About

Aaron Goldstein is an assistant department head for undergraduate studies and an associate professor in the Department of Chemical Engineering at Virginia Tech. He earned his Ph.D. from Carnegie Mellon University in 1997 and his B.S. from the University of California, Berkeley, in 1990. His research interests include biomedical and tissue engineering as well as interfacial phenomena in bioengineering. He is involved in the research group focused on these areas and is based in Goodwin Hall at Virginia Tech, located at 635 Prices Fork Road, Blacksburg, VA 24061.

Research topics

• Medicine
• Internal medicine
• Biochemistry
• Cell biology
• Biology
• Pathology
• Intensive care medicine
• Composite material
• Engineering
• Immunology
• Biomedical engineering
• Anesthesia
• Materials science
• Chemistry
• Cardiology
• Cancer research

Selected publications

• Latency-Bounded Retrieval Indexing for Real-Time Specialized Vision-Language-Action Inference on Warehouse Robots

  2026-01-01

  articleOpen access
  PublisherDOI

• Classification of electrospun material composition using early dynamic response of cantilever sensors and feature-based machine learning

  Sensors and Actuators A Physical · 2026-04-07

  article
  PublisherDOI

• Predictors of 90-day mortality in elderly and unfit AML patients treated with single-agent hypomethylating agents or in combination with venetoclax

  Blood · 2025-11-03

  articleOpen access

  Abstract Background: Hypomethylating agents (HMAs) alone or in combination with venetoclax (VEN) offer a lower-intensity and tolerable treatment option for older and unfit acute myeloid leukemia (AML) patients. However, despite improved response rates, early mortality remains problematic. In the VIALE-A trial (DiNardo et al., 2020), nearly 20% of patients receiving either HMA alone or HMA-VEN died within 90 days, often due to treatment-related complications. High morbidity and mortality within the first three cycles underscore the need to identify patients at highest-risk for early treatment-related death to better tailor therapy. Methods: We performed a retrospective cohort study to identify clinical and molecular predictors of 90-day mortality in patients treated with HMA ± VEN between 2016 and 2025; 592 patients were identified by querying AML ICD-10-CM codes within the Northwell Health EMR. A cohort of 338 patients with AML (defined by the 2022 International Consensus Classification) received upfront HMA ± VEN induction (~76% received HMA+VEN). Univariate analysis (Mann-Whitney, chi-square, Fisher's exact tests) was conducted across 100 variables; p < 0.25 was used in the selection process for building the multivariable logistic regression model. Variables included age, ECOG performance status (PS), AML subtype, initial bone marrow findings, Day –7 to Day +1 pre-treatment transfusion dependence, laboratories, comorbidities, and various aspects of social support. In the multivariable analysis, adjusted odds ratios (aOR) and 95% confidence intervals (CI) were reported to quantify the strength and direction of associations. P<0.05 was considered statistically significant. Results: Among 338 patients (median age 79.9 years; range 42–102; 48% female; 50% White), the overall 90-day mortality rate was 34%. Univariate analysis identified 14 variables associated with early mortality: ECOG PS ≥ 2 (n=338; p < 0.001), adequate social support (n=260; p = 0.090), chronic kidney disease (Stage IIIb-V) (n=37; p = 0.085), dementia (n=14; p = 0.085), obstructive sleep apnea (n=15; p = 0.113), complex karyotype (n=127; p = 0.026), KMT2A rearrangement by FISH (n=19; p = 0.006), del(5q) by FISH (n=55; p = 0.112), blast % on marrow or peripheral flow cytometry (n=260; p = 0.034), increasing WBC count (n=334; p = 0.147) and LDH (n=313; p < 0.001) as continuous variables, RBC transfusion (n=135; p = 0.087) and platelet transfusion (n=134; p = 0.239) needs within 7 days of induction, and an active infection on day 1 of induction therapy (n=85; p<0.001). TP53 mutation status (n=78; p = 0.54) and age (n=338; p-value 0.219) were not significantly associated with 90-day mortality. The final analysis identified five independent predictors of 90-day mortality: KMT2A rearrangement (aOR=3.4 [95% CI:1.1-10.1]), infection requiring hospitalization on day 1 of induction therapy (aOR=3.1 [95% CI:1.7-5.4]), ECOG PS (2 vs 0–1: aOR=2.9 [95%CI: 1.6-5.1]; 3–4 vs 0-1: aOR=4.9 [95% CI: 2.1-11.1]), complex karyotype (aOR=2.4 [95% CI:1.4-4.2]), and WBC (25-100 × 10⁹/L vs. 0-25: aOR=0.9 [95% CI: 0.5-1.9]; >100 × 10⁹/L vs. 0-25: aOR=5.9 [95% CI: 2.1-16.7]). Discussion: Previously these five predictors have been linked to poor HMA ± VEN outcomes. Elevated WBC ≥ 25 × 10⁹/L is linked to a high complication risk, VEN resistance, and early relapse and death (Maiti et al., 2021). KMT2A rearrangements are also associated with poor response rates and inferior survival (Montalban-Braco et al., 2020). Similarly, complex karyotype also confers a poor prognosis (Papaemmanuil et al., 2016). Our data support the existing body of evidence that WBC count, KMT2A, and complex karyotype are associated with poor outcomes. We further add that these are associated with early death within 90-days. Interestingly, TP53 mutations, present in approximately 20% of our cohort, were not predictive of 90-day mortality in this population. As PS correlates with early mortality, most trials exclude patients ≥75 years old with an ECOG PS ≥3 (DiNardo et al., 2019; Wei et al., 2020). However, our data does not identify age as a predictor of outcome; therefore, future prospective clinical trials for older patients should prioritize PS over age for eligibility. Expansion of the dataset is ongoing to enhance model accuracy and evaluate additional predictors, with the aim of improving personalized care and minimizing harm in vulnerable AML populations.

  PublisherOA PDFDOI

• Can Language Models Improve the Performance of SVD-based Recommender Systems?

  Proceedings of the ... International Florida Artificial Intelligence Research Society Conference · 2025-05-14

  articleOpen access1st authorCorresponding

  Traditional recommendation algorithms cannot provide personalized recommendations based on user preferences provided through text, e.g., “I like movies which take me into a dreamland”. Large Language Models(LLMs) have emerged as one of the most promising tools for natural language processing in recent years.This research proposes a framework that leverages the capabilities of LLMs to enhance movie recommendation systems by refining the recommendations of traditional recommendation systems and integrating them with language-based user preference inputs. We employ a Singular Value Decomposition (SVD) algorithm to generate initial movie recommendations. The base SVD algorithm is implemented from the Surprise Python library and trained on the MovieLens 32M dataset.

  PublisherOA PDFDOI

• Abstract A083: BRAF Inhibitors in BRAF V600E-Mutated Solid Tumors: A Real World Single Center Experience Post-Tumor-Agnostic FDA Approval

  Molecular Cancer Therapeutics · 2025-10-22

  article

  Abstract Background: The 2022 FDA tumor-agnostic approval of dabrafenib plus trametinib expanded targeted therapy for BRAF V600E-mutated unresectable or metastatic solid tumors beyond traditional indications. However, real-world evidence, especially in diverse histologies such as CNS, lung, and non-CRC GI cancers, remains limited. Methods: We conducted a retrospective review of patients with BRAF V600E-mutated cancers treated with BRAF inhibitors (BRAFi, often combined with MEK or EGFR inhibitors) at a large academic hospital from 2013 to 2025. Key data collected included demographics, treatment line and duration, radiographic responses (complete response [CR], partial response [PR], stable disease, or progression of disease [POD]), and progression-free survival (PFS, defined as time from BRAFi start to POD or death). We calculated objective response rate (ORR = CR + PR), disease control rate (DCR = ORR + stable disease), and median PFS, both overall and by histology: CNS tumors, lung cancers, GI malignancies (pancreas, sinonasal, neuroendocrine), and colorectal cancer (CRC). Results: Among 30 patients (CNS, n=3; lung, n=7; non-CRC GI, n=3; CRC, n=17), median age was 69 years (range 5-81), with 63% female, 57% deceased, and median BRAFi treatment line of 2 (range 1-5). Regimens primarily involved encorafenib-based combinations (50%) or dabrafenib/trametinib (40%). Overall, ORR was 20%, DCR 57%, median treatment duration 7.5 months, and median PFS 6.9 months (63% events). By histology, outcomes varied markedly: CNS showed robust activity (ORR 67%, DCR 100%, median PFS 22.8 months), suggesting durable BRAF targeting in brain tumors; lung demonstrated moderate efficacy (ORR 43%, DCR 71%, median PFS 10.8 months), indicating potential early-line benefits; non-CRC GI cancers exhibited poor responses (ORR 0%, DCR 0%, median PFS 4.6 months), highlighting limited utility; and CRC had modest control (ORR 6%, DCR 53%, median PFS 6.5 months), reflecting partial synergies with EGFR inhibitors but frequent resistance. Conclusions: This real-world series reveals histology-dependent BRAFi efficacy post-tumor-agnostic approval, with encouraging and prolonged PFS in CNS and lung tumors—potentially guiding expanded use in these contexts - but suboptimal outcomes in GI and CRC, underscoring possible resistance mechanisms and the value of biomarker-driven selection. These insights advocate for larger, prospective registries to refine precision oncology strategies. Citation Format: Sharon Santhosh, Srinivas Govindan, Adit Singhal, Aaron Goldstein, Jervon Wright, Xinhua Zhu. BRAF Inhibitors in BRAF V600E-Mutated Solid Tumors: A Real World Single Center Experience Post-Tumor-Agnostic FDA Approval [abstract]. In: Proceedings of the AACR-NCI-EORTC International Conference on Molecular Targets and Cancer Therapeutics; 2025 Oct 22-26; Boston, MA. Philadelphia (PA): AACR; Mol Cancer Ther 2025;24(10 Suppl):Abstract nr A083.

  PublisherDOI

• A Language-Driven Framework for Improving Personalized Recommendations: Merging LLMs with Traditional Algorithms

  ArXiv.org · 2025-07-09

  preprintOpen access1st authorCorresponding

  Traditional recommendation algorithms are not designed to provide personalized recommendations based on user preferences provided through text, e.g., "I enjoy light-hearted comedies with a lot of humor". Large Language Models (LLMs) have emerged as one of the most promising tools for natural language processing in recent years. This research proposes a novel framework that mimics how a close friend would recommend items based on their knowledge of an individual's tastes. We leverage LLMs to enhance movie recommendation systems by refining traditional algorithm outputs and integrating them with language-based user preference inputs. We employ Singular Value Decomposition (SVD) or SVD++ algorithms to generate initial movie recommendations, implemented using the Surprise Python library and trained on the MovieLens-Latest-Small dataset. We compare the performance of the base algorithms with our LLM-enhanced versions using leave-one-out validation hit rates and cumulative hit rates. Additionally, to compare the performance of our framework against the current state-of-the-art recommendation systems, we use rating and ranking metrics with an item-based stratified 0.75 train, 0.25 test split. Our framework can generate preference profiles automatically based on users' favorite movies or allow manual preference specification for more personalized results. Using an automated approach, our framework overwhelmingly surpassed SVD and SVD++ on every evaluation metric used (e.g., improvements of up to ~6x in cumulative hit rate, ~3.7x in NDCG, etc.), albeit at the cost of a slight increase in computational overhead.

  PublisherOA PDFDOI

• A Rare Case of Argatroban-Induced Anaphylaxis in a Patient With Intermediate-High Risk Pulmonary Embolism

  Cureus · 2024

  • Medicine
  • Cardiology
  • Internal medicine

  Heparin-induced thrombocytopenia is a rare and potentially devastating complication of heparin therapy. Patients with an absolute indication for anticoagulation, such as those with significant pulmonary embolism, must be switched to a different anticoagulant, such as argatroban, a direct thrombin inhibitor. We report a case of anaphylaxis to argatroban in a patient who was initially on heparin for intermediate-high risk pulmonary embolism but developed suspected type II heparin-induced thrombocytopenia. This case highlights the significance of recognizing and treating anaphylactic reactions and the diagnostic challenges associated with heparin-induced thrombocytopenia.

  PublisherOA PDFDOI

• Keratose Hydrogel Drives Differentiation of Cardiac Vascular Smooth Muscle Progenitor Cells: Implications in Ischemic Treatment

  Stem Cell Reviews and Reports · 2023 · 3 citations

  • Medicine
  • Cell biology
  • Cancer research

  signaling is involved in KOS hydrogel-preferred VSMC differentiation and that enhanced blood flow are likely resulted from angiogenesis and/or arteriogenesis induced by transplanted VSMCs.

  PublisherDOI

• Review for "Biomimetic in situ tracheal microvascularization for segmental tracheal reconstruction in one‐step"

  2023-01-09

  peer-review1st authorCorresponding
  PublisherDOI

• Review for "Biomimetic in situ tracheal microvascularization for segmental tracheal reconstruction in one‐step"

  2022-11-17

  peer-review1st authorCorresponding
  PublisherDOI

Recent grants

• NIH Grant R21AR051945

  NIH · $300k · 2008

• NIH Grant R03DE015915

  NIH · $150k · 2007

• Collaborative Research: Ligament Tissue Engineering

  NSF · $229k · 2009–2013

• NIH Grant R21AR055200

  NIH · $365k · 2011

Frequent coauthors

• Patrick Thayer

  15 shared

• Scott A. Guelcher

  Vanderbilt University Medical Center

  12 shared

• Chris A. Bashur

  Florida Institute of Technology

  11 shared

• Brian J. Love

  University of Michigan–Ann Arbor

  11 shared

• Jenni R. Popp

  National Institute of Standards and Technology

  10 shared

• Linda A. Dahlgren

  Virginia–Maryland College of Veterinary Medicine

  9 shared

• Satyavrata Samavedi

  Indian Institute of Technology Hyderabad

  6 shared

• Christine E. Schmidt

  The Graduate Center, CUNY

  6 shared