Monday, July 31, 2017

HealthLandscape Releases 5th Geospatial Brief

Today, HealthLandscape is releasing our fifth Geospatial Brief, “Where are Areas in Greatest Need of New HealthCenters? A Spatial Empirical Bayes Approach.” [MT1] Brief #5 uses 2015 data from the UDS Mapper (www.udsmapper.org) to explore areas with high rates and numbers of low-income population that are not being served by the Health Center Program. Specifically, we use a spatial empirical Bayes approach to create smoothed rates for low-income penetration and weighted estimates for the number of low-income population not being served by a health center. This blog discusses the importance of identifying priority areas of need and provides more details about our approach.

Prioritizing areas in highest need of safety-net health services and allocating resources in the most efficient way possible will be increasingly important in the current political context. The American Health Care Act (AHCA) and the Better Care Reconciliation Act (BRCA) drastically reduce funding for Medicaid, while the president’s proposed budget slashes funding for the Food and Drug Administration (FDA), the Centers for Disease Control (CDC), and National Institutes of Health (NIH). Despite the growing opioid epidemic, even the Substance Abuse and Mental Health Services Administration (SAMHSA) budget is the target of proposed cuts (Jost, 2017). While the funding for the federally-funded Health Center Program (HCP) appears to be steady, the funding cuts in these other areas have the potential to be particularly impactful on the HCP, which relies on the federal government and Medicaid as important revenue sources (Han et al., 2017).

Health Centers (HCs) that are part of the HCP are a vital part of the health care safety net. Established more than 50 years ago to provide primary medical, dental and behavioral care, the approximately 1,400 HCs operate more than 9,800 clinic sites to meet the needs of the medically underserved population in the US. The HCP serves roughly 25 million high-need people annually. Research has shown that HCs have a positive impact in terms of improved access to care, reduced avoidable hospitalizations, and better health outcomes (e.g., Evans, et al., 2015). The Health Resources and Services Administration (HRSA) and Bureau of Primary Health Care (BPHC) fund new sites based on need within a community, established by the applicant based on service-area based statistics.

In general, identifying priority areas of need presents several challenges. First, one could choose from a number of different ways to define need. Is poverty a sufficient measure to define need?  What about a composite measure of social deprivation?  These are questions that are still being sorted out.  In the context of the HCP, using low-income penetration or poverty rates will assure you prioritize areas with highest rates of need, but doesn’t consider the number of people in need. This issue could be dealt with by using the number of low-income people that are not served; however, this would lead to only densely populated urban areas being identified as areas in need, and would exclude rural and isolated areas.

There is also the matter of attempting to define the geography in need, which comes back to the age-old modifiable areal unit problem (MAUP). Should neighborhoods be used as the unit of analysis?  If so, is it appropriate to define neighborhoods based on census tracts? In the health center world, ZIP Code tabulation areas (ZCTAs) are typically used because of data availability (and that is how data are reported in the UDS Mapper), but need may exist within particular parts of a ZCTA or be spread across parts of multiple ZCTAs.

To address the issues above, we used a spatial empirical Bayes approach to identify high-need ZCTAs. In general, empirical Bayesian approaches help deal with the instability in population over large areas by adjusting or smoothing rates based on population size toward the overall mean. For example, using a Bayesian approach to estimate low-income penetration rates would adjust ZCTAs with small low-income populations more towards the overall mean, while high low-income population ZCTAs would be adjusted less.

While using a standard empirical Bayesian approach would help smooth out ZCTAs with very small populations, the approach does not account for regional variations, and would likely over-adjust small, isolated rural areas with high need. For these reasons, we used a spatial (also called local) empirical Bayes approach to estimate low-income penetration rates. The spatial empirical Bayes approach still adjusts the rates based on the size of the low-income population, but instead of adjusting based on the overall mean, it adjusts the rates for ZCTAs based on a local mean, which is the average rate of each ZCTA and its contiguous neighbors.
 
We defined high-need ZCTAs as those with a smoothed penetration rate of less than one percent and ranked them based on a weighted estimate of low-income population not served by health centers. The weighted unserved low-income estimate was calculated by taking the average number unserved for each ZCTA and its contiguous neighbors. The geospatial brief focuses on the top 500 ZCTAs in need based on our criteria; we chose 500 because between 400 and 700 new HCs have been funded annually over the past seven years.

Again, it is important to note that our approach for defining high-need areas for the HCP is just one of many possibilities. It is not intended to be used as a rule for funding allocations for the HCP, but only as a starting point to help overcome the challenges of identifying areas most in need. The primary objective of the geospatial brief is to highlight the potential of geospatial and Bayesian methods for overcoming these challenges to identify priority areas.

References

Evans, Christopher S., Sunny Smith, Leslie Kobayashi, and David C. Chang. (2015). The Effect of Community Health Center (CHC) Density on Preventable Hospital Admissions in Medicaid and Uninsured Patients. Journal of Health Care for the Poor and Underserved 26(3): 839-851.

Han, Xinxin, Qian Luo, & Leighton Ku. 2017. Medicaid Expansion and Grant Funding Increases Helped Improve Community Health Center Capacity. Health Affairs 36(1): 49-56.

Jost, Timothy. “Trump Budget Proposes Big Health Cuts.” May 23, 2017. Health Affairs Blog
Available at http://healthaffairs.org/blog/2017/05/23/trump-budget-proposes-big-health-cuts/


 [MT1]Link to Brief

Wednesday, July 26, 2017

Geospatial Research Brief: Where are Areas of Greatest Need of New Health Centers? A Spatial Empirical Bayes Approach

Our previous geospatial briefs have shown the potential of geospatial hot-spotting methods to identify priority areas of need (e.g., Topmiller, 2016). Geospatial methods can also be integrated with Bayesian approaches to account for spatial variation and variance instability in regards to population. This brief illustrates the use of a spatial empirical Bayes approach to identify high-need areas based on low-income populations not served by the federally-funded Health Center Program (HCP).

The HCP is part of the health care safety net and includes approximately 1,400 grantees operating more than 9,800 clinic sites to meet the needs of the medically underserved population in the US (HRSA, 2016). The HCP experienced substantial growth under both the George W. Bush and Obama administrations, with recent literature demonstrating the positive impacts of this expansion (Iglehart 2008; Han et al., 2017; Cole et al., 2017). In an uncertain health care climate, where health centers (HCs) are seen as important cornerstones of the health care safety net, it is important to understand where HCs are currently serving patients and where there may be continued or growing need.

Using 2015 data from the Uniform Data System (UDS) Mapper (UDS, 2015), we utilized a variety of geospatial approaches to identify ZIP Code Tabulation Areas (ZCTAs)[i] most in need based on low-income penetration rates and the number of low-income population not being served[ii]. Because health centers serve patients living outside of the ZCTA in which they are located, it is necessary to create weighted or smoothed estimates of penetration rates which incorporate need in contiguous ZCTAs. Smoothed low-income penetration rates were estimated using a spatial empirical Bayes approach[iii], which adjusts or smooths the rates for each ZCTA based on its low-income population and the local mean, which is the average of contiguous ZCTAs’ rates; ZCTAs with smaller low-income populations will have their rates adjusted more towards the local mean (Cromley & McLafferty, 2012). We also created spatially-weighted estimates for the total number of low-income population by calculating the average of each ZCTA and its contiguous neighbors.

High-need ZCTAs were identified by (1) removing all ZCTAs that had zero low-income population; (2) creating smoothed low-income penetration rates[iv] , low-income rates, and poverty rates using a spatial empirical Bayes approach; (3) creating spatially-weighted estimates for the number of low income population not being served by health centers (weighted average of ZCTA and its neighbors); (4) identifying all ZCTAs with smoothed low-income penetration rates of less than 1 percent; (5) finally, focusing only on those ZCTAs with smoothed low-income penetration rates of less than 1 percent, we identified the top 500 ZCTAs[v] ranked by the highest number of spatially-weighted low-income population not being served by health centers.


Smoothed & Weighted Characteristics of High-Need ZCTAs

All ZCTAs^
(N=32,166)
All High-Need (N=500)
High-Need Urban (N=381)
High-Need Large Rural (N=76)
High-Need Small or Isolated Rural (N=15)
High-Need
Puerto Rico (N=28)
Unserved Low-Income Population (% of High-Need)   
84,303,659
3,378,283
2,355,967
(69.7%)
385,493
(11.4%)
64,975
(1.9%)
571,848
(16.9%)
Poverty Rates (Below 100% FPL)
14.9%
16.5%
14.2%
18.8%
20.0%
38.5%
Low-Income Rates (Below 200% FPL)
34.5%
35.8%
32.0%
41.8%
47.4%
65.1%
Sources: UDS, 2015; American Community Survey, 2010-2014
^Not weighted



The weighted unserved low-income population for the top 500 ZCTAs ranges from 3,497 to 39,455, with a total of almost 3.4 million people living in these ZCTAs. More than one-quarter of these 3.4 million people live in the top 50 ZCTAs in need, where almost half (24) of the ZCTAs are located in Puerto Rico, and more than one-fourth (13) are located in Texas. More than one-half of the high-need population is located in just four states or territories, with Texas (21.5%), Puerto Rico (16.9%), Pennsylvania (8.6%), and North Carolina (7.6%) having the largest populations in high-need areas. Fifteen states have no high-need ZCTAs based on our criteria. Less than 20% of the high-need ZCTAs and about 13% of the high-need population are located in rural areas, where low-income rates exceed 40%, compared to 32% for urban high-need ZCTAs, and 34.5% for all ZCTAs. Poverty and low-income rates are highest in small and isolated rural ZCTAs, as well as in Puerto Rico. About 60% of high-need ZCTAs are located in Medically Underserved Areas/Population (MUA/Ps) and more than half (276) are being served by at least one health center, with almost 25% (119) already being served by at least two health centers. The map below displays the number of high-need ZCTAs by State.


This brief highlights innovative geospatial approaches for identifying priority areas in the context of the federally-funded Health Center Program.  These methods could be used as a starting point for identifying priority areas for future health center funding, as they incorporate need across contiguous ZCTAs and account for variations in population. This research is subject to several limitations. First, we defined our high-need criteria based on low-income penetration rates; there are many other ways we could have defined high-need ZCTAs, such as excluding ZCTAs with relatively low rates of low-income population or focusing only on ZCTAs that are located in MUA/Ps. In addition, the UDS data do not include patients being served by non-HC providers, such as free clinics or private physicians. Thus, it is possible that some of the high-need areas that we identified are being served by other providers. Ideas for future research include using other criteria and exploring differences among high-need areas and comparing the location of high-need areas with the actual locations of new health centers.

Michael Topmiller, PhD, Health GIS Research Specialist
Jennifer Rankin, PhD, Director of Research and Product Services


References
Cole, Megan B., Omar Galarraga, Ira B. Wilson, Brad Wright, & Amal N. Trivedi. 2017. At Federally Funded Health Centers, Medicaid Expansion Was Associated with Improved Quality of Care. Health Affairs  36(1): 40-48.

Cromley, Ellen K., & Sara L. McLafferty. 2012. GIS and Public Health. New York, NY. Guilford Press.

Han, Xinxin, Qian Luo, & Leighton Ku. 2017. Medicaid Expansion and Grant Funding Increases Helped Improve Community Health Center Capacity. Health Affairs  36(1): 49-56.

Health Resources and Services Administration (HRSA), 2016. The Health Center Program.    

Iglehart, John K. 2008. Spreading the Safety Net — Obstacles to the Expansion of Community Health Centers.  New England Journal of Medicine 358:1321-1323.

Lu, Yao, David Slusky. 2016. Impact of Women’s Health Clinic Closures on Preventive Care. Applied Economics 8(3): 100-124.

Topmiller, Michael, 2016. “Where are “Hot Spots” of Medicare Spending and Preventable Hospitalizations and “Cold Spots” for Preventive Care?”  HealthLandscape Geospatial Research Brief #2. 

Accessed at http://www.healthlandscape.org/Geospatial-Analysis.cfm
Uniform Data System (UDS), 2015. Available at http://www.udsmapper.org.


[i] Health Centers report data by ZIP Code, but the data are converted to ZCTAs in order to compare health center patient data to area demographic data.  ZCTAs are created for the decennial census approximating the area covered by a ZIP Code.
[ii] Low-income includes those people who live below 200% of the Federal Poverty Level.  This population is used by the Health Center Program as an approximation of the medically underserved population.
[iii] Spatially smoothed rates were calculated using GeoDa (Anselin, et al., 2006).
[iv] Health centers do not report patients by ZIP Code by income level, therefore low-income penetration rates are calculated by dividing the number of health center patients from a ZCTA divided by the total low-income population living in a ZCTA.  Nationally in 2015, 93% of all health center patients were low income. For more information, see http://www.udsmapper.org/about.cfm.
[v] The HC Program has added between 400 and 700 new health centers annually over the past seven years.

Thursday, July 20, 2017

New Versions of HealthLandscape Tools Coming Soon!

This summer the HealthLandscape team is hard at work rebuilding our suite of tools in JavaScript. During the course of this modernization, we are making improvements to the look, feel, and friendliness of our products. Today I want to let you know how you can get ready for the first of our tools to be rebuilt, the UDS Mapper.

The UDS Mapper is funded by the Health Resources Services Administration (HRSA). The intended and most common users are Federally Qualified Health Center (FQHC) professionals researching their region, creating Service Area Maps, and identifying areas for potential expansion. The broader audience is diverse and includes, but is not limited to, academics, health care consultants, and even non-medical private businesses.

The UDS Mapper was originally developed in Adobe Flash and has served us well, but many browsers will soon stop supporting this platform. The new version, written in JavaScript, will have the features the Mapper is well-known for; Patient Origin maps, Medically Underserved Areas/Populations layers, the drawing tool, as well as the ability to map your own data via QuickGeocodes.

Currently when one is developing a map the user needs to complete work, including adding any points, labels, or other annotations, in one online session. That changes with the new UDS Mapper where you will be able to save your project, log out, and return to complete it at another time. In addition to the anticipated saving feature, we will be adding new ways to share your maps with colleagues.

An important part of utilizing the new UDS Mapper is attending one of our free "What's New with the UDS Mapper" webinars in September of this year. Our usual two webinars per month will increase to four during this month to support this roll-out. You can register now for them here.

What's New with the UDS Mapper?Wednesday09/06/20172:00 PM ET» Register «
What's New with the UDS Mapper?Tuesday09/12/20172:00 PM ET» Register «
What's New with the UDS Mapper?Monday09/18/201711:00 AM ET» Register «
What's New with the UDS Mapper?Thursday09/28/20172:00 PM ET» Register «
In these sessions the UDS Mapper team will help orient you to the new tool and will be taking your questions. This is a great opportunity to introduce your colleagues to this resource if they have not yet been trained.

The HealthLandscape team is very excited to get these improved tools released. Be sure to bookmark this blog, like us on Facebook, or follow us on Twitter to get updates on your favorite HealthLandscape tools.

Monday, July 10, 2017

We're Off to the Esri User Conference

It's my favorite time of year!


HealthLandscape is headed to the Esri User Conference in San Diego for a week of paper sessions, learning labs, technical demonstrations, and connecting with industry peers. We're excited to get a closer look at ArcGIS Pro, try out Insights for ArcGIS, pick up some new Python and Leaflet skills, and, of course, wake up at 5:30 on Wednesday morning to run the Esri 5k along the San Diego waterfront.

If you're heading to the conference, come check out our two maps featured in the Map Gallery. Our small-format submission features the changes scientists have seen in mosquito season, while our Story Map entry is a Swipe Map showing the significant impact of the Affordable Care Act on insurance coverage across the country.

















While you're there, be sure to stop by the User Applications Fair to get a demo of our newest site - the PHIT Communities Population Health Improvement Tool. Mark Carrozza and I are looking forward to showing how we're helping communities improve population health by leveraging GIS and local, small area data.

 PHIT Communities allows users to map small area estimates for chronic disease risk factors and health outcomes for the largest 500 cities across the U.S. To address areas of need and development population health improvements plans, PHIT Communities also includes data on clinical preventive services, healthy lifestyle choices, and access to primary and preventive care to help guide local decision making. 


We're honored that PHIT Communities has been selected as one of the eight applications featured in the User Applications Fair this year, and hoping that we'll take home a top three finish for the seventh year in a row (vote for us!).

Can't make the conference? We'll be tweeting about the cool things we learn while we're there, so follow us @healthlandscape to keep up!

Hope to see you there!



Jené Grandmont
Senior Manager, Application Development & Data Services