The project's goal is to help local health departments in the United States target their testing efforts to identify the most vulnerable people to lead poisoning.
This is a terminal based program that provides a ranking of lead poisoning risk among children in a user chosen jurisdiction. The program allows the user to focus on regions from the State level to the Census block group level. A text file of the ranked regions and a map is provided.
Reduce-the-lead relies on some libraries
If you want to download data yourself, you will need a US Census API key.
You will also need to create a file called census_api.ini file. In
this file, do this
[census_api]
key: <<YOUR-CENSUS-KEY>>
After creating the census_api.ini file, you can run
data_download.py to do its magic. Or you can unzip data.zip.
To run Reduce-the-lead, simply type python reducethelead.py at the
command line.
To see how reducethelead.py operates, look at the
example_reducingthelead.ipynb IPython notebook.
The motivation for this analysis is the point that in the United States there is an inadequate organisation of testing. Two examples highlight this point. First, 21 States do not provide information to the Center for Disease Control and Prevention (CDC). Second, of those that do send information, the information sent is inadequate. It is now acknolowedged that there is no safe threshold BLL in children. Many states only send BLLs for children above thresholds. We need to test and organise information better.
I rely on the CDC to determine which groups are at risk.
The CDC defines children from the following groups as at a higher risk
- poor
- members of racial-ethnic minority groups
- recent immigrants
- live in older, poorly maintained rental properties
- refugees
- children adopted from other countries
- have parents who are exposed to lead at work, or
- pregnant woman
In this analysis, I will collect data that best approximates these groups.
The data comes from the five year US Census American Community Survey ending in 2013. This data source provides the highest possible spatial resolution, allowing for ranking of Census block groups. A typical block group contains between 600 and 3000 people. The cost is that the information is not immediate. The data can be interpreted as five year averages. Reduce-the-lead results should be used with caution in rapidly changing communities.
In this version, I have data that approximates the first six of these groups. Data on parents' jobs or pregnancy was not available at a low enough spatial resolution.
A problem with providing information along 6 dimensions is that we're unlikely to obtain a single ranking of districts to visit. As a first attempt at resolving this, I will provide a ranking based on the linear combination of these dimensions that explains most of the variation in the data. Specifically, I'll take the first component of a principal component analysis and rank regions. This is not a 'best' ranking, the aim is to start the conversation.
Reduce-the-lead can augment the work of health departments that have good testing procedures. For instance, it is plausible that testing may miss some groups at risk. In addition, if we obtained data from regions with good testing, we could validate the performance of this analysis. Specifically, we could measure the performance of the maps: how accurate was Reduce-the-lead? How sensitive was Reduce-the-lead? This information will help improve Reduce-the-lead and in turn, improve targeting of lead poisoning tests at the local level.