SCIENTIFIC INVESTIGATIONS A Feedback-Controlled Mandibular ... · 871 Journal of Clinical Sleep Medicine, Vol. 13, No. 7, 2017 Study Objectives: Mandibular protruding oral appliances

871 Journal of Clinical Sleep Medicine, Vol. 13, No. 7, 2017

Study Objectives: Mandibular protruding oral appliances represent a potentially important therapy for obstructive sleep apnea (OSA). However, their clinical utility is limited by a less-than-ideal efficacy rate and uncertainty regarding an efficacious mandibular position, pointing to the need for a tool to assist in delivery of the therapy. The current study assesses the ability to prospectively identify therapeutic responders and determine an efficacious mandibular position.Methods: Individuals (n = 202) with OSA participated in a blinded, 2-part investigation. A system for identifying therapeutic responders was developed in part 1 (n = 149); the predictive accuracy of this system was prospectively evaluated on a new population in part 2 (n = 53). Each participant underwent a 2-night, in-home feedback-controlled mandibular positioner (FCMP) test, followed by treatment with a custom oral appliance and an outcome study with the oral appliance in place. A machine learning classification system was trained to predict therapeutic outcome on data obtained from FCMP studies on part 1 participants. The accuracy of this trained system was then evaluated on part 2 participants by examining the agreement between prospectively predicted outcome and observed outcome. A predicted efficacious mandibular position was derived from each FCMP study.Results: Predictive accuracy was as follows: sensitivity 85%; specificity 93%; positive predictive value 97%; and negative predictive value 72%. Of participants correctly predicted to respond to therapy, the predicted mandibular protrusive position proved efficacious in 86% of cases.Conclusions: An unattended, in-home FCMP test prospectively identifies individuals with OSA who will respond to oral appliance therapy and provides an efficacious mandibular position.Clinical Trial Registration: The trial that this study reports on is registered on www.clinicaltrials.gov, ID NCT03011762, study name: Feasibility and Predictive Accuracy of an In-Home Computer Controlled Mandibular Positioner in Identifying Favourable Candidates for Oral Appliance TherapyKeywords: home sleep test, obstructive sleep apnea, oral appliance therapy, predictive accuracy, sleep-disordered breathingCitation: Remmers JE, Topor Z, Grosse J, Vranjes N, Mosca EV, Brant R, Bruehlmann S, Charkhandeh S, Jahromi SA. A feedback-controlled mandibular positioner identifies individuals with sleep apnea who will respond to oral appliance therapy. J Clin Sleep Med. 2017;13(7):871–880.

INTRODUCTION

Obstructive sleep apnea (OSA) is a chronic disease caused by repeated narrowing or closure of the pharynx during sleep. The disease compromises quality of life,1,2 increases risk of accidents,3–5 and is associated with an increased risk of car-diovascular events6 and reduced survival rate.7,8 Continuous positive airway pressure (CPAP), the standard therapy for this disease, is efficacious and benign. However, long-term compli-ance with CPAP is low,9,10 and this may contribute to its inef-fectiveness in preventing adverse cardiovascular outcomes.11–14 Oral appliance therapy is an alternative therapy that protrudes the mandible, thereby opening the pharynx.15–17 Though indi-viduals generally prefer oral appliances, their efficacy rate is less than ideal.9,18–24

Prospective selection of individuals who respond to oral ap-pliance therapy should increase the clinical utility of the ther-apy, and research has focused on predicting outcome with oral appliance therapy.25,26 Disease severity, body position during sleep, age, sex, weight, and other anatomical features and clini-cal variables have been investigated and found to correlate with outcome.26–37 Similarly, encouraging results have been reported

SCIENTIF IC INVESTIGATIONS

A Feedback-Controlled Mandibular Positioner Identifies Individuals With Sleep Apnea Who Will Respond to Oral Appliance TherapyJohn E. Remmers, MD1,2; Zbigniew Topor, PhD1,2; Joshua Grosse, MMath2; Nikola Vranjes, DDS3; Erin V. Mosca, PhD2; Rollin Brant, PhD4; Sabina Bruehlmann, PhD2; Shouresh Charkhandeh, DDS2,3; Seyed Abdolali Zareian Jahromi, PhD1,2

1University of Calgary, Calgary, Canada, 2Zephyr Sleep Technologies, Calgary, Canada, 3The Snore Centre, Calgary, Canada, 4University of British Columbia, Vancouver, Canada

pii: jc-17-00007 ht tp://dx.doi.org/10.5664/jcsm.6656

with pharyngeal imaging and nasoendoscopy.29,38–41 However, none of these approaches has been shown in prospective stud-ies to identify individuals who will experience therapeutic suc-cess with oral appliance therapy. Use of a remotely-controlled mandibular positioner (RCMP) used during sleep has been demonstrated to accurately identify such individuals.42–46 This test uses temporary dental trays and is carried out in a polysom-nographic setting, where a technologist advances the mandible

BRIEF SUMMARYCurrent Knowledge/Study Rationale: Oral appliances that protrude the mandible are a potentially important therapy for the treatment of obstructive sleep apnea. However, only 50% to 70% of individuals with sleep apnea experience a satisfactory therapeutic response to the therapy. Hence, we need a test that prospectively identifies therapeutic responders.Study Impact: The current study describes a feedback-controlled mandibular positioner for use at home and demonstrates that it prospectively identifies individuals with obstructive sleep apnea who will respond to oral appliance therapy. Additionally, the feedback-controlled mandibular positioner test provides a mandibular position that is efficacious in almost all responders.

872Journal of Clinical Sleep Medicine, Vol. 13, No. 7, 2017

JE Remmers, Z Topor, J Grosse, et al. In-Home Test to Identify Oral Appliance Responders

to eliminate respiratory events (REs). Although the test appears to be safe and well tolerated, its cost and inconvenience com-promise its clinical utility. Addition of computerized feedback control to the aforementioned system would enable a compara-ble unattended home test that might circumvent the deficiencies associated with the polysomnographic test.

We have developed a feedback-controlled mandibular posi-tioner that identifies REs in real time and protrudes the man-dible according to predetermined algorithms. We have carried out unattended tests on a group of individuals with OSA while asleep. Data derived from these tests, combined with knowl-edge of therapeutic outcome with a custom oral appliance, were used to train a machine learning system to predict therapeutic outcome. The trained system was then prospectively evaluated on a separate group of individuals. The current study describes the performance and accuracy of the titration/prediction sys-tem in predicting therapeutic outcome and in determining an efficacious mandibular position.

METHODS

GeneralThis study was approved by the Conjoint Health Research Eth-ics Board of the University of Calgary, and informed consent was obtained from all participants. Participants were recruited from the population of patients seen by the principal investiga-tor or other sleep physicians at the outpatient clinic of the Foot-hills Medical Center Sleep Disorders Center (Calgary, Alberta, Canada), and by the dental co-investigators at The Snore Centre

(Calgary, Alberta, Canada). Most of the individuals referred to the recruitment sites were referred for snoring or suspected sleep-disordered breathing; all potential study participants re-ceived an in-home cardiorespiratory evaluation during sleep (Snore Sat Recorder, Sagatech Electronics, Inc., Calgary, Al-berta, Canada) to document their oxygen desaturation index (ODI) and oxyhemoglobin indices. All study participants were required to meet the following eligibility criteria: ability to un-derstand and provide informed consent, age between 21 and 80 years, OSA (ODI > 10 events/h), body mass index (BMI) < 45 kg/m2, neck circumference < 50 cm, absence of severe oxyhe-moglobin desaturation during sleep (indicated by a mean oxy-hemoglobin saturation > 90%), mandibular range of motion > 5 mm, adequate dentition (10 upper and 10 lower teeth), ability to breathe comfortably through the nose, absence of central sleep apnea, absence of anticipated changes in medical therapy that could alter OSA severity during the study, absence of antici-pated change in body weight (≥ 5%) during the study, absence of symptomatic nonrespiratory sleep disorders (eg, restless legs syndrome, chronic insomnia), absence of severe respira-tory disorder(s) other than sleep-disordered breathing, and ab-sence of loose teeth or advanced periodontal disease.

A 2-part investigation was performed involving 2 separate groups of participants. Part 1 was carried out on a first group of participants (n = 149) to develop a predictive method. Part 2 was carried out on a second, different group of participants (n = 53) to evaluate prospectively the accuracy of the predic-tive method. Although the participants in part 1 and part 2 were different individuals, the procedures and protocol were the same in both parts of the study. Specifically, all partici-pants underwent a 2-night feedback-controlled mandibular po-sitioning (FCMP) study, and all received a custom treatment appliance. All participants underwent outcome testing with the treatment appliance in the mouth for 2 nights in the home using a validated monitor47 that automatically calculates a 4% ODI. Therapeutic success or failure was defined as a mean ODI value from the 2 outcome studies of less or greater than 10 events/h, respectively.

Feedback-Controlled Mandibular PositionerThe FCMP system received and processed nasal airflow and oxyhemoglobin saturation (O2 Sat) signals and identified REs in real time at the participant’s bedside. Signal processing was performed by a laptop computer that formed part of the FCMP system (Figure 1). O2 Sat was derived from a pulse oximeter (MS-2040 Masimo Corporation, Irvine, California) using a finger hood probe. Nares pressures were recorded separately from each naris using a dual catheter system, with each naris cannula leading to a separate pressure transducer. Airflow in each naris was calculated as the square root of naris pressure, with total respiratory airflow being the sum of the 2 airflows. Tidal volume was calculated as the integral of inspiratory air-flow, and breath-by-breath ventilation as the ratio of tidal vol-ume to respiratory period. REs were detected in real time and were stratified in relation to the magnitude of decrease in O2 Sat and ventilation by the FCMP device.

The FCMP device also included a computer-controlled man-dibular positioner consisting of a linear actuator attached to

Figure 1—Feedback-controlled mandibular positioner device.

The feedback-controlled mandibular positioner device was composed of a laptop computer for signal processing, a pulse oximeter and nasal cannula to collect physiological parameters of oxygen saturation and airflow, and a mandibular positioner attached to temporary dental trays to move the mandible in response to respiratory events.



temporary dental trays (Figure 1). The mandibular positioner and temporary dental trays were identical to those used in pre-vious studies of the RCMP device.42–45 Briefly, upper and lower trays were filled with impression material and attached by an-terior brackets to a linear actuator that moved the mandible in the anterior-posterior axis. The FCMP study comprised 2 full-night sequences. At the beginning of each night, the test was initiated with the dental trays set at the participant’s habitual bite and was allowed to protrude only to the previously mea-sured maximum voluntary protrusion for the participant. At the test start, or following interruption of operation by the par-ticipant, the mandible was held at a static position for 30 min-utes before the controlling algorithms began active adjustment.

The first night entailed a dynamic interaction between the controller and observed REs. The size and timing of the pro-trusive step depended on the number and grade of REs, and ranged from 0.2 mm of movement after a 3-minute delay to 1.4 mm of immediate movement. Successive protrusive move-ments were separated by at least 3 minutes with the distance moved being the sum of all singular steps associated with in-dividual REs detected during this period. Each singular step was 0.2 mm with a minimum 5-second delay to ensure a slow rate of adjustment. If no REs were observed during a particu-lar 3-minute interval, small perturbations were carried out in which the mandible was protruded in two 0.3-mm steps sepa-rated by a 2- to 5-second delay, then returned 0.6 mm to the prior position if the preceding protrusion did not result in a significant increase (≥ 30%) in respiratory airflow. Conversely, the mandible was retruded 0.6 mm and held at that position if the airflow did not decrease. If the retrusive step caused a

decrease in airflow, the mandible was returned to the position prior to the retrusive step. Perturbation activity was abandoned upon detection of an RE. The results of this first study night were used to automatically calculate a provisional prediction of oral appliance therapy outcome and a predicted efficacious protrusive position at which to start the next night of study.

The second night was a quasi-static study involving smaller, less frequent adjustments in mandibular position in order to evaluate the efficacy of the predicted efficacious protrusive po-sition derived from the dynamic study. The movement of the mandible was restricted at or near the provisionally predicted efficacious protrusive position in order to specifically predict outcome at that position. The mandible was slowly moved from the individual’s habitual bite to the predicted efficacious pro-trusive position derived from the dynamic study night, and a continuous ODI calculation (number of REs/elapsed time) com-menced in real time. Once the predicted efficacious protrusive position was reached, the mandible was not moved unless the ODI exceeded 10 events/h for 2 hours or more. In that case, the mandible was advanced 0.6 mm and the continuous ODI calculation was reinitiated. The mandible was held in this posi-tion or advanced 0.6 mm if the ODI value exceeded 10 events/h after 1 hour, and so on. Though the FCMP test was designed to find an efficacious protrusive position, not necessarily the minimum efficacious protrusive position, the system automati-cally prompted a third study night at a starting protrusive value of 70% if the results from the second study night predicted an efficacious protrusive position > 80%. This was done in an at-tempt to find a lower efficacious protrusive position than the one tested during the second study night.

Figure 2—Classification error versus number of trees in the random forest.

Classification error is a prediction of the accuracy of the random forest when it is applied to a population other than the one it was trained on. Note that the error stabilizes at approximately 250 trees; 400 trees were used to optimize accuracy.



Machine Learning SystemA random forest (RF) machine learning method was developed in part 1 of the investigation. The RF machine was trained by providing data from the FCMP study carried out on each part 1 participant, together with the binary outcome for that partici-pant, ie, therapeutic success or failure derived from the home sleep study with the oral appliance in place. Specifically, 266 features were extracted from both nights of each FCMP study. The RF machine uses multiple decision trees, and we evaluated the dependence of classification error on the number of trees using the “out of bag” method.48 Figure 2 shows the relation between classification error and number of trees. A 400-tree RF system was selected as this number provided a classifica-tion error that was clearly at a minimum value. In order to pre-dict the overall outcome, each of the 400 decision trees gave a prediction of either therapeutic success or failure. All decision trees were polled, and the majority poll determined the final

prediction for that study. Therapeutic success was predicted if more than 50% of decision trees predicted success and thera-peutic failure if fewer than 50% of decision trees predicted suc-cess. Prediction of therapeutic outcome was performed solely by the RF machine, without human intervention, based only on data derived from the FCMP test.

The internal predictive accuracy of the trained RF machine was assessed by three tenfold cross-validation tests applied to part 1 participants. The external predictive accuracy of the trained RF machine was evaluated by applying it to the FCMP test data from part 2 participants, none of whom had partici-pated in part 1. The observed outcomes of part 2 participants were not supplied to the machine. Rather, using the FCMP data alone, the RF machine prospectively predicted the thera-peutic outcome (success or failure) for each part 2 participant. Each prediction was then compared to the actual therapeutic outcome (success or failure) observed for that participant. A binary agreement analysis was carried out by calculating sen-sitivity, specificity, positive predictive value (PPV), and nega-tive predictive value (NPV).

Research ProtocolThe research protocol is outlined in Figure 3. Participants un-derwent a 2-night respiratory home evaluation during sleep with a Snore Sat Recorder to determine baseline ODI and vis-ited the dentist for fitting of the temporary trays and for deter-mination of the habitual bite and maximal voluntary protrusive position, as described previously.43 These readings were used as range of motion limits during the test. Before the night 1 FCMP test, a technician visited the home and instructed the participant on the use of the bedside equipment and self-appli-cation of the nasal cannula, O2 Sat probes, and temporary den-tal trays. On the morning following each study night, data from the FCMP system were uploaded, analyzed, and used to set the parameters for the next night of study. Upon completion of the study, the FCMP system automatically calculated a prediction of either therapeutic success (outcome ODI < 10 events/h) or failure (outcome ODI > 10 events/h) for each participant. For those individuals predicted to experience therapeutic success, the FCMP system automatically calculated a predicted effica-cious protrusive position.

After completing the FCMP test, each participant re-ceived a custom treatment appliance, either a SomnoDent G2 (SomnoMed, Sydney, New South Wales, Australia) in part 1 or a MicrO2 (ProSomnus Sleep Technologies, Pleasanton, California, United States) in part 2. Though the dentists were blinded to the FCMP prediction, they were provided with a target protrusive position for each participant’s custom oral appliance. The target was either the predicted efficacious pro-trusive position for those individuals who were predicted to experience therapeutic success, or, for individuals predicted by the FCMP test to experience therapeutic failure, a default value of 70% of the protrusive range. The custom oral appliances were set to the target protrusive position (either immediately or over a period of a few weeks if required) and a 2-night home sleep study with the custom appliance was performed. If the ODI value was < 10 events/h, the individual was considered a therapeutic responder. If the ODI value was > 10 events/h, the

Figure 3—Study flowchart for parts 1 and 2.

Overview of participant recruitment, flow of study events, and study endpoints. FCMP = feedback-controlled mandibular positioner; OA = oral appliance; ODI = oxygen desaturation index.



appliance was advanced progressively with periodic respira-tory home evaluations during sleep until therapeutic success was observed or maximum protrusion was reached.

RESULTS

As shown in Figure 3, 202 individuals meeting the acceptance criteria were recruited for the study: 149 in part 1 and 53 in part 2. Five individuals from part 1 and two individuals from part 2 had inconclusive FCMP tests (ie, no prediction) and declined further testing. Thirteen individuals from part 1 and three in-dividuals from part 2 were lost to follow-up and did not com-plete the study protocol. The final study population, therefore, consisted of 179 individuals who completed the study protocol: 131 in part 1 and 48 in part 2.

As shown in Table 1, participants were, on average, middle age, predominantly male, obese, and had moderate sleep ap-nea. The mean baseline physiological characteristics, includ-ing sex distribution, adjusted neck circumference,49 arterial oxygen saturation, or SaO2, time spent at < 90% SaO2, and sleep apnea scales (Epworth Sleepiness Scale,50 Sleep Apnea Quality of Life Index,51 and Pittsburgh Sleep Quality Index52) for participants in the 2 parts were comparable except that the mean BMI and mean baseline ODI were significantly greater in part 2 than part 1. A plot of participants’ baseline ODI in re-lation to their BMI, shown in Figure 4, reveals an overlapping distribution of points for participants in parts 1 and 2.

The overall oral appliance success rate was 76.3% for phase 1 and 70.8% for phase 2. The difference between the 2 val-ues was not statistically significant (P = .44; Fisher exact test). The distribution of baseline ODI in relation to BMI for thera-peutic responders and nonresponders (Figure 4) reveals that therapeutic outcome cannot clearly be predicted by the ODI/BMI relationship. Of the 101 obese (BMI > 30 kg/m2) partici-pants who completed either part of the study, 68.3% experi-enced therapeutic success (outcome ODI < 10 events/h). Of the 51 individuals with severe OSA (ODI > 30 events/h) who

completed either part of the study, 52.5% experienced thera-peutic success.

The final outcome response to custom oral appliance ther-apy for FCMP predicted responders and for FCMP predicted nonresponders in part 2 is shown in Figure 5. With one excep-tion, all predicted responders experienced therapeutic success (outcome ODI < 10 events/h), and 13 of 18 predicted nonre-sponders experienced therapeutic failure (outcome ODI > 10 events/h). These data are summarized in Table 2, showing

Table 1—Baseline data for participants in study parts 1 and 2.Part 1 (n = 131) Part 2 (n = 48) P Value

Sex Female: n = 21;Male: n = 110

Female: n = 8;Male: n = 40

.541†

Age (y) 49.8 (24–76) 48.4 (26–70) .409Adjusted neck circumference (cm) 48.0 (35.5–59.5) 49.1 (39–63) .207BMI (kg/m2) 30.3 (20.1–39.1) 33.3 (23.6–45.9) .001*Baseline ODI (events/h) 25.5 (10.5–65.1) 31.1 (10.3–74.6) .046*SaO2 (%) 92.1 (88.7–96.3) 91.9 (87.1–94.7) .242% of time at < 90% SaO2 13.0 (0.1–57.3) 16.5 (0.7–84.9) .148Epworth Sleepiness Scale score 9.7 (0–23) 9.7 (0–23) .986Sleep Apnea Quality of Life Index 4.6 (1.7–6.6) 4.3 (2.0–6.7) .131Pittsburgh Sleep Quality Index 8.1 (1–19) 8.1 (3–19) .965

* = statistical significance. † = Fisher exact test. Independent samples t tests used for remainder of variables. Data are presented as mean (range). Individuals with inconclusive feedback-controlled mandibular positioner tests and those lost to follow-up (not included in the table) had baseline characteristics similar to individuals who completed the study. BMI = body mass index, ODI = oxygen desaturation index, SaO2 = arterial oxygen saturation.

Figure 4—Distribution of ODI and BMI for all study participants (n = 179).

Participants who experienced therapeutic success (outcome ODI < 10 events/h) are represented by a closed circle for part 1 or a closed square for part 2, while those who experienced therapeutic failure (outcome ODI > 10 events/h) are represented by an open circle for part 1 or an open square for part 2. BMI = body mass index, ODI = oxygen desaturation index.



an overall error rate of 12% and the following statistical pa-rameters: sensitivity 85%; specificity 93%; positive predictive value 97%, and negative predictive value 72%. The machine learning system was re-trained on part 1 data using a crite-rion of therapeutic success of outcome ODI < 10 events/h and a 50% reduction from baseline ODI. When applied to part 2 data, the < 10 events/h and 50% predictive accuracy remained high, with the following statistical parameters: sensitivity 81%, specificity 94%, positive predictive value 96%, and negative predictive value 71%.

Figure 6 shows the relationship between predicted effica-cious protrusive position and the final efficacious protrusive position at outcome for part 2 participants predicted to be a therapeutic success. The median predicted efficacious protru-sive position for all predicted responders in part 2 was 63% of full protrusive range. The third night study was prompted on 15 participants (31%) and reduced the predicted efficacious

protrusive position in almost all cases (13 of 15; 87%) by an average of 2.5 ± 0.8 mm. Of the 29 participants who were cor-rectly predicted to be therapeutic successes, 1 participant was not tested at the predicted efficacious protrusive position due to a protocol discrepancy, 24 (86%) responded with an ODI < 10 events/h at the predicted efficacious protrusive position, and the remaining 4 participants required an additional 1 to 3 mm of protrusion to achieve therapeutic success. The median value for final efficacious protrusive position at outcome was 74% of full protrusive range. Efficacious protrusive position did not correlate with ODI (r2 = 0.11) or BMI (r2 = 0.17).

Over 500 nights of FCMP testing were completed during both parts of the current study. After each FCMP study, some participants reported having tooth and/or gum discomfort dur-ing the FCMP test. These symptoms were transient, however, and did not persist beyond a few hours. No significant safety concerns or adverse events were reported by the participants or study coordinators. All participants received a clinical evalua-tion by one of the study dentists within 3 weeks of completing the FCMP test. These examinations revealed no injury or dam-age to teeth, tongue, or mucosa, and no dental pain.

DISCUSSION

The current investigation supports the feasibility of an unat-tended sleep study using a feedback-controlled mandibular positioner. The results indicate that analysis of test data us-ing prospectively established decision methods predicts binary therapeutic outcome and identifies an efficacious protrusive position with reasonable accuracy.

Both part 1 and part 2 study populations displayed a broad range of baseline ODI values, with mild, moderate, and severe

Figure 5—Baseline and outcome ODI for part 2 participants.

Participants predicted by feedback-controlled mandibular positioner automatic prediction to experience therapeutic success (outcome ODI < 10 events/h) are shown in panel A, and those predicted to experience therapeutic failure (outcome ODI > 10 events/h) are shown in panel B. Of the 30 participants predicted to achieve therapeutic success, only 1 was a therapeutic failure. Of the 18 participants predicted to be therapeutic failures, 5 achieved therapeutic success. The horizontal dashed line depicts the border between therapeutic failure and therapeutic success. ODI = oxygen desaturation index.

Table 2—Sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) for part 2 data.

Predicted Success

Predicted Failure

Therapeutic Success 29 5 Sensitivity

85%Therapeutic

Failure 1 13 Specificity 93%

PPV 97% NPV 72%

Predictions of therapeutic success or failure were made automatically by the random forest machine using feedback-controlled mandibular positioner test data only.



OSA groups being comparably represented. Similarly, values of BMI were evenly distributed between 25 and 45 kg/m2. BMI and baseline ODI were significantly lower in part 1, where the RF was trained, than in part 2, where the predictive method was validated. That the predictive accuracy of the RF was high despite this difference in the training and validation popula-tions suggests that the method may be relatively insensitive to such population differences. The overall therapeutic suc-cess rates for the 2 populations, 71% and 76%, is somewhat greater than some published reports,21,24,53,54 but comparable to others.19,20,23,27,33

ODI was selected as the measure by which to assess OSA in all baseline and outcome studies, all of which were home sleep tests. Over the course of the study, more than 800 full-night tests were carried out (approximately 2 baseline and 2 outcome tests for each participant). This large number of tests dictated that we use automatic scoring in the interest of efficiency and reliability. The Snore Sat Recorder used in this study has been shown to correlate well with a polysomnogram-determined respiratory disturbance index (r2 = 0.97)47 suggesting that the recorder provides a reliable index of the full-night frequency of respiratory disturbances. Furthermore, the oxygen compo-nent used in the scoring of ODI is less prone to uncontrolled variability and noise than the ventilatory component used in the scoring of AHI. A comparison of both ODI and AHI (au-toscored by a home sleep recorder) to PSG data found oximetry to be more reliable than nasal airflow measurements.55

The accuracy of the FCMP in predicting oral appliance therapy outcome in the home is comparable to that previously reported for in-laboratory, attended studies using an RCMP.43,45 Specifi-cally, sensitivity and specificity for the 2 previous investigations of RCMP were 86% and 92%,43 and 82% and 93%, respectively.45 The values for FCMP in the current study were 85% and 93%, re-spectively. Compared to the previous studies, the PPV value from the current study was higher (97% versus 93%43 and 90%45) and the NPV was lower (72% versus 83%43 and 87%45). These differ-ences probably reflect the higher therapeutic success rate in the current study compared to the previous studies.43,45 The Wiscon-sin Sleep Cohort reveals that approximately 30% of all individu-als with sleep apnea are female,8 whereas our representation in the study was 16%. This may influence the predictive accuracy, and may or may not influence the overall oral appliance efficacy rate. A recent study by Sutherland et al. showed no correlation of sex with oral appliance outcome.26 We did not put a quota on the relative proportion of males and females recruited into the study; however, we acknowledge that the low number of females in our study population is a shortcoming.

RF machine learning appears to be a suitable method for predicting therapeutic outcome. This ensemble method uti-lizes values for a large number (n = 266) of weak predictors derived from the 2 nights of testing. These values are analyzed by a forest of 400 individual decision trees to arrive at a bi-nary prediction of therapeutic outcome. Because the method’s elemental analysis is a single decision tree classification, RF classification would appear to be intrinsically appropriate for predicting a binary outcome. The use of many variables and many decision trees should result in a more generalizable, ro-bust prediction method.

Individuals with mild and supine predominant OSA have been reported to be favorable candidates for oral appliance therapy.26,29,30,44 Additional individual-specific factors includ-ing younger age, female sex, and moderate obesity (charac-terized by lower BMI and neck circumference) may predict a successful outcome with oral appliance therapy.22,27,29,30,39 How-ever, complete resolution of OSA by oral appliance therapy has been reported in individuals with severe OSA and obesity.43,54,56 Craniofacial features including shorter soft palate length,35,57,58 higher hyoid bone position,58,59 greater angle between the cra-nial base and mandibular plane,34 and retrognathic mandible31 have also been reported to predict favorable outcome of oral appliance therapy. However, a recent review noted that cepha-lometric parameters alone are not able to reliably predict out-come of oral appliance therapy.60 A recent study by Edwards et al. has demonstrated the importance of control system char-acteristics, such as loop gain, in explaining treatment failure of oral appliance therapy,61 indicating that factors other than anatomical features play a role in explaining oral appliance treatment failure.

New techniques such as nasoendoscopy during drug-induced sleep or while awake,40,41 cone-beam computed tomography scans,38,62 multisensor catheters,27,39 and nasopharyngoscopy29 that assess anatomical and functional characteristics of up-per airways may ultimately prove to provide clinical utility. Currently, however, none have been tested prospectively and none would appear to convey the predictive accuracy, conve-nience, and applicability documented in the current study for the FCMP test.

Another proposed approach for predicting suitability of oral appliance therapy is multinight self-titration using an

Figure 6—Efficacious versus predicted target protrusive positions.

Positions are represented as a percent of full protrusion. Data points that fall on the diagonal line represent participants whose target protrusive position was efficacious. The dashed horizontal line depicts the median predicted efficacious protrusion (63%). The shaded area denotes participants with a correctly predicted target protrusion less than 70% who would have been overprotruded with a common titration protocol.



adjustable temporary appliance.63,64 Here the appliance is set at a fixed protrusion for 1 or more nights, and advanced in pre-selected steps between nights in response to individual and/or bed partner report. Such a method would seem to suffer from variability introduced by numerous multinight comparisons, from placebo effects, and from the intrinsic unreliability and inconvenience of such reporting. Currently, no published re-ports document the predictive accuracy of the method.

Prediction of an efficacious mandibular position is an aspect of a feedback-controlled mandibular positioning test that may contribute to clinical utility. Our results support the notion that the predicted protrusive position provided by the FCMP test is efficacious in most cases. In addition to improving efficiency of delivering oral appliance therapy, provision of an effective protrusive position derived from the FCMP test may reduce the long treatment time associated with self-titration. Further, for a substantial number of participants, the test identified an effective therapeutic position at protrusive distances less than 70% of full protrusion. This implies that use of the efficacious protrusive position may lessen the final protrusive distance compared to traditional approaches, many of which begin pro-trusive adjustment of the custom treatment appliance at 70% of full protrusion. Thus, information derived the FCMP study may improve tolerance and reduce side effects associated with unnecessarily high protrusions.

In conclusion, the results of the current investigation dem-onstrate feasibility and reasonable predictive accuracy of an in-home sleep test performed with a feedback-controlled man-dibular positioner. The test appears to have the capability to predict a binary outcome of oral appliance therapy as well as an efficacious mandibular position. The only other test, to our knowledge, shown prospectively to accurately make these predictions is the antecedent test, the RCMP in the polysom-nographic setting.43 Although the antecedent test requires identification of sleep stages and body position to predict out-come of oral appliance therapy, the feedback-controlled system achieves comparable predictive accuracy using only informa-tion related to respiratory status and mandibular position. This likely reflects the design of the 2-night, dynamic FCMP test where all variables, including sleep stage and body position, are appropriately and automatically taken into account by the machine learning method. The feasibility and predictive ac-curacy of the feedback-controlled mandibular positioner home sleep test described in the current report suggest that such a test may usefully contribute to clinical management of OSA.

ABBRE VI ATIONS

BMI, body mass indexCPAP, continuous positive airway pressureFCMP, feedback-controlled mandibular positionerNPV, negative predictive valueO2 Sat, oxyhemoglobin saturationODI, oxygen desaturation indexOSA, obstructive sleep apneaPPV, positive predictive valueRCMP, remotely-controlled mandibular positioner

REs, respiratory eventsRF, random forestSaO2, arterial oxygen saturation

REFERENCES

1. D’Ambrosio C, Bowman T, Mohsenin V. Quality of life in patients with obstructive sleep apnea: effect of nasal continuous positive airway pressure. Chest. 1999;115(1):123–129.

2. Engleman HM, Douglas NJ. Sleep. 4: Sleepiness, cognitive function, and quality of life in obstructive sleep apnoea/hypopnoea syndrome. Thorax. 2004;59(7):618–622.

3. George CFP. Sleep. 5: Driving and automobile crashes in patients with obstructive sleep apnoea/hypopnoea syndrome. Thorax. 2004;59(9):804–807.

4. Horstmann S, Hess CW, Bassetti C, Gugger M, Mathis J. Sleepiness -related accidents in sleep apnea patients. Sleep. 2000;23(3):383–389.

5. Tregear S, Reston J, Schoelles K, Phillips B. Obstructive sleep apnea and risk of motor vehicle crash: systematic review and meta-analysis. J Clin Sleep Med. 2009;5(6):573–581.

6. Shamsuzzaman ASM, Gersh BJ, Somers VK. Obstructive sleep apnea: implications for cardiac and vascular disease. JAMA. 2003;290(14):1906–1914.

7. Punjabi NM, Caffo BS, Goodwin JL, et al. Sleep-disordered breathing and mortality: a prospective cohort study. PLoS Med. 2009;6(8):e1000132.

8. Young T, Finn L, Peppard PE, et al. Sleep disordered breathing and mortality: eighteen-year follow-up of the Wisconsin sleep cohort. Sleep. 2008;31(8):1071–1078.

9. Phillips CL, Grunstein RR, Darendeliler MA, et al. Health outcomes of continuous positive airway pressure versus oral appliance treatment for obstructive sleep apnea: a randomized controlled trial. Am J Respir Crit Care Med. 2013;187(8):879–887.

10. Weaver TE, Sawyer A. Management of obstructive sleep apnea by continuous positive airway pressure. Oral Maxillofac Surg Clin North Am. 2009;21(4):403–412.

11. McEvoy RD, Antic NA, Heeley E, et al. CPAP for prevention of cardiovascular events in obstructive sleep apnea. N Engl J Med. 2016;375(10):919–931.

12. Barbe F, Sa M, Carmona C, Marı JM. Effect of continuous positive airway pressure on the incidence of hypertension and cardiovascular events in nonsleepy patients with obstructive sleep apnea. JAMA. 2012;307(20):2161–2168.

13. Parra O, Sánchez-Armengol A, Bonnin M, et al. Early treatment of obstructive apnoea and stroke outcome: a randomised controlled trial. Eur Respir J. 2011;37(5):1128–1136.

14. Peker Y, Glantz H, Eulenburg C, Wegscheider K, Herlitz J, Thunström E. Effect of positive airway pressure on cardiovascular outcomes in coronary artery disease patients with non-sleepy obstructive sleep apnea: the RICCADSA randomized controlled trial. Am J Respir Crit Care Med. 2016;194(5):613–620.

15. Ferguson KA. The role of oral appliance therapy in the treatment of obstructive sleep apnea. Clin Chest Med. 2003;24(2):355–364.

16. Isono S, Tanaka A, Sho Y, Konno A, Nishino T. Advancement of the mandible improves velopharyngeal airway patency. J Appl Physiol. 1995;79(6):2132–2138.

17. Kato J, Isono S, Tanaka A, et al. Dose-dependent effects of mandibular advancement on pharyngeal mechanics and nocturnal oxygenation in patients with sleep-disordered breathing. Chest. 2000;117(4):1065–1072.

18. Aarab G, Lobbezoo F, Hamburger HL, Naeije M. Effects of an oral appliance with different mandibular protrusion positions at a constant vertical dimension on obstructive sleep apnea. Clin Oral Investig. 2010;14(3):339–345.

19. Andrén A, Hedberg P, Walker-Engström M-L, Wahlén P, Tegelberg Å. Effects of treatment with oral appliance on 24-h blood pressure in patients with obstructive sleep apnea and hypertension: a randomized clinical trial. Sleep Breath. 2012;17(2):705–712.

20. Bloch KE, Iseli A, Zhang JN, et al. A randomized, controlled crossover trial of two oral appliances for sleep apnea treatment. Am J Respir Crit Care Med. 2000;162(1):246–251.



21. Ferguson KA, Ono T, Lowe AA, Al-Majed S, Love LL, Fleetham JA. A short-term controlled trial of an adjustable oral appliance for the treatment of mild to moderate obstructive sleep apnoea. Thorax. 1997;52(4):362–368.

22. Fukuda T, Tsuiki S, Kobayashi M, Nakayama H, Inoue Y. Selection of response criteria affects the success rate of oral appliance treatment for obstructive sleep apnea. Sleep Med. 2014;15(3):367–370.

23. Gagnadoux F, Fleury B, Vielle B, et al. Titrated mandibular advancement versus positive airway pressure for sleep apnoea. Eur Respir J. 2009;34(4):914–920.

24. Petri N, Svanholt P, Solow B, Wildschiødtz G, Winkel P. Mandibular advancement appliance for obstructive sleep apnoea: results of a randomised placebo controlled trial using parallel group design. J Sleep Res. 2008;17(2):221–229.

25. Okuno K, Pliska BT, Hamoda M, Lowe AA, Almeida FR. Prediction of oral appliance treatment outcomes in obstructive sleep apnea: a systematic review. Sleep Med Rev. 2016;30:25–33.

26. Sutherland K, Takaya H, Qian J, Petocz P, Ng ATM. Oral appliance treatment response and polysomnographic phenotypes of OSA. J Clin Sleep Med. 2015;11(8):861–868.

27. Bosshard V, Masse J-F, Sériès F. Prediction of oral appliance efficiency in patients with apnoea using phrenic nerve stimulation while awake. Thorax. 2011;66(3):220–225.

28. Cartwright RD. Predicting response to the tongue retaining device for sleep apnea syndrome. Arch Otolaryngol. 1985;111(6):385–388.

29. Chan ASL, Lee RWW, Srinivasan VK, Darendeliler MA, Grunstein RR, Cistulli PA. Nasopharyngoscopic evaluation of oral appliance therapy for obstructive sleep apnoea. Eur Respir J. 2010;35(4):836–842.

30. Chung JW, Enciso R, Levendowski DJ, Morgan TD, Westbrook PR, Clark GT. Treatment outcomes of mandibular advancement devices in positional and nonpositional OSA patients. Oral Surg Oral Med Oral Pathol Oral Radiol Endod. 2010;109(5):724–731.

31. Hoekema A, Doff MHJ, de Bont LGM, et al. Predictors of obstructive sleep apnea-hypopnea treatment. J Dent Res. 2007;86(12):1181–1186.

32. Lee C-H, Jung HJ, Lee WH, et al. The effect of positional dependency on outcomes of treatment with a mandibular advancement device. Arch Otolaryngol Head Neck Surg. 2012;138(5):479–483.

33. Marklund ME, Stenlund H, Franklin KA. Mandibular advancement devices in 630 men and women with obstructive sleep apnea and snoring. Chest. 2004;125(4):1270–1278.

34. Mehta A, Qian J, Petocz P, Darendeliler MA, Cistulli PA. A randomized, controlled study of a mandibular advancement splint for obstructive sleep apnea. Am J Respir Crit Care Med. 2001;163(6):1457–1461.

35. Ng ATM, Darendeliler MA, Petocz P, Cistulli PA. Cephalometry and prediction of oral appliance treatment outcome. Sleep Breath. 2012;16(1):47–58.

36. Shen H-L, Wen Y-W, Chen N-H, Liao Y-F. Craniofacial morphologic predictors of oral appliance outcomes in patients with obstructive sleep apnea. J Am Dent Assoc. 2012;143(11):1209–1217.

37. Zeng B, Ng ATM, Darendeliler MA, Petocz P, Cistulli PA. Use of flow–volume curves to predict oral appliance treatment outcome in obstructive sleep apnea. Am J Respir Crit Care Med. 2007;175(7):726–730.

38. De Backer JW, Vanderveken OM, Vos WG, et al. Functional imaging using computational fluid dynamics to predict treatment success of mandibular advancement devices in sleep-disordered breathing. J Biomech. 2007;40(16):3708–3714.

39. Ng ATM, Qian J, Cistulli PA. Oropharyngeal collapse predicts treatment response with oral appliance therapy in obstructive sleep apnea. Sleep. 2006;29(5):666–671.

40. Vanderveken OM, Vroegop AV, van de Heyning PH, Braem MJ. Drug-induced sleep endoscopy completed with a simulation bite approach for the prediction of the outcome of treatment of obstructive sleep apnea with mandibular repositioning appliances. Oper Tech Otolaryngol Head Neck Surg. 2011;22(2):175–182.

41. Vroegop AV, Vanderveken OM, Dieltjens M, et al. Sleep endoscopy with simulation bite for prediction of oral appliance treatment outcome. J Sleep Res. 2013;22(3):348–355.

42. Tsai WH, Vazquez JC, Oshima T, et al. Remotely controlled mandibular positioner predicts efficacy of oral appliances in sleep apnea. Am J Respir Crit Care Med. 2004;170(4):366–370.

43. Remmers JE, Charkhandeh S, Grosse J, et al. Remotely controlled mandibular protrusion during sleep predicts therapeutic success with oral appliances in patients with obstructive sleep apnea. Sleep. 2013;36(10):1517–1525.

44. Dort LC, Hajduk E, Remmers JE. Mandibular advancement and obstructive sleep apnoea: a method for determining effective mandibular protrusion. Eur Respir J. 2006;27(5):1003–1009.

45. Sutherland K, Ngiam J, Cistulli PA. Performance of remotely controlled mandibular protrusion sleep studies for prediction of oral appliance treatment response. J Clin Sleep Med. 2017;13(3):411–417.

46. Kastoer C, Dieltjens M, Oorts E, et al. The use of remotely controlled mandibular positioner as a predictive screening tool for mandibular advancement device therapy in patients with obstructive sleep apnea through single-night progressive titration of the mandible: a systematic review. J Clin Sleep Med. 2016;12(10):1411–1421.

47. Vázquez JC, Tsai WH, Flemons WW, et al. Automated analysis of digital oximetry in the diagnosis of obstructive sleep apnea. Thorax. 2000;55(4):302–307.

48. Breiman L. Random forests. Mach Learn. 2001;45(1):5–32.49. Flemons WW, Whitelaw WA, Brant R, Remmers JE. Likelihood ratios

for a sleep apnea clinical prediction rule. Am J Respir Crit Care Med. 1994;150(5 Pt 1):1279–1285.

50. Johns MW. A new method for measuring daytime sleepiness: the Epworth sleepiness scale. Sleep. 1991;14(6):540–545.

51. Flemons WW, Reimer MA. Development of a disease-specific health-related quality of life questionnaire for sleep apnea. Am J Respir Crit Care Med. 1998;158(2):494–503.

52. Buysse DJ, Reynolds CF, Monk TH, et al. The Pittsburgh Sleep Quality Index: a new instrument for psychiatric practice and research. Psychiatry Res. 1989;28(2):193–213.

53. Engleman HM, McDonald JP, Graham D, et al. Randomized crossover trial of two treatments for sleep apnea/hypopnea syndrome: continuous positive airway pressure and mandibular repositioning splint. Am J Respir Crit Care Med. 2002;166(6):855–859.

54. Vecchierini MF, Attali V, Collet JM, et al. A custom-made mandibular repositioning device for obstructive sleep apnoea-hypopnoea syndrome: the ORCADES study. Sleep Med. 2016;19:131–140.

55. Chai-Coetzer CL, Antic NA, Rowland LS, et al. A simplified model of screening questionnaire and home monitoring for obstructive sleep apnoea in primary care. Thorax. 2011;66(3):213–219.

56. Walker-Engström M-L, Ringqvist I, Vestling O, Wilhelmsson B, Tegelberg A. A prospective randomized study comparing two different degrees of mandibular advancement with a dental appliance in treatment of severe obstructive sleep apnea. Sleep Breath. 2003;7(3):119–130.

57. Lee CH, Kim J-W, Lee HJ, et al. Determinants of treatment outcome after use of the mandibular advancement device in patients with obstructive sleep apnea. Arch Otolaryngol Head Neck Surg. 2010;136(7):677–681.

58. Mayer G, Meier-Ewert K. Cephalometric predictors for orthopaedic mandibular advancement in obstructive sleep apnoea. Eur J Orthod. 1995;17(1):35–43.

59. Eveloff SE, Rosenberg CL, Carlisle CC, Millman RP. Efficacy of a Herbst mandibular advancement device in obstructive sleep apnea. Am J Respir Crit Care Med. 1994;149(4 Pt 1):905–909.

60. Denolf PL, Vanderveken OM, Marklund ME, Braem MJ. The status of cephalometry in the prediction of non-CPAP treatment outcome in obstructive sleep apnea patients. Sleep Med Rev. 2016;27:56–73.

61. Edwards BA, Andara C, Landry S, et al. Upper-airway collapsibility and loop gain predict the response to oral appliance therapy in obstructive sleep apnea patients. Am J Respir Crit Care Med. 2016;194(11):1413–1422.

62. Van Leeuwen M, Martinez-Ferrate R, Preble D, Hanewinkel W, Gleeson M, Andra J. A two-dimensional gauge and protocol for fitting oral appliances used in treating sleep breathing disorders. Compend Contin Educ Dent. 2015;36(2):140–145.

63. Levendowski DJ, Westbrook P, Morgan TD. Initial evaluation of a titration appliance for temporary treatment of obstructive sleep apnea. J Sleep Disord Ther. 2012;1(1).



64. Morgan TD. Novel approaches to the management of sleep-disordered breathing: reviewing the past and looking into the future. Sleep Med Clin. 2016;11(2):xiii-xiv.

ACKNOWLEDGMENTSThe authors thank the participants, SomnoMed Ltd. for the complementary provision of SomnoDent appliances, ProSomnus Sleep Technologies for the complementary provision of MicrO2 appliances, and Zephyr Sleep Technologies for the provision of the FCMP device. Funding for the trial was provided by Alberta Innovates Technology Futures, Zephyr Sleep Technologies and the NRC-IRAP Industrial Assistance Fund.

SUBMISSION & CORRESPONDENCE INFORMATIONSubmitted for publication January 6, 2017Submitted in final revised form March 28, 2017Accepted for publication May 3, 2017

Address correspondence to: Dr. Sabina Bruehlmann, Director, Technology, Zephyr Sleep Technologies Inc., 610A 70th Ave SE, Calgary, AB, Canada; Tel: (587) 332-0285; Fax: (587) 332-0208; Email: [email protected]

DISCLOSURE STATEMENTWork for this study was completed at Zephyr Sleep Technologies and The Snore Centre, located in Calgary, Alberta, Canada. The study involved an investigational medical device. All authors have reviewed and approved the manuscript. J.E. Remmers, Z. Topor, J. Grosse, E.V. Mosca, S. Bruehlmann, S. Charkhandeh, and S.A. Zareian Jahromi are employees of Zephyr Sleep Technologies. R. Brant previously received warrants for work done for Zephyr Sleep Technologies. N. Vranjes has no conflicts of interest to report. Financial support was provided by Alberta Innovates Technology Futures, Zephyr Sleep Technologies, and the NRC-IRAP Industrial Assistance Fund. Equipment was provided by SomnoMed Ltd., ProSomnus Sleep Technologies, and Zephyr Sleep Technologies.

Documents

SCIENTIFIC INVESTIGATIONS A Feedback-Controlled Mandibular ... · 871 Journal of Clinical Sleep Medicine, Vol. 13, No. 7, 2017 Study Objectives: Mandibular protruding oral appliances