Counting 3,146 Big Data & Machine Learning Frameworks, Toolsets, and Examples...
Suggestion? Feedback? Tweet @stkim1

Author
Contributors
Last Commit
Jul. 3, 2018
Created
May. 30, 2018

Mortality Explorer

Explore US Mortality data using the GRAND stack. Work on this is ongoing and contributions are welcome!

Current status:

  • Data from 2015 is now modeled in Neo4j
  • Data is validated and errors logged
  • Need mortality data from other years
  • Need to link ICD values in mortality CSV to ICD nodes
  • Currently building GraphQL Schema and node server
  • No work done on React front end yet

Data Sources

Original data set is from this Kaggle project, which in turn is based on public CDC data from 2015.

ICD codes are taken from this Github repo and are from 2018. 2015 codes are available here but only in XML form. I will convert to CSV only if the 2018 codes have changed significantly enough to make the 2015 mortality data problematic.

Technical details

Data model

NOTE: the data file 2015_data.csv is too large to store in this repo. Please download it from this Kaggle project.

The data itself is rendered as a graph in Neo4j version 3.4. APOC plugin is required. Current strategy is to render all mortality dimensions (e.g. sex, race, education level) as specific "dimension nodes", rather than properties. Each "Death" node links to dimension nodes via specific relationships. For example:

(:Death) -[:HAS_SEX]-> (:Sex {key: "M", value: "Male"})

The general form is:

(:Death) -[:HAS_<upper-cased dimension>]->(:<Capitalized dimension> {key: <key>, value: "<value>"})

Note: when the dimension name starts with a number (e.g. 39_cause_recode), we prefix it with an underscore when we create the label for the dimension node.

I'm doing it this way because just having a huge list of Death nodes with all attributes stored as properties is too similar to what it would look like in a relational DB, and doesn't leverage the advantages of a graph DB. The relational model might ultimately be the best way to store and query this data, but then I wouldn't get to play with Neo4J :-)

Identifying individual death nodes

One challenge is that there is no uniquely identifying data for a given death in the data files. As Neo4J does not provide any sort of auto-increment, we need to reference the line number in the CSV as the unique identifier. This is why the apoc.load.csv call is required to load the data, rather than LOAD CSV Cypher construct.

Logging misses

The dimension data comes from Kaggle as well (see data/2015_dimensions.json). ICD codes are also externally sourced. This means that as we import the death data, references to this external data may fail either due to errors in the data or because the external references don't exist.

In either case, we want to know if our death data is failing to create a relationship to a dimension node.

Example queries

Count how many people died of emphysema:

MATCH (:_358_cause_recode {key: 266})<-[r:HAS_358_CAUSE_RECODE]-(:Death) RETURN count(r);

Count how many men vs women died of emphysema:

MATCH (:_358_cause_recode {key: 266})<-[:HAS_358_CAUSE_RECODE]-(:Death)-[r:HAS_SEX]->(s) RETURN s, count(r) AS count ORDER BY count DESC;

Count of different activity codes for homicides/suicides:

MATCH (:Manner_of_death {value: "Homicide"})<-[:HAS_MANNER_OF_DEATH]-(:Death)-[r:HAS_ACTIVITY_CODE]->(a) RETURN a, count(r) AS count ORDER BY count DESC;

MATCH (:Manner_of_death {value: "Suicide"})<-[:HAS_MANNER_OF_DEATH]-(:Death)-[r:HAS_ACTIVITY_CODE]->(a) RETURN a, count(r) AS count ORDER BY count DESC;

Count of causes of death for sports-related injuries:

MATCH (:Activity_code {key: 0})<-[:HAS_ACTIVITY_CODE]-(:Death)-[r:HAS_358_CAUSE_RECODE]->(c) RETURN c, count(r) AS count ORDER BY count DESC;

Count of causes of death for work-related injuries:

MATCH (:Injury_at_work {key: "Y"})<-[:HAS_INJURY_AT_WORK]-(:Death)-[r:HAS_358_CAUSE_RECODE]->(c) RETURN c, count(r) AS count ORDER BY count DESC;

Accidental deaths that occured on a Monday, returning counts of all dimensions:

MATCH (:Day_of_week_of_death {value: "Monday"}) <-[:HAS_DAY_OF_WEEK_OF_DEATH]- (d:Death) -[:HAS_MANNER_OF_DEATH]-> (:Manner_of_death {value: "Accident"}) WITH d MATCH (d)-[*]->(dim) RETURN labels(dim)[0] AS dimension, dim.value AS value, count(dim.value) as count ORDER BY dimension, count DESC;

People with a doctorate degree who were the victims of homicide via gun violence:

MATCH (:Education_2003_revision {key: 8}) <-[:HAS_EDUCATION_2003_REVISION]- (d:Death) -[:HAS_358_CAUSE_RECODE]->(:_358_cause_recode {key: 435}) WITH d MATCH (d)-[*]->(dim) RETURN labels(dim)[0] AS dimension, dim.value AS value, count(dim.value) as count ORDER BY dimension, count DESC;

People who died of herpes:

MATCH (:Death)-[r:HAS_358_CAUSE_RECODE]->(c {key: 42}) RETURN c, count(r) AS count ORDER BY count DESC;

API / backend

API is https://graphql.org/, running on a serverless backend (Node locally).

Front-end

Front-end is React.