Sociodemographic Factors that Impact Vaccine Uptake

COVID-19 has made a great impact and led to the loss of lives worldwide. To best serve the public and reduce COVID-19 cases, it is important to study vaccine uptake and identify some of the factors that give rise to hesitancy. In this research, we analyzed aggregated datasets from the CDC for percent vaccinated at the county level as of July 21st, and US Census data containing variables attributed to social vulnerability to quantify how various sociodemographic factors are related to vaccine uptake. We then used LISA (Local Indicator for Spatial Autocorrelation) to observe how vaccine uptake varies across the US and found geographic clusters of hot spots and cold spots. We computed a baseline model to predict vaccine uptake using linear regression. However, we found that this model was unable to explain vaccine uptake in some locations such as Virginia and Georgia, so we used multivariate geographically weighted regression (MGWR). We found four primary variables of interest including per capita income, percentage of Republican voters, percentage of individuals 65+, and percentage of minorities, which best-predicted vaccine uptake using Spearman’s correlations. Linear regression results have an R^2 value of 0.510 whereas MGWR has an R^2 of 0.753, indicating that geographically weighted regression can better predict vaccine uptake using just four variables and geographic coordinates. Future work will focus on using the local regression models from MGWR to estimate vaccine uptake at the census block group (CBG) level and identify vulnerable areas due to low vaccine uptake.

WEBVTT 252 00:25:41.520 –> 00:25:45.960 Shivani Gurrapu: Social, economic factors that affect vaccine uptake and acceptance. 253 00:25:51.600 –> 00:26:03.090 Shivani Gurrapu: has led to a loss of many lives across the world and access to vaccines, has been a green changing tool, while the world recovers from the disease so understanding that. 254 00:26:03.630 –> 00:26:12.000 Shivani Gurrapu: vaccine acceptance and uptake correlations can be highly beneficial for the government and the Community to reallocate resources accordingly. 255 00:26:15.810 –> 00:26:31.830 Shivani Gurrapu: So our research questions were how have social economic factors influence access acceptance and how do they relate to access optic and which counties are at high risk based on socio economic factors and vaccine update. 256 00:26:33.750 –> 00:26:43.350 Shivani Gurrapu: So for our research we used three main data sets which included Facebook data which is compromised of aggregated data that uses to. 257 00:26:44.490 –> 00:26:56.460 Shivani Gurrapu: estimate vaccine acceptance and it is updated daily and has data across 700 counties and more and beginning from December 2020. 258 00:26:57.240 –> 00:27:06.870 Shivani Gurrapu: And the second one that we used was the CDC vaccine uptake data, and that includes records of vaccine administration, since January 2021. 259 00:27:07.380 –> 00:27:17.880 Shivani Gurrapu: And that includes data for across 3000 and more counties and also the percent of people who have received at least one dose of coven 19 vaccine. 260 00:27:18.480 –> 00:27:31.140 Shivani Gurrapu: And data across different age groups, so the last data set that we used was the social vulnerability index data set and they use this 15 us census variables in order to identify communities. 261 00:27:32.280 –> 00:27:43.380 Shivani Gurrapu: and main themes for in this dataset worthy socio economic status household composition and disability minority status and language and housing and transportation. 262 00:27:45.900 –> 00:27:54.900 Leela Yaddanapudi: After we received our data, we had to do some data cleaning, including changing the formatting and much so data types of each of the columns. 263 00:27:55.440 –> 00:28:05.700 Leela Yaddanapudi: From strings to integers and vice versa, our data was also very sparse so we had to fill in some missing data using machine learning Kanan amputation. 264 00:28:06.690 –> 00:28:16.230 Leela Yaddanapudi: We also use standardized residuals to detect outliers and filtered out any residuals that were more than two standard deviations away from the mean. 265 00:28:17.010 –> 00:28:31.020 Leela Yaddanapudi: In some cases of our analysis on models and to combine our data between Facebook CDs, to be an spi on the index the phipps code column, which represents each county and combined the data based on. 266 00:28:35.730 –> 00:28:43.890 Leela Yaddanapudi: These are the results of our linear regression after doing multiple correlations We found that four variables, where the strongest and. 267 00:28:44.370 –> 00:28:53.940 Leela Yaddanapudi: We also had a mentor who specializes in epidemiology on our team, and he also suggested using these four variables and our linear regression model. 268 00:28:54.510 –> 00:29:11.850 Leela Yaddanapudi: And R squared value for per capita income percentage republican presenter people over 65 and percentage of mine already was there a point 510 and this linear regression model was used to predict vaccine outtake based on these four variables. 269 00:29:15.300 –> 00:29:21.510 Jack Blumstein: Here we have a map of a vaccine updating, these are the percentage of people who are fully vaccinated across the US. 270 00:29:22.800 –> 00:29:24.630 Jack Blumstein: So we can just get a good look at. 271 00:29:25.710 –> 00:29:27.330 Jack Blumstein: Our data here, and the next slide. 272 00:29:28.920 –> 00:29:39.030 Jack Blumstein: One of the things we did to try and improve upon our linear regression was to perform multi variable geographically weighted regression basically what this does is it. 273 00:29:41.250 –> 00:29:50.910 Jack Blumstein: waits their aggression geographically, so we can see that different variables have different coefficients across the United States like if we look at this map of. 274 00:29:51.330 –> 00:30:01.590 Jack Blumstein: The coefficient of the estimated percentage of the minority population and the South East areas such as like Georgia, you can see that it’s more strongly negatively correlated and then. 275 00:30:02.250 –> 00:30:20.400 Jack Blumstein: As we move to the North West, you can see that the correlation is not as strong and it becomes more positive and then every can see here with the coefficient of estimated per capita income, there are some clusters of more heavily negatively correlated areas and some pluses and more. 276 00:30:21.750 –> 00:30:22.320 Jack Blumstein: Higher. 277 00:30:23.970 –> 00:30:27.540 Jack Blumstein: more positively correlated areas and then next slide. 278 00:30:30.330 –> 00:30:38.910 Jack Blumstein: Another thing we did was performed a Lisa analysis which basically just finds look for outliers you can see, on the map on the left is a map remote percent. 279 00:30:39.420 –> 00:30:43.980 Jack Blumstein: update and then a map on the right, it has the least amount on the Lucy map, we have a. 280 00:30:44.310 –> 00:30:58.470 Jack Blumstein: hotspots don’t as cold spots and diamonds hotspots are high Arctic area surrounded by high object donors are low uptake surrounded by high tech cold spots are low uptick area surrounded by la tech and diamonds are high Arctic area surrounded by low update. 281 00:30:59.580 –> 00:31:00.120 Jack Blumstein: An excellent. 282 00:31:01.290 –> 00:31:11.430 Shivani Gurrapu: So next steps are for our research include using a multivariate geographically weighted regression in order to predict the vaccine uptake on the senses block group level. 283 00:31:11.820 –> 00:31:25.470 Shivani Gurrapu: and finding outliers in the vaccine uptake on the CBD level to determine the CB geez that are vulnerable and, lastly, is to find a public or publish findings in a scientific journal, thank you.