HSI Spring 2021 Research Updates

Progress reports on HSI-funded research by Professors Jónas Jónasson, Nikos Trichakis, Georgia Perakis, Erin Kelly, and  Catherine Tucker.

Identifying High-Risk Patients with Substance Use Disorder

Professors Jónas Jónasson and Nikos Trichakis are continuing their opioid use disorder work at the Staten Island Performing Provider System (SI PPS). In 2018, Staten Island’s 28.7 per 100,000 opioid death rate was the highest of all the five boroughs of New York City and nearly double the country’s rate of 14.6 deaths per 100,000.  Last semester, Jonasson and Trichakis reported the results of their initial analyses, which aimed to identify patients at the highest risk of opioid overdose.

For Spring 2021, the researchers expanded the model to include all adverse opioid events, such as dependence, abuse and death. Now, their algorithm can identify the top 1% of the highest risk patients, which includes 68% of adverse opioid events. Addiction treatment and healthcare organizations can target selected interventions on those patients most at risk. This is fantastic news for resource-constrained or limited resource agencies.

Even better, the researchers tested their model under real-world conditions and confirmed that their model can easily handle data limitations such as a lag in obtaining data. The researchers note that effective treatment means both identifying patients and implementing effective interventions. Their research focuses on the former, not the latter. Yet, their work on identification solves one of the most difficult issues for healthcare organizations serving this population.

Recommending Optimal Treatment Paths

Professor Georgia Perakis continues to lead her research team (consisting of Professor Dessislava Pachamanova, and PhD students Amine Bennouna, Omar Skali Lami, and Asterios Tsiourvas) in refining their algorithm for predicting the optimal treatment path for patients in the healthcare setting. Their model is interpretable by a healthcare clinician, meaning it suggests an optimal series of steps for a specific cohort of patients. After analyzing the theoretical framework and proofs in the Fall of 2020, the team’s goals for the Spring of 2021 include studying the performance of the algorithm on real patient data.

In 2020, they reported the successful initial test with two data points with diabetes patients to recommend the optimal insulin dose. Now, the research team has extended the model to deal with real-world issues such as incomplete data sets. They intend to test the enhanced model with a much larger, more complex data set, MIMIC-III, which is a comprehensive database of nearly 90,000 patient admissions to critical care units between 2001 and 2012. 

There are other algorithms that try to predict patient risk and optimal treatment. However, this team’s model’s interpretability is a key differentiator that makes it more likely to be adopted in practice. Applying their work to real patient data has the potential to change how patients are treated and save lives.

Improving Worker Wellbeing

More emphasis is being placed on employee population health – both mental and physical – now than before the pandemic. This project recognizes work as one of the social determinants of health and has the potential to improve population health by changing workplace policies and practices in deliberate and meaningful ways. 

Professor Erin Kelly and her team are partnering with the e-commerce division of a national retail firm to evaluate a participatory workplace intervention and its impact on both fulfillment center employees’ health and wellbeing, and on key organizational outcomes. Specifically, the researchers are testing whether workers in a fulfillment center that establishes a Health and Well-being Committee (HaWC), composed of frontline workers and middle managers who work together to address workplace concerns, fare better than their counterparts in other fulfillment centers at the same company. 

Since the start of 2021, the team co-developed the HaWC intervention with managers’ and frontline workers’ input, which was focused on improving the work environment to support both physical and mental health. Through the process of raising and prioritizing concerns at the HaWC meetings, the committee decided on developing a building-wide music playlist, drawing on music recommendations from the frontline workers, as their first project. The committee saw this project as a quick win with high visibility and plan to use it to build momentum to take on more complex concerns in the future. In addition to this project, the HaWC revived the monthly safety audit, which had not been conducted since March 2020.  

Following the success of this pilot, the researchers are now focusing on refining and formalizing the design and implementation procedures. Based on what they have learned, they are developing a three-session training for the co-leads that will lay out the procedures and equip co-leads with facilitation skills to start and run a successful committee. Embedded in this design are practical steps for keeping the local managers apprised of and supporting the improvement projects that the HaWC is prioritizing while maintaining a highly participatory process that encourages more worker involvement. 

COVID-19 Reporting Delays May Skew Policy

COVID-19 data collection, verification and forecasting have been at the center of several research projects. Policy decisions are based, in part, on data so problems with data accuracy may result in misguided policy. Professor Catherine Tucker adds to this discussion with her recent work focused on delays in reporting, specifically the difference in time between when a COVID case is diagnosed and when it is reported.  

Tucker presents significant reporting delays in state-level COVID-19 data and shows the diversity of delays across states and over time. It is not enough, however, to prove that there are reporting delays. For this work to have an impact on policy, reporting delays need to have some sort of deleterious effect. Tucker makes a point of addressing this question as well.  

Tucker focused on six policies: stay-at-home/shelter-in-place orders, mandating face mask use by all individuals in public spaces, closing non-essential businesses, closing K-12 schools, closing restaurants except take-out, and closing bars. Restaurants and bars are regarded as essential businesses in many states and therefore they are treated as separate variables.  

Her research shows that accounting for reporting delays in statistical analysis leads to sharp changes in estimated causal effects of almost every policy investigated, including mask mandates, closing businesses and K-12 schools, and shelter-in-place orders. Therefore, it is critical to account for reporting delays when considering whether to implement COVID-19 restrictions as well as to evaluate the effectiveness of those policies. Overall, this work shows that it is imperative to consider data management practices during pandemics. Moreover, the results from this study also offer broader insights for thinking generally about the consequences of delays in data inputs for information systems.