Developing and Implementing Predictive Models in a Learning Healthcare System: Traditional and Artificial Intelligence Approaches in the Veterans Health Administration.

Predicting clinical risk is an important part of healthcare and can inform decisions about treatments, preventive interventions, and provision of extra services. The field of predictive models has been revolutionized over the past two decades by electronic health record data; the ability to link such data with other demographic, socioeconomic, and geographic information; the availability of high-capacity computing; and new machine learning and artificial intelligence methods for extracting insights from complex datasets. These advances have produced a new generation of computerized predictive models, but debate continues about their development, reporting, validation, evaluation, and implementation. In this review we reflect on more than 10 years of experience at the Veterans Health Administration, the largest integrated healthcare system in the United States, in developing, testing, and implementing such models at scale. We report lessons from the implementation of national risk prediction models and suggest an agenda for research.

The effects of department of Veterans Affairs medical centers on socio-economic outcomes: Evidence from the Paycheck Protection Program.

Do medical facilities also help advance improvements in socio-economic outcomes? We focus on Veterans, a vulnerable group over the COVID-19 pandemic who have access to a comprehensive healthcare network, and the receipt of funds from the Paycheck Protection Program (PPP) between April and June as a source of variation. First, we find that Veterans received 3.5% more loans and 6.8% larger loans than their counterparts (p < 0.01), controlling for a wide array of zipcode characteristics. Second, we develop models to predict the number of PPP loans awarded to Veterans, finding that the inclusion of local VA medical center characteristics adds almost as much explanatory power as the industry and occupational composition in an area and even more than the education, race, and age distribution combined. Our results suggest that VA medical centers can play an important role in helping Veterans thrive even beyond addressing their direct medical needs.

When houses of worship go empty: The effects of state restrictions on well-being among religious adherents

Using data on over 50,000 individuals between March 2020 and May 2021 in the United States, this paper investigates the effects of state-level house of worship restrictions on subjective well-being (SWB). My identification strategy exploits plausibly exogenous variation in the timing of these policies on religious adherents with their non-religious counterparts before versus after the adoption of the state restrictions. The adoption of these restrictions led to a 0.117 standard deviation reduction in current life satisfaction and a 4.8 percentage point rise in self-isolation among the religious, relative to their counterparts. Numeric caps (i.e., restrictions on exactly how many people can gather) are more harmful for SWB than percentage caps (i.e., restrictions on the percent of occupancy rates during “normal times” as set out for the building). The results are robust to a wide array of controls, including income, political affiliation, economic sentiment, industry, and occupation. Moreover, they are robust to state × time fixed effects, which exploit variation between religious and non-religious adherents after controlling for all shocks common in the same state over time. Finally, there is almost no evidence that these restrictions have any public health benefits.

Sectoral digital intensity and GDP growth after a large employment shock: A simple extrapolation exercise

We introduce a state‐dependent algorithm with minimal data requirements for predicting output dynamics as a function of employment across industries and locations. The method generalizes insights of Okun (1963) by leveraging measures of industry heterogeneity. We use the algorithm to examine gross domestic product (GDP) dynamics following the COVID‐19 pandemic of 2020, delivering informative projections of aggregate and sectoral output. Because the pandemic curtailed the ability to perform certain tasks at work, our application examines whether greater reliance on digital technologies can mediate employment and productivity losses. We use industry‐level indices of digital task intensity and ability to work from home, together with publicly available data on employment and GDP for Canada, to document that: (i) employment responses after the shock's onset are milder in digitally intensive sectors and (ii) conditional on the size of employment changes, GDP responses are less extreme in digitally intensive sectors. Our projections indicate a return to pre‐crisis aggregate output within eight quarters of the initial shock with significant heterogeneity in recovery patterns across sectors.

Correction: How social capital helps communities weather the COVID-19 pandemic.

[This corrects the article DOI: 10.1371/journal.pone.0245135.].

Does marriage protect mental health? Evidence from the COVID-19 pandemic.

Using weekly variation from April 23 to June 23 2020, we exploit the surge in unemployment over the coronavirus pandemic to identify the effects on mental health outcomes and the role of marital status as a protective factor for households. We find that married respondents are 1-2 percentage points less likely, relative to their unmarried counterparts, to experience mental health problems following declines in work-related income since the start of the pandemic. Our results suggest that the combination of intrafamily substitution and the psychological benefits of marriage helps insure against unanticipated fluctuations in job and income loss.

Designing COVID-19 mortality predictions to advance clinical outcomes: Evidence from the Department of Veterans Affairs.

Using administrative data on all Veterans who enter Department of Veterans Affairs (VA) medical centres throughout the USA, this paper uses artificial intelligence (AI) to predict mortality rates for patients with COVID-19 between March and August 2020. First, using comprehensive data on over 10 000 Veterans' medical history, demographics and lab results, we estimate five AI models. Our XGBoost model performs the best, producing an area under the receive operator characteristics curve (AUROC) and area under the precision-recall curve of 0.87 and 0.41, respectively. We show how focusing on the performance of the AUROC alone can lead to unreliable models. Second, through a unique collaboration with the Washington D.C. VA medical centre, we develop a dashboard that incorporates these risk factors and the contributing sources of risk, which we deploy across local VA medical centres throughout the country. Our results provide a concrete example of how AI recommendations can be made explainable and practical for clinicians and their interactions with patients.

Pandemic response policies' democratizing effects on online learning.

The COVID-19 pandemic has changed peoples' lives in unexpected ways, especially how they allocate their time between work and other activities. Demand for online learning has surged during a period of mass layoffs and transition to remote work and schooling. Can this uptake in online learning help close longstanding skills gaps in the US workforce in a sustainable and equitable manner? We answer this question by analyzing individual engagement data of DataCamp users between October 2019 and September 2020 (n = 277,425). Exploiting the staggered adoption of actions to mitigate the spread of COVID-19 across states, we identify the causal effect at the neighborhood level. The adoption of nonessential business closures led to a 38% increase in new users and a 6% increase in engagement among existing users. We find that these increases are proportional across higher- and lower-income neighborhoods and neighborhoods with a high or low share of Black residents. This demonstrates the potential for online platforms to democratize access to knowledge and skills that are in high demand, which supports job security and facilitates social mobility.

How Much Does the (Social) Environment Matter? Using Artificial Intelligence to Predict COVID-19 Outcomes with Socio-demographic Data.

While the coronavirus pandemic has affected all demographic brackets and geographies, certain areas have been more adversely affected than others. This paper focuses on Veterans as a potentially vulnerable group that might be systematically more exposed to infection than others because of their co-morbidities, i.e., greater incidence of physical and mental health challenges. Using data on 122 Veteran Healthcare Systems (HCS), this paper tests three machine learning models for predictive analysis. The combined LASSO and ridge regression with five-fold cross validation performs the best. We find that socio-demographic features are highly predictive of both cases and deaths-even more important than any hospital-specific characteristics. These results suggest that socio-demographic and social capital characteristics are important determinants of public health outcomes, especially for vulnerable groups, like Veterans, and they should be investigated further.

The impact of COVID-19 on the U.S. child care market: Evidence from stay-at-home orders.

Stay-at-home orders (SAHOs) were implemented in most U.S. states to mitigate the spread of COVID-19. This paper quantifies the impact of these containment policies on a measure of the supply of child care. The supply of such services may be particularly vulnerable to a SAHO-type policy shock, given that many providers are liquidity-constrained. Using plausibly exogenous variation from the staggered adoption of SAHOs across states, we find that online job postings for early care and education teachers declined by 16% after enactment. This effect is driven exclusively by private-sector services. Indeed, hiring by public programs like Head Start and pre-kindergarten has not been influenced by SAHOs. We also find that ECE job postings increased dramatically after SAHOs were lifted, although the number of such postings remains 4% lower than that during the pre-pandemic period. There is little evidence that child care search behavior among households was altered by SAHOs. Because forced supply-side changes appear to be at play, our results suggest that households may not be well-equipped to insure against the rapid transition to the production of child care. We discuss the implications of these results for child development and parental employment decisions.

How social capital helps communities weather the COVID-19 pandemic.

Why have the effects of COVID-19 been so unevenly geographically distributed in the United States? This paper investigates the role of social capital as a mediating factor for the spread of the virus. Because social capital is associated with greater trust and relationships within a community, it could endow individuals with a greater concern for others, thereby leading to more hygienic practices and social distancing. Using data for over 2,700 US counties, we investigate how social capital explains the level and growth rate of infections. We find that moving a county from the 25th to the 75th percentile of the distribution of social capital would lead to a 18% and 5.7% decline in the cumulative number of infections and deaths, as well as suggestive evidence of a lower spread of the virus. Our results are robust to many demographic characteristics, controls, and alternative measures of social capital.

Did the Paycheck Protection Program Hit the Target?

From the Abstract: This paper takes an early look at the Paycheck Protection Program (PPP), a large and novel small business support program that was part of the initial policy response to the COVID-19 [coronavirus disease 2019] pandemic. We use new data on the distribution of the first round of PPP loans and high-frequency microlevel employment data to consider two dimensions of program targeting. First, we do not find evidence that funds flowed to areas more adversely affected by the economic effects of the pandemic, as measured by declines in hours worked or business shutdowns. If anything, funds flowed to areas less hard hit. Second, we find significant heterogeneity across banks in terms of disbursing PPP funds, which does not only reflect differences in underlying loan demand. The top-4 banks alone account for 36% of total pre-policy small business loans, but disbursed less than 3% of all PPP loans in the first round. Areas that were significantly more exposed to low PPP banks received much lower loan allocations. We do not find evidence that the PPP had a substantial effect on local economic outcomes--including declines in hours worked, business shutdowns, initial unemployment insurance claims, and small business revenues--during the first round of the program. Firms appear to use first round funds to build up savings and meet loan and other commitments, which points to possible medium-run impacts. As data become available, we will continue to study employment and establishment responses to the program and the impact of PPP support on the economic recovery. Measuring these responses is critical for evaluating the social insurance value of the PPP and similar policies.COVID-19 (Disease)