Developmental Stuttering Screening Instrument: evidence of sensitivity and accuracy measures

ABSTRACT Purpose: to verify the sensitivity and accuracy measures of the Developmental Stuttering Screening Instrument (DSSI). Methods: the DSSI was administered to 30 parents/guardians of children aged 2 to 5 years and 11 months with and without complaint of stuttering. The instrument administration was timed. The sensitivity analysis used the Weight of Evidence (WoE) binary classification model to verify the strength level of the items. The cutoff scores were established with grouping analysis with the k-means cluster method, based on the minimum and maximum values of each identified group’s scores. The data were analyzed with the SPSS statistical software (version 20.0) and were considered significant with p ≤ 0.05. Results: the interviews lasted an overall mean of 17 minutes. The WoE model revealed that the items with the greatest predictive strength for risk of stuttering were the social reaction to their speech, the physical concomitants, and the comprehension of the child’s speech. The correspondence analysis showed a strong association between “having complaints” and “high total score”, as well as between “not having complaints” and “low total score”, indicating that the parents’ complaints are a factor that leads to high scores in the instrument. “Sex” had little predictive effect for risk. The grouping analysis enabled the stratification of subjects into three risk levels: “not at risk”, “under observation”, and “at risk”. Conclusion: the instrument presented the first evidence of sensitivity and accuracy measures, thus, making the identification of risk of developmental stuttering in preschoolers, possible.摘要 目的：验证发育性口吃筛查工具（Developmental Stuttering Screening Instrument，DSSI）的敏感性与准确性指标。方法：选取30名年龄介于2岁至5岁11个月、有/无口吃主诉的儿童的家长或监护人作为研究对象，对其施测DSSI并记录施测时长。敏感性分析采用证据权重（Weight of Evidence，WoE）二分类模型检验各条目的强度水平；临界得分通过k-means聚类分组分析确定，基于各分组得分的最小值与最大值。采用SPSS统计软件（版本20.0）进行数据分析，以p≤0.05作为显著性判定标准。结果：所有施测访谈的总平均时长为17分钟。WoE模型显示，对口吃风险预测能力最强的三个条目分别为对儿童言语的社交反应、言语伴随的躯体动作以及儿童言语的可理解性。对应分析结果表明，“存在口吃主诉”与“总得分较高”存在强关联，“无口吃主诉”与“总得分较低”亦存在强关联，提示家长的主诉是导致工具得分偏高的影响因素。“性别”对口吃风险的预测作用较弱。聚类分析将研究对象划分为三个风险层级：“无风险”“需观察”与“有风险”。结论：本研究首次为DSSI提供了敏感性与准确性指标的实证依据，使其可用于学龄前儿童发育性口吃风险的识别。