An additive score optimized by a genetic learning algorithm predicts readmission risk after glioblastoma resection

Arka N. Mallela, Prateek Agarwal, Nicholas J. Goel, Joseph Durgin, Mohit Jayaram, Donald M. O'Rourke, Steven Brem, Kalil G. Abdullah

Research output: Contribution to journalArticlepeer-review

2 Scopus citations


Thirty-day readmission following glioblastoma (GBM) resection is not only correlated with decreased overall survival but also increasingly tied to quality metrics and reimbursement. This study aimed to determine factors linked with 30-day readmission to develop a simple risk stratification score. From 2005 to 2016, 666 unique resections (467 patients) of primary/recurrent tissue-confirmed glioblastoma were retrospectively identified. We recorded patient demographics and medical history, tumor characteristics, post-operative complications and 30-day readmission. Univariate and multivariate logistic regression, optimized using a genetic learning algorithm, were used to determine factors associated with readmission. The multivariate model was converted to a simple additive score. The 30-day readmission rate was 20.3% in our cohort of 666 unique resections (60.7% first resection). Lower pre/post-operative KPS, recurrent resection, surgical-site infection, post-operative VTE, post-operative VPS, and discharge to a rehabilitation facility were significantly associated with an increased readmission risk (p < 0.05). MGMT methylation and chemoradiation were associated with decreased readmission risk (p < 0.05). Medical co-morbidities and past medical history, location of tumor in eloquent areas of the brain, and length of ICU/hospital stay did not predict readmission. The Glioblastoma Readmission Risk Score, developed from the multivariate model, accounts for increased BMI, decreased pre-operative KPS, current smoking, post-operative complications, MGMT methylation, and post-operative radiation. This risk score can be routinely used to stratify risk and assist in clinical decision making and outcome analyses.

Original languageEnglish (US)
Pages (from-to)1-5
Number of pages5
JournalJournal of Clinical Neuroscience
StatePublished - Oct 2020


  • Glioblastoma
  • Predictive analytics
  • Readmission
  • Risk score

ASJC Scopus subject areas

  • Surgery
  • Neurology
  • Clinical Neurology
  • Physiology (medical)


Dive into the research topics of 'An additive score optimized by a genetic learning algorithm predicts readmission risk after glioblastoma resection'. Together they form a unique fingerprint.

Cite this