Predicting a Successful Response to Oral Appliance Therapy: Advancing Knowledge One Model at a Time

Aaron B. Holley MD; Joseph W. Dombrowsky MD

摘要

When prescribing an oral appliance (OA) for the treatment of obstructive sleep apnea (OSA), the American Academy of Sleep Medicine (AASM) and American Academy of Dental Sleep Medicine (AADSM) guidelines recommend providers use a custom-made OA, fabricated by a licensed dentist.1 The device and the dental appointments associated with fabrication and titration are expensive, there are logistical challenges with reimbursement, and treatment delays while the device is being titrated. A recent review estimated that OAs provide a complete response (apnea-hypopnea index [AHI] 5 events/h) only 48% of the time.2 Ideally, sleep medicine providers would have a method for identifying those who would be sucessfully treated with an OA prior to fabrication, so that the monetary and time costs could be avoided when treatment is unlikely to be effective.The clinical practice guideline published by the AASM and AADSM recommends OA treatment for patients who prefer an OA or are intolerant of positive airway pressure (PAP) therapy.1 In clinical practice, the sleep medicine provider should also consider polysomnography data, clinical variables, patient occupation, physical examination findings, and likelihood of PAP adherence. Incidentally, some of the predictors for low PAP adherenceyounger age, lower body mass index (BMI) and AHI3are also associated with OA success.That said, predicting OA response is difficult. Individual variables associated with success include AHI, BMI, Mallampati score, nasopharyngoscopy, spirometry, and craniofacial photography, among others.1,2 As is often the case, no one variable performs well enough to impact clinical decision making, so investigators have turned to integrating multiple variables via statistical modeling to improve performance. Identifying the optimal model is not easy. The existing literature consists of multiple derivation studies without successful validation in an external patient sample. These studies lack consistency in the definition used to define OA response. They rarely include overlapping variables, so we often cannot tell whether a given predictor improves the performance of those previously identified. Authors of a recent systematic review attempted to synthesize the existing data, but of the 17 studies they analyzed, only two used the same index test, methodology, and diagnostic thresholds.4 In short, the derivation studies rarely include the same predictors (independent variables) or outcome definition (dependent variable), and we do not have external validation. This makes choosing the right model, and improving on what we know, hard to do.Enter the study by Sutherland et al., published in this issue of the Journal of Clinical Sleep Medicine.5 The authors hypothesized that combining multiple, awake assessments of upper airway function would produce an optimized model for predicting OA response. The multimodal phenotyping assessments they studied were nasopharyngoscopy, spirometry, and craniofacial photography. They also assessed the effects that age, BMI, sex, AHI, and waist and neck circumference have on model performance. They included three separate definitions for OA response and created a separate model for each. They concluded that awake multimodal phenotyping assessments do not improve predictive accuracy when added to models based on clinical variables alone. Therefore, there is no reason to use nasopharyngoscopy, spirometry, or craniofacial photography when deciding whether to prescribe an OA.Because it is negative, this study is unlikely to change clinical practice. However, that is precisely why it is important. The assessments they studied performed well. The authors could have stopped there, submitted for publication, and attempted to convince us that the awake assessments they did should be considered in clinical practice. This would be very misleading, but we would not necessarily know that. We would have no way of judging their model against the others that exist, and clinical practice guidelines would continue to limit their recommendations given the poor quality of existing evidence.Instead, they took the additional step of testing whether awake multimodal phenotyping improves on what we already know, by adding each variable one at a time to a model with clinical predictors. They made sure to include several definitions for OA response as the dependent variable for their models. Instead of finding that awake multimodal phenotyping predicts OA response by one specific definition, the authors proved these tests add nothing to standard clinical variables, no matter how OA response is defined. Moving forward, investigators can switch their focus away from nasopharyngoscopy, spirometry, and craniofacial photography.We still do not know which model to use, and all models lack external validation. However, the authors, and the editors at the Journal of Clinical Sleep Medicine, deserve credit for publishing a negative study. Unfortunately, far too often we

机译：在开具用于治疗阻塞性睡眠呼吸暂停（OSA）的口腔用具（OA）时，美国睡眠医学学会（AASM）和美国牙科睡眠医学学会（AADSM）指南建议提供者使用由以下人员制造的定制OA： 1该设备以及与制造和滴定相关的牙科预约很昂贵，在报销方面存在后勤方面的挑战，并且在滴定该设备时会延误治疗。最近的一项评估估计，OA仅在48％的时间内可提供完全缓解（呼吸暂停-低通气指数[AHI] 5个事件/小时）。2理想情况下，睡眠药物提供者应具有一种方法，可以识别出那些将被成功治疗的人。在制造之前进行OA，以便在不太可能有效的治疗方法时避免金钱和时间成本.AASM和AADSM公布的临床实践指南建议对喜欢OA或不耐受气道正压的患者进行OA治疗（ PAP疗法。1在临床实践中，睡眠医学提供者还应考虑多导睡眠图数据，临床变量，患者职业，体格检查结果以及PAP坚持的可能性。顺便说一句，低年龄段的PAP依从性较低，体重指数（BMI）和AHI3的某些预测因素也与OA成功有关，也就是说，预测OA反应很困难。与成功相关的各个变量包括AHI，BMI，Mallampati评分，鼻咽镜检查，肺活量测定和颅面摄影等1,2。在通常情况下，没有一个变量的表现足以影响临床决策，因此研究者转向通过统计建模对多个变量进行积分以提高性能。确定最佳模型并不容易。现有文献包括多个衍生研究，但未在外部患者样本中成功验证。这些研究在用于定义OA反应的定义中缺乏一致性。它们很少包含重叠的变量，因此我们通常无法分辨给定的预测变量是否可以提高先前确定的预测变量的性能。最近的系统评价的作者试图综合现有数据，但是在他们分析的17个研究中，只有两个使用相同的指标检验，方法和诊断阈值。4简而言之，推导研究很少包含相同的预测变量（独立变量））或结果定义（因变量），我们没有外部验证。这使得选择正确的模型以及改进我们所知的方法变得困难。进入Sutherland等人的研究，该研究发表在本期《临床睡眠医学杂志》上。5作者假设结合了多个清醒的评估上呼吸道功能将产生用于预测OA反应的优化模型。他们研究的多模式表型评估是鼻咽镜检查，肺活量测定和颅面摄影。他们还评估了年龄，BMI，性别，AHI以及腰围和颈围对模型表现的影响。他们为OA响应包括了三个单独的定义，并为每个创建了一个单独的模型。他们得出结论，当将清醒的多模式表型评估添加到仅基于临床变量的模型中时，并不能提高预测准确性。因此，在决定是否开OA时无需使用鼻咽镜检查，肺活量测定或颅面摄影检查，因为它是阴性的，因此该研究不太可能改变临床实践。但是，这就是为什么它很重要的原因。他们研究的评估表现良好。作者可能会在那里停下来，提交发表，并试图说服我们在临床实践中应考虑他们所做的清醒评估。这将是非常误导的，但我们不一定知道这一点。我们无法将他们的模型与其他模型进行比较，鉴于现有证据的质量较差，临床实践指南将继续限制他们的建议，相反，他们采取了额外的步骤来测试清醒的多峰表型是否可以改善我们的研究成果。已经知道，通过一次将每个变量添加到具有临床预测变量的模型中。他们确保将OA响应的几个定义作为其模型的因变量。作者证明，无论如何定义OA响应，这些测试并没有发现清醒的多峰表型可以通过一个特定的定义预测OA响应，而是证明这些测试对标准临床变量没有任何影响。展望未来，研究人员可以将重点从鼻咽镜检查，肺活量测定和颅面摄影术转移出去。我们仍然不知道使用哪种模型，并且所有模型都没有外部验证。但是，《临床睡眠医学杂志》的作者和编辑因发表负面研究而值得赞扬。不幸的是，我们经常

Predicting a Successful Response to Oral Appliance Therapy: Advancing Knowledge One Model at a Time

摘要

著录项

相似文献

相关主题

期刊订阅