A Predictive Formula for Dicult Endotracheal Intubation in the Emergency Department

Background Current predictors for evaluating dicult endotracheal intubation had poor accessibility or sensitivity at the emergency department, so we evaluated the incidence and predictive factors, then built an easy-to-use predictive formula. Methods For the 110 patients, dicult airway was dened as Cormack & Lehane classication grade III and IV at rst attempt of intubation. The univariate associations between patient characteristics and dicult endotracheal intubation were then analyzed, and the signicantly associated factors were included in a multivariate binary logistic regression model then a predictive formula was generated. Generalized association plot (GAP) was used to show the relationship between each variables. Results The incidence of dicult intubation in our study was 35.5%. In the dicult airway group, signicantly higher rates (p<0.05) of high body mass index (BMI); double chin; thick, short neck; Mallampati diculty; smaller inter-incisors distance; smaller thyromental distance; and upper airway obstruction were noted. Finally, a predictive formula for dicult intubation was successfully established by the combination of four predictors: BMI (odds ratio [OR]=1.270), thyromental distance (OR=0.614), upper airway obstruction (OR=4.038), and Mallampati diculty (OR=5.163). A cut-off score of 4 provided the best sensitivity (79.5%) and specicity (81.7%)(95% CI: 0.794 to 0.938). Conclusions Our predictive formula could be used by emergency physicians to quickly identify and carefully manage patients with potentially dicult intubation. Early expert consultation could be sought when necessary.


Background
Airway assessment is important for emergency physicians because identifying patients with di cult airways allows for more careful intubation and early expert consultation; decreased intubation attempts and adverse events such as airway trauma, esophageal intubation, aspiration, hypoxemia, hypotension, dysrhythmia, and even cardiac arrest (1).
For this unmet need, researchers have tried to identify the predictors of di cult intubation. Because of the urgency in the emergency department (ED), the predictive method should be as simple as possible. The Mallampati score ( Fig. 2A) is the most commonly used bedside tool for predicting di cult endotracheal intubation. While it offers easy preoperative performance, it has limited accuracy and only poor to moderate discriminative power when used alone (2). LEMON score is another predictor, which includes look externally, the 3-3-2 rule, Mallampati score more than 3, obstruction, and neck mobility. However, the judgment of "look externally" is too subjective and the 3-3-2 rule is based on nger-widths, which vary among operators.
To address the limitations of the Mallampati and LEMON score, researchers have proposed the use of the upper lip bite test, thyromental distance, inter-incisor gap, sternomental distance (3), thyromental height (4), acromio-axillo-suprasternal notch index (5), and lower jaw protrusion maneuverability (2). However, not all of these measurements are universal; for example, thyromental height is not a strong predictor of di cult visualization of the larynx in Japanese patients (6). Other researchers have proposed combinations of these measures for precise prediction. Srivilaithona et al. showed that the weighted combination of ve independent predictors (male sex, large tongue, limited mouth opening, poor neck mobility, and obstructed airway) helped to discriminate di cult intubation patients (7); however, the use of so many predictors is not suitable for ED. In this study, we evaluated the incidence and predictive factors of di cult intubation and developed an easy-to-use predictive formula for emergency physicians. physicians were asked for enrollment to this study. These participants consented to join this study and wrote informed consent by themselves or were volunteered by their delegates. The exclusion criterions included patient age under 20 years old, intubated by non-participant physicians, intubation via videolaryngoscope, or refused to join the study by themselves or delegates. Comatose patients were included if their delegate agreed after routine intubation.

De nition and data collection
First intubation attempt that could be classi ed as Cormack & Lehane classi cation grade III or IV were termed di cult (Fig. 2B), and Cormack & Lehane classi cation grade I or II were termed non-di cult.
These scores were assigned during the rst view by direct laryngoscopy. Patients were categorized into two groups for comparison and evaluation.
The anatomical distances measured included the inter-incisor, thyromental, sternomental, and thyrosternal distances (Fig. 3). The inter-incisor distance was de ned as the distance between the upper and lower incisors, with the patient's mouth most widely opened. The other distances were measured with the patient's neck in full extension. As shown in Fig. 3, thyromental distance was measured from the bony, prominent point of the mentum to the thyroid notch with the patient's mouth closed; thyrosternal distance was measured from the thyroid notch to sternal notch and the sternomental distance was measured from the mentum to the sternal notch. These three distances formed a triangle. The thyromental and thyrosternal distances together were longer than the sternomental distance except when the mentum, thyroid notch and sternal notch were in a straight line. Rulers were provided to all participant physicians for easy access.
When the patient's mouth was open, the physicians performed initial foreign body removal or sputum suction so that the airway was well-prepared. After this preprocessing, the remaining narrowing of the airway was classi ed as sputum, food, or blood impaction.
Generally, the initial intubation was attempted by the chief residents or attending staffs, whose extensive experience with emergency intubation could prevent bias due to inexperience, and all of them are emergency medicine specialists. Each chief resident or staff member used a specially designed form to check and record the patient's airway condition, including number of attempts, and the attending staff made all clinical decisions regarding airway management. A research assistant synthesized and code the raw data, also checked for missing values regularly, and contacted the original intubating physician for clari cation or re-measurement of the missing values, nally, the data would be sent to the statistics center of our hospital. The datasets used and analysed during the current study are available from the corresponding author on reasonable request.

Statistical analysis
SPSS Statistics for Windows, version 17.0 (SPSS Inc., Chicago, IL) was used for data management and statistical analysis. Categorical variables were described using frequency distributions and reported as n (%), while continuous variables were reported as mean ± standard deviation.
Univariate analysis was used to evaluate the association between patient characteristics and di cult endotracheal intubation. Categorical variables were assessed by Pearson's chi-squared or Fisher's exact tests, while continuous variables were assessed by Student's t tests. Statistical signi cance was set at an alpha level of 0.05. Odds ratios (ORs) and 95% con dence intervals (CIs) of individual patient characteristics were also computed to assess the potential risk factors for di cult endotracheal intubation.
All individual risk factors signi cantly associated with di cult endotracheal intubation in the univariate analysis were included in a multivariate binary logistic regression model to reduce the type I error, except for intubation injury and the mean number of attempts, because these factors were consequences of successful intubation; thus, they could not serve as predictive factors.
Based on the results of the logistic regression model, a predictive formula comprising the β coe cient of no more than 4 independent predictors due to one-in-ten rule was established. To calculate predictive score, each β coe cient was multiplied by its independent predictor value and the results were summed as the nal score.
To investigate the relationship between each variables, a correlation matrix with hierarchical dendrogram was performed with RStudio package (RStudio: Integrated Development for R. RStudio, Inc., Boston, MA.) and generalized association plots (GAP) (39,40).

Results
In this 17-month duration, up to 1960 patients were intubated in emergency department of MacKay Memorial Hospital, but 316 patients were intubated by non-participant physicians, 21 patients were less than 20-year-old, and 1513 patients were not consentable, so nally 110 adult endotracheal intubation patients (66 men and 44 women) met our criteria and were enrolled in this study, and there were no missing data from them ( Fig. 1). Their ages ranged from 21 to 96 years. The incidence of di cult endotracheal intubation was 35.5%. The di cult airway group had higher BMI than the non-di cult airway group, but no signi cant differences in age, sex, and di cult mask ventilation were observed between two groups (Table 1A).
The causes of intubation (Table 1B) included sepsis, lung disease, heart disease, renal disease, inhospital cardiac arrest, out-of-hospital cardiac arrest, neurological problems such as intracranial hemorrhage/cerebrovascular accident, and facial trauma (in one case); however, causes of intubation did not differ signi cantly between two groups (Table 1B).
The presentations of intubated patients are shown in Table 1C. The prevalence of double chin; sunken cheeks; thick, short neck; and Mallampati di culty (grade 3 or 4) differed signi cantly between the two groups. However, the prevalence of receding mandible, snoring, lack of teeth, and poor neck mobility did not. Table 1D shows the four anatomical distances in the head and neck. Two of the variables, maximal interincisor distance and thyromental distance, differed signi cantly between the groups (p = 0.014). However, the other two variables, sternomental distance and thyrosternal distance, had no signi cant difference between the groups which means the two distances were almost equal between two groups.
Upper airway obstruction is shown in Table 1E. The three subitems of upper airway obstruction represented different agents for impaction: sputum, food, and blood. Only upper airway obstruction and sputum impaction differed signi cantly. Table 1F shows the methods of intubation, related injuries, and the mean number of attempts. Before anesthesia induction, 6.4% (7/110) of patients were fully conscious (Glasgow Coma Scale 15), half (55/110) were confused, and 26.3% (29/110) required cardiorespiratory resuscitation. Because not all our patients were comatose, two methods of intubation were used: rapid sequence or coma and sedationonly without paralysis. Neither of the two methods showed signi cant differences. Intubation injury and the mean number of attempts were also important factors, but only the mean number of attempts differed signi cantly. Table 2 shows the multivariate ORs of di cult endotracheal intubation. For every one-point increase in BMI, the chance of di cult intubation increased by 27.0% (OR 1.270, 95% CI, 1.111 to 1.451, p < 0.001); for every 1-cm increase in thyromental distance, it decreased by 38.6% (OR 0.614, 95% CI, 0.395 to 0.955, p = 0.03). Patients with upper airway obstruction were 4.04 times more likely to have di cult intubation than those with non-upper airway obstruction (OR 4.038, 95% CI, 1.456 to 11.200, p = 0.007). Patients with Mallampati grades of 3 or 4 were 5.16 times more likely to have di cult intubation than those with Mallampati grades of 1 or 2 (OR 5.163, 95% CI, 1.895 to 14.066, p = 0.001).  Receiver operating characteristic (ROC) curves were drawn to identify the optimal cut-off point (Fig. 4).
The area under the ROC curve was 0.866 (95% CI: 0.794 to 0.938, p < 0.001), and an integer cut-off score of 4 showed a sensitivity of 79.5% and speci city of 81.7%, demonstrating perfect classi cation with the highest Youden's index of 61.18% (Table 3).  correlation and blue indicated negative one. The darker the color, the higher the Pearson product-moment correlation coe cient. Hierarchical clustering tree was also used to sort the matrix that put the variables which were highly related together, such as thyrosternal distance and sternomental distance, which had a high positive correlation with more than 0.7 of Pearson coe cient, and the two distances also had a moderate positive correlation with inter-incisor distance, however, compared the three distances with thyromental distance, there was only 0.2 of Pearson coe cient. In the hierarchical clustering tree, age and lack of teeth, also sunken cheeks and di cult mask ventilation were proximate, and both the two pairs of variables had approximately 0.3 of Pearson coe cient. The variable of operator was far to not only intubation injury, but also number of attempts, and even di cult intubation in the hierarchical clustering tree, and in the matrix, the colors for its relationship to each variable were almost light.

Discussion
Anatomical changes such as dental loss, head and neck joint changes (8-11) affect airway management, including intubation and ventilation (12). A prospective study reported that head and neck movement, thyromental distance, and inter-incisor gap decreased with age; however, the dentition grade, Mallampati score, and cervical joint rigidity increased. Therefore, the authors concluded that middle-aged or elderly adults had a higher risk of di cult endotracheal intubation (13). However, we observed no signi cant difference in the proportion of elderly patients between the di cult and non-di cult intubation groups. This difference may be because our age cutoff was 65 years, while Rose and Cohen showed that patients in the 40-59-year age range were at risk for di cult endotracheal intubation (8); we de ned this middle-aged group as "non-elderly," which may have masked the effect of aging on di cult intubation.
We recorded the indications for intubation because they may have been key factors; for example, head and neck trauma may decrease neck mobility and even affect the measurement of thyromental, thyrosternal, and sternomental distances. Endotracheal intubation is often considered a contraindication in cases with maxillofacial trauma as maxillomandibular xation may be disturbed (14). Therefore, it was recommended that intubation of the laryngeal mask airway be included in the algorithm for these patients (15). However, our study included only one patient with facial trauma, which limited our evaluation. No other indications for intubation, including lung, heart, and renal disease, differed signi cantly between two groups. Therefore, the cause of intubation may not be a major factor related to di cult intubation.
The lack of signi cant differences in some presentation variables may have been due to trouble in measurement or de nition such as lack of teeth and poor neck mobility. Varying presentations were possible for the "lack of teeth"; "lack" could range from one to all. If all the teeth are lost, it will be di cult for upper and lower gums to come into contact, limiting temporomandibular joint movement (16). The location of tooth loss also requires clari cation. Evidence has revealed that the loss of posterior teeth results in over-closure of the oral cavity (17) while the loss of anterior teeth may allow easy access to the airway and prevent intubation-related teeth loss (18,19). Regarding poor neck mobility, despite Brit Long's report of the positive likelihood ratio of impaired neck mobility for predicting di cult endotracheal intubation (20), and in Wilson's di cult endotracheal intubation predictive model, head and neck movement was divided into three levels to calculate the risk (21). Nevertheless, measurement is di cult in humpbacked patients because their kyphotic deformities are related to spinal osteoporosis and degeneration of intervertebral discs (17); as the neck of patients cannot fully extend, clearly discriminating neck and back stiffness from non-precise measurement is di cult, as shown by our raw data.
Intubation injuries included mechanical damage to the patients' teeth and/or airways, hematoma formation, and even aspiration of gastric contents (29)(30)(31). The most common cause of postintubation injuries was over lled cuffs (31). A systematic review reported a high prevalence of intubation-related laryngeal injury (83%) (32). However, in our study, the rates were 2.6% and 5.6% in the di cult and nondi cult endotracheal intubation groups, respectively; both were signi cantly lower than previously reported, which may be due to the high consensus among the participant physicians in the present study. Thus, they were likely more careful and attentive when evaluating and intubating these patients.
The four key predictive factors in our formula were reasonable; the combination of Mallampati classi cation and thyromental distance is preferable for airway assessment because of its better speci city and positive predictive value over either alone (33). However, this combination had low sensitivity, indicating the need for additional predictive factors; thus, the results of our study may be a possible solution to this obstacle.
The previously reported multi-factor predictive models included Naguib's new model, a predictive formula developed using logistic regression that includes thyromental distance, Mallampati score, inter-incisor gap, and height (34). While the rst two factors were also included in our algorithm, inter-incisor gap was not, despite its signi cant association with di cult intubation in the univariate analysis. Moreover, our formula used BMI rather than height. Other studies have also reported that di cult intubation is more common among obese patients (35), indicating that BMI is a more powerful predictor.
In our formula, Mallampati score had the largest coe cient, making it the most important factor. A previous study reported no signi cant association between an increased Mallampati score and di cult intubation; thus, their revised LEMON methods excluded the weight of Mallampati score (36). However, they also reported that the Mallampati score was not easily available and was obtained in only 57% of patients. In comparison, we assessed all patients successfully; thus, their negative result may have been a result of too many missing Mallampati scores.
Based on our ndings, we recommend the routine assessment of Mallampati score in all patients regardless of whether the patient has been admitted to the ED or general ward. Because the Mallampati score had the largest β coe cient value, it can allow early detection of those with a high risk of di cult endotracheal intubation; when these patients require intubation, the necessary time for evaluation may not be enough, so pre-assessment will allow physicians to be well-prepared for this challenge.
In our study, all patients were pre-treated with initial sputum suction and foreign body removal; therefore, recorded upper airway obstruction indicated at least middling to severe obstruction due to residual or newly produced issues. As a component of the modi ed LEMON criteria, the sensitivity and negative predictive value of upper airway obstruction are well-validated (37). Our results showed that none of the upper airway obstruction subclasses were signi cant. Previous studies did not evaluate upper airways obstruction subclasses separately or only reported the percentages of each kind of impaction overall (38). To our knowledge, our study is the rst to show that individual evaluation of each kind of impaction may show non-signi cant results, but grouping all impaction types into a single variable can reveal signi cant differences.
Sternomental and thyrosternal distances did not differ signi cantly in our study. GAP analysis implied that although sternomental, thyrosternal, and thyromental were anatomical distances, but thyromental distance had its distinctness due to the smaller Pearson coe cient to all the other distances. Ramadhani reported sternomental distance as an indicator of head and neck mobility in 1996 (22). One study reported a signi cant difference in this anatomical distance between the single and multiple-attempt laryngoscopy groups (23). However, our study did not compare single and multiple attempts, and several factors, such as the di culty in evaluating neck mobility, may affect the results. Another study also did not recommend sternomental distance as the sole predictor (24), indicating its weaker predicting role. In contrast, the non-signi cant difference in thyrosternal distance was consistent with the ndings of previous studies. Unlike thyromental distance, thyrosternal distance was not a good bedside parameter for predicting di cult endotracheal intubation (25)(26)(27)(28).
In previous studies, it has been proven that the number of remained dentate decreased when people become older and older, and the elderly also had higher proportion of becoming completely edentulous (41,42), our GAP analysis also found the same trend, the variables of age and lack of teeth were moderately correlated. Besides, the moderately correlation of sunken cheeks and di cult mask ventilation can also be shown by GAP analysis, which matches the result of previous studies, sunken cheeks was found to independently identify di cult mask ventilation (43,44), so by the above two associations, this visualization tool is reasonable and reliable. However, we found sunken cheeks was a predictor of non-di cult endotracheal intubation in this study, it may imply less intubation resistant from facial muscle.
The training level of operator should be considered, so similar study excluded the case which is intubated by low experience operators, including last year of medical student and rst-year internist in general practice (45). Therefore, we only included the patients who were intubated by chief residents and attending physicians, and the GAP analysis also demonstrated that the variable of operator has almost zero correlation to all of the adverse index such as intubation injury, number of attempts, and even di cult intubation, implied that our chief residents and attending physicians have almost equal level of the skill of endotracheal intubation, so there was less experience related bias in our study.
Our study had some limitations. First, our formula directly used the β coe cient and the four coe cients were not integers because we did not adjust them to avoid decreasing the sensitivity or speci city. While this prevents easy calculation by physicians, the derived formula can be entered into a computer as a simple program or application to solve this problem. Second, because our emergency physicians had heavy work, they could only enroll patients when they were relatively unbusy, therefore, we included only 110 patients, which was far fewer than the numbers included in other similar studies, however, as approximately 1000 patients are intubated in the emergency department of MacKay Memorial Hospital each year, our study of 110 patients could be considered a sampling survey, and we also plan to enroll a new group of patients to validate our derived formula. Furthermore, the prospective design meant that we could avoid missing data, unlike in retrospective studies. Third, most of the agreements were signed by the patient's delegates after successful intubation; an increasing number of intubation attempts would likely increase the di culty of enrollment due to patient family anxiety and irritation. However, the participant physicians participating in this study may be more careful and have highly consensus about intubation; therefore, the incidence of di cult intubation may not have been underestimated.

Conclusions
In this study, di cult intubation was associated with increased BMI, lesser thyromental distance, upper airway obstruction, and Mallampati di culty (grade 3 or 4), our predictive formula which based the above variables allowed careful management and early expert consultation when necessary. Moreover, GAP analysis strengthened the special role of thyromental distance among other anatomical distances, and also shown the interactive relationship between each variables by data visualization. Availability of data and materials The datasets are available from the corresponding author on reasonable request. All participants signed informed consent by themselves or were volunteered by their delegates.

Consent for publication
All participants provided their consent by themselves or their delegates.

Figure 1
The study ow diagram. In this duration, we had 1960 endotracheal intubated patients who were potentially eligible study subjects, but after excluded the patients who were intubated by non-participant physicians, underage patients, and non-consentable ones, nally, 110 patients were included for further analysis.  De nitions of thyromental, thyrosternal, and sternomental distances. The three anatomical landmarks of airway evaluation include the bony prominent point of the mentum and the thyroid notch and sternal notch. The combination of two of these three landmarks allows measurement of the three distances (A, B, C), which form a triangle. In most patients, A plus B is larger than C, but some patients' neck could be extended so that C is equal to A plus B.

Figure 4
Receiver operating characteristic (ROC) curve of our predictive formula for di cult endotracheal intubation. ROC curves depicting the relationship between the sensitivity and speci city of our model for predicting di cult endotracheal intubation. The area under the ROC (AuROC) shows the discriminating power. AuROC=0.866 (95% con dence interval; 0.794, 0.938); Cut-off score = 4; Sensitivity=79.5%; Speci city=81.7%.

Figure 5
Proximity matrix with hierarchical clustering tree between each variable of demographic factors and physical ndings. The color in proximity matrix (or the heatmap) showed that the Pearson's correlation was positive (Red) or negative (Blue) between each pair of variables, and the hierarchical tree made the related variables became a cluster, the lower hierarchy meant the closer relationship, which could correspond to the correlation coe cient scale on the top of the tree.