Safety and Efficacy of Drug-Induced Sleep Endoscopy Using a Probability Ramp Propofol Infusion System in Patients with Severe Obstructive Sleep Apnea

Abstract

BACKGROUND: 

Drug-induced sleep endoscopy (DISE) uses sedative-hypnotics to induce moderate obstruction in sleep apnea patients, thereby facilitating anatomic assessment of obstructive physiology. Implementation of DISE with propofol requires a dosing strategy that reliably and efficiently produces obstruction while minimizing oxygen desaturation.

METHODS: 

The surgeon in a prospective study of transoral robotic resection of the tongue base enrolled 97 patients with obstructive sleep apnea confirmed by polysomnography who failed continuous positive airway pressure. All patients were screened by DISE. Propofol dose was determined using custom software written in MATLAB, which has been previously described. Studies were performed in an operating room with standard monitors and resuscitation equipment. No topical anesthesia was used, and no IV drugs other than propofol were used. All patients received 2 L/min supplemental oxygen via a nasal cannula placed in the mouth. After initiation of propofol sedation, a pediatric bronchoscope was positioned via the naris to observe the velopharynx. The sedation sequence was continued until the clinical end point of obstruction onset was noted. Observation of the pharynx was performed for a sufficient period to obtain images of the anatomic site(s) of obstruction. The infusion was then terminated. Statistical analysis was performed with MATLAB (MathWorks, version 2012b). Comparison of saturation nadirs between DISE and subject sleep studies was performed with both the paired and unpaired Student t test.

RESULTS: 

The subject population was characterized by a median body mass index of 32.1 (interquartile range [IQR] 6.8) kg/m² and apnea-hypopnea index of 48 (IQR 32). All patients demonstrated obstruction within the design variables. Obstruction was observed after 236 (±57.9) seconds at an estimated effect-site concentration of 4.2 ± 1.3 mcg/mL. The median saturation nadir during DISE was significantly higher (91.4% (IQR 5.1)) than that during standard sleep studies (81.0% [IQR 11.2], P < 0.0001). Ninety-five percent confidence intervals for correlations between DISE saturation nadir and body mass index, age, apnea-hypopnea index, or administered propofol dose included zero in all cases.

CONCLUSIONS: 

A propofol infusion strategy that requires limited experience with propofol dose selection and only 1 pump dosing change reliably produced airway obstruction in patients with severe sleep apnea. Clinical obstruction was achieved faster than target-controlled infusion-based systems for similar procedures reported in the literature. The observed degree of oxygen desaturation in the model system was within a clinically acceptable range.

In Brief

Published ahead of print July 25, 2014.

Drug-induced sleep endoscopy (DISE) uses sedative-hypnotics to induce moderate obstruction to facilitate anatomic differentiation of obstructive physiology. The epidemic of obstructive sleep apnea (OSA) coupled with a substantial continuous positive airway pressure failure rate and new surgical approaches have promoted the use of DISE as a diagnostic tool. Palatal collapse at the velopharynx, lateral collapse of the pharyngeal walls, and obstruction at the tongue base are common sites of dynamic obstruction.¹ In contrast to other diagnostic approaches, real-time visual observation of the airway during DISE can differentiate the relative role of each mechanism in a given patient.² DISE may also provide prognostic information regarding therapeutic interventions such as mandibular advancement devices.³

Obstruction commonly, but not always, occurs contemporaneously with loss of consciousness. Anesthetic induction with propofol to loss of consciousness is readily accomplished; however, reliable prediction of the necessary propofol dose to achieve and maintain obstruction without causing prolonged loss of consciousness or oxygen desaturation is challenging. Manual bolus approaches require an experienced practitioner to obtain consistent results. Overshoots can invoke clinically concerning oxygen desaturation and aborted procedures while underdosing may add substantial time to the procedure and discomfort to the patient. Target-controlled infusion (TCI) may reduce the variability of propofol titration,⁴ but TCI is not available in North America. Moreover, titration by TCI requires small serial increments in the target to achieve the desired clinical end point, a time-consuming process that is not well suited to high-volume throughput or patient comfort. Manual and TCI approaches assume that the anesthesiologist can surmise the appropriate target level for each patient. We have previously demonstrated in simulation a method for producing a continuously increasing probability of loss of consciousness with a simple sequence of infusion rates.⁵ We hypothesized that such an approach, with a hybrid pharmacokinetic/pharmacodynamic model, would permit reliable and efficient titration of propofol to an end point of visible airway collapse and/or loss of genioglossus tone in a population with severe OSA. We hypothesized that the risk of oxygen desaturation during DISE would be no worse than that observed during sleep studies.

METHODS

The IRB of the University of Pennsylvania, Perelman School of Medicine, approved this study, and written informed patient consent was provided. Ninety-seven patients were enrolled in a prospective study of transoral robotic resection of the tongue base (ClinicalTrials.gov #NCT01187160); this study was a secondary outcome measure of that study. All patients had previously undergone an overnight polysomnography study in a sleep laboratory and had attempted CPAP as a therapeutic approach but found this not to be helpful. The apnea-hypopnea index (AHI) for each patient was recorded from the sleep study. For each patient, the saturation nadirs from the sleep study and DISE were paired for statistical analysis. All patients were screened with DISE with propofol infusions determined by custom software written in MATLAB, which has been previously described.⁵ The system uses the Cortínez et al.⁶ pharmacokinetic model and the Johnson pharmacodynamic model⁷ to determine an infusion sequence comprised a bolus, initial infusion, secondary infusion, and a time for transition from the initial to secondary infusion using the age and weight of the individual patient. This sequence minimizes the difference between the predicted probability of loss of responsiveness and a linear ramp. We have termed this probability ramp control (PRC). A brief description of the mathematical approach can be found in the Appendix. The MATLAB program (heretofore referred to as “the system”) performed the calculations, indicated the time to perform the transition to the second infusion, and logged the time at which obstruction was noted. Patient characteristics are presented in Table 1.

Table 1: Demographics of Analyzed Patients

All DISE studies were performed in an operating room with standard monitors and resuscitation equipment. A single otorhinolaryngologist (Erica R. Thaler, MD) performed each nasopharyngoscopy. Propofol was administered by an anesthesiologist (JEM or JHA). No topical anesthesia was used, and no IV drugs other than propofol were used. All DISE patients received 2 L/min supplemental oxygen via a nasal cannula placed in the mouth. Supplemental oxygen was not used during polysomnography studies (as is typical for this procedure). Pulse oximetry data were recorded for subsequent analysis from a Nellcor pulse oximeter (Covidien, Mansfield, MA) at 30-second intervals by the DocuSys anesthesia record keeping system. Sedation was provided with propofol infused through a pigtail side-port adapter with a free-flowing IV catheter. A 60-mL Becton-Dickinson syringe was loaded with 40 mL propofol, and the line primed until propofol was visibly present at the hub. A Baxter AS50 pump (Baxter, Deerfield, IL) was programmed with the initial bolus and infusion rates as determined by the control system. After this initial bolus of propofol, an Olympus model BF-3C160 pediatric bronchoscope (Olympus Corporation, Center Valley, PA) was passed via the naris. With the bronchoscope in position to observe the velopharynx, the sedation sequence proceeded until the onset of obstruction was noted. This was identified as the obstruction clinical end point. Observation of the pharynx was performed for a sufficient period to obtain images of the anatomic site(s) of obstruction. The infusion was then terminated, and the patient was allowed to recover. Patient characteristics and derived pharmacokinetic measures were assessed for normal distribution using the Lilliefors test at 5% significance level using the Statistics Toolbox of MATLAB 8.0 (MathWorks, Natick, MA). Saturation nadirs were assessed with the Lilliefors test, which rejected the hypothesis that they were normally distributed. Comparison of saturation nadirs from DISE and polysomnography was performed with both the paired and unpaired Student t test. Correlation of DISE saturation nadir and body mass index (BMI), AHI, propofol effect-site concentration at obstruction, and age were assessed by Spearman ρ, with Fisher transformation applied to determine 95% confidence intervals (CIs). No power analysis was performed for this study, because it was a secondary outcome measure for the larger study of transoral robotic surgery.

RESULTS

The subject population was characterized by a median BMI of 32.1 (± 6.8) kg/m² and median AHI of 48 (±32). The median time to obstruction, as determined by the otorhinolaryngologist, was 3.8 (interquartile range [IQR] 1.2) minutes. The mean, predicted effect-site concentration of propofol at obstruction was 4.2 ± 1.3 mcg/mL. The median saturation nadir during DISE was significantly higher (91.4 ± 5.1% IQR) than that during standard sleep studies (81.0% ± 11.2% IQR, both paired and unpaired t test P < 0.0001). Figure 1 depicts the histogram plot of the distribution of saturation nadirs during polysomnography and DISE.

Figure 1: Histogram of nadir oxygen saturation during polysomnography and drug-induced sleep endoscopy for analyzed patients (n = 97). Bin widths are 4.2% for polysomnography and 2.6% for DISE.

Saturation nadirs were lower during polysomnography than DISE in all but 7 patients; in that subcohort, median nadirs were 89.3% and 85%, respectively. Correlation analysis was performed with Spearman ρ for saturation nadir during DISE and BMI, AHI, age, and propofol effect-site concentration at obstruction. The 95% CIs included zero in all cases, but the correlation between predicted propofol effect-site concentration and saturation could be as low as −0.43.

All patients completed DISE, and there were no adverse events associated with the study.

DISCUSSION

DISE with midazolam was introduced >20 years ago in the United Kingdom.⁸ There are no standardized criteria for DISE interpretation, but test-retest reliability⁹ and interrater reliability¹⁰ have been reported. Consistency of interpretation is high among experienced otolaryngologists.

A variety of anesthetic approaches have been described, with most relying on modifications of the originally reported midazolam regimen or propofol by TCI or manual bolus. In children, dexmedetomidine after sevoflurane induction has been reported, but no comparison studies on differential results between anesthetic techniques have been published.¹¹

DISE is a niche procedure with a growing number of clinical reports in the literature. In the typical scenario, a surgeon decides to implement the practice and requests anesthesiology to perform the anesthesia within a narrow range of anesthetic depth. The average practicing anesthesiologist will have limited experience with the gradual titration of propofol to the specific end point of airway obstruction without unintentional overdose necessitating airway intervention. Our study population includes patients with high AHI who failed continuous positive airway pressure therapy. According to guidelines issued by the American Academy of Sleep Medicine, patients with an AHI greater than 30 (i.e., >30 >10-second episodes of apnea or hypoventilation per hour) have severe OSA. Previously described DISE approaches with propofol are not without risk. Reports of desaturation requiring bag-mask ventilation are not uncommon with manual propofol titration even in patients with only moderate disease. Indeed, in a recent study by De Vito et al.¹² comparing TCI with manual propofol bolus, 65% of patients in the manual group demonstrated “sedation instability” compared with 5% in the TCI group. In contrast to the present study, those investigators excluded higher risk patients with BMI >30 kg/m², presumably over concern for oxygen desaturation, and the mean AHI of study subjects was 21 ± 7. Most reports using propofol for DISE have focused on surgical assessment or diagnostic utility and have not reported oxygen saturation profiles to the extent characterized here. The effective sedation seen in this study with a low rate of desaturation and infrequent need for airway support is an important result. The lowest saturation measured in our study (patients received oxygen at 2 L/min via oral cannula) was 74%. The mean oxygen saturation value was significantly higher than the mean during polysomnography and was comparable with that reported by De Vito during TCI infusion. Supplemental oxygen is rarely used during polysomnography studies. We acknowledge that the use of supplemental oxygen decreases the sensitivity of pulse oximetry to detect hypoventilation, but neither oxygen saturation nor hypoventilation was used as detection criteria during DISE. Few anesthesiologists would withhold supplemental oxygen from obese patients with severe sleep apnea during deep sedation with propofol without a compelling reason to do so. Conversely, few sleep centers would insist on supplemental oxygen during a sleep study as an essential safety measure. The comparison is made to demonstrate that the risk of desaturation during DISE is lower than that during a sleep study under normal practices. The extent of desaturation in obese patients with OSA is multifactorial. In addition to obstruction or apnea, loss of functional residual capacity during sleep and the balance of metabolic rate and oxygen delivery under the effects of propofol also play a role.

Controlled infusion of propofol based on several competing pharmacokinetic models has been proposed to reduce the variability of anesthetic technique and rate of desaturation.¹³ TCI approaches have generally evidenced good reliability but require slow titration, and the optimal pharmacokinetic model in the severe OSA population has not been rigorously demonstrated. Predicted effect-site concentrations between 2.0 and 4.8 μg/mL to achieve obstruction have been reported with the Schnider or Marsh propofol models.^12,14 The predicted effect-site concentration for obstruction we observed (4.2 ± 1.3 μg/mL) when using the Cortínez model (and adjoining an effect compartment with a k_e0 determined to yield the time to peak propofol effect of 1.6 minutes) is in a range of similar magnitude. There may well be a significant difference between the “true” effect-site concentrations and those modeled with our system. Since the predicted effect-site concentration is also likely to be rapidly changing at the time of obstruction, the precision of the predicted concentration may also be reduced. It is important to note that the primary intent of the study was not to determine the “true” effect-site concentrations or even a target that could be generalized to other systems. Rather, we aimed to evaluate the fidelity of our hybrid Cortínez-Johnson pharmacokinetic/pharmacodynamic model to efficiently and reproducibly obtain conditions for DISE assessment while avoiding significant oxygen desaturation. Since unlike TCI, the practitioner need not make a preliminary guess as to the appropriate target, an effective strategy is more important than an accurately predicted effect-site concentration.

TCI may yield results that are superior to bolus techniques and may be easier to standardize.¹⁵ An important premise of the overall project is to create a system that reduces the reliance on provider experience to estimate manual titration requirements in the context of the lack of TCI approved devices in North America. We addressed this issue by devising a system that required few interactions with the pump, making it feasible for the anesthesiologist to observe the patient while titrating propofol to the clinical end point. The modeling system does not require a physical connection to the pump, since the user does the programming. The system is thus compatible with any pump capable of delivering a propofol bolus and infusion. For clinical efficiency and patient comfort, the time to complete the procedure is also an important variable.

In 1 study by De Vito et al.¹² the average time for the procedures was 15.2 minutes in the TCI group and 6.2 minutes in the manual control group. The mean time to obstruction employing our control system was 3.9 minutes, which is comparable with the manual approach and substantially faster than with TCI control in published studies.

Our system is able to achieve the end point of obstruction in a time similar to that reported by De Vito for bolus propofol, without the associated incidence of desaturation requiring intervention that was noted.

It is critical to point out that the system described remains investigational and is not approved by any regulatory agency for clinical use. The authors encourage readers interested in utilizing the system in an IRB-approved protocol to contact us. A limitation of our study is a lack of comparison to a control group receiving the standard of care for DISE in our institution for the simple reason that we had no experience with DISE before initiating this study. Rather than embarking on a learning curve for manual DISE, we chose to use this opportunity to further develop probability ramp control in a challenging patient population. While it is possible that an experienced clinician might obtain results superior to those we report, the automated approach holds more promise for improving care and minimizing variance between providers. The study was not designed to assess the pharmacokinetic rigor of the hybrid Johnson-Cortínez model that was implemented. As such, a further limitation of the study is the inability to report on the error between actual and predicted effect-site and/or plasma concentrations of propofol. Thus, the effect-site concentrations necessary to achieve obstruction reported here may not translate to similar values when other systems or models are used. In future studies, measurement of plasma concentrations would facilitate model refinement.

CONCLUSIONS

A propofol infusion strategy that requires limited experience with propofol dose selection and only 1 pump dosing change reliably produced airway obstruction in patients with severe sleep apnea undergoing DISE. The reported system produced clinical obstruction faster than TCI-based systems for similar procedures reported in the literature with a clinically acceptable rate of desaturation when compared with polysomnography in the same individuals. The ability to generalize the described strategy to the inexperienced clinician in a high-volume setting will need further study.

APPENDIX: CONTROL SYSTEM IMPLEMENTATION

The purpose of the control system is to determine an infusion sequence that will cause a linear increase in the probability of sedation over a defined interval that will be similar for all weights and ages of patients, as illustrated in Appendix Figure 1. The sequence comprised an initial bolus, an initial infusion rate, a secondary infusion rate, and a time at which this transition occurs. This permits the operator to set up the pump with the first 2 values before initiation of sedation and only make a single change in infusion rate at a specified time to complete the procedure.

Appendix: Figure 1. Sedation of a 48-year-old patient weighing 100 kg. The propofol infusion rate is depicted in green: a bolus of 295 μg/kg (29.5 mg over 29 seconds) followed by an initial infusion of 189 μg/kg/min for 116 seconds (delivering 36.5 mg), followed by an increase to 486 μg/kg/min. The probability of unresponsiveness for the patient is depicted in red. The dotted line is the target (the desired trajectory of response probability).

To accomplish this, 3 components are used:

• A pharmacokinetic model of propofol
• A pharmacodynamics model of propofol
• A minimization of trajectory error

PROPOFOL PHARMACOKINETIC MODEL

The purpose of the pharmacokinetic model is to translate drug administration into effect-site concentration. While several models of propofol have been described, the model of Cortínez et al.⁶ is used, as it does not suffer from the limitations of the calculation of lean body mass. Model parameters were taken from Table 2 of that publication. A biophase compartment is adjoined to this model, calculating k_e0 so that the time to peak effect is 1.6 minutes. The MATLAB code for the state space model with observation of the effect-site concentration can be found on the OpenTCI website.^a Given an infusion sequence I (comprised a bolus (B), an initial infusion (I₁), a second infusion (I₂), and a time for transition from I₁ to I₂ (T₁), the effect-site concentration for propofol is given by equation A1:

where C_e(t) is the effect-site concentration at time t and PK is the pharmacokinetic model with the adjoined effect-site compartment.

PROPOFOL PHARMACODYNAMICS MODEL

The purpose of the pharmacodynamic model is to determine the probability of a clinical event given an estimated effect-site concentration. The model of Johnson et al.⁷ is employed. This model considers both propofol and remifentanil and provides response probability predictions for 4 different levels of stimulation; the parameters for loss of responsiveness were employed with remifentanil set to zero, as shown in equation A2:

where C₅₀ is the effect-site concentration associated with a 50% probability of loss of responsiveness and n is the steepness of the curve.

MINIMIZATION OF TRAJECTORY ERROR

The trajectory is a line of increasing probability, indicated as “Target” in Appendix Figure 1. The line starts at a 10% probability at 80 seconds and ends at 90% probability at 200 seconds. The trajectory error is the difference between the predicted response probability for a given infusion sequence and the trajectory. The infusion sequence (B, I₁, I₂, and T₁) that results in the minimum value for trajectory error is identified by a simplex minimization using the MATLAB Optimization Toolbox.

Any open loop method will be affected by modeling errors. Consider the result of varying all the parameters of the nominal Cortínez model⁶ by half of the 95% CIs, yielding 2 additional models termed Upper CI and Lower CI. The parameters of these models are listed in Appendix Table 1. Assume that the end point of obstruction will be achieved at an effect-site concentration of 2.5 μg/mL. The application of the infusion sequence determined for the nominal parameters to the 3 models is depicted in Appendix Figure 2. The lower CI model responds more quickly (having lower volumes and clearances), and obstruction will be observed after 1.8 minutes, which would correspond to an effect-site concentration of 1.8 μg/mL under the assumption that the patient followed the nominal model. Similarly, the upper CI model will not reach obstruction until 3.4 minutes, corresponding to an effect-site concentration of 3.4 μg/mL. Compare this result with application of a TCI designed for the nominal model. Starting with an effect-site target of 1.6 μg/mL, we increase the target by 0.2 μg/mL every 3 minutes, as depicted in Appendix Figure 3. The lower CI model will achieve obstruction after the second step, corresponding to an effect-site concentration of 1.8 μg/mL, while the upper CI model will achieve obstruction after the ninth step, corresponding to an effect-site concentration of 3.2 μg/mL. Although the impact of parameter error on the error in the identified effect-site concentration is minimal, the time required to arrive at obstruction is considerably longer: 6.2 to 27.25 minutes with the TCI approach versus 1.8 to 3.4 minutes with PRC. In addition, the TCI approach requires multiple adjustments of the target concentration, while PRC requires only a single adjustment. Thus, PRC reduces the effort of both the endoscopist and the anesthesiologist without a significant reduction in precision in the determination of the effect-site concentration associated with obstruction.

Appendix Table 1: Model Parameters for the 3 Sim ulated Patients Used in Appendix Figures 2 and 3

Appendix: Figure 2. Application of the infusion sequence designed for a patient with nominal parameters to the nominal (red), upper CI (green), and lower CI (blue) patients. Obstruction occurs at 2.5 μg/mL, but the onset is at 1.8 minutes in lower CI and 3.4 minutes in upper CI, leading to misidentification of the effect-site concentration associated with obstruction.

Appendix: Figure 3. Application of a target-controlled infusion designed for a patient with nominal parameters to the nominal (red), upper CI (green), and lower CI (blue) patients. Obstruction occurs at 2.5 μg/mL, but the onset follows the second step in lower CI and the ninth step in upper CI, leading to misidentification of the effect-site concentration associated with obstruction.

DISCLOSURES

Name: Joshua H. Atkins, MD, PhD.

Contribution: This author helped design and conduct the study, analyze the data, and write the manuscript.

Attestation: Joshua H. Atkins has seen the original study data, reviewed the analysis of the data, approved the final manuscript, and is the author responsible for archiving the study files.

Name: Jeff E. Mandel, MD, MS.

Contribution: This author helped design and conduct the study, analyze the data, and write the manuscript.

Attestation: Jeff E. Mandel has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.

Name: Giulia Rosanova, BA.

Contribution: This author helped analyze the data.

Attestation: Giulia Rosanova has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.

This manuscript was handled by: Peter S. A. Glass, MB, ChB.

ACKNOWLEDGMENTS

We acknowledge the excellent research support provided by Mary Hammond, MSN, from the Department of Anesthesiology and Critical Care. We acknowledge the contributions of Elie Sarraf, MD, to system design and Erica R. Thaler, MD, who performed the DISE procedures. We thank Rebecca Speck, PhD, for statistical review and helpful suggestions.