Introduction

The reason this repository exists is to create a series of tools for HCSRN programmers to use VDW Census Demographics data consistently. Please feel free to give them a try and post feedback directly on github. If you are not quite sure on how to clone a repository, please send an email to alphonse.derus@kp.org.

When you want to get the latest and greatest macros/updates here, just do a "git pull".

Use cases

We tend to see a few use generic cases for VDW Census:

1. Estimate SES for a recruitment study or a data only study of current patients.
   1. Example: Estimating SES of a population of diabetics.
   2. Suggested Parameters:
      1. CENSUS_YEAR - Use the most current ACS CENSUS_YEAR (currently 2021).
      2. GEOCODE_BOUNDARY_YEAR census_demog_acs - Use the GEOCODE_BOUNDARY_YEAR associated with the most current ACS (2020).
      3. GEOCODE_BOUNDARY_YEAR census_loc - Use the GEOCODE_BOUNDARY_YEAR associated with the most current ACS (2020).
2. Estimate SES for a data only study with a limited timeframe.
   1. Example: Estimating SES of ER visits between 2015 and 2020.
   2. Suggested Parameters:
      1. CENSUS_YEAR - Use the CENSUS_YEAR that most aligns with the sample frame. CENSUS_YEAR=2020 was measured from 2016-2020. This seems like a reasonable tolerance level as only one year of ACS sampling was not included in the ACS sampling.
      2. GEOCODE_BOUNDARY_YEAR census_demog_acs - Use the GEOCODE_BOUNDARY_YEAR associated with the selected ACS (2020).
      3. GEOCODE_BOUNDARY_YEAR census_loc - Use the GEOCODE_BOUNDARY_YEAR associated with the selected ACS (2020).
3. Estimate SES for a data only study with a long timeframe.
   1. Example: Estimating SES of ER visits between 2006 and 2020.
   2. Suggested Parameters:
      1. CENSUS_YEAR - Use the CENSUS_YEAR that most aligns a single index date like a date of diagnosis or a procedure. ACS values are smoothed out over time, but you might want to get to "close enough" or a tolerance based on the index date. If you have something very old, it can be reasonable to use, but also, there comes a point at which a study says, "this record exceeds the tolerance to be an acceptable match".
         1. record A - Index date of 01jan2008 would use CENSUS_YEAR=2012, GEOCODE_BOUNDARY_YEAR=2010
         2. record B - Index date of 01jan2007 would use CENSUS_YEAR=2012, GEOCODE_BOUNDARY_YEAR=2010 assuming a 1 year tolerance
         3. record C - Index date of 01jan2000 would be left missing because it exceeds a tolerance threshold.
         4. record D - Index date of 12dec2021 would use CENSUS_YEAR=2021, GEOCODE_BOUNDARY_YEAR=2020

Here are some things worth considering:

1. Census is not individualized data. The aggregated values can closely resemble collected data (including race), but it is helpful to conceptualize.
2. Avoid using correlated/component variables in models. This includes the component values of computed indexes using the same underlying data (e.g., SVI, ADI, NDI, RUCA).

Specifications

Census Location - The person + location + time table.

Census Demographics - Where we get sociodemographics.

Macros

%assess_census_demog_acs

This macro assesses the local copy of _&_vdw_census_demogacs and returns a printout of the (census_year, geocode_boundary_year) values available.

%join_census_loc

This macro performs a left join from an input population (MRN) with a specified index date (either a variable in the dataset or a hard coded date. Default value is today()).

Example 1 - Explicit index_date out of enrollment {#join_ex_1}

join_census_loc(in_dset, out_dset, index_date, days_tolerance_pre=0, days_tolerance_post=0, debug=false);
%join_census_loc(
                &_vdw_enroll(where=('31jan2022'd between enr_start and enr_end) obs=10000) /*in_dset*/
                ,SampleJoin                     /*outdset*/
                , index_date = '31jan2022'd     /*explicit index_date */
                , days_tolerance_pre=90         /*how far back can we look for a location period? */
                , days_tolerance_post=90        /*how far forward can we look for a location period? */
                , debug=true                    /*do we want to look at some key frequencies */
                );

Example 2 - Explicit index_date {join_ex_2}

* make a temp table
data _tmp;
    set &_vdw_enroll(where=(enr_start between '01JUN2018'd and '01JAN2019'd) obs=10000);
    format idx yymmddd10.;
    do i='01jan2017'd to '31dec2021'd by 15;
        idx = i;
        output;
    end;
run;

%join_census_loc(_tmp
                ,SampleJoin_2
                , index_date = idx
                , debug=false);

Example 3 - Explicit index_date {join_ex_3}

%join_census_loc(_tmp
                ,SampleJoin_3
                , debug=true);

%fetch_census_demog

This macro takes a series of parameters and returns a dataset with the original dataset + whatever census demographics are available (i.e., not dependent on any particular implementation).

NOTE: THIS IS PROBABLY BROKEN AT MOST SITES AND NEEDS TO BE FIXED WITH 2020 CHANGES!

Example usage below:

* define target dataset;
%let acs5yr = acs.acs_demog_v;
%fetch_census_demog(
                input_ds = &_vdw_census_loc. (obs=100) /*YOUR data*/
                , idvar = mrn /*Have it in your dataset-  Do not change unless you are using study_id or something*/
                , geocode_var = geocode /*Have it in your dataset- join your dataset to &_vdw_census_loc where  indexdate between loc_start and loc_end - do not touch*/
                , index_date = today() /*You should change this - needs to be a DATE*/
                , years_prior_tolerance = 5 /*Recommended settings*/
                , years_after_tolerance = 3 /*Recommended settings*/
                , demog_data_src = &_vdw_census_demog_acs. /*switch to new file*/
                , demog_geo_var = geocode /*Do not change*/
                , census_yr_var = census_year /*Do not change*/
                , outds = work.outds /*Where do you want it to go? */
                ) ;

%fetch_ruca_2010_from_usda

This macro takes two parameters and returns a dataset containing the USDA Rural-Urban Commuting Area (RUCA) codes. RUCA codes are typically generated 3 years after the last decennial, so 2020 codes could be available in 2023. This could have been parameterized to take previous RUCA vintages, but the grain is very different, and currently VDW only has 2010-2019 data. It might be the case that we want to have a more abstract solution in the future.

The preview parameter runs a proc contents and prints the first five observations of the returned dataset.

Please reference values found in the Documentation.

Example usage below:

%let outdset = work.ruca2010;
* this reads as run the macro, store the result in the default work library, and output to the ruca2010 dataset in the work library. Allow the default preview and infomode values so I can see a contents and top 5 values from the output dataset and have an INFO statement in the log that tells me about the documentation;
%fetch_ruca_2010_from_usda(&outdset);