Negin Ashrafi

Negin Ashrafi

University of Southern California, Los Angeles, California, USA
orcid

Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
16 August 2024 in BMC Medical Informatics and Decision Making
Verified
Journal Article
Open access
PDF file
JGJiayi Gao
Jiayi Gao
YLYuying Lu
Yuying Lu
NANegin Ashrafi
Negin Ashrafi
IDIan Domingo
Ian Domingo
KAKamiar Alaei
Kamiar Alaei
MPMaryam Pishgar
Maryam Pishgar
Problem: Sepsis, a life-threatening condition, accounts for the deaths of millions of people worldwide. Accurate prediction of sepsis outcomes is crucial for effective treatment and management. Previous studies have utilized machine learning for prognosis, but have limitations in feature sets and model interpretability. Aim: This study aims to develop a machine learning model that enhances prediction accuracy for sepsis outcomes using a reduced set of features, thereby addressing the limitations of previous studies and enhancing model interpretability. Methods: This study analyzes intensive care patient outcomes using the MIMIC-IV database, focusing on adult sepsis cases. Employing the latest data extraction tools, such as Google BigQuery, and following stringent selection criteria, we selected 38 features in this study. This selection is also informed by a comprehensive literature review and clinical expertise. Data preprocessing included handling missing values, regrouping categorical variables, and using the Synthetic Minority Over-sampling Technique (SMOTE) to balance the data. We evaluated several machine learning models: Decision Trees, Gradient Boosting, XGBoost, LightGBM, Multilayer Perceptrons (MLP), Support Vector Machines (SVM), and Random Forest. The Sequential Halving and Classification (SHAC) algorithm was used for hyperparameter tuning, and both train-test split and cross-validation methodologies were employed for performance and computational efficiency. Results: The Random Forest model was the most effective, achieving an area under the receiver operating characteristic curve (AUROC) of 0.94 with a confidence interval of ±0.01. This significantly outperformed other models and set a new benchmark in the literature. The model also provided detailed insights into the importance of various clinical features, with the Sequential Organ Failure Assessment (SOFA) score and average urine output being highly predictive. SHAP (Shapley Additive Explanations) analysis further enhanced the model’s interpretability, offering a clearer understanding of feature impacts. Conclusion: This study demonstrates significant improvements in predicting sepsis outcomes using a Random Forest model, supported by advanced machine learning techniques and thorough data preprocessing. Our approach provided detailed insights into the key clinical features impacting sepsis mortality, making the model both highly accurate and interpretable. By enhancing the model’s practical utility in clinical settings, we offer a valuable tool for healthcare professionals to make data-driven decisions, ultimately aiming to minimize sepsis-induced fatalities.

4 September 2024 in PLOS ONE
Verified
Journal Article
Open access
PDF file
HLHexin Li
Hexin Li
NANegin Ashrafi
Negin Ashrafi
CKChris Kang
Chris Kang
GZGuanlan Zhao
Guanlan Zhao
YCYubing Chen
Yubing Chen
MPMaryam Pishgar
Maryam Pishgar
Background: Mechanical ventilation (MV) is vital for critically ill ICU patients but carries significant mortality risks. This study aims to develop a predictive model to estimate hospital mortality among MV patients, utilizing comprehensive health data to assist ICU physicians with early-stage alerts. Methods: We developed a Machine Learning (ML) framework to predict hospital mortality in ICU patients receiving MV. Using the MIMIC-III database, we identified 25,202 eligible patients through ICD-9 codes. We employed backward elimination and the Lasso method, selecting 32 features based on clinical insights and literature. Data preprocessing included eliminating columns with over 90% missing data and using mean imputation for the remaining missing values. To address class imbalance, we used the Synthetic Minority Over-sampling Technique (SMOTE). We evaluated several ML models, including CatBoost, XGBoost, Decision Tree, Random Forest, Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Logistic Regression, using a 70/30 train-test split. The CatBoost model was chosen for its superior performance in terms of accuracy, precision, recall, F1-score, AUROC metrics, and calibration plots. Results: The study involved a cohort of 25,202 patients on MV. The CatBoost model attained an AUROC of 0.862, an increase from an initial AUROC of 0.821, which was the best reported in the literature. It also demonstrated an accuracy of 0.789, an F1-score of 0.747, and better calibration, outperforming other models. These improvements are due to systematic feature selection and the robust gradient boosting architecture of CatBoost. Conclusion: The preprocessing methodology significantly reduced the number of relevant features, simplifying computational processes, and identified critical features previously overlooked. Integrating these features and tuning the parameters, our model demonstrated strong generalization to unseen data. This highlights the potential of ML as a crucial tool in ICUs, enhancing resource allocation and providing more personalized interventions for MV patients.

25 June 2024 in medRxiv
Verified
Preprint
Open access
PDF file
JZJiahong Zhang
Jiahong Zhang
HLHexin Li
Hexin Li
NANegin Ashrafi
Negin Ashrafi
ZYZhijiang Yu
Zhijiang Yu
GPGreg Placencia
Greg Placencia
MPMaryam Pishgar
Maryam Pishgar
Heart failure affects millions of people worldwide. It greatly reduces quality of life and is associated with high mortality rates. Despite extensive research, the statistical connection between heart failure and mortality rates for ICU patients remains underexplored, indicating the need for improved prediction models. This study identified 1,177 patients over 18 years old from the MIMIC-III database using ICD-9 codes. Preprocessing consisted of handling missing data, deleting duplicates, treating skewness, and oversampling to alleviate data imbalances. 18 features were selected within a LightGBM model by checking Variance Inflation Factor (VIF) values, LASSO Regression, and univariate analysis. The final output of the LASSO Logistic Regression model had the highest test AUC-ROC of 0.8766 (95% CI 0.8065 - 0.9429) and accuracy of 0.7291 compared to other baseline models, including Logistic Regression, Random Forest, LightGBM, Support Vector Machine (SVM), and Decision Trees. All models demonstrated good calibration with relatively low Brier scores, highlighting their reliability in predicting in-hospital mortality. Our models predicted deaths of heart failure ICU patients better than the best results found in both literature and baseline models. These results were based on preprocessing missing values via improved imputation strategies and improved feature selection based on an expanded literature search and improved experiences selecting key features. With the Grid-Search, we had a near-perfect predictive model. These methods greatly increased the predictive accuracy of in-hospital mortality in ICU patients with heart failure.

6 August 2025 in IoT
Verified
Journal Article
Open access
PDF file
LSLi Sun
Li Sun
NANegin Ashrafi
Negin Ashrafi
MPMaryam Pishgar
Maryam Pishgar
Urban public transportation systems face increasing pressure from shifting travel patterns, rising peak-hour demand, and the need for equitable and resilient service delivery. While complex network theory has been widely applied to analyze transit systems, limited attention has been paid to behavioral segmentation within such networks. This study introduces a frequency-based framework that differentiates high-frequency (HF) and low-frequency (LF) passengers to examine how distinct user groups shape network structure, congestion vulnerability, and robustness. Using over 20 million smart-card records from Beijing’s multimodal transit system, we construct and analyze directed weighted networks for HF and LF users, integrating topological metrics, temporal comparisons, and community detection. Results reveal that HF networks are densely connected but structurally fragile, exhibiting lower modularity and significantly greater efficiency loss during peak periods. In contrast, LF networks are more spatially dispersed yet resilient, maintaining stronger intracommunity stability. Peak-hour simulation shows a 70% drop in efficiency and a 99% decrease in clustering, with HF networks experiencing higher vulnerability. Based on these findings, we propose differentiated policy strategies for each user group and outline a future optimization framework constrained by budget and equity considerations. This study contributes a scalable, data-driven approach to integrating passenger behavior with network science, offering actionable insights for resilient and inclusive transit planning.

16 March 2024 in medRxiv
Verified
Preprint
Open access
PDF file
JGJiayi Gao
Jiayi Gao
YLYuying Lu
Yuying Lu
NANegin Ashrafi
Negin Ashrafi
IDIan Raymond Domingo
Ian Raymond Domingo
KAKamiar Alaei
Kamiar Alaei
MPMaryam Pishgar
Maryam Pishgar
Background Sepsis, a life-threatening condition, is the cause of a large number of mortalities worldwide. Accurate prediction of sepsis outcomes is crucial for effective treatment and management. Previous studies have explored machine learning for prognosis but have limitations in feature sets and model interpretability. Methods This study analyzes intensive care patient outcomes using the MIMIC-IV database, focusing on adult sepsis cases. Employing the latest data extraction tools, such as Google Big Query, and following stringent selection criteria, we selected 38 features in this study. This selection is also informed by a comprehensive literature review and clinical expertise. We used statistical methods to handle the imbalances inherent in healthcare datasets. Our modeling focused on various classification techniques, with a train-test split preferred over cross-validation for its superior performance and computational efficiency. Results The Random Forest model emerged as the most effective, achieving an AUROC of 0.94 with a confidence interval of 0.01, significantly outperforming other baseline models, and the best result in our literature review. This study not only yields a high-performing model but also provides granular insights into the factors affecting mortality, and demonstrates the value of advanced analytics in critical care. Conclusions The study shows significant improvement in predicting sepsis outcomes, indicating the potential of machine learning in critical care. By enhancing model accuracy and stability, this research contributes to clinical decision-making, offering a pathway for data-driven approaches to reduce sepsis-induced fatalities. What is New? Use of advanced data preprocessing techniques addressing missing or duplicate values, and regrouping categorical variables. Improved feature selection process, achieving more accurate predictions. Use of additional scoring tools, allowing for a more comprehensive assessment of the proposed methodology’s performance. Better interpretability of the model’s predictive outcomes through the use of SHAP analysis. What are the Clinical Implications? The use of machine learning methods in medicine allows for an immediate and accurate second opinion, serving as an alternate source of confirmation for medical professionals. Mortality predictions are a valuable asset to resource management in hospitals. Knowing a patient’s potential mortality outcome allows hospitals to refactor their resources to help those in a more desperate state. Predictive models help in making more efficient use of healthcare services, allowing for patients at greater risk of death to be treated more urgently, helping save the most lives.

16 September 2025 in IEEE Journal of Biomedical and Health Informatics
Verified
Journal Article
Open access
HJHouji Jin
Houji Jin
NANegin Ashrafi
Negin Ashrafi
KAKamiar Alaei
Kamiar Alaei
EPElham Pishgar
Elham Pishgar
GPGreg Placencia
Greg Placencia
MPMaryam Pishgar
Maryam Pishgar
Sepsis is a major cause of ICU mortality, where early recognition and effective interventions are essential for improving patient outcomes. However, the vasoactive-inotropic score (VIS) varies dynamically with a patient's hemodynamic status, complicated by irregular medication patterns, missing data, and confounders, making sepsis prediction challenging. To address this, we propose a novel Teacher–Student multitask framework with self-supervised VIS pretraining via a Masked Autoencoder (MAE). The teacher model performs mortality classification and severity-score regression, while the student distills robust time-series representations, enhancing adaptation to heterogeneous VIS data. Compared to LSTM-based methods, our approach achieves an AUROC of 0.829 on MIMIC-IV 3.0 (9,476 patients), outperforming the baseline (0.74). SHAP analysis revealed that SOFA score (0.147) had the greatest impact on ICU mortality, followed by LODS (0.033), single marital status (0.031), and Medicaid insurance (0.023), highlighting the role of sociodemographic factors. SAPSII (0.020) also contributed significantly. These findings suggest that clinical and social factors should be considered in ICU decision-making. Our multitask and distillation strategies enable earlier identification of high-risk patients, improving prediction accuracy and disease management, offering new tools for ICU decision support.