DMDonald R. Miller

Donald R. Miller

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Donald R. Miller has a total of 226 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Association of Military Branch and Rank With Amyotrophic Lateral Sclerosis Incidence Among United States Veterans

    by Andrea L. Roberts, Alexandre D. Vilela Braga, Ian W. Tang, Ran S. Rotem, Lauren R. Moo, Donald R. Miller, Steven D. Shirk, Marc G. Weisskopf

    Military service is one of the most consistent risk factors for the development of amyotrophic lateral sclerosis (ALS), but little is known about what aspects of military service matter. Understanding the distribution of ALS risk in military personnel may help identify key risk factors. We aimed to ascertain the association of United States (US) military branch and rank with incident ALS. We conducted a longitudinal cohort study including all veterans with a Veterans Health Administration encounter from January 1, 2000, through October 1, 2024, and more than 2 years of follow-up. Incident ALS was identified through medical records. Exposures included military branch (Army, Navy, Air Force, Marine Corps, Coast Guard, or multiple branches), military rank (officer, enlisted, or both), and length of service. Risk was estimated with Cox proportional hazards models, controlled for age, adjusted for race/ethnicity, and stratified by sex. The analytic sample comprised 9,157,938 men (13,935 incident ALS cases, mean age = 56.9 years) and 784,941 women (484 cases, mean age = 42.4 at start of follow up). Among men, compared with service in the Army, service in the Air Force (hazard ratio [HR] = 1.29, 95% CI 1.23-1.35), Navy (HR = 1.15, 95% CI 1.10-1.20), and Coast Guard (HR = 1.26, 95% CI 1.06-1.50) was associated with higher rates of ALS and service in the Marines a lower rate (HR = 0.78, 95% CI 0.68-0.88). Officers had higher rates than enlisted personnel (men, HR = 1.64, 95% CI 1.53-1.74; women, HR = 1.72, 95% CI 1.31-2.27). Longer service was associated with lower rates of ALS. In age-stratified models, differences in hazard rates between Army veterans and Air Force, Navy, and Coast Guard veterans were greatest for the youngest men (17-61 years, HR range = 1.26 to 1.51) and smallest for the oldest (>75-104 years, HR range = 0.99 to 1.12). ALS rates varied considerably by branch and rank in the US military, with larger differences among younger veterans. Officers and those serving in the Air Force, Navy, or Coast Guard may incur exposure to military environments that increase risk of ALS. It is imperative to identify relevant exposures to reduce harm to service members and, potentially, civilians with similar exposures.


  • Open Access

    Detecting Severe Hypoglycemia in Electronic Health Record Notes of Type 2 Diabetes Outpatients: A Hybrid Natural Language Processing Approach (Preprint)

    by Glenn T. Gobbel, Donald R. Miller, Richard Noriega, Sarah E. McDannold, Michael E. Matheny, Ann M. Borzecki

    BACKGROUND Optimal glycemic control is essential for limiting diabetes–related complications, but the risk of hypoglycemia constrains treatment intensification. Current hypoglycemia surveillance methods relying on diagnosis codes substantially underreport events, and improved methods for identifying hypoglycemia from electronic health records (EHRs) are needed. OBJECTIVE To develop and validate an artificial intelligence (AI) model that can use EHR clinical notes to retrospectively identify severe hypoglycemia events among outpatients with type 2 diabetes (T2D) presenting to the Emergency Department (ED) or requiring an Inpatient Admission (IP). METHODS Using Veterans Health Administration data, we identified a sample of T2D patients in 2019 and classified hypoglycemia via ICD-10 codes or outpatient glucose <70 mg/dL (Hypo+) versus controls (Hypo−). Text corpora included clinical notes near ED encounters (±24 hours) or admission (−12 to +24 hours). We created a hybrid classifier to identify hypoglycemic patients by combining rule-based feature extraction with random forest–based machine learning. A rule–based concept extraction module within the RapTAT natural language processing system generated 44 assertion-aware features. Random forest models were developed and trained using these features and leave-one-out cross-validation and then tested on held-out sets. We created one model for hypoglycemia during ED visits and a second for IP admissions with hypoglycemia. Feature importance was evaluated across both models, and performance was evaluated across different cohorts (2018 vs. 2019) and clinical contexts (presence of low blood glucose <70 mg/dL, hypoglycemia treatment). RESULTS When evaluated across all clinical contexts, the ED model achieved a normalized positive predictive value (PPVN) of 82%-88% and sensitivity of 88%-92%, with specificity of 88% (2018) and 80% (2019). Performance was highest in the context of documented low glucose and hypoglycemia treatment. The IP model performed better overall: PPVN 89%–94%, sensitivity 85%-93%, and specificity 93%-94%. Requiring ≥5 documents improved IP accuracy. Feature importance rankings correlated across models (r=0.74); top predictors included counts and ratios of low glucose phrases and document–level coverage of hypoglycemia concepts. Hypoglycemia cases skewed older, male, and had higher representation of African American patients. CONCLUSIONS AI in the form of a hybrid classification model applied to unstructured EHR notes accurately identified severe hypoglycemia in outpatient T2D populations, outperforming historically reported diagnosis code–based sensitivity and specificity. Models were most effective for admission–related hypoglycemia and when corroborating glucose values and treatment were present. These tools enable scalable surveillance, that may inform safer diabetes management, and support quality improvement.

  • Open Access

    Comparative identification of Veterans with mild cognitive impairment or Alzheimer’s disease extracted from clinical notes and administrative data

    by Dan Berlowitz, Byron J. Aguilar, Xuyang Li, Ekaterina Shishova, Guneet Jasuja, Peter J Morin, Donald R Miller, Maureen K. O'COnnor, Andrew H. Nguyen, Raymond Zhang, et al.

    Background: Registries that identify people with mild cognitive impairment (MCI) or Alzheimer’s dementia (AD) will be critical in targeting interventions to delay disease progression. Optimal methods to identify people for inclusion in such registries remain uncertain. We now compare approaches that rely on International Classification of Diseases (ICD)‐10 codes to the use of natural language processing (NLP) of clinicians’ notes.Method: We used data from the Department of Veterans Affairs (VA) considering all Veterans aged 50 or older from the fiscal year 2019. People for the “ICD” cohorts were identified based on at least one code for either AD or MCI. For the “NLP” cohorts, we used electronic clinician notes and previously validated algorithms that search for keywords ("Alz*” or “Mild Cognitive Impairment”), excluding false positive cases identified through text screening such as “family history.” The number and overlap of people identified by the ICD‐ and NLP‐based approaches were compared.Result: Overall, 144,942 Veterans documented with either ICD coding or clinical notes using NLP as probable AD patients were identified; 32,498 (22.4%) people were identified by ICD codes and 141,816 (97.8%) people by clinical notes. Only 20.3% of the Veterans were identified by both clinical notes and ICD; 77.5% were identified by clinical notes alone and only 2.2% by ICD alone. In contrast, for the 134,699 Veterans with MCI, 95,324 (70.8%) people were identified by ICD codes and 79,232 (58.8%) people by clinical notes. Veterans identified by both ICD and clinical notes consisted of 29.6% of the sample, compared to 41.2% by ICD and 29.2% by clinical notes.Conclusion: Identification of patients with MCI or AD in electronic clinical notes using NLP is an alternative approach for developing a disease registry to the traditional methodology based on ICD coding in administrative databases. The differences in case identification between NLP‐ and ICD‐based approach may have reflected a clinical practice pattern in making clinical judgments regarding MCI and AD and documenting MCI and AD assessments in clinical notes as opposed to administrative records among the US Veterans.

  • Open Access

    Estimating the prevalence of mild cognitive impairment from clinical notes in U.S. Veterans

    by Guneet Jasuja, Xuyang Li, Byron J. Aguilar, Ekaterina Shishova, Peter J Morin, Dan Berlowitz, Donald R Miller, Maureen K. O'COnnor, Andrew H. Nguyen, Raymond Zhang, et al.

    Background: Individuals with mild cognitive impairment (MCI), a noticeable decline in cognitive abilities, are at increased risk of developing Alzheimer’s disease (AD) or other dementia. Risk factors for dementia, such as obesity, stroke, depression are highly prevalent among U.S. veterans We report on the trends in the prevalence of MCI using clinical notes over the past 5 years in the Veteran Affairs, the largest integrated healthcare system in the U.S.Method: Prevalent cases of MCI were identified from VA clinical notes for all Veterans aged 50 years and older in VA care from fiscal years (FYs) 2015‐19. VA clinical notes are maintained in searchable text integration utility files (TIU) available in the VA Corporate Data Warehouse (CDW). Pharmacy clinical notes were excluded. “Mild cognitive impairment” keyword was searched in clinical notes to identify patients with MCI. We validated the keyword “mild cognitive impairment” in 199 clinical notes. The positive predictive value for this validation was approximately 91%. For each FY, age and sex adjusted rates directly standardized to the 2010 U.S. Census general population were estimated.Result: A total of 308,930 prevalent cases of MCI were identified from VA clinical notes from FYs 2015‐19. Standardized prevalence rates for MCI increased steeply, with prevalence of 0.9% in FY 2015, increasing to 1.31% in FY 2019. Standardized age adjusted MCI prevalence rates also increased sharply for male (0.77% in FY 2015 to 1.10% in FY 2019) and female veterans (1.01% in FY 2015 to 1.50% in FY 2019) with rates higher for females than males for each FY.Conclusion: We observed recent trends of increasing prevalence rates for MCI among veterans identified from VA clinical notes. Considering the increased risk of dementia among individuals with MCI, our study finding of increasing prevalence of MCI among veterans highlights the critical need for effective interventions and treatment to target veterans with MCI.


  • Open Access

    Diagnosis of Severity of Alzheimer’s Disease from Clinical notes in the Veterans Affairs Healthcare System

    by Byron J. Aguilar, Ekaterina Shishova, Xuyang Li, Andrew H. Nguyen, Peter J Morin, Dan Berlowitz, Maureen K. O'COnnor, Donald R Miller, Guneet Jasuja, Raymond Zhang, et al.

    Background: Alzheimer’s disease (AD) typically progresses in three stages: mild, moderate, and severe. Early detection improves patient care related to identification of suitable interventions and end‐of‐life planning. However, AD severity diagnosis is uncommon. The aim of this study is to identify Veterans in various stages of AD using clinical notes in the Veterans Affairs Healthcare System.Method: Notes containing “Alz*” from fiscal year 2019 were identified then screened for information indicating a clinical diagnosis of probable AD. Screened notes with AD severity keywords (mild, moderate, severe, early, late, advanced) within four words of “Alz*” were considered as a severity diagnosis. Notes containing a single AD severity diagnosis were included in the study.Result: 141,816 Veterans with 547,318 notes containing “Alz*” were identified as probable AD cases. Roughly 9% of notes contained at least one severity keyword and 6% of notes contained a single severity keyword. 34,181 notes with a single severity keyword from 14,148 Veterans were included in this study. The cohort comprised of 96% males and 78% identified as White with a mean age of 79. Ten percent of Veterans with an AD note included severity diagnosis. Approximately 19%, 12%, 18%, 20%, 3%, and 28% of severity notes contained mild, moderate, severe, early, late, and advanced, respectively. The mean age for Veterans with a mild, moderate, or severe AD diagnosis increased from 78 to 81. The mean age for early, late, and advanced AD patients increased from 74, 79, and 81, respectively. AD severity diagnosis was found in notes related to primary care (11%), internal medicine (10%), psychiatry (8%), mental health (7%), neurology (6%), geriatric (5%), neuropsychology (1%), and psychology (1%).Conclusion: AD severity diagnosis is uncommon. Identification of patients in various stages of AD will lead to a better understanding of transition from one severity to another and facilitate the coordination of services to improve patient care. Specifically, identification of AD patients in the early stage may identify candidates for novel therapeutics focused on delaying AD progression.

  • Open Access

    Development and validation of a natural language processing system that extracts cognitive test results from clinical notes

    by Jinying Chen, Xuyang Li, Byron J. Aguilar, Ekaterina Shishova, Peter J Morin, Dan Berlowitz, Donald R Miller, Maureen K. O'COnnor, Andrew H. Nguyen, Raymond Zhang, et al.

    Background: Cognitive test results from electronic health records (EHRs) are key information for assessing the severity and progression of patients with mild cognitive impairment (MCI) and Alzheimer’s’ disease (AD). However, such information is often recorded in unstructured clinical notes rather than in an administrative database. We developed and validated a natural language processing (NLP) system to extract cognitive test results from clinical notes in the Veterans Affair (VA) Healthcare System.Method: An NLP system was developed using regular expression‐based rules and Python to extract results for six tests that have been used most frequently in VA: Mini‐Mental State Exam (MMSE), Montreal Cognitive Assessment (MoCA), Saint Louis University Mental Status Examination (SLUMS), Mini‐cog, Boston Naming Test (BNT), and Benton Visual Retention Test (BVRT). The system extracted test results from each note in two steps: (1) searched a test name, or variations and abbreviation of the test name and, if successful, (2) searched the quantitative results (e.g., 12/30 for MMSE, 74th percentile for BNT) and/or descriptive results (e.g., “borderline impairment” for BVRT) within 5 words or one sentence before or after the test name. To balance the system performance and speed, we developed 3‐8 extraction rules per test based on a manual review of 30‐50 notes for each test. We further validated NLP performance on 6 held‐out datasets (50‐200 notes/test). We automatically sampled the development and held‐out notes by searching the test name and its variations/abbreviation to increase positive cases, i.e., notes that contained the above test results.Result: The NLP system achieved 0.72‐0.92 predictive positive values (PPV), 0.96‐1.00 recall, and 0.83‐0.95 F1 scores on the validation test sets (Table 1). In addition, it demonstrated a scalable performance (processed 200,000 notes in 7 min), allowing an extraction of millions of notes from ∼1 million patients within hours.Conclusion: Rule‐based NLP can extract cognitive test results with adequate performance and scalable capability for clinical notes from patients with MCI or AD within the VA Healthcare System.

  • Top Cited Papers
    Open Access

    Genomics and phenomics of body mass index reveals a complex disease network

    by Jie Huang, Jennifer E. Huffman, Yunfeng Huang, Ítalo Do Valle, Themistocles L. Assimes, Sridharan Raghavan, Benjamin F. Voight, Chang Liu, Albert-László Barabási, Rose D. L. Huang, et al.

    Elevated body mass index (BMI) is heritable and associated with many health conditions that impact morbidity and mortality. The study of the genetic association of BMI across a broad range of common disease conditions offers the opportunity to extend current knowledge regarding the breadth and depth of adiposity-related diseases. We identify 906 (364 novel) and 41 (6 novel) genome-wide significant loci for BMI among participants of European (N~1.1 million) and African (N~100,000) ancestry, respectively. Using a BMI genetic risk score including 2446 variants, 316 diagnoses are associated in the Million Veteran Program, with 96.5% showing increased risk. A co-morbidity network analysis reveals seven disease communities containing multiple interconnected diseases associated with BMI as well as extensive connections across communities. Mendelian randomization analysis confirms numerous phenotypes across a breadth of organ systems, including conditions of the circulatory (heart failure, ischemic heart disease, atrial fibrillation), genitourinary (chronic renal failure), respiratory (respiratory failure, asthma), musculoskeletal and dermatologic systems that are deeply interconnected within and across the disease communities. This work shows that the complex genetic architecture of BMI associates with a broad range of major health conditions, supporting the need for comprehensive approaches to prevent and treat obesity.


  • Open Access

    Factors Associated With Declining Lung Cancer Screening After Discussion With a Clinician in a Cohort of US Veterans

    by Eduardo R. Núñez, Tanner J. Caverly, Sanqian Zhang, Mark E. Glickman, Shirley X. Qian, Jacqueline H. Boudreau, Donald R. Miller, Christopher G. Slatore, Renda Soylemez Wiener

    Lung cancer is the leading cause of death from cancer worldwide, and lung cancer screening (LCS) has the potential to reduce mortality from lung cancer by approximately 20% among individuals at high risk.1,2 Lung cancer screening has been recommended by the US Preventive Services Task Force since 2013.3,4 However, fewer than 10% of eligible individuals have been screened, with lower rates of LCS in racial and ethnic minority populations.5-8


  • Top Cited Papers

    Large-scale genome-wide association study of coronary artery disease in genetically diverse populations

    by Catherine Tcheandjieu, Xiang Zhu, Austin T. Hilliard, Shoa L. Clarke, Valerio Napolioni, Shining Ma, Kyung Min Lee, Huaying Fang, Fei Chen, Yingchang Lu, et al.

    We report a genome-wide association study (GWAS) of coronary artery disease (CAD) incorporating nearly a quarter of a million cases, in which existing studies are integrated with data from cohorts of white, Black and Hispanic individuals from the Million Veteran Program. We document near equivalent heritability of CAD across multiple ancestral groups, identify 95 novel loci, including nine on the X chromosome, detect eight loci of genome-wide significance in Black and Hispanic individuals, and demonstrate that two common haplotypes at the 9p21 locus are responsible for risk stratification in all populations except those of African origin, in which these haplotypes are virtually absent. Moreover, in the largest GWAS for angiographically derived coronary atherosclerosis performed to date, we find 15 loci of genome-wide significance that robustly overlap with established loci for clinical CAD. Phenome-wide association analyses of novel loci and polygenic risk scores (PRSs) augment signals related to insulin resistance, extend pleiotropic associations of these loci to include smoking and family history, and precisely document the markedly reduced transferability of existing PRSs to Black individuals. Downstream integrative analyses reinforce the critical roles of vascular endothelial, fibroblast, and smooth muscle cells in CAD susceptibility, but also point to a shared biology between atherosclerosis and oncogenesis. This study highlights the value of diverse populations in further characterizing the genetic architecture of CAD.