Data-Driven Justice Initiative

Data-Driven Justice Initiative: Identifying Frequent Users of Multiple Public Systems for More Effective Early Assistance


The Challenge

Local governments across the country have recognized that a number of individuals repeatedly cycle through jails, emergency rooms, mental-health facilities, and social services, with poor outcomes for the individuals and high costs to the local government. Typically, these individuals struggle with multi-dimensional problems — homelessness, mental illness, substance abuse, and chronic health conditions — but they are not tracked across systems, making it difficult to identify these individuals and provide coordinated care. Early identification of vulnerable individuals and cross-system initiatives to provide comprehensive treatment would reduce their total service needs and prevent future incarceration.

How Can Data Driven Approaches Help?

By combining data from multiple separate systems, officials can understand and address each individual’s challenges and underlying conditions. By building predictive models on top of combined data, officials can identify persons at risk early enough to intervene. This has two benefits:

  • Better Resource Allocation: Enable jurisdictions to target scarce resources to individuals who most need additional support by giving each individual a risk score to intervene before long-term harm is done (e.g. before getting a jail record).
  • Reduce future interactions: Proactively target relevant interventions to reduce the probability that an individual has a future interaction with public systems. One example of a data-driven approach in Johnson County, Kansas would be providing mental health co-responders and case managers a list of individuals in need of support to reach out to them proactively, before a “crisis intervention” response

Our Solution

[Video]

An accurate Early Intervention System (EIS) will efficiently identify individuals at risk of contact with public systems so local agencies can provide appropriate services and interventions. By analyzing a jurisdiction’s mental health, criminal justice, emergency medical response, and social services data, a data-driven EIS can help target resources for individuals who would benefit most from those interventions. This system can:

  • Identify which individuals are most likely to enter or re-enter the criminal justice system, the mental health system, the social services system, or receive emergency services
  • Identify which individuals are cycling through combinations of these systems

What We’ve Done

DSaPP has partnered with Johnson County and Salt Lake County to build a prototype Early Intervention System (EIS) for vulnerable individuals who cycle through multiple public systems. By combining data across multiple systems, we are able to match individuals across multiple data sets and identify which individuals have already interacted with multiple systems. We have also identified common sequences of interactions that occur and are leading indicators that predict interaction with multiple systems.

We have developed predictive models that assign risk scores of making future contact with another system to individuals who have made contact with one system. These models will produce ranked lists of individuals at risk who may receive follow-up care or interventions. We hope that the models will also provide proactive risk warnings at points of contact (e.g., EMS dispatch, jail bookings).

Early Results

By matching data across multiple systems from Johnson County, Kansas, a municipality of approximately 575,000 people, we were able to do early analysis on the individuals interacting with multiple systems and the negative outcomes associated. 127,000 individuals in Johnson County have had an interaction with at least one of MedAct (Johnson’s County’s EMS), County Mental Health Services, or the County Criminal Justice System. Of those individuals, there are over 12,000 people that intersect with two or all three departments; 4,430 have encountered Mental Health and the criminal justice systems, and over 1,000 have interacted with all three systems. Looking solely at EMS, fewer than 40% of the individuals who call EMS have called EMS more than once over the period of our data. For individuals who have interacted with EMS, Criminal Justice, and Mental Health, more than 60% have more than one EMS call. 20% of the individuals who have interacted with all three systems have 5 or more EMS calls.

We generated a list of the 200 people most at risk of entering the criminal justice system. If, in 2015, the Johnson County co-responders had called the individuals on the list, 1 in 2 calls they made would have put them in contact with someone who ended up going to jail in 2015. Within that list of 200 people, 104 went to jail, for an average of two months each, for a total of 6,987 days between those 104 individuals. On average, it had been 2 years since these individuals’ last contact with the mental health center at the time of their bookings. The minimum cost to the county for this jail time, based on incremental costs per person, is $255,44.72. This presents an opportunity for Johnson County to prevent jail time and reinvest the money saved in preventative measures.

What We’re Doing

  • Improving our models – We’re adding new predictors and new algorithms to the DDJ toolkit.
  • Improving the code – We are improving the speed, performance, and reliability of the code we’ve written for this project. We are also creating a common schema that will make it easier for additional jurisdictions to use our work and improving our modeling webapp to help our partners understand how model performance compares.
  • Adding data from Johnson County and Salt Lake County – Each jurisdiction is gathering new data for the models.

What We Will Do

  • Field Validation – We will work with Johnson County to validate the model’s predictions.
  • Add jurisdictions – We’re looking for new jurisdictions to join the project. Adding jurisdictions helps us build better tools and helps our partners learn valuable lessons from one another.

There are many challenges and barriers for potential partners to consider with this work, but the potential payoff is large and we are willing to help solve them as much as possible:

  • Legal – To share the data with the University and within the county, there are legal issues and agreements to be addressed; Healthcare and Mental Health data is particularly sensitive.
  • Organizational – Often, data and interventions sit in different organizations within the county, with different stakeholders. Many stakeholders must co-operate to be successful, including leadership, database owners,and the front line individuals responsible for data collection and intervention.
  • TechnicalCounties have limited technology resources, both infrastructure and individuals who support the data. Matching individuals and de-duplicating records across systems is non-trivial.

Ready to Contact Us?

We have posted several resources to help you prepare to join the project:

If you think you fit, please let us know.