About me
I'm an Assistant Professor in the Department of Environmental Health and Engineering at Johns Hopkins University. My research involves policy-relevant issues at the intersection of AI/data science, environmental justice, and health equity. Recent areas of focus include planetary health, humanitarian operations, and climate adaptation.
I completed my PhD in Biomedical Data Science at Stanford and obtained a BS in Statistics at the University of Chicago. Previously, I've held data science roles at the World Health Organization and Médecins Sans Frontières, and was a Climate Security Fellow at the Center for Climate and Security. In 2024, I was named to the Grist 50 list of climate leaders and Business Insider's AI Power List.
Feel free to reach out if you'd like to discuss any shared interests or potential collaborations. My email is bhuynh {at} jhu {dot} edu. My last name can approximately be pronounced "hwin" or "win," and my pronouns are he/him.
Selected publications
For a full list of publications, see here: Google Scholar.
Estimated Childhood Lead Exposure From Drinking Water in Chicago
JAMA Pediatrics, 2024
Huynh BQ, Chin ET, & Kiang MV. [The Guardian] [The Washington Post] [CNN] [NPR] [Chicago Sun-Times] [CBS] [NBC] [ABC] [Show abstract]
Importance: There is no level of lead in drinking water considered to be safe, yet lead service lines are still commonly used in water systems across the US.
Objective: To identify the extent of lead-contaminated drinking water in Chicago, Illinois, and model its impact on children younger than 6 years.
Design, Setting, and Participants: For this cross-sectional study, a retrospective assessment was performed of lead exposure based on household tests collected from January 2016 to September 2023. Tests were obtained from households in Chicago that registered for a free self-administered testing service for lead exposure. Machine learning and microsimulation were used to estimate citywide childhood lead exposure.
Exposure: Lead-contaminated drinking water, measured in parts per billion.
Main Outcomes and Measures: Number of children younger than 6 years exposed to lead-contaminated water.
Results: A total of 38 385 household lead tests were collected. An estimated 68% (95% uncertainty interval, 66%-69%) of children younger than 6 years were exposed to lead-contaminated water, corresponding to 129 000 children (95% uncertainty interval, 128 000-131 000 children). Ten-percentage-point increases in block-level Black and Hispanic populations were associated with 3% (95% CI, 2%-3%) and 6% (95% CI, 5%-7%) decreases in odds of being tested for lead and 4% (95% CI, 3%-6%) and 11% (95% CI, 10%-13%) increases in having lead-contaminated drinking water, respectively.
Conclusions and Relevance: These findings indicate that childhood lead exposure is widespread in Chicago, and racial inequities are present in both testing rates and exposure levels. Machine learning may assist in preliminary screening for lead exposure, and efforts to remediate the effects of environmental racism should involve improving outreach for and access to lead testing services.
Mitigating allocative tradeoffs and harms in an environmental justice data tool
Nature Machine Intelligence, 2024
Huynh BQ*, Chin ET*, Koenecke A, Ouyang D, Ho DE, Kiang MV, & Rehkopf DH. *Co-first author. [arXiv] [Grist, SF Chronicle, SFGate, Long Beach Post, Jefferson Public Radio, and more via CalMatters] [Show abstract]
Neighborhood-level screening algorithms are increasingly being deployed to inform policy decisions. However, their potential for harm remains unclear: algorithmic decision-making has broadly fallen under scrutiny for disproportionate harm to marginalized groups, yet opaque methodology and proprietary data limit the generalizability of algorithmic audits. Here we leverage publicly available data to fully reproduce and audit a large-scale algorithm known as CalEnviroScreen, designed to promote environmental justice and guide public funding by identifying disadvantaged neighborhoods. We observe the model to be both highly sensitive to subjective model specifications and financially consequential, estimating the effect of its positive designations as a 104% (62-145%) increase in funding, equivalent to $2.08 billion ($1.56-2.41 billion) over four years. We further observe allocative tradeoffs and susceptibility to manipulation, raising ethical concerns. We recommend incorporating technical strategies to mitigate allocative harm and accountability mechanisms to prevent misuse.
AI for Anticipatory Action: Moving Beyond Climate Forecasting
Proceedings of the 2023 AAAI Fall Symposia
Huynh BQ & Kiang MV. [Show abstract]
Disaster response agencies have been shifting from a paradigm of climate forecasting towards one of anticipatory action: assessing not just what the climate will be, but how it will impact specific populations, thereby enabling proactive response and resource allocation. Machine learning models are becoming exceptionally powerful at climate forecasting, but methodological gaps remain in terms of facilitating anticipatory action. Here we provide an overview of anticipatory action, review relevant applications of machine learning, identify common challenges, and highlight areas where machine learning can uniquely contribute to advancing disaster response for populations most vulnerable to climate change.
Public health impacts of an imminent Red Sea oil spill
Nature Sustainability, 2021
Huynh BQ, Kwong LH, Kiang MV, Chin ET, Mohareb AM, Jumaan AO, Basu S, Geldsetzer P, Karaki FM, & Rehkopf DH. [Invited commentary] [Nature research highlight] [Stanford press release] [BBC] [CNN] [The Economist] [The Guardian] [Al Jazeera] [The New Yorker] [The World] Update: disaster averted! [Show abstract]
The possibility of a massive oil spill in the Red Sea is increasingly likely. The Safer, a deteriorating oil tanker containing 1.1 million barrels of oil, has been deserted near the coast of Yemen since 2015 and threatens environmental catastrophe to a country presently in a humanitarian crisis. Here, we model the immediate public health impacts of a simulated spill. We estimate that all of Yemen’s imported fuel through its key Red Sea ports would be disrupted and that the anticipated spill could disrupt clean-water supply equivalent to the daily use of 9.0–9.9 million people, food supply for 5.7–8.4 million people and 93–100% of Yemen’s Red Sea fisheries. We also estimate an increased risk of cardiovascular hospitalization from pollution ranging from 5.8 to 42.0% over the duration of the spill. The spill and its potentially disastrous impacts remain entirely preventable through offloading the oil. Our results stress the need for urgent action to avert this looming disaster.
Routine asymptomatic testing strategies for airline travel during the COVID-19 pandemic: a simulation analysis
The Lancet Infectious Diseases, 2021
Kiang MV, Chin ET, Huynh BQ, Chapman LAC, Rodríguez-Barraquer I, Greenhouse B, Rutherford G, Bibbins-Domingo K, Havlir D, Basu S, & Lo NC. (Editor's choice.) [UCSF press release] [NPR] [ABC News] [SF Chronicle] [Cited in CDC report] [Cited in UK guidance] [Show abstract]
Background: Routine viral testing strategies for SARS-CoV-2 infection might facilitate safe airline travel during the COVID-19 pandemic and mitigate global spread of the virus. However, the effectiveness of these test-and-travel strategies to reduce passenger risk of SARS-CoV-2 infection and population-level transmission remains unknown.
Methods: In this simulation study, we developed a microsimulation of SARS-CoV-2 transmission in a cohort of 100 000 US domestic airline travellers using publicly available data on COVID-19 clinical cases and published natural history parameters to assign individuals one of five health states of susceptible to infection, latent period, early infection, late infection, or recovered. We estimated a per-day risk of infection with SARS-CoV-2 corresponding to a daily incidence of 150 infections per 100 000 people. We assessed five testing strategies: (1) anterior nasal PCR test within 3 days of departure, (2) PCR within 3 days of departure and 5 days after arrival, (3) rapid antigen test on the day of travel (assuming 90% of the sensitivity of PCR during active infection), (4) rapid antigen test on the day of travel and PCR test 5 days after arrival, and (5) PCR test 5 days after arrival. Strategies 2 and 4 included a 5-day quarantine after arrival. The travel period was defined as 3 days before travel to 2 weeks after travel. Under each scenario, individuals who tested positive before travel were not permitted to travel. The primary study outcome was cumulative number of infectious days in the cohort over the travel period without isolation or quarantine (population-level transmission risk), and the key secondary outcome was the number of infectious people detected on the day of travel (passenger risk of infection).
Findings: We estimated that in a cohort of 100 000 airline travellers, in a scenario with no testing or screening, there would be 8357 (95% uncertainty interval 6144–12831) infectious days with 649 (505–950) actively infectious passengers on the day of travel. The pre-travel PCR test reduced the number of infectious days from 8357 to 5401 (3917–8677), a reduction of 36% (29–41) compared with the base case, and identified 569 (88% [76–92]) of 649 actively infectious travellers on the day of flight; the addition of post-travel quarantine and PCR reduced the number of infectious days to 1474 (1087–2342), a reduction of 82% (80–84) compared with the base case. The rapid antigen test on the day of travel reduced the number of infectious days to 5674 (4126–9081), a reduction of 32% (26–38) compared with the base case, and identified 560 (86% [83–89]) actively infectious travellers; the addition of post-travel quarantine and PCR reduced the number of infectious days to 2518 (1935–3821), a reduction of 70% (67–72) compared with the base case. The post-travel PCR alone reduced the number of infectious days to 4851 (3714–7679), a reduction of 42% (35–49) compared with the base case.
Interpretation: Routine asymptomatic testing for SARS-CoV-2 before travel can be an effective strategy to reduce passenger risk of infection during travel, although abbreviated quarantine with post-travel testing is probably needed to reduce population-level transmission due to importation of infection when travelling from a high to low incidence setting.
Frequency of routine testing for COVID-19 in high-risk environments to reduce workplace outbreaks
Clinical Infectious Diseases, 2020
Chin ET*, Huynh BQ*, Chapman LAC, Murrill M, Basu S, & Lo NC. *Co-first author. [Cited in CDC guidance] [Cited in Africa CDC Guidance] [Cited in UC System-wide testing recommendations] [Show abstract]
Routine asymptomatic testing strategies for COVID-19 have been proposed to prevent outbreaks in high-risk healthcare environments. We used simulation modeling to evaluate the optimal frequency of viral testing. We found that routine testing substantially reduces risk of outbreaks, but may need to be as frequent as twice weekly.
Projected geographic disparities in healthcare worker absenteeism from COVID-19 school closures and the economic feasibility of child care subsidies: a simulation study
BMC Medicine, 2020
Chin ET*, Huynh BQ*, Lo NC, Hastie T, & Basu S. *Co-first author. [Cited in WHO report] [Show abstract]
Background: School closures have been enacted as a measure of mitigation during the ongoing coronavirus disease 2019 (COVID-19) pandemic. It has been shown that school closures could cause absenteeism among healthcare workers with dependent children, but there remains a need for spatially granular analyses of the relationship between school closures and healthcare worker absenteeism to inform local community preparedness.
Methods: We provide national- and county-level simulations of school closures and unmet child care needs across the USA. We develop individual simulations using county-level demographic and occupational data, and model school closure effectiveness with age-structured compartmental models. We perform multivariate quasi-Poisson ecological regressions to find associations between unmet child care needs and COVID-19 vulnerability factors.
Results: At the national level, we estimate the projected rate of unmet child care needs for healthcare worker households to range from 7.4 to 8.7%, and the effectiveness of school closures as a 7.6% and 8.4% reduction in fewer hospital and intensive care unit (ICU) beds, respectively, at peak demand when varying across initial reproduction number estimates by state. At the county level, we find substantial variations of projected unmet child care needs and school closure effects, 9.5% (interquartile range (IQR) 8.2–10.9%) of healthcare worker households and 5.2% (IQR 4.1–6.5%) and 6.8% (IQR 4.8–8.8%) reduction in fewer hospital and ICU beds, respectively, at peak demand. We find significant positive associations between estimated levels of unmet child care needs and diabetes prevalence, county rurality, and race (p<0.05). We estimate costs of absenteeism and child care and observe from our models that an estimated 76.3 to 96.8% of counties would find it less expensive to provide child care to all healthcare workers with children than to bear the costs of healthcare worker absenteeism during school closures.
Conclusions: School closures are projected to reduce peak ICU and hospital demand, but could disrupt healthcare systems through absenteeism, especially in counties that are already particularly vulnerable to COVID-19. Child care subsidies could help circumvent the ostensible trade-off between school closures and healthcare worker absenteeism.
Forecasting internally displaced population migration patterns in Syria
and Yemen
Disaster Medicine and Public Health Preparedness, 2019
Huynh BQ & Basu, S. [Pre-print PDF] [Cited in UN report] [Show abstract]
Objectives: Armed conflict has contributed to an unprecedented number of internally displaced persons (IDPs), individuals who are forced out of their homes but remain within their country. IDPs often urgently require shelter, food, and healthcare, yet prediction of when IDPs will migrate to an area remains a major challenge for aid delivery organizations. We sought to develop an IDP migration forecasting framework that could empower humanitarian aid groups to more effectively allocate resources during conflicts.
Methods: We modeled monthly IDP migration between provinces within Syria and within Yemen using data on food prices, fuel prices, wages, location, time, and conflict reports. We compared machine learning methods with baseline persistence methods of forecasting.
Results: We found a machine learning approach that more accurately forecast migration trends than baseline persistence methods. A random forest model outperformed the best persistence model in terms of root mean square error of log migration by 26% and 17% for the Syria and Yemen datasets, respectively.
Conclusions: Integrating diverse data sources into a machine learning model appears to improve IDP migration prediction. Further work should examine whether implementation of such models can enable proactive aid allocation for IDPs in anticipation of forecast arrivals.
A deep feature fusion methodology for breast cancer diagnosis demonstrated on three imaging modality datasets
Medical Physics, 2017
Antropova NO*, Huynh BQ*, & Giger ML. *Co-first author. (Editor's choice.) [Show abstract]
Background: Deep learning methods for radiomics/computer-aided diagnosis (CADx) are often prohibited by small datasets, long computation time, and the need for extensive image preprocessing.
Aims: We aim to develop a breast CADx methodology that addresses the aforementioned issues by exploiting the efficiency of pre-trained convolutional neural networks (CNNs) and using pre-existing handcrafted CADx features.
Materials & Methods: We present a methodology that extracts and pools low- to mid-level features using a pretrained CNN and fuses them with handcrafted radiomic features computed using conventional CADx methods. Our methodology is tested on three different clinical imaging modalities (dynamic contrast enhanced-MRI [690 cases], full-field digital mammography [245 cases], and ultrasound [1125 cases]).
Results: From ROC analysis, our fusion-based method demonstrates, on all three imaging modalities, statistically significant improvements in terms of AUC as compared to previous breast cancer CADx methods in the task of distinguishing between malignant and benign lesions. (DCE-MRI [AUC = 0.89 (se = 0.01)], FFDM [AUC = 0.86 (se = 0.01)], and ultrasound [AUC = 0.90 (se = 0.01)]).
Discussion/Conclusion: We proposed a novel breast CADx methodology that can be used to more effectively characterize breast lesions in comparison to existing methods. Furthermore, our proposed methodology is computationally efficient and circumvents the need for image preprocessing.
Digital mammographic tumor classification using transfer learning from deep convolutional neural networks
Journal of Medical Imaging, 2016
Huynh BQ, Li H, & Giger ML. [Show abstract]
Convolutional neural networks (CNNs) show potential for computer-aided diagnosis (CADx) by learning features directly from the image data instead of using analytically extracted features. However, CNNs are difficult to train from scratch for medical images due to small sample sizes and variations in tumor presentations. Instead, transfer learning can be used to extract tumor information from medical images via CNNs originally pretrained for nonmedical tasks, alleviating the need for large datasets. Our database includes 219 breast lesions (607 full-field digital mammographic images). We compared support vector machine classifiers based on the CNN-extracted image features and our prior computer-extracted tumor features in the task of distinguishing between benign and malignant breast lesions. Five-fold cross validation (by lesion) was conducted with the area under the receiver operating characteristic (ROC) curve as the performance metric. Results show that classifiers based on CNN-extracted features (with transfer learning) perform comparably to those using analytically extracted features [area under the ROC curve (AUC)=0.81]. Further, the performance of ensemble classifiers based on both types was significantly better than that of either classifier type alone (AUC=0.86 versus 0.81, p=0.022). We conclude that transfer learning can improve current CADx methods while also providing standalone classifiers without large datasets, facilitating machine-learning methods in radiomics and precision medicine.