Our study validated the models published by Khalafallah et al. to validate their calculator’s (https://neurooncsurgery.shinyapps.io/brain_mets_calculator/) ability to predict poor outcomes regarding extended LOS and non-routine discharge disposition for BM patients who undergo neurosurgical procedures [9] 9. We found high accuracy of the proposed predictive models in our population, which included a greater range of cranial procedures. Each independent predictor in our multivariate analyses was analyzed, and notably the mFI-5 score was not a strong predictor of outcome, even after subgroup analyses of craniotomy only patients.
Our patient cohort was a similar in number compared to Khalafallah et al.’s (244 patients vs. 235, respectively) and showed similar demographic profiles, cancer types, and proportion of patients with a solitary BM, supratentorial tumors, and extracranial metastases [9]. Our mean KPS and mFI-5 scores were also similar. We did have more patients with a history of prior radiation therapy (46.3% vs. 27.2%). Our cohort had a higher percentage of patients who were dependent with their activities of daily living (ADLs) (23.4% vs. 7.2%). Khalafallah et al.’s group had more patients who underwent craniotomy (99.6% vs 75.8%), while we had a portion of patients who underwent less invasive procedures (11.9% LITT and biopsy, 9% biopsy alone).
Our post-operative complication rate was lower than that of Khalafallah et al.’s cohort (15.6% vs. 36.6%), even accounting for neurosurgical procedure type (our complication rate for craniotomy-only patients was 14.1%) [9] 9. Identifying the reason for this difference is beyond the scope of this study, although it would be interesting to determine in future studies given the similarity of our overall patient cohorts.
In terms of outcomes, Khalafallah et al.’s cohort had a higher percentage of patients with an extended LOS (25.1% vs. 8.6%, Khalafallah vs. ours) and non-routine discharge disposition (22.6% vs 11.1%) [9]. Khalafallah et al.’s cohort also had a higher mean length of hospital stay than our cohort (6.17 days vs. 3.6 days). There are several possible explanations for these differences. First, the majority of Khalafalleh et al.’s patients underwent a craniotomy and our neurosurgical population included patients who underwent procedures other than craniotomy, which are generally more minimally invasive procedures. Our data supports that those who undergo craniotomies have longer LOS than those who undergo biopsy or LITT with biopsy procedure. Another factor to consider is that the greater number of patients with non-routine discharge disposition in their study may also account for the higher percentage of patients requiring extended LOS, as sometimes discharge is delayed due to logistical delays finding suitable accepting facilities or number of ancillary team members involved in discharge planning (such as case managers). Lastly, there may also be institutional differences in terms of definition of “readiness of discharge,” which are beyond the scope of our study. One potential explanation for Khalafallah et al.’s models being better predictors of extended LOS is that they had a broader range of LOS (mean 6.17 +/- SD 6.41 days vs. ours mean 3.6 +/- SD 2.72 days), and thus with greater differences in outcomes than ours, there would be more variation to predict.
Our areas under the ROC curves (0.8427 and 0.8422 for extended LOS and non-routine discharge disposition, respectively) support high accuracy of the models proposed by Khalafallah et al. for these outcomes [9]. However, mFI-5 was not significant in our regression models, even though our mean mFI-5 was similarly as low as the Khalafallah et al.’s cohort (1.1 vs. 1.0). When we excluded the mFI-5 from the regressions, the ROC curve for extended LOS did improve, though it remained the same for non-routine discharge disposition. We initially hypothesized the significant of the mFI-5 was affected by the greater range of neurosurgical procedures performed at our institution compared to Khalafallah et al.’s. The presence of comorbidities may more greatly impact the outcomes of patients who undergo craniotomies, which are more invasive compared to other neurosurgical procedures. However, in a separate subgroup analysis of our craniotomy only patients, the mFI-5 was still not significant (Supplementary Table 1). To further investigate why the mFI-5 was not significant in our cohort, we performed separate analyses on components of the mFI-5. We noted similar percentages of comorbidities between our group and Khalafallah et al.’s, except our patient cohort had more patients who were dependent (23.4% vs. 7.2%). We thus sought whether dependent functional status was its own independent risk factor for poor outcome. Surprisingly, dependent functional status was significant in its lower association with extended LOS (0.15, p=0.0338). One explanation for this finding is that given these patients were dependent before admission, they had likely either come from home with home health services previously arranged or a non-home admission source, which would have facilitated quicker discharge given arrangements already being in place. We also included hypertension in this separate analysis since it was the comorbidity that was most prevalent (51.6%). Hypertension was not a significant variable when analyzed independently. Overall, we were not able to validate the mFI-5 as a significant predictor of outcome, and the significance of the mFI-5 in surgical outcomes remains an area of further study.
As far as other independent predictors of outcome, three of four predictors of extended LOS were similarly significant in our analyses: non-Caucasian race, non-home admission source, and KPS, while mFI-5 was not. As far as independent risk factors for non-routine discharge disposition, age and non-home admission source were significant predictors in our initial multivariate analysis. In our model, age seemed to have been a negative predictor of non-routine discharge disposition (OR 0.95, 95% CI 0.91-0.99), while it was a positive predictor in the Khalafallah et al. study (OR 1.07, 95% CI 1.02-1.11) [10, 11]. When we performed the regression without the mFI-5, age was no longer significant which made the initial difference irrelevant. In contrast to Khalafallah et al., neither race nor KPS were independent risk factors for non-routine discharge disposition [9]. Similar to Khalafallah et al., marital status was not a significant predictor of this outcome. In addition, since our cohort included a greater variety of type of neurosurgical procedures performed, we did a separate analysis including type of procedure in our regressions and found it to be an independent predictor of LOS. Overall, our study supports the model for extended LOS as a strong predictive model and that the model for non-routine discharge disposition is heavily influenced by non-home admission source, and perhaps other factors not included originally.
This study is limited by its retrospective design, single institution design, and limited sample size. Its results inspire further research using this online calculator for predicting post-operative outcomes and informing treatment plans. Notably, our study was more inclusive of patients who underwent a larger range of neurosurgical procedures offered to BM patients. Accurate predictive calculators of neurosurgical outcomes are valuable. For BM patients at various time points of disease and treatment courses with many prognostic uncertainties, this additional piece of information can improve their care decisions.