Chicago Lead Poisoning Prevention2018-10-17T21:36:27+00:00

    

Lead Poisoning Prevention

Partner(s): Chicago Department of Public Health, Chicago Department of Innovation and Technology, AllianceChicago
Status: Completed July 2018
Github repo: https://github.com/Chicago/lead-model
Team: Eric Potash, Avishek Kumar, Joe Walsh, and Rayid Ghani

Summary
Lead poisoning afflicts hundreds of thousands of children in the United States every year, leading to decreased IQ points, increased prevalence of ADHD, and irreversible neurological consequences. A common approach to identifying lead hazards is to test all children for elevated blood lead levels and then investigate and remediate the homes of children with elevated tests. This can prevent lead exposure for future residents, but only after the initial child has been poisoned. DSaPP developed a machine learning-based predictive system that identifies a child’s risk of lead poisoning before it occurs so that it can be prevented. Healthcare providers are using DSaPP’s system to intervene on young patients during clinical visits through the Lead Safe API, and the City of Chicago is using it to conduct preventive inspections.  

Background
Lead poisoning imposes lifelong health and economic costs on hundreds of thousands of people every year in the United States. Exposure to lead is associated with premature birth, edema, herniation, atrophy and white-matter degeneration, and elevated blood lead levels (BLL) are associated with lower IQ in children as well as with poorer achievement on reading and math standardized tests. Lead-related child health issues conservatively bear a price tag of over $40 billion annually. Despite these significant costs, the current approach to identifying and remediating lead hazards is to test all children for elevated blood lead levels and then investigate and remediate the indicated homes. This helps prevent exposure for future residents but comes too late for poisoned children, who disproportionately come from low socioeconomic status and minority groups.

Scope
This project enables health officials to move from reacting to lead-poisoning to prevention by remediating hazards before children get poisoned. The DSaPP model provides accurate and actionable risk scores for children or homes so departments of public health and clinicians can use their limited resources to help the children at highest risk of lead poisoning before they get poisoned. The model can produce granular risk assessments for targeting interventions, from providing informational brochures (cheap and easy to provide to almost everyone) to remediating homes (costly and difficult to provide to many), depending on available resources. This prevention stance will serve to better level the playing field by remediating a core determinant of health for low-income and at-risk children nationwide.

Data
DSaPP’s model uses five types of data to generate risk scores:

  • Individual blood lead level test results. These are typically available from the STELLAR or HLPPS databases run by state departments of public health.
  • Home lead inspection records. These include the inspectors’ findings as well as the date of inspection and date of compliance.
  • WIC records. We can’t estimate risk for children we don’t know exist. CDPH uses WIC records to find potentially vulnerable children before they become mobile and start putting things in their mouths.
  • Information about each house or apartment building in the jurisdiction, such as address, type of building, and age of the structure.
  • American Community Survey data from the US Census.

The data need to cover at least three to five years and be stored electronically.

Analysis
DSaPP tested many models with varying combinations of data and algorithms. Testing simulated the past, using only the information that would have been available at the time to make a prediction to see how accurate the predictions would have been. DSaPP selected the best model based on the proportion of children with risk scores in the top 5% (the proportion that CDPH estimated it could proactively inspect) were later found to have high blood lead levels. We performed a randomized controlled trial with CDPH.

Results
The best model used random forests and provided a three-fold lift over a random baseline (confirmed by the RCT) and a two-fold lift over a policy targeting the community areas with the highest rates of lead poisoning. Although CDPH had blood tests going back to the early 1990s, only recent test results helped, thanks to changing poisoning patterns.

We helped the Chicago Department of Innovation and Technology transfer the model to their servers and build an API for EMRs to securely obtain risk assessments for their young patients. Four AllianceChicago clinics are using the API, with more Chicago-area healthcare providers ready to join. CDPH has used the model to conduct outreach calls and preventive inspections.  

CDPH has described this work as pioneering in the use of machine learning and predictive analytics in public health and as having the potential to have a significant impact on both health and economic outcomes for communities across the US. The Milbank Memorial Fund and AcademyHealth recognized this project with the 2018 State and Local Innovation Prize.

Future Plans and Areas for Improvement
This project is ready to scale. Only one healthcare system was given access to the model’s API while it was being developed, but now that it is deployed, the City will allow more to join.

DSaPP and CDPH have partnered with the Sinai Urban Health Institute to test whether trained community health workers can effectively conduct home lead hazard inspections. CHWs would significantly increase the number of preventive inspections that could be conducted.

Publications
Predictive Modeling for Public Health: Preventing Childhood Lead Poisoning
Eric Potash, Joe Brew, Alexander Loewi, Subhabrata Majumdar, Andrew Reece, Joe Walsh, Eric Rozier, Emile Jorgenson, Raed Mansour, Rayid Ghani. Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2015.

Randomization Bias in Field Trials to Evaluate Targeting Methods.
Eric Potash. Economics Letters, 2018.

Media Coverage
Could Chicago prevent childhood lead poisoning before it happens? Chicago Tribune
1,000 kids diagnosed with lead poisoning in 2014, Fox 32 Chicago

Videos
DSSG 2014 Final Presentation: https://www.youtube.com/watch?v=DbplLXRQquI
DSSG 2014 Project Overview: https://www.youtube.com/watch?v=Fc4IHhJSV3I
DSSG 2018 Update: https://www.youtube.com/watch?v=i_maI-WwahM
Chi Hack Night 2014: https://youtu.be/shBRBWSsBTI?t=1850
Chi Hack Night 2018: https://www.youtube.com/watch?v=FoYR3uz_qro
KDD 2015: https://www.youtube.com/watch?v=iYdHDiR3Klo