Angel Claudio, Bonnie Cooper, Manolis Manoli, Magnus Skonberg, Christian Thieme, Leo Yi
"2021-05-23"
The global COVID-19 pandemic has greatly impacted disease beyond direct cases of COVID-19.
Reports have identified statistically significant decreases in diagnoses for communicable diseases such as pneumonia and influenza.\( ^{1,2,3} \)
Hypothesis 1: Disease mitigation efforts (e.g. masks, social distancing) have directly reduced non-COVID disease cases.
Other reports show that while disease diagnostic rates have decreased for diseases such as heart attack and stroke, mortality rates for these diseases have increased during COVID.\( ^{4,5,6} \)
Hypothesis 2: Changes in patient behavior and health care availability during the pandemic have led to missed cases of disease.
Our Goal: Use U.S. state-level disease data on pneumonia and influenza in conjunction with data on state government enforced COVID regulations to look for an association between disease and disease mitigation efforts.
If government restriction data significantly impacts the modeling of pneumonia and influenza mortality data, we interpret this as support for Hypothesis 1

We measured the excess PI mortality as:
\[ \mbox{Excess PI} = \\ \mbox{max}( \mbox{PI}_{weekly} - \mbox{expected}_{weekly}) \]
Excess PI for the 2020-21 flu season is significantly different when compared to all other flu seasons in our dataset
Next, we look at the relationship of excess PI deaths and COVID related deaths
Figure A) state-level excess PI
Figure B) Tukey Honest Significant Differences of Means test
We observe a strong correlation between excess PI deaths and COVID related deaths

To test Hypothesis 1 we incorporated state-level government restriction data

To model the impact of government restriction data on excess PI mortality, we used a simple linear regression model as a baseline (Model 1) and developed four additional models:
| Model | Method | Var_Num | R2_train | RMSE_train | R2_test | RMSE_test |
|---|---|---|---|---|---|---|
| 1 | Simple Linear | 1 | 0.8056 | 2.3288 | 0.8056 | 2.5144 |
| 2 | Mult. Linear | 7 | 0.8207 | 2.2255 | 0.8207 | 2.3869 |
| 3 | Mult. Linear | 8 | 0.8434 | 2.0714 | 0.8434 | 2.3098 |
| 4 | Ridge | 6 | 0.8403 | 2.1129 | 0.7875 | 2.3333 |
| 5 | Lasso | 6 | 0.8403 | 2.1129 | 0.7874 | 2.3341 |
Table 1: Model performance metrics used to evaluate and select from different modeling approaches
Although our model improves upon the description of excess PI mortality, there is room for improvement: diagnostic plots reveal that residual variance is not constant across the model fit (left panel) and the normality plot of residuals suggests high-skewed distribution
