The American College of Surgeons National Surgical Quality Improvement Project (NSQIP) 2012-2015 databases were obtained for this analysis. These databases contain validated, multi-institutional data collected in a standardized method across institutions by trained coders (21). The NSQIP databases contain cases from private and academic centers. In 2012, 374 hospitals provided data, which rose to 603 by 2015.

The NSQIP database was filtered by relevant criteria. Inclusion required a billing code for brain metastasis (ICD9 198.3 or ICD10 C79.31), and a neurosurgical CPT code of 61510, 61518, 61520, 61521, 61526, or 61530. Patients were included if they were at least 18 years old, had a neurosurgeon as the primary attending surgeon for their case, and received a procedure under general anesthesia. These criteria yielded 3567 cases. Of these, 67 were excluded for operative time under 30 minutes or post-operative length of stay of 0 days since these were assumed to be part of an aborted or miscoded procedure.

Pertinent preoperative and perioperative co-variates were obtained for each case. Sex, age, body mass index (BMI), case urgency, pre-operative co-morbidities, American Society of Anesthesiologist (ASA) class, and pre-operative laboratory abnormalities were included. Tumor location was extrapolated based on the associated current procedural terminology (CPT) code, with infratentorial (CPT 61518, 61520, 61521, 61526, or 61530) and supratentorial (CPT 61510) strata. These codes were initially derived to stratify procedures for billing purposes. Operative time was binned into quartiles.

The modified frailty index (mFI) with 5 co-variates (mFI-5) was calculated for each case as an estimate of patients’ chronic disease burden with one point assigned for congestive heart failure (CHF), chronic obstructive pulmonary disease (COPD), diabetes mellitus (DM), hypertension (HTN), and non-independent functional status. This scale is a modification of the validated mFI-11, with strong concordance with the mFI-11 in prior nationwide analyses (16, 22). A mFi-5 score of 0 was considered non-frail, 1 as pre-frail, and 2+ as frail.

NSQIP databases encode postoperative outcomes up to 30 days, including surgical site infections (SSIs), wound dehiscence, pneumonia, reintubation, deep vein thrombosis (DVT)/pulmonary embolism (PE), urinary tract infections (UTI), cerebrovascular accidents (CVAs), cardiac arrests, myocardial infarction (MI), transfusions, sepsis, extended length of stay defined as over 1 week, reoperation, unplanned readmission, and death. Outcomes of interest were further stratified by whether they occurred during the index hospitalization or after initial discharge. Analyses of post-discharge complications excluded patients who died during the index hospitalization.

Complications encoded on the day of discharge were considered post-discharge as they would likely have prolonged hospitalizations had they occurred prior to a discharge. In contrast, death that was encoded on the day of discharge was considered pre-discharge given that a critical patient at risk of death would likely not have been discharged that same day.

The R Package for Statistical Computing version 3.5.1 was utilized for all analyses. Cases with missing co-variate data were binned into “missing” categories and treated as a separate independent variable during all analyses. The impact of frailty and tumor location on all postoperative outcomes was calculated as odds ratios (OR) through multivariate regression with adjustment for sex, age, BMI, ASA class, transfer from outside hospital, tumor location, smoking, steroid usage, case urgency, pre-operative lab abnormalities, operative time, use of operative microscope and stereotactic navigation. A two-sided probability value (p) of under 0.05 was utilized as the threshold for statistical significance.

A predictive model for 30-day death was also developed. A univariate screen of all co-variates associated with death, excluding any co-variates with fewer than 0.1% of any level associated with a death, was conducted using a χ² statistical significance cut-off of p < 0.10. The screened co-variates were included in a multivariate logistic regression with backwards stepwise variable elimination. The Hosmer-Lemeshow test and area under the receiver-operator curve (AUROC) concordance statistic was utilized to assess calibration and discrimination, respectively. The pre-determined value for an acceptable calibration for the model was p ≥ 0.05.

Within the study population, 3500 patients with brain metastases were included, with 55% being female ( Table 1 ). Median age of this cohort was 61 years (Interquartile range [IQR] 54-69). Infratentorial lesions represented 24% of cases. Patients were considered non-frail (n=1672, 48%), pre-frail (n=1245, 36%), or frail (n=583, 17%) as quantified by the mFI-5 index. Most patients were admitted to the same hospital where they presented (n=2722, 78%), and 50% of cases were electively scheduled. Median operative length was 139 minutes (IQR 98-193).

We first examined post-operative medical complications and interrogated their temporal patterns in an effort to identify risk factors for poor or unexpected outcomes. Median length of stay following operation was 3 days (IQR 2-5); 14% of patients were hospitalized for longer than 7 days postoperatively. The most frequent postoperative medical complications overall were venous thromboembolism (VTE; including DVTs and PEs), pneumonia, UTI, and sepsis ( Table 2 ). Stratifying complications by whether they occurred during index hospitalization or after initial discharge revealed temporal trends for specific complications ( Table 2 ). Cardiac events (myocardial infarctions and cardiac arrests) and CVAs tended to occur within 0-2 days postoperatively and prior to discharge ( Figure 1A ). In contrast, the median time to occurrence for in-hospital DVTs, infectious complications, and reoperations tended to be 4-6 days. Complications that arose after discharge occurred with a wide temporal spread ( Figure 1A ). Superficial site infections and sepsis tended to be more common after discharge than before discharge.

Reoperation occurred in 5% of patients (n=170). Reoperations were equally likely to occur during index hospitalization as after discharge ( Figure 1B ). The most frequently coded reoperation within 30 days was hematoma evacuation (n=33, 1% of all patients), which occurred a median of 2 days post-operatively (IQR 0-6), followed by the placement of external ventricular drain (EVD) or cerebrospinal fluid (CSF) shunt (n=22, 0.6%), which occurred after a median of 3 days (range 0-6) and 13 days (IQR 6-16) respectively. Among hematoma evacuations, 28% occurred on the same day as the index procedure while another 21% occurred on the first post-operative day.

Unplanned readmission was observed in 12% of cases (n=427). The most frequent indications surgical site infections (SSI) (n=39, 1%), non-infectious cardiac or pulmonary diseases (n=28, 0.8%), oncological management of metastasis (n=28, 0.8%), and pneumonia (n=27, 0.8%).

Death within 30 days of a craniotomy occurred in 4% of patients and demonstrated wide temporal trends related to discharge. While the majority of 30-day deaths occurred after discharge (72%), over a quarter were within the index hospitalization. Amongst patients with tracked mortality in this cohort, 66% were male and the median age was 63 years (IQR: 55 – 71). These patients were sicker than those who did not die: 32% of patients who died within 30 days were frail compared to 16% of those who did not. Furthermore, death after discharge occurred in 7% of patients requiring a reoperation compared to 3% of patients who did not (OR 2.4, p = 0.006). Similarly, death after discharge occurred in 12% of readmitted patients compared with 2% of non-readmitted patients (OR 7.7, p < 0.001).

We hypothesized that tumor location might affect the risk for a post-operative complication and mortality. Patients with infratentorial metastasis demonstrated similar demographic and co-morbid characteristics as those with supratentorial metastases, including age, gender, BMI, prior smoking history, CHF, COPD, hypertension, and diabetes. Craniotomies for infratentorial lesions were associated with an average operative time 33 minutes longer than supratentorial approaches (p < 0.001).

Infratentorial location was found to be associated with increased odds for multiple 30-day medical complications, reoperation, and unplanned readmission (all p < 0.05) ( Table 3 ). In particular, it was associated with increased odds of SSI (OR 2.9), pneumonia (OR 1.6), and reintubation (OR 1.9). It was also associated with length of stay above 7 days (OR 1.3) ( Figure 2 ). Interestingly, infratentorial location was associated with decreased odds of VTE (OR 0.6).

Focusing on those cases with reoperation, infratentorial location was associated specifically with reoperations during the index hospitalization but not after discharge ( Table 3 ). Amongst infratentorial lesions, 27% of reoperations were for CSF shunting (not including lumbar drains), 15% entailed evacuation of a hematoma, and 11% were for placement of EVD. Placement of EVD or CSF shunt after the initial craniotomy for tumor was significantly more likely for infratentorial metastases than supratentorial metastases (OR 7.1).

Concordantly, infratentorial location was associated with significant risk of unplanned readmission for hydrocephalus (OR 5.1) ( Table 3 ). The presence of hydrocephalus correlated with reoperation for CSF shunting or EVD placement for infratentorial lesions: 75% of hydrocephalus cases received either CSF shunt or EVD compared to 0.4% of cases without hydrocephalus (χ², p < 0.001). Readmissions for electrolyte/metabolic derangements (OR 4.5) and surgical site infections (OR 2.5) were also more common following infratentorial approaches. There was no statistically significant association between tumor location and death (p = 0.77).

We further assessed the association of frailty, as a composite reflection of chronic disease burden, and postoperative outcomes. Frail patients had significantly higher likelihood of developing pneumonia (OR 2.8) ( Table 4 ). Frailty was also associated with increased risk of reoperation on the index hospitalization compared to non-frail patients (OR 2.0), but not reoperation after discharge ( Figure 2 ). Frail patients were especially at risk for surgical evacuation of hematoma (OR 3.6). Frailty was not associated with unplanned readmission overall or any specific reasons for readmission. Pre-frailty and frailty had elevated odds ratios of 1.8 and 2.6 for death compared to non-frail patients, respectively. On temporal analysis, frailty was associated with death after discharge, but not before discharge.

A multivariate logistic regression analysis with stepwise variable elimination was performed to construct a predictive model for death within 30 days after surgery for brain metastasis ( Table 5 ). Notably, frailty and pre-frailty were significant predictors in this model. Thrombocytopenia below 100,000 cells/dL had the highest strength of association of any variable. Additional predictors within the model included ASA score of 4/5, male sex, age over 75 years, other preoperative laboratory abnormalities, non-elective case designation, and preoperative steroid use. The model demonstrated a fair ability to correctly classify patients who die prematurely (AUROC 0.75).