Designing Difference-in-Difference Studies with Staggered Treatment Adoption: Key Concepts and Practical Guidelines.

Difference-in-difference (DID) estimators are a valuable method for identifying causal effects in the public health researcher's toolkit. A growing methods literature points out potential problems with DID estimators when treatment is staggered in adoption and varies with time. Despite this, no practical guide exists for addressing these new critiques in public health research. We illustrate these new DID concepts with step-by-step examples, code, and a checklist. We draw insights by comparing the simple 2 × 2 DID design (single treatment group, single control group, two time periods) with more complex cases: additional treated groups, additional time periods of treatment, and treatment effects possibly varying over time. We outline newly uncovered threats to causal interpretation of DID estimates and the solutions the literature has proposed, relying on a decomposition that shows how the more complex DIDs are an average of simpler 2 × 2 DID subexperiments.

Polluting under the Radar: Emissions, Inequality, and Concrete Batch Plants in Houston.

Small industrial sources collectively release large amounts of pollution, including particulate matter (PM) that contributes to air quality problems in the United States and elsewhere. We study one such type of industrial facility, concrete batch plants, and analyze PM emissions and siting patterns of 131 plants located in Harris County, Texas. We find that concrete batch plants in Harris County are collectively a major pollution source, contributing between 38 and 111 tons of primary PM2.5 emissions (between 26%-76% of PM2.5 from the median Texas oil refinery) and between 109 and 493 tons of primary PM10 emissions (between 64%-290% of PM10 from the median refinery). Estimates from an integrated assessment model suggest that health damages from the PM2.5 emissions alone amount to $29 million annually, reflecting two additional premature deaths per year. We further find that concrete batch plants in Harris County are disproportionately located in census tracts with more low-income, Hispanic, and Black populations, thereby raising important environmental justice questions. On the basis of these findings, we argue that small pollution sources require more air quality monitoring and emissions reporting and that regulatory agencies should consider cumulative environmental and health impacts of these sources as part of the permitting process.

Understanding Excess Emissions from Industrial Facilities: Evidence from Texas.

We analyze excess emissions from industrial facilities in Texas using data from the Texas Commission on Environmental Quality. Emissions are characterized as excess if they are beyond a facility's permitted levels and if they occur during startups, shutdowns, or malfunctions. We provide summary data on both the pollutants most often emitted as excess emissions and the industrial sectors and facilities responsible for those emissions. Excess emissions often represent a substantial share of a facility's routine (or permitted) emissions. We find that while excess emissions events are frequent, the majority of excess emissions are emitted by the largest events. That is, the sum of emissions in the 96-100th percentile is often several orders of magnitude larger than the remaining excess emissions (i.e., the sum of emissions below the 95th percentile). Thus, the majority of events emit a small amount of pollution relative to the total amount emitted. In addition, a small group of high emitting facilities in the most polluting industrial sectors are responsible for the vast majority of excess emissions. Using an integrated assessment model, we estimate that the health damages in Texas from excess emissions are approximately $150 million annually.

A multimodal neural network with gradient blending improves predictions of survival and metastasis in sarcoma.

The objective of this study is to develop a multimodal neural network (MMNN) model that analyzes clinical variables and MRI images of a soft tissue sarcoma (STS) patient, to predict overall survival and risk of distant metastases. We compare the performance of this MMNN to models based on clinical variables alone, radiomics models, and an unimodal neural network. We include patients aged 18 or older with biopsy-proven STS who underwent primary resection between January 1st, 2005, and December 31st, 2020 with complete outcome data and a pre-treatment MRI with both a T1 post-contrast sequence and a T2 fat-sat sequence available. A total of 9380 MRI slices containing sarcomas from 287 patients are available. Our MMNN accepts the entire 3D sarcoma volume from T1 and T2 MRIs and clinical variables. Gradient blending allows the clinical and image sub-networks to optimally converge without overfitting. Heat maps were generated to visualize the salient image features. Our MMNN outperformed all other models in predicting overall survival and the risk of distant metastases. The C-Index of our MMNN for overall survival is 0.77 and the C-Index for risk of distant metastases is 0.70. The provided heat maps demonstrate areas of sarcomas deemed most salient for predictions. Our multimodal neural network with gradient blending improves predictions of overall survival and risk of distant metastases in patients with soft tissue sarcoma. Future work enabling accurate subtype-specific predictions will likely utilize similar end-to-end multimodal neural network architecture and require prospective curation of high-quality data, the inclusion of genomic data, and the involvement of multiple centers through federated learning.

Extraction and classification of structured data from unstructured hepatobiliary pathology reports using large language models: a feasibility study compared with rules-based natural language processing.

AIMS: Structured reporting in pathology is not universally adopted and extracting elements essential to research often requires expensive and time-intensive manual curation. The accuracy and feasibility of using large language models (LLMs) to extract essential pathology elements, for cancer research is examined here. METHODS: Retrospective study of patients who underwent pathology sampling for suspected hepatocellular carcinoma and underwent Ytrrium-90 embolisation. Five pathology report elements of interest were included for evaluation. LLMs (Generative Pre-trained Transformer (GPT) 3.5 turbo and GPT-4) were used to extract elements of interest. For comparison, a rules-based, regular expressions (REGEX) approach was devised for extraction. Accuracy for each approach was calculated. RESULTS: 88 pathology reports were identified. LLMs and REGEX were both able to extract research elements with high accuracy (average 84.1%-94.8%). CONCLUSIONS: LLMs have significant potential to simplify the extraction of research elements from pathology reporting, and therefore, accelerate the pace of cancer research.

Can policy affect initiation of addictive substance use? Evidence from opioid prescribing.

Drug control policy can have unintended consequences by pushing existing users to alternative, possibly more dangerous substances. Policies that target only new users may therefore be especially promising. Using commercial insurance claims data, we provide the first evidence on a set of new policies intended to reduce opioid initiation in the form of limits on initial prescription length. We also provide the first evidence on the impact of must-access prescription drug monitoring programs (MA-PDMPs), laws that do not target new users, on initial opioid use. Although initial limit policies reduce the average length of initial prescriptions, they do so primarily by raising the frequency of short prescriptions, resulting in increases in opioids dispensed to new users. In contrast, we find that MA-PDMPs reduce opioids dispensed to new users, even though they do not explicitly set out to do so. Neither policy significantly affects extreme use such as doctor shopping among new patients, because such behavior is very rare.

Care management reduced infant mortality for Medicaid managed care enrollees in Ohio.

OBJECTIVES: In 2012, the Ohio Department of Medicaid introduced requirements for enhanced care management to be delivered by Medicaid managed care organizations (MCOs). This study evaluated the impact of care management on reducing infant mortality in the largest Medicaid MCO in Ohio. STUDY DESIGN: Observational study using infant and maternal individual-level enrollment and claims data (2009-2015), which used a quasi-experimental research design built on a sibling-comparison approach that controls for within-family confounders. METHODS: Using individual-level data from the largest MCO in Ohio, we estimated linear probability models to examine the effect of infant engagement in care management on infant mortality. We used a within-family fixed-effects research design to determine if care management reduced infant mortality and estimated models separately for healthy infants and nonhealthy infants. RESULTS: Infant engagement in care management was associated with a reduction of 7.4 percentage points (95% CI, -10.7 to -4.1; P <.001) in infant mortality among the most vulnerable infants, those identified as not well at birth. This effect was larger in recent years and likely driven by new statewide enhanced care management requirements. Infant mortality was unchanged for healthy infants engaged in care management (coefficient = 0.03; 95% CI, -0.01 to 0.08). CONCLUSIONS: This study provides evidence that care management can be effective in reducing infant mortality among Medicaid MCO enrollees, a population at high risk of mortality. Few infants were engaged in care management, suggesting to policy makers that there is room for many additional infants to benefit from this intervention.

Association Between Medicaid Expansion and Rates of Opioid-Related Hospital Use.

Importance: The rate of opioid-related emergency department (ED) visits and inpatient hospitalizations has increased rapidly in recent years. Medicaid expansions have the potential to reduce overall opioid-related hospital events by improving access to outpatient treatment for opioid use disorder. Objective: To examine the association between Medicaid expansions and rates of opioid-related ED visits and inpatient hospitalizations. Design, Setting, and Participants: A difference-in-differences observational design was used to compare changes in opioid-related hospital events in US nonfederal, nonrehabilitation hospitals in states that implemented Medicaid expansions between the first quarter of 2005 and the last quarter of 2017 with changes in nonexpansion states. All-payer ED and hospital discharges from 45 states in the Healthcare Cost and Utilization Project FastStats were included. Exposures: State implementation of Medicaid expansions between 2005 and 2017. Main Outcomes and Measures: Rates of all opioid-related ED visits and inpatient hospitalizations, measured as the quarterly numbers of treat-and-release ED discharges and hospital discharges related to opioid abuse, dependence, and overdose, per 100 000 state population. Results: In the 46 states and District of Columbia included in the study, 1524 observations of emergency department data and 2219 observations of opioid-related inpatient hospitalizations were analyzed. The post-2014 Medicaid expansions were associated with a 9.74% (95% CI, -18.83% to -0.65%) reduction in the rate of opioid-related inpatient hospitalizations. There appeared to be no association between the pre-2014 or post-2014 Medicaid expansions and the rate of opioid-related ED visits (post-2014 Medicaid expansions, -3.98%; 95% CI, -14.69% to 6.72%; and pre-2014 Medicaid expansions, 1.02%; 95% CI, -5.25% to 7.28%). Conclusions and Relevance: Medicaid expansion appears to be associated with meaningful reductions in opioid-related hospital use, possibly attributable to improved care for opioid use disorder in other settings.

Gains in health insurance coverage explain variation in Democratic vote share in the 2008-2016 presidential elections.

In the last decade, health care reform has dominated U.S. public policy and political discourse. Double-digit rate increases in premiums in the Health Insurance Marketplaces established by the Affordable Care Act (ACA) in 2018 make this an ongoing issue that could affect future elections. A seminal event that changed the course of policy and politics around health care reform is the 2016 presidential election. The results of the 2016 presidential election departed considerably from polling forecasts. Given the prominence of the Affordable Care Act in the election, we test whether changes in health insurance coverage at the county-level correlate with changes in party vote share in the presidential elections from 2008 through 2016. We find that a one-percentage-point increase in county health insurance coverage was associated with a 0.25-percentage-point increase in the vote share for the Democratic presidential candidate. We further find that these gains on the part of the Democratic candidate came almost fully at the expense of the Republican (as opposed to third-party) presidential candidates. We also estimate models separately for states that did and did not expand Medicaid and find no differential effect of insurance gains on Democratic vote share for states that expanded Medicaid compared to those that did not. Our results are consistent with the hypothesis that outcomes in health insurance markets played a role in the outcome of the 2016 presidential election. The decisions made by the current administration, and how those decisions affect health insurance coverage and costs, may be important factors in future elections as well.

Macroeconomic conditions and opioid abuse.

We examine how deaths and emergency department (ED) visits related to use of opioid analgesics (opioids) and other drugs vary with macroeconomic conditions. As the county unemployment rate increases by one percentage point, the opioid death rate per 100,000 rises by 0.19 (3.6%) and the opioid overdose ED visit rate per 100,000 increases by 0.95 (7.0%). Macroeconomic shocks also increase the overall drug death rate, but this increase is driven by rising opioid deaths. Our findings hold when performing a state-level analysis, rather than county-level; are primarily driven by adverse events among whites; and are stable across time periods.

Opposition to Obamacare: A Closer Look.

Prior telephone surveys have reported two main reasons for opposition to the Affordable Care Act (ACA): distrust of government and opposition to the universal coverage mandate. The authors set out to elucidate the reasons for this opposition. This article describes how the authors used qualitative methods with semistructured interviewing as a principal investigative method to gather information from people they met while bicycling across the United States from April through July 2016. During this time, the authors conducted open-ended, semistructured conversations with people they met as they rode their bicycles from Washington, DC, to Seattle, Washington. Informants were chosen as a convenience sample. One hundred sixteen individuals participated as informants. The majority of comments were negative toward the ACA. Conversations were categorized into four themes, which included the following: (1) The ACA has increased the cost of health insurance; (2) government should not tell people what to do; (3) responsibility for ACA problems is diffuse; and (4) the ACA should not pay for other people's problems. These face-to-face conversations indicated that opposition to the ACA may be due to the fact that many Americans have experienced an increase in the cost of insurance either through increased premiums or greatly increased deductibles. They blame this increase in cost on the ACA, President Obama, the government or insurance companies, and the inclusion of "others" in insurance plans. The authors discuss how these findings can influence medical education curricula to better prepare future physicians to discuss health policy issues with patients.