This study is a retrospective cohort design and was approved by the Naval Medical Center Portsmouth’s Institutional Review Board (NMCP.2021.0054).

Data from the ACS NSQIP database during the years 2015-2019 were obtained. These data come from over 700 hospitals and are collected using well-described methods to assure a high level of validity [18]. Noncardiac surgical procedures were included using current procedural terminology (CPT) codes 10000-32999 and 34000-69999. Patients undergoing cardiac surgery were excluded from this analysis due to the unique risks associated with that patient population, including the risks associated with cardiopulmonary bypass. Similar to previous investigations [19], we excluded minor surgeries such as endoscopies (CPT 43200-43272, 45300-45392, 46600-46608) and minor musculoskeletal procedures (CPT 29000-29750). Additionally, patients were excluded if they underwent emergency surgery.

The following demographic and health data were collected for each patient: CPT code, age, race, ethnicity, height, weight, sex assigned in the medical record, functional status, American Society of Anesthesiologists (ASA) Physical Score, sodium level, hematocrit, creatinine, steroid use, ascites, sepsis or septic shock, ventilator dependence, disseminated cancer, diabetes, hypertension, weight loss (at least 10% in the past year), congestive heart failure (CHF), dyspnea, smoking, chronic obstructive pulmonary disease, and dialysis. Patient records were included based on the following criteria: sodium, hematocrit, and creatinine assessment <5 days prior to surgery; BMI of >12 and <60; ages 18 to 89 years; hematocrit of >21% and <50%; sodium level of ≥130 mEq/L and ≤150 mEq/L; creatinine level of ≥0.5 mg/dL and ≤4.0 mg/dL; and undergoing surgery under a primary CPT listed in at least 50 patient records.

The primary exposure was the preoperative sodium level. A priori, the serum sodium level of 140 mEq/L was empirically determined to be the reference value for the development of statistical models.

The 2 primary outcomes were defined as aggregate morbidity within 30 days of index surgery and mortality within 30 days of index surgery. Aggregate morbidity included any of the following: cardiac arrest, myocardial infarction, cerebrovascular accident, deep vein thrombosis, pulmonary embolism, postoperative sepsis or septic shock, renal insufficiency or failure, reintubation, failure to wean from the ventilator, pneumonia, wound dehiscence, or surgical site infection (including superficial, deep, or organ space). Details regarding the standardized definitions of these variables have been previously published [19].

Of the 4,551,726 patient records available, 1,003,956 met all inclusion criteria. Most patient records were excluded due to laboratory assessments occurring more than 4 days from surgery or not at all (n=3,388,178), continuous variables outside of the prespecified ranges (n=145,458), and a primary CPT that was not represented in at least 50 patient records (n=14,134). Bivariate analyses indicated that those included versus excluded differed across all identified demographic characteristics and medical comorbidities (Multimedia Appendix 1). In the included sample, 15,474 (0.3%) patient records had missing data; therefore, no imputation was performed. Morbidity and mortality rates in the included cohort were 8.5% and 1.3%, respectively. Descriptive statistics are reported (Table 1). Morbidity (Table 2) and mortality status (Table 3) are also reported. Bivariate test results indicated that all demographic characteristics and medical comorbidities were associated with morbidity and mortality status. As such, all of these factors were included as covariates in the GAMs.

In both outcome GAMs, all continuous covariates (age, BMI, sodium, hematocrit, creatinine, and percent CPT morbidity or mortality) were modeled as smooth terms and were substantially associated with both morbidity and mortality. Across both models, patients assigned male in the medical record with an elevated ASA status, steroid use, sepsis or septic shock, cancer, a positive smoking status, chronic obstructive pulmonary disease, renal failure, and CHF were more likely to experience morbidity and mortality compared to their reference counterparts (Multimedia Appendix 2). When controlling for other demographic characteristics and medical comorbidities, patients whose race and ethnicity were listed as Asian or Latino had a lower probability of morbidity and mortality relative to White patients. Similarly, White patients had a greater probability of morbidity relative to Native Hawaiian and Pacific Islander patients, but White patients had a lower probability of morbidity than patients of unknown race and ethnicity. Patients with diabetes and hypertension had a greater probability of morbidity relative to those without these conditions. The odds of morbidity and mortality across sodium levels of 130-150 mEq/L relative to a sodium level of 140 mEq/L followed a U-shaped curve (Figure 1).

The E-values corresponding to the RR of morbidity and mortality across sodium levels are shown in Figures S1A and S1B in Multimedia Appendix 3. For example, at a sodium level of 135 mEq/L, the morbidity (RR 1.07, 95% CI 1.07-1.07) and mortality (RR 1.30, 95% CI 1.30-1.30) would be rendered null if an unmeasured confounder was associated with both sodium and the outcome by a RR of 1.35-fold (lower 95% CI 1.35) and 1.93-fold (lower 95% CI 1.92), respectively. For reference, these E-values are similar to the effects of ASA I versus III on morbidity (RR 1.33, 95% CI 1.32-1.33) and CHF on mortality (RR 1.97, 95% CI 1.94-1.99). The RR (95% CI) for fixed effects on both outcomes are reported in Multimedia Appendix 2.

Lastly, GAMs evaluating morbidity and mortality were conducted on a subsample of the included group with the lowest tercile propensity scores (n=333,701). Model results were similar to the main analysis, such that the effect of sodium was significant (P<.001), and the nonlinear pattern followed a U-shape (Figures S1C and S1D in Multimedia Appendix 3).

This exploratory analysis calls into question the current understanding of the “normal” range of serum sodium levels (135-145 mEq/L) within the context of perioperative care, as values of serum sodium concentration within this range of normal values were associated with 30-day aggregate morbidity and mortality. By examining preoperative sodium levels in over 1 million patients as a continuous variable instead of the commonly used categories (eg, hyponatremic, eunatremic, and hypernatremic), this study provides improved granularity on the association between small deviations in sodium and perioperative outcomes. As such, what is considered “normal” sodium values in the general population may not be normal in patients undergoing elective noncardiac surgery.

As health care shifts to value-based care, these findings may also play a role in evaluating value-based perioperative practices. For example, recent evidence using NSQIP data indicates that preoperative laboratory assessment is not associated with the odds of postoperative complications and readmission in patients undergoing ambulatory surgery with an ASA I or II status, thereby suggesting the low value of preoperative laboratory assessment [31]. However, such findings may be premised on clinician practices that are contingent on a definition of “normal” that is, per these findings, associated with increased risk of aggregate morbidity and mortality (eg, ~135 mEq/L). Given the potential impact of these findings, combined with the lack of causal assumptions that can be made, future work is needed to assess whether clinical intervention addressing high- and low-normal sodium serum concentrations improves clinical outcomes and value-based care.

This study possessed several strengths. Though the main variable of interest was serum sodium concentrations, models included many demographic characteristics and medical comorbidities that have previously been shown to be substantially associated with aggregate morbidity and mortality risk. These factors included other laboratory values (eg, creatinine and hematocrit) that may also warrant further inspection, given their relationship with postoperative outcomes. By controlling for these covariates and using a weighted approach based on the inverse probability of record inclusion, the results of this study are likely generalizable to adult patients undergoing any elective, noncardiac surgery in the United States. We restricted records to those with laboratory results collected less than 5 days before surgery, thereby increasing the likelihood that the recorded values actually reflected serum sodium levels at the time of surgery.

This study was tempered by several limitations. First, no causal conclusions can be drawn from the study due to the retrospective, associative nature of the study design and analytic approach. Additionally, there may be several covariates, including specific health conditions, medication receipt (both in the days leading up to surgery and perioperatively), preoperative recommendations (eg, fasting), and prior health care received, that are neither collected in the NSQIP database nor included in the analysis but could be associated with morbidity and mortality. While this database is a robust and extensive collection of surgical outcome data in the United States [18], the inclusion of other covariates mentioned above could serve to refine this model and provide more specific areas of research to explore. Examples of other potential confounders include medications, preoperative fasting, and certain comorbidities, which themselves may be associated with abnormal sodium levels. When considering the potential impact of missing confounders on model results, E-values indicated that any confounder would need to surpass the strength of most fixed covariates within our models and account for unique variance not otherwise accounted for by current covariates to render the effect of sodium null.

This analysis indicated that both preoperative hyponatremia and preoperative hypernatremia were associated with an increased risk of 30-day aggregate morbidity and mortality. The relationship was nonlinear, such that the risk increased with further deviation from a serum sodium concentration of 140. While prior investigations have demonstrated that dysnatremia is a modifiable risk factor and optimization of preoperative serum sodium levels may represent an opportunity for a reduction in both perioperative morbidity and mortality [31], this study suggests that preoperative serum sodium levels that are within the currently accepted upper and lower limits of normal are likely indicative of elevated risk. As such, future prospective studies are needed to better confer sodium level ranges associated with optimized outcomes after surgery, as well as the potential to directly alter patients’ serum sodium concentrations to improve postoperative outcomes.