing aids or glasses, and that they perceive the cause of their stuttering as psychological, attributing their speech difficulties to a lack of effort [47]. Due to the strong stigma, family members may be inclined to conceal stuttering within the family and may be reluctant to report it even to researchers.

Logistic regression analysis also revealed that guardians’ various concerns about their child’s development, as well as a diagnosis of diseases and/or disabilities in the child, predicted higher odds of stuttering. In this context, approximately 40% of guardians in the “stuttering” group expressed concern about their child’s development, which is relatively consistent with previous studies reporting that 44.0–62.6% of young children to adolescents who stutter experience concomitant speech, language, and/or emotional difficulties [30, 31]. However, in this study, the percentages (ranging from 5.5% to 12.6%) of children in the stuttering group who reported having language development, articulation, or emotional control problems were not statistically significantly different from those (ranging from 4.7 to 7.5%) in the non-stuttering group. Only the percentage of guardians who answered “yes” to “not playing with friends” was significantly higher in the stuttering group, a finding that warrants further investigation. These concerns may also be strongly triggered by the experience of stuttering in their child. The retrospective nature of the study may have also influenced the data on possible predictive factors for stuttering after experiencing stuttering over time. However, it cannot be ruled out that they may be caused by co-occurring problems such as delayed development that precedes stuttering. The present study is unable to definitively distinguish between these two possibilities. Although significantly more children in the stuttering group exhibited diseases and/or disabilities compared to the non-stuttering group, no significant distribution bias was found between the groups regarding Down syndrome and ADHD, which have been reported to be correlated with stuttering [33, 49, 50]. On the other hand, the present study found that significantly more children with a medical history of KD presented and/or had exhibited stuttering at the age of 3 years. However, this result may be false positive because a recent study of KD in Japan reported its annual incidence to be 0.26% [51], whereas the mean estimated annual incidence of KD reported in our survey was only 0.12%. Hence, we cannot definitively claim an association between stuttering and KD. Similarly, only one out of 1,983 guardians reported that their child had Down’s syndrome or ADHD, preventing any meaningful interpretation.

The present study found no association between being a male, having advanced language skills, or maternal education and stuttering onset. This differs from some previous findings [24, 25]. A significant sex bias is known to exist among schoolchildren who stutter, with a male-to-female ratio of approximately 3:1 [18]. However, considerably smaller sex differences have been revealed in younger children when assessed soon after stuttering onset [5, 20, 24]. Among these previous studies, Yairi & Ambrose [20] reported a 2 to 1 ratio near onset, whereas Reilly et al. [24] found a male-to-female ratio of 1.58:1 in their stuttering group. In the present study, the male-to-female ratio was even smaller, 1.33:1. The reason for this low sex ratio, which caused the logistic regression analysis to reveal that the male sex was not associated with the onset of stuttering, remains unknown given the significant number of child participants. However, we note in Table 3 that a sex ratio close to the typical distribution (64.6/35.4%) is reflected in the subgroup of children who recovered from stuttering.

The relationship between stuttering onset and language development remains unclear. Reilly et al. [24] and Reilly et al. [25] reported that higher vocabulary scores at 2 years were significantly associated with the onset of stuttering. Similarly, Watkins [52] reported that the language skills of children who stutter were higher than average. In contrast, other studies have reported that various aspects of language proficiency were lower among children who stutter than among those who do not [28, 53‒55]. These inconsistent findings could be attributed to differences in the children’s ages, the specific language aspects assessed, the timing of assessment, as well as the fact that we inquired about “the age at which two-word sentences appeared” in order to assess language development easily. Finally, as the numbers of responders to the queries regarding the educational and economic status of guardians were rather small, our finding of non-significant relationships of these aspects with stuttering contributes little to the inconsistent findings of previous studies regarding the influence of the socioeconomic and educational attainment of guardians on stuttering in children [10, 13, 24, 25, 56].

As participation was voluntary, parents who observed disfluencies in their children may have been more motivated to respond to the questionnaire than those who did not. This could have resulted in an overestimation of the incidence and prevalence. Conversely, some guardians of stuttering children may have avoided answering the screening questionnaire due to the strong stigma against stuttering in certain areas of Japan [28], which might have lowered the estimated incidence and prevalence.

The current study had the risk that a part of the assessment for determining prevalence or incidence did not include direct observation of all suspected cases and relied mostly on reports from parents. Where direct observation was not possible, efforts were made to obtain solid data through interaction with guardians, asking for specific examples of stuttering symptoms. However, it cannot be claimed that assessments based solely on reports from parents, lacking specialist knowledge, are entirely free from errors.

As the determination of the past history of stuttering was partially based on retrospective information (although short-term), the data’s accuracy may have been influenced by the informants’ ambiguous memory. However, given that most stuttering onsets occur after the age of 24 months, the responses we received about the child’s disfluencies could be based on memory from just the last year or so, which is less likely to involve errors than retrospective data based on older memories. Regarding predictive factors for stuttering at the age of 3 years, considering that our data collection was not prospective, predictors, which are prone to change over time, such as guardians’ concerns, may have been affected by the presence or absence of stuttering at the time of the survey.

Since children over the age of four are not included in this survey sample, it is not possible to capture all cases of early childhood-onset stuttering. However, the health checkup involves children just before the age of 4, which represents the upper limit of the period in which stuttering occurs. Therefore, it is possible to provide a lower limit for the cumulative incidence in early childhood. We believe that the current study provides the best body of information regarding early childhood stuttering in Japan. It is a significant step toward developing knowledge about this worldwide disorder.

The present study found that the stuttering prevalence among 3-year-old children in Japan was 6.5% (95% CI: 5.4–7.7%) and that the cumulative incidence of stuttering up to the national 3-year-old health checkup was estimated to be 8.9% (95% CI: 7.7–10.2%), which is consistent with recent findings in other countries and cultures. The present study also found that a family history of stuttering was associated with the risk of stuttering at the age of 3 years, which is consistent with previous clinical findings. To establish an effective public health service, future follow-up survey is necessary to determine how many of these children will recover, what characteristics these children possess that contribute to recovery, and how long it takes for them to recover.

The authors appreciate the work of all of the staff engaged in data collection at the health checkup sites including Ms. Kumiko Takeuchi (manager of a public health center), Ms. and Mr. Yuri Sasaki, Yoko Mizuto, Rika Nakano, Yumi Yamaguchi, Akiko Ishihara, Yuko Ohara-Tanemura, Toshihito Aoki, and Miyuki Ito (SLHTs), Mr. Katsunori Nakamura (special education school teacher), and graduate and undergraduate students from the University of Tsukuba, Kitasato University, and Kanazawa University, as well as our research assistants, including Ms. Hiromi Hanaoka, Michiko Nishimura, Keiko Sako, Tomoko Inoue, Makiko Nishida, Chiharu Shimizu, Yukiko Kaneko, and Takako Doira, who were involved in data entry. We are also grateful to Professor Yoshikazu Nakamura of the Jichi Medical University of Public Health for fruitful discussions that helped promote this cohort study, and to Professor Ehud Yairi for his constructive suggestions. We also thank all the participating children and their guardians.

This study was approved by the Human Research Ethics Committees of the institutions with which the first to fifth authors were affiliated. These Ethics Committees were: (1) the National Rehabilitation Center for Persons with Disabilities Review Board, (2) Ethics Committee, Faculty of Human Sciences, University of Tsukuba, (3) Ethics Review Board of the School of Allied Health Sciences, Kitasato University, (4) Kyushu University Clinical Ethics Review Committee, and (5) the School of Teacher Education Ethics Committee, the Institute of Human and Social Sciences, Kanazawa University. The ethics approval reference numbers were: (1) 28-05, (2) 筑28-141, (3) 2016-032, (4) 28-129, and (5) 2016-6, respectively. Written informed consent was obtained from all parents/guardians in accordance with the guidelines established by the committees.

The authors have no conflict of interest to declare.

This study was performed by a grant-in-aid from Japan Agency for Medical Research and Development (AMED) under Grant No. JP18dk0310066. This study was partially supported by JSPS KAKENHI Grant Number JP23K02583 (Naomi Sakai).

Naomi Sakai: conceiving and designing the study, obtaining ethics approval, analyzing the data, interpreting the data, and writing the manuscript. Shoko Miyamoto, Yuki Hara, Yoshikazu Kikuchi, and Hiroaki Kobayashi: designing the study, obtaining ethics approval, and collecting the data of survey. Takaaki Takeyama and Jiro Udaka: collecting the data of survey. Daisuke Sudo: analyzing the data and interpreting the data. Koichi Mori: conceiving and designing the study, AMED grant application as a chief investigator, interpreting the data, and final approval of the version to be submitted. All of the authors discussed the results and commented on this paper.

The data that support the findings of this study are openly available in “figshare” at 10.6084/m9.figshare.25858480. Further inquiries can be directed to the corresponding author.