About

Kevin Sullivan is an Associate Professor at the University of Virginia School of Engineering and Applied Science. He obtained his background in computer science from Tufts University in 1987, working closely with David Krumme, and completed his graduate studies at the University of Washington, earning both his MS and PhD in 1994 under the supervision of David Notkin. His graduate work focused on the evolvability and evolution of software, emphasizing the importance of design modularity and the challenges of maintaining software properties in a 'soft' medium. Since joining UVa as an Assistant Professor of Computer Science in 1994, Sullivan has worked extensively in areas including software evolvability, value-based software engineering, and formal methods for software and systems assurance. His research explores the options value of modularity in design, reconciling the potential of aspect-oriented programming with the need for abstraction and information hiding to preserve evolvability. He is also engaged in establishing a discipline of cyber-social learning systems, aiming to inform the design of 21st-century service systems across sectors such as healthcare, education, and defense. Sullivan's overarching goal is to catalyze a self-regenerative transformation of critical societal sectors into more effective cyber-social learning systems.

Research topics

• Medicine
• Internal medicine
• Intensive care medicine
• Gerontology
• Geography
• Psychology
• Neuroscience
• Audiology
• Computational biology
• Biology
• Genetics
• Immunology
• Cartography
• Pediatrics

Selected publications

• Elevated Rack Height to Control Biofouling on an Intertidal Oyster Farm: Efficacy and Economics

  Journal of Shellfish Research · 2026-01-03

  articleSenior author

  Biofouling is an ever-present problem for shellfish farmers. Among the many biofouling agents are mud worms, genus Polydora. Mud worms can cause significant economic loss on oyster farms. This project evaluated the efficacy of elevated rack height as a strategy to control mud worm biofouling on an intertidal oyster farm employing rack and bag culture methods. During the study oyster production and cost data were collected for oysters grown at each rack height (15″, 20″, and 30″) in an experimental farm field trial. This information, along with past business records for the rack and bag farm operation, was used to inform an economic cost model. Rack height had a significant effect on oyster growth, mortality, and mud worm fouling during the experiment with higher growth, mortality, and fouling with decreasing rack heights. Oyster condition, shell strength, shell height, shell width, and shell depth did not significantly differ among rack heights. Economic analyses indicated that each rack height is practicable and can be financially viable. Some labor cost was saved by using elevated racks because of reduced biofouling; however, labor increased overall because of the increased time to grow-out on the elevated racks. Higher capital and labor costs, and lower growth rates associated with higher rack heights, were more than offset by increases in survival, thus increasing overall revenue and profits when using higher rack heights. The economic model presented here suggests that even small improvements in survival can greatly improve profitability. Conversely, small reductions in survival could lead to consequential cash flow problems. This study combined experimental results and economic modeling to demonstrate that for this intertidal farm increasing the height of oyster racks might prove an effective strategy to control fouling pests.

  PublisherDOI

• Effects of Hearing Intervention on Blood-Based Biomarkers of Neurodegeneration: A Secondary Analysis of the ACHIEVE Randomised Controlled Trial

  SSRN Electronic Journal · 2026-01-01

  preprintOpen access
  PublisherDOI

• Mid-Life Vascular Risk and Rate of Physical Function Decline Among Older Adults: The Atherosclerosis Risk in Communities (ARIC) Study

  UNC Libraries · 2026-04-22

  articleOpen access

  BACKGROUND: Physical function and its decline in older age may be connected to treatable vascular risk factors in mid-life. This study aimed to evaluate whether these factors affect the underlying rate of decline. METHODS: This prospective cohort included 5 481 older adults aged 67-91 in the Atherosclerosis Risk in Communities Study (mean [standard deviation {SD}] age = 75.8 [5.0], 58% women, 21% Black race) without a history of stroke. The main outcome was the rate of Short Physical Performance Battery (SPPB) decline over a median late-life follow-up of 4.8 years. Primary mid-life (aged 45-64) exposures were Visit 1 hypertension (>140/90 mm Hg or treatment), diabetes (>126 mg/dL or treatment), high cholesterol (>240 mg/dL or treatment), and smoking, and number of decades of vascular risk exposure across Visits 1-4. RESULTS: The average adjusted rate of SPPB decline (points per 5 years) for older adults was -0.79 (confidence interval [CI]: -0.87, 0.71) and was accelerated by mid-life hypertension (+57% decline vs normotension: additional decline of -0.47, 95% CI: -0.64, -0.30), diabetes (+73% decline vs no diabetes: additional decline of -0.67, 95% CI: -1.09, -0.24), elevated systolic blood pressure (+17% decline per SD: -0.16, 95% CI: -0.23, -0.10), and elevated fasting blood glucose (+16% decline per SD: -0.015, 95% CI: -0.24, -0.06). Each decade greater mid-life exposure to hypertension (+32% decline: -0.93, 95% CI: -1.25, -0.61) and diabetes (+35% decline: -1.03, 95% CI: -1.68, -0.38) was associated with faster SPPB decline. CONCLUSIONS: Mid-life control of blood pressure and diabetes may offset aging-related functional decline.

  PublisherDOI

• Moderating Effects of Endurance on Gait Speed Associations With Incident Mild Cognitive Impairment and Dementia

  Innovation in Aging · 2025-12-01

  articleOpen access

  Abstract Although slower gait speed is associated with cognitive impairment, endurance may be more predictive in older adults with faster gait speed. We examined moderating effects of endurance on the relationship of gait speed with incident mild cognitive impairment (MCI) or dementia in cognitively normal ARIC participants (n = 2242, mean age=78.6 + 4.3, 58% female, 19% Black). At the sixth exam, endurance was assessed as distance walked at a fast pace (Two Minute Walk, 2MW); usual 4-meter gait speed was also assessed. Cognitive status was adjudicated during the 6th-9th exams (mean follow-up 5.1 years). Multinomial logistic regression models with a 2MW-x-gait speed interaction term estimated relative risk ratios (RRR) of MCI and dementia relative to normal cognition per SD higher gait speed or 2MW distance, adjusting for demographic and cardiovascular factors. Results used representative continuous gait speed and continuous 2MW distance values. Longer 2MW distance was associated with lower risk of MCI and dementia at gait speed <1m/s [e.g. gait speed=0.8: MCI RRR=0.59, 95%CI(0.42, 0.83); dementia RRR=0.29, (0.13, 0.63)] and with dementia at gait speed>1m/s [gait speed=1.2: RRR=0.28, (0.09, 0.81)], but not with MCI (RRR=0.85, (0.57, 1.26)]. Conversely, faster gait speed was associated with lower risk of MCI at low 2MW [100 m; RRR=0.84, (0.74, 0.94)], but was attenuated at high 2MW (150 m; RRR=0.91, (0.84, 1.00)]. A 2MW-x-gait speed interaction term was supported (p = 0.04), primarily driven by differential MCI associations. Combining endurance and gait speed may be especially helpful in identifying cognitively normal older adults at risk for impending early-stage cognitive impairment.

  PublisherOA PDFDOI

• A Learning Curve is Associated With Combined Hybrid Procedure and Single Ventricle-Ventricular Assist Device Insertion in Neonates With Hypoplastic Left Heart Syndrome

  World Journal for Pediatric and Congenital Heart Surgery · 2025-09-24

  article

  Objectives A minority of patients with hypoplastic left heart syndrome (HLHS) are at extremely high risk for staged palliation and can be bridged-to-heart transplantation with bilateral pulmonary artery bands, ductal stenting, and single ventricle-ventricular assist device insertion (HYBRID + sVAD). The purpose of this analysis is to assess our learning curve associated with our first ten patients with functionally univentricular ductal-dependent systemic circulation who were supported with primary HYBRID + sVAD as bridge-to-heart transplantation. Methods Patients were temporally separated into two cohorts: the first five and second five. Demographic, perioperative, and outcome data were collected. Continuous variables are described as median [IQR](range). Categorical variables are described as N (%). P values were calculated using Fisher exact t test for categorical variables and unpaired t tests for continuous variables. Results Ten patients underwent HYBRID + sVAD operations for HLHS (2017-2022). Patients in the initial cohort and the most recent cohort were similar in age and weight. Liver dysfunction and renal dysfunction were more common in the first five patients (2/5 = 40%) versus the next five patients (0/5 = 0%). Length of sVAD support was longer in the most recent five patients (98 days [64-138] vs 154 days [134-225], P = .08); however, no increase in sVAD-associated stroke or bleeding was seen in the most recent five patients. Despite very similar demographic and preoperative profiles, only two of the first five patients (2/5 = 40%) survived to heart transplantation, while all of the next 5 (5/5 = 100%) were successfully bridged-to-cardiac transplantation with HYBRID + sVAD and are alive today. Conclusions Our experience with primary HYBRID + sVAD as bridge-to-heart transplantation in neonates with HLHS demonstrates an important learning curve associated with this operation and approach.

  PublisherDOI

• Effects of hearing intervention on falls in older adults: findings from a secondary analysis of the ACHIEVE randomised controlled trial

  The Lancet Public Health · 2025-05-27

  articleOpen access

  BACKGROUND: Hearing loss is highly prevalent among older adults and has been associated with an increased likelihood of falling. We aimed to examine the effect of a hearing intervention on falls over 3 years among older adults in a secondary analysis of the ACHIEVE study. METHODS: The Aging and Cognitive Health Evaluation in Elders (ACHIEVE) study was a 3-year, unmasked, randomised controlled trial of adults aged 70-84 years at enrolment with untreated hearing loss and without substantial cognitive impairment. Participants were recruited at four US community-based field sites from two study populations: (1) an ongoing observational study of cardiovascular health (Atherosclerosis Risk in Communities [ARIC] study), and (2) de novo from the community. Participants were randomly assigned (1:1) to a hearing intervention (audiological counselling and provision of hearing aids) or a health education control (didactic education and enrichment activities covering chronic disease prevention topics). A prespecified exploratory outcome was falls. Self-reported falls in the past 12 months were assessed at baseline and annually for 3 years, and analysed by intention to treat with covariate adjustment. The study was registered with ClinicalTrials.gov, NCT03243422, and is completed. FINDINGS: Between Nov 9, 2017, and Oct 25, 2019, 3004 individuals were screened for eligibility and 977 (238 [24%] from the ARIC study and 739 [76%] de novo) were randomly assigned, with 490 (50%) in the hearing intervention group and 487 (50%) in the health education control group. Overall mean age was 76·8 years (SD 4·0), 523 (54%) participants were female and 454 (46%) were male, and 112 (11%) were Black, 858 (88%) were White, and seven (1%) were other race. In adjusted analyses, the intervention group had a 27% reduction in the mean number of falls over 3 years compared with the control group (intervention group: 1·45 [95% CI 1·28 to 1·61]; control group: 1·98 [1·82 to 2·15]; mean difference: -0·54 [95% CI -0·77 to -0·31]). This 3-year effect of hearing intervention was consistent across both the ARIC and de novo study populations. INTERPRETATION: Hearing intervention versus a health education control was associated with a reduction in the mean number of falls over 3 years in older adults. Ongoing follow-up of ACHIEVE participants in a separate follow-up study (NCT05532657) will enable examination of the longer term effects of hearing intervention on falls. FUNDING: US National Institutes of Health.

  PublisherOA PDFDOI

• Endurance and Gait Speed Associations with Incident Dementia in Older Adults: The ARIC Study

  Innovation in Aging · 2025-12-01

  articleOpen access

  Abstract Faster gait speed and better endurance have been cross-sectionally and jointly associated with a lower likelihood of dementia. However, gait speed associations plateau above 1m/s, suggesting gait speed may be less predictive at near normal speeds. This study examined prospective associations of gait speed, endurance, and moderating effects of endurance on gait speed associations with incident dementia among ARIC participants without dementia at Visit 6 (2016-2017) who completed assessments of endurance ( 2- Minute Walk (2MW), meters) and usual-pace 4-meter gait speed (m/s) (n = 2767, mean age 78+/- 4.3 years, 43% male, 18% Black). Dementia was adjudicated using neuropsychological tests, functional assessments, informant interviews, and medical/vital record surveillance over a median of 3.3 years follow-up. Cox regression models estimated the hazard ratio (HR, 95% CI) for dementia, incorporating gait speed, 2MW, gait speed-by-2MW interaction, demographic, and cardiovascular factors. In separate models, each standard deviation higher 2MW [HR = 0.62, (0.50,0.76)] and faster gait speed [HR = 0.63, (0.50,0.81)] were associated with lower incident dementia risk. However, when jointly adjusting for 2MW and gait speed, higher 2MW was associated with a 32% lower incident dementia risk [HR = 0.68,(0.51, 0.90)], but faster gait speed was not statistically associated with incident dementia [HR = 0.85, (0.62,1.16)]. Moderating effects were not supported (gait speed-x-2MW interaction, p = 0.14). Although both endurance and gait speed predict dementia, endurance may be more informative across the gait speed spectrum, whereas gait speed may be more feasible for clinical settings. Longer follow-up and more dementia events may better inform differential utilities of gait speed and endurance measures.

  PublisherOA PDFDOI

• Effects on blood neurodegeneration biomarkers of a randomized hearing rehabilitation intervention: ACHIEVE Clinical Trial update

  Alzheimer s & Dementia · 2025-12-01

  articleOpen access

  BACKGROUND: The Aging and Cognitive Health Evaluation in Elders (ACHIEVE) randomized trial (n = 977;ClinicalTrials.gov:NCT03243422) demonstrated that hearing intervention slowed 3-year cognitive decline by 48% among a subgroup of participants. To further investigate differences by treatment with respect to cognitive benefit, we tested the hypothesis that hearing intervention is associated with improved neurodegeneration blood biomarkers 3-years post-randomization and assessed whether combining biomarkers and cognitive results improved power in the trial design. METHOD: The ACHIEVE study is a multicenter, parallel-arm, randomized trial (hearing intervention vs health education control) on 3-year cognitive decline among adults 70-84years with untreated hearing loss and without substantial cognitive impairment. Participants were recruited from two populations: (1) a long-standing cardiovascular health observational study (Atherosclerosis Risk in Communities [ARIC]), and (2) de novo community volunteers. Plasma was collected in year 3 in a subsample (n = 540) while ARIC participants (n = 164) also had baseline plasma. Glial fibrillary acidic protein (GFAP) and neurofilament light (NfL) neurodegenerative biomarkers were derived using the Alamar CNS protein panel. Regression models estimated the association of treatment with 3-year biomarkers and global cognitive differences in the combined sample and by recruitment source and annualized change from baseline in the ARIC sample. We included a simple composite score of the standardized outcomes (Table 2). RESULT: Among the subsample, baseline characteristics were balanced by treatment. There were no significant differences on 3-year neurodegeneration biomarker levels by treatment in the combined (Table 1) or de novo groups (Table 2). Among ARIC participants, intervention resulted in lower 3-year GFAP (mean:-0.415;95%CI:-0.712,-0.118) and NfL (mean:-0.349;95%CI:-0.652,-0.046) (Table 2). Annualized change from baseline analyses among ARIC participants (Table 3) revealed intervention was associated with a slower rate of cognitive decline (difference-in-means:0.059;95%CI:0.018,0.100) and a slower rate of increase in GFAP (difference-in-means: -0.060;95%CI:-0.112,-0.009). Differences in estimates suggest combined score, relative to cognition alone, may increase power in clinical trials that show a positive effect (ARIC subgroup: p = 0.02vs. 0.08[Table 2] and p = 0.0009vs.0.048[Table 3]). CONCLUSION: Hearing intervention was associated with positive 3-year effects on neurodegeneration blood biomarkers in ARIC participants which parallels cognitive decline results. Results provides objective evidence of brain changes following hearing intervention and potential signal of a more powerful combined outcome approach for future brain health clinical trials.

  PublisherOA PDFDOI

• Association of LDL cargo proteins with white matter hyperintensity volume in older adults from the atherosclerosis risk in communities study

  Journal of Alzheimer s Disease · 2025-12-12

  articleOpen access

  Background Low-density lipoprotein cholesterol (LDL-C) has been associated with Alzheimer's disease (AD) pathology and other neuroimaging measures, such as brain volume and white matter hyperintensity (WMH) volume. Objective In this exploratory study, we examined cross-sectional associations between LDL cargo proteins and AD-related outcomes. Methods We randomly selected 65 participants without dementia with amyloid PET and brain MRI data available in the Atherosclerosis Risk in Communities. We used a mass spectrometry-based technique to quantify proteins in LDL isolated from plasma collected at the ARIC Visit 5. Linear or logistic regression was applied to evaluate the associations between individual LDL protein or LDL-C and (1) brain amyloid deposition (yes or no), (2) temporal-parietal meta-ROI brain volume (a biomarker of AD-related neurodegeneration), or (3) WMH volume, adjusting for covariates. Results Participants' average age was 76.3 years with a standard deviation of 5.4, and females comprised 53.8% (35 out of 65). The estimated effect sizes of many LDL protein's associations with these neuroimaging measures were larger than that of LDL-C. The strongest association was higher LDL apolipoprotein C1 (apoC1) and lower WMH volume. Each SD increment of LDL apoC1 LDL was associated with a lower WMH volume of 7243.1 mm 3 (95% CI [−10482.0, −4004.1]; a BH-adjusted p = 0.002). In comparison, each SD increment of LDL-C (in mg/dL) was associated with a lower WMH volume of 1676.1 mm 3 (95% CI [−5425.9, 2073.7]; a BH-adjusted p of 0.90) . Conclusions This study suggests that increased LDL apoC1 was linked to decreased WMH volume in older adults without dementia.

  PublisherOA PDFDOI

• Plasma Neurodegeneration and AD Biomarkers Identify Populations With Rapid Cognitive Decline

  Innovation in Aging · 2025-12-01

  articleOpen access

  Abstract Predicting rapid cognitive decline is valuable for patient counseling and targeting clinical trials to slow down this rate of decline. We used neurodegeneration and AD biomarkers measured in the ARIC cohort (amyloid-β 42 to amyloid-β 40 ratio, p-tau 181, NFL, and GFAP on frozen plasma collected in 2011-13 using the Quanterix SiMoA platform, N = 1825) to develop a risk score for the rate of decline in global cognition factor score and validated its performance among participants of the ACHIEVE clinical trial (same 4 biomarkers assayed using the Alamar CNS protein panel, N = 502 at the 3-year follow-up). Biomarkers were standardized within platforms using inverse-normal transformation. Cognitive decline risk scores developed in ARIC were applied to ACHIEVE for validation. Biomarker risk scores developed in ARIC validated in ACHIEVE with the upper quintile having a mean (95% CI) progression rate of -0.123 (-0.137, -0.108) in ARIC, -0.111 (-0.158, -0.065) in ACHIEVE (compared to mean of -0.012 in the total ACHIEVE validation sample). Adding age and Mini Mental State Exam did not improve the risk scores. The power of a clinical trial would change from 5% to 96% as mean annualized cognitive decline increases from -0.012 to -0.111 (selected top risk quintile; a progression SD of 0.25) in a trial of 800 participants and a 20% reduction in cognitive decline from an intervention. In conclusion, risk scores based on plasma biomarkers allow for identification of subgroups likely to experience rapid cognitive decline and may increase the power of cognitive decline clinical trials.

  PublisherOA PDFDOI

Recent grants

• EAGER: Software Engineering Research for Societal Grand Challenge Problems

  NSF · $100k · 2010–2012

• Collaborative Research: SoD-TEAM: Representations for a Science of Design

  NSF · $460k · 2006–2010

• Collaborative Research: Developing a Constructive Logic-Based Theory of Value-Based Systems Engineering

  NSF · $150k · 2014–2017

• SHF: Small: Explicating and Exploiting the Physical Semantics of Code

  NSF · $515k · 2019–2024

• A Logic-Based, Value-Oriented Science of Design

  NSF · $150k · 2005–2007

Frequent coauthors

• Rebecca F. Gottesman

  276 shared

• David S. Knopman

  Mayo Clinic

  213 shared

• Michael Griswold

  Jackson Memorial Hospital

  167 shared

• B. Gwen Windham

  University of Mississippi Medical Center

  151 shared

• Jeannette Simino

  University of Mississippi Medical Center

  134 shared

• A. Richey Sharrett

  Johns Hopkins University

  126 shared

• Chad Blackshear

  University of Mississippi Medical Center

  121 shared

• Thomas H. Mosley

  Jackson Memorial Hospital

  117 shared

Labs

• Kevin Sullivan LabPI

Education

• M.S.

  University of Washington

  1994

• Ph.D.

  University of Washington

  1994

• B.S.

  Tufts University

  1987