Zreik, Jad

Purpose Aneurysmal rupture and subarachnoid hemorrhage (SAH) have an exceptionally high mortality and morbidity burden. The aim of this study was to develop interpretable machine learning models for predicting short-term poor outcomes defined by the National Inpatient Sample Subarachnoid Hemorrhage Outcome Measure (NIS-SOM). Methods The National Inpatient Sample (NIS) database was queried from 2008 to 2018 to identify patients diagnosed with SAH who had undergone endovascular coiling or clipping for intracranial aneurysm. Demographic, comorbidity, risk factor, and hospital characteristic variables were recorded. Variables were preprocessed, and the feature space was reduced to include the most important features. To predict poor outcomes, machine learning models were trained and cross-validated before being evaluated on a separate testing set. Shapley Additive exPlanations of the best performing model was used for general and local model interpretation. Results Among 18,149 admissions (mean age 55 ± 14 years, 68.8% women), 52.9% had a poor outcome. Test-set AUCs ranged 0.74–0.80; a multilayer perceptron performed best (AUC 0.80, precision 0.74, recall 0.82). SHAP ranked the ten most influential variables: age, neurological comorbidity, paralysis, Medicare insurance, smoking status, Elixhauser burden, fluid-electrolyte disorders, weight loss, arrhythmia, and heart failure. Conclusions The modeling predicted nationwide aSAH prognosis with decent accuracy and highlighted clinical, socioeconomic, and system-level drivers of determinants of poor short-term outcome. These results support the potential of explainable ML tools as complementary tools for early risk stratification, guiding resource allocation, and informing prospective multi-center validation and implementation studies.

OBJECTIVE Although it has been shown that surgery for glioblastoma (GBM) at high-volume facilities (HVFs) may be associated with better postoperative outcomes, the use of such hospitals may not be equally distributed. The authors aimed to evaluate racial and socioeconomic differences in access to surgery for GBM at high-volume Commission on Cancer (CoC)–accredited hospitals. METHODS The National Cancer Database was queried for patients with GBM that was newly diagnosed between 2004 and 2015. Patients who received no surgical intervention or those who received surgical intervention at a site other than the reporting facility were excluded. Annual surgical case volume was calculated for each hospital, with volume ≥ 90th percentile defined as an HVF. Multivariable logistic regression was performed to identify patient-level predictors for undergoing surgery at an HVF. Furthermore, multiple subgroup analyses were performed to determine the adjusted odds ratio of the likelihood of undergoing surgery at an HVF in 2016 as compared to 2004 for each patient subpopulation (by age, race, sex, educational group, etc.). RESULTS A total of 51,859 patients were included, with 10.7% (n = 5562) undergoing surgery at an HVF. On multivariable analysis, Hispanic White patients (OR 0.58, 95% CI 0.49–0.69, p < 0.001) were found to have significantly lower odds of undergoing surgery at an HVF (reference = non-Hispanic White). In addition, patients from a rural residential location (OR 0.55, 95% CI 0.41–0.72, p < 0.001; reference = metropolitan); patients with nonprivate insurance status (Medicare [OR 0.78, 95% CI 0.71–0.86, p < 0.001], Medicaid [OR 0.68, 95% CI 0.60–0.78, p < 0001], other government insurance [OR 0.68, 95% CI 0.52–0.86, p = 0.002], or who were uninsured [OR 0.61, 95% CI 0.51–0.72, p < 0.001]); and lower-income patients ($50,354–$63,332 [OR 0.68, 95% CI 0.63–0.74, p < 0.001], $40,227–$50,353 [OR 0.84, 95% CI 0.76–0.92, p < 0.001]; reference = ≥ $63,333) were also found to be significantly associated with a lower likelihood of surgery at an HVF. Subgroup analyses revealed that elderly patients (age ≥ 65 years), both male and female patients and non-Hispanic White patients, and those with private insurance, Medicare, metropolitan residential location, median zip code–level household income in the first and second quartiles, and educational attainment in the first and third quartiles had increased odds of undergoing surgery at an HVF in 2016 compared to 2004 (all p ≤ 0.05). On the other hand, patients with other governmental insurance, patients with a rural residence, and those from a non-White racial category did not show a significant difference in odds of surgery at an HVF over time (all p > 0.05). CONCLUSIONS The present analysis from the National Cancer Database revealed significant disparities in access to surgery at an HVF for GBM within the United States. Furthermore, there was evidence that these racial and socioeconomic disparities may have widened between 2004 and 2016. The findings should assist health policy makers in the development of strategies for improving access to HVFs for racially and socioeconomically disadvantaged populations.

Currently, large numbers of clinical trials are performed to investigate different forms of experimental therapy for patients suffering from chronic spinal cord injury (SCI). However, for the enrollment process, there are different views on how the time period between injury and interventions should be determined. Herein, we sought to evaluate the impact of time-to-enrollment in chronic SCI clinical trials. A data set comprising 957 clinical studies from clinicalTrials.gov was downloaded and analyzed focusing on the eligibility criteria for post-injury time-to-enrollment. We also aggregated individual patient data from nine clinical trials of regenerative interventions for chronic SCI selected by a systematic literature search from 1990 to 2018. Characteristics of the studies were assessed and compared by dividing into three groups based on time-to-enrollment (group 1 ≤ 12 months, group 2 = 12–23 months and group 3 ≥ 24 months). In ClinicalTrials.gov registry, 445 trials were identified for chronic SCI where 87% (385) were unrestricted in the maximum post-injury time for trial eligibility. From systematic literature search, nine studies and 156 patients (group 1 = 30, group 2 = 55 and group 3 = 71) were included. The range of time-to-enrollment was 0.5 to 321 months in those studies. We also observed various degrees of motor and sensory improvement in between three time-to-enrollment groups. Our results indicate that enrolling wide ranges of time-to-enrollment in a group may present imprecise outcomes. Clinical trial designs should consider appropriate post-injury time frames to evaluate therapeutic benefit.

OBJECTIVE Spinal procedures are increasingly conducted as outpatient procedures, with a growing proportion conducted in ambulatory surgery centers (ASCs). To date, studies reporting outcomes and cost analyses for outpatient spinal procedures in the US have not distinguished the various outpatient settings from each other. In this study, the authors used a state-level administrative database to compare rates of overnight stays and nonroutine discharges as well as index admission charges and cumulative 7-, 30-, and 90-day charges for patients undergoing outpatient lumbar decompression in freestanding ASCs and hospital outpatient (HO) settings. METHODS For this project, the authors used the Florida State Ambulatory Surgery Database (SASD), offered by the Healthcare Cost and Utilization Project (HCUP), for the years 2013 and 2014. Patients undergoing outpatient lumbar decompression for degenerative diseases were identified using CPT (Current Procedural Terminology) and ICD-9 codes. Outcomes of interest included rates of overnight stays, rate of nonroutine discharges, index admission charges, and subsequent admission cumulative charges at 7, 30, and 90 days. Multivariable analysis was performed to assess the impact of outpatient type on index admission charges. Marginal effect analysis was employed to study the difference in predicted dollar margins between ASCs and HOs for each insurance type. RESULTS A total of 25,486 patients were identified; of these, 7067 patients (27.7%) underwent lumbar decompression in a freestanding ASC and 18,419 (72.3%) in an HO. No patient in the ASC group required an overnight stay compared to 9.2% (n = 1691) in the HO group (p < 0.001). No clinically significant difference in the rate of nonroutine discharge was observed between the two groups. The mean index admission charge for the ASC group was found to be significantly higher than that for the HO group ($35,017.28 ± $14,335.60 vs $33,881.50 ± $15,023.70; p < 0.001). Patients in ASCs were also found to have higher mean 7-day (p < 0.001), 30-day (p < 0.001), and 90-day (p = 0.001) readmission charges. ASC procedures were associated with increased charges compared to HO procedures for patients on Medicare or Medicaid (mean index admission charge increase $4049.27, 95% CI $2577.87–$5520.67, p < 0.001) and for patients on private insurance ($4775.72, 95% CI $4171.06–$5380.38, p < 0.001). For patients on self-pay or no charge, a lumbar decompression procedure at an ASC was associated with a decrease in index admission charge of −$10,995.38 (95% CI −$12124.76 to −$9866.01, p < 0.001) compared to a lumbar decompression procedure at an HO. CONCLUSIONS These “real-world” results from an all-payer statewide database indicate that for outpatient spine surgery, ASCs may be associated with higher index admission and subsequent 7-, 30-, and 90-day charges. Given that ASCs are touted to have lower overall costs for patients and better profit margins for physicians, these analyses warrant further investigation into whether this cost benefit is applicable to outpatient spine procedures.

OBJECTIVEThe National Surgical Quality Improvement Program (NSQIP) and National Readmissions Database (NRD) are two widely used databases for research studies. However, they may not provide generalizable information in regard to individual institutions. Therefore, the objective of the present study was to evaluate 30-day readmissions following anterior cervical discectomy and fusion (ACDF) and posterior lumbar fusion (PLF) procedures by using these two national databases and an institutional cohort.METHODSThe NSQIP and NRD were queried for patients undergoing elective ACDF and PLF, with the addition of an institutional cohort. The outcome of interest was 30-day readmissions following ACDF and PLF, which were unplanned and related to the index procedure. Subsequently, univariable and multivariable analyses were conducted to determine the predictors of 30-day readmissions by using both databases and the institutional cohort.RESULTSAmong all identified risk factors, only hypertension was found to be a common risk factor between NRD and the institutional cohort following ACDF. NSQIP and the institutional cohort both showed length of hospital stay to be a significant predictor for 30-day related readmission following PLF. There were no overlapping variables among all 3 cohorts for either ACDF or PLF. Additionally, the national databases identified a greater number of risk factors for 30-day related readmissions than did the institutional cohort for both procedures.CONCLUSIONSOverall, significant differences were seen among all 3 cohorts with regard to top predictors of 30-day unplanned readmissions following ACDF and PLF. The higher quantity of significant predictors found in the national databases may suggest that looking at single-institution series for such analyses may result in underestimation of important variables affecting patient outcomes, and that big data may be helpful in addressing this concern.