Effects of Intra-operative Cardiopulmonary Variability On Post-operative Pulmonary Complications in Major Non-cardiac Surgery: A Retrospective Cohort Study

Abstract

Intraoperative cardiopulmonary variables are well-known predictors of postoperative pulmonary complications (PPC), traditionally quantified by median values over the duration of surgery. However, it is unknown whether cardiopulmonary instability, or wider intra-operative variability of the same metrics, is distinctly associated with PPC risk and severity. We leveraged a retrospective cohort of adults (n = 1202) undergoing major non-cardiothoracic surgery. We used multivariable logistic regression to evaluate the association of two outcomes (1)moderate-or-severe PPC and (2)any PPC with two sets of exposure variables- (a)variability of cardiopulmonary metrics (inter-quartile range, IQR) and (b)median intraoperative cardiopulmonary metrics. We compared predictive ability (receiver operating curve analysis, ROC) and parsimony (information criteria) of three models evaluating different aspects of the intra-operative cardiopulmonary metrics: Median-based: Median cardiopulmonary metrics alone, Variability-based: IQR of cardiopulmonary metrics alone, and Combined: Medians and IQR. Models controlled for peri-operative/surgical factors, demographics, and comorbidities. PPC occurred in 400(33%) of patients, and 91(8%) experienced moderate-or-severe PPC. Variability in multiple intra-operative cardiopulmonary metrics was independently associated with risk of moderate-or-severe, but not any, PPC. For moderate-or-severe PPC, the best-fit predictive model was the Variability-based model by both information criteria and ROC analysis (area under the curve, AUC_{Variability-based} = 0.74 vs AUC_Median-based = 0.65, p = 0.0015; AUC_{Variability-based} = 0.74 vs AUC_Combined = 0.68, p = 0.012). For any PPC, the Median-based model yielded the best fit by information criteria. Predictive accuracy was marginally but not significantly higher for the Combined model (AUC_Combined = 0.661) than for the Median-based (AUC_Median-based = 0.657, p = 0.60) or Variability-based (AUC_{Variability-based} = 0.649, p = 0.29) models. Variability of cardiopulmonary metrics, distinct from median intra-operative values, is an important predictor of moderate-or-severe PPC.

Introduction

Postoperative pulmonary complications (PPC) are a key quality metric for intraoperative care [1], with an incidence of up to 33% [1, 2] and an associated 40-fold increase in 30-day post-operative mortality [3]. Clinical trials aiming to decrease PPC in general surgical patients [4] have led to the widespread adoption of lung-protective intra-operative ventilation strategies [5–7].

Extensive prior work has established the relationship between median values of intraoperative respiratory variables and PPC [8–10]. However, few studies explored the effects of intraoperative variability of cardiopulmonary variables as predictors or potentially modifiable risk factors for PPC. Data on the association between intraoperative cardiopulmonary variability and severity of PPC are also scarce.

Studies on cardiopulmonary variability to date have been focused on intraoperative hemodynamic variability as an exposure that affects postoperative cardiovascular [11], renal [12], and pulmonary [13] outcomes, as well as mortality [14, 15]. By contrast, the relationship between intraoperative respiratory variability and postoperative pulmonary outcomes remains poorly studied. Yet, such a relationship is biologically plausible and clinically meaningful given the strong association between intraoperative mechanical ventilation and PPC [1, 5]. Evidence of an association between respiratory variability and postoperative pulmonary outcomes could inform clinical decision-making and guide interventions to reduce such variability [16], particularly with the future adoption of real-time decision-support tools or closed-loop systems that allow for tighter control of intraoperative hemodynamic or respiratory variability.

We hypothesized that the variability of intraoperative cardiopulmonary variables, distinct from median values, is independently associated with the frequency and severity of PPC. To test our hypothesis, we analyzed data from a well-characterized prospective cohort of patients undergoing non-cardiothoracic surgery. Our aims were (1) to infer individual risk factors for both any PPC and moderate and severe PPC, and (2) to compare predictive models of each clinical outcome based on either the median values of intraoperative cardiopulmonary variables, variability in intraoperative cardiopulmonary variables, or both.

Methods

Ethics

Ethical approval was not required for this secondary retrospective analysis of an IRB-approved prospective multicenter study cohort [2].

Study population

We analyzed 1202 adult major non-cardiothoracic surgery patients in seven U.S. institutions (May—November 2014) [2]. Inclusion criteria were: age ≥ 18 years; ASA Physical Status III; elective or emergency non-cardiac, non-aortic surgery; general anesthesia with endotracheal intubation; and expected duration of surgery ≥ 2 h. Exclusion criteria were: preoperative ventilatory support or oxygen dependency; tracheostomy; pregnancy; life expectancy ≤ 30 days as estimated based on the clinician’s notes in the patient’s medical record.

Outcomes

We examined two co-primary, prospectively collected outcomes: (1) moderate or severe PPC and (2) any PPC during the first seven days post-operatively. Moderate or severe PPC included pneumonia, pneumothorax, ARDS, and re-intubation, while any PPC also included mild complications: prolonged O₂ requirement (> 1 day of post-operative nasal cannula or facemask), post-operative atelectasis, pleural effusion, and bronchospasm as previously defined [3]. Mechanical ventilation was considered a PPC if it was required for respiratory failure. The components of our primary outcome were based on determinants of postoperative pulmonary complications [3, 17].

Data collection

We conducted a secondary retrospective analysis ofrdir data from a large prospective study. The primary data collection has previously been described in detail [2]. Three broad categories of data were available: baseline patient demographics and comorbidities, surgical factors and peri-operative variables (e.g. pre-operative hemoglobin, pre-operative SpO_2, type and duration of surgery), and intra-operative cardiopulmonary variables [2]. Surgical/peri-operative variables were recorded as a composite value over the case duration (e.g. total desaturation time (defined at an oxygen saturation threshold less than 90%), estimated blood loss, urine output, Table 1). Intra-operative cardiopulmonary variables included mean arterial pressure (MAP), end-tidal carbon dioxide content (EtCO₂), oxygen saturation by pulse oximetry (SpO₂), respiratory rate, driving pressure, heart rate, and temperature. Cardiopulmonary variables were recorded as the intraoperative 1st (Q1), 2nd (median,Q2) and 3rd (Q3) quartiles over the duration of surgery, and the inter-quartile range (IQR) of each metric was calculated as a measure of variability. In this study, driving pressure was approximated as the difference between peak airway pressure and positive end-expiratory pressure (PEEP).

Table 1.

Baseline patient demographic characteristics

	All	Mild PPC	Moderate or severe PPC	P value
	N = 1202	N = 309	N = 91
Demographic characteristics
Male gender	0.53 (0.50)	0.44 (0.50)	0.66 (0.48)	< 0.001
Age	62.1 (13.8)	64.3 (12.9)	67.1 (10.5)	< 0.001
BMI	30.0 (7.53)	30.5 (7.94)	29.4 (7.77)	0.341
History of CVD	103 (8.57%)	27 (8.74%)	6 (6.59%)	0.783
History of HTN	790 (65.7%)	216 (69.9%)	68 (74.7%)	0.017
History of CAD	246 (20.5%)	68 (22.0%)	22 (24.2%)	0.601
Missing data	1 (0.08%)	0 (0.00%)	0 (0.00%)
History of CHF	68 (5.66%)	23 (7.44%)	8 (8.79%)	0.136
Missing	1 (0.08%)	0 (0.00%)	0 (0.00%)
History of valvular heart disease	72 (5.99%)	21 (6.80%)	6 (6.59%)	0.784
Missing	1 (0.08%)	0 (0.00%)	0 (0.00%)
History of COPD	103 (8.57%)	34 (11.0%)	17 (18.7%)	< 0.001
Missing	1 (0.08%)	1 (0.32%)	0 (0.00%)
History of asthma	172 (14.3%)	39 (12.6%)	13 (14.3%)	0.607
Current smoking	161 (13.4%)	44 (14.2%)	15 (16.5%)	0.665
Missing	2 (0.17%)	0 (0.00%)	0 (0.00%)
Former smoking	478 (39.8%)	139 (45.0%)	37 (40.7%)
Missing	64 (5.32%)	14 (4.53%)	4 (4.40%)
History of cancer	499 (41.5%)	137 (44.3%)	49 (53.8%)	0.013
History of GERD	431 (35.9%)	120 (38.8%)	42 (46.2%)	0.052
Missing	2 (0.17%)	1 (0.32%)	0 (0.00%)
History of CKD	245 (20.4%)	64 (20.7%)	26 (28.6%)	0.115
History of liver disease	146 (12.1%)	37 (12.0%)	20 (22.0%)	0.044
Missing	2 (0.17%)	0 (0.00%)	0 (0.00%)
History of diabetes	301 (25.0%)	86 (27.8%)	24 (26.4%)	0.366
History of chronic alcohol use	92 (7.65%)	24 (7.77%)	11 (12.1%)	0.382
Missing	1 (0.08%)	0 (0.00%)	0 (0.00%)
History of anemia	124 (10.3%)	41 (13.3%)	9 (9.89%)	0.138

BMI body mass index, CVD cerebrovascular disease, HTN hypertension; CAD coronary artery disease; CHF chronic heart failure; COPD chronic obstructive pulmonary disease; GERD gastro-esophageal reflux disease; CKD chronic kidney disease

Statistical analysis and model inference

Patient characteristics and intra-operative variables were summarized as the mean and standard deviation among the study population (continuous variables) or as percentages (categorical variables, Table 1). Missing data were imputed using Multivariate Imputation via Chained Equations. Additional details about the data are provided in the Supplementary Material (section S1).

We aimed to understand the distinct effect of variability in cardiopulmonary metrics, in addition to median values, on PPC risk. We therefore used multivariable logistic regression models to evaluate the association of our two outcomes: (1) moderate or severe PPC and (2) any PPC - with median intraoperative cardiopulmonary metrics and variability of cardiopulmonary metrics (the interquartile range). We first inferred the association of each outcome with individual demographic, peri-operative, and intra-operative cardiopulmonary risk factors, using the complete set of available data for each patient (Table 3). To limit multi-collinearity, a representative from pairs of interdependent variables (e.g. driving pressure and peak pressure) was heuristically chosen for inclusion in the regression model, and variance inflation factors were calculated for each predictor variable (Supplemental Material, section S1, Table S2, Fig. S3). The full set of predictor variables used in the logistic regression models is provided in Table 4. Adjusted odds ratios (OR) and 95% confidence intervals (95% CI) were reported from the multivariate logistic regression models, with Bonferroni correction for multiple comparisons. Significant independent risk factors for each outcome were defined as variables with adjusted odds ratios and 95% confidence intervals greater than 1.0, while protective factors were defined as variables with adjusted odds ratios and 95% confidence intervals less than 1.0.

Table 3.

Postoperative pulmonary complications according to outcome severity

Complications	Mild PPC (N = 309)	Mod-Sev PPC (N = 91)	Any Complication (N = 400)
Atelectasis	154 (49.8%)	51 (56.0%)	205 (51.2%)
Prolonged O2 assistance (> 1d)
By nasal cannula	207 (67.0%)	28 (30.8%)	235 (58.8%)
By face mask	5 (16.2%)	7 (7.7%)	12 (3.0%)
Bronchospasm	10 (3.2%)	3 (3.3%)	13 (3.3%)
PO noninvasive ventilation	0 (0.0%)	46 (50.5%)	46 (11.5%)
Pneumonia	0 (0.0%)	22 (24.2%)	22 (5.5%)
Pneumothorax	0 (0.0%)	4 (4.4%)	4 (1.0%)
Pleural effusion	75 (24.3%)	41 (45.1%)	116 (28.9%)
PO mechanical ventilation	0 (0.0%)	33 (36.3%)	33 (8.3%)
ARDS	0 (0.0%)	2 (2.2%)	2 (0.5%)

PO post-operative, ARDS Acute Respiratory Distress Syndrome

Table 4.

Model coefficients (adjusted odds ratio and 95% confidence interval) for multivariable logistic regression models of the two study outcomes - any PPC and moderate or severe PPC

Variable	Odds ratio Any PPC	[95% CI] Moderate-Severe PPC
Demographic characteristics
Age	1.53 [1.28, 1.83]	1.83 [1.30, 2.59]
Male gender	0.59 [0.43, 0.81]	1.46 [0.84, 2.59]
BMI	1.18 [1.00, 1.41]	0.87 [0.62, 1.20]
History of anemia	0.86 [0.48, 1.53]	0.71 [0.24, 1.96]
Duration of anesthesia	1.50 [1.28, 1.76]	1.41 [1.12, 1.78]
History of asthma	0.88 [0.58, 1.33]	1.43 [0.67, 2.88]
History of CAD	0.99 [0.68, 1.42]	0.84 [0.44, 1.55]
History of cancer	1.08 [0.81, 1.44]	1.62 [0.97, 2.71]
History of COPD	1.61 [0.99, 2.62]	2.67 [1.28, 5.38]
History of CVD	1.09 [0.65, 1.80]	0.65 [0.21, 1.66]
History of diabetes	1.04 [0.75, 1.44]	0.90 [0.49, 1.59]
History of GERD	1.25 [0.93, 1.67]	1.50 [0.90, 2.49]
History of CHF	2.04 [1.13, 3.67]	1.87 [0.68, 4.62]
History of valvular disease	0.96 [0.53, 1.71]	0.65 [0.21, 1.66]
History of HTN	1.05 [0.75, 1.47]	1.37 [0.75, 2.57]
History alcohol use	1.20 [0.70, 2.05]	1.53 [0.64, 3.37]
History of liver disease	1.04 [0.68, 1.60]	1.59 [0.80, 3.03]
History of CKD	1.18 [0.83, 1.67]	1.37 [0.76, 2.44]
Current smoker	1.34 [0.87, 2.03]	1.25 [0.59, 2.51]
Former smoker	1.16 [0.86, 1.55]	0.67 [0.40, 1.12]
Perioperative and surgical factors
Emergency surgery	3.10 [1.62, 6.00]	4.14 [1.75, 9.36]
Intra-abdominal surgery	2.63 [1.95, 3.57]	1.44 [0.83, 2.52]
EBL per kg IBW	1.18 [1.01, 1.42]	1.31 [1.09, 1.60]
Minutes of desaturation	1.08 [0.93, 1.27]	1.08 [0.90, 1.29]
Pre-operative hemaglobin	0.85 [0.70, 1.03]	0.98 [0.70, 1.37]
Pre-operative SpO2	0.81 [0.70, 0.94]	1.08 [0.84, 1.41]
Urine output (mL/kg/h)	0.97 [0.82, 1.13]	1.02 [0.76, 1.30]
Vasopressin use	1.83 [0.72, 4.79]	1.10 [0.25, 3.91]
Ventillator_mode: PC	1.57 [0.99, 2.50]	0.74 [0.33, 1.62]
Vent_mode: VC	1.23 [0.85, 1.79]	0.54 [0.29, 1.03]
MED opioids per kg IBW	1.09 [0.94, 1.27]	1.21 [0.96, 1.51]
Ketamine dosage	1.19 [0.99, 1.48]	1.13 [0.95, 1.33]
Cardiopulmonary variables
IQR driving pressure IQR	1.02 [0.87, 1.19]	0.68 [0.48, 0.93]
Median driving pressure	0.87 [0.73, 0.99]	1.35 [1.03, 1.79]
ETCO2_IQR	0.98 [0.82, 1.16]	0.73 [0.49, 1.01]
Median ETCO2	1.08 [0.93, 1.26]	0.91 [0.70, 1.19]
FiO2_IQR	1.04 [0.89, 1.20]	1.10 [0.84, 1.41]
Median FiO2	1.16 [1.01, 1.34]	1.10 [0.85, 1.41]
IQR of HR	1.06 [0.90, 1.25]	1.10 [1.03, 1.24]
Median HR	1.21 [1.03, 1.42]	1.12 [0.85, 1.47]
MAP IQR	1.06 [0.92, 1.23]	1.07 [0.75, 1.30]
Median MAP	0.89 [0.77, 1.04]	0.88 [0.67, 1.15]
Respiratory rate IQR	0.94 [0.79, 1.11]	1.38 [1.05, 1.79]
Median respiratory rate	0.93 [0.79, 1.09]	1.07 [0.80, 1.42]
SpO2 IQR	1.10 [0.93, 1.31]	1.22 [1.04, 1.60]
Median SpO2	1.02 [0.86, 1.22]	0.82 [0.62, 1.09]
IQR temperature	1.12 [0.97, 1.30]	0.93 [0.73, 1.18]
Median temperature	0.90 [0.77, 1.06]	0.95 [0.71, 1.27]

BMI body mass index, CVD cerebrovascular disease, HTN hypertension, CAD coronary artery disease, CHF chronic heart failure, COPD chronic obstructive pulmonary disease, GERD gastro-esophageal reflux disease, CKD chronic kidney disease, IBW adjusted body weight, MED morphine equivalent dosage, PC pressure control, VC volume control, hb hemoglobin, IQR interquartile range, HR heart rate, SpO2 peripheral oxygen saturation, RR respiratory rate, FiO2 inspired oxygen nitrationation, ETCO2 end- tidal concentration of carbon dioxide

Next, we generated predictive models to evaluate the relative contributions of median values and variability (IQR) of intraoperative cardiopulmonary metrics to PPC risk. For each outcome (any vs. moderate or severe PPC) we compared three multivariable logistic regression models including different aspects of the intra-operative cardiopulmonary dynamics:

Median-based model: median values of cardiopulmonary variables.

Variability-based model: IQR cardiopulmonary variables.

Combined model: median and IQR of cardiopulmonary variables.

All three models included the full set of demographic, peri-operative, and surgical factors and baseline comorbidities as predictor variables (Table 2). Models were compared using measures of predictive ability (receiver operating curve analysis) and model parsimony (Akaike Information Criteria [AIC] and Bayesian Information Criteria [BIC]). For ROC analysis, AUC values were generated by bootstrapping. All statistical analysis were conducted using the R statistical programming environment (version 4.0.2).

Table 2.

Baseline perioperative/surgical factors and intra-operative cardiopulmonary variables

	All	Mild PPC	Moderate or severe PPC	P value
	N = 1202	N = 309	N = 91
Perioperative and surgical factors
Desaturation time (min)	0.61 (2.89)	0.67 (2.55)	1.25 (4.33)	0.061
Required blood transfusion	0.11 (0.31)	0.15 (0.36)	0.26 (0.44)	< 0.001
Required vasopressin	0.02 (0.15)	0.04 (0.19)	0.04 (0.21)	0.014
Emergency surgery	61 (5.07%)	22 (7.12%)	12 (13.2%)	< 0.001
Missing	2 (0.17%)	1 (0.32%)	0 (0.00%)
Volume crystalloid per kg IBW (L/kg)	0.10 (0.07)	0.11 (0.07)	0.11 (0.08)	0.003
Volume colloid per kg IBW (mL/kg)	0.00 (0.01)	0.00 (0.01)	0.00 (0.01)	0.714
MED opioids (mg/kg IBW)	0.46 (0.30)	0.50 (0.34)	0.54 (0.43)	< 0.001
Duration anesthesia (h)	4.64 (2.22)	5.05 (2.35)	5.96 (3.22)	< 0.001
Vent mode				0.002
Other	327 (27.2%)	75 (24.3%)	32 (35.2%)
PC	178 (14.8%)	64 (20.7%)	16 (17.6%)
VC	697 (58.0%)	170 (55.0%)	43 (47.3%)
Urine output (ml/kg/h)	1.24 (1.54)	1.14 (1.23)	1.22 (1.41)	0.449
Preoperative hb (mg/dL)	12.4 (2.07)	12.1 (2.00)	12.3 (2.30)	0.034
Preoperative SpO2	97.2 (2.13)	96.8 (2.33)	96.9 (2.39)	< 0.001
EBL per kg IBW (mL/kg)	5.45 (10.4)	6.52 (8.59)	11.5 (22.3)	< 0.001
Intra-operative cardiopulmonary variables
Median ETCO2	35.1 (3.13)	35.3 (2.68)	35.0 (3.53)	0.548
Median FiO2	54.4 (13.9)	56.3 (14.9)	55.7 (14.4)	0.009
Median HR (bpm)	69.8 (12.3)	71.6 (12.7)	72.4 (12.3)	< 0.001
Median MAP (mm Hg)	82.6 (10.5)	82.5 (9.92)	81.3 (10.2)	0.443
Median RR	12.0 (2.19)	11.8 (2.13)	12.3 (2.69)	0.057
Median SpO2	98.8 (1.36)	98.8 (1.30)	98.5 (1.71)	0.055
Median temperature	36.6 (1.25)	36.6 (1.11)	36.6 (1.13)	0.757
Median driving pressure (mm Hg)	16.2 (5.11)	15.9 (4.92)	17.0 (5.85)	0.173
MAP IQR	14.9 (10.1)	15.7 (16.3)	15.4 (6.22)	0.237
ETCO2 IQR	3.61 (2.69)	3.60 (2.43)	3.55 (2.01)	0.958
FiO2 IQR	9.67 (11.8)	9.93 (11.2)	11.6 (14.4)	0.234
SpO2 IQR	0.95 (1.04)	0.98 (1.03)	1.38 (1.18)	< 0.001
HR IQR	9.05 (5.76)	9.69 (7.08)	10.2 (5.67)	0.005
Driving pressure IQR	2.79 (3.12)	2.97 (3.06)	2.52 (2.64)	0.406
Respiratory rate IQR	1.69 (1.66)	1.59 (1.42)	2.10 (2.02)	0.038
Temperature IQR	0.47 (0.37)	0.51 (0.41)	0.53 (0.44)	0.021

IBW adjusted body weight, MED morphine equivalent dosage, PC pressure control, VC volume control, hb hemoglobin, IQR interquartile range, HR heart rate, SpO2 peripheral oxygen saturation, RR respiratory rate, FiO2 inspired oxygen nitrationation, ETCO2 end-tidal concentration of carbon dioxide

Results

Study population and outcomes

1202 patients were included in the analysis. The participants’ average age was 62 + −14 years and their average BMI 30.5 + −8 kg/m², with high rates of pre-existing hypertension and previous history of malignancy (Table 1). Post-operative pulmonary complications (PPC) occurred in 400 patients, with 91 patients experiencing moderate or severe outcomes (Table 3). Among patients with mild PPC (n = 309), 67.0% (n = 207) required supplemental oxygen via nasal cannula for longer than one day post-operatively, and 49.8% (n = 154) experienced post-operative atelectasis (Table 3). Among patients with moderate or severe PPC, the most common outcomes were post-operative non-invasive ventilation (n = 46, 50.5%), post-operative mechanical ventilation (n = 33, 36.3%), and post-operative pneumonia (n = 22, 24.2%).

Cardiopulmonary variability and outcome severity

We observed higher variability of multiple intra-operative cardiopulmonary variables among patients with moderate or severe PPC when compared to patients with any PPC (Table 2). For example, (1) the median of the heart-rate variability (defined as the IQR of intraoperative heart rate) was 5.67 +− 10.2 for moderate or severe PPC vs. 5.76 +− 9.05 for any PPC and 7.08 +− 9.69 for all patients included in the study (p = 0.005), (2) the median of respiratory rate variability was 2.02 +− 2.10 for moderate or severe PPC vs. 1.42 +− 1.59 for any PPC and 1.66 +− 1.69 for all patients (p = 0.04), (3) the median variability in SpO2 1.18 +− 1.38 for moderate or severe PPC vs. 0.98 +− 1.03 for any PPC and 0.95 +− 1.04 for all study patients (p < 0.001). Variability in other cardiopulmonary metrics, including MAP, ETCO2, and driving pressure, was similar across study outcomes (Table 2).

Multivariable inferential modelling

First, we used comprehensive multivariable regression models to infer individual risk factors for each of our clinical outcomes – any and moderate or severe PPC – from the available set of patient characteristics, perioperative and surgical factors, and intra-operative cardiopulmonary variables (Table 4).

Patient characteristics, perioperative and surgical factors

We inferred individual risk factors for each of our two outcomes—any and moderate or severe PPC – from the full available set of demographic data, perioperative data, and intraoperative cardiopulmonary data. The distinct risk factors that we identified differed according to outcome severity (Table 4, Fig. 1). Notably, male gender and history of CHF were associated with increased risk of any PPC, while prior history of COPD significantly increased the risk of moderate or severe PPC. Among peri-operative and surgical factors, intra-abdominal surgery and decreased pre-operative SpO₂ were significant risk factors for any PPC, while increasing estimated blood loss was a significant risk factor for moderate or severe PPC.

Fig. 1.

Association between intraoperative exposure variables and post-operative pulmonary outcomes. Abbreviations: PPC, postoperative pulmonary complications; M-S, moderate or severe; , direct relationship; , inverse relationship; -, no significant relationship

Intra-operative variables: Median values

Median values of intra-operative cardiopulmonary variables were associated with PPC risk in multivariate regression analysis of both study outcomes (Fig. 1, Table 4). For the outcome of moderate or severe PPC, median driving pressure (OR 1.35, 95% CI 1.03 to 1.79) was a significant respiratory risk factor. For the outcome of any PPC, median heart rate (OR 1.21, 95% CI 1.03 to 1.42) was a significant hemodynamic risk factor, while median driving pressure yielded a modestly significant decrease in risk of any PPC (OR 0.87, 95% CI 0.73 to 0.99). Of note, this effect on the composite outcome was driven entirely by prolonged use of nasal cannula (see further explanation in the Supplemental Material, section S2).

Intra-operative variables: Variability

The interquartile ranges of multiple cardiopulmonary variables, our measure of intra-operative variability, were significantly associated with risk of moderate or severe PPC in multivariable regression analyses. Increasing variability (IQR) of intra-operative SpO2 (OR 1.22, 95% CI 1.04 to 1.60), respiratory rate (OR 1.38, 95% CI 1.05 to 1.79), and heart rate (OR 1.10, 95% CI 1.03 to 1.24]), were significant independent risk factors for moderate or severe PPC (Table 4, Fig. 1). Variability in driving pressure was associated with a modest but significant decrease in risk of moderate or severe PPC (OR 0.68, 95% CI 0.48 to 0.93). Conversely, this association did not hold for the outcome of any PPC (Fig. 1, Table 4).

Multivariable predictive modelling and model comparison

Next, we compared three distinct predictive models to evaluate the relative contributions of median values and variability (IQR) of intraoperative cardiopulmonary metrics on the risk of our two study outcomes—any and moderate-or-severe PPC. To mitigate confounding, every model adjusted for the full set of demographic, peri-operative, and surgical factors and baseline comorbidities (Methods, Tables 1 and 2). For each outcome, we compared three multivariable logistic regression models, evaluating different aspects of the intraoperative cardiopulmonary metrics:

(Median-based model): Median-only (median values of cardiopulmonary variables).

(Variability-based Model): IQR-only (IQR of cardiopulmonary variables).

(Combined model): Median and IQR (both median values and IQR of cardiopulmonary variables).

Notably, the best-fit model for the outcome of moderate-or-severe PPC included only variability (IQR) in cardiopulmonary variables (Variability-based model). This finding was present according to both information criteria (AIC and BIC, Supplementary Material Table S4) and ROC analysis (Fig. 2B). Indeed, predictive ability for moderate-or-severe PPC was significantly higher for the Variability-based model (area under the curve, AUC = 0.75, Fig. 2B) than for the Median-based model (AUC_{variability-based} = 0.73 with 95% CI 0.70 to 0.77,.vs AUC_median-based = 0.66, with 95% CI 0.63 to 0.68, p = 0.0007 by bootstrapping) or for the combined model (AUC_{variability-based} = 0.73 vs AUC_combined = 0.68 with 95% CI 0.64 to 0.70, p = 0.039 by bootstrapping).

Fig. 2.

Comparison of predictive ability for models of ‘any’ (A) and ‘severe’ (B) PPC by receiver operating curve (ROC) analysis. AUC: area under the curve. M1: model including only medians of intra-operative cardiopulmonary variables. M2: model including only IQRs of intra-operative cardiopulmonary variables. M3: model including both medians and IQRs of intra-operative cardiopulmonary variables

Conversely, for the outcome variable of any PPC, the best-fit (most parsimonious) model by both AIC and BIC was the Median-based model (Supplementary Material Table S4). Predictive accuracy was marginally higher for the Median-based model (AUC_median-based = 0.702 with 95% CI 0.67 to 0.73, Fig. 2A) than for the Variability-based model (AUC_{variability-based} = 0.694 with 95% CI 0.66 to 0.72) or the combined model (AUC_comhined = 0.699 with 95% CI 0.67 to 0.72), but the difference was not statistically significant (p = 0.62 and p = 0.46, respectively).

Discussion

In this large, multicenter cohort of major noncardiac surgery patients, we identified distinct PPC risk factors among demographic, peri-operative, and intra-operative cardiopulmonary variables. These risk factors differed according to outcome severity. Furthermore, we found that variability in intra-operative cardiopulmonary variables, distinct from median values, is a significant predictor of moderate-or-severe PPC. Overall, our results suggest that targeting cardiopulmonary stability over the course of surgery, distinct from targeting absolute threshold goals in hemodynamic and respiratory parameters, may be key to mitigating the risk of moderate-or-severe PPC.

Anesthesiologists strive to minimize intraoperative cardiopulmonary variability in their daily practice, understanding the physiologic basis for the relationship between acute hemodynamic fluctuations and adverse outcomes such as stroke or renal injury. Yet, in the absence of closed-loop systems, our results provide quantitative support for the association between postoperative pulmonary risk and intraoperative cardiopulmonary variability, further supporting clinical efforts to minimize cardiopulmonary fluctuations around a patient’s baseline.

Our findings suggest that intra-operative cardiopulmonary variability is a major determinant of PPC severity.

In comprehensive multivariable regression models, variability in multiple intraoperative cardiopulmonary variables was independently associated with the risk of moderate-or-severe PPC (Table 4). Furthermore, comparison of models including cardiopulmonary variable medians, variability (IQR), or both together demonstrated that for moderate-or-severe PPC, the model including only variability was the most parsimonious and had the highest predictive accuracy. Importantly, all models controlled for a wide range of potentially confounding demographic, peri-operative, and surgical variables (Table 1). Prior studies have also identified the effects of intraoperative hemodynamic variability primarily on severe postoperative outcomes such as stroke, acute kidney injury, and death [12]. Thus, our observations may reflect the fact that moderate-or-severe PPC are more likely to represent true deviations from a normal postoperative course and serve as a more accurate measure of an adverse outcome. By contrast, some components of the composite outcome of any PPC, such as supplemental oxygen or pleural effusion may reflect local care practices [18] or expected postoperative course [19] and thus are not necessarily significant complications.

This study characterizes the previously unknown relationship between intraoperative respiratory variability and pulmonary outcomes.

Prior publications focused exclusively on hemodynamic rather than respiratory variability as an intraoperative exposure and assessed predominantly cardiovascular outcomes or mortality [11, 14, 15]. The only study to date exploring the effects of intraoperative variability on postoperative pulmonary outcomes demonstrated a relationship between intraoperative blood pressure variability and failure to extubate [13]. Our results expand upon these earlier observations to identify the effects of variability in multiple individual hemodynamic and respiratory metrics on PPC risk stratified by outcome severity.

In a multivariable analysis of the risk of moderate-or-severe PPC, variability in intraoperative SpO2, respiratory rate, and heart rate demonstrated the largest effects.

Such findings are biologically plausible since major deviations of these variables from normal could indicate clinical deterioration and/or directly result in organ damage. As a marker of clinical derangements, variable SpO2 suggests varying or periodically compromised intraoperative pulmonary gas exchange. Our results indicate that even if median values are acceptable, the presence of such intraoperative variations by themselves heralds significantly worse respiratory outcomes. Increased variability of respiratory rate may indicate ventilator dys-synchrony or mechanical ventilation adjustments made to compensate for pulmonary dysfunction. Moreover, higher variability of respiratory rate may directly lead to lung injury through hyperinflation and increased lung strain [20]. As a component of increased hemodynamic variability, tachycardia may suggest hypovolemia/hypotension, arrhythmias, administration of sympathomimetic agents, but could also lead to direct myocardial damage, especially with pre-existing coronary artery disease. Intraoperative bradycardia may also indicate arrhythmias, but more likely baroreceptor reflex from treatment of hypotension with phenylephrine or administration of beta-blockers for hypertension or tachyarrhythmia. The above scenarios, although cardiac in nature, may lead to transient or persistent left heart strain or overload and manifest as pulmonary complications mediated by pulmonary congestion.

Contrary to our expectations, variability in driving pressure independently decreased the risk of severe PPC in our model.

The mechanisms for such an effect are unclear. We found no significant correlation between median driving pressure and driving pressure IQR (Supplemental Material Fig. S3). One speculative explanation is that higher driving pressure variability is a marker of pulmonary recruitment manoeuvres that decrease the risk of PPC.

Given that intraoperative cardiopulmonary variability in this study and hemodynamic variability in prior studies [21] have been associated with, and may plausibly cause, adverse outcomes, they may be viewed as targets for clinical intervention.

Anesthesia providers typically minimize intraoperative fluctuations of hemodynamic and respiratory metrics. However, competing clinical needs of the patient and cognitive load on the provider make it challenging or even impossible for a clinician to minimize hemodynamic and respiratory variability. By contrast, automatization of hemodynamic and respiratory management based on real-time data is a promising aid to support the clinicians’ efforts aimed at minimizing intraoperative cardiopulmonary variability. Closed loop hemodynamic management systems have demonstrated clinically meaningful reduction in intraoperative fluctuations of blood pressure [13, 22]. A combination of such systems with closed loop respiratory management [16] has the potential to address both cardiac and pulmonary fluctuations that, based on our findings, are associated with increased risk of moderate-or-severe PPC. Our study opens a new avenue for clinical investigation of interventions that target intraoperative hemodynamic and respiratory variability to reduce postoperative pulmonary complications. If our results are corroborated by other studies, intra-operative hemodynamic and respiratory variability could be useful as quality or intraoperative real-time metrics. However further study is needed.

The results of this study should be interpreted in the context of its limitations. Our findings are primarily applicable to major abdominal surgery patients. We did not include several other subgroups of patient at high risk for pulmonary complications such as cardiac, thoracic and aortic surgery patients – populations that are likely to be particularly vulnerable to the potential adverse effects of intraoperative cardiopulmonary variability. Our study was not designed to address causality or the extent to which the identified predictive hemodynamic and respiratory variability is modifiable. While our multivariable models control for a wide range of demographic, perioperative, and intra-operative factors, we cannot ignore the possibility of residual confounding from unobserved covariables. We only had access to summary measures of intraoperative variables, such as medians, first and third quartiles, rather than true time series data. We were therefore unable to assess other measures of variability. Finally, our study was not designed to evaluate the relationship between pre-operative cardiopulmonary instability on intra-operative or post-operative outcomes, but this could represent an important area of future investigation. Future studies are needed to understand whether patients who exhibit pre-operative fluctuations around baseline hemodynamic and respiratory parameters are prone to more severe intra-operative and post-operative pulmonary complications.

Conclusions

We have demonstrated that the variability of intraoperative SpO2, respiratory rate and heart rate predict the development of significant (moderate or severe) postoperative pulmonary complications with higher reliability than their median values. Our hypothesis-generating findings suggest that intraoperative fluctuations of hemodynamic and respiratory variables may serve as targets for clinical interventions and measures of quality of anesthesia care.

Data Availability

Data will be made available by reasonable request to the corresponding author.