Residential scene classification for gridded population sampling in developing countries using deep convolutional neural networks on satellite imagery.

BACKGROUND: Conducting surveys in low- and middle-income countries is often challenging because many areas lack a complete sampling frame, have outdated census information, or have limited data available for designing and selecting a representative sample. Geosampling is a probability-based, gridded population sampling method that addresses some of these issues by using geographic information system (GIS) tools to create logistically manageable area units for sampling. GIS grid cells are overlaid to partition a country's existing administrative boundaries into area units that vary in size from 50 m × 50 m to 150 m × 150 m. To avoid sending interviewers to unoccupied areas, researchers manually classify grid cells as "residential" or "nonresidential" through visual inspection of aerial images. "Nonresidential" units are then excluded from sampling and data collection. This process of manually classifying sampling units has drawbacks since it is labor intensive, prone to human error, and creates the need for simplifying assumptions during calculation of design-based sampling weights. In this paper, we discuss the development of a deep learning classification model to predict whether aerial images are residential or nonresidential, thus reducing manual labor and eliminating the need for simplifying assumptions. RESULTS: On our test sets, the model performs comparable to a human-level baseline in both Nigeria (94.5% accuracy) and Guatemala (96.4% accuracy), and outperforms baseline machine learning models trained on crowdsourced or remote-sensed geospatial features. Additionally, our findings suggest that this approach can work well in new areas with relatively modest amounts of training data. CONCLUSIONS: Gridded population sampling methods like geosampling are becoming increasingly popular in countries with outdated or inaccurate census data because of their timeliness, flexibility, and cost. Using deep learning models directly on satellite images, we provide a novel method for sample frame construction that identifies residential gridded aerial units. In cases where manual classification of satellite images is used to (1) correct for errors in gridded population data sets or (2) classify grids where population estimates are unavailable, this methodology can help reduce annotation burden with comparable quality to human analysts.

Evaluating Policies to Decrease the Risk of Introducing SARS-CoV-2 Infections to Nursing Home Facilities.

We used an individual-based microsimulation model of North Carolina to determine what facility-level policies would result in the greatest reduction in the number of individuals with SARS-CoV-2 entering the nursing home environment from 12/15/2021 to 1/3/2022 (e.g., Omicron variant surge). On average, there were 14,287 (Credible Interval [CI]: 13,477-15,147) daily visitors and 17,168 (CI: 16,571-17,768) HCW coming from the community into 426 nursing home facilities. Policies requiring a negative rapid test or vaccinated status for visitors resulted in the greatest reduction in the number of individuals with SARS-CoV-2 infection entering the nursing home environment with a 29.6% (26.9%-32.0%) and 24.0% (CI: 22.2%-25.5%) reduction, respectively. Policies halving visits (21.2% [20.0%-28.2%]), requiring all vaccinated HCW to receive a booster (7.8% [CI: 7.4%-8.7%]), and limiting visitation to a primary visitor (6.5% [CI: 3.5%-9.7%]) reduced infectious contacts to a lesser degree.

Estimate of undetected severe acute respiratory coronavirus virus 2 (SARS-CoV-2) infection in acute-care hospital settings using an individual-based microsimulation model.

OBJECTIVE: Current guidance states that asymptomatic screening for severe acute respiratory coronavirus virus 2 (SARS-CoV-2) prior to admission to an acute-care setting is at the facility's discretion. This study's objective was to estimate the number of undetected cases of SARS-CoV-2 admitted as inpatients under 4 testing approaches and varying assumptions. DESIGN AND SETTING: Individual-based microsimulation of 104 North Carolina acute-care hospitals. PATIENTS: All simulated inpatient admissions to acute-care hospitals from December 15, 2021, to January 13, 2022 [ie, during the SARS-COV-2 ο (omicron) variant surge]. INTERVENTIONS: We simulated (1) only testing symptomatic patients, (2) 1-stage antigen testing with no confirmatory polymerase chain reaction (PCR) test, (3) 1-stage antigen testing with a confirmatory PCR for negative results, and (4) serial antigen screening (ie, repeat antigen test 2 days after a negative result). RESULTS: Over 1 month, there were 77,980 admissions: 13.7% for COVID-19, 4.3% with but not for COVID-19, and 82.0% for non-COVID-19 indications without current infection. Without asymptomatic screening, 1,089 (credible interval [CI], 946-1,253) total SARS-CoV-2 infections (7.72%) went undetected. With 1-stage antigen screening, 734 (CI, 638-845) asymptomatic infections (67.4%) were detected, with 1,277 false positives. With combined antigen and PCR screening, 1,007 (CI, 875-1,159) asymptomatic infections (92.5%) were detected, with 5,578 false positives. A serial antigen testing policy detected 973 (CI, 845-1,120) asymptomatic infections (89.4%), with 2,529 false positives. CONCLUSIONS: Serial antigen testing identified >85% of asymptomatic infections and resulted in fewer false positives with less cost per identified infection compared to combined antigen plus PCR testing.

Impacts of timing, length, and intensity of behavioral interventions to COVID-19 dynamics: North Carolina county-level examples.

We sought to examine how the impact of revocable behavioral interventions, e.g., shelter-in-place, varies throughout an epidemic, as well as the role that the proportion of susceptible individuals had on an intervention's impact. We estimated the theoretical impacts of start day, length, and intensity of interventions on disease transmission and illustrated them on COVID-19 dynamics in Wake County, North Carolina, to inform how interventions can be most effective. We used a Susceptible, Exposed, Infectious, and Recovered (SEIR) model to estimate epidemic curves with modifications to the disease transmission parameter (ß). We designed modifications to simulate events likely to increase transmission (e.g., long weekends, holiday seasons) or behavioral interventions likely to decrease it (e.g., shelter-in-place, masking). We compared the resultant curves' shape, timing, and cumulative case count to baseline and across other modified curves. Interventions led to changes in COVID-19 dynamics, including moving the peak's location, height, and width. The proportion susceptible, at the start day, strongly influenced their impact. Early interventions shifted the curve, while interventions near the peak modified shape and case count. For some scenarios, in which the transmission parameter was decreased, the final cumulative count increased over baseline. We showed that the timing of revocable interventions has a strong impact on their effect. The same intervention applied at different time points, corresponding to different proportions of susceptibility, resulted in qualitatively differential effects. Accurate estimation of the proportion susceptible is critical for understanding an intervention's impact. The findings presented here provide evidence of the importance of estimating the proportion of the population that is susceptible when predicting the impact of behavioral infection control interventions. Greater emphasis should be placed on the estimation of this epidemic component in intervention design and decision-making. Our results are generic and are applicable to other infectious disease epidemics, as well as to future waves of the current COVID-19 epidemic. Developed into a publicly available tool that allows users to modify the parameters to estimate impacts of different interventions, these models could aid in evaluating behavioral intervention options prior to their use and in predicting case increases from specific events.

Enhancing the prediction of hospitalization from a COVID-19 agent-based model: A Bayesian method for model parameter estimation.

Agent-based models (ABMs) have become a common tool for estimating demand for hospital beds during the COVID-19 pandemic. A key parameter in these ABMs is the probability of hospitalization for agents with COVID-19. Many published COVID-19 ABMs use either single point or age-specific estimates of the probability of hospitalization for agents with COVID-19, omitting key factors: comorbidities and testing status (i.e., received vs. did not receive COVID-19 test). These omissions can inhibit interpretability, particularly by stakeholders seeking to use an ABM for transparent decision-making. We introduce a straightforward yet novel application of Bayes' theorem with inputs from aggregated hospital data to better incorporate these factors in an ABM. We update input parameters for a North Carolina COVID-19 ABM using this approach, demonstrate sensitivity to input data selections, and highlight the enhanced interpretability and accuracy of the method and the predictions. We propose that even in tumultuous scenarios with limited information like the early months of the COVID-19 pandemic, straightforward approaches like this one with discrete, attainable inputs can improve ABMs to better support stakeholders.

Incorporation of near-real-time hospital occupancy data to improve hospitalization forecast accuracy during the COVID-19 pandemic.

Public health decision makers rely on hospitalization forecasts to inform COVID-19 pandemic planning and resource allocation. Hospitalization forecasts are most relevant when they are accurate, made available quickly, and updated frequently. We rapidly adapted an agent-based model (ABM) to provide weekly 30-day hospitalization forecasts (i.e., demand for intensive care unit [ICU] beds and non-ICU beds) by state and region in North Carolina for public health decision makers. The ABM was based on a synthetic population of North Carolina residents and included movement of agents (i.e., patients) among North Carolina hospitals, nursing homes, and the community. We assigned SARS-CoV-2 infection to agents using county-level compartmental models and determined agents' COVID-19 severity and probability of hospitalization using synthetic population characteristics (e.g., age, comorbidities). We generated weekly 30-day hospitalization forecasts during May-December 2020 and evaluated the impact of major model updates on statewide forecast accuracy under a SARS-CoV-2 effective reproduction number range of 1.0-1.2. Of the 21 forecasts included in the assessment, the average mean absolute percentage error (MAPE) was 7.8% for non-ICU beds and 23.6% for ICU beds. Among the major model updates, integration of near-real-time hospital occupancy data into the model had the largest impact on improving forecast accuracy, reducing the average MAPE for non-ICU beds from 6.6% to 3.9% and for ICU beds from 33.4% to 6.5%. Our results suggest that future pandemic hospitalization forecasting efforts should prioritize early inclusion of hospital occupancy data to maximize accuracy.

Lessons learned from the rapid development of a statewide simulation model for predicting COVID-19's impact on healthcare resources and capacity.

The first case of COVID-19 was detected in North Carolina (NC) on March 3, 2020. By the end of April, the number of confirmed cases had soared to over 10,000. NC health systems faced intense strain to support surging intensive care unit admissions and avert hospital capacity and resource saturation. Forecasting techniques can be used to provide public health decision makers with reliable data needed to better prepare for and respond to public health crises. Hospitalization forecasts in particular play an important role in informing pandemic planning and resource allocation. These forecasts are only relevant, however, when they are accurate, made available quickly, and updated frequently. To support the pressing need for reliable COVID-19 data, RTI adapted a previously developed geospatially explicit healthcare facility network model to predict COVID-19's impact on healthcare resources and capacity in NC. The model adaptation was an iterative process requiring constant evolution to meet stakeholder needs and inform epidemic progression in NC. Here we describe key steps taken, challenges faced, and lessons learned from adapting and implementing our COVID-19 model and coordinating with university, state, and federal partners to combat the COVID-19 epidemic in NC.

Modeling inpatient and outpatient antibiotic stewardship interventions to reduce the burden of Clostridioides difficile infection in a regional healthcare network.

Antibiotic exposure can lead to unintended outcomes, including drug-drug interactions, adverse drug events, and healthcare-associated infections like Clostridioides difficile infection (CDI). Improving antibiotic use is critical to reduce an individual's CDI risk. Antibiotic stewardship initiatives can reduce inappropriate antibiotic prescribing (e.g., unnecessary antibiotic prescribing, inappropriate antibiotic selection), impacting both hospital (healthcare)-onset (HO)-CDI and community-associated (CA)-CDI. Previous computational and mathematical modeling studies have demonstrated a reduction in CDI incidence associated with antibiotic stewardship initiatives in hospital settings. Although the impact of antibiotic stewardship initiatives in long-term care facilities (LTCFs), including nursing homes, and in outpatient settings have been documented, the effects of specific interventions on CDI incidence are not well understood. We examined the relative effectiveness of antibiotic stewardship interventions on CDI incidence using a geospatially explicit agent-based model of a regional healthcare network in North Carolina. We simulated reductions in unnecessary antibiotic prescribing and inappropriate antibiotic selection with intervention scenarios at individual and network healthcare facilities, including short-term acute care hospitals (STACHs), nursing homes, and outpatient locations. Modeled antibiotic prescription rates were calculated using patient-level data on antibiotic length of therapy for the 10 modeled network STACHs. By simulating a 30% reduction in antibiotics prescribed across all inpatient and outpatient locations, we found the greatest reductions on network CDI incidence among tested scenarios, namely a 17% decrease in HO-CDI incidence and 7% decrease in CA-CDI. Among intervention scenarios of reducing inappropriate antibiotic selection, we found a greater impact on network CDI incidence when modeling this reduction in nursing homes alone compared to the same intervention in STACHs alone. These results support the potential importance of LTCF and outpatient antibiotic stewardship efforts on network CDI burden and add to the evidence that a coordinated approach to antibiotic stewardship across multiple facilities, including inpatient and outpatient settings, within a regional healthcare network could be an effective strategy to reduce network CDI burden.

Creation of a Geospatially Explicit, Agent-based Model of a Regional Healthcare Network with Application to Clostridioides difficile Infection.

Agent-based models (ABMs) describe and simulate complex systems comprising unique agents, or individuals, while accounting for geospatial and temporal variability among dynamic processes. ABMs are increasingly used to study healthcare-associated infections (ie, infections acquired during admission to a healthcare facility), including Clostridioides difficile infection, currently the most common healthcare-associated infection in the United States. The overall burden and transmission dynamics of healthcare-associated infections, including C difficile infection, may be influenced by community sources and movement of people among healthcare facilities and communities. These complex dynamics warrant geospatially explicit ABMs that extend beyond single healthcare facilities to include entire systems (eg, hospitals, nursing homes and extended care facilities, the community). The agents in ABMs can be built on a synthetic population, a model-generated representation of the actual population with associated spatial (eg, home residence), temporal (eg, change in location over time), and nonspatial (eg, sociodemographic features) attributes. We describe our methods to create a geospatially explicit ABM of a major regional healthcare network using a synthetic population as microdata input. We illustrate agent movement in the healthcare network and the community, informed by patient-level medical records, aggregate hospital discharge data, healthcare facility licensing data, and published literature. We apply the ABM output to visualize agent movement in the healthcare network and the community served by the network. We provide an application example of the ABM to C difficile infection using a natural history submodel. We discuss the ABM's potential to detect network areas where disease risk is high; simulate and evaluate interventions to protect public health; adapt to other geographic locations and healthcare-associated infections, including emerging pathogens; and meaningfully translate results to public health practitioners, healthcare providers, and policymakers.