Risk Analysis Index Predicts Complications and Discharge Outcomes after Brain Tumor Resection Better than Patient Age and Modified Frailty Index-5

Data Source

The NSQIP is a nationally validated, multi-institutional database whose characteristics have been summarized before.^16–20 Our data were extracted from the NSQIP database from the years 2015–2019. This study was performed under the data user agreement (DUA) of the American college of Surgeons (ACS) with our institution and was considered exempt by our Institutional Review Board (Study ID 21–315).

Patient population and baseline characteristics

International Classification of Diseases (ICD)-9 and ICD-10 diagnostic codes were used to identify brain tumor patients (over 18 years of age) in the NSQIP dataset (Supplemental Table 1). Current Procedural Terminology (CPT) codes (Supplemental Table 2) were then used to identify all BTR patients. In addition to the overall BTR cohort, we subdivided patients by tumor type to compare RAI scoring within each homogeneous type of brain tumor pathology: meningiomas, primary malignant brain neoplasms, and metastatic brain tumors. Each cohort’s selection criteria were based on the appropriate ICD-9 and − 10 codes (Supplemental Tables 3–5).

Risk Analysis Index-Adminstrative (RAI-A) Scoring

Retrospective RAI-A scoring adapted from the original scoring system proposed by Hall and colleagues is shown in Supplemental Tables 6 & 7.¹⁰

Modified frailty index-5 (mFI-5)

The mFI-5 is comprised of five NSQIP variables: diabetes mellitus, hypertension, dependent functional status, chronic obstructive pulmonary disease (COPD), and congestive heart failure (CHF) (Supplemental Table 8).⁸ The mFI-5 scores range from 0 to 5, where a score of 0 is “non-frail,” 1 is defined as “pre-frail,” 2 as “frail,” and a score of 3 or more as “severely frail”, as the presence of each mFI-5 variable receives a point.

Outcome measures

The outcome measures included in-hospital mortality, major complications, Clavien-Dindo Grade IV complications, unplanned readmission, unplanned reoperation, extended length of stay (eLOS, > 75th percentile of LOS), and non-routine discharge destination (discharge to non-home location).

Statistical Analysis

We conducted statistical analyses using Statistical Package for Social Sciences (SPSS) version 28 (IBM Corp., Armonk, NY). The univariate and multivariate regression analysis were conducted to discern predictive ability of patient age, RAI-A, and mFI-5 on each outcome measure of interest. The effect sizes for dichotomous outcomes were summarized by odds ratio (OR) and associated 95% confidence intervals (95% CI). The association of age, categories of mFI-5 (with reference to ‘no frailty’) and RAI-A (with reference to RAI-A 16–20) with primary outcomes in patients undergoing craniotomy for all cohort of brain tumors is illustrated in forest plot infographics (Fig. 2). For all purposes, p < 0.05 was considered statistically significant.

Study population characteristics

There were 30,951 patients that met inclusion criteria and underwent craniotomy for BTR. The median age of our data sample was 59 years old (interquartile range, IQR 47–68) and 47.8% of the patients were male. Key patient demographics are summarized in Table 1. The median BMI was 27.7 (IQR 24.2–32.2) kg/m². Tumor distribution type was 40.5% primary malignant brain tumors, 23.4% meningiomas, 23.1% metastatic brain tumors, 9.3% benign brain tumors, and 3.7% other tumors of the brain. The majority of cases were non-elective (59.8%) and the other 40.1% were elective surgeries. The distribution of mFI-5 scores was 56% non-frail (0 point), 31.2% pre-frail (1 point), 11.3% frail (2 points), and 1.5% severely frail (≥3 points). The distribution of frailty scores within each diagnostic pathology subtype differed between groups with the metastatic cohort with the greatest incidence of severe frailty scores (Supplemental Tables 9–11). The minimum RAI-A score range was 16 since all patients had cancer as defined by RAI-A scoring (Supplemental Tables 6 and 7). In the overall cohort, median RAI-A was 25 (IQR 20–26) where 34.1% of patients had scores 16–20, 37.2% had scores 21–25, 15% had scores 26–30, 10.9% had scores 31–35, 1.3% had scores 36–40, 1.2% had scores 41–45, and only 0.3% had scores ≥ 46. Incidence of RAI-A scores differed between different diagnostic cohorts. The meningioma cohort had a median RAI-A score of 20 (IQR 20–25), the primary malignant brain tumor cohort had a median RAI-A score of 25 (IQR 20–26), and metastatic tumor cohort had a median RAI-A score of 26 (21–29) (Supplemental Tables 9–11).

 

The most common reported comorbidity was hypertension (37.7%), followed by disseminated cancer 23.2%. Nearly one in five patients (17.8%) were transferred from an outside facility and nearly one in every five (17.6%) were smokers (Table 1). The median operative time was 182 minutes (IQR 122–269), with a median hospital LOS of 4 days (IQR 3–8). The operative time and median hospital LOS were, as expected, dependent on the specific pathology subtype (Supplemental Tables 9–11).

The overall in-hospital mortality rate was 1.1% (333 patients), however, the pathology subtype determined the mortality rate: 0.7% for meningiomas,, 1.1% for primary malignant brain tumor, and 1.6% for metastatic brain tumor (Supplemental Tables 9–11). The overall readmission and reoperation rates were 10.7% and 5.3%, respectively. The overall major complication rate was 10.1% whereas the minor complication rate was 5.4%. The non-home destination discharge rate was 22.1%. Incidence of these outcomes can be seen for each individual cohort subtype in Supplemental Tables 9–11. Figure 1 depicts the incidence of major outcomes in-hospital mortality, major complications, Clavien-Dindo Grade IV, eLOS, and non-routine discharge by age group, mFI-5 frailty, and RAI-A score tiers. RAI-A scores and mFI-5 tiers show a stepwise distribution of major outcome incidence with increasing severity across all outcomes, whereas age does not.

Univariate analysis of age, mFI-5, and RAI-A on surgical outcomes

Univariate analysis demonstrated that increasing RAI-A score tiers offered greater discrimination compared to mFI-5, predictive for all surgical outcomes (Supplemental Table 12). Age alone did not reflect increased odds of poor outcomes. This stepwise incremental increase in odds ratios for RAI-A tiers was consistent in the overall brain tumor cohort as well as each sub-cohort univariate analysis for meningioma, primary malignant brain tumors, and metastatic tumors of the brain, with increasingly worse odds for the metastatic tumor group (Supplementary Tables 13–15).

Multivariate analysis of age, mFI-5, and RAI-A on surgical outcomes

Multivariable regression analysis (adjusting for age, sex, BMI, non-elective surgery status, race, and ethnicity) demonstrated that RAI-A was an independent predictor of all BTR outcomes (Fig. 2). In the forest plot infographic, increasing RAI-A score categories (with reference to RAI-A 16–20) were found to be more predictive in an overall incremental, stepwise fashion, to demonstrate increased risk with each increased RAI-A level for mortality, major complications, Clavien-Dindo Grade IV, ELOS, and nonroutine discharge. Figure 2 depicts the effect sizes of RAI-A compared to mFI-5 for predicting surgical outcomes in all BTR patients. Increasingly high RAI-A score tiers are better able to predict the extent of risk preoperatively compared to the previously used mFI-5 frailty scores. As shown in Fig. 2, the highest mFI-5 scores never have OR greater than 3, whereas RAI-A score tiers exceed this OR with increasing scoring. Table 3 summarizes the multivariate effect sizes of RAI-A analysis for the entire cohort of all brain tumor resection patients and for each pathology subtype of meningiomas, primary malignant brain tumors, and metastatic brain tumors (Supplemental Tables 16–18).

Overall cohort

To the best of our knowledge, this is the first neurosurgical report investigating frailty with the RAI-A. We compared the RAI-A to the mFI-5 in BTR patients and demonstrate that increasing RAI-A scores were an independent risk factor for increased mortality and worse outcomes where RAI-A was also more discriminative than mFI-5. Greater RAI-A scores had discrete, incrementally increasing associations with worse outcomes and was also able to accurately predict BTR mortality rates. This mortality prediction occurred even with the limited in-hospital window, which is noteworthy, as some other surgical studies have found that the RAI-A does a better job predicting increased mortality with a one-year follow-up period.²¹ The RAI-A’s discriminative power in independently predicting mortality rates and outcomes in BTR patients provides critical information for risk assessment and pre-operative patient and family counselling that has not been available previously.

Increasing patient age was not an independent risk factor for increased mortality and worse outcomes in BTR patients. Although neurosurgical evidence reporting that frailty matters more than advanced patient age is accumulating rapidly, this is still a newer concept since advanced patient age has long been used as justification for not offering surgical interventions in some patients. We found the increasing, proportional incremental relationship between increasing RAI and worse outcomes, including increased mortality. A previous BTR study that examined frailty with the mFI-5 frailty scale, noted the limitation of mFI-5 as a frailty tool due to its inability to capture functional impairment associated with the true classical frailty phenotype of limited mobility, impaired ADLs, and limited physiologic reserve.^6,22 The mFI-5 performs better as a comorbidity index, delineating which patients have more comorbidities, and would be expected to have worse outcomes as a result.^10,23

Since this is the first neurosurgical report using RAI, and because it is new for all specialties, the RAI frailty tiers and terminology have not been definitively established.^{11–13, 24} Our analysis was conducted using 5-point scoring intervals for the distribution of scores in our study. This enabled us to evaluate what tier scoring intervals may be appropriate for clinical consideration after statistical analysis was performed. Our data aligns with the recommendation by George et al. that patients with a score of 31–39 should be considered “frail” and patients scoring over 40 as “severely frail”.¹¹ However, we suggest that patients scoring 21–30 be considered “pre-frail,” as we found that “normal” does not reflect their increased likelihood of poor outcomes compared to those patients with a score < 20. We agree that patients with scores < 20 may be considered robust.¹¹ These terms provide meaningful clinical reference points, since increasing RAI-A score is clearly an independent predictor of worse postoperative morbidity and mortality, after controlling for age, mFI-5 score, race, ethnicity, BMI, sex, and elective surgical status. This retrospective RAI-A assessment will be confirmed with prospective studies (RAI-C), which are underway at our institution, with 15 months of RAI-C data for all neurosurgery patients collected.

Sub-cohort Analysis

BTR patients comprise a very heterogeneous group of patients.²⁵ Therefore, we evaluated the three largest pathology subtypes, i.e., meningiomas, primary malignant brain tumors, and brain metastases (n = 7,240, n = 12,550, and n = 7,149 respectively). For each sub-cohort, RAI-A’s superior discriminative ability over mFI-5 in outcome prediction was preserved, but there were different effect sizes with the OR for different outcomes, as would be expected when comparing a benign brain tumor (meningioma) to brain cancers (primary or metastatic). Importantly, increasing patient age was again not predictive of adverse surgical outcomes (OR of ~ 1).

Meningiomas

Meningiomas comprise 39.2% of all brain tumors.²⁶ Though most meningiomas have low postoperative mortality rates and are often curative, they can be associated with significant peri- and post-operative complications when involving cranial nerves, important arteries, and are along the skull base with limited access.^27–30 The most severe frailty group included all patients with RAI-A scores ≥ 41 due to the small number of patients with scores ≥ 46. Increasing RAI-A scores had the highest odds ratios for eLOS and discharge to non-home location. The risk for mortality, major complication, and Clavien-Dindo Grade IV risk was lower compared to the overall study population in the “severely frail” group of ≥ 41, but the OR for ELOS and non-home discharge were high. Frail patients (RAI-A: 36–40) had a significant increase in mortality risk, while increasing mFI-5 scores had no association with increased mortality. Nevertheless, the strongest predictive power of RAI-A was with eLOS and non-home discharge supporting the key findings from a previous meningioma frailty paper that looked at large-data and mFI-5.³¹ However, this study highlights the RAI-A’s superior discriminative power in outcome prediction in this large dataset analysis.

Primary Malignant Tumors of the Brain

Primary malignant tumors make up 29.1% of all primary brain tumors.²⁶ The epidemiology of this group is approximately 61% glioblastoma and 39% other malignant gliomas.²⁶ This cohort also had fewer patients with RAI scores ≥ 46, thus for analysis patients with scores ≥ 41 were combined into one tier of “severely frail”. On multivariate analysis, RAI-A was again independently predictive of stepwise increase in postoperative risk of major adverse outcomes, with a variance in the OR compared to all patients with brain cancer. In the case of primary malignant neoplasms of the brain, the RAI-A best predicted increased mortality and eLOS. In this cohort, the “severely frail” group of RAI ≥ 41 had high OR for mortality and ELOS. Our RAI findings were superior to mFI again and predicts adverse outcomes better than any previous report on older patients with malignant cerebral neoplasm or glioblastoma.^3,32 As primary malignant brain tumors such as glioblastomas have a higher disease burden on the growing elderly population, the benefits of an accurate RAI score prediction will help inform surgeons and families regarding preoperative risk management.^32,33

Metastatic Brain Tumors

Metastatic brain tumors affect 20% of cancer patients; lung cancer patients have the highest incidence of intracranial metastasis, followed by breast cancer and melanoma.^34–38 The incidence of brain metastasis has increased with improved cancer survival and significant advancements in neuroradiological imaging modalities and surveillance techniques.³⁹ Unsurprisingly, metastatic cancer patients are a vulnerable and frail population with increased morbidity and mortality rates with surgical interventions, when compared to patients without metastatic disease. More than half of metastatic brain tumor patients die following progression of their brain lesions, however as surgical treatments combined with radiation therapy regimens both continue to improve and whole-brain radiation use and toxicity decrease, outcomes have been improving.^40–42 Expectedly, this sub-cohort had the greatest increase in OR associated with increasing RAI-A scores. The metastatic group also had increased OR for Clavien-Dindo Grade IV complications in comparison to the other disease cohorts irrespective of RAI score. Previous studies have demonstrated that the mFI-5 and Johns Hopkins Adjusted Clinical frailty tools frailty scales have good ability to predict adverse outcomes for metastatic brain tumor patients, but the RAI has superior discrete stepwise prediction of adverse outcomes.^4,5

Limitations

The primary limitations of this study are inherent to analysis based on any large national database, and the results need to be interpreted accordingly. Due to the predetermined standard NSQIP variables, we modified the RAI-A score to exclude the missing preoperative cognitive decline variable as has been done in other studies that demonstrated significant discriminative ability in other surgical specalties.^10–14 Additionally, weight loss and poor appetite are not discrete variables in NSQIP variable recording structure and thus all patients were recorded having + 9 for presence of the affirmative weight loss variable alone, since these two variables are often both present and since the NSQIP does not differentiate them. Despite these NSQIP RAI-A scoring limitations, we were able to generate a modified RAI-A scoring system that was superior to anything published, which can be used for future retrospective cohort studies. Another limitation was that meningioma patients were given a baseline score of 16–20 points based on age with cancer, so that they could be compared directly with the other two brain cancer cohorts as ‘cancer’ as a term is defined by the Hall et al. study as “any diagnosis of cancer excluding non-melanoma skin cancers […] defined as any cancer diagnosis with or without metastasis, and thus includes a wide range of disease processes” and therefore reasonably includes meningiomas.¹⁰

There are also limitations given the lack of many variables critical in neurooncology clinical outcomes research that are not recorded in the NSQIP such as tumor size, surgical approach, and extent of resection. There may be inherent selection bias present in any retrospective study, even when the data is prospectively collected. Despite these limitations, this study is robust, with a large sample size of almost 31,000 patients, providing the necessary statistical power to compare the RAI-A score to mFI-5 score as a significant predictor of postoperative outcomes in BTR patients. Furthermore, our study will be followed by a separate prospective RAI-C report with over a year of data collection already ongoing and where the stratification of frailty tiers and effect sizes for postoperative outcomes will be analyzed. We anticipate there may be some differences in longer term outcomes, as our NSQIP RAI-A studies are only able to measure short-term postoperative outcomes, whereas prospective studies will be able to study 180-day and 1-year outcomes.²¹ Nevertheless, the 30-day limitation inherent to the NSQIP database is irrelevant, particularly with the consideration that other RAI studies found increasing RAI scores to be more predictive of mortality even when the 180 and 1-year follow-ups were included.^11,13,21