The principal goal was to identify risk factors for persistent stuttering in the areas of epidemiology, symptoms and etiology. Several factors were identified which suggested how differential diagnosis of stuttering from fluent speech could be made. Details are given here of on-going work that uses these and other factors to predict persistence and recovery of stuttering employing structural equation modeling (SEM).
One of the main findings of this study was that there was remarkably little difference between children who persisted or recovered from stuttering in the age range 8-12 plus in epidemiology. There was a chance of around 50% of persistence or recovery and this did not depend on gender nor on age of attendance at clinic. Recovery rate was lower in this study than in others with younger children (Andrews & Harris, 1964
; Yairi & Ambrose, 2005
), which is consistent with the view that most recovery happened when the children were young. It is also of note that, at 53.9%, the estimate of recovery rate was close to the 47% reported by Fritzell (1976)
for children with similar ages to those in the current study. Age of onset was roughly the same for both genders and did not depend on age of attendance at clinic. Also, age of onset of four years seven months was comparable with those reported in other studies, including those with younger children (Andrews & Harris, 1964
; Yairi & Ambrose, 2005
). Age of recovery was the same for gender and age of referral groups, although there was a hint that later attendance at clinic was associated with lower recovery rates. The reason for late attendance at clinic for the children in the current study was not clear, given that onset age was comparable to that reported by Yairi and Ambrose (2005)
. The children who recovered had stuttered for nearly nine years and they had been followed up, on average, for nearly six years to ensure that there was no relapse (the time the children were followed up was longer than the 40-month period reported by Yairi and Ambrose, 2005
). In fact, the only epidemiological factor that was significant was that more boys were affected than girls (5.33:1). This has been reported elsewhere and is considered as a major risk factor in other research. The sex ratio in the current study was higher than in Andrews and Harris (1964)
. This was expected, as females recover at a younger age than males (Andrews & Harris, 1964
) so an older sample should consist predominantly of males.
A score of 16 points on SSI-3 at intake separated persistent and recovered children from 77% of the age-matched fluent controls. At the final assessment, the SSI-3 scores of the recovered group had reduced and they were not statistically distinguishable from their age-matched controls. At this later age, the SSI-3 scores of the persistent group were significantly higher than those for both the recovered and the age-matched control groups. Using persistent/recovered designations determined from the information obtained at later attendance, showed that the recovered speakers' SSI-3 scores were about seven points lower than those of the persistent speakers at intake. Though there was overlap in SSI-3 scores for the two groups of children, the SSI-3 scores at intake could be used in conjunction with other measures for assessing risk of persistent stuttering. SSI-3 scores have been used for contrasting the fluency of speakers who stutter and control speakers by our own and other research groups (Arnold, Conture & Ohde, 2005
; Davis, Shisca & Howell, in press
; Howell, 2007
; Howell et al., 2006
). In addition to the symptom changes that were apparent in SSI-3 scores, there was also a change in counts of different dysfluency types for the persistent speakers. Part-word dysfluencies increased relative to stallings for the persistent group of speakers. Both types of dysfluency reduced in speakers who recovered, and the counts of the two types were not distinguishable from age-matched fluent controls at the earlier or later age. Therefore, the increase in counts of part-word repetitions across sessions is another risk factor for persistent stuttering, as noted elsewhere (Conture, 1990
; Howell, 2007
Of the etiological factors examined as potentials for risk, persistence and recovery showed no tendency to run in families. The reason why the current study failed to find fewer persistent relatives in the recovered than the persistent proband group could be due to Ambrose et al.'s (1997)
sample being younger than the current one, so recovery of the speakers was not fully resolved in their study. That is, a shift of some speakers from their recovered group who had high numbers of recovered relatives to the persistent group who had few recovered relatives would reduce the reported differences between proband groups. Family history (genetics) was not a risk factor for persistence for speakers in this age range although there are indications from other literature (Yairi & Ambrose, 2005
) that it would be useful, particularly when working with younger children.
In all but one case where etiological factors for persistence/recovery were identified, the persistent group performed differently from the recovered and the control groups. The exception was for the temperamental dimension of adaptability where recovered speakers were less adaptable than control speakers or persistent speakers (tests made from age 10 years and upwards). This temperament dimension was the only one that was significant and in the same direction in three previous studies (Anderson et al., 2003
; Embrechts et al., 2000
; Howell et al., 2004
). This was the only temperamental variable that was specific to recovery. In addition, approach and threshold distinguished all the speakers who stuttered from controls. Thus these temperamental variables were not sensitive to persistence/recovery although they were for stuttering in general. They should be included in a model intended to distinguish the gross fluency groups.
At the time when persistence and recovery were established, backward masking thresholds were about 10dB worse for persistent speakers than for recovered speakers, as also found by Howell et al. (2006)
. Signal intensity needed to be about 10 dB higher to be detected by persistent participants compared to recovered ones. Backward masking threshold is an additional risk factor that should be used for predicting risk of persistence. It does not appear useful for distinguishing children who stutter in general from fluent speakers, as the scores for the control speakers were not statistically different from the scores of either of the groups of children who stuttered, as reported elsewhere (Howell & Williams, 2004
Assessments of cerebellar performance were also made at the times when persistence and recovery were established. Variability when balancing with arms by the side was the only factor that differed between persistent and recovered children (recovered children were poorer), and between persistent and controls (persistent children were poorer). Poor performance on balance when arms were held by the side appears to be a risk factor for persistent stuttering. Deficits on the Dow-Moruzzi battery that distinguished between the three fluency groups potentially implicate different cerebellar control in stuttering, as predicted by EXPLAN theory. Past pointing looks like a candidate for differentiating controls from all children who stutter, but not for persistent/recovered groups. Apart from these findings on cerebellar tasks, there were only two other significant effects (postural stability with arms by the side for persistent versus recovered speakers, and balance time with arms outstretched for persistent versus control speakers).
Future work and limitations
The primary goal of the current research was to assess demographics, symptoms and etiological variables as risk factors for persistent stuttering. Demographics and symptomatology were examined on all participants. Etiological factors were examined only for subsamples of the participants. This allowed a wide range of potential risk factors to be explored. More remains to be done with the current data. In particular, examination of the linguistic properties of the language sample is ongoing.
The approach of trawling a wide range of factors in sub-samples allowed those that are associated with a high risk of persistence to be identified. Others that do not appear to be important for this purpose do not need to be collected further. The current results suggest most epidemiological factors are not relevant for risk of persistence, nor are many of the temperament dimensions and standard measures of auditory function. There are other factors that the literature suggests may also be relevant for predicting persistence, such as state anxiety, which has been reported to differ between persistent and recovered speakers at 12 plus (Davis et al., in press
). Anxiety could only be a risk factor for persistence if it was present at an early stage. To establish whether this is the case or not, suitable tests that can be employed for measuring children's anxiety would need to be developed; parental report tests have been used by others (Yairi & Ambrose, 2005
). Speakers who go on to persist in their stuttering would have to show higher anxiety levels than recovered speakers at the earlier age.
After the set of risk factors has been defined and data obtained, their success at predicting persistence and recovery can be assessed retrospectively (on available data) and prospectively (on newly-collected data). One appropriate technique is SEM (Levine, Petrides, Davis, Jackson & Howell, 2005
) which allows models to be set up in terms of how variables relate to each other in the form of a path diagram. The pattern of correlations between variables can then be inspected to see whether the model is a reasonable fit to the data. SEM distinguishes observed variables (actually measured) from latent variables (hypothetical constructs). If EXPLAN were set up as an SEM model for predicting persistence versus recovery, the latent variables would be: 1) linguistic planning; and 2) motor programming and execution. Language factors that are measured should correlate together and with the planning latent variable. Similarly, motor factors should correlate together and with the execution latent variable. According to EXPLAN, fluency problems arise from poor coordination between the two processes inherent in the latent variables. Thus observed variables that affect both latent variables (correlate with planning and execution) should occur. One method for locating variables that affect both latent variable would be to insert a phrase in a syntactically easy or complex frame (Kleinow & Smith, 2000
) and assess variability in how the utterance is spoken at two speech rates (Smith & Kleinow, 2000
). The condition where linguistic and motor processing demands are high should correlate with planning and execution variables. The condition where linguistic demand is high but motor demand is low should correlate with the planning variable alone. The condition where linguistic demand is low but motor demand is high should correlate with the execution variable.
Specification of an SEM model also focuses attention on what observable output variables are appropriate for determining the state of stuttering as persistent or recovered). Some authors advocate speech measures as paramount (Wingate, 2001
), whereas clinicians often emphasize the importance of a person's self esteem in speaking situations. Some fundamental questions about outcome measures can be addressed using SEM techniques: Can you have cases where self esteem is improved but speech performance is not and vice versa? Does persistent stuttering affect one of these variables (speech performance) and recovered forms affect the other (self esteem)?
Apart from modeling persistence and recovery, these data are also valuable for: 1) sub-typing different forms of stuttering; and 2) comparison with other groups of individuals reported to be stuttering. An example of the former would be to divide the sample into stutterers and clutterers and see whether they show the same pattern of epidemiology, symptomatology and etiology to answer questions about the relationship between these forms of dysfluent speech. Examples of the latter would be comparison with individuals whose primary home language is not English to determine whether they are truly stuttering.
The main limitation in these data is that many of the tests cannot be conducted at early and late ages. This was mentioned in connection with anxiety, where there are no tests suitable for young children. Similar limitations apply to sensory and motor tasks. Availability of tests that can be carried out at a range of ages would allow research questions raised by this study to be examined. For example, they could be used to address why recovered speakers acquired a problem in past pointing (a complex movement task). Hypothetically, it is possible that maintaining fluency focuses attention on speech control. This may be at the expense of other complex movement tasks like past pointing. This possibility is worth further investigation as, if it is correct, it would indicate that achieving fluency can affect movement complexity tasks.
The current study has also been limited to assessment of persistence and recovery. All individuals had treatment to keep this factor constant. Thus, there were no individuals who had no treatment. Therefore, the impact of treatment versus lack of it on recovery rates cannot be determined.
Clinicians may monitor the following six factors (identified in the current study) that put a speaker at risk of persisting in stuttering: 1) gender; 2) SSI-3 scores (those of persistent stutterers were higher than those who recovered); 3) part-word repetition rate and how it changes over time (this increased in persistent speakers); 4) persistent speakers had more adaptable temperaments than recovered speakers at intake; 5) the persistent speakers had worse backward masking thresholds than the recovered speakers at 12 plus; 6) balance with arms by the side was more variable in recovered than persistent speakers at 12 plus. Bear in mind that factors 4-6 were identified from randomly selected subsets of the overall sample rather than the entire sample.
At the first recording all children scored 16 or higher on the SSI-3, and 72 out of the 76 (95%) scored 21 or higher, which would be classified as moderate stuttering using SSI-3. The demographics section showed that the children were, generally speaking, well past the age where they reported stuttering had started and that chance of persistence did not depend on age group. It is surprising that these children attended late at clinic given that they clearly had a problem, as revealed by the SSI-3 scores at their first attendance and the earlier they attended the less time they stuttered. The late attendance may be, as Yairi and Ambrose (2005)
observed, due to the oft-repeated cliché: leave stuttering alone and it will go away. Progressive speech, motor, sensory and temperamental changes occur. The main lesson for the pathologist is, then, how to get the message over not to delay treatment.
It seems important to attempt a comparison of the present results for older children who stutter, with those of Yairi and Ambrose (2005)
for young, preschool-age children. To begin, the children in current study all received treatment, as did 89.5% of the participants in Yairi and Ambrose. In both studies, children who persisted did not respond to treatment. For obvious reasons, it is not known whether the persistent children in the present study would have recovered if they had received treatment earlier. In addition, it is not known whether some of the persistent children in Yairi and Ambrose's study showed unassisted recovery at later ages. The most cautious assumption, given the different possible pathways that persistence and recovery may take in older and younger children, is that the factors that put a child at risk for persistent stuttering apply at all ages.
It is also worth considering how both age and developmental changes across the children in both studies, along with the risk factors for persistence and recovery may have influenced the outcome of the treatment received. One possibility assumes that divergence between recovered and persistent children occurs at older ages (i.e. the present sample), but does not differ in form from that reported by Yairi and Ambrose for younger children. In the present study, children who recovered across the range of ages exhibited similar proportions of stalling and advancing dysfluency types. In this sense, they did not change their pattern of stuttering. Conversely, children who persisted changed their patterns of stuttering over time, as did the younger children in Yairi and Ambrose's study.
The observation that changes in stuttering type and proportion over time characterize persistence may suggest that protracted and perhaps habituated production of stuttering leads to adaptations in the central nervous system, and that such adaptations reduce the chance of recovery, either naturally or as the result of therapy (e.g., Howell et al., 2000
). Such adaptations may arise from the cerebellum which has traditionally been linked to sensory-motor integration (Stein & Glickstein, 1992
). Adaptations in other areas of the brain, such as the arcuate fasciculus (Watkins, Smith, Davis & Howell, in press
), might also potentially be involved. It may be the case that adaptive modifications in central nervous system functioning occurring up to and beyond age 12 yield more intransigent stuttering behavior that requires different treatments from those that are effective for younger ages. Reports in the clinical literature suggest that this is the case.
A second possibility is that a positive family history of persistent stuttering is associated with a genetic predisposition for no, or limited, response to treatment. Presently, this speculation is in need of empirical support. It is possible that children in families with a history of persistence may exhibit temperament, speech and central nervous system behaviors and adaptations that interact through family history, and that therapy specifically designed for such intransigent forms is required.