The Association Between Atypical Speech Development and Adolescent Self-Harm

Abstract

Background:

Adolescent self-harm is a major public health issue internationally. Various factors associated with adolescent self-harm have been identified, including being bullied and experiencing mental health problems. Stuttering and speech sound disorder are associated with both of these factors. It was hypothesized that both stuttering and speech sound disorder would be associated with self-harm. This is the first study to explore the relationship between communication disorders and adolescent self-harm.

Method:

Secondary analysis of a large, longitudinal, prospective, community sample, the Avon Longitudinal Study of Parents and Children, was carried out. Clinicians identified children who stuttered or exhibited speech sound disorder at the age of 8 years. When the cohort members were 16 years old, they were asked to complete a questionnaire about self-harm. Multinomial logistic regression was used to examine the associations between stuttering and speech sound disorder and the self-harm outcomes, adjusting for other relevant factors.

Results:

Of 3,824 participants with data for both speech status and self-harm, 94 (2.5%; 95% confidence interval [CI; 2.0, 3.0]) stuttered at 8 years of age and 127 (3.3%; 95% CI [2.8, 3.9]) displayed speech sound disorder. Speech sound disorder at the age of 8 years was associated with self-harm with suicidal intent in both unadjusted and adjusted models. Differences between the adjusted and unadjusted models were small, suggesting that speech sound disorder is largely an independent risk factor for self-harm with suicidal intent. Stuttering at the age of 8 years was not associated with adolescent self-harm, and there was no association between speech sound disorder and self-harm without suicidal intent.

Conclusion:

Compared with individuals without speech sound disorder, adolescents with speech sound disorder at the age of 8 years have twice the risk of reporting self-harm with suicidal intent, even when other important predictors are taken into account.

Supplemental Material:

https://doi.org/10.23641/asha.22573030

Self-harm (deliberate self-injury via methods such as cutting, burning, or poisoning) is a serious public health issue (World Health Organization, 2016). As well as being an inherently injurious act, self-harm is of particular concern because of its strong association with attempted and completed suicide. Prevalence of self-harm without suicidal intent is worryingly high across all age groups, but evidence suggests it is higher in young people. A meta-analysis of 128 studies of self-harm without suicidal intent (Swannell et al., 2014) provided a pooled prevalence estimate of 17.2% among adolescents, compared with 13.4% in young adults and 5.5% in adults. Moreover, there is evidence that incidence in younger children is increasing with variation across the sexes. Marchant et al. (2020) found that, between 2003 and 2015, there were increases in attendances at emergency departments for self-harm for young people and that females were more likely than males to be admitted.

A history of self-harm without suicidal intent has been shown to be strongly associated with an increased likelihood of attempting suicide (Mars et al., 2019), and a history of self-harm is the strongest predictor of completed suicide (Geulayov et al., 2019). Foster et al. (1997) reported that, among those who had died by suicide, 52% had a history of self-harm. According to theoretical accounts of self-harm, individuals may use it as a means of expressing, reducing, numbing, or distracting from psychological distress such as anxiety or emotional pain (e.g., Chapman et al., 2006; Linehan, 1993; Madge et al., 2011; O'Connor et al.,2010; Valencia-Agudo et al., 2018). Indeed, in a study of adolescent thoughts about self-harm and suicide attempts across 17 European counties, lifetime prevalence of suicide attempts was on average 10.5% (Kokkevi et al., 2012).

Risk factors for self-harm, with and without suicidal intent, are wide ranging and include being bullied and experiencing mental health problems such as depression, anxiety, low self-esteem, and poor self-concept (Madge et al., 2008; Mars et al., 2014; Monto et al., 2018; Valencia-Agudo et al., 2018). In a systematic review by Evans et al. (2004), stress factors that were identified as having strong evidence of association with self-harm in adolescents included poor peer relationships and family discord.

These are factors that have also been associated with atypical speech development. Atypical speech development in this context is used as an overarching label to cover two broad types of difficulty, which can occur in children: stuttering, where speech is characterized by unintentional repetitions, prolongations, and interruptions, and speech sound disorder (SSD), where the production of speech sounds is affected by substitutions, omissions, distortions, and additions of phonemes as a consequence of either articulatory or phonological problems, or both. Each of these types of difficulty is regarded as persistent if it continues into middle childhood. Dworzynski et al. (2007) reported that just over 1% of 7-year-olds exhibited persistent stuttering, whereas Wren et al. (2016) estimated the prevalence of SSD at 8 years to be 3.6%.

Atypical speech development impacts a child's ability to make themselves understood. It is widely known that this can influence educational attainment (Anthony et al., 2011; Wren et al., 2021) with social and economic consequences, which can continue into adulthood. It has also been associated with psychosocial factors such as negative peer reactions (Langevin et al., 2009; Overby et al., 2007) and bullying (Blood & Blood, 2004, 2007; Daniel & McLeod, 2017; Davis et al., 2002; Langevin et al., 2009; McCormack et al., 2011, 2012; McLeod, Daniel, & Barr, 2013; Reilly et al., 2015; Sweeting & West, 2001).

Stuttering and Self-Harm

Evidence suggests anxiety disorders (especially social anxiety disorder) may be more prevalent among adults who stutter than in the general population, at least among those seeking treatment for stuttering (Blumgart et al., 2010; Iverach, Jones, et al., 2009; Iverach, O'Brian, et al., 2009; Iverach & Rapee, 2014; Menzies et al., 2008). There is some evidence that social anxiety disorder emerges during childhood in people who stutter: Iverach et al. (2016) reported a 6-fold increase in the prevalence of social anxiety disorder among 7- to 12-year-olds who stuttered compared with controls who did not stutter. By contrast, Messenger et al. (2015) found that social anxiety scores for children and adolescents who stuttered were within normal limits on the Revised Children's Manifest Anxiety Scale (RCMAS; Reynolds & Richmond, 1978). However, they also found that boys who stuttered had high scores on the RCMAS Lie Scale, suggesting that they might have answered questions elsewhere on the RCMAS in such a way as to present themselves in a more positive light, for example, by concealing their anxieties about their speech.

Using data from a large national longitudinal study, Briley et al. (2021) reported higher levels of depression in young people who stutter. They also found that boys who stuttered were more likely to experience suicidal ideation than their fluent peers, but the same finding was not observed in girls. Further supporting evidence comes from a large population study that found that behavioral, emotional, and social difficulties may be more common among children who stutter than typically developing children (McAllister, 2016).

Reports of self-harm in this population are rarer. A review of anxiety, social phobia, depression, and suicide among people who stutter by Rezaeian et al. (2020) found no adequate research on suicide and stuttering. More recently, however, Al-Ghamdi et al. (2022), in their study of 59 males who stuttered, have found a higher risk of developing negative thoughts leading to suicidal attempts and reported that 33.9% of their participants had attempted suicide at least once. Similarly, Boyle (2015b) reported unpublished research by Kuster (2012) and Kuster et al. (2013), which provided evidence from the writings of people who stutter that suicidal thoughts occur in this population, and described two who took their own lives. This devastating outcome may be related to factors other than stuttering, but nevertheless, Rezaeian et al. (2020) concede that there is a known association between anxiety and suicide. Their review found a consistent reporting of an association between stuttering and anxiety.

A weakness of many of these studies was the age of the populations observed. Rezaeian et al. (2020) in their review noted that most of the 34 articles reported on adults. However, the more recent study of Al-Ghamdi et al. (2022) included males from the age of 11 years and up: 94% of the sample was aged over 16 years and 49% over 21 years. There is a lack of knowledge, therefore, in the relationship between stuttering and self-harm in younger populations and whether the findings so far are incidental or suggestive of a pattern.

A variety of explanations have been provided in the literature for why children who stutter also experience problems with bullying and mental health, which could, potentially, lead to self-harming behavior. McAllister (2016) considered that Boyle's (2015a) model of self-stigmatization may offer a partial explanation for the emergence of psychological difficulties in some children who stutter. Boyle's model hypothesizes that once a child becomes aware that they are stigmatized by others, they start to apply the same negative attitudes to themselves, leading to problems with their well-being.

Blood et al. (2011) compared 54 students who stuttered with 54 who did not stutter and found significantly greater rates of victimization in those who stuttered compared to those who did not. This was combined with lower self-esteem and a less optimistic life orientation in the group who stuttered. They concluded that increased vigilance and early intervention are vital for this population.

SSD and Self-Harm

Children with SSD may have higher levels of stress, which may be linked to increased levels of self-harm. Evans et al. (2004) identify stress in a variety of forms, including poor peer relationships and family discord, as associated with an increased likelihood of self-harm in adolescents who show vulnerabilities such as a family history of suicidal behavior and poor communication with family. Children with SSD are known to have poor peer relationships compared to their nonaffected friends (McCormack et al., 2011, 2012; Wren et al., 2023), and difficulties in making themselves understood could contribute to poor communication and discord within the family. This may be exacerbated by evidence that suggests children with SSD have problems handling stress, as reported by speech-language pathologists and parents of children with SSD (McCormack et al., 2010).

Children with SSD have also been shown to have higher rates of emotional problems and other mental disorders compared to their nonaffected peers in a large-scale national study with 12,388 participants in Australia based on parent report (Keating et al., 2001), though what is included within the labels of emotional problems and mental disorders in this investigation is unclear. In a small-scale qualitative study, the manifestation of these emotional problems for two children with SSD is reported by mothers who commented on the distress observed in their sons at not being able to be understood and in response to bullying (McCormack et al., 2012). An extreme example is provided by Carrigg et al. (2015) who reported a single case study of a man aged 22 years with severe SSD who had presented to emergency and mental health services 4 times during his adolescence as a result of self-harming and suicidal ideation.

There are conflicting findings, however. Beitchman et al.'s (2001) study of 105 young people with a history of speech and language impairment at the age of 5 years found that the 38 who had problems with speech only (i.e., SSD) were no more likely than controls to show psychiatric problems at the age of 19 years. In addition, while Wren et al. (2023) found problems with peer relationships and emotionality as reported by teachers in their 8-year-olds with SSD, their sample did not demonstrate any greater levels of depression than nonaffected peers.

Understanding why children with SSD might self-harm is important to consider in exploring whether an association exists. McCormack et al. (2010) found that parents reported that their children often withdrew from communication interactions, perhaps fearful that they would not be understood. They considered that the frustrations that children may feel as a result of communication breakdown may contribute to their difficulties in handling stress.

Aims of This Study

Investigations into risk factors for self-harm in children and young people have, to date, focused on factors relating to demographics (e.g., sex and socioeconomic status), the individual (e.g., IQ, personality, mental health, and behavior), and the environment (e.g., exposure to self-harm, early adverse experiences, stressful life events, substance use, online social media use, and cyberbullying; Carballo et al., 2020; Gillies et al., 2018; Liu et al., 2018; Mars et al., 2014, 2019; Memon et al., 2018). The communication challenges that occur as a result of atypical speech development have not been considered in such analyses, yet we have demonstrated that there is an association with many of these risk factors.

Studies of atypical speech and self-harm have, to date, focused on older populations rather than children and young people and have tended toward small samples. Al-Ghamdi et al. (2022) recommended that future studies should be designed to establish the relationship between atypical speech and self-harm. Large sample sizes are needed to determine if the associations observed in individuals are idiosyncratic or representative of a pattern. Such studies have the advantages of increased precision of results and, in a sample of the general population, of containing a reasonable number of participants with the exposures and outcomes of interest. If children with atypical speech development are at greater risk of self-harm, those working with these children in a professional capacity in schools and health roles and parents should be aware of this. Moreover, the impact of interventions should be measured not only on speech production but also on risk of self-harm.

The aim of this study was therefore to determine whether children with atypical speech development were more likely to engage in self-harming behavior, with or without suicidal intent. Few studies have sufficient data to explore this question as it is rare for longitudinal population studies to include information on children's speech development. The Avon Longitudinal Study of Parents and Children (ALSPAC) is an exception, however. Used by Mars et al. (2014, 2019) to investigate differences in etiology and prognosis for young people who self-harm with and without suicidal intent, this study includes data collected by postal questionnaire to 4,799 participants on a range of issues including self-harm and data from direct assessment of 7,390 children's speech at the age of 8 years.

The specific questions addressed in this work are as follows.

1. Are children who stutter more likely than their nonstuttering peers to self-harm, with or without suicidal intent?

2. Are children with speech sound production difficulties more likely than their unaffected peers to self-harm, with or without suicidal intent?

Method

Participants

ALSPAC is a large, longitudinal, prospective, community-sampled study of pregnancy, child health, and lifespan development based in the former county of Avon in the United Kingdom. The original sample consisted of 14,541 pregnant women with expected delivery dates between April 1, 1991, and December 31, 1992 (Boyd et al., 2013; Fraser et al., 2013). These pregnancies resulted in 14,676 live births, with 13,988 children alive at 1 year of age. At the age of 7 years, a further 548 children joined this study; these children were eligible from birth, but their mothers had not been recruited during the initial sweep. Data collection has occurred regularly since the start of this study. Please note that the study website contains details of all the data that are available through a fully searchable data dictionary and variable search tool; see http://www.bristol.ac.uk/alspac/researchers/our-data/. Ethical approval for this study was obtained from the ALSPAC Ethics and Law Committee and the local research ethics committee.

Not all participants completed all aspects of data collection. A total of 7,390 participants completed the speech assessment measures at the age of 8 years, and 4,799 participants completed the self-harm questionnaire at the age of 16 years. Only those participants who had data on both measures were included in the analysis (n = 3,824). Figure 1 provides a summary of participant numbers through this study. Children with comorbidities in the sample were not excluded from either of the case groups or the control group. This was because recruitment was from a population study in which all children were eligible for inclusion provided they were born during the recruitment period and in the location for recruitment. Removing those children with known comorbidities would risk inclusion of those with unknown comorbidities, potentially biasing the sample. Using a normative population sample ensured that comorbidities would appear within the sample in the same proportions as in the population as a whole.

Figure 1.

Summary of case identification (see text for further details). CASE = Child & Adolescent Self-harm in Europe questionnaire (Madge et al., 2008); SSD = speech sound disorder.

Variables

Data on speech fluency and speech sound production, self-harm, and known confounders were collected at various times during the ALSPAC study.

Speech Fluency and Speech Sound Production

Data on atypical speech development were collected at the “Focus at 8” clinic. All 13,314 cohort members who were still alive, consenting, and contactable were invited to the Focus at 8 clinic, which took place when they were 8 years 6 months old. The clinic, which was designed to explore the children's physical and mental health, as well as cognitive and social development, was attended by 7,488 children (M_age = 103.8 months, SD = 3.92; 50.1% boys, 49.9% girls). Although the original sample was diverse in terms of socioeconomic status, those who attended the Focus at 8 clinic were predominantly White (96.1%) and more likely to own their own homes compared to nonattendees (83.3% vs. 61.2%) and the mothers were more likely to have continued education to at least the age of 18 years (43.3% vs. 24.9%), suggesting a bias toward a higher socioeconomic status in the attendees (Wren et al., 2013). Reasons for nonattendance are not available, but given the nature of this study as an inclusive longitudinal cohort study, it was anticipated that some families would cease to maintain their involvement over time. The rate of attrition is similar to other longitudinal cohort studies (Gustavson et al., 2012). Details of the measures collected at the clinic are available at http://www.bris.ac.uk/media-library/sites/alspac/documents/researchers/clinics/focusclinicsessions.pdf.

The clinic included a 20-min direct assessment of the child's speech and language, which 7,390 children completed. No targeted speech assessments were carried out in clinic. Digital recordings were made of the children's oral responses to the Wechsler Objective Language Dimensions (WOLD; Rust, 1996) and an adapted version of the Nonword Repetition Test (Gathercole et al., 1994), and these provided the speech samples that were used to identify the child's speech status. Three subsamples of speech were collected: isolated words produced during a confrontation naming task (WOLD), a connected speech sample generated from three picture description tasks (WOLD), and isolated nonwords that the children repeated after listening to an audio recording of each one (Nonword Repetition Test).

The collection of data from these 8-year-olds was carried out using an agreed protocol. In the majority of cases (85.9%), speech data collection was carried out by U.K.-trained speech and language therapists (SLTs) who were native British English speakers; the remaining samples were collected by psychologists following training from the SLTs. The team carrying out data collection noted any fluency or speech features that they considered atypical in terms of fluency or speech sound production, taking into account the child's age and accent. Children were considered to have typical speech if their speech was consistent with a regional accent and they produced only isolated errors that were not indicative of a system-wide error.

The recordings of those children who were considered to have atypical fluency during the clinic session and those whose parents were concerned about their child's fluency were listened to by members of the SLT assessment team. These recordings were then orthographically transcribed by the assessment team with disfluencies written as they were heard, for example, “bu, bu, bu, but” or “‘but, but, but.” A 4-point rating scale was piloted by two SLTs with considerable experience in working with children who stutter. Firstly, they independently rated 27 samples and then compared and discussed their interpretations of the scale. A further 45 samples were then independently rated, resulting in 98% agreement between raters. Samples rated 3 or 4 indicated stuttering. A rating scale was used for this analysis to include children who had occasional moments of severe stuttering, who would not meet the criteria for 3% stutter-like disfluencies (Yairi & Ambrose, 2005). The choice of a rating scale was influenced by the relatively small speech samples, which also lacked the range of speaking tasks found in more formal stuttering assessments.

These recordings and transcription were used as the basis to categorize participants into those who stuttered and those who did not. On the basis of these ratings, 173 children out of a total of 382 were identified as stuttering. There were 126 (73%) boys and 47 (27%) girls, and the mean sample length was 222 syllables.

A total of 991 children were classified as having atypical speech sound production. Those whose errors were limited to what Shriberg (1993) described as common clinical distortions (N = 580) were identified and excluded from further analysis. Common clinical distortions are defined by Shriberg (1993) as labialized and velarized /l/; labialized, velarized, and derhotacized /r/; and dentalized and lateralized /s/, /z/, /ʃ/, /ʒ/, /tʃ/, and /dʒ/. This is because children with these errors at this age are not typically seen for intervention in the United Kingdom as these productions are considered representative of speech difference rather than speech disorder. Of the remaining 411, five recordings were missing or corrupted and could not be transcribed; the data for these individuals were also excluded from any further analysis. The recordings of the remaining 406 participants with atypical speech sound production together with a randomly selected control group of 50 participants from the rest of the cohort were listened to, transcribed using narrow transcription, and analyzed in terms of error type and percentage consonants correct (PCC).

The continuous speech samples of these 456 participants contained a mean of 141.7 word tokens (SD = 61.4). Two of the PCC measures, PCC Late 8 and PCC Adjusted (Shriberg, 1993), were selected to be used for the purposes of identifying true cases of SSD from within the sample of children with atypical speech sound production as identified by assessors during the assessment. The PCC Late 8 measure was selected as it was considered to be sensitive to later speech sound development, which was important because the samples were past the age of typical speech sound acquisition. The PCC Adjusted was selected as this measure accepts common clinical distortions as acceptable productions, which was consistent with our exclusion of children whose speech contained only these errors.

The recordings from a sample of 50 children selected at random from the total of 456 were retranscribed by a second transcriber. Point-to-point interrater reliability was 92.3% for these samples. As this was carried out post hoc, it was not possible to resolve discrepancies for the 7.7% where differences were observed and the original transcriptions were used in further analyses.

The PCC scores for the control group revealed that three children had scores that were considered outliers. Therefore, the means for the remaining 47 were used in analysis. The mean PCC Late 8 score for the 47 controls for both boys and girls was 96.7% with an SD of 4.0 (boys: 95.8%, SD = 4.3; girls: 97.6%, SD = 3.6). The mean PCC Adjusted score for the 47 controls for both boys and girls was 98.1% with an SD of 1.7 (boys: 97.8%, SD = 1.6; girls: 98.5%, SD = 1.7).

The PCC scores for the children with atypical speech sound development were compared against these controls. Two children were shown to have 100% PCC Adjusted, suggesting that their speech errors were limited to common clinical distortions. These children were excluded from any further analysis.

The criterion for identifying an individual as an SSD case was a score at least 1.2 SDs below the mean for the control participants. The choice for this cutoff was based on the work of Records and Tomblin (1994). They identified that clinicians' decisions regarding diagnosis based on clinical judgment were consistent with a cutoff of −1.2 SDs on standardized tests. This process identified 263 individuals who met the diagnostic criterion for SSD; one cohort member subsequently withdrew consent, leaving 262 SSD cases potentially included in our analyses if they also had self-harm data. More details on the process of assessment and analysis of these recordings are provided in the work of Wren et al. (2016). Six children appeared in both the children who stutter and the SSD group. They were included in all analyses.

Self-harm. When participants were 16 years old, they were asked to complete a questionnaire about self-harm based on items from the Child & Adolescent Self-harm in Europe study (Madge et al., 2008). Compared with nonresponders (N = 4,528), those who returned the questionnaire (N = 4,799) were more likely to be female, have a higher level of education, and have a mother with a lower class of occupation (Mars et al., 2014). Presence of a history of self-harm was identified via the question, “Have you ever hurt yourself on purpose in any way (e.g., by taking an overdose of pills or by cutting yourself)?” Those who were identified as having self-harmed with suicidal intent responded to the item, “Do any of the following explain why you hurt yourself on [the most recent occasion]?” by selecting, “I wanted to die” or responding “Yes” to the item, “On any of the occasions when you have hurt yourself on purpose, have you seriously wanted to kill yourself?” On the basis of responses to these three items, three groups were identified: those who reported that they had not deliberately self-harmed, those who reported they have self-harmed without suicidal intent, and those who reported self-harm with suicidal intent (SHSI).

Covariates

Additional factors that were controlled for in the analyses were those that have been identified as important in previous investigations of self-harm (Mars et al., 2014). They included variables relating demographic factors (sex, income, and socioeconomic status), child IQ, body dissatisfaction, mental health, substance use, sensation-seeking behavior, childhood sexual abuse, cruelty to children in the household, bullying, exposure to self-harm, and parental suicide attempts. Table 1 provides details on the processes used to collect these data and information on when they were collected.

Table 1.

List of covariate variables included in the analysis.

Grouped variables	Variable	Method of data collection	Timing of data collection
Demographic	Biological sex	Birth records from midwife	Birth
	Equivalized incomea	Questionnaire to mother	Age 33 and 37 months
	Parental social class	Questionnaire to mother	Pregnancy
	Level of maternal education (< O-level/O-level/> O-level)b	Questionnaire to mother	Pregnancy
Personal, mental health, behavior	IQ (Wechsler Intelligence Scale for Children–Third Edition (WISC III; Wechsler, 1991)	Direct assessment	Age 8 years
	Impulsivity (stop-signal task, Logan et al., 1984)	Direct assessment	Age 10 years
	Body dissatisfaction (happy or unhappy over the past year)	Questionnaire to child	Age 13 years
	Mental health (Short Moods and Feelings Questionnaire)	Questionnaire to child	Age 14 years
	Depressive or anxiety disorder (DAWBA; Goodman et al., 2000)	Interview to child	Age 15 years
	Substance use (alcohol, cannabis, smoking)	Questionnaire to child	Age 15 years
	Sensation seeking (novelty and intensity subscales of the Arnett Inventory of Sensation Seeking; Arnett, 1994)	Questionnaire to child	Age 16 years
Early adverse experiences	Childhood sexual abuse	Questionnaire to mother	Measure repeated 7 times from birth to age 8 years
	Cruelty to children in household	Questionnaire to mother	Measure repeated 8 times from birth to age 11 years
	Bullying (modified version of the bullying and friendship interview schedule; Woods & Wolke, 2003)	Interview to child	Age 12 years
	Parental suicide attempt	Questionnaire to mother	Measure repeated 8 times from birth to age 11 years
	Exposure to self-harm in friends, mother, and father	Questionnaire to child	Age 16 years

^a Average weekly household disposable income recorded at the ages of 3 and 4 years, divided into quintiles and rescaled to account for family size, composition, and estimated housing benefits.

^b “O-level” was the qualification obtained at the age of 16 years when the parents of the cohort were at school.

Statistical Analysis

Self-harm was analyzed as a three-level outcome: no self-harm, self-harm without suicidal intent, and SHSI. Multinomial logistic regression, which produces relative risk ratio (RRRs), was used to examine the associations between the exposures of stuttering and SSD, and the self-harm outcome variable, adjusting for other important covariates. All of the factors that Mars et al. (2014) examined were associated with the outcome in models adjusting for sex only. All covariates were adjusted for, whether or not they were associated with stuttering or SSD. Using covariates that are not associated with the exposures of interest (stuttering and SSD), but that are good predictors of the outcome, increases the precision of the estimates. Five models were generated: one modeling stuttering and SSD alone, one adjusted for demographic factors (sex and measures of socioeconomic position), one adjusted for personal factors (IQ, body dissatisfaction, measures of mental health, and of substance use), one adjusted for early adverse experiences (measures of victimization and self-harm in friends and family), and a final model including all variables mentioned. We carried out the analyses in Stata (Version 16).

Missing data imputation. This analysis was based on a data set consisting of 3,824 individuals with stuttering, SSD, and self-harm data. The percentage of missing values for covariates varied from 0% for sex to 27% for parental cruelty to children in the household. Values for missing covariates were imputed using Multivariate Imputation by Chained Equations (van Buuren, 2018) and analyzed via the “mi” command in Stata. The variables and imputation models employed were those used by Mars et al. (2014), who imputed missing values from the variables in the analysis plus a number of auxiliary variables (variables associated with their exposures of interest and earlier/later measures of the same exposures). Two hundred imputed data sets were generated.

Results

Of the 3,824 participants who provided data for both speech status at the age of 8 years and self-harm at the age of 16 years, 94 (2.5%; 95% confidence interval [CI; 2.0, 3.0]) were assessed to be stuttering at the age of 8 years and 127 (3.3%; 95% CI [2.8, 3.9]) were classified as having SSD. While these figures represent a classification of the sample based on observation of behaviors that are consistent with stuttering or speech sound production problems, they should not be considered prevalence rates, as cutoffs based on specific metrics have not been used. There were six participants who were evaluated as having both. This was considered too few to investigate a possible interaction between the stutter and SSD variables, so these interactions were not included in the models used.

The results for the variables measured compared to the self-harm outcome were very similar to those reported in the work of Mars et al. (2014), which is to be expected as a large subset (3,824/4,799; 79.6%) of their observations was used. The 975 missing participants either lacked stuttering and SSD measures or had withdrawn consent. The main difference was that parental social class was not significant in this analysis (p = .178), whereas it was significant for Mars et al. (p = .004). Parental social class was included in the multivariable analyses. Because of the ALSPAC Executive requirement noted above, a supplementary descriptive table showing the variables included in the models broken down by self-harm status was not provided, as this would be disclosive in conjunction with Table 2.

Table 2.

Descriptive table for variables by stutter and SSD status.

Measure (age of assessment)	No stutter (n = 3,730)		Stutter (n = 94)		p valuea	No SSD (n = 3,697)		SSD (n = 127)		p valuea
Outcome
Self-harm (age 16 years)					≥ .05					.046
No self-harm	3,025	(81.1%)	81	(86.7%)		3,009	(81.4%)	97	(76.4%)
Self-harm without suicidal intent	459	(12.3%)	≥ 9 & ≤ 13	(≥ 9.6% & ≤ 13.8%)		455	(12.3%)	15	(11.8%)
Self-harm with suicidal intent	246	(6.6%)	< 5	(≥ 0% & ≤ 4.3%)		233	(6.3%)	15	(11.8%)
Any self-harm	705	(18.9%)	13	(13.8%)		688	(18.6%)	30	(23.6%)
Demographic
Female sex, n (%)	2,202	(59.0%)	29	(30.9%)	< .001	2,174	(58.8%)	57	(44.9%)	.002
Equivalized income (33 and 47 months), n (%)b					.425					.670
1st quintile (highest)	947	(27.4%)	28	(31.5%)		945	(27.6%)	30	(25.4%)
2nd quintile	824	(23.8%)	25	(28.1%)		822	(24.0%)	27	(22.9%)
3rd quintile	681	(19.7%)	18	(20.2%)		679	(19.8%)	20	(17.0%)
4th quintile	589	(17.0%)	12	(13.5%)		578	(16.9%)	23	(19.5%)
5th quintile (lowest)	415	(12.0%)	6	(6.7%)		403	(11.8%)	18	(15.3%)
Parent social class (pregnancy), n (%)c					.046					.076
Professional/managerial	2,361	(66.6%)	69	(76.7%)		2,360	(67.1%)	70	(59.3%)
Other	1,182	(33.4%)	21	(23.3%)		1,155	(32.9%)	48	(40.7%)
Mother's education (pregnancy), n (%)					.002					.008
Degree	784	(21.4%)	31	(33.3%)		794	(21.9%)	21	(16.5%)
A-level	1,083	(29.6%)	34	(36.6%)		1,085	(29.9%)	32	(25.2%)
O-level	1,201	(32.8%)	20	(21.5%)		1,181	(32.5%)	40	(31.5%)
< O-level	597	(16.3%)	8	(8.6%)		571	(15.7%)	34	(26.8%)
Total IQ (age 8 years), M (SD)	107.4	(16.1)	112.5	(17.5)	.002	107.7	(16.0)	101.1	(19.3)	< .001
Personal, mental health, behavior
Impulsivity (age 10 years), stop-signal task, mean number of trials correct at 250-ms delay (SD)	13.69	(2.59)	14.09	(2.25)	.175	13.71	(2.56)	13.43	(3.17)	.278
Body dissatisfaction (age 13 years), n (%)	1,020	(31.9%)	21	(26.9%)	.350	1,004	(31.8%)	37	(32.7%)	.826
Mental health
Depressive symptoms (age 14 years), SMFQ score 11+, n (%)	351	(11.0%)	9	(10.6%)	.901	340	(10.7%)	20	(20.2%)	.003
Depressive disorder (age 15 years), DAWBA, n (%)	41	(1.3%)	< 5	(≥ 0% & ≤ 4.3%)	≥ .05d	41	(1.4%)	< 5	(≥ 0% & ≤ 3.1%)	≥ .05d
Anxiety disorder (age 15 years), DAWBA, n (%)	47	(1.5%)	< 5	(≥ 0% & ≤ 4.3%)	≥ .05d	47	(1.6%)	< 5	(≥ 0% & ≤ 3.1%)	≥ .05d
Substance use (age 15 years)
Alcohol, heavy drinking, n (%)	546	(18.7%)	13	(16.3%)	.584	546	(18.8%)	13	(13.7%)	.211
Cannabis, at least occasional use, n (%)	243	(8.1%)	6	(7.1%)	.732	244	(8.2%)	5	(5.1%)	.275
Smoking, at least weekly, n (%)	229	(7.6%)	< 5	(≥ 0% & ≤ 4.3%)	≥ .05	226	(7.5%)	5	(5.0%)	.347
Sensation-seeking (age 16 years), M (SD)
Arnett intensity subscale	25.92	(4.52)	27.12	(4.50)	.012	25.97	(4.52)	25.48	(4.78)	.236
Arnett novelty subscale	25.93	(4.24)	27.07	(4.21)	.015	25.98	(4.25)	25.43	(4.03)	.157
Early adverse experience
Victimization
Sexual abuse (from birth to age 8 years), n (%)	17	(0.5%)	< 5	(≥ 0% & ≤ 4.3%)	≥ .05d	17	(0.5%)	< 5	(≥ 0% & ≤ 3.1%)	≥ .05d
Parental cruelty to children (from birth to age 11 years), n (%)	121	(4.4%)	< 5	(≥ 0% & ≤ 4.3%)	≥ .05d	123	(4.6%)	< 5	(≥ 0% & ≤ 3.1%)	≥ .05d
Being bullied (age 12 years), n (%)	834	(25.0%)	22	(26.5%)	.758	819	(24.8%)	37	(34.3%)	.025
Self-harm in friends and family
Parent suicide attempt (from birth to age 11 years), n (%)	48	(1.4%)	< 5	(≥ 0% & ≤ 4.3%)	≥ .05d	41	(1.2%)	7	(5.8%)	.001d
Mother self-harm (age 16 years), n (%)e	44	(1.2%)	< 5	(≥ 0% & ≤ 4.3%)	≥ .05d	42	(1.1%)	< 5	(≥ 0% & ≤ 3.1%)	≥ .05d
Father self-harm (age 16 years), n (%)e	22	(0.6%)	< 5	(≥ 0% & ≤ 4.3%)	≥ .05d	21	(0.6%)	< 5	(≥ 0% & ≤ 3.1%)	≥ .05d
Self-harm in friends (age 16 years), n (%)e	1,472	(39.7%)	33	(35.5%)	.414	1,456	(39.6%)	49	(38.6%)	.816

Note. Number of respondents with missing data: 279 for income, 191 for social class, 66 for maternal education, 44 for total IQ, 550 for body dissatisfaction, 506 for impulsivity, 104 for the intensity subscale of the Arnett's sensation-seeking scale, 117 for the novelty subscale of the Arnett's sensation-seeking scale, 553 for SMFQ score, 687 for DAWBA depression, 686 for DAWBA anxiety, 819 for heavy alcohol, 733 for cannabis, 714 for smoking, 227 for sexual abuse, 1,028 for physical cruelty to child, 408 for being bullied, 282 for parent suicide attempt (parent rated), 21 for mother self-harm (child rated), 21 for father self-harm (child rated), and 21 for self-harm in friends (child rated). SSD = speech sound disorder; SMFQ = short mood and feelings questionnaire; DAWBA = development and well-being assessment.

^a Chi-square/Fisher's exact test of the association between the speech and categorical variables and t-test for differences in means for continuous variables.

^b Quintiles represent highest to lowest household income. Quintiles were derived from income measures at ages 33 and 47 months on a larger subset of the cohort; hence, in the present sample, numbers are not evenly distributed.

^c Highest social class of mother and father.

^d Fisher's exact test was used because one of the expected values was less than 5.

^e Child rated.

Table 2 shows the variables included in the models grouped by the stutter/no stutter (the latter including SSD participants) and SSD/no SSD measures (the latter including stuttering participants). This table has been edited so that the exact values for cells < 5 are not shown and any information from which these values could be calculated has also been modified. This includes the corresponding p values, which are given as < .05 or ≥ .05. In this raw analysis, there was no difference between the rates of the different type of self-harm in participants with and without stuttering (p ≥ .05; see Table 2). There was, however, a difference between the SSD and non-SSD groups, with rates of 12.3% and 11.8% for self-harm without suicidal intent, and 6.3% and 11.8% for SHSI, respectively (p = .046; see Table 2).

The stutter and SSD groups both had a smaller proportion of girls. The stutter group was more likely to have parents from a professional/managerial class, has more educated mothers, and has a higher mean IQ than the nonstuttering group, whereas the SSD group was more likely to have less educated mothers and has a lower mean IQ than the non-SSD group. Participants in the stuttering group had higher mean sensation-seeking scores (both intensity and novelty) than the nonstuttering group, but no such difference was seen for the SSD group. The SSD group were more likely to have depressive symptoms and to have experienced a parental suicide attempt. Other factors (income, body dissatisfaction, anxiety disorder, substance use, measures of victimization, and other self-harm in friends and family) did not show an association with stuttering or with SSD.

There was a great deal of consistency between the different models used to examine the relationship between stuttering and SSD and the three self-harm categories (see Table 3). There was no association between stuttering and self-harm without suicidal intent (vs. no self-harm) in any of the models. There was a consistent but nonsignificant lower risk of SHSI (compared to no self-harm) and of SHSI (vs. self-harm without such intent) for those who stuttered in all models. There was no association between SSD and self-harm without suicidal intent and a consistent but nonsignificant positive association between SSD and SHSI versus without in all models. The only association that was significant was between SHSI (vs. no self-harm) for the SSD group compared to the non-SSD group. This was seen for all models. The fully adjusted model produced an RRR of 2.57 (95% CI [1.31, 5.03]) for SHSI and SSD. This RRR can be broken down as

Table 3.

Associations between speech variables and self-harm with and without suicidal intent, with adjustments for other variables (n = 3,824).

Model	Omnibus test for variable	Self-harm without suicidal intent versus no self-harm			Self-harm with suicidal intent versus no self-harm			Self-harm with suicidal intent versus self-harm without suicidal intent
	p value	OR	95% CI	p value	OR	95% CI	p value	OR	95% CI	p value
Stutter
Model 1	.228	0.89	[0.47, 1.69]	.731	0.29	[0.07, 1.20]	.089	0.33	[0.07, 1.50]	.150
Model 2	.409	1.24	[0.64, 2.39]	.520	0.44	[0.11, 1.83]	.260	0.36	[0.08, 1.64]	.185
Model 3	.328	0.86	[0.44, 1.68]	.654	0.34	[0.08, 1.45]	.145	0.40	[0.09, 1.84]	.239
Model 4	.277	0.92	[0.47, 1.81]	.812	0.30	[0.07, 1.31]	.109	0.33	[0.07, 1.53]	.157
Final model	.500	1.10	[0.54, 2.27]	.787	0.42	[0.09, 2.00]	.277	0.38	[0.08, 1.91]	.241
SSD
Model 1	.044	1.03	[0.59, 1.78]	.929	2.04	[1.16, 3.57]	.013	1.99	[0.95, 4.14]	.067
Model 2	.010	1.26	[0.71, 2.22]	.426	2.44	[1.37, 4.34]	.002	1.94	[0.92, 4.06]	.080
Model 3	.068	1.10	[0.62, 1.96]	.739	2.06	[1.11, 3.83]	.022	1.87	[0.88, 4.00]	.105
Model 4	.080	1.11	[0.62, 2.00]	.723	2.04	[1.09, 3.80]	.025	1.83	[0.87, 3.88]	.113
Final model	.020	1.43	[0.78, 2.64]	.248	2.57	[1.31, 5.03]	.006	1.79	[0.82, 3.91]	.145

Note. Model 1: Stutter and SSD variables only. Model 2: Stutter, SSD, gender, equivalized income, parental social class, mother's education. Model 3: Stutter, SSD, total IQ, body dissatisfaction, Arnett intensity subscale, Arnett novelty subscale, depressive symptoms (SMFQ score 11+), DAWBA depression, DAWBA anxiety, alcohol (heavy drinking), cannabis (occasional), smoking (weekly). Model 4: Stutter, SSD, childhood sexual abuse, cruelty to children in household, being bullied. Final model: All variables listed in the previous models. OR = odds ratio; CI = confidence interval; SSD = speech sound disorder; SMFQ = short mood and feelings questionnaire; DAWBA = development and well-being assessment.

(The relative risk of SHSI compared to no self-harm in participants with SSD)/(The relative risk of SHSI compared to no self-harm in participants without SSD),

adjusted for the other variables in the model. The RRRs did not change greatly between the variously adjusted models. The full results for all variables included in the models are presented Supplemental Materials S1–S6.

The Type I error was fixed at 5% (i.e., statistical significance of 2.5%, two tailed) in keeping with convention. The Type II error is dependent upon sample size, which was fixed in this case. Given the complexity of the modeling, it was not possible to provide any a priori power calculations, and this therefore may have been low as indicated in the discussion. However, once parameter estimates are made, the precision of the estimates can be seen from CIs.

Discussion

This is the first study to explore the relationship between atypical speech development (stuttering and SSD) and adolescent self-harm with or without suicidal intent. Based on results of earlier studies indicating that experiencing bullying and mental health difficulties predict the likelihood of self-harm (e.g., Madge et al., 2011; Valencia-Agudo et al., 2018) and that those with a developmental history of stuttering or SSD are more likely than typically developing peers to experience such issues (e.g., Blumgart et al., 2010; Iverach, Jones, et al., 2009; Iverach, O'Brian, et al., 2009; Iverach & Rapee, 2014; Keating et al., 2001; McAllister, 2016; McCormack et al., 2012; McLeod, Daniel, & Barr, 2013; Menzies et al., 2008; Sullivan et al., 2016), we hypothesized that participants with stuttering or SSD would be more likely than those without these speech problems to report self-harm with or without suicidal intent.

Our hypothesis was partially supported. Individuals with SSD were more than twice as likely as controls to SHSI, an effect that was observed in all of our statistical models, which controlled for a wide range of factors that have previously been shown to be associated with self-harm (Mars et al., 2014). The absence of large differences between the adjusted and unadjusted models may suggest that SSD is largely an independent risk factor for SHSI. The exact relationship among SHSI, suicide attempts, SSD, and other predictors is likely to be a complex one that warrants further investigation.

SSD and Self-Harm

Despite the small absolute number of cases of SHSI among those with SSD in our sample, taken together with prior findings cited in the Introduction, clinicians should take the risk seriously. First, the prevalence of SSD is relatively high, estimated at 3.6% at the age of 8 years (Wren et al., 2016). Second, the consequences of SHSI, not just for the individual but for their family and friends, are potentially devastating. A history of self-harm is the strongest predictor of completed suicide (Geulayov et al., 2019), and as many as 52% of those who die by suicide may have a history of self-harm (Foster et al., 1997).

McCormack et al. (2010) reported on the challenges faced by children if the impact of their SSD is such that they are unintelligible. Frustration and a tendency to become withdrawn as a response to communication breakdown and associated problems with making and sustaining friendships, as reported across other large epidemiological studies (McCormack et al., 2011), could be a contributory factor in why some children with SSD suffer with mental health and well-being issues (Keating et al., 2001; McCormack et al., 2012; McLeod, Daniel, & Barr, 2013; McLeod, Harrison, et al., 2013; Wren et al., 2023) and have difficulty handling stress (McCormack et al., 2010). The evidence from this study suggests that this impact may go beyond such negative feelings and into actions, specifically SHSI.

These impacts may also be experienced in those children whose SSD does not affect intelligibility but does mark them out as different. Interviews with seven adults who self-identified as having a lisp, a misarticulation primarily affecting only the sibilant consonants rarely impacting on intelligibility, by Lockenvitz et al. (2022) found that participants reported experiences of both internalized and public stigma. The SSD group in this study included children for whom low intelligibility may not have been a factor. Nevertheless, they may be vulnerable to the self-stigmatization described by Boyle (2015a) and the internalized and public stigma experienced by the participants in the study of Lockenvitz et al. (2022).

Stigma can be perceived by individuals if they are exposed to negative reactions from others in relation to their speech. In a study of audience response to child disordered speech, Strömbergsson et al. (2021) found that listeners reacted more often to speech disrupted by acceptability (i.e., it sounded odd) than by speech disrupted by problems with intelligibility. It is not clear from the work reported in this article whether the association between SSD and SHSI is an impact of unacceptability or unintelligibility of speech. Further research is needed using a narrower definition of SSD such that only children whose intelligibility is likely to have been affected are included. This will help to determine whether it is communication breakdown as a function of low intelligibility, which is the important factor, or problems with acceptability of speech. Similarly, investigations into the trajectory of SSD and associations with self-harm are needed to determine if the relationship is one of severity and persistence of SSD rather than one of intelligibility and acceptability of speech.

Stuttering and Self-Harm

No relationship was observed between stuttering and SHSI. The absence of this relationship was unexpected in view of the large number of studies indicating an association between stuttering and both being bullied and mental health problems, which are known risk factors for self-harm (Blood & Blood, 2004, 2007; Daniel & McLeod, 2017; Davis et al., 2002; Langevin et al., 2009; McCormack et al., 2011, 2012; Reilly et al., 2015; Sweeting & West, 2001). This is encouraging and suggests that, although there is evidence that, in adulthood, this population may be more likely to suffer from anxiety (Blumgart et al., 2010; Iverach, Jones, et al., 2009; Iverach, O'Brian, et al., 2009; Iverach & Rapee, 2014; Menzies et al., 2008); depression (Briley et al., 2021), and other social, emotional, and behavioral difficulties (Blood et al., 2011; Boyle, 2015a; McAllister, 2016), the evidence from this study suggests that these features, when experienced in childhood, do not necessarily progress to self-harming behavior, either with or without suicidal intent.

Nevertheless, it is important that alternative possible explanations for these findings are explored. First, this study may be insufficiently powered to detect such a relationship, given the relatively small number of participants who stuttered and completed the self-harm questionnaire. Moreover, this sample contained far fewer females than those in the control group, as would be expected given the higher prevalence of stuttering in males. As self-harm is more common in females (Marchant et al., 2020), it is possible that the lack of association observed is a function of the gender differences between the two samples rather than presence or absence of stuttering, although all models except Model 1 were adjusted for gender. Another explanation could be that the evidence base for mental health issues among people who stutter has mainly involved participants who were seeking treatment for stuttering; there is limited evidence drawn from community samples such as the one used in this study. Furthermore, few studies of mental health and stuttering have focused on childhood and adolescence, which are time windows that might be more relevant to this study. Finally, although recovery from stuttering is most likely in the preschool years, it is possible that some individuals may have recovered after the age of 8 years in the current sample (Yairi & Ambrose, 2013), which might have influenced the outcome (assuming that there was a relationship between stuttering and self-harm in the first place). Further research is needed to investigate these possibilities.

A strength of this study is that it used a large community-based sample. As Mars et al. (2014) pointed out in their study, use of a community sample is important because most self-harm is not seen by specialist services. A further strength is the availability of data from direct assessment of speech, an attribute that ALSPAC shares with very few other large birth cohorts. The richness of the ALSPAC data set also enabled us to control for a wide range of other factors associated with self-harm. Despite the large sample size in this study, a potential weakness was the lack of statistical power, particularly for stuttering. Use of a larger community sample would have been desirable; however, it is worth pointing out that ALSPAC is the only community study of this size and breadth that contains both speech data and information about self-harm and was thus the optimal choice among currently available resources.

Limitations

Large-scale longitudinal population studies like ALSPAC are rare. Even more unusual is the availability of speech data, which have been collected through direct assessment rather than relying solely on parent report, as is the case with other similar studies. However, there are also limitations. One of these is that the data available for use are predetermined, sometimes years in advance of when the data are used. Thus, some variables used in the analysis may represent behaviors or circumstances that can change over time, yet data are only available at a single time point. For example, the level of maternal education recorded in pregnancy may have changed by the time the speech data were collected. Similarly, measures of personal, mental health, and behavior can change over time, and participants' responses may be different if asked about their body dissatisfaction or substance use, for example, when older or younger.

Moreover, the types of assessment and analyses that can be carried out on large population samples of over 7,000 participants are different from those that can be used with smaller samples. Time and cost limitations for this study meant that all measures of speech and language had to be completed within 20 min. The most efficient way to collect data therefore was to use assessments that could provide data for both speech and language analysis. It was for this reason that language assessments were used as the basis for the speech sample used in the analyses reported in this article.

However, these limitations must be balanced with the strength that comes with a large sample size and the confidence that we can have in findings. There may be fluctuations in individuals' responses over time, but these are unlikely to result in major differences to results. Moreover, given the age of the children, phonetic transcription of continuous speech samples would arguably provide richer and more naturalistic data with greater sensitivity to difference than had a single-word phonemically balanced picture description task been used.

Attrition is another challenge with longitudinal research. However, use of imputation by another study using the same data set (Mars et al., 2014) suggests that the data included in the analysis reported in this article were robust. Reporting accuracy is also a potential source of bias. Participants may fail to report sensitive issues such as SHSI accurately, for example, because self-reports may be affected by current mood state; nonetheless, self-report is considered to be more reliable than interviewing, which is the method used in some other studies. Sexual abuse may be inaccurately reported, in part due to not only sensitivity but also imperfect knowledge when, as in this study, a parent rather than the victim is the source of the data.

The self-harm data do not allow researchers to pinpoint the timing of reported incidents relative to the influence of predictors. It is therefore difficult to determine with any certainty whether some predictors predated the reported self-harm and, therefore, whether there could be a causal relationship with the outcome. While this is particularly the case for variables collected relatively close to the age of 16 years when the self-harm questionnaire was administered, Mars et al. (2014) have pointed out that since self-harm rarely occurs before the age of 12 years, predictors gathered before this age almost certainly predate the self-harm events; importantly for this study, this includes diagnosis of stuttering and SSD.

Clinical Implications

As part of a broader self-harm prevention strategy, practitioners need to be aware of the diverse range of factors, including speech sound production difficulties, that are associated with and may contribute causally to SHSI among adolescents (Madge et al., 2011). Speech and language therapy services need to ensure that provision is in place for older children with speech sound production difficulties and preschoolers. This provision needs to include awareness that children with a history of speech sound production difficulties—who may not present with these currently—are at higher risk than their peers of SHSI. It should also be considered that severity may not be a factor in determining the likelihood of observing self-harm behavior and that problems with acceptability of speech may be as important (or possibly even more important) as intelligibility. Other professionals who need to be aware of the findings reported here include general practitioners and other clinicians, teachers, and those in the voluntary, community, and social enterprise sectors who provide counseling for children and young people, both those targeting self-harm and suicide prevention and more wide-ranging services such as Childline in the United Kingdom.

Considering that a high percentage of adolescents who attempt suicide have previously self-harmed with suicidal intent (Brausch & Gutierrez, 2010; Muehlenkamp & Gutierrez, 2007; Zetterqvist et al., 2013), professionals need to be vigilant for risk factors that are particularly associated with suicide attempts, such as self-harm in family or friends or the presence of anxiety or behavioral disorders (Mars et al., 2019). There are differing views about the effectiveness of interventions to address self-harm in adolescents (Hawton et al., 2015; Ougrin et al., 2015). Further research needs to develop the therapeutic evidence base and, in the context of this study, explore further the relationship between atypical speech development and self-harm and develop appropriate intervention approaches.

Conclusions

Compared with individuals without SSD, adolescents who had persistent SSD at the age of 8 years have twice the risk of reporting SHSI versus no self-harm, and this effect may be independent of other predictors of the outcome. The relationship among these other predictors, SSD, SHSI, and actual suicide attempts may be a complex one that requires further investigation. This finding has important implications for service delivery for this client group.

Data Availability Statement

Information about how to access the data used in the study reported here can be found at http://www.bristol.ac.uk/alspac/researchers/access/.

Funding Statement

This article presents independent research funded by the National Institute for Health Research (NIHR) Collaboration for Leadership in Applied Health Research & Care East of England, at Cambridgeshire and Peterborough National Health Service (NHS) Foundation Trust. The views expressed in this study are those of the author(s) and not necessarily those of the NHS, the NIHR, or the Department of Health. The U.K. Medical Research Council and Wellcome (Grant 217065/Z/19/Z) and the University of Bristol provide core support for ALSPAC. This publication is the work of the authors, and they will serve as guarantors for the contents of this article. A comprehensive list of grant funding is available on the ALSPAC website (http://www.bristol.ac.uk/alspac/external/documents/grant-acknowledgements.pdf); this research was specifically funded by Wellcome Trust and MRC (Grant 076467/Z/05/Z).