Application of machine learning algorithms for functional outcome prediction after aneurysmal subarachnoid hemorrhage - a single-center retrospective study

Aug 5, 2026

Upload by Quinn

This single-center retrospective study applies machine learning algorithms to predict functional outcome after aneurysmal subarachnoid hemorrhage. The work collects admission and clinical assessment data, including Glasgow Coma Scale, WFNS and Hunt & Hess grades, and evaluates complications. Prognostic models are built with multiple algorithm families, followed by performance assessment using AUROC/AUPR, calibration, and decision curve analysis, supported by forest plot and nomogram. Statistical analyses quantify model discrimination and clinical usefulness.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages67

Reading Time2 h 49 min

AccessFree to download

Abstract

Introduction

Epidemiology of spontaneous subarachnoid hemorrhage (SAH)

Pathophysiology of spontaneous subarachnoid hemorrhage

Machine learning

Objective

Methods

Patients

Data processing

Construction of prognostic models

Evaluation metrics and statistical analysis

Results

Patient characteristics

LR analysis

Evaluation of the models

Discussion

DISSERTATION Application of machine learning algorithms for functional outcome prediction after aneurysmal subarachnoid hemorrhage: a single-center retrospective study Anwendung von “machine learning“ Algorithmen für die Vorhersage der funktionellen Erholung nach subarachnoidaler Blutung-eine monozentrische retrospektive Studie zur Erlangung des akademischen Grades Doctor medicinae (Dr. med. ) vorgelegt der Medizinischen Fakultät Charité – Universitätsmedizin Berlin von Han Wang Erstbetreuung: [Prof. Dr. med. Andreas Patzak](Prof. Dr. med. Andreas Patzak) Datum der Promotion: 15. September 2025 Table of contents List of tables ................................................................................................................... iv List of figures ................................................................................................................... v List of abbreviations ........................................................................................................ vi Abstract ........................................................................................................................... 1 1 Introduction ...............................................................................................................3 1.1 Epidemiology of spontaneous subarachnoid hemorrhage (SAH) .......................3 1.2 Pathophysiology of spontaneous subarachnoid hemorrhage .............................3 1.2.1 Increased intracranial pressure .......................................................................3 1.2.2 Early brain injury ..............................................................................................4 1.2.3 Delayed cerebral ischemia ..............................................................................4 1.3 Machine learning ................................................................................................5 1.3.1 The history of machine learning.......................................................................5 1.3.2 The application of machine learning in medicine .............................................6 1.4 The current challenges of existing prognostic models for aSAH ........................6 1.5 Objective ............................................................................................................7 2 Methods....................................................................................................................8 2.1 Patients ..............................................................................................................8 2.2 Clinical data collection ........................................................................................8 2.3 The detail of clinical assessments upon admission ............................................9 2.3.1 Glasgow Coma Scale (GCS) ...........................................................................9 2.3.2 World Federation of Neurosurgical Societies (WFNS) grades.........................9 2.3.3 Hunt & Hess (H&H) grades ........................................................................... 10 2.4 The diagnosis of complications ........................................................................ 10 2.5 Functional outcome assessment ...................................................................... 11 2.6 Data processing................................................................................................ 11 2.7 Construction of prognostic models ................................................................... 12 2.7.1 LR models ..................................................................................................... 12 2.7.2 KNN models .................................................................................................. 13 2.7.3 XGB models ...........................................................................................

FAQs

Which data sources and clinical assessments are used for model building?

The study uses clinical data collection and admission assessments, including Glasgow Coma Scale, WFNS grades, and Hunt & Hess grades, together with data processing steps and complication diagnosis.

What machine learning model families are constructed for prognostic prediction?

Prognostic models are built using LR, KNN, XGB, RF, and ANN approaches, followed by selection of parameters for advanced algorithms.

How is the predictive performance and clinical value of the models evaluated?

Evaluation includes AUROC and AUPR, calibration curves, receiver operating characteristic and precision-recall curves, and decision curve analysis (DCA) with clinical impact considerations.

Aug 5, 2026

Upload by Quinn