See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/340754458

An empirical investigation of factors driving lactate levels in sepsis patients

using machine learning methods

Conference Paper · March 2020

CITATIONS READS

0 150

12 authors, including:

Saraswathi Sathees Shamsudheen Marakkar

University of Connecticut Cochin University of Science and Technology
2 PUBLICATIONS   2 CITATIONS    7 PUBLICATIONS   5 CITATIONS   

SEE PROFILE SEE PROFILE

Ramya Dhatri Vunikili Shabeeb Kannattikuni

Siemens Healthineers Georgetown University
11 PUBLICATIONS   41 CITATIONS    2 PUBLICATIONS   0 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Genetic resilience View project

An empirical investigation of factors driving lactate levels in sepsis patients using machine learning methods View project

All content following this page was uploaded by Saraswathi Sathees on 19 April 2020.

The user has requested enhancement of the downloaded file.

 An empirical investigation of factors driving lactate levels in
sepsis patients using machine learning methods
Saraswathi Sathees1^, Shamsudheen Marakkar2, Ramya Vunikili3, Shabeeb Kannattikuni4, Kipp W Johnson5,
Benjamin S Glicksberg6, Joseph Scarpa7, Anoop Vadakkan Cherian8, Oommen K Mathew9,
Lakshminarayanan Subramanian3, Joel T Dudley4 and Khader Shameer10*
1University of Connecticut, Storrs, CT; 2Molecular Robotics, Kerala, India; 3Courant Institute, New York University, NY;
4Georgetown University, Washington, DC; 5Institute for Next Generation Healthcare, New York, NY; 6Bakar Computational
Health Sciences Institute, UCSF, CA.; 7Cornell University, New York, NY; 8Max Planck Institute for Heart and Lung Research,
Germany; 9Indian Institute of Information Technology, Kottayam; 10Northwell Health, New York, NY
*Corresponding author; ^ Presenting author
Introduction: Sepsis is a life-threatening disease caused as a result of the body's response to infection [1].
According to the World Health Organization, it is estimated that more than 30 million people are affected
by sepsis worldwide every year, potentially leading to 6 million deaths. Early diagnoses of sepsis are critical
to reducing clinical acuity of sepsis. Recent studies have shown that an increase in lactate level rate as a
crucial biomarker to diagnose sepsis. However, it is unclear how lactate level increase with other routinely
measured clinical biomarkers or survival rate in sepsis population. We hypothesize that lactate dynamics
will have an impact on other clinical biomarkers, and changes in lactate levels could be detected using a
combinatorial, digital biomarker composed of
routinely measured laboratory tests.
Methods: Briefly, we compiled a training cohort
and testing cohort with patients diagnosed with
sepsis and associated laboratory data from
MIMIC databases. A validation cohort was
compiled using phenomic data in UK Biobank
for validation. Collectively, essential variables
affecting the lactate levels are identified. A
library of features was evaluated for correct
classification type (lactate increase, lactate
decrease, lactate increase with sepsis test definite
and lactate increase with sepsis test negative) and
adjusted if necessary. In the Model stage, after
partitioning the dataset, a GridSearchCV on at
least three different ML algorithms are run to
Figure 1:Analytics Workflow
identify the best parameters to use for a model.
Three model interpretability plots, such as partial dependence plot, variable importance plot, and
scatterplots from the best fitting model, were compiled to represent the variable importance. Lastly, in the
Analyze stage- Error metrics table (i.e., RMSE, MAE, standard deviation, MAPE, bias), the variable
importance plot of different algorithms is compared. The best model was chosen from the models run based
on comparing the model accuracy and error metrics. Figure 1 shows the workflow diagram of the analytic
approaches used in the study and can be extended to other data-driven analytics investigations.

Results: It is imperative to treat sepsis in its early stages due to its fatality. Analyzing the phenotype
trajectories and applying machine learning algorithms, we can study how the lactate level rate affects the
sepsis patients. Additionally, this could result in better healthcare delivery systems for the population-scale
risk stratification of sepsis and improve patient outcomes.
References:
1. Liu AC, et.al. Sepsis in the era of data-driven medicine: personalizing risks, diagnoses, treatments and
prognoses. Briefings in bioinformatics. 2019: bbz059.

966

View publication stats