Perioperative Mortality of the COVID-19 Recovered Patient Compared to a Matched Control: A Multicenter Retrospective Cohort Study

This was a retrospective cohort study utilizing propensity-matched observational data collected as part of an established research consortium, the Multicenter Perioperative Outcomes Group (Ann Arbor, Michigan). Institutional review board approval was obtained by all participating sites to contribute deidentified data to the Multicenter Perioperative Outcomes Group with a waiver of written informed consent, and the primary study center (Oregon Health & Science University, Portland, Oregon) obtained institutional review board approval to study this limited dataset. Our report adhered to the REporting of studies Conducted using Observational Routinely-collected Data (RECORD) statement extension of the Strengthening of the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for reporting of observational studies as well as extension of STROBE for reporting of propensity-matched studies.^16,17  Before the data were accessed, the study proposal including the data analysis and statistical plan was presented to the Multicenter Perioperative Outcomes Group perioperative clinical research committee, and underwent revision and final approval with study registration and posting on May 10, 2021.¹⁸  After study approval and registration, data were extracted from the Multicenter Perioperative Outcomes Group dataset. Methods for local electronic health record data acquisition, validation, mapping to semantically interoperable universal Multicenter Perioperative Outcomes Group concepts, and secure transfer to the coordinating center have been previously described.¹⁹ 

The study period for the exposure cohort was from April 2020 until April 2021 and included Multicenter Perioperative Outcomes Group participating centers that had (1) reported outcome and exposure variables without gaps in the study period; (2) submitted laboratory findings with mapped variables to the COVID-19 polymerase chain reaction, nucleic acid, or antigen tests; (3) contributed more than 20 exposure cases to the dataset; and (4) reported mortality outcomes.

We included patients age greater than 18 yr with at least one laboratory COVID-19 test (positive or negative result) undergoing inpatient surgical procedures. We excluded cases with American Society of Anesthesiologists (ASA; Schaumburg, Illinois) emergency classification, ASA Physical Status V or VI, ambulatory procedures, vascular surgery, liver transplant surgery, lung transplant surgery, nonoperative procedures, magnetic resonance imaging, cardiac surgery, and cases started between the hours of 8:00 pm and 6:00 am. These exclusions were aimed at targeting an elective inpatient population undergoing surgery not directly attributable to sequelae from a potential COVID-19 diagnosis.

The primary analysis compared all outcomes to patients in the same time period who had negative COVID-19 tests before surgery. A preplanned sensitivity analysis was performed, comparing COVID-19–exposed patients to matched patients having surgery during the year before the pandemic (April 2019 to March 2020).

The exposure cohort time period occurred during the first identified COVID-19 strains as well as the alpha variant. The delta variant did not appear in the United States during this study period. By the end of the observation period (April 2021), many Americans had received some COVID-19 vaccination.

The primary exposure variable studied was COVID-19 diagnosis ascertained by preoperative laboratory polymerase chain reaction, nucleic acid, or antigen test any time before the surgical procedure. “Detected” or “positive” was considered positive while “not detected” or “negative” was considered negative. Cases with positive tests noted only after the surgical procedure were excluded.

The primary outcome was all-cause 30-day mortality defined using a combination of recorded date of procedure and date of mortality. The secondary outcome of AKI was defined by the Kidney Disease: Improving Global Outcomes guidelines as serum creatinine of at least 0.4 mg/dl within 48 h after surgery or an increase of at least 50% from baseline within a 7-day period.²⁰  The secondary outcome of pulmonary complications is defined broadly by the presence of new pulmonary International Classification of Diseases, Tenth Revision codes within 90 days (pneumonia, respiratory failure, pneumothorax, pulmonary embolus) not present on admission.

For descriptive purposes and confounder adjustment of variables potentially associated with the COVID-19 exposure and study outcomes, perioperative characteristics were included as covariates. These included patient demographic, anthropometric, medical history, anesthetic, surgical, and laboratory testing variables as available within the Multicenter Perioperative Outcomes Group database (tables 1 to 3).

We performed a descriptive analysis of patient and procedure characteristics, including the overall cohort and cohorts with and without the exposure variable. As part of our descriptive analysis, we assessed missingness in our variables, which informed the analytic approach: we conducted a complete case analysis if less than 5% of the data was missing in all the variables; otherwise, we implemented multiple imputation by chained equations or omitted a variable if the rate of missingness was disproportionally greater than the other variables. A missing data report is included in the Supplemental Digital Content (https://links.lww.com/ALN/D352).

The timing between the positive test and procedure date and the 30-day postoperative mortality after elective surgery in patients with a previous COVID-19 diagnosis was described using descriptive statistics and data visualization (e.g., histograms and bar charts).

We compared the 30-day postoperative mortality and the perioperative complications of surgical patients who recovered from a previous COVID-19 infection, presenting for unrelated surgery, to a cohort that tested negative for COVID-19. For our primary analysis, we compared the exposed cohort to a contemporaneously matched control cohort with laboratory-confirmed negative tests during the same time period. We also conducted a sensitivity analysis with a matched control from the year before (April 2019 to March 2020). We expected patients who recovered from COVID-19 infection might differ from uninfected patients, and in order to compare outcomes in these two groups, we performed exact matching based on the institution, age, ASA classification, race, sex, and solid tumor (without metastasis). We then used inverse probability weighting to further reduce the potential bias and confounding in our comparisons. In this study, the propensity score was the conditional probability that a patient had a previous laboratory-confirmed COVID-19 infection, given a set of covariates. For each patient in our sample, we estimated the propensity for COVID-19 infection using a nonparametric gradient boosted model to account for possible nonlinear relationships between the covariates and propensity scores.^21,22  The model was executed to estimate the propensity for COVID-19 in the entire cohort and to evaluate the quality of covariate balance between the two groups before and after weighting. The returned propensity scores were then transformed into inverse probability weights and used to construct a weighted cohort of patients who were diagnosed with COVID-19 who were similar to patients with a negative result, based on the observable characteristics. The variables listed in tables 1 to 3 were considered for estimating the propensity scores.

To evaluate balance across groups, we utilized plots and tables to compare the absolute standardized mean difference, defined as the difference between the group means divided by their pooled SD, as an effect size of covariate imbalance between groups. We adopt the position that a standardized difference of 0.1 represents a meaningful imbalance.²³  The P value of the two-sample chi-square test for covariates between the unweighted and weighted results was also calculated (Supplemental Digital Content, https://links.lww.com/ALN/D352). Briefly, the chi-square test examines whether two samples come from the same distribution by comparing observed counts with expected counts under the null hypothesis of a common distribution. A P value greater than 0.05 suggests that there is insufficient evidence of a difference and thus implies a similar covariate distribution between treatment groups.

Common support was examined with density plots to check for overlap across the groups’ propensity score distributions, and then trimming was performed so that group propensities shared a common range.

To compare 30-day postoperative mortality and perioperative complications in recovered COVID-19 patients versus never diagnosed patients, we used a generalized linear mixed model with binomial family and logit link function. The cohort was weighted by the inverse probability weighting to balance the groups in terms of their likelihood of a COVID-19 infection and included a categorical variable for previous COVID-19 diagnosis, as described above. No additional covariates were added, as balance was adequately achieved between diagnosis groups. Two-tailed Wald tests of the categorical variable for previous COVID-19 diagnosis were used to formally test the effect of previous infection at a 5% significance level. The same modeling approach was implemented to compare the perioperative complication rates for AKI and pulmonary complications. We also conducted a sensitivity analysis comparing patients with a previous COVID-19 diagnosis to matched controls from the year before (April 2019 to March 2020) to assess the robustness of the conclusions and to ensure appropriate interpretation of the results. For this matched group, we similarly performed exact matching followed by inverse probability weighting.

To determine if the 30-day postoperative mortality rate was related to the length of time from COVID-19 diagnosis to surgery, we conducted an analysis of patients who took a COVID-19 test before their procedure date. We implemented a generalized linear mixed model with 30-day postoperative mortality as the outcome, regressed on COVID-19 diagnosis, length of time from diagnosis to surgery, and their interaction. Length of time from diagnosis to surgery was binned into 2-week segments: 0 to 2 weeks, 2 to 4 weeks, 4 to 6 weeks, 6 to 8 weeks, and greater than 8 weeks. We then formed comparisons of mortality rates between COVID-19 diagnosis levels within each week segment and used a Holm–Bonferroni correction to adjust for multiple comparisons. All analyses were performed utilizing R 4.2.1 and are available in the Supplemental Digital Content (https://links.lww.com/ALN/D352).

Based on a cohort discovery tool in the Multicenter Perioperative Outcomes Group, there were 77,876 patients with a negative COVID-19 laboratory test and 2,116 confirmed positive patients. Including 2,000 cases in each arm would achieve power to detect a 2.4-fold difference in mortality between groups. This sample size calculation was computed from a two-proportion test with alpha set to 0.05. We did not a priori define a minimally significant difference in mortality rates.

The final analysis included 37 medical centers, yielding 237,126 patients with a preoperative COVID-19 test, each undergoing one index surgery. COVID-19 was detected in 5,361 patients undergoing elective surgery (2.3%). After matching and propensity score trimming, 30-day in-hospital postoperative mortality occurred in 229 of 4,951 (4.6%). Among those reporting creatinine laboratory outcomes, AKI was observed in 172 of 1,814 (9.5%). Among those reporting diagnosis codes, pulmonary complications were observed in 237 of 1,637 (14%).

As described in table 1, our study population had a median age between 54 and 60 yr, was majority white race, was majority female, and majority had an ASA Physical Status score of III. The most common comorbidity was hypertension without complication.

Each positive case was exact-matched to both a contemporaneous and historical control on institution, age, ASA classification, race, sex, and solid tumor (without metastasis). After exact-matching, propensity score weights were constructed using observed presurgical covariates. The resulting covariate balances after weighting are displayed in tables 1 to 3.

After exact matching and inverse probability weighting, COVID-19 exposure was associated with an increased 30-day mortality risk compared to contemporaneous matches (odds ratio, 1.63; 95% CI, 1.38 to 1.91) and historical matches (odds ratio, 1.89; 95% CI, 1.49 to 2.39). COVID-19 exposure was associated with an increased risk of pulmonary complications in contemporaneous matches (odds ratio, 1.60; 95% CI, 1.36 to 1.88) and historical matches (odds ratio, 1.83; 95% CI, 1.53 to 2.18). COVID-19 exposure was not associated with an increased risk of AKI in either contemporaneous matches (odds ratio, 1.03; 95% CI, 0.87 to 1.22) or historical matches (odds ratio, 0.97; 95% CI, 0.81 to 1.16).

The effect of time between COVID-19 infection and time of surgery is shown in figure 1. The 0- to 2-week time frame was associated with a significantly increased risk of mortality, but that effect became nonsignificant after 2 weeks. Figure 2 similarly demonstrates the effect of interval between infection and time of surgery on risk of pulmonary complications. The effect of pulmonary complications was increased in the 0- to 2-week period but also became nonsignificant after 2 weeks. Finally, figure 3 represents the effect of this interval on risk of AKI, demonstrating no increased risk from COVID-19 exposure at any interval.

In this matched and weighted observational study of patients testing positive for COVID-19 before elective inpatient surgery, COVID-19 exposure early in the pandemic was associated with an increased 30-day mortality compared to a contemporaneous matched cohort as well as a historical cohort. Exposure was also associated with an increased risk of postoperative pulmonary complications but not postoperative AKI. Surgery performed within 2 weeks of a positive test was associated with elevated risk of mortality and pulmonary complications, but the effect was not observed beyond 2 weeks.

Previous studies have attempted to define the perioperative mortality of patients infected with COVID-19.^{1–3,7,9,10,24,25}  Many of these studies encompass international surveys and the experience of countries with a different patient demographic than the United States. Most United States studies are limited to isolated surgical populations or do not investigate mortality. One large observational study from the American College of Surgeons (Chicago, Illinois) similarly identified a higher mortality rate, but most of the effect was observed in patients diagnosed with COVID-19 after surgery.²⁶  Our findings give more granular context to the United States experience. Our findings add to the existing literature by assessing risk in the elective surgery population, where there may be greater flexibility with respect to timing of the procedure. Our findings are novel as they compare patients to two sets of rigorously matched controls: contemporaneous controls and historical controls, with confounding further addressed via inverse probability weighting. As the COVID-19 pandemic has changed the healthcare landscape, both sets of controls were important to mitigate confounding due to secular trends as well as risk of occult or unreported infections in control subjects. A major strength of the study is the use of data from multiple centers with shared reporting structure, mitigating reporting bias, and the inclusion in the contemporaneous cohort of only those patients with the presence of at least one COVID-19 test result. Other studies have included exposed patients who were tested for symptomatic reasons rather than a cohort that was tested as a routine.

The risks of surgery after COVID-19 infection must be balanced against the risks of delaying surgery; these data will help to guide the shared decision-making and discussions that should occur among the patient, the surgeon, and the anesthesiologist. In contrast to previous work indicating that the risk of mortality is significant up to 7 weeks,³  we observed that this risk tapered off after 2 weeks. Another study demonstrated that a composite risk of complications and death declined with time after infection.²⁷  This has important clinical practice implications as patients may be able to proceed with surgery after a brief waiting period. Current recommendations advise a waiting period up to 7 weeks for higher-risk patients. National organizations should consider these data when advising how long to delay elective surgery should COVID-19 variants become more lethal and for emerging viral respiratory infections.

Our finding of elevated risk of pulmonary complications carries significant biologic plausibility and is also consistent with previous work demonstrating increased risks both with non–COVID-19 respiratory infections²⁸  and with COVID-19.⁹  As these risks become nonsignificant after 2 weeks from time of test to surgery, our data again may offer guidance for key risk–benefit discussions regarding surgical delay.

The risk of perioperative AKI in the setting of antecedent COVID-19 infection has not been well-studied. AKI was frequently observed in hospitalized COVID-19 patients,²⁹  and an increased perioperative risk has been observed with urgent and emergent surgeries.²⁴  One survey-based study did find a higher rate of AKI in COVID-19 patients undergoing surgery early in the pandemic compared to a matched control 8 yr earlier.³⁰  It is reassuring that we did not identify an increased risk of AKI in the perioperative period among a population of patients undergoing elective surgery. In any event, this finding of the current study deserves further attention in future dedicated studies.

The cohort identified in this study was exposed to previous variants with a known higher risk of mortality compared to more recent omicron variants.³¹  While vaccination data were not available in this study, the sample included a time frame up to April 2021 when most Americans had not yet been vaccinated, but vaccination had begun. Thus, our findings may not be generalizable to vaccinated patients or patients exposed to more recent variants. Indeed, vaccination does appear to confer a lower risk of perioperative mortality.³²  Even so, our findings have important healthcare implications for understanding the risks associated with respiratory infections in general. Much of our guidance on how to manage COVID-19 came from previous work on severe acute respiratory syndrome and Middle Eastern respiratory syndrome. In the event of another respiratory pandemic or evolution of COVID-19, these findings will add to the understanding of perioperative risks to patients undergoing elective surgery.

Our study has several important limitations to consider. First, there was variability in preoperative asymptomatic testing requirements across the United States. However, national standards recommended such practice, and we only included centers that provided robust exposure and outcome data for each covariate of interest. It is likely that the sampled cohort had routine preoperative testing and variable mapping that included the outcomes and exposures of interest. During the epoch sampled, most COVID-19 tests were polymerase chain reaction tests. As such, a proportion of patients with positive COVID-19 samples will represent patients with postinfectious remote viral shedding rather than acute infection. Conversely, antigen tests being less sensitive may have misclassified some infected patients as false negatives. Additionally, the preoperative symptom profiles of the COVID-19–positive patients were not available in this dataset, limiting our ability to discriminate between positivity due to symptomatic infections versus asymptomatic acute infections or remote postinfectious positivity. As discussed, these findings from an early cohort may not generalize to vaccinated patients, patients who have suffered from previous COVID-19 infection, or those exposed to more recent COVID-19 variants. It is also plausible that surgeries performed in a shorter interval from infection were simply more urgent, so there is likely residual confounding by indication. The exposure outcome of pulmonary complications is broad and includes diagnosis codes with disparate mechanistic sources. Finally, there may be underreporting bias in the outcomes of interest due to lack of integration across American healthcare systems, such that patients undergoing surgery at one institution may present elsewhere for care related to, e.g., postoperative pulmonary complications or AKI. Similarly, an institution may not have been made aware of postdischarge mortality events.

In summary, we found that patients with positive test for COVID-19 before surgery early in the pandemic have an elevated risk of perioperative mortality and pulmonary complications, as compared to both matched test-negative contemporaneous controls as well as matched historical controls presenting before the pandemic. We did not observe an effect of COVID-19 on postoperative AKI risk. The span of time from positive test to time of surgery affected the mortality and pulmonary risk, which subsided after 2 weeks and 6 weeks, respectively.

The authors gratefully acknowledge the valuable contributions to protocol and final manuscript review by the Multicenter Perioperative Outcomes Group collaborators, listed in the appendix; and Ngoc Wasson, M.P.H., Anesthesiology and Perioperative Medicine, Oregon Health & Science University, Portland, Oregon, for editing this manuscript. The study is registered at https://mpog.org/currentprojects/.

Funding was provided by departmental and institutional resources at each contributing site. In addition, partial funding to support underlying electronic health record data collection into the Multicenter Perioperative Outcomes Group registry was provided by Blue Cross Blue Shield of Michigan/Blue Care Network (Detroit, Michigan) as part of the Blue Cross Blue Shield of Michigan/Blue Care Network Value Partnerships program. Although Blue Cross Blue Shield of Michigan/Blue Care Network and Multicenter Perioperative Outcomes Group work collaboratively, the opinions, beliefs, and viewpoints expressed by the authors do not necessarily reflect the opinions, beliefs, and viewpoints of Blue Cross Blue Shield of Michigan/Blue Care Network or any of its employees.

Dr. Mathis receives funding from the U.S. National Institutes of Health (Bethesda, Maryland; K01HL141701, R01DK133226) unrelated to the current work. The other authors declare no competing interests.

Supplemental Digital Content, https://links.lww.com/ALN/D352

Supplemental Figure 1: Density plots of propensity scores unweighted (A) and weighted (B). The distribution of the propensity scores between patients diagnosed with COVID-19 (red) and those without known previous diagnosis (blue) were different in the unweighted sample, but more similar in the inverse probability weight–weighted sample.

Supplemental Table 1: Covariate Balance Mortality

Supplemental Figure 1.2: Balance of individual covariates in the primary analysis of mortality in the unweighted and weighted samples.

Supplemental Table 2: Covariate Balance Mortality Sensitivity Outcome

Supplemental Figure 2.2: Balance of individual covariates in the sensitivity (historical) analysis of mortality in the unweighted and weighted samples.

Supplemental Figure 3: Density plots of propensity scores unweighted (A) and weighted (B).

Supplemental Table 3: Covariate Balance: Pulmonary Outcome

Supplemental Figure 3.2: Balance of individual covariates in the primary analysis of pulmonary complications in the unweighted and weighted sample.

Supplemental Figure 4: Density plots of propensity scores unweighted (A) and weighted (B).

Supplemental Table 4: Covariate Balance Pulmonary Sensitivity Outcome

Supplemental Figure 4.2: Balance of individual covariates in the sensitivity (historical) analysis of pulmonary complications in the unweighted and weighted samples.

Supplemental Figure 5: Density plots of propensity scores unweighted (A) and weighted (B).

Supplemental Table 5: Covariate Balance Acute Kidney Injury

Supplemental Figure 5.2: Balance of individual covariates in the primary analysis of acute kidney injury in the unweighted and weighted samples.

Supplemental Figure 6: Density plots of propensity scores unweighted (A) and weighted (B).

Supplemental Table 6: Covariate Balance Acute Kidney Injury Sensitivity Outcome

Supplemental Figure 6.2: Balance of individual covariates in the sensitivity (historical) analysis of acute kidney injury in the unweighted and weighted samples.

Supplemental Table 7: Missing Data Analysis

Supplemental Table 8: Statistical Package Code for Weighting and Matching

Multicenter Perioperative Outcomes Group Investigators:

Nonauthor collaborators who did not meet criteria for authorship but contributed to study design and/or manuscript review have given permission to be listed as collaborators on this article.

The additional Multicenter Perioperative Outcomes Group Investigators for this study are as follows:

Bhiken I. Naik, M.B.B.Ch., M.S.C.R.: Anesthesiology, University of Virginia, Charlottesville, Virginia.

Brad M. Taicher, D.O., M.B.A.: Department of Anesthesiology, Duke University Medical Center, Durham, North Carolina.

Douglas A. Colquhoun, M.B.Ch.B., M.Sc., M.P.H.: Department of Anesthesiology, University of Michigan, Ann Arbor, Michigan.

Robert Freundlich, M.D., M.S., M.S.C.I.: Anesthesiology, Vanderbilt University Medical Center, Nashville, Tennessee.

Robert B. Schonberger, M.D., M.H.S., M.H.C.D.S.: Anesthesiology, Yale School of Medicine, New Haven, Connecticut.

Karen B. Domino, M.D., M.P.H.: Anesthesiology & Pain Medicine, University of Washington, Seattle, Washington.

Lee-lynn Chen, M.D.: Department of Anesthesia and Perioperative Care, University of California-San Francisco, San Francisco, California.

Nathan Pace, M.D., M.Stat.: Anesthesiology, Perioperative and Pain Medicine, University of Utah, Salt Lake City, Utah.