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