Data were collected from the ACS-NSQIP dataset from 2017 to 2019. The ACS-NSQIP database contains prospectively collected data from over 600 hospitals internationally, detailing surgical patient demographics, procedures, and adverse events within 30 days. This database has been widely used for outcomes research in the spine surgery literature [9–11]. The ACS-NSQIP data are audited for inter-observer agreement, with an overall disagreement rate of 2.3% [12]. Since the data are de-identified and publicly available, this study is exempt from review by our institutional review board, and patient consent was neither required nor sought. The dataset encodes diagnoses using the International Classification of Diseases, 10th Revision, Clinical Modification (ICD-10) codes and procedures using the Current Procedural Terminology (CPT) system.

Patients were included if they had CPT code 22551 (anterior interbody fusion, with discectomy and decompression; cervical below C2) or 22554 (arthrodesis, anterior interbody technique, including minimal discectomy to prepare interspace [other than for decompression]; cervical below C2).

Individuals were excluded if they (1) were under 18 years of age; (2) underwent a revision procedure; (3) had an American Society of Anesthesiologists (ASA) classification of 4 or higher (since ASA >3 is considered a contraindication to discharge on postoperative day 1); (4) underwent an emergency or non-elective procedure; (5) received a multiple-level procedure, as identified by CPT codes 22552 or 22585; (6) had an ICD-10 or CPT code indicating deformity, infection, malignancy, dislocation, discectomy without fusion, osteotomy, non-anterior procedures, or non-cervical procedures; (7) had the procedure performed by someone other than a neurosurgeon or orthopedic surgeon; (8) received bone morphogenetic protein-2 (since the U.S. Food and Drug Administration issued a warning against its use in the cervical spine in 2008); (9) were hospitalized for more than 30 days (as ACS-NSQIP does not collect information on complications beyond that point); or (10) had missing data regarding discharge date.

We collected information regarding patient age, sex, race, height, weight, ASA classification, functional status (independent or dependent), diabetes, smoking (within 1 year), chronic obstructive pulmonary disease (COPD), congestive heart failure, hypertension, corticosteroid use for a chronic condition, hematocrit, sodium, partial thromboplastin time (PTT), international normalized ratio (INR), platelets, leukocytes, surgeon specialty, operation length, and use of a graft (identified using CPT codes 20930, 20931, and 20936–20938).

Age was categorized at the 25th and 75th percentiles for the eligible study population, resulting in age groups of 18–57, 58–65, and >65 years. Anemia was defined as hematocrit below 41% in males and below 36% in females. Platelets were categorized as <150,000/μL or ≥150,000/μL; leukocytes as ≥12,000/μL, <4,000/μL, or in between; albumin as ≥3 g/dL or <3 g/dL; sodium as ≥135 mEq/L or <135 mEq/L; PTT as >40 seconds or ≤40 seconds; and INR as >1.4 or ≤1.4. Height and weight were used to calculate body mass index in kg/m², and were then categorized into World Health Organization (WHO) class I obesity (30–34.9 kg/m²), WHO class II obesity or above (≥36 kg/m²), and non-obese. Graft type was categorized as allograft or autograft. Categories for operative time were determined by selecting the 30-minute intervals nearest to the 50th and 75th percentiles of the eligible study population (120 and 150 minutes, respectively).

The primary outcome of interest was LOS, defined as the number of calendar days between the day of the operation and the day of discharge. To investigate the patient characteristics associated with LOS >1 day, we divided the study population into cohorts of individuals with LOS ≤1 day and LOS >1 day. The LOS ≤1 cohort includes individuals discharged on the same day as the procedure or after one night of observation.

We were also interested in outcomes and complications within 30 days for each cohort. These included rates of unplanned readmission, reoperation, blood transfusion, wound-related infection (both superficial and deep surgical site infections), deep venous thrombosis (DVT), pulmonary embolism, sepsis, pneumonia, urinary tract infection, myocardial infarction, wound dehiscence, and acute renal failure. To understand which complications were experienced by patients with a longer hospital stay and which potentially precluded discharge, we studied rates of pre-discharge complications. Similarly, we compared rates of complications occurring after discharge to identify delayed complications that may not be immediately recognized or treated in patients discharged early or those who underwent procedures in the ambulatory setting.

Missing data in the ACS-NSQIP are typically due to unmeasured laboratory values or missing comorbidity information not recorded by the participating institution. Baseline rates of missing data were assessed for each laboratory value, and those with <40% missing data were imputed using a non-parametric multiple imputation method (missForest), based on a random forest machine learning approach [13]. Variables imputed in this manner include hematocrit, leukocytes, platelets, sodium, and INR.

As all covariates were categorical, we first performed unadjusted comparisons between patients with LOS >1 day and LOS ≤1 day using Pearson’s chi-square test. To evaluate predictors of LOS >1, we conducted separate univariate logistic regression analyses for each covariate. Covariates with a p-value <0.20 on univariate analysis and present in >1.0% of the study population were entered into a final multivariate logistic regression model to avoid skewing the results with unrelated predictors. Variables that met these criteria and were incorporated into the model include age, sex, race, ASA classification, functional status, obesity, diabetes, hypertension, COPD, steroid use, anemia, platelets, leukocytes, sodium, operative length, and use of allograft. To determine whether individuals discharged within 1 day were at greater risk for pre- or post-discharge complications, we used a propensity score matching algorithm with a balanced nearest-neighbor approach, incorporating all collected covariates. We compared baseline characteristics in the matched groups using Pearson’s chi-square tests to ensure that differences were mitigated by matching. We also performed forward and backward stepwise logistic regression analysis over all collected covariates, using 30-day readmission as the outcome, to identify risk factors for readmission in patients discharged within 1 day.

An α of 0.05 was determined a priori to represent statistical significance. Due to the large number of independent comparisons, Bonferroni multiple comparisons correction was applied, resulting in an adjusted α=0.002 for the entire study [14]. Bonferroni correction is a statistical method used to correct for the random chance of finding a significant comparison at a given α value when performing numerous independent comparisons. Accordingly, we reported 99.8% confidence intervals (CIs) for all odds ratios (OR). All statistical analysis was performed using R ver. 3.6.3 (R Foundation for Statistical Computing, Vienna, Austria), stepwise logistic regression was performed using the R package MASS, and propensity score matching was performed using MatchIt ver. 3.0.2 (https://cran.r-project.org/package=MatchIt) [15,16].

Preoperative factors associated with LOS >1 for single-level ACDF included age >65 years, female sex, non-White identity, ASA class 3, and dependent functional status. Increased age has been associated with extended LOS, readmission, and complications following ACDF in other studies, and was present in our analysis. Older patients tend to have more comorbidities, fewer physiological reserves, and decreased access to personal and social resources [17]. Female sex is a significant marker for prolonged LOS. In a case series of 4,995 patients undergoing transforaminal lumbar interbody fusion, female sex was associated with increased rates of admission [18]. Furthermore, prior reports have found that females had higher utilization of healthcare resources before surgery, although some studies have reported contrasting findings, indicating that this result may require further investigation [18,19]. Non-White race as a factor for LOS >1 may be attributed to decreased socioeconomic status. Existing studies have linked socioeconomic factors, such as insurance status, marital status, and race, to differing surgical outcomes [8,20]. In a case series of 1,896 patients undergoing ACDF, non-White race was significantly associated with increased LOS [8]. A separate case series of 278 patients found non-White race associated with longer LOS and a larger percentage of extended stays [20]. However, the impact of race on surgical outcomes and LOS is controversial, as other studies have found no differences in outcomes [21]. ASA class is a general marker for health and comorbidities, used to evaluate patients before surgery [22]. Higher ASA class has been shown in other spine studies to be associated with extended LOS and worse outcomes [8,23]. Dependent functional status is a significant marker of extended LOS. These patients are typically associated with more medical comorbidities and a diminished ability to perform activities of daily living and instrumental activities of daily living [24]. In one national database study of 17,075 patients undergoing elective cervical spine surgery, dependent functional status was associated with increased mortality and 30-day complications, including sepsis, pulmonary, renal, and cardiac complications [24].

Intraoperative factors associated with LOS >1 included operative length >120 minutes and perioperative blood transfusion. Increased operative length suggests higher case complexity and may be associated with a greater risk of complications and blood loss [25]. Additionally, patients who receive perioperative blood transfusions are often associated with higher estimated blood loss and lower preoperative hemoglobin levels, suggesting more comorbidities [25]. Prior reports have also found that patients undergoing cervical surgery who receive perioperative blood transfusions are at higher risk of extended LOS and perioperative complications [26,27]. However, the use of allografts was significantly associated with decreased LOS. In comparison, autografts require additional incisions in the iliac crest to harvest graft material, which has previously been associated with increased donor site pain and may be relevant in this cohort [8]. Consequently, these patients may have required more robust pain management before discharge.

Any clinical complication requiring inpatient management will extend LOS. However, in this analysis, pre-discharge complications associated with extended LOS included reoperation and pneumonia. Immediate reoperation after the initial surgery may result from life-threatening complications such as hemorrhage or infection [28]. Specifically, in ACDF, esophageal perforation, instrumentation failure, expanding cervical hematoma, and CSF leaks have been previously reported as complications necessitating reoperation [29]. Another complication unique to ACDF is dysphagia and recurrent laryngeal nerve injury due to traction and the proximity of the esophagus and other structures [30]. These patients are more susceptible to aspiration pneumonia due to impaired swallowing, requiring longer inpatient management to treat their complications.

As this study utilized a large, retrospective national database, there are limitations. The ACS-NSQIP database variables were not selected specifically to examine risk factors and complications associated with single-level ACDF. As such, many complications associated with ACDF, including dysphagia and pseudarthrosis, are not quantified in the database. These would provide further clarity on the causes of reoperation and extended LOS. Perioperative complications, such as DVT, pulmonary embolism, sepsis, myocardial infarction, and excessive wound drainage, may require further inpatient management after surgery and would be associated with longer LOS. However, these values, particularly for pulmonary embolism, trended towards significance, indicating that this study would benefit from increased power. Additionally, a significant degree of variation in surgeon preferences regarding decisions such as surgical drain placement, direct or indirect decompression, and discharge criteria is not captured in this analysis. Regardless, with rising costs and complications associated with extended LOS, these results may help identify patients who would benefit from surgery in the ambulatory setting.