Eileen Chen (ezchen) & Ron Chew (rchew)
67-364 Practical Data Science: Project 1
COVID-19’s Negative Impact on Mental Health and Access to Care
We present our key findings and insights from this project in this video presentation.
"How are you doing?" The typical answer to that question is "good", "not bad", or something of the neutral, yet positive adjectives. More and more frequently recently, we have heard answers to this ubiquitous question over Zoom like "eh, same old", "tired", or just grimaces on people's faces. It has gotten to a point where most people we talk to are drained, dejected, and have a grim outlook on their current situation or the state of the world. We did not actively find this problem. News articles filled with seemingly perpetually negative outlooks on the state of the country and the pandemic, social media posts uncovering the wide systemic racism that has been occurring in the US for a long time, and calls with friends who tell us they are feeling down or depressed lately trigger us every day and remind us of how the pandemic has shaken lives and affected people's well-being.
Since the COVID-19 pandemic started in March 2020, many of us have experienced a wide array of negative feelings and moods, whether that is feeling worried about a family member that is sick, feeling distraught about the hate crimes against black and asian communities, feeling stressed about getting laid off, or feeling lonely without the ability to see loved ones and friends as often or at all. We as students have definitely experienced a rollercoaster of emotions since the pandemic started and felt that it has taken a mental toll on us. Thus, we are interested in how the pandemic has impacted everyone else's mental health, which groups were impacted the most, and how Americans can be supported during this time with early identification of anxiety and depression symptoms.
The mental health during COVID-19 dataset we use was produced from the National Center for Health Statistics' (NCHS) partnership with the U.S. Census Bureau on the Household Pulse Survey. The specifics of the Household Pulse Survey can be found here. The data is updated bi-weekly. It currently has 5,154 rows and 14 columns. The dataset is structured by the weekly time periods. Each time period is divided depending on the indicator (depressive disorder, anxiety disorder, or both). Within each indicator the data is grouped by National estimate, by age, by gender, by race/Hispanic ethnicity, by education, or by state. The data seems clean, but it is definitely not tidy. We cannot tidy up the rows to be individual observations as the data is inherently grouped. We will thus use filtering of the grouped data to extract different demographics.
We also use the mental health care received or not received dataset also from NCHS and the Census Bureau. The structure of the data is very similar with 3,540 rows and 15 columns. The column labels are the same besides the indicator categories being "Took Prescription Medication for Mental Health, Last 4 Weeks", "Received Counseling or Therapy, Last 4 Weeks", "Took Prescription Medication for Mental Health And/Or Received Counseling or Therapy, Last 4 Weeks", and "Needed Counseling or Therapy But Did Not Get It, Last 4 Weeks". There is also an additional supression flag column that has no values. This data starts with phase 2 beginning Aug 19.
Finally, we use the United States COVID-19 Cases and Deaths by State over Time dataset from the CDC, updated daily. From the CDC page, some notes about the data to keep in mind is that "this aggregate dataset is structured to include daily numbers of confirmed and probable case and deaths reported to CDC by states over time. Because these provisional counts are subject to change, including updates to data reported previously, adjustments can occur. These adjustments can result in fewer total numbers of cases and deaths compared with the previous data, which means that new numbers of cases or deaths can include negative values that reflect such adjustments."
Our Initial Questions
How has the US population's mental health issues and receipt of mental healthcare changed during the pandemic?
Are certain genders, races, age groups, or states seeing different impacts on mental health and receipt of mental health care during the pandemic? Do the rates of change of issues vs care differ within groups?
Does the state's rate of new COVID-19 cases or rate of COVID-19 related deaths affect their mental health and receipt of mental healthcare? Are certain states more "mentally resilient"?
We break down our analysis and questions by joining the data sets in different ways. Thus, the structure of our data analysis flows as follows:
- Question 1 & Question 2 (Symptoms Data & Mental Health Care Data)
a. Data Cleaning
b. Exploratory Analysis of National Depression & Anxiety Trends
c. Deep Dive into Mental Health Care & Age
- Question 3 (Mental Health Care Data & COVID-19 Cases and Deaths Data)
a. Data Cleaning
b. National Average, Breakdowns by State, Changes over Time of COVID-19 incidence
c. Relating COVID-19 incidence with Mental Health Symptoms
How has the US population's mental health issues and receipt of mental healthcare changed during the pandemic?
The symptoms of anxiety and depression data as well as the mental health care data were already pretty clean. We just had to convert the dates to the datetime type so that Altair could work with it.
Exploratory Analysis of National Depression & Anxiety Trends
To get a feel for the data we are working with, we do some exploratory analysis of the data to reveal the general trends over time and demographic variances.
Figure 1. Starting from late April of 2020 to late Februrary of 2021, there is a slight gradual increase in the percentage of people experiencing symptoms of depressive and/or anxiety disorder. The two breaks in the graph indicate periods where data was not collected. The greatest percentage of people experience symptoms of anxiety or depressive disorder (around 35-43%) while the least percentag of people experience symptoms of depressive disorder. All three indicators rise and flow at a similar rate. It seems that there was a rise around the end of summer when phase 1 was ending and phase 2 began. There seemingly was a dip during the time that data was not collected and the second rise spans across the beginning of phase 2 in mid August to the middle of phase 3 and holiday season.
In January to June of 2019, 11% experienced symptoms of anxiety or depression. Even compared to the lowest national average of 34% during the pandemic, that is a 209% increase in the reported symptoms. With this in mind, we were curious as to which demographic groups were most impacted mentally, starting with age.
Figure 2. 44.87% of 18-29 year olds reported symptoms of depression or anxiety and there is noticeable decrease with every increase in age range with 16.8% of 80 year olds and above reporting symptoms. Both 18-29 year olds and 30-39 year olds report a significant percentage above the average as denoted by the dotted red line.
We conjecture that 18-29 year olds may have been experiencing the most shifts and changes in their lives, going from seeing their friends often and having in person classes to being forced to lockdown at home and limit their social interaction. No in person events also means that everything is virtual, often on social media such as Instagram and Tik Tok, which this meta-analysis by NIH proves to negatively impact young adults' mental health.
Figure 3. An average of 35.91% of females experience symptoms of depression or anxiety vs 29.02% of males. However, given that in general "about twice as many women as men experience depression" (Mayo Clinic), we do not view this difference as anything abnormal.
Figure 4. Non-Hispanic people of other races or multiple races report the highest percentage of symptoms of anxiety or depression at 40.52%. However, Hispanic or Latino and Black people also report a high level symptoms with all of those groups reporting higher than the average of 34.18%.
From news and research such as this one from the CDC, we know that Black and Hispanic communities experience a disproportionately high rate of coronavirus cases and deaths, and believe it may contribute to the stresses and grievances in those communities.
Figure 5. Respondants with less than a high school diploma report the most symptoms of depression or anxiety and those with a Bachelor's degree or higher report the least.
We speculate that this may be caused by mass layoffs (especially in the in person services industries like restaurants) during the pandemic and thus financial struggles among those with less than a high school diploma. Those with a bachelor's degree or higher may hold jobs that are less dispensible or they are able to work from home, meaning they are less prone to layoffs and the stresses of looking for a new job. People with financial stability also more often than not have greater access to mental health care.
Figure 6. Differences among states for percentages of people with symptoms are lower as compared to some of the other demographic groupings. Louisiana, Nevada, and Mississippi report the highest percentages while South and North Dakota report the lowest percentages.
With these alarmingly high percentages of symptoms of anxiety and depression across the board, we turn to data that demonstrates whether or not Americans are receiving the necessary mental health care.
Deep Dive into Mental Health Care & Age
Figure 7. It seems that the lowest percentages of people received counseling or therapy and the highest percentage of people took prescription medication for mental health and/or received counseling or therapy. There is a slight positive correlation between the number of people who received counseling or therapy and the number of people who could not receive it from mid-October to mid-November. Before holiday season (end of November and December) there is a dramatic decrease in the percent of people who took prescription medication and/or received counseling or therapy. The number of people who needed counseling or therapy but did not get it also went down before both increased again ramping up to Christmas season. There is also a sharp increase from 10.3% to 25.3% for people who took prescription medication during January.
With this general trend graph, we cannot come to any strong conclusions about mental health care received vs not received. However, from the pure percentages alone, it is alarming that at one point it came close to 1 out of 5 Americans who could not receive adequate access to counseling or therapy when needed.
Based on the symptoms of anxiety and depression we analyzed, we as college students were most interested in the age demographic. Thus, we first separately analyzed the care received and not received for different age groups.
Figure 8. 18-59 year olds all received more mental health care than 60+ year olds. This makes sense given that these younger groups experienced the most anxiety and depression symptoms. 18-29 year olds, although the group reporting the highest percentage of symptoms, did not receive the most care.
These percentages of people who received care do not provide much solice since 19.44% of 18-29 year olds received care but 44.87% reported symptoms of depresison or anxiety. We believe that this may be in part due to the stigma there is around getting mental health care as well as the lack of access young adults and people in general have to counseling and therapy, as we explore in the next graph.
Figure 9. We clearly see here that 18-29 year olds have a starkingly high percentage (19.58%) who did not receive mental health care but needed it at the time. Young adults are not receiving the care they need.
Having analyzed both symptoms of anxiety and depression as well as mental health care data, we now want to examine them closely together, again narrowing down on the 18-29 age group. We thus change our initial more generic inquiries about all demographics to focusing on age.
Are certain ages seeing different impacts on mental health and receipt of mental health care during the pandemic? Do the rates of change of issues vs care differ within age groups?
It is of note that mental health care data began to be collected during phase 2 in mid-August while symptoms data began to be collected during phase 1 in late April. Thus, when combining the data, symptoms data during phase 1 is dropped. Here, we first take a look at national trends for comapring percentages of people with symptoms of anxiety or depression and the care that Americans received and did not receive.
National Average of Symptoms vs Care
We first find the Nataional difference of citizens experiencing mental health symptoms vs those that did or did not receive care.
Figure 10. Even with the growing amount of people with symptoms of anxiety and/or depression, the amount of people receiving mental health care stays the same (with a dip around late November)and the amount of people not able to receive the care needed increases.
Below, we perform the same analysis and present a simlar graph, but filtering on just 18-29 year olds.
Figure 11. Unlike the national average data, for 18-29 year olds, the percentage of young adults who could not receive therapy or counseling but needed it started to surpass the percentage of young adults who received care.
We were curious about the exactly comparisons of the national average data to 18-29 year olds, so we overlay them below.
Figure 12. The figure above demonstrates that there is a greater access gap to mental health care for 18-29 year olds than the national average. The percentages of young adults who experience symptoms but are not receiving proper care are glaringly high. Young adults are also receiving around the same amount of care as the national average throughout the pandemic despite having much higher percentages of anxiety and depression symptoms.
There is a spike in symptoms among 18-29 year olds from mid to late October as well as a slight increase in the number receiving and not receiving care.
With a good understanding of the mental health symptoms and healthcare situation during the COVID-19 pandemic, we want to see if the pandemic itself played a role in causing additional mental distress and health issues. Our hypothesis is that the pandemic played a direct role in increasing the prevalence of symptoms, and we should see a direct correlation to the pandemic.
Does the incidence of COVID-19 affect the prevalence of mental health symptoms in the US?
Does the state's rate of new COVID-19 cases or rate of COVID-19 related deaths affect the mental health of their citizens?
First, we'll need to import, clean, and tidy the daily state COVID-19 cases and deaths data from the CDC.
Because the data is in a daily format per state (24720 rows), we will need to aggregrate the data into the weekly (5 days in Phase 1) or fornightly (12 days in Phase 2 and 3) buckets as per the mental health data.
We'll need to identify these buckets from the symptoms dataset, then attach them to the the cases/death data.
With the additional date bucket labels, we can now group the data by time to see the national average for new COVID-19 cases and deaths across the study periods.
National Averages of COVID-19 incidents
Charting the national average during the study period, we can see that there were a few spikes in new COVID-19 cases and deaths per day. Cases and deaths both spiked in the summer months of June - August 2020, and during the winter months November 2020 - January 2021. There was an initial spike of deaths in April - May 2020 due to the overwhelming of healthcare facilities as well.
Focusing on these particular time periods will be the most helpful in understanding any relationships between the COVID-19 pandemic and mental health symptoms/care. According to our hypothesis, we should see mental health symptoms rise during the same spikes, and states with higher spike should see more symptoms.
Figure 13. The figure above shows the spikes in COVID-19 cases in the US, especially during the months of Summer 2020 and Holiday 2020.
Figure 14. The figure above shows the spikes in COVID-19 deaths in the US, especially during the months of Summer 2020 and Holiday 2020.
Finding States with the Lowest/Highest Incidents
With the national average found, we can now compare the same data between states to identify those with the highest/lowest incidents. We first take an overall view of each state during the pandemic in order to identify ends/outliers, then idenitfy how they have changed over time.
In order to fairly compare between the different states, we need to account for the differing population sizes. We will compare the rates per 1M population instead. First, we need to import population data from the US Census. Because this dataset uses full states names, we'll need to convert these too with another simple dataset.
We can now divide the cases/deaths by population to find the number of incidents per 1M population.
From the charts, we can see that Hawaii and Vermont have the lowest COVID-19 cases and deaths. Meanwhile, South Dakota and Rhode Island have the highest number of incidents. We will focus on these states to see if the number of incidents affect the mental health of citizens in terms of symptoms and care received. Additionally we will also inspect Louisiana and Nevada as they had the highest percentage of citizens with mental health symptoms.
(We take a different approach here to picking the "highest" states as we want to find states with both high cases and deaths, i.e. not North Dakota as it has high cases, but not the highest deaths.)
Figure 13. The figure above shows the states of ND, SD, TN, UT and RI having the highest number of cases of COVID-19 during the pandemic, with VT and HI having the lowest.
Figure 14. The figure above shows the states of SD, MS, AZ and RI having the highest number of deaths from COVID-19 during the pandemic, with VT and HI having the lowest.
Changes in State incidents over time
With those specific states identified, we can now plot the time series data for those states to see how their cases/deaths changed during the pandemic. As previously, we will need to account both the duration of each study period and the population size of each state.
From the following charts we can see that all the states do follow the general trend with the National average, having risen and fallen at around the same peak periods. However, we can see that the states of South Dakota (SD) and Rhode Island (RI) seem to have much more exaggerated changes, especially during the Holiday 2020 period. We can analyze this change to see if mental health symptoms experienced by those states also rise as much during that time to prove our hypothesis.
We can now plot these changes along with the mental health symptoms to see if there are any of those correlations.
Figure 15. The figure above shows the states's new COVID-19 cases per 1M population, per day during the pandemic. SD and RI overtake the other states in terms of cases during the Holiday 2020 months.
Figure 16. The figure above shows the states's new COVID-19 deaths per 1M population, per day during the pandemic. Similar to cases, SD and RI overtake the other states in terms of deaths during the Holiday 2020 months.
Using Time Series to find a relationship between COVID-19 Incidence and Mental Health Symptoms
In order to plot both COVID-19 cases/deaths and the mental health symptoms data, the mental health data first needs to be grouped by the same time period labels and the abbreviated state name. They can then be joined and plotted together.
From these last charts we can see that the prevalence of mental health symptoms increased with the spikes in COVID-19 incidence during Summer 2020 and Holiday 2020 across all states. As identified earlier, we look to South Dakota's (SD) and Rhode Island's (RI) extreme rise in cases and deaths during Holiday 2020 to see if there is indeed a direct correlation and propotionality to mental health symptoms.
However, we can see that while their incidence shot much higher than other states, the prevalence of mental health symptoms did not rise by a proportionate amount. In fact, South Dakota actually had the lower or the lowest percentage of citizens with mental health symptoms compared to other states. This shows that other factors that are different between states are causing the increases in mental health symptoms, and not the incidence of COVID-19.
Figure 17. The figure above shows the states's new COVID-19 cases per 1M population, per day plotted against the percentage of residents with mental health symptoms. There is a small correlation, but no direct proportion between cases and deaths.
Figure 18. The figure above shows the states's new COVID-19 deaths per 1M population, per day plotted against the percentage of residents with mental health symptoms. As with cases, there is a small correlation, but no direct proportion between cases and deaths.
In conclusion, we were unable to prove to that COVID-19 incidence is directly correlated and proporitonal to the prevalence of mental health symptoms due to the different rates of change in symptoms vs cases/deaths between different states. We suspect that other different factors between states are causing these changes in mental health symptoms, such as the social distancing or business closure policies enacted by each state.
Suggestions & Future Work
From our analysis, we ultimately found that:
The pandemic and the worsening of it greatly increased depression/anxiety among Americans, especially impacting 18-29 year olds.
However, there is an increasing lack of access to mental health care.
Although there is no correlation between the incidence within a state and the mental health of its citizens, there may be other factors during COVID-19 such as speicific state policies, financial instability or a lack of in person interaction that leads to a rise in mental health issues.
We suggest the following for schools, workplaces, governments, and people in general to be aware of and/or implement:
Decrease barriers to mental health care such as increasing in-network mental health care or building in mental health days at universities
Decrease the social stigma around mental health and receiving mental health care such as therapy with awareness and empowerment
Increase access and knowledge of where to access mental health care
Build supportive local and global communities around mental health, whether at universities or in the workplace