More than 6 million American people (mostly from the older population) have heart failures (HF), and this number keeps increasing each year. On average, we see around 870 thousand new cases of HF each year. 1 The current estimated cost for managing HF is 30 billion dollars. With this rate, HF is expected to reach at least a 3% prevalence with a projected cost of $53 billion by 2030. 3 In Texas, HF is the number 1 cause of death for adults over 40 years old. 2 According to the American Health Ranking by United Foundation, Texas is placed at number 15 for heart disease among all the states in the U.S.9 Therefore, preventing HF hospitalizations can help to reduce both the morbidity & mortality rates due to heart disease in the state of Texas and save a large number of medical costs for local hospitals (by avoiding federal penalties for high readmission rates [13]).

However, early prevention of HF hospitalization is not an easy task. A good solution targeting HF should be community-based since they prove to be effective in targeting a wide range of high-risk populations. 5 In order to develop an effective management plan, The Texas Health Department and local hospitals/organizations need to understand deeply the core characteristic of this disease throughout Texas geographic and over time. Since Texas is a large state, the distribution of demographics across the states will greatly vary. Based on the number in 2018, Texas has 10 counties in the U.S. with the largest shares of Hispanic residents. This makes Texas become the state with the 2nd largest Hispanic population in the whole nation. 11 In addition, Texas is also in top ten states with largest African American population according to the U.S. Census 2010.[12] A large proportion of the Hispanic and African American populations in Texas have social, geographical, economic, or other risk factors that contribute to the increased risk for HF. Among those factors, diabetes, obesity, and physical inactivity are among the highest risk factors for heart disease according to a report from the American Heart Association.10 Hence, from the previous facts and my exploratory analysis in this project, I find that there is an indication of spatial variation and an increasing trend of HF hospitalizations in Texas over the years. Therefore, to gain more insights into the HF hospitalization problems in Texas, I did a spatio-temporal analysis to identify the high-risk areas and contributing factors to HF.

Data description:

The data source is from and some other public health websites. I collected both the GIS data for map visualization and the raw data tables of risk factors and HF hospitalization in all Texas counties over the 10 year periods ("2005-2007", "2006-2008", "2007-2009", "2008-2010", "2009-2011", "2010-2012", "2011-2013", "2012-2014", "2013-2015", "2014-2016") for analysis. Therefore, this is a retrospective study design to identify the significant risk factors affecting the HF hospitalization rates from data in past years. After cleaning and combining the raw tables, my data has the sample size N = 2540 and includes 9 variables as described in the following table:

Table 1: Descriptions of the dataset Alt text


For this project, I used R Statistical Computing Software (version 3.5.2). I first visualized the SIR from the cleaned data to check for the spatial variability and time trend. The map plot and time trend were created using the ggplot package, and they are reported in fig. 1 and 2. From the map and the trend plot, we can see that the HF rates (or SIRs) slightly decrease from 2005 to 2014. Then, the rates increase rapidly from 2014 to 2016. When I highlighted the trend of Harris county alone, it also shows a similar pattern with the rest of counties in Texas.

Alt text

Fig. 1: Maps of SIR in Texas counties.

Alt text Fig. 2: a) Trend of SIR in Texas counties over different year periods; b) Trend of SIR in Harris county over different year periods

After checking the spatial and temporal trend, I found a spatio-temporal model would be appropriate for this project. Hence, I used the Bernardinelli model (Bernardinelli et al, 1995) from INLA package to estimate the relative risk (RR) of HF hospitalizations for each Texas county and year since it is a spatial-temporal model. Then, my model\ is expressed as:

Alt text


Table 2: Output from Bernardinelli model: Alt text

My model is expressed as: Alt text

From the final model, we can interpret the effect of each variable in the risk factor list on the prevalence of HF hospitalization. Looking at the exp(coef) in table 2, we can see that an increase in the percentage of any of the risk factors will increase the relative risk of HF hospitalization by 1 unit when all other variables remain constant. We can use the 95% confidence interval (range between the lower bound and upper bound in table 2) of the posterior means to judge which parameters may be different from zero. From table 2, I can only see that the CI of Poverty is away from 0. Thus, we can say that percentages of poverty are significant in the spatial-temporal fixed effect model of predicting RR of HF hospitalizations. On the other hand, the rest of the CIs of other variables includes 0 in them. This indicates that these variables don’t contribute significantly to the model. I dropped all the insignificant variables (keeping only the Poverty variable) and then compare the new model with the original one. The new WAIC is 8797.23 and the new DIC is 9195.50, which are slightly higher than the original criterions. Therefore, I decided to choose the original model as my final model. Now, I could map the final RR and the exceedance probabilities calculated by the final model over time using ggplot. I also created an animated version of this map using gganimate package. The results are as followings:
Fig. 3: Maps of RR in Texas counties over the from 2005-1016 Alt text

Fig. 4: Texas counties with probabilities of exceedance of 2 from 2005-2016. Alt text


From the exceeded probability map, we can see that, after 2012, more and more counties reach the RR of HF hospitalization over 2. Therefore, the Texas Health Department can use this map to identify the high-risk (red) areas for HF hospitalization to develop programs for early prevention of HF hospitalization. From my inference analysis, I find that the percentage of poverty is a significantly high-risk factor contributing to the increase in HF hospitalizations in Texas. This impact of poverty on HF hospitalization matches with the previously reported results from other published papers. 14,15 A good solution for the reduction of HF hospitalization rate should be community-based targeting the population with a high percentage in poverty, and it should help to minimize health disparities caused by this socioeconomic factor.

Limitation and What's next for HFMap:

From fig. 2, we all can see a dramatic increase from 2012 until 2016. Thus, a change-point model will be more appropriate to fit this temporal trend, and this method will be explored in my next steps for this project. All 4 figures reported in this paper show that there is a dramatic increase in HF hospitalization from the spatial-temporal plot. However, we don’t know what changes in the risk factors causing this increase. Although I explored the effects of common risk factors on the prevalence of HF hospitalizations in the analysis, I haven’t investigated the temporal-spatial trends of each risk factor (especially the significant risk-factor of Poverty) due to the time limitation. Thus, drawing the temporal-spatial maps and trend plots from the significant risk factors will provide us with more insights on the rates of HF hospitalization. These additional visualizations will be considered in my future work.


1. Heart Failure. (n.d.). Retrieved from 
2. Konstam, M. A. (2018, April 30). Heart Failure Costs, Minority Populations, and Outcomes. Retrieved from
3. Texas Heart Disease and Stroke Program - Home. (n.d.). Retrieved from
4. Roger, V. L. (2010, April). The heart failure epidemic. Retrieved from Walton-Moss, B., Samuel, L., Nguyen, T. H., 
5. Commodore-Mensah, Y., Hayat, M. J., & Szanton, S. L. (2014). Community-Based Cardiovascular Health Interventions in Vulnerable Populations. The Journal of Cardiovascular Nursing, 29(4), 293–307. doi: 10.1097/jcn.0b013e31828e2995
4. ICD - ICD-9-CM - International Classification of Diseases, Ninth Revision, Clinical Modification. (2015, November 6). Retrieved from
5. Division of Diabetes Translation. (2020, March 25). Retrieved from
6. Heart Disease Maps and Data Sources. (2019, December 2). Retrieved from
7. Heart Disease by State. (n.d.). Retrieved April 29, 2020, from
8. Texas Fact Sheet. (n.d.). Retrieved April 29, 2020, from
9. Schaeffer, K. (2019, November 20). In a rising number of U.S. counties, Hispanic and black Americans are the majority. Retrieved April 29, 2020, from
10. The Black Population: 2010 - (n.d.). Retrieved April 29, 2020, from
11. McIlvennan, C., Eapen, Z., & Allen, L. (2015, May 19). Hospital readmissions reduction program. Retrieved April 29, 2020, from
12. Ahmad, K., Suboc, T., Nazir, U., Nitti, K., Andrade, A., & Cotts, W. (2017). Impact of Poverty and Rural Status on Heart Failure Mortality. Journal of Cardiac Failure, 23(8). doi:10.1016/j.cardfail.2017.07.208
13. Ahmad, K., Chen, E. W., Nazir, U., Cotts, W., Andrade, A., Trivedi, A. N., . . . Wu, W. (2019). Regional Variation in the Association of Poverty and Heart Failure Mortality in the 3135 Counties of the United States. Journal of the American Heart Association, 8(18). doi:10.1161/jaha.119.012422

Built With

  • gganimate
  • ggplot
  • inla
  • r
  • rgdal
  • rstudio
Share this project: