A capture-recapture modeling framework emphasizing expert opinion in disease surveillance

In disease surveillance, capture-recapture methods are commonly used to estimate the number of diseased cases in a defined target population. Since the number of cases never identified by any surveillance system cannot be observed, estimation of the case count typically requires at least one crucial assumption about the dependency between surveillance systems. However, such assumptions are generally unverifiable based on the observed data alone. In this paper, we advocate a modeling framework hinging on the choice of a key population-level parameter that reflects dependencies among surveillance streams. With the key dependency parameter as the focus, the proposed method offers the benefits of (a) incorporating expert opinion in the spirit of prior information to guide estimation; (b) providing accessible bias corrections, and (c) leveraging an adapted credible interval approach to facilitate inference. We apply the proposed framework to two real human immunodeficiency virus surveillance datasets exhibiting three-stream and four-stream capture-recapture-based case count estimation. Our approach enables estimation of the number of human immunodeficiency virus positive cases for both examples, under realistic assumptions that are under the investigator's control and can be readily interpreted. The proposed framework also permits principled uncertainty analyses through which a user can acknowledge their level of confidence in assumptions made about the key non-identifiable dependency parameter.

GIS-Based Assessments of Neighborhood Food Environments and Chronic Conditions: An Overview of Methodologies

The industrial revolution and urbanization fundamentally restructured populations' living circumstances, often with poor impacts on health. As an example, unhealthy food establishments may concentrate in some neighborhoods and, mediated by social and commercial drivers, increase local health risks. To understand the connections between neighborhood food environments and public health, researchers often use geographic information systems (GIS) and spatial statistics to analyze place-based evidence, but such tools require careful application and interpretation. In this article, we summarize the factors shaping neighborhood health in relation to local food environments and outline the use of GIS methodologies to assess associations between the two. We provide an overview of available data sources, analytical approaches, and their strengths and weaknesses. We postulate next steps in GIS integration with forecasting, prediction, and simulation measures to frame implications for local health policies.

Joint effects of air pollution and neighborhood socioeconomic status on cognitive decline - Mediation by depression, high cholesterol levels, and high blood pressure

Air pollution and neighborhood socioeconomic status (N-SES) are associated with adverse cardiovascular health and neuropsychiatric functioning in older adults. This study examines the degree to which the joint effects of air pollution and N-SES on the cognitive decline are mediated by high cholesterol levels, high blood pressure (HBP), and depression. In the Emory Healthy Aging Study, 14,390 participants aged 50+ years from Metro Atlanta, GA, were assessed for subjective cognitive decline using the cognitive function instrument (CFI). Information on the prior diagnosis of high cholesterol, HBP, and depression was collected through the Health History Questionnaire. Participants' census tracts were assigned 3-year average concentrations of 12 air pollutants and 16 N-SES characteristics. We used the unsupervised clustering algorithm Self-Organizing Maps (SOM) to create 6 exposure clusters based on the joint distribution of air pollution and N-SES in each census tract. Linear regression analysis was used to estimate the effects of the SOM cluster indicator on CFI, adjusting for age, race/ethnicity, education, and neighborhood residential stability. The proportion of the association mediated by high cholesterol levels, HBP, and depression was calculated by comparing the total and direct effects of SOM clusters on CFI. Depression mediated up to 87 % of the association between SOM clusters and CFI. For example, participants living in the high N-SES and high air pollution cluster had CFI scores 0.05 (95 %-CI:0.01,0.09) points higher on average compared to those from the high N-SES and low air pollution cluster; after adjusting for depression, this association was attenuated to 0.01 (95 %-CI:-0.04,0.05). HBP mediated up to 8 % of the association between SOM clusters and CFI and high cholesterol up to 5 %. Air pollution and N-SES associated cognitive decline was partially mediated by depression. Only a small portion (<10 %) of the association was mediated by HBP and high cholesterol.

Disparities in Unmet Health Care Needs Among US Children During the COVID-19 Pandemic

PURPOSE

The COVID-19 pandemic disrupted pediatric health care in the United States, and this disruption layered on existing barriers to health care. We sought to characterize disparities in unmet pediatric health care needs during this period.

METHODS

We analyzed data from Wave 1 (October through November 2020) and Wave 2 (March through May 2021) of the COVID Experiences Survey, a national longitudinal survey delivered online or via telephone to parents of children aged 5 through 12 years using a probability-based sample representative of the US household population. We examined 3 indicators of unmet pediatric health care needs as outcomes: forgone care and forgone well-child visits during fall 2020 through spring 2021, and no well-child visit in the past year as of spring 2021. Multivariate models examined relationships of child-, parent-, household-, and county-level characteristics with these indicators, adjusting for child's age, sex, and race/ethnicity.

RESULTS

On the basis of parent report, 16.3% of children aged 5 through 12 years had forgone care, 10.9% had forgone well-child visits, and 30.1% had no well-child visit in the past year. Adjusted analyses identified disparities in indicators of pediatric health care access by characteristics at the level of the child (eg, race/ethnicity, existing health conditions, mode of school instruction), parent (eg, childcare challenges), household (eg, income), and county (eg, urban-rural classification, availability of primary care physicians). Both child and parent experiences of racism were also associated with specific indicators of unmet health care needs.

CONCLUSIONS

Our findings highlight the need for continued research examining unmet health care needs and for continued efforts to optimize the clinical experience to be culturally inclusive.

How Socio-economic Inequalities Cluster People with Diabetes in Malaysia: Geographic Evaluation of Area Disparities Using a Non-parameterized Unsupervised Learning Method

Accurate assessments of epidemiological associations between health outcomes and routinely observed proximal and distal determinants of health are fundamental for the execution of effective public health interventions and policies. Methods to couple big public health data with modern statistical techniques offer greater granularity for describing and understanding data quality, disease distributions, and potential predictive connections between population-level indicators with areal-based health outcomes. This study applied clustering techniques to explore patterns of diabetes burden correlated with local socio-economic inequalities in Malaysia, with a goal of better understanding the factors influencing the collation of these clusters. Through multi-modal secondary data sources, district-wise diabetes crude rates from 271,553 individuals with diabetes sampled from 914 primary care clinics throughout Malaysia were computed. Unsupervised machine learning methods using hierarchical clustering to a set of 144 administrative districts was applied. Differences in characteristics of the areas were evaluated using multivariate non-parametric test statistics. Five statistically significant clusters were identified, each reflecting different levels of diabetes burden at the local level, each with contrasting patterns observed under the influence of population-level characteristics. The hierarchical clustering analysis that grouped local diabetes areas with varying socio-economic, demographic, and geographic characteristics offer opportunities to local public health to implement targeted interventions in an attempt to control the local diabetes burden.

A Design and Analytical Strategy for Monitoring Disease Positivity and Biomarker Levels in Accessible Closed Populations

In this paper, we advocate and expand upon a previously described monitoring strategy for efficient and robust estimation of disease prevalence and case numbers within closed and enumerated populations such as schools, workplaces, or retirement communities. The proposed design relies largely on voluntary testing, which is notoriously biased (e.g., in the case of coronavirus disease 2019) due to nonrepresentative sampling. The approach yields unbiased and comparatively precise estimates with no assumptions about factors underlying selection of individuals for voluntary testing, building on the strength of what can be a small random sampling component. This component enables the use of a recently proposed "anchor stream" estimator, a well-calibrated alternative to classical capture-recapture (CRC) estimators based on 2 data streams. We show that this estimator is equivalent to a direct standardization based on "capture," that is, selection (or not) by the voluntary testing program, made possible by means of a key parameter identified by design. This equivalency simultaneously allows for novel 2-stream CRC-like estimation of general mean values (e.g., means of continuous variables like antibody levels or biomarkers). For inference, we propose adaptations of Bayesian credible intervals when estimating case counts and bootstrapping when estimating means of continuous variables. We use simulations to demonstrate significant precision benefits relative to random sampling alone.

Liquefied Petroleum Gas or Biomass Cooking and Severe Infant Pneumonia

BACKGROUND

Exposure to household air pollution is a risk factor for severe pneumonia. The effect of replacing biomass cookstoves with liquefied petroleum gas (LPG) cookstoves on the incidence of severe infant pneumonia is uncertain.

METHODS

We conducted a randomized, controlled trial involving pregnant women 18 to 34 years of age and between 9 to less than 20 weeks' gestation in India, Guatemala, Peru, and Rwanda from May 2018 through September 2021. The women were assigned to cook with unvented LPG stoves and fuel (intervention group) or to continue cooking with biomass fuel (control group). In each trial group, we monitored adherence to the use of the assigned cookstove and measured 24-hour personal exposure to fine particulate matter (particles with an aerodynamic diameter of ≤2.5 μm [PM2.5]) in the women and their offspring. The trial had four primary outcomes; the primary outcome for which data are presented in the current report was severe pneumonia in the first year of life, as identified through facility surveillance or on verbal autopsy.

RESULTS

Among 3200 pregnant women who had undergone randomization, 3195 remained eligible and gave birth to 3061 infants (1536 in the intervention group and 1525 in the control group). High uptake of the intervention led to a reduction in personal exposure to PM2.5 among the children, with a median exposure of 24.2 μg per cubic meter (interquartile range, 17.8 to 36.4) in the intervention group and 66.0 μg per cubic meter (interquartile range, 35.2 to 132.0) in the control group. A total of 175 episodes of severe pneumonia were identified during the first year of life, with an incidence of 5.67 cases per 100 child-years (95% confidence interval [CI], 4.55 to 7.07) in the intervention group and 6.06 cases per 100 child-years (95% CI, 4.81 to 7.62) in the control group (incidence rate ratio, 0.96; 98.75% CI, 0.64 to 1.44; P = 0.81). No severe adverse events were reported to be associated with the intervention, as determined by the trial investigators.

CONCLUSIONS

The incidence of severe pneumonia among infants did not differ significantly between those whose mothers were assigned to cook with LPG stoves and fuel and those whose mothers were assigned to continue cooking with biomass stoves. (Funded by the National Institutes of Health and the Bill and Melinda Gates Foundation; HAPIN ClinicalTrials.gov number, NCT02944682.).

Association Between Clinician-Level Factors and Patient Outcomes in Virtual and In-Person Outpatient Treatment for Substance Use Disorders: Multilevel Analysis

BACKGROUND

The use of virtual treatment services increased dramatically during the COVID-19 pandemic. Unfortunately, large-scale research on virtual treatment for substance use disorder (SUD), including factors that may influence outcomes, has not advanced with the rapidly changing landscape.

OBJECTIVE

This study aims to evaluate the link between clinician-level factors and patient outcomes in populations receiving virtual and in-person intensive outpatient services.

METHODS

Data came from patients (n=1410) treated in a virtual intensive outpatient program (VIOP) and an in-person intensive outpatient program (IOP), who were discharged between January 2020 and March 2021 from a national treatment organization. Patient data were nested by treatment providers (n=58) examining associations with no-shows and discharge with staff approval. Empathy, comfort with technology, perceived stress, resistance to change, and demographic covariates were examined at the clinician level.

RESULTS

The VIOP (β=-5.71; P=.03) and the personal distress subscale measure (β=-6.31; P=.003) were negatively associated with the percentage of no-shows. The VIOP was positively associated with discharges with staff approval (odds ratio [OR] 2.38, 95% CI 1.50-3.76). Clinician scores on perspective taking (β=-9.22; P=.02), personal distress (β=-9.44; P=.02), and male clinician gender (β=-6.43; P=.04) were negatively associated with in-person no-shows. Patient load was positively associated with discharge with staff approval (OR 1.04, 95% CI 1.02-1.06).

CONCLUSIONS

Overall, patients in the VIOP had fewer no-shows and a higher rate of successful discharge. Few clinician-level characteristics were significantly associated with patient outcomes. Further research is necessary to understand the relationships among factors such as clinician gender, patient load, personal distress, and patient retention.

Fairness as an afterthought: An American perspective on fairness in model developer-clinician user collaborations

Numerous ethics guidelines have been handed down over the last few years on the ethical applications of machine learning models. Virtually every one of them mentions the importance of "fairness" in the development and use of these models. Unfortunately, though, these ethics documents omit providing a consensually adopted definition or characterization of fairness. As one group of authors observed, these documents treat fairness as an "afterthought" whose importance is undeniable but whose essence seems strikingly elusive. In this essay, which offers a distinctly American treatment of "fairness," we comment on a number of fairness formulations and on qualitative or statistical methods that have been encouraged to achieve fairness. We argue that none of them, at least from an American moral perspective, provides a one-size-fits-all definition of or methodology for securing fairness that could inform or standardize fairness over the universe of use cases witnessing machine learning applications. Instead, we argue that because fairness comprehensions and applications reflect a vast range of use contexts, model developers and clinician users will need to engage in thoughtful collaborations that examine how fairness should be conceived and operationalized in the use case at issue. Part II of this paper illustrates key moments in these collaborations, especially when inter and intra disagreement occurs among model developer and clinician user groups over whether a model is fair or unfair. We conclude by noting that these collaborations will likely occur over the lifetime of a model if its claim to fairness is to advance beyond "afterthought" status.

Space-time clustering of COVID-19 cases in the United States veteran population

PURPOSE

To assess the distribution and clustering of coronavirus disease 2019 (COVID-19) testing and incidence over space and time, U.S. Department of Veteran's Affairs (VA) data were used to describe where and when veterans experienced highest proportions of test positivity.

METHODS

Data for 6,342,455 veterans who utilized VA services between January 1, 2018, and September 30, 2021, were assessed for COVID-19 testing and test positivity. Testing and positivity proportions by county were mapped and focused-cluster tests identified significant clustering around VA facilities. Spatial cluster analysis also identified where and when veterans experienced highest proportions of test positivity.

RESULTS

Within the veterans study population and our time window, 21.3% received at least one COVID-19 test, and 20.4% of those tested had at least one positive test. There was statistically significant clustering of testing around VA facilities, revealing regional variation in testing practices. Veterans experienced highest test positivity proportions between November 2020 and January 2021 in a cluster of states in the Midwest, compared to those who received testing outside of the identified cluster (RR: 3.45).

CONCLUSIONS

Findings reflect broad regional trends in COVID-19 positivity which can inform VA policy and resource allocation. Additional analysis is needed to understand patterns during Delta and Omicron variant periods.

Enhanced Inference for Finite Population Sampling-Based Prevalence Estimation with Misclassification Errors

Epidemiologic screening programs often make use of tests with small, but non-zero probabilities of misdiagnosis. In this article, we assume the target population is finite with a fixed number of true cases, and that we apply an imperfect test with known sensitivity and specificity to a sample of individuals from the population. In this setting, we propose an enhanced inferential approach for use in conjunction with sampling-based bias-corrected prevalence estimation. While ignoring the finite nature of the population can yield markedly conservative estimates, direct application of a standard finite population correction (FPC) conversely leads to underestimation of variance. We uncover a way to leverage the typical FPC indirectly toward valid statistical inference. In particular, we derive a readily estimable extra variance component induced by misclassification in this specific but arguably common diagnostic testing scenario. Our approach yields a standard error estimate that properly captures the sampling variability of the usual bias-corrected maximum likelihood estimator of disease prevalence. Finally, we develop an adapted Bayesian credible interval for the true prevalence that offers improved frequentist properties (i.e., coverage and width) relative to a Wald-type confidence interval. We report the simulation results to demonstrate the enhanced performance of the proposed inferential methods.

Machine learning to predict risk for community-onset Staphylococcus aureus infections in children living in southeastern United States

Staphylococcus aureus (S. aureus) is known to cause human infections and since the late 1990s, community-onset antibiotic resistant infections (methicillin resistant S. aureus (MRSA)) continue to cause significant infections in the United States. Skin and soft tissue infections (SSTIs) still account for the majority of these in the outpatient setting. Machine learning can predict the location-based risks for community-level S. aureus infections. Multi-year (2002-2016) electronic health records of children <19 years old with S. aureus infections were queried for patient level data for demographic, clinical, and laboratory information. Area level data (Block group) was abstracted from U.S. Census data. A machine learning ecological niche model, maximum entropy (MaxEnt), was applied to assess model performance of specific place-based factors (determined a priori) associated with S. aureus infections; analyses were structured to compare methicillin resistant (MRSA) against methicillin sensitive S. aureus (MSSA) infections. Differences in rates of MRSA and MSSA infections were determined by comparing those which occurred in the early phase (2002-2005) and those in the later phase (2006-2016). Multi-level modeling was applied to identify risks factors for S. aureus infections. Among 16,124 unique patients with community-onset MRSA and MSSA, majority occurred in the most densely populated neighborhoods of Atlanta's metropolitan area. MaxEnt model performance showed the training AUC ranged from 0.771 to 0.824, while the testing AUC ranged from 0.769 to 0.839. Population density was the area variable which contributed the most in predicting S. aureus disease (stratified by CO-MRSA and CO-MSSA) across early and late periods. Race contributed more to CO-MRSA prediction models during the early and late periods than for CO-MSSA. Machine learning accurately predicts which densely populated areas are at highest and lowest risk for community-onset S. aureus infections over a 14-year time span.

Direct mosquito feedings on dengue-2 virus-infected people reveal dynamics of human infectiousness

Dengue virus (DENV) transmission from humans to mosquitoes is a poorly documented, but critical component of DENV epidemiology. Magnitude of viremia is the primary determinant of successful human-to-mosquito DENV transmission. People with the same level of viremia, however, can vary in their infectiousness to mosquitoes as a function of other factors that remain to be elucidated. Here, we report on a field-based study in the city of Iquitos, Peru, where we conducted direct mosquito feedings on people naturally infected with DENV and that experienced mild illness. We also enrolled people naturally infected with Zika virus (ZIKV) after the introduction of ZIKV in Iquitos during the study period. Of the 54 study participants involved in direct mosquito feedings, 43 were infected with DENV-2, two with DENV-3, and nine with ZIKV. Our analysis excluded participants whose viremia was detectable at enrollment but undetectable at the time of mosquito feeding, which was the case for all participants with DENV-3 and ZIKV infections. We analyzed the probability of onward transmission during 50 feeding events involving 27 participants infected with DENV-2 based on the presence of infectious virus in mosquito saliva 7-16 days post blood meal. Transmission probability was positively associated with the level of viremia and duration of extrinsic incubation in the mosquito. In addition, transmission probability was influenced by the day of illness in a non-monotonic fashion; i.e., transmission probability increased until 2 days after symptom onset and decreased thereafter. We conclude that mildly ill DENV-infected humans with similar levels of viremia during the first two days after symptom onset will be most infectious to mosquitoes on the second day of their illness. Quantifying variation within and between people in their contribution to DENV transmission is essential to better understand the biological determinants of human infectiousness, parametrize epidemiological models, and improve disease surveillance and prevention strategies.

Fidelity and adherence to a liquefied petroleum gas stove and fuel intervention: The multi-country Household Air Pollution Intervention Network (HAPIN) trial

BACKGROUND

Reducing household air pollution (HAP) to levels associated with health benefits requires nearly exclusive use of clean cooking fuels and abandonment of traditional biomass fuels.

METHODS

The Household Air Pollution Intervention Network (HAPIN) trial randomized 3,195 pregnant women in Guatemala, India, Peru, and Rwanda to receive a liquefied petroleum gas (LPG) stove intervention (n = 1,590), with controls expected to continue cooking with biomass fuels (n = 1,605). We assessed fidelity to intervention implementation and participant adherence to the intervention starting in pregnancy through the infant's first birthday using fuel delivery and repair records, surveys, observations, and temperature-logging stove use monitors (SUMs).

RESULTS

Fidelity and adherence to the HAPIN intervention were high. Median time required to refill LPG cylinders was 1 day (interquartile range 0-2). Although 26% (n = 410) of intervention participants reported running out of LPG at some point, the number of times was low (median: 1 day [Q1, Q3: 1, 2]) and mostly limited to the first four months of the COVID-19 pandemic. Most repairs were completed on the same day as problems were reported. Traditional stove use was observed in only 3% of observation visits, and 89% of these observations were followed up with behavioral reinforcement. According to SUMs data, intervention households used their traditional stove a median of 0.4% of all monitored days, and 81% used the traditional stove < 1 day per month. Traditional stove use was slightly higher post-COVID-19 (detected on a median [Q1, Q3] of 0.0% [0.0%, 3.4%] of days) than pre-COVID-19 (0.0% [0.0%, 1.6%] of days). There was no significant difference in intervention adherence pre- and post-birth.

CONCLUSION

Free stoves and an unlimited supply of LPG fuel delivered to participating homes combined with timely repairs, behavioral messaging, and comprehensive stove use monitoring contributed to high intervention fidelity and near-exclusive LPG use within the HAPIN trial.

Impacts of census differential privacy for small-area disease mapping to monitor health inequities

The U.S. Census Bureau will implement a modernized privacy-preserving disclosure avoidance system (DAS), which includes application of differential privacy, on publicly released 2020 census data. There are concerns that the DAS may bias small-area and demographically stratified population counts, which play a critical role in public health research, serving as denominators in estimation of disease/mortality rates. Using three DAS demonstration products, we quantify errors attributable to reliance on DAS-protected denominators in standard small-area disease mapping models for characterizing health inequities. We conduct simulation studies and real data analyses of inequities in premature mortality at the census tract level in Massachusetts and Georgia. Results show that overall patterns of inequity by racialized group and economic deprivation level are not compromised by the DAS. While early versions of DAS induce errors in mortality rate estimation that are larger for Black than non-Hispanic white populations in Massachusetts, this issue is ameliorated in newer DAS versions.

Tailoring capture-recapture methods to estimate registry-based case counts based on error-prone diagnostic signals

Surveillance research is of great importance for effective and efficient epidemiological monitoring of case counts and disease prevalence. Taking specific motivation from ongoing efforts to identify recurrent cases based on the Georgia Cancer Registry, we extend recently proposed "anchor stream" sampling design and estimation methodology. Our approach offers a more efficient and defensible alternative to traditional capture-recapture (CRC) methods by leveraging a relatively small random sample of participants whose recurrence status is obtained through a principled application of medical records abstraction. This sample is combined with one or more existing signaling data streams, which may yield data based on arbitrarily non-representative subsets of the full registry population. The key extension developed here accounts for the common problem of false positive or negative diagnostic signals from the existing data stream(s). In particular, we show that the design only requires documentation of positive signals in these non-anchor surveillance streams, and permits valid estimation of the true case count based on an estimable positive predictive value (PPV) parameter. We borrow ideas from the multiple imputation paradigm to provide accompanying standard errors, and develop an adapted Bayesian credible interval approach that yields favorable frequentist coverage properties. We demonstrate the benefits of the proposed methods through simulation studies, and provide a data example targeting estimation of the breast cancer recurrence case count among Metro Atlanta area patients from the Georgia Cancer Registry-based Cancer Recurrence Information and Surveillance Program (CRISP) database.

A Dynamic Spatial Factor Model to Describe the Opioid Syndemic in Ohio

BACKGROUND

The opioid epidemic has been ongoing for over 20 years in the United States. As opioid misuse has shifted increasingly toward injection of illicitly produced opioids, it has been associated with HIV and hepatitis C transmission. These epidemics interact to form the opioid syndemic.

METHODS

We obtain annual county-level counts of opioid overdose deaths, treatment admissions for opioid misuse, and newly diagnosed cases of acute and chronic hepatitis C and newly diagnosed HIV from 2014 to 2019. Aligned with the conceptual framework of syndemics, we develop a dynamic spatial factor model to describe the opioid syndemic for counties in Ohio and estimate the complex synergies between each of the epidemics.

RESULTS

We estimate three latent factors characterizing variation of the syndemic across space and time. The first factor reflects overall burden and is greatest in southern Ohio. The second factor describes harms and is greatest in urban counties. The third factor highlights counties with higher than expected hepatitis C rates and lower than expected HIV rates, which suggests elevated localized risk for future HIV outbreaks.

CONCLUSIONS

Through the estimation of dynamic spatial factors, we are able to estimate the complex dependencies and characterize the synergy across outcomes that underlie the syndemic. The latent factors summarize shared variation across multiple spatial time series and provide new insights into the relationships between the epidemics within the syndemic. Our framework provides a coherent approach for synthesizing complex interactions and estimating underlying sources of variation that can be applied to other syndemics.

Sensitivity and Uncertainty Analysis for Two-stream Capture-Recapture Methods in Disease Surveillance

Capture-recapture methods are widely applied in estimating the number ( ) of prevalent or cumulatively incident cases in disease surveillance. Here, we focus the bulk of our attention on the common case in which there are 2 data streams. We propose a sensitivity and uncertainty analysis framework grounded in multinomial distribution-based maximum likelihood, hinging on a key dependence parameter that is typically nonidentifiable but is epidemiologically interpretable. Focusing on the epidemiologically meaningful parameter unlocks appealing data visualizations for sensitivity analysis and provides an intuitively accessible framework for uncertainty analysis designed to leverage the practicing epidemiologist's understanding of the implementation of the surveillance streams as the basis for assumptions driving estimation of . By illustrating the proposed sensitivity analysis using publicly available HIV surveillance data, we emphasize both the need to admit the lack of information in the observed data and the appeal of incorporating expert opinion about the key dependence parameter. The proposed uncertainty analysis is a simulation-based approach designed to more realistically acknowledge variability in the estimated associated with uncertainty in an expert's opinion about the nonidentifiable parameter, together with the statistical uncertainty. We demonstrate how such an approach can also facilitate an appealing general interval estimation procedure to accompany capture-recapture methods. Simulation studies illustrate the reliable performance of the proposed approach for quantifying uncertainties in estimating in various contexts. Finally, we demonstrate how the recommended paradigm has the potential to be directly extended for application to data from >2 surveillance streams.

Fidelity and adherence to a liquefied petroleum gas stove and fuel intervention: the multi-country Household Air Pollution Intervention Network (HAPIN) trial

Background

Reducing household air pollution (HAP) to levels associated with health benefits requires nearly exclusive use of clean cooking fuels and abandonment of traditional biomass fuels.

Methods

The Household Air Pollution Intervention Network (HAPIN) trial randomized 3,195 pregnant women in Guatemala, India, Peru, and Rwanda to receive a liquefied petroleum gas (LPG) stove intervention (n=1,590), with controls expected to continue cooking with biomass fuels (n=1,605). We assessed fidelity to intervention implementation and participant adherence to the intervention starting in pregnancy through the infant's first birthday using fuel delivery and repair records, surveys, observations, and temperature-logging stove use monitors (SUMs).

Results

Fidelity and adherence to the HAPIN intervention were high. Median time required to refill LPG cylinders was 1 day (interquartile range 0-2). Although 26% (n=410) of intervention participants reported running out of LPG at some point, the number of times was low (median: 1 day [Q1, Q3: 1, 2]) and mostly limited to the first four months of the COVID-19 pandemic. Most repairs were completed on the same day as problems were reported. Traditional stove use was observed in only 3% of observation visits, and 89% of these observations were followed up with behavioral reinforcement. According to SUMs data, intervention households used their traditional stove a median of 0.4% of all monitored days, and 81% used the traditional stove <1 day per month. Traditional stove use was slightly higher post-COVID-19 (detected on a median [Q1, Q3] of 0.0% [0.0%, 3.4%] of days) than pre-COVID-19 (0.0% [0.0%, 1.6%] of days). There was no significant difference in intervention adherence pre- and post-birth.

Conclusion

Free stoves and an unlimited supply of LPG fuel delivered to participating homes combined with timely repairs, behavioral messaging, and comprehensive stove use monitoring contributed to high intervention fidelity and near-exclusive LPG use within the HAPIN trial.

Fusing an agent-based model of mosquito population dynamics with a statistical reconstruction of spatio-temporal abundance patterns

The mosquito Aedes aegypti is the vector of a number of medically-important viruses, including dengue virus, yellow fever virus, chikungunya virus, and Zika virus, and as such vector control is a key approach to managing the diseases they cause. Understanding the impact of vector control on these diseases is aided by first understanding its impact on Ae. aegypti population dynamics. A number of detail-rich models have been developed to couple the dynamics of the immature and adult stages of Ae. aegypti. The numerous assumptions of these models enable them to realistically characterize impacts of mosquito control, but they also constrain the ability of such models to reproduce empirical patterns that do not conform to the models' behavior. In contrast, statistical models afford sufficient flexibility to extract nuanced signals from noisy data, yet they have limited ability to make predictions about impacts of mosquito control on disease caused by pathogens that the mosquitoes transmit without extensive data on mosquitoes and disease. Here, we demonstrate how the differing strengths of mechanistic realism and statistical flexibility can be fused into a single model. Our analysis utilizes data from 176,352 household-level Ae. aegypti aspirator collections conducted during 1999-2011 in Iquitos, Peru. The key step in our approach is to calibrate a single parameter of the model to spatio-temporal abundance patterns predicted by a generalized additive model (GAM). In effect, this calibrated parameter absorbs residual variation in the abundance time-series not captured by other features of the mechanistic model. We then used this calibrated parameter and the literature-derived parameters in the agent-based model to explore Ae. aegypti population dynamics and the impact of insecticide spraying to kill adult mosquitoes. The baseline abundance predicted by the agent-based model closely matched that predicted by the GAM. Following spraying, the agent-based model predicted that mosquito abundance rebounds within about two months, commensurate with recent experimental data from Iquitos. Our approach was able to accurately reproduce abundance patterns in Iquitos and produce a realistic response to adulticide spraying, while retaining sufficient flexibility to be applied across a range of settings.

Space-Time Trends of Community Onset Staphylococcus aureus Infections in Children: A Group Based Trajectory Modeling Approach

PURPOSE

Staphylococcus aureus (S. aureus) remains a serious cause of infections in the U.S. and worldwide. In the U.S., methicillin resistant S. aureus (MRSA) is the leading cause of skin and soft tissue infections. This study identifies 'best' to 'worst' infection trends from 2002 to 2016, using group-based trajectory modelling approach.

METHODS

Electronic health records of children living in the southeastern U.S. with S. aureus infections from 2002-2016 were retrospectively studied, by applying a group-based trajectory model to estimate infection trends (low, high, very high), and then assess spatial significance of these trends at the census tract level; we focused on community onset (CO) infections and not those considered healthcare acquired.

RESULTS

Three methicillin sensitive (MSSA) infection trends (low, high, very high) and three MRSA trends (low, high, very high) were identified from 2002-2016. Among census tracts with community onset (CO) S. aureus cases, 29% of tracts belonged to the best trend (low infection) for both MRSA and MSSA; higher proportions occurring in the less densely populated areas. Race disparities were seen with the worst MRSA infection trends and were more often in urban areas.

CONCLUSIONS

Group based trajectory modeling identified unique trends of S. aureus infection rates over time and space, giving insight into the associated population characteristics which reflect these trends of community onset infection.

Study design and rationale for the PAASIM project: a matched cohort study on urban water supply improvements and infant enteric pathogen infection, gut microbiome development and health in Mozambique

INTRODUCTION

Despite clear linkages between provision of clean water and improvements in child health, limited information exists about the health impacts of large water infrastructure improvements in low-income settings. Billions of dollars are spent annually to improve urban water supply, and rigorous evaluation of these improvements, especially targeting informal settlements, is critical to guide policy and investment strategies. Objective measures of infection and exposure to pathogens, and measures of gut function, are needed to understand the effectiveness and impact of water supply improvements.

METHODS AND ANALYSIS

In the PAASIM study, we examine the impact of water system improvements on acute and chronic health outcomes in children in a low-income urban area of Beira, Mozambique, comprising 62 sub-neighbourhoods and ~26 300 households. This prospective matched cohort study follows 548 mother-child dyads from late pregnancy through 12 months of age. Primary outcomes include measures of enteric pathogen infections, gut microbiome composition and source drinking water microbiological quality, measured at the child's 12-month visit. Additional outcomes include diarrhoea prevalence, child growth, previous enteric pathogen exposure, child mortality and various measures of water access and quality. Our analyses will compare (1) subjects living in sub-neighbourhoods with the improved water to those living in sub-neighbourhoods without these improvements; and (2) subjects with household water connections on their premises to those without such a connection. This study will provide critical information to understand how to optimise investments for improving child health, filling the information gap about the impact of piped water provision to low-income urban households, using novel gastrointestinal disease outcomes.

ETHICS AND DISSEMINATION

This study was approved by the Emory University Institutional Review Board and the National Bio-Ethics Committee for Health in Mozambique. The pre-analysis plan is published on the Open Science Framework platform (https://osf.io/4rkn6/). Results will be shared with relevant stakeholders locally, and through publications.

Inapparent infections shape the transmission heterogeneity of dengue

Transmission heterogeneity, whereby a disproportionate fraction of pathogen transmission events result from a small number of individuals or geographic locations, is an inherent property of many, if not most, infectious disease systems. For vector-borne diseases, transmission heterogeneity is inferred from the distribution of the number of vectors per host, which could lead to significant bias in situations where vector abundance and transmission risk at the household do not correlate, as is the case with dengue virus (DENV). We used data from a contact tracing study to quantify the distribution of DENV acute infections within human activity spaces (AS), the collection of residential locations an individual routinely visits, and quantified measures of virus transmission heterogeneity from two consecutive dengue outbreaks (DENV-4 and DENV-2) that occurred in the city of Iquitos, Peru. Negative-binomial distributions and Pareto fractions showed evidence of strong overdispersion in the number of DENV infections by AS and identified super-spreading units (SSUs): i.e. AS where most infections occurred. Approximately 8% of AS were identified as SSUs, contributing to more than 50% of DENV infections. SSU occurrence was associated more with DENV-2 infection than with DENV-4, a predominance of inapparent infections (74% of all infections), households with high Aedes aegypti mosquito abundance, and high host susceptibility to the circulating DENV serotype. Marked heterogeneity in dengue case distribution, and the role of inapparent infections in defining it, highlight major challenges faced by reactive interventions if those transmission units contributing the most to transmission are not identified, prioritized, and effectively treated.

Monovalent Rotavirus Vaccine Efficacy Against Different Rotavirus Genotypes: A Pooled Analysis of Phase II and III Trial Data

BACKGROUND

Rotavirus vaccine performance appears worse in countries with high rotavirus genotype diversity. Evidence suggests diminished vaccine efficacy (VE) against G2P[4], which is heterotypic with existing monovalent rotavirus vaccine formulations. Most studies assessing genotype-specific VE have been underpowered and inconclusive.

METHODS

We pooled individual-level data from 10 Phase II and III clinical trials of rotavirus vaccine containing G1 and P[8] antigens (RV1) conducted between 2000 and 2012. We estimated VE against both any-severity and severe (Vesikari score ≥11) rotavirus gastroenteritis (RVGE) using binomial and multinomial logistic regression models for non-specific VE against any RVGE, genotype-specific VE, and RV1-typic VE against genotypes homotypic, partially heterotypic, or fully heterotypic with RV1 antigens. We adjusted models for concomitant oral poliovirus and RV1 vaccination and the country's designated child mortality stratum.

RESULTS

Analysis included 87 644 infants from 22 countries in the Americas, Europe, Africa, and Asia. For VE against severe RVGE, non-specific VE was 91% (95% confidence interval [CI]: 87-94%). Genotype-specific VE ranged from 96% (95% CI: 89-98%) against G1P[8] to 71% (43-85%) against G2P[4]. RV1-typic VE was 92% (95% CI: 84-96%) against partially heterotypic genotypes but 83% (67-91%) against fully heterotypic genotypes. For VE against any-severity RVGE, non-specific VE was 82% (95% CI: 75-87%). Genotype-specific VE ranged from 94% (95% CI: 86-97%) against G1P[8] to 63% (41-77%) against G2P[4]. RV1-typic VE was 83% (95% CI: 72-90%) against partially heterotypic genotypes but 63% (40-77%) against fully heterotypic genotypes.

CONCLUSIONS

RV1 VE is comparatively diminished against fully heterotypic genotypes including G2P[4].

Multi-Year Comparison of Community- and Species-Level West Nile Virus Antibody Prevalence in Birds from Atlanta, Georgia and Chicago, Illinois, 2005-2016

West Nile virus (WNV) is prevalent in the United States but shows considerable variation in transmission intensity. The purpose of this study was to compare patterns of WNV seroprevalence in avian communities sampled in Atlanta, Georgia and Chicago, Illinois during a 12-year period (Atlanta 2010-2016; Chicago 2005-2012) to reveal regional patterns of zoonotic activity of WNV. WNV antibodies were measured in wild bird sera using ELISA and serum neutralization methods, and seroprevalence among species, year, and location of sampling within each city were compared using binomial-distributed generalized linear mixed-effects models. Seroprevalence was highest in year-round and summer-resident species compared with migrants regardless of region; species explained more variance in seroprevalence within each city. Northern cardinals were the species most likely to test positive for WNV in each city, whereas all other species, on average, tested positive for WNV in proportion to their sample size. Despite similar patterns of seroprevalence among species, overall seroprevalence was higher in Atlanta (13.7%) than in Chicago (5%). Location and year of sampling had minor effects, with location explaining more variation in Atlanta and year explaining more variation in Chicago. Our findings highlight the nature and magnitude of regional differences in WNV urban ecology.

Quantifying heterogeneities in arbovirus transmission: Description of the rationale and methodology for a prospective longitudinal study of dengue and Zika virus transmission in Iquitos, Peru (2014-2019)

Current knowledge of dengue virus (DENV) transmission provides only a partial understanding of a complex and dynamic system yielding a public health track record that has more failures than successes. An important part of the problem is that the foundation for contemporary interventions includes a series of longstanding, but untested, assumptions based on a relatively small portion of the human population; i.e., people who are convenient to study because they manifest clinically apparent disease. Approaching dengue from the perspective of people with overt illness has produced an extensive body of useful literature. It has not, however, fully embraced heterogeneities in virus transmission dynamics that are increasingly recognized as key information still missing in the struggle to control the most important insect-transmitted viral infection of humans. Only in the last 20 years have there been significant efforts to carry out comprehensive longitudinal dengue studies. This manuscript provides the rationale and comprehensive, integrated description of the methodology for a five-year longitudinal cohort study based in the tropical city of Iquitos, in the heart of the Peruvian Amazon. Primary data collection for this study was completed in 2019. Although some manuscripts have been published to date, our principal objective here is to support subsequent publications by describing in detail the structure, methodology, and significance of a specific research program. Our project was designed to study people across the entire continuum of disease, with the ultimate goal of quantifying heterogeneities in human variables that affect DENV transmission dynamics and prevention. Because our study design is applicable to other Aedes transmitted viruses, we used it to gain insights into Zika virus (ZIKV) transmission when during the project period ZIKV was introduced and circulated in Iquitos. Our prospective contact cluster investigation design was initiated by detecttion of a person with a symptomatic DENV infection and then followed that person's immediate contacts. This allowed us to monitor individuals at high risk of DENV infection, including people with clinically inapparent and mild infections that are otherwise difficult to detect. We aimed to fill knowledge gaps by defining the contribution to DENV transmission dynamics of (1) the understudied majority of DENV-infected people with inapparent and mild infections and (2) epidemiological, entomological, and socio-behavioral sources of heterogeneity. By accounting for factors underlying variation in each person's contribution to transmission we sought to better determine the type and extent of effort needed to better prevent virus transmission and disease.

An integrated abundance model for estimating county-level prevalence of opioid misuse in Ohio

Opioid misuse is a national epidemic and a significant drug related threat to the United States. While the scale of the problem is undeniable, estimates of the local prevalence of opioid misuse are lacking, despite their importance to policy-making and resource allocation. This is due, in part, to the challenge of directly measuring opioid misuse at a local level. In this paper, we develop a Bayesian hierarchical spatio-temporal abundance model that integrates indirect county-level data on opioid-related outcomes with state-level survey estimates on prevalence of opioid misuse to estimate the latent county-level prevalence and counts of people who misuse opioids. A simulation study shows that our integrated model accurately recovers the latent counts and prevalence. We apply our model to county-level surveillance data on opioid overdose deaths and treatment admissions from the state of Ohio. Our proposed framework can be applied to other applications of small area estimation for hard to reach populations, which is a common occurrence with many health conditions such as those related to illicit behaviors.

581. Monovalent rotavirus vaccine efficacy against different rotavirus genotypes: a pooled analysis of Phase II and III trial data

Background

Rotavirus vaccine effects appear lower in countries with high child mortality rates and high diversity of rotavirus genotypes. Some evidence suggests diminished vaccine efficacy (VE) against the G2P[4] genotype, which is heterotypic with existing monovalent rotavirus vaccine formulations. Most studies assessing genotype-specific VE have been underpowered and inconclusive.

Methods

We pooled individual-level data from ten Phase II and III clinical trials of monovalent rotavirus vaccine containing G1 and P[8] antigens (RV1). We estimated VE against any-severity and severe (Vesikari score ≥11) rotavirus gastroenteritis (RVGE) using binomial and multinomial logistic regression models for three types of VE: non-specific VE against any RVGE; genotype-specific VE against specific genotypes; and RV1-typic VE against genotypes homotypic, partially heterotypic, or fully heterotypic with the RV1 G1 and P[8] antigens. Models were adjusted for oral poliovirus vaccination concomitant with RV1 vaccination and the country’s child mortality stratum.

Results

A total of 87,644 infants from 22 countries in the Americas, Europe, Africa, and Asia were included in analysis. VE against severe RVGE was 91% (95% confidence interval (CI): 87-94%). Genotype-specific VE ranged from 96% (95% CI: 89-98%) against homotypic G1P[8] to 71% (95% CI: 43-85%) against fully-heterotypic G2P[4]. VE against severe RVGE caused by partially heterotypic genotypes (92% (95% CI: 84-96%)) was similar to VE against the homotypic genotype, but VE against fully heterotypic genotypes was lower (83% (95% CI: 67-91%)). VE against any-severity RVGE was 82% (95% CI: 75-87%). Genotype-specific VE estimates against any-severity RVGE ranged from 94% (95% CI: 86-97%) against G1P[8] to 63% (95% CI: 41-77%) against G2P[4]. VE against any-severity RVGE was lower (83% (95% CI: 72-90%) against partially heterotypic genotypes, but lowest (63% (95% CI: 40-77%)) against fully heterotypic genotypes.

Conclusion

RV1 VE is diminished against fully heterotypic genotypes including G2P[4].

Disclosures

Benjamin Lopman, PhD, Epidemiological Research and Methods, LLC: Advisor/Consultant.

Liquefied Petroleum Gas or Biomass for Cooking and Effects on Birth Weight

BACKGROUND

Exposure during pregnancy to household air pollution caused by the burning of solid biomass fuel is associated with adverse health outcomes, including low birth weight. Whether the replacement of a biomass cookstove with a liquefied petroleum gas (LPG) cookstove would result in an increase in birth weight is unclear.

METHODS

We performed a randomized, controlled trial involving pregnant women (18 to <35 years of age and at 9 to <20 weeks' gestation as confirmed on ultrasonography) in Guatemala, India, Peru, and Rwanda. The women were assigned in a 1:1 ratio to use a free LPG cookstove and fuel (intervention group) or to continue using a biomass cookstove (control group). Birth weight, one of four prespecified primary outcomes, was the primary outcome for this report; data for the other three outcomes are not yet available. Birth weight was measured within 24 hours after birth. In addition, 24-hour personal exposures to fine particulate matter (particles with a diameter of ≤2.5 μm [PM2.5]), black carbon, and carbon monoxide were measured at baseline and twice during pregnancy.

RESULTS

A total of 3200 women underwent randomization; 1593 were assigned to the intervention group, and 1607 to the control group. Uptake of the intervention was nearly complete, with traditional biomass cookstoves being used at a median rate of less than 1 day per month. After randomization, the median 24-hour personal exposure to fine particulate matter was 23.9 μg per cubic meter in the intervention group and 70.7 μg per cubic meter in the control group. Among 3061 live births, a valid birth weight was available for 94.9% of the infants born to women in the intervention group and for 92.7% of infants born to those in the control group. The mean (±SD) birth weight was 2921±474.3 g in the intervention group and 2898±467.9 g in the control group, for an adjusted mean difference of 19.6 g (95% confidence interval, -10.1 to 49.2).

CONCLUSIONS

The birth weight of infants did not differ significantly between those born to women who used LPG cookstoves and those born to women who used biomass cookstoves. (Funded by the National Institutes of Health and the Bill and Melinda Gates Foundation; HAPIN ClinicalTrials.gov number, NCT02944682.).

Novel application of one-step pooled molecular testing and maximum likelihood approaches to estimate the prevalence of malaria parasitaemia among rapid diagnostic test negative samples in western Kenya

Background

Detection of malaria parasitaemia in samples that are negative by rapid diagnostic tests (RDTs) requires resource-intensive molecular tools. While pooled testing using a two-step strategy provides a cost-saving alternative to the gold standard of individual sample testing, statistical adjustments are needed to improve accuracy of prevalence estimates for a single step pooled testing strategy.

Methods

A random sample of 4670 malaria RDT negative dried blood spot samples were selected from a mass testing and treatment trial in Asembo, Gem, and Karemo, western Kenya. Samples were tested for malaria individually and in pools of five, 934 pools, by one-step quantitative polymerase chain reaction (qPCR). Maximum likelihood approaches were used to estimate subpatent parasitaemia (RDT-negative, qPCR-positive) prevalence by pooling, assuming poolwise sensitivity and specificity was either 100% (strategy A) or imperfect (strategy B). To improve and illustrate the practicality of this estimation approach, a validation study was constructed from pools allocated at random into main (734 pools) and validation (200 pools) subsets. Prevalence was estimated using strategies A and B and an inverse-variance weighted estimator and estimates were weighted to account for differential sampling rates by area.

Results

The prevalence of subpatent parasitaemia was 14.5% (95% CI 13.6–15.3%) by individual qPCR, 9.5% (95% CI (8.5–10.5%) by strategy A, and 13.9% (95% CI 12.6–15.2%) by strategy B. In the validation study, the prevalence by individual qPCR was 13.5% (95% CI 12.4–14.7%) in the main subset, 8.9% (95% CI 7.9–9.9%) by strategy A, 11.4% (95% CI 9.9–12.9%) by strategy B, and 12.8% (95% CI 11.2–14.3%) using inverse-variance weighted estimator from poolwise validation. Pooling, including a 20% validation subset, reduced costs by 52% compared to individual testing.

Conclusions

Compared to individual testing, a one-step pooled testing strategy with an internal validation subset can provide accurate prevalence estimates of PCR-positivity among RDT-negatives at a lower cost.

Negative Control Exposures: Causal Effect Identifiability and Use in Probabilistic-bias and Bayesian Analyses With Unmeasured Confounders

BACKGROUND

Probabilistic bias and Bayesian analyses are important tools for bias correction, particularly when required parameters are nonidentifiable. Negative controls are another tool; they can be used to detect and correct for confounding. Our goals are to present conditions that assure identifiability of certain causal effects and to describe and illustrate a probabilistic bias analysis and related Bayesian analysis that use a negative control exposure.

METHODS

Using potential-outcome models, we characterized assumptions needed for identification of causal effects using a dichotomous, negative control exposure when residual confounding exists. We defined bias parameters, characterized their relationships with the negative control and with specified causal effects, and described the corresponding probabilistic-bias and Bayesian analyses. We present analytic examples using data on hormone therapy and suicide attempts among transgender people. To address possible confounding by healthcare utilization, we used prior tetanus-diphtheria-pertussis (TdaP) vaccination as a negative control exposure.

RESULTS

Hormone therapy was weakly associated with risk (risk ratio [RR] = 0.9). The negative control exposure was associated with risk (RR = 1.7), suggesting confounding. Based on an assumed prior distribution for the bias parameter, the 95% simulation interval for the distribution of confounding-adjusted RR was (0.17, 1.6), with median 0.5; the 95% credibility interval was similar.

CONCLUSIONS

We used dichotomous negative control exposure to identify causal effects when a confounder was unmeasured under strong assumptions. It may be possible to relax assumptions and the negative control exposure could prove helpful for probabilistic bias analyses and Bayesian analyses.

Cross-sectional analysis of the association between personal exposure to household air pollution and blood pressure in adult women: Evidence from the multi-country Household Air Pollution Intervention Network (HAPIN) trial

Elevated blood pressure (BP) is a leading risk factor for the global burden of disease. Household air pollution (HAP), resulting from the burning of biomass fuels, may be an important cause of elevated BP in resource-poor communities. We examined the exposure-response relationship of personal exposures to HAP -fine particulate matter (PM_2.5), carbon monoxide (CO), and black carbon (BC) - with BP measures in women aged 40-79 years across four resource-poor settings in Guatemala, Peru, India and Rwanda. BP was obtained within a day of 24-h personal exposure measurements at baseline, when participants were using biomass for cooking. We used generalized additive models to characterize the shape of the association between BP and HAP, accounting for the interaction of personal exposures and age and adjusting for a priori identified confounders. A total of 418 women (mean age 52.2 ± 7.9 years) were included in this analysis. The interquartile range of exposures to PM_2.5 was 42.9-139.5 μg/m³, BC was 6.4-16.1 μg/m³, and CO was 0.5-2.9 ppm. Both SBP and PP were positively associated with PM_2.5 exposure in older aged women, achieving statistical significance around 60 years of age. The exact threshold varied by BP measure and PM_2.5 exposures being compared. For example, SBP of women aged 65 years was on average 10.8 mm Hg (95% CI 1.0-20.6) higher at 232 μg/m³ of PM_2.5 exposure (90th percentile) when compared to that of women of the same age with personal exposures of 10 μg/m³. PP in women aged 65 years was higher for exposures ≥90 μg/m³, with mean differences of 6.1 mm Hg (95% CI 1.8-10.5) and 9.2 mm Hg (95% CI 3.3-15.1) at 139 (75th percentile) and 232 μg/m³ (90th percentile) respectively, when compared to that of women of the same age with PM_2.5 exposures of 10 μg/m³. Our findings suggest that reducing HAP exposures may help to reduce BP, particularly among older women.

Using Capture-Recapture Methodology to Enhance Precision of Representative Sampling-Based Case Count Estimates

The application of serial principled sampling designs for diagnostic testing is often viewed as an ideal approach to monitoring prevalence and case counts of infectious or chronic diseases. Considering logistics and the need for timeliness and conservation of resources, surveillance efforts can generally benefit from creative designs and accompanying statistical methods to improve the precision of sampling-based estimates and reduce the size of the necessary sample. One option is to augment the analysis with available data from other surveillance streams that identify cases from the population of interest over the same timeframe, but may do so in a highly nonrepresentative manner. We consider monitoring a closed population (e.g., a long-term care facility, patient registry, or community), and encourage the use of capture-recapture methodology to produce an alternative case total estimate to the one obtained by principled sampling. With care in its implementation, even a relatively small simple or stratified random sample not only provides its own valid estimate, but provides the only fully defensible means of justifying a second estimate based on classical capture-recapture methods. We initially propose weighted averaging of the two estimators to achieve greater precision than can be obtained using either alone, and then show how a novel single capture-recapture estimator provides a unified and preferable alternative. We develop a variant on a Dirichlet-multinomial-based credible interval to accompany our hybrid design-based case count estimates, with a view toward improved coverage properties. Finally, we demonstrate the benefits of the approach through simulations designed to mimic an acute infectious disease daily monitoring program or an annual surveillance program to quantify new cases within a fixed patient registry.

Comparing and linking machine learning and semi-mechanistic models for the predictability of endemic measles dynamics

Measles is one the best-documented and most-mechanistically-studied non-linear infectious disease dynamical systems. However, systematic investigation into the comparative performance of traditional mechanistic models and machine learning approaches in forecasting the transmission dynamics of this pathogen are still rare. Here, we compare one of the most widely used semi-mechanistic models for measles (TSIR) with a commonly used machine learning approach (LASSO), comparing performance and limits in predicting short to long term outbreak trajectories and seasonality for both regular and less regular measles outbreaks in England and Wales (E&W) and the United States. First, our results indicate that the proposed LASSO model can efficiently use data from multiple major cities and achieve similar short-to-medium term forecasting performance to semi-mechanistic models for E&W epidemics. Second, interestingly, the LASSO model also captures annual to biennial bifurcation of measles epidemics in E&W caused by susceptible response to the late 1940s baby boom. LASSO may also outperform TSIR for predicting less-regular dynamics such as those observed in major cities in US between 1932–45. Although both approaches capture short-term forecasts, accuracy suffers for both methods as we attempt longer-term predictions in highly irregular, post-vaccination outbreaks in E&W. Finally, we illustrate that the LASSO model can both qualitatively and quantitatively reconstruct mechanistic assumptions, notably susceptible dynamics, in the TSIR model. Our results characterize the limits of predictability of infectious disease dynamics for strongly immunizing pathogens with both mechanistic and machine learning models, and identify connections between these two approaches.

Machine learning techniques in infectious disease modeling have grown in popularity in recent years. However, systematic investigation into the comparative performance of these approaches with traditional mechanistic models are still rare. In this paper, we compare one of the most widely used semi-mechanistic models for measles (TSIR) with a commonly used machine learning approach (LASSO), comparing performance and limits in predicting short to long term outbreaks of measles, one of the best-documented and most-mechanistically-studied non-linear infectious disease dynamical systems. Our results show that in general the LASSO outperform TSIR for predicting less-regular dynamics, and it can achieve similar performance in other scenarios when compared to the TSIR. The LASSO also has the advantages of not requiring explicit demographic data in model training. Finally, we identify connections between these two approaches and show that the LASSO model can both qualitatively and quantitatively reconstruct mechanistic assumptions, notably susceptible dynamics, in the TSIR model.

Neighborhood characteristics as confounders and effect modifiers for the association between air pollution exposure and subjective cognitive functioning

BACKGROUND

Air pollution has been associated with cognitive function in the elderly. Previous studies have not evaluated the simultaneous effect of neighborhood-level socioeconomic status (N-SES), which can be an essential source of bias.

OBJECTIVES

We explored N-SES as a confounder and effect modifier in a cross-sectional study of air pollution and subjective cognitive function.

METHODS

We included 12,058 participants age 50+ years from the Emory Healthy Aging Study in Metro Atlanta using the Cognitive Function Instrument (CFI) score as our outcome, with higher scores representing worse subjective cognitive function. We estimated 9-year average ambient carbon monoxide (CO), nitrogen oxides (NOx), and fine particulate matter (PM2.5) concentrations at residential addresses using a fusion of dispersion and chemical transport models. We collected census-tract level N-SES indicators and created two composite measures via principal component analysis and k-means clustering. Associations between pollutants and CFI and effect modification by N-SES were estimated via linear regression models adjusted for age, education, race and N-SES.

RESULTS

N-SES confounded the association between air pollution and CFI, independent of individual characteristics. We found significant effect modifications by N-SES for the association between air pollution and CFI (p-values<0.001) suggesting that effects of air pollution differ depending on N-SES. Participants living in areas with low N-SES were most vulnerable to air pollution. In the lowest N-SES urban areas, interquartile range (IQR) increases in CO, NOx, and PM2.5 were associated with 5.4% (95%-confidence interval, -0.2,11.3), 4.9% (-0.4,10.4), and 9.8% (2.2,18.0) changes in CFI, respectively. In lowest N-SES suburban areas, IQR increases in CO, NOx, and PM2.5 were associated with higher changes in CFI, namely 13.0% (0.9,26.5), 13.0% (-0.1,27.8), and 17.3% (2.5,34.2), respectively.

DISCUSSION

N-SES is an important confounder and effect modifier in our study. This finding could have implications for studying health effects of air pollution and identifying susceptible populations.

The complex relationship of air pollution and neighborhood socioeconomic status and their association with cognitive decline

BACKGROUND

Air pollution and neighborhood socioeconomic status (nSES) have been shown to affect cognitive decline in older adults. In previous studies, nSES acts as both a confounder and an effect modifier between air pollution and cognitive decline.

OBJECTIVES

This study aims to examine the individual and joint effects of air pollution and nSES on cognitive decline on adults 50 years and older in Metro Atlanta, USA.

METHODS

Perceived memory and cognitive decline was assessed in 11,897 participants aged 50+ years from the Emory Healthy Aging Study (EHAS) using the cognitive function instrument (CFI). Three-year average air pollution concentrations for 12 pollutants and 16 nSES characteristics were matched to participants using census tracts. Individual exposure linear regression and LASSO models explore individual exposure effects. Environmental mixture modeling methods including, self-organizing maps (SOM), Bayesian kernel machine regression (BKMR), and quantile-based G-computation explore joint effects, and effect modification between air pollutants and nSES characteristics on cognitive decline.

RESULTS

Participants living in areas with higher air pollution concentrations and lower nSES experienced higher CFI scores (beta: 0.121; 95 % CI: 0.076, 0.167) compared to participants living in areas with low air pollution and high nSES. Additionally, the BKMR model showed a significant overall mixture effect on cognitive decline, suggesting synergy between air pollution and nSES. These joint effects explain protective effects observed in single-pollutant linear regression models, even after adjustment for confounding by nSES (e.g., an IQR increase in CO was associated with a 0.038-point lower (95 % CI: -0.06, -0.01) CFI score).

DISCUSSION

Observed protective effects of single air pollutants on cognitive decline can be explained by joint effects and effect modification of air pollutants and nSES. Researchers must consider nSES as an effect modifier if not a co-exposure to better understand the complex relationships between air pollution and nSES in urban settings.

Understanding Variation in Rotavirus Vaccine Effectiveness Estimates in the United States: The Role of Rotavirus Activity and Diagnostic Misclassification

BACKGROUND

Estimates of rotavirus vaccine effectiveness (VE) in the United States appear higher in years with more rotavirus activity. We hypothesized rotavirus VE is constant over time but appears to vary as a function of temporal variation in local rotavirus cases and/or misclassified diagnoses.

METHODS

We analyzed 6 years of data from eight US surveillance sites on 8- to 59-month olds with acute gastroenteritis symptoms. Children's stool samples were tested via enzyme immunoassay (EIA); rotavirus-positive results were confirmed with molecular testing at the US Centers for Disease Control and Prevention. We defined rotavirus gastroenteritis cases by either positive on-site EIA results alone or positive EIA with Centers for Disease Control and Prevention confirmation. For each case definition, we estimated VE against any rotavirus gastroenteritis, moderate-to-severe disease, and hospitalization using two mixed-effect regression models: the first including year plus a year-vaccination interaction, and the second including the annual percent of rotavirus-positive tests plus a percent positive-vaccination interaction. We used multiple overimputation to bias-adjust for misclassification of cases defined by positive EIA alone.

RESULTS

Estimates of annual rotavirus VE against all outcomes fluctuated temporally, particularly when we defined cases by on-site EIA alone and used a year-vaccination interaction. Use of confirmatory testing to define cases reduced, but did not eliminate, fluctuations. Temporal fluctuations in VE estimates further attenuated when we used a percent positive-vaccination interaction. Fluctuations persisted until bias-adjustment for diagnostic misclassification.

CONCLUSIONS

Both controlling for time-varying rotavirus activity and bias-adjusting for diagnostic misclassification are critical for estimating the most valid annual rotavirus VE.

Optimizing laboratory-based surveillance networks for monitoring multi-genotype or multi-serotype infections

With the aid of laboratory typing techniques, infectious disease surveillance networks have the opportunity to obtain powerful information on the emergence, circulation, and evolution of multiple genotypes, serotypes or other subtypes of pathogens, informing understanding of transmission dynamics and strategies for prevention and control. The volume of typing performed on clinical isolates is typically limited by its ability to inform clinical care, cost and logistical constraints, especially in comparison with the capacity to monitor clinical reports of disease occurrence, which remains the most widespread form of public health surveillance. Viewing clinical disease reports as arising from a latent mixture of pathogen subtypes, laboratory typing of a subset of clinical cases can provide inference on the proportion of clinical cases attributable to each subtype (i.e., the mixture components). Optimizing protocols for the selection of isolates for typing by weighting specific subpopulations, locations, time periods, or case characteristics (e.g., disease severity), may improve inference of the frequency and distribution of pathogen subtypes within and between populations. Here, we apply the Disease Surveillance Informatics Optimization and Simulation (DIOS) framework to simulate and optimize hand foot and mouth disease (HFMD) surveillance in a high-burden region of western China. We identify laboratory surveillance designs that significantly outperform the existing network: the optimal network reduced mean absolute error in estimated serotype-specific incidence rates by 14.1%; similarly, the optimal network for monitoring severe cases reduced mean absolute error in serotype-specific incidence rates by 13.3%. In both cases, the optimal network designs achieved improved inference without increasing subtyping effort. We demonstrate how the DIOS framework can be used to optimize surveillance networks by augmenting clinical diagnostic data with limited laboratory typing resources, while adapting to specific, local surveillance objectives and constraints.

The US Coronavirus Disease 2019 (COVID-19) Surveillance Environment: An Ecological Analysis of the Relationship of Testing Adequacy in the Context of Vaccination

BACKGROUND

Coronavirus disease 2019 (COVID-19) testing is a critical component of public health surveillance and pandemic control, especially among the unvaccinated, as the nation resumes in-person activities. This study examined the relationships between COVID-19 testing rates, testing positivity rates, and vaccination coverage across US counties.

METHODS

Data from the Health and Human Services' Community Profile Report and 2016-2020 American Community Survey 5-Year Estimates were used. A total of 3114 US counties were analyzed from January through September 2021. Associations among the testing metrics and vaccination coverage were estimated using multiple linear regression models with fixed effects for states and adjusted for county demographics. COVID-19 testing rates (polymerase chain reaction [PCR] testing per 1000), testing positivity (percentage of all PCR tests that were positive), and vaccination coverage (percentage of county population that was fully vaccinated) were determined.

RESULTS

Nationally, median daily COVID-19 testing rates were highest in January and September (35.5 and 34.6 tests per capita, respectively) and lowest in July (13.2 tests per capita). Monthly testing positivity was between 0.03 and 0.12 percentage points lower for each percentage points of vaccination coverage, and monthly testing rates were between 0.08 and 0.22 tests per capita higher for each percentage point of vaccination coverage.

CONCLUSIONS

The quantity of COVID-19 testing was associated with vaccination coverage, implying counties having populations with relatively lower protection against the virus are conducting less testing than counties with relatively more protection. Monitoring testing practices in relation to vaccination coverage may be used to monitor the sufficiency of COVID-19 testing based on population susceptibility to the virus.

Home-to-Hospital Distance and Outcomes Among Community-Acquired Sepsis Hospitalizations

PURPOSE

To examine the hypothesis that longer distance from home-to-hospital is associated with worse outcomes among hospitalizations for community-acquired sepsis.

METHODS

A secondary analysis of data from the REasons for Geographic and Racial Differences in Stroke (REGARDS) prospective cohort of 30,239 white and black US adults ≥ 45 years old was conducted. Self-reported hospitalizations for serious infection between 2003-2012 fulfilling 2/4 systemic inflammatory response syndrome criteria were included. Estimated driving distance was derived from geocoded data and evaluated continuously and as quartiles of very close, close, far, very far (<3.1, 3.1-5.8, 5.9-11.5, and >11.5 miles respectively). The primary outcome was 30-day mortality while the secondary outcome was sequential organ failure assessment (SOFA) score on arrival.

RESULTS

912 hospitalizations for community-acquired sepsis had adequate data for analysis. The median (interquartile range) estimated driving distance was 5.8 miles (3.1,11.7), and 54 (5.9%) experienced the primary outcome. Compared to living very close, participants living very far had a mortality odds ratio of 1.30 (95% CI 0.64,2.62) and presenting SOFA score difference of 0.33 (95% CI -0.03,0.68).

CONCLUSIONS

Among a national sample of community-acquired sepsis hospitalizations, there was no significant association between home-to-hospital distance and either 30-day mortality or SOFA score on hospital presentation.

Telehealth Services for Substance Use Disorders During the COVID-19 Pandemic: Longitudinal Assessment of Intensive Outpatient Programming and Data Collection Practices

BACKGROUND

The onset of the COVID-19 pandemic necessitated the rapid transition of many types of substance use disorder (SUD) treatments to telehealth formats, despite limited information about what makes treatment effective in this novel format.

OBJECTIVE

This study aims to examine the feasibility and effectiveness of virtual intensive outpatient programming (IOP) treatment for SUD in the context of a global pandemic, while considering the unique challenges posed to data collection during an unprecedented public health crisis.

METHODS

The study is based on a longitudinal study with a baseline sample of 3642 patients who enrolled in intensive outpatient addiction treatment (in-person, hybrid, or virtual care) from January 2020 to March 2021 at a large substance use treatment center in the United States. The analytical sample consisted of patients who completed the 3-month postdischarge outcome survey as part of routine outcome monitoring (n=1060, 29.1% response rate).

RESULTS

No significant differences were detected by delivery format in continuous abstinence (χ²₂=0.4, P=.81), overall quality of life (F_2,826=2.06, P=.13), financial well-being (F_2,767=2.30, P=.10), psychological well-being (F_2,918=0.72, P=.49), and confidence in one's ability to stay sober (F_2,941=0.21, P=.81). Individuals in hybrid programming were more likely to report a higher level of general health than those in virtual IOP (F_2,917=4.19, P=.01).

CONCLUSIONS

Virtual outpatient care for the treatment of SUD is a feasible alternative to in-person-only programming, leading to similar self-reported outcomes at 3 months postdischarge. Given the many obstacles presented throughout data collection during a pandemic, further research is needed to better understand under what conditions telehealth is an acceptable alternative to in-person care.

Statistical implications of endogeneity induced by residential segregation in small-area modelling of health inequities

Health inequities are assessed by health departments to identify social groups disproportionately burdened by disease and by academic researchers to understand how social, economic, and environmental inequities manifest as health inequities. To characterize inequities, group-specific small-area health data are often modeled using log-linear generalized linear models (GLM) or generalized linear mixed models (GLMM) with a random intercept. These approaches estimate the same marginal rate ratio comparing disease rates across groups under standard assumptions. Here we explore how residential segregation combined with social group differences in disease risk can lead to contradictory findings from the GLM and GLMM. We show that this occurs because small-area disease rate data collected under these conditions induce endogeneity in the GLMM due to correlation between the model's offset and random effect. This results in GLMM estimates that represent conditional rather than marginal associations. We refer to endogeneity arising from the offset, which to our knowledge has not been noted previously, as "offset endogeneity". We illustrate this phenomenon in simulated data and real premature mortality data, and we propose alternative modeling approaches to address it. We also introduce to a statistical audience the social epidemiologic terminology for framing health inequities, which enables responsible interpretation of results.

LPG stove and fuel intervention among pregnant women reduce fine particle air pollution exposures in three countries: Pilot results from the HAPIN trial

The Household Air Pollution Intervention Network trial is a multi-country study on the effects of a liquefied petroleum gas (LPG) stove and fuel distribution intervention on women's and children's health. There is limited data on exposure reductions achieved by switching from solid to clean cooking fuels in rural settings across multiple countries. As formative research in 2017, we recruited pregnant women and characterized the impact of the intervention on personal exposures and kitchen levels of fine particulate matter (PM2.5) in Guatemala, India, and Rwanda. Forty pregnant women were enrolled in each site. We measured cooking area concentrations of and personal exposures to PM2.5 for 24 or 48 h using gravimetric-based PM2.5 samplers at baseline and two follow-ups over two months after delivery of an LPG cookstove and free fuel supply. Mixed models were used to estimate PM2.5 reductions. Median kitchen PM2.5 concentrations were 296 μg/m3 at baseline (interquartile range, IQR: 158-507), 24 μg/m3 at first follow-up (IQR: 18-37), and 23 μg/m3 at second follow-up (IQR: 14-37). Median personal exposures to PM2.5 were 134 μg/m3 at baseline (IQR: 71-224), 35 μg/m3 at first follow-up (IQR: 23-51), and 32 μg/m3 at second follow-up (IQR: 23-47). Overall, the LPG intervention was associated with a 92% (95% confidence interval (CI): 90-94%) reduction in kitchen PM2.5 concentrations and a 74% (95% CI: 70-79%) reduction in personal PM2.5 exposures. Results were similar for each site. CONCLUSIONS: The intervention was associated with substantial reductions in kitchen and personal PM2.5 overall and in all sites. Results suggest LPG interventions in these rural settings may lower exposures to the WHO annual interim target-1 of 35 μg/m3. The range of exposure contrasts falls on steep sections of estimated exposure-response curves for birthweight, blood pressure, and acute lower respiratory infections, implying potentially important health benefits when transitioning from solid fuels to LPG.

Considerations for Improving Reporting and Analysis of Date-Based COVID-19 Surveillance Data by Public Health Agencies

More than a year after the first domestic COVID-19 cases, the United States does not have national standards for COVID-19 surveillance data analysis and public reporting. This has led to dramatic variations in surveillance practices among public health agencies, which analyze and present newly confirmed cases by a wide variety of dates. The choice of which date to use should be guided by a balance between interpretability and epidemiological relevance. Report date is easily interpretable, generally representative of outbreak trends, and available in surveillance data sets. These features make it a preferred date for public reporting and visualization of surveillance data, although it is not appropriate for epidemiological analyses of outbreak dynamics. Symptom onset date is better suited for such analyses because of its clinical and epidemiological relevance. However, using symptom onset for public reporting of new confirmed cases can cause confusion because reporting lags result in an artificial decline in recent cases. We hope this discussion is a starting point toward a more standardized approach to date-based surveillance. Such standardization could improve public comprehension, policymaking, and outbreak response. (Am J Public Health. 2021;111(12):2127-2132. https://doi.org/10.2105/AJPH.2021.306520).

Developing a synthetic control group using electronic health records: Application to a single-arm lifestyle intervention study

The electronic health records (EHR) infrastructure offers a tremendous resource for identifying controls who match the characteristics of study participants in a single-arm trial. The objectives are to (1) demonstrate the feasibility of curating a synthetic control group for an existing study cohort through EHR data extraction and (2) evaluate the effect of a lifestyle intervention on selected cardiovascular health metrics. A total of 711 university employees were recruited between 2008 and 2012 to participate in a health partner intervention to improve cardiovascular health and were followed for five years. Data of nearly 8000 eligible subjects were extracted from the EHR to create a synthetic control cohort during the same study period. To minimize confounding, crude comparison, exact matching, propensity score matching, and doubly robust estimation were used to compare the selected cardiovascular health metrics at 1 and 5 years of follow-up. Blood pressure and body mass index improved in the intervention group compared to the EHR synthetic controls. The findings of changes in lipid measurements were somewhat unexpected. When analyzing the subgroup without lipid-lowering medications, the intervention group exhibited better control of cholesterol levels over time than did our synthetic controls. Some measurements in the EHR system may be more robust for synthetic selection than others. EHR synthetic controls can provide an alternative to estimate intervention effects appropriately in single-arm studies for these measurements.

A MULTIVARIATE SPATIOTEMPORAL CHANGE-POINT MODEL OF OPIOID OVERDOSE DEATHS IN OHIO

Ohio is one of the states most impacted by the opioid epidemic and experienced the second highest age-adjusted fatal drug overdose rate in 2017. Initially it was believed prescription opioids were driving the opioid crisis in Ohio. However, as the epidemic evolved, opioid overdose deaths due to fentanyl have drastically increased. In this work we develop a Bayesian multivariate spatiotemporal model for Ohio county overdose death rates from 2007 to 2018 due to different types of opioids. The log-odds are assumed to follow a spatially varying change point regression model. By assuming the regression coefficients are a multivariate conditional autoregressive process, we capture spatial dependence within each drug type and also dependence across drug types. The proposed model allows us to not only study spatiotemporal trends in overdose death rates but also to detect county-level shifts in these trends over time for various types of opioids.