Patient and Operative Factors Predict Risk of Discretionary Prolonged Postoperative Mechanical Ventilation in a Broad Surgical Cohort

Abstract

Background:

Patients undergoing surgery with general anesthesia and endotracheal intubation are ideally extubated upon case completion, as Prolonged Postoperative Mechanical Ventilation (PPMV) has been associated with poor outcomes. However, some patients require PPMV for surgical reasons, such as airway compromise, while others remain intubated at the discretion of the anesthesia provider. Incidence and risk factors for discretionary PPMV (DPPMV) have been described in individual surgical subspecialties and intensive care unit (ICU) populations, but are relatively understudied in a broad surgical cohort. The present study seeks to fill this gap and identify the perioperative risk factors that predict DPPMV.

Methods:

After obtaining IRB exemption, existing electronic health record databases at our large referral center were retrospectively queried for adult surgeries performed between January 2018 and December 2020 with general anesthesia, endotracheal intubation, and by surgical services that do not routinely leave patients intubated for surgical reasons. Patients who arrived to the ICU intubated after surgery were identified as experiencing DPPMV. Selection of candidate risk factors was performed with LASSO-regularized logistic regression, and surviving variables were used to generate a multivariable logistic regression model of DPPMV risk.

Results:

A total of 32,915 cases met inclusion criteria, of which 415 (1.26%) experienced DPPMV. Compared to extubated patients, those with DPPMV were more likely to have undergone emergency surgery (42.9% versus 3.4%, p < 0.001), surgery during an existing ICU stay (30.8% versus 2.8%, p < 0.001), and have 20 out of the 31 Elixhauser comorbidities (p < 0.05 for each comparison), amongst other differences. A risk model with twelve variables, including ASA Physical Classification Status, emergency surgery designation, four Elixhauser comorbidities, surgery during an existing ICU stay, surgery duration, estimated number of intraoperative handoffs, and vasopressor, sodium bicarbonate, and albuterol administration, yielded an area under the receiver operating characteristic curve of 0.97 (95% confidence interval, 0.96–0.97) for prediction of DPPMV.

Conclusions:

DPPMV was uncommon in this broad surgical cohort but could be accurately predicted using readily available patient-specific and operative factors. These results may be useful for preoperative risk stratification, postoperative resource allocation, and clinical trial planning.

Introduction

Prolonged Postoperative Mechanical Ventilation (PPMV) has been associated with adverse outcomes, including both short- and long-term mortality and increased length of stay in the Intensive Care Unit (ICU).^1,2 Failure of initial extubation (reintubation) has also been associated with adverse outcomes, including increased mortality and prolonged hospitalization.^3,4 The need to weigh the relative impact of these risks can complicate the decision to extubate at the end of surgery.^5,6

Risk factors for PPMV have been evaluated for individual surgeries and in surgical subspecialties (e.g., cardiac,^7,8 thoracic,⁹ and spine surgery)¹⁰, as well as in ICU populations.⁵ Commonly implicated risk factors include patient characteristics (e.g., age, American Society of Anesthesiologists (ASA) Physical Classification Status, and medical comorbidities) as well as operative characteristics (e.g., surgery type, surgery duration, and surgery end time). Staffing data, such as number of attending handoffs, have also been associated with delayed extubation.¹¹ Although composite outcomes of PPMV and postoperative reintubation have been assessed in broad surgical cohorts,^12,13 risk factors for PPMV alone are relatively understudied in this population.

The ability to predict PPMV across a multitude of surgeries has important implications for preoperative risk assessment and postoperative resource allocation.¹⁴ While there are no universal or fixed criteria for PPMV, indications can be broadly grouped into two categories: those that obligate endotracheal intubation and mechanical ventilation, such as airway obstruction or management of cerebral edema, and those acted upon at the discretion of the anesthesia provider, such as varying degrees of acid-base disturbance or hemodynamic instability. While conditions that obligate PPMV are usually anticipated during preoperative planning, the need for discretionary PPMV (DPPMV) is more difficult to identify in advance of the decision to extubate. Our objective is therefore to describe the incidence of DPPMV, and identify the patient and perioperative risk factors that predict the need for DPPMV.

Methods

The Institutional Review Board (IRB) at Vanderbilt University Medical Center (VUMC) approved an exemption for this study including a waiver of the requirement for written informed consent (#191068). The Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) guidelines were used in the preparation of this manuscript.¹⁵

Eligibility & Data Collection

Data of patients undergoing surgeries between January 1, 2018 and December 31, 2020 at Vanderbilt University Medical Center (VUMC), a quaternary care center, were retrospectively extracted from the VUMC Perioperative Data Warehouse (Nashville, TN) using structured query language (SQL). All available cases were assessed in order to maximize the number of observations per degrees of freedom in the final model. Adults (age ≥ 18 years on date of surgery) who underwent surgery with general anesthesia and endotracheal intubation in one of Vanderbilt’s main operating rooms were eligible for inclusion. Procedures at off-site ambulatory surgery centers were excluded. Cases with intraoperative death, deceased-donor organ donation (ASA Physical Classification Status 6), or secondary events (e.g. imaging) between operating room departure and Post-Anesthesia Care Unit (PACU) or Intensive Care Unit (ICU) arrival were also excluded. All patients had to be free from mechanical ventilation for at least twenty-four hours preoperatively. Cases were also excluded if they were boarded under emergency general surgery, trauma surgery, otolaryngology, oral and maxillofacial surgery, cardiac surgery, thoracic surgery, hepatobiliary surgery, or neurosurgery, as patients in these services frequently require obligatory PPMV for surgical reasons. Consequently, PPMV in the remaining patients was considered to occur at the discretion of the anesthesia provider (i.e., DPPMV).

DPPMV was specifically operationalized as an arrival to the ICU intubated after surgery. Patients could be transported directly from the operating room or via the PACU in the event of ICU overflow. No minimum or maximum duration of mechanical ventilation was required. For cases meeting criteria for DPPMV, manual chart review was performed by a study author (MGC) and ambiguous cases were discussed with an attending anesthesiologist (REF, DAM) to confirm accuracy. All other patients were extubated in the operating room at the end of their operation, or in the PACU shortly thereafter.

Candidate risk factors were identified based on literature review and clinical experience, then screened for physiologic probability, availability in the electronic health record (EHR) as structured data, and independence from the other factors. For each eligible case, we collected patient age, sex, ASA Physical Classification Status, ASA emergency case designation, surgery duration, surgery end time, presence of a difficult airway chart flag, Elixhauser comorbidities (as assessed at the end of the patient’s hospitalization using International Classification of Diseases, Ninth Revision and Tenth Revision codes), number of distinct attending and non-attending anesthesia staff assigned to the case, selected medication administration data (for vasopressors, inotropes, albuterol, and sodium bicarbonate), airway placement and removal times, and, if applicable, ICU admission time. The fluid and electrolyte disorders Elixhauser comorbidity included ICD-9 codes 253.6 (syndrome of inappropriate antidiuretic hormone secretion) and 276.0–276.9 (electrolyte abnormalities, acid-base disturbances, and volume overload/depletion). The number of intraoperative handoffs was estimated as the sum of distinct anesthesia attendings in excess of one and distinct non-attendings (anesthesia residents and CRNAs) in excess of one. Accordingly, a case with one attending and one resident would have zero handoffs, while a case with one attending and two residents would have one handoff. Staff providing coverage for breaks were included in these calculations. Intraoperative medications were categorized as inotropes, vasopressors, or assigned to both categories based upon their physiologic effects. Dobutamine and milrinone were categorized as inotropes; phenylephrine, norepinephrine, and vasopressin were categorized as vasopressors; and dopamine and epinephrine were categorized as both inotropes and vasopressors. In all, forty-four candidate risk factors were assessed.

Outcomes data, including in-hospital mortality, thirty-day mortality, and need for a Rapid Response Team (RRT) activation within seven days of surgery, were also retrieved for each eligible patient.

Statistical Analysis

The candidate risk factors were summarized in the DPPMV and control (extubated) groups using medians with interquartile ranges (IQRs) for continuous variables and counts with percentages for categorical variables. Nonparametric univariate analysis was accomplished using Mann-Whitney U tests for continuous variables, chi-squared tests for categorical variables (or Fisher exact tests for counts less than five), and Kruskal-Wallis tests for ordinal variables.^16,17

We implemented a two-step regression procedure to develop and validate the model for predicting DPPMV risk. In the first step, we performed variable selection with a Least Absolute Shrinkage and Selection Operator (LASSO)-regularized logistic regression. Ten-fold cross validation was used to tune the regularization parameter $\widehat{\lambda }$, and an optimal penalty value was determined with area under the receiver operating characteristic curve (AUC) as the objective function to maximize. In order to minimize the risk of overfitting while maintaining near-optimal model performance, λ was chosen as the maximum value at which the mean AUC was within one standard error of the estimate at $\widehat{\lambda }$. Of note, for modeling purposes, ASA physical classification status was treated as a categorical variable to avoid the additional model complexity associated with an ordinal representation. In the second step, variables that survived at the chosen λ were used to create a separate multivariable logistic regression model of DPPMV risk. The associations were then summarized and reported using odds ratios (ORs) with 95% confidence intervals (CIs) and Wald tests. Performance of this model was assessed using 2000 bootstraps of a Receiver Operating Characteristic (ROC) curve. In each iteration, AUC was recorded, and sensitivity and specificity were assessed at the point maximizing Youden’s J statistic (sensitivity + specificity − 1). Averages and 95% confidence intervals for these parameters are reported. A calibration plot and Brier score were computed for the final model. Additionally, a scaled Brier score was computed, defined as 1 – [(Brier score)/(null model Brier score) where the null model contains no predictors. By this definition, a score of 1 is ideal, while scores ≤ 0 represent poor or very poor performance.¹⁸

Furthermore, to assess practical utility of the model in the preoperative time period, we conducted two sensitivity analyses. The first consisted of a model without any of the surviving Elixhauser comorbidity terms, as these data might only be available postoperatively. The second consisted of a model without any of the surviving intraoperative terms, such as surgery duration or intraoperative medication administration, that might not be known until the end of surgery. Performance of each model was assessed in an identical manner to that described above.

We also performed a supplemental unadjusted post-hoc analysis of all-cause reintubation rates within the cohort. We selected a time period of twenty-four hours following initial extubation for its relevance to the decision to extubate at the end of the case. Reintubation rates, as well as differences in in-hospital mortality, thirty-day mortality, and need for a Rapid Response Team (RRT) activation within seven days of surgery, were compared between the two groups using chi-squared tests, or Fisher exact tests for counts less than five.

All statistical analyses were performed using R (R Foundation for Statistical Computing, Vienna, Austria; RStudio 1.3.1093, Boston, Massachusetts). The LASSO-regularized regression was accomplished with the glmnet package (version 4.1) and the calibration plot was produced with the rms package (version 6.1-1). The significance threshold was set at α = 0.05.

Results

The final cohort included 32,915 cases, of which 440 met criteria for DPPMV. Reassignment of 25 false positives (5.7%) identified during manual review yielded 415 confirmed cases of DPPMV (Figure 1). Incidence of DPPMV in this cohort was therefore 1.26%. Of all included cases, the majority were performed by renal and urologic surgery (N = 8,042, 24.4%), orthopedic surgery (N = 6,630, 20.1%), or general surgery (N = 5,321, 16.2%). For cases of DPPMV, the most common services were vascular surgery (N = 132, 31.8%) and orthopedic surgery (N = 106, 25.5%), followed by renal and urologic surgery (N = 47, 11.3%) and general surgery (N = 40, 9.6%).

Figure 1.

Cohort and case selection. * Note that a case may meet more than one exclusion criterion.

Results of the univariate analysis are shown in Table 1. The risk of DPPMV was associated with older patients (median 58.6 versus 55.6 years, p = 0.011), male sex (58.6% versus 48.0%, p < 0.001), higher ASA Physical Status Classification (p < 0.001), and 20 of the 31 Elixhauser comorbidities (p < 0.05). DPPMV was also strongly associated with emergency surgery (42.9% versus 3.4%, p < 0.001). There was no difference between the groups in presence of a known difficult airway (1.7% versus 1.4%, p = 0.712).

Table 1:

Univariate analysis.

Variable	Extubated	DPPMV	P Value
	N=32,500	N= 415
Demographics
Age, yr - median (IQR)	55.6 (26.2)	58.6 (23.8)	0.011a ***
Male - no. (%)	15614 (48.0%)	243 (58.6%)	<0.001b ***
ASA class - no. (%)			<0.001d ***
1	1035 (3.2%)	2 (0.5%)
2	10132 (31.2%)	33 (8.0%)
3	19623 (60.4%)	195 (47.0%)
4	1704 (5.2%)	168 (40.5%)
5	6 (0.0%)	17 (4.1%)
6	0 (0%)	0 (0%)
ASA class E - no. (%)	1115 (3.4%)	178 (42.9%)	<0.001b ***
Difficult airway flag- no. (%)	440 (1.4%)	7 (1.7%)	0.712b
Elixhauser total score - median (IQR)	0 (5)	6 (14.5)	<0.001a ***
Congestive heart failure - no. (%)	1911 (5.9%)	84 (20.2%)	<0.001b ***
Arrhythmia - no. (%)	5440 (16.7%)	262 (63.1%)	<0.001b ***
Valvular disease - no. (%)	439 (1.4%)	21 (5.1%)	<0.001b ***
Pulmonary circulation disease - no. (%)	520 (1.6%)	38 (9.2%)	<0.001b ***
Peripheral vascular disease - no. (%)	1533 (4.7%)	115 (27.7%)	<0.001b ***
Hypertension, uncomplicated - no. (%)	13737 (42.3%)	228 (54.9%)	<0.001b ***
Hypertension, complicated - no. (%)	2820 (8.7%)	68 (16.4%)	<0.001b ***
Paralysis - no. (%)	317 (1.0%)	14 (3.4%)	<0.001b ***
Neurological disease - no. (%)	868 (2.7%)	18 (4.3%)	0.053b
Chronic pulmonary disease - no. (%)	4890 (15.0%)	107 (25.8%)	<0.001b ***
Diabetes, uncomplicated - no. (%)	3679 (11.3%)	66 (15.9%)	0.004b ***
Diabetes, complicated - no. (%)	3578 (11.0%)	107 (25.8%)	<0.001b ***
Hypothyroidism - no. (%)	3518 (10.8%)	47 (11.3%)	0.805b
Renal failure - no. (%)	3860 (11.9%)	105 (25.3%)	<0.001b ***
Liver disease - no. (%)	1371 (4.2%)	43 (10.4%)	<0.001b ***
Peptic ulcer disease - no. (%)	142 (0.4%)	3 (0.7%)	0.435c
AIDS/HIV - no. (%)	180 (0.6%)	3 (0.7%)	0.504c
Lymphoma - no. (%)	140 (0.4%)	2 (0.5%)	0.701c
Metastatic cancer - no. (%)	2157 (6.6%)	37 (8.9%)	0.08b
Solid tumor - no. (%)	7409 (22.8%)	71 (17.1%)	0.007b ***
Rheumatoid arthritis - no. (%)	946 (2.9%)	26 (6.3%)	<0.001b ***
Coagulopathy - no. (%)	1090 (3.4%)	132 (31.8%)	<0.001b ***
Obesity - no. (%)	6529 (20.1%)	94 (22.7%)	0.218b
Weight loss - no. (%)	1409 (4.3%)	116 (28%)	<0.001b ***
Fluid/electrolyte disorder - no. (%)	3506 (10.8%)	306 (73.7%)	<0.001b ***
Blood loss anemia - no. (%)	334 (1.0%)	11 (2.7%)	0.003b ***
Deficiency anemia - no. (%)	759 (2.3%)	11 (2.7%)	0.796b
Alcohol abuse - no. (%)	483 (1.5%)	20 (4.8%)	<0.001b ***
Drug abuse - no. (%)	884 (2.7%)	38 (9.2%)	<0.001b ***
Psychoses - no. (%)	142 (0.4%)	4 (1.0%)	0.114c
Depression - no. (%)	4204 (12.9%)	63 (15.2%)	0.201b
Operative Characteristics
Case during ICU stay - no. (%)	922 (2.8%)	128 (30.8%)	<0.001b ***
OR duration, min. - median (IQR)	165 (128)	249 (248)	<0.001a ***
Case end 1601-0659 - no. (%)	9975 (30.7%)	283 (68.2%)	<0.001b ***
E,t,imat.e,d handoff, - no. (%)			<0.001d ***
0	24929 (76.7%)	191 (46%)
1+	7571 (23.3%)	224 (54%)
Intraoperative Medications
Vasopressor infusion - no. (%)	5779 (17.8%)	171 (41.2%)	<0.001b ***
Inotrope infusion - no. (%)	28 (0.1%)	3 (0.7%)	0.007c ***
Albuterol administration - no. (%)	1155 (3.6%)	51 (12.3%)	<0.001b ***
Bicarbonate administration - no. (%)	193 (0.6%)	75 (18.l%)	<0.001b ***

IQR: interquartile range; aMann-Whitney U; bChi squared; cFisher exact; dKruskal-Wallis;

^“***”denotes statistical significance.

Compared to patients who were extubated, patients with DPPMV were much more likely to have their operation during an existing ICU stay (30.8% versus 2.8%, p < 0.001), and tended to have longer surgeries (median 249 versus 165 minutes, p < 0.001) that were more likely to end after normal working hours (68.2% versus 30.7%, p < 0.001) and involve a higher number of estimated intraoperative handoffs (p < 0.001). Use of vasopressors (p < 0.001), inotropes (p = 0.007), albuterol (p < 0.001), and sodium bicarbonate (p < 0.001) were more common in cases with DPPMV.

Twelve variables survived LASSO regularization, including ASA Physical Classification Status, ASA emergency status, four Elixhauser comorbidities (peripheral vascular disease, arrhythmia, fluid/electrolyte disorder, and coagulopathy), case during an existing ICU stay, surgery duration, estimated number of intraoperative handoffs, vasopressor infusion, sodium bicarbonate administration, and albuterol administration. Results are shown in Table 2. The model intercept, β₀, was −8.46 (95% CI, −9.94--6.98; p<0.001). The binary predictors most strongly associated with the risk of DPPMV were ASA Class 5 (adjusted odds ratio, 18.4; 95% CI, 2.91–116; p = 0.002) and ASA emergency status (adjusted odds ratio, 12.1; 95% CI, 9.10–16.0; p < 0.001). ASA Class 4 (adjusted odds ratio, 5.09; 95% CI, 1.16–22.4; p = 0.031), case during an existing ICU stay (adjusted odds ratio, 4.96; 95% CI, 3.72–6.62; p < 0.001), intraoperative sodium bicarbonate administration (adjusted odds ratio, 4.84; 95% CI, 3.23–7.25; p < 0.001), and presence of a fluid/electrolyte disorder (adjusted odds ratio, 4.63; 95% CI, 3.53–6.07; p < 0.001) were also found to be strongly associated with increased risk of DPPMV.

Table 2:

Multivariable logistic regression model of DPPMV risk.

Variable	aOR	95% CI	P Value
Demographics
ASA Class
ASA Class 2	1.11	[0.25, 4.98]	0.894
ASA Class 3	1.82	[0.42, 7.93]	0.423
ASA Class 4	5.09	[1.16, 22.4]	0.031 ***
ASA Class 5	18.4	[2.91, 116.]	0.002 ***
ASA E Designation	12.1	[9.10, 16.0]	<0.001 ***
Elixhauser comorbidities
Peripheral vascular disease	1.23	[0.90, 1.68]	0.186
Arrhythmia	1.63	[1.26, 2.11l	<0.001 ***
Coagulopathy	2.46	[1.84, 3.28]	<0.001 ***
Fluid/electrolyte disorder	4.63	[3.53, 6.07]	<0.001 ***
Operative Characteristics
Case during ICU stay	4.96	[3.72, 6.62]	<0.001
OR duration, min.	1.01	[1.00, 1.01]	<0.001 ***
Estimated handoffs	1.29	[1.15, 1.46]	<0.001 ***
Intraoperative Medications
Vasopressor infusion	2.09	[1.64, 2.67]	<0.001
Albuterol administration	3.35	[2.28, 4.90]	<0.001 ***
Bicarbonate administration	4.84	[3.23, 7.25]	<0.001 ***

aOR: Adjusted Odds Ratio; Cl: Confidence Interval;

^“***”denotes statistical significance.

The model was able to accurately predict DPPMV in this cohort. Bootstrapping revealed an AUC of 0.97 (95% CI, 0.96–0.97), sensitivity of 0.93 (95% CI, 0.89–0.95), and specificity of 0.89 (95% CI, 0.87–0.93). A calibration plot is shown in Figure 2. The Brier score was 0.009, and the scaled Brier score was 0.256.

Figure 2:

Calibration plot for the multivariable logistic regression model. A nonparametric fitted curve is shown alongside the ideal calibration line. The relative distribution of predicted risks is displayed along the horizontal axis.

Dropping the Elixhauser comorbidity terms only minimally affected model performance: AUC decreased to 0.95 (95% CI, 0.94–0.96), sensitivity decreased to 0.89 (95% CI, 0.85–0.95), and specificity was unchanged at 0.89 (95% CI, 0.82–0.92). The final model also performed well when the intraoperative terms (i.e. surgery duration, estimated number of intraoperative handoffs, vasopressor infusion, sodium bicarbonate administration, and albuterol administration) were dropped: The AUC decreased to 0.94 (95% CI 0.92–0.95), sensitivity decreased to 0.88 (95% CI 0.86–0.92), and specificity decreased to 0.87 (95% CI 0.84–0.87).

Airway placement and removal data were available for 96.8% of the cohort (31,866 of 32,915 cases), and availability was similar between patients who underwent DPPMV (401 of 415; 96.6%) and those who did not (31,465 of 32,500; 96.8%). Overall, 235 patients (0.74%) experienced at least one reintubation in the 24-hours following their initial extubation. The rate of reintubation was significantly higher in patients who underwent DPPMV compared to those who were extubated at the end of their case (6.5% versus 0.66%, p < 0.001 by Chi-square test).

Patients who experienced DPPMV were also more likely to experience in-hospital mortality (7.5% versus 0.3%, p < 0.001) and thirty-day mortality (9.4% versus 0.6%, p < 0.001) than extubated patients. No difference was observed between the groups in need for RRT activation within seven days of surgery (0.5% versus 0.2%, p = 0.202).

Discussion

We have determined the incidence of, and risk factors for, DPPMV in a broad surgical cohort. Although relatively uncommon, occurrence of DPPMV could be accurately predicted using a small number of readily available patient-specific and operative factors.

The incidence of DPPMV observed in this cohort (1.26%) is substantially lower than that reported amongst patients with postoperative ICU admissions (28%).⁵ It is also lower than rates observed in cardiac surgeries,^7,8 as well as certain thoracic⁹ and spine surgeries.¹⁰ This reflects the broadly inclusive criteria used to construct the cohort, which captured both small procedures and major surgeries across a wide range of surgical services.

We failed to detect a between-group difference in presence of a known difficult airway in this study, consistent with findings reported in spine surgery¹⁰ and in patients with postoperative ICU admission.⁵ Presence of a difficult airway can be an indication for DPPMV, especially when delaying extubation could reduce risk of reintubation or permit availability of more experienced providers.¹⁹ The relative subjectivity of this candidate risk factor raises the question of over- or under-reporting and use of inconsistent definitions.²⁰ Importantly, incidence of difficult airway in the DPPMV group (1.7%) was similar to a separate cohort of patients admitted to the ICU for PPMV (1.7%)⁵ and, as expected, slightly lower than in larger cohorts with otolaryngology, oral and maxillofacial surgery, and cardiac surgery cases (4.9%).²¹ Moreover, variation in detection or reporting would likely be unbiased between the two groups.

The multivariable DPPMV risk prediction model performed very well in this cohort, achieving high sensitivity and specificity. Aggressive variable selection was employed to mitigate risk of overfitting, and a favorable ratio of observations to degrees of freedom was achieved in the final model. The relatively small number of surviving variables improves suitability for clinical adoption.²² Clinical utility is further increased by the completeness and availability of model inputs, a consequence of standardized charting practices and use of structured data from the electronic health record.

A potential criticism of the model is the use of Elixhauser comorbidity data, which are frequently assessed postoperatively upon discharge from the hospital. This could theoretically limit usability of the model in the pre- and intra-operative periods. Firstly, we note that the majority of the Elixhauser comorbidities are chronic conditions and are therefore likely to be present on hospital admission if noted upon hospital discharge. Secondly, the model performs well even without Elixhauser data; estimates of AUC and sensitivity decreased by only 2–4% when the Elixhauser terms were dropped, and specificity was unchanged. We therefore demonstrate that DPPMV risk can be accurately assessed using only preoperative and intraoperative data. Moreover, we demonstrate that model performance is relatively preserved even when intraoperative terms (such as surgery duration and intraoperative medication administration) are dropped, further supporting the preoperative utility of our model.

DPPMV risk stratification has the potential to benefit patients, providers, and health care institutions. For patients, understanding the risk of an adverse outcome improves the informed consent process, and could be used to guide preoperative risk mitigation strategies.²³ For providers, risk stratification can be used to contextualize and compare postoperative outcomes. For health care institutions, advanced knowledge of patients likely to require ICU-level care could be used for operational planning and resource allocation. ICU resources are finite and expensive, making utilization optimization an important priority.¹⁴ Furthermore, delays in ICU transfer have been associated with increased hospital length of stay and mortality,²⁴ and could potentially be anticipated or even mitigated with advance notice of a probable admission. Resource scheduling is complex and multifactorial, and additional work would be required to characterize the utility, if any, of these models in improving operational efficiency and related patient outcomes. Finally, DPPMV risk prediction could be used to identify high-risk populations for inclusion in prospective studies.

There are several limitations to this study. Although generalizability of the results is limited by the single center design, our study is strengthened by its inclusion of a broad surgical cohort and the relative recency of study data. Vanderbilt is a large quaternary care academic medical center and may therefore have higher patient and case complexity than other health systems. Consequently, our estimate of DPPMV incidence may be higher than at other centers. On the other hand, the broad availability of staff trained in advanced airway management at our institution may increase our risk tolerance at time of extubation, thereby decreasing the rate of DPPMV relative to centers without these resources. Notably, however, the rate of 24-hour all-cause reintubation in the DPPMV group (6.5%) is similar to pooled rates from other ICU populations, suggesting that our single-center risk tolerance for reintubation (and consequently our risk tolerance at time of initial extubation) are congruent with other centers.²⁵ While this study does extend findings at other institutions,⁵ additional external validation is warranted, and future studies should ideally include multiple centers and incorporate surgical services that were excluded from this work (e.g. cardiac, thoracic, or trauma surgery). Overall, clinical utility of a DPPMV risk prediction model may be lower at sites where DPPMV is rare or ICU resources are abundant. Lastly, this study is limited by its retrospective design, and inability to establish causal relationships between risk factors and outcomes.

Conclusion

DPPMV was uncommon in this broad surgical cohort but could be accurately predicted using readily available patient-specific and operative factors. These results may be useful for preoperative risk stratification, postoperative resource allocation, and clinical trial planning.

KEY POINTS.

Question:

Are patient and operative factors associated with the risk of discretionary prolonged postoperative mechanical ventilation?

Findings:

In the cohort of 32,917 cases, a risk model with thirteen variables was developed to predict the risk of discretionary prolonged postoperative mechanical ventilation.

Meaning:

Accurately predicting discretionary prolonged postoperative mechanical ventilation may be useful for preoperative risk assessment and postoperative resource allocation.

Glossary of Terms

PPMV

Prolonged Postoperative Mechanical Ventilation

ICU

Intensive Care Unit

ASA

American Society of Anesthesiologists

DPPMV

Discretionary PPMV

IRB

Institutional Review Board

TRIPOD

Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis

VUMC

Vanderbilt University Medical Center

SQL

Structured Query Language

PACU

Post-Anesthesia Care Unit

RRT

Rapid Response Team