| JOURNAL HOME | CME HOME | THIS MONTH | PAST ISSUES | ETOC | COLLECTIONS |
| AUTHORS | REVIEWERS | EDITORIAL BOARD | FEEDBACK | RSS | HELP |
| ||||||||||||||
| ||||||||||||||
|
|
|
|
||||||||||||
|
||||||||||||||
From the Departments of *Anesthesiology and Bioengineering, University of Utah, Salt Lake City, Utah.
Address correspondence and reprint requests to Dwayne R. Westenskow, PhD, Department of Anesthesiology, University of Utah, 3C 444 SOM, Salt Lake City, UT 84132. Address e-mail to Dwayne.Westenskow{at}hsc.utah.edu.
 |
Abstract |
|---|
 Top Abstract IntroductionMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
50% and model’s resistance element (R2) changed 
10-fold from baseline. Testing this rule on 40 cases of SCW, four false positives were obtained. During SCW, R1 and R2 increased, whereas C2 decreased significantly from baseline. This preliminary study is a promising step toward noninvasive, real-time detection of EBI to aid clinicians in decision making. |
Introduction |
|---|
|
TopAbstractIntroductionMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
Inadvertent EBI occurs frequently [9%–10% of all intubations (8)] in operating rooms (ORs) and intensive care units (ICUs) and may go unrecognized at onset. In a study of 100 emergency patients, inadvertent EBI, determined using anteroposterior chest radiographs, went unrecognized by the physicians and anesthesiologists in 7% of the cases (2). According to the Australian Incident Monitoring Study, of the 182 incidents related to tracheal tubes, 42% were EBI (4). In another study (8), there were 154 incidents of EBI in 3947 anesthesia-related critical events reported to the Australian Incident Monitoring Study. Though easy to correct once detected, it takes longer (105 s) to detect EBI than to detect cardiac arrest (7 s) and breathing circuit malfunction (21 s) (9).
Current techniques for diagnosing EBI are time consuming, nonspecific, unreliable, or requiring additional equipment. The conventional chest auscultation technique is not automated, requires manual expertise, and is unreliable (10). In a recent study, only two of 16 EBI events were detected with chest auscultation performed by experienced anesthesiologists (11). The pulse oximeter reports a decrease in arterial O2 saturation during EBI, but only when Fio2 0.3 (12,13). The capnogram remains normal in about 89% of the EBI cases (8). End-tidal CO2 decreases during EBI; however, a similar decrease can be caused because of kinking of the endotracheal tube, an obstruction, or an excessive leak of air around the endotracheal tube cuff (14), thus leading to false positives. High airway pressure is a nonspecific indicator of EBI because high pressure may also be caused by several other complications, such as obstructed airway, bronchospasm, and stiff chest wall (SCW) (15,16). An error in differential diagnosis may lead to unnecessary extubations and reintubations (17).
Novel methods to detect correct placement of the endotracheal tube in the trachea are based on 1) acoustic reflectometry, 2) light transmission, 3) chest radiography, and 4) bronchoscopy. The acoustic reflectometry method requires specific calibration for each endotracheal tube, and is intended for use only in apneic patients. Background noise may affect the performance of the device (14). The method based on light transmission requires training and has false positives in obese or dark-skinned patients or in patients with neck edema, tumors, or abnormal upper airway anatomy (14). A chest radiograph (7,18) or bronchoscopy can detect EBI, but requires equipment that is too expensive to use routinely. Thus, there is a need for a method that uses information from noninvasive sensors that are routinely used in the clinical setting and that can accurately identify EBI within seconds of onset.
Recently, we used a model-based approach to automatically detect an obstructed endotracheal tube using airway pressure and flow signals (19). It was shown that during an endotracheal tube obstruction, the model’s parameters changed such that the event could be diagnosed noninvasively, automatically, and accurately. In the present study, the same five-element lumped model (19), based on the signals from airway flow and pressure sensors, was used to automatically detect EBI. The five elements of the model were iteratively identified in real-time by matching model impedance to the experimentally measured impedance.
From the case reports (3), it was observed that during EBI, when only one lung was ventilated, clinicians thought that the increased difficulty of manually ventilating the lungs indicated the loss of lung compliance. Therefore, it was hypothesized that the model’s compliance elements will decrease in value during EBI, providing specific and sensitive detection. To measure the specificity of the method, events of SCW in the absence of EBI were produced to provide potentially false positive conditions, and it was hypothesized that the compliance would decrease and resistance would increase during SCW. The current study is a preliminary evaluation of the model.
|
METHODS |
|---|
|
TopAbstractIntroductionMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
Three conditions were studied for each animal. The control condition consisted of appropriate endotracheal tube placement after general anesthesia. EBI was produced by advancing the endotracheal tube from the trachea into the right mainstem bronchus. EBI was confirmed by the absence of chest movement and the absence of breath sounds on the left side. A SCW event in the absence of EBI was simulated to provide a condition that produces flow and pressure changes that could mimic EBI. The chest wall was stiffened by tightly wrapping a 20-cm wide pressure cuff (Dura-CuffTM; Critikon – Johnson and Johnson; Tampa, FL) around the chest of each dog. The pressure in the cuff was increased in steps to 20, 40, 60, and 80 mm Hg and maintained at each step for 1 min. Weights were placed on the abdomen to prevent it from distending.
During each induced condition (control, EBI, and SCW), the airway pressure and flow signals were sampled at a frequency of 60 Hz for 70.83 s (NICO; Novametrix; Wallingford, CT). Respiratory input impedance was calculated from measured pressure and flow for each experimental animal. Random initial parameters were chosen (within limits) and a model (Fig. 1) iterated through stepwise changes of all parameters (using optimization function) until a fit (assessed by root mean square error [RMSE]) between the modeled impedance and the experimentally measured impedance was achieved. After creating an EBI or SCW, the model was again iterated in a stepwise fashion until a second fit was achieved. The baseline parameters were calculated using data collected 2 min before producing a complication. The changes from baseline in the parameters of the model were then compared across conditions and used to form a predictive rule.
|
Parameter Identification
The model parameters (R1, C1, R2, C2, and L) were iteratively adjusted until the RMSE between the model-predicted impedance and the measured impedance was minimized.
Optimization
To iteratively determine the model parameters, an optimization function "fminimax" was used (available in the optimization toolbox of the software Matlab by Mathworks) (19). The constraints for the model parameters were set as physiologically possible values over the following ranges: 0–1000 cm H2O · L–1 · s–1 for R1 and R2, 0–10 L per cm H2O for C1 and C2, and 0–10 cm H2O · L–1 · s–2 for L.
Formulation of Rule
A rule to identify EBI was formulated by analyzing the behavior of the model parameters during EBI and SCW. Repeated measures analysis of variance (ANOVA) was used to determine the model parameters that significantly (P < 0.05) changed from baseline during each complication (EBI and four levels of SCW). The conditions on the model parameters that 1) significantly changed from baseline and/or 2) behaved differently during the two complications were set to formulate the rule to detect EBI.
The entire signal processing and optimization took 10–15 s using a 677 MHz, 256 MB RAM PC to analyze 1 min of data to identify the five model parameters.
Statistics
Repeated measures ANOVA was used to determine a statistically significant change from control in each model parameter during each complication.
|
RESULTS |
|---|
|
TopAbstractIntroductionMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
|
|
|
As C2 decreased significantly in both complications, it was not included in the rule. In nine of 10 cases, it was observed that the ratio of C1 during EBI to C1 during baseline was <0.5. In 36 of 40 cases of SCW, it was observed that the ratio of C1 during SCW to C1 during baseline was > 0.5. A significant change in R2 (>10-fold) from baseline during SCW was observed. On the contrary, the change in R2 was <10-fold from baseline during EBI. Therefore, R2 was included in the rule. Thus, an event is classified as EBI if 
C1 0.5 and R2 10, where indicates ratio of the particular parameter during complication to that during baseline.
Based on this rule, the model identified EBI in nine of 10 cases (Table 1). There were four false positives for EBI of 40 cases of SCW (Table 1). Thus, both the sensitivity and specificity of the model for detection of EBI are 90%.
|
|
DISCUSSION |
|---|
|
TopAbstractIntroductionMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
The changes that occurred in the model’s parameters during EBI and SCW can be explained based on the relationship between the model’s parameters and the lung’s mechanical properties. In our previous work (19), we speculated that C1 represents compliance due to alveolar gas compression, R1 represents the resistance of airways and R2–C2 represent lung and chest wall properties. L, over the frequency range investigated, represents the mass of the column of gas in the airways oscillating against the lung compliance. During EBI, as only one lung is ventilated, compliance of the respiratory system is reduced, causing the model C2 to decrease. The tidal volume normally distributed between the two lungs during endotracheal intubation is forced into one lung during EBI. This results in compression of the alveolar gas, causing the model C1 to decrease. No significant change was observed in the modeled R2 because the EBI is not known to change the viscoelastic properties of the lung and chest wall, as would be changed in certain pulmonary diseases (e.g., emphysema). The total cross-sectional area of the airways decreases during one-lung ventilation, and hence, R1 should increase. In the present study, an increase in R1 was observed during EBI, although the difference was not statistically significant, possibly because the statistical analysis was performed using ANOVA in which the EBI and SCW groups were included for comparison. Change in R1 during EBI may be relatively less significant than change in R1 during SCW, resulting in an insignificant P value for EBI.
During a SCW event, to make the chest wall stiffer, a pressure cuff was wrapped around the chest of the animal and was inflated stepwise to higher and higher pressures (20, 40, 60, and 80 mm Hg). The pressure retarded the expansion of the lung/chest during inspiration and increased force on the lung/chest during expiration, making the chest wall stiffer and thus decreasing C2. The pressure may also have constricted the lower airways, reducing their cross-sectional areas and thereby increasing R2. The possible explanation for an increase in airway resistance (R1) during SCW is that the pressure in the cuff wrapped around the chest of the animal collapsed, or at least reduced, the cross-sectional area of the upper airways, thus increasing their resistance. Moreover, the increased peak expiratory flow rate in response to the applied pressure may increase the frictional resistance of the airways, and may also be responsible for the increase in R1. However, further investigation is necessary to validate our speculations.
The method is based on the change in the model parameters from baseline, and therefore, it requires baseline data (when both lungs are ventilated). In the ICU and OR, the clinician can initially confirm the correct placement of the endotracheal tube in the trachea, and this can be used as the baseline for the model. The displacement of the tube in the bronchi at a later time in the ICU or during surgery can be detected with the changes in model parameters from the baseline. While intubating an emergency patient, or during the very first intubation in the OR/ICU, the normal range of chest wall compliance and resistance based on the weight, age, and gender of the subject (21–23) may be used as the baseline. Moreover, EBI is significantly more common during gynecological, neurological, and laparoscopic surgeries (4–7) and about 50% of the EBI cases are associated with trunk surgery and 30% with head and neck surgery (8). This indicates that identification of EBI during surgery is as important as is identifying EBI at the beginning of surgery, immediately after intubation.
The model identified four of the 40 events of SCW as EBI. All these false positives occurred in the same animal at each level of the chest wall stiffness (20, 40, 60, and 80 mm Hg cuff pressure). In a different animal, the event of EBI was unidentified by the model. There were no noticeable differences in the physical (body weight and core temperature), pulmonary (end-tidal CO2 and peak inspiratory pressure), or hemodynamic (heart rate, arterial O2 saturation, cardiac output) characteristics at baseline in these animals when compared with other animals in the study. The failure of the model to correctly identify events in these animals may be a limitation of the model or, possibly, because these animals had preexisting cardiovascular or pulmonary disease. The sensitivity and specificity of the model for detection of EBI and differentiating from SCW are based on the dataset from the current study and may change for data from a different study.
In our previous work (19), we investigated the influence of positive end-expiratory pressure on the model parameters, and no significant change in model resistance and compliance was observed. This suggests that the application of positive end-expiratory pressure in the clinical setting may not interfere with the proposed method. The present preliminary study indicates that the model may indeed have diagnostic ability in OR/ICU and emergency care. However, additional studies are required before proceeding to a bedside application of the model.
The rule formulated to identify EBI is based on the collected data from 10 paralyzed dogs under controlled experimental conditions, and may not apply in a clinical setting. The limited size of the data set should be considered before making inferences. Noise in a clinical environment (motion artifacts, spontaneous breathing, cough, etc.) may cause false detection of EBI, and therefore, the raw airway pressure and flow signals may need to be filtered to remove noise and reject artifacts. Further studies are needed to observe the model’s behavior during multiple events and alter the rule based on the results. However, the preliminary results obtained from the work are promising and justify further study.
In the future, if the method replaces the current alarm system then it may minimize false positive alarms. On the basis of the proposed rule, an EBI may be identified, and the clinician may be warned by a text or voice message or pulmonary display.
We conclude that the proposed five-element lumped pulmonary model can identify EBI in real time using noninvasive measurement of airway pressure and flow. The model can successfully differentiate EBI from SCW with both sensitivity and specificity of 90%. A decrease in compliance without any significant change in resistance is specific to EBI, whereas a decrease in compliance accompanied with an increase in resistance is specific to SCW. The practical benefit of the model-based detection to the clinicians is that perhaps the model may be integrated with alarms and displays in ICUs and ORs to specifically identify EBI. The present study is a promising step toward noninvasive, real-time detection of EBI to aid clinicians in decision making.
|
ACKNOWLEDGMENTS |
|---|
|
Footnotes |
|---|
Supported by NIH-NIBIB grant R01 EB000294-05.
|
REFERENCES |
|---|
|
TopAbstractIntroductionMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  S. Tejman-Yarden, A. Zlotnik, L. Weizman, J. Tabrikian, A. Cohen, N. Weksler, and G. M. Gurman Acoustic Monitoring of Lung Sounds for the Detection of One-Lung Intubation Anesth. Analg., August 1, 2007; 105(2): 397 - 404. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
) measured at the mouth are inputs to the model.
