About

Thom Miller received his BA in Theatre and MFA in Playwriting from Southern Illinois University, Carbondale. He also holds a professional certificate in screenwriting from the UCLA. As a writer, his first full-length play, "Three,” won the Christian H. Moe award. His plays Untitled/Love/Affair and The Audition had their world premieres at the Pittsburgh New Works Festival. He has twice participated in the Kennedy Center Playwright’s Intensive and served as the New Play Program Chair for Region III of the Kennedy Center American College Theatre Festival. Currently, he is the Theatre Studies BFA Coordinator for the Department of Theatre at the University of Illinois Urbana-Champaign and has received the FAA Specialized Faculty Award for Excellence. He teaches courses including Introduction to Theatre Arts, Text to Stage, Introduction to Playwriting, Senior Projects, Introduction to Screenwriting, Playwrights Workshop, and Advanced Screenwriting, and is the producer of Illinois Theatre.

Research topics

• Data Mining
• Computer Science
• Machine Learning
• Natural Language Processing
• Artificial Intelligence
• Nuclear physics
• Algorithm
• Statistics
• Speech recognition
• Physics
• Optics
• Engineering
• Atomic physics
• Mathematics

Selected publications

• Cross-site predictions of readmission after psychiatric hospitalization with mood or psychotic disorders (Preprint)

  2025-01-22

  preprintOpen access

  <sec> <title>BACKGROUND</title> Patients with mood or psychotic disorders experience high rates of unplanned hospital readmissions. Predicting the likelihood of readmission can guide discharge decisions and optimize patient care. </sec> <sec> <title>OBJECTIVE</title> The purpose of this study is to evaluate the predictive power of structured variables from electronic health records (EHRs) for all-cause readmission across multiple sites within the Mass General Brigham (MGB) health systems and to assess the transportability of prediction models between sites. </sec> <sec> <title>METHODS</title> This retrospective, multi-site study analyzed structured variables from EHRs separately for each site to develop in-site prediction models. The transportability of these models was evaluated by applying them across different sites. The predictive performance was measured using the F1 score, and additional adjustments were made to account for differences in predictor distributions. </sec> <sec> <title>RESULTS</title> The study found that the relevant predictors of readmission varied significantly across sites. For instance, the length of stay was a strong predictor at only three of the four sites. In-site prediction models achieved an average F1 score of 0.666, whereas cross-site predictions resulted in a lower average F1 score of 0.551. Efforts to improve transportability by adjusting for differences in predictor distributions did not lead to better performance. </sec> <sec> <title>CONCLUSIONS</title> The findings indicate that individual site-specific models are necessary to achieve reliable prediction accuracy. Furthermore, the results suggest that the current set of predictors may be insufficient for cross-site model transportability, highlighting the need for more advanced predictor variables and predictive algorithms to gain robust insights into the factors influencing early psychiatric readmissions. </sec>

  PublisherDOI

• Aspect-Oriented Summarization for Psychiatric Short-Term Readmission Prediction

  arXiv (Cornell University) · 2025-02-14

  preprintOpen accessSenior author

  Recent progress in large language models (LLMs) has enabled the automated processing of lengthy documents even without supervised training on a task-specific dataset. Yet, their zero-shot performance in complex tasks as opposed to straightforward information extraction tasks remains suboptimal. One feasible approach for tasks with lengthy, complex input is to first summarize the document and then apply supervised fine-tuning to the summary. However, the summarization process inevitably results in some loss of information. In this study we present a method for processing the summaries of long documents aimed to capture different important aspects of the original document. We hypothesize that LLM summaries generated with different aspect-oriented prompts contain different information signals, and we propose methods to measure these differences. We introduce approaches to effectively integrate signals from these different summaries for supervised training of transformer models. We validate our hypotheses on a high-impact task -- 30-day readmission prediction from a psychiatric discharge -- using real-world data from four hospitals, and show that our proposed method increases the prediction performance for the complex task of predicting patient outcome.

  PublisherOA PDFDOI

• Cross-site predictions of readmission after psychiatric hospitalization with mood or psychotic disorders

  medRxiv · 2024-08-26

  preprintOpen accessCorresponding

  Abstract Patients with mood or psychotic disorders have high rates of unplanned readmission, and predicting readmission likelihood may guide discharge decisions. In this retrospective, multi-site study, we assess the predictive power of various structured variables from electronic health records for all-cause readmission in each site separately and evaluate the generalizability of the in-site prediction models across sites. We find that the set of relevant predictors vary significantly across. For example, length of stay is strongly predictive of readmission at only three out of the four sites. We also find a general lack of cross-site generalizability of the in-site prediction models, with in-site predictions having an average F1 score of 0.666, compared to an average F1 score of 0.551 for cross-site predictions. The generalizability cannot be improved even after adjusting for differences in the distributions of predictors. These results indicate that, with this set of predictors, fitting individual models at each site is necessary to achieve reasonable prediction accuracy. Additionally, they suggest that more sophisticated predictors variables or predictive algorithms are needed to develop generalizable models capable of extracting robust insights into the root causes of early psychiatric readmissions.

  PublisherOA PDFDOI

• Lessons Learned on Information Retrieval in Electronic Health Records: A Comparison of Embedding Models and Pooling Strategies

  arXiv (Cornell University) · 2024-09-23

  preprintOpen access

  Objective: Applying large language models (LLMs) to the clinical domain is challenging due to the context-heavy nature of processing medical records. Retrieval-augmented generation (RAG) offers a solution by facilitating reasoning over large text sources. However, there are many parameters to optimize in just the retrieval system alone. This paper presents an ablation study exploring how different embedding models and pooling methods affect information retrieval for the clinical domain. Methods: Evaluating on three retrieval tasks on two electronic health record (EHR) data sources, we compared seven models, including medical- and general-domain models, specialized encoder embedding models, and off-the-shelf decoder LLMs. We also examine the choice of embedding pooling strategy for each model, independently on the query and the text to retrieve. Results: We found that the choice of embedding model significantly impacts retrieval performance, with BGE, a comparatively small general-domain model, consistently outperforming all others, including medical-specific models. However, our findings also revealed substantial variability across datasets and query text phrasings. We also determined the best pooling methods for each of these models to guide future design of retrieval systems. Discussion: The choice of embedding model, pooling strategy, and query formulation can significantly impact retrieval performance and the performance of these models on other public benchmarks does not necessarily transfer to new domains. Further studies such as this one are vital for guiding empirically-grounded development of retrieval frameworks, such as in the context of RAG, for the clinical domain.

  PublisherOA PDFDOI

• Prevalence and clinicopathological features of incidentally detected TRBC1-dim populations in peripheral blood flow cytometry

  Leukemia & lymphoma/Leukemia and lymphoma · 2024-05-15 · 1 citations

  article
  PublisherDOI

• Clinicopathologic features of human monkeypox lymphadenitis

  Histopathology · 2023-02-03 · 2 citations

  letter

  To the Editor: In July 2022, an outbreak of human monkeypox was reported in several nonendemic countries. Even though enlarged lymph nodes and tonsils are commonly identified in patients diagnosed with monkeypox,1, 2 descriptions of lymph node pathology in human monkeypox infection are lacking. Herein, we present the first pathologic description of monkeypox lymphadenitis in humans. A 46-year-old man presented with progressive throat pain, dysphagia, and dyspnea for 2 weeks, with subsequent development of a vesiculopustular rash on his trunk and upper extremities. His medical history was notable for HIV-1 infection with a CD4 count of 816 cells/μl and undetectable HIV-1 RNA in plasma. He had an enlarged right tonsil for several months and recent unprotected receptive oral sex with another man. Nasolaryngoscopic examination revealed a profoundly enlarged right tonsil covered in yellow exudate and bilateral cervical lymphadenopathy, which were confirmed on computed tomography (Figure 1A,B). The patient was admitted to intensive care for airway observation and further diagnostic workup. A skin lesion swab screened positive for monkeypox virus DNA using a nonvariola Orthopoxvirus quantitative polymerase chain reaction (qPCR), and infection was confirmed with a separate clade 2/3 (West African) monkeypox virus qPCR.3, 4 At the time, malignancy was suspected clinically, as the patient's throat pain and dysphagia preceded any skin manifestations and in part due to the lack of institutional experience with monkeypox infection. Fine-needle aspiration (FNA) of the dominant right level 2A lymph node was performed, revealing a predominance of small to intermediate-sized mature lymphocytes, scattered plasma cells, and foci of necrosis (Figure 1C–E). By immunohistochemistry, scattered CD30+ immunoblasts were present (Figure 1F). TdT and HHV-8 stains and in situ hybridization for EBV-encoded RNA (EBER) were negative. Flow cytometry detected, of all CD45+ events, 44% CD3+ T-cells, 36% CD19+ B-cells, and fewer granulocytes, NK-cells, plasma cells, and monocytes. B-cells and plasma cells showed a skewed kappa:lambda ratio of approximately 1:1. Plasma cells uniformly expressed CD117 and demonstrated downregulation of CD27 and CD81. T- and NK-cells showed no immunophenotypic abnormalities. Representative flow cytometry plots are shown in Figure 2A. The lymph node aspirate was monkeypox DNA-positive via nonvariola Orthopoxvirus qPCR and clade 2/3 monkeypox virus qPCR (Figure 2B), supporting a diagnosis of necrotizing lymphadenitis secondary to monkeypox infection. Given the initial uncertainty of the diagnosis and prognosis, intubation and tracheostomy were considered, but neither was ultimately necessary. The patient was treated with dexamethasone and tecovirimat while inpatient and was provided an additional 9-day course of tecovirimat at discharge (to complete 14 days total). While the tonsillar hypertrophy persisted, the dysphonia and odynophagia improved after dexamethasone treatment. The patient was offered follow-up with Otolaryngology and Infectious Diseases, but did not return following discharge. The variola (smallpox) and monkeypox viruses are related viruses in the Orthopoxvirus genus and have similar cutaneous disease presentation and progression. While mucosal involvement of smallpox is typically described as an enanthem of the oral cavity, monkeypox has been associated with oropharyngotonsillitis.1, 5 Additionally, prominent peripheral lymphadenopathy (particularly cervical and inguinal) has been cited as the singular clinical feature that may distinguish monkeypox from smallpox.6 To our knowledge, this is the first description of lymph node pathology in human monkeypox infection and contributes to the list of possible causes of necrotizing lymphadenitis. This case also illustrates that monkeypox lymphadenitis can be associated with abnormal flow cytometry findings, including lambda-skewed B-cells and plasma cells and aberrant expression of CD117 on plasma cells. However, light chain-skewing without monotypia and CD117 expression on plasma cells are not considered specific markers of malignancy and, in this setting, are consistent with the reactive etiology. Overall, it is important for both the treating clinician and pathologist to recognize necrotizing lymphadenitis as a manifestation of human monkeypox infection, especially if lymphadenopathy precedes the rash or if the rash is not disseminated. Furthermore, this case supports FNA as a suitable diagnostic modality for confirming monkeypox lymphadenitis and excluding other diagnostic possibilities, such as malignancy. ZJQ, RG, and HDL conceived, drafted, critically revised, and approved the final version of the report. DSM, KW, and SK critically revised the clinical portions of the report and approved the final version. TM and OS critically revised the hematopathology portions of the report and approved the final version. VN and BAP provided substantial contributions to the acquisition of qPCR data, critically revised the virology portions of the report, and approved the final version. The authors have no conflicts of interest to declare. Institutional Review Board approval is not required for this case report, as it is not considered research. Informed consent is not required for this case report, as no identifiable information is included. Data sharing not applicable to this article as no datasets were generated or analysed during the current study.

  PublisherDOI

• A 1.9GHz 0.57V Vmin 576Kb embedded product-ready L2 cache in 5nm FinFET technology

  2023-06-11 · 5 citations

  article

  A product-ready L2 cache (L2C) design based on 6T ultra-dense SRAM cells with novel circuits capable of boosting word-line, cell, and, bit-line supplies independently using single supply and metal coupling capacitance is demonstrated for the first time in 5nm technology. A metal short detection circuit is provided to increase the robustness of the design. Hardware data shows that L2C operates with a minimum supply of 0.57V and reaches a maximum operating frequency of 1.9GHz at 1.1V.

  PublisherDOI

• Natural Language Processing to Automatically Extract the Presence and Severity of Esophagitis in Notes of Patients Undergoing Radiotherapy

  JCO Clinical Cancer Informatics · 2023-07-01 · 19 citations

  articleOpen access

  PURPOSE: Radiotherapy (RT) toxicities can impair survival and quality of life, yet remain understudied. Real-world evidence holds potential to improve our understanding of toxicities, but toxicity information is often only in clinical notes. We developed natural language processing (NLP) models to identify the presence and severity of esophagitis from notes of patients treated with thoracic RT. METHODS: Our corpus consisted of a gold-labeled data set of 1,524 clinical notes from 124 patients with lung cancer treated with RT, manually annotated for Common Terminology Criteria for Adverse Events (CTCAE) v5.0 esophagitis grade, and a silver-labeled data set of 2,420 notes from 1,832 patients from whom toxicity grades had been collected as structured data during clinical care. We fine-tuned statistical and pretrained Bidirectional Encoder Representations from Transformers-based models for three esophagitis classification tasks: task 1, no esophagitis versus grade 1-3; task 2, grade ≤1 versus >1; and task 3, no esophagitis versus grade 1 versus grade 2-3. Transferability was tested on 345 notes from patients with esophageal cancer undergoing RT. RESULTS: Fine-tuning of PubMedBERT yielded the best performance. The best macro-F1 was 0.92, 0.82, and 0.74 for tasks 1, 2, and 3, respectively. Selecting the most informative note sections during fine-tuning improved macro-F1 by ≥2% for all tasks. Silver-labeled data improved the macro-F1 by ≥3% across all tasks. For the esophageal cancer notes, the best macro-F1 was 0.73, 0.74, and 0.65 for tasks 1, 2, and 3, respectively, without additional fine-tuning. CONCLUSION: To our knowledge, this is the first effort to automatically extract esophagitis toxicity severity according to CTCAE guidelines from clinical notes. This provides proof of concept for NLP-based automated detailed toxicity monitoring in expanded domains.

  PublisherOA PDFDOI

• Natural language processing to automatically extract the presence and severity of esophagitis in notes of patients undergoing radiotherapy

  arXiv (Cornell University) · 2023-03-24

  preprintOpen access

  Radiotherapy (RT) toxicities can impair survival and quality-of-life, yet remain under-studied. Real-world evidence holds potential to improve our understanding of toxicities, but toxicity information is often only in clinical notes. We developed natural language processing (NLP) models to identify the presence and severity of esophagitis from notes of patients treated with thoracic RT. We fine-tuned statistical and pre-trained BERT-based models for three esophagitis classification tasks: Task 1) presence of esophagitis, Task 2) severe esophagitis or not, and Task 3) no esophagitis vs. grade 1 vs. grade 2-3. Transferability was tested on 345 notes from patients with esophageal cancer undergoing RT. Fine-tuning PubmedBERT yielded the best performance. The best macro-F1 was 0.92, 0.82, and 0.74 for Task 1, 2, and 3, respectively. Selecting the most informative note sections during fine-tuning improved macro-F1 by over 2% for all tasks. Silver-labeled data improved the macro-F1 by over 3% across all tasks. For the esophageal cancer notes, the best macro-F1 was 0.73, 0.74, and 0.65 for Task 1, 2, and 3, respectively, without additional fine-tuning. To our knowledge, this is the first effort to automatically extract esophagitis toxicity severity according to CTCAE guidelines from clinic notes. The promising performance provides proof-of-concept for NLP-based automated detailed toxicity monitoring in expanded domains.

  PublisherOA PDFDOI

• Correction to “Imaging Nanometer Phase Coexistence at Defects During the Insulator–Metal Phase Transformation in VO₂ Thin Films by Resonant Soft X-ray Holography”

  Nano Letters · 2021-08-19

  articleOpen access

  In our original Letter, we used spectral holography to identify the growth of metallic domains in insulating VO2. In this work, we identified three phases based on spectroscopy: M1, M2, and R. However, spectroscopy based on the holographic method, as implemented in our paper did not give the true spectra of the sample due to the presence of the beam block. Whereas the M1 and R phases could be assigned based on the low-temperature and high-temperature states, assignment to the M2 phase was speculative. We have recently performed quantitative coherent diffractive imaging with initial holographic phasing on the same sample including low momentum transfer information,1 which enabled us to extract the actual spectra of the sample at all points in space. In doing this, we discovered that the observation that was assigned to the M2 phase is, in fact, an artifact. The origin of the artifact is discussed in detail in ref 1 but relates to an ambiguity in determining the reconstruction plane when the exact structure of the material is unknown. As a result, no M2 phase was observed in this sample, and the regions that were assigned M2 are, in fact, M1.

  PublisherOA PDFDOI

Recent grants

• Automated domain adaptation for clinical natural language processing

  NIH · $1.2M · 2018–2023

Frequent coauthors

• T.‐C. Chiang

  114 shared

• Aaron M. Lindenberg

  SLAC National Accelerator Laboratory

  41 shared

• Joshua S. Wittenberg

  SLAC National Accelerator Laboratory

  32 shared

• Erzsi Szilagyi

  Stanford University

  31 shared

• J. Safranek

  SLAC National Accelerator Laboratory

  27 shared

• David A. Reis

  Stanford University

  27 shared

• Jeff Corbett

  SLAC National Accelerator Laboratory

  27 shared

• Mariano Trigo

  26 shared

Education

• Ph.D., Theatre

  University of Illinois at Urbana-Champaign

  2000

• M.A., Theatre

  University of Illinois at Urbana-Champaign

  1996

• B.A., Theatre

  University of Illinois at Urbana-Champaign

  1994

Awards & honors

• FAA Specialized Faculty Award for Excellence
• Christian H. Moe award