Data Utilization and Preparation

In this section, we will discuss the data layer of the app. It will be consisted of two parts:

• Data Sets Utilized
• Data Processing

Primarily there are two data sets used in this app: CDC WONDER Detailed Mortality Data and County Health Ranking Social Determinants Data. The former one is consisted only with county level mortality rate while the later one is actually an aggregate of county-level data from 20 sources curated by the Robert Wood Johnson Foundation, consisting of more than 160 factors.

Data Sources

CDC WONDER Mortality

The county level mortality data is grouped in three-year-block. More detailed discussion of why we are grouping mortality rate in such a manner can be found in Other Restrictions under Data Sources section. The primary reason is to deal with missing values, e.g. the mortality rate could be suppressed if it is not prevalent enough. Also, method used for imputing missing values can be found in Data Preparation section under Missing Data Imputation. Here we will only briefly discuss the reason for choosing a specific subset to mortality data.

To understand the major causes of mortality, we selected the leading causes of death and their definitions were take from Stein et.al. "The Epidemic of Despair Among White Americans: Trends in the Leading Causes of Premature Death, 1999–2015", American Journal of Public Health, 107(10):1541–1547, 2017.

This version of Mortality Minder (MM) focuses on midlife deaths attributed to leading causes of death including deaths of Despair, Cardiovascular Diseases, Cancer and All Cause, but the approach can be readily generalized to any health problem of interest.

Link to the mortality data website can be found here: CDC WONDER Multiple Causes of Deaths

+ ICD-10 Code

The cause of death is selected based on the ICD-10 code. Currently, the selected causes of deaths are: Despair (F10-F19, X40-X49, X60-X84, Y10-Y19), Cancer (C00-C97), Cardiovascular Disease (I00-I99), Assault (X85-Y09), and All Cause death as defined by CDC Wonder. Other causes of death could be added but care must be taken that they are prevalent enough so that CDC Wonder does not suppress rates for many counties. For less prevalent causes in the future, the mortality rates could be calculate for 4 or 5 year chunks, instead of 3, to ensure more reliable mortality rate estimation.

+ Age Range

In order to correspond with the result reported by the paper mentioned above, we only choose a subset of the population. Only reported death of 25-64 were selected for further analysis and visualization. MM aims to capture the actual experience of mortality of American ages 25 to 64 at a community-level, so our analysis does not age-adjusted. All deaths in a community are treated equally. In the future, age-adjustment could easily be added to MM.

County Health Ranking Social Determinants

+ Sources

To investigate the possible causes of mortality rate trends behind, We gathered factors addressing health behaviors, clinical care, education, employment, social supports, community safety and physical environment domains from County Health Ranking website. Though the data set is downloaded from a single website, it is actually an aggregation of multiple data sources like CDC Wonder, Census Bureau, and USDA Food Environment Atlas. Links to the primary sources can be found under Data Sources section.

+ Further Filtering

We filtered the original 168 factors from CHR to a set of approximately 70 factors that were relevant to at least one cause of death at the national level. The primary criteria for filtering is:

• Not directly related to mortality, e.g. car accident mortality.
• Pass significant threshold (p < 0.05) based on corrected p-values for multiple hypothesis testing using the Benjamin-Hochberg Method.

For detailed description, please refer to Data Preparation under Developer Guide section.

Data Processing

Before carrying out analysis and corresponding data visualization, we need to use loaders to transform the raw data files first. There are several reasons for doing this instead of importing the raw data files directly:

• Common factors or identifiers, e.g. county names, have to have the same string representations as well as the same data types.
• Missing values should be either removed or imputed.
• Not all of the social determinants data are relevant. So some filtering operations, either by algorithm or by hand, are needed.

For more information, please refer to the Data Preparation section.