In NeuroICU, treatments typically adhere to guidelines based on scientific consensus.
Despite this, the prognosis for patients with intracranial hemorrhages has not
significantly improved over recent decades, resulting in generally unsatisfactory
outcomes. While randomized controlled trials (RCTs) are considered the gold standard for
clinical research, they can be expensive and ethically challenging to conduct.
Observational studies provide an alternative method, offering larger datasets covering
longer periods, which can be more beneficial and feasible for certain research endeavors.
Machine Learning (ML) algorithms, unlike classical statistical methods, have the
capability to process a vast number of variables, offering a personalized and dependable
approach for healthcare providers in patient management. Recognizing the necessity for
well-designed studies to identify optimal therapeutic strategies for neurocritical
patients and acknowledging the limitations of existing guidelines, we aim to leverage ML
programs to develop an advanced system capable of uncovering data patterns and linking
them to potential outcomes.
The BLANDISH project focuses on patients with spontaneous intracerebral hemorrhage
(sICH), a condition lacking proven beneficial treatment. By collecting and analyzing data
from sICH patients admitted to neuroICU, the project aims to develop a supervised ML
algorithm named BLANDISH.
This algorithm will stratify patients based on prognosis, identifying those at highest
risk of death and secondary brain injuries. By guiding each patient towards the most
targeted therapeutic strategy, the algorithm could help improve patient outcomes and
assess the effectiveness of current clinical practices. Additionally, it may enable
healthcare staff to better allocate resources and introduce individualized therapeutic
programs based on precision medicine, potentially reducing hospitalization times and
healthcare costs.
The initial step in developing the BLANDISH algorithm involves collecting clinical data,
stored in a structured datalake, which serves as a data repository. This platform will
gather information on the clinical course of sICH patients admitted to neuroICU. After
data collection, the next steps include preprocessing, variable selection correlated with
patient mortality, and algorithm training with input data and internal validation to
control its behavior. External validation will follow, involving data collection from
various clinical centers to assess the algorithm's reliability and generalization
capacity. A multicenter observational clinical study will then be conducted to validate
the BLANDISH algorithm, aiming to determine its impact on sICH patient outcomes. This
final phase includes a survival study comparing patients in the experimental group (whose
treatment is guided by BLANDISH) with those following standard clinical practice. The
project aims to demonstrate the superiority of the ML approach over current guidelines,
evaluating the accuracy and potential improvements in patient management across different
care settings.