Extracted main text — title through conclusion, appendix excluded. This is what our citation measures are computed over, published so the extraction can be checked by eye.
13,236 characters · 9 sections · 15 citation commands
Estimating County-Level COVID-19 Exponential Growth Rates Using Generalized Random Forests
Early and accurate detection of community outbreaks is critical to address the threat of resurgent waves of COVID-19. Specifically, an epidemic outbreak is confirmed when its incident cases are estimated to grow exponentially. Furthermore, the potential impact of an outbreak is also measured by its exponential growth rate, as higher rates indicate more rapid disease spread. Last but not least, for a fixed epidemiology model (e.g. SIR, SEIR), there is an one-to-one correspondence between the exponential growth rate of an epidemic outbreak and its basic reproduction number $R_0$, a common measure of intensity of epidemic outbreaks lipsitch2003transmission. Therefore, the exponential growth rate of an epidemic outbreak's incident cases is the most important "model-free" parameter to estimate for detecting the outbreak chowell2003sars. It remains an epidemiological challenge to obtain accurate exponential growth rate estimates for disease outbreaks ma2014estimating. Specifically, it is difficult to choose the fitting window size for the exponential growth rate estimation to balance the speed and accuracy of outbreak detection. On one hand, a longer fitting window is preferable as larger sample size would reduce variance of the exponential growth rate estimates of outbreaks. On the other hand, shorter fitting windows are better at detecting early-stage outbreaks, especially if these outbreaks were driven by recent policy changes such as school reopening. In the current practice, this fitting window size is treated as a hyperparameter that is either directly specified by the user (c.f. Mel) or determined by some cross validation methods (c.f. chowell2007comparative).
This paper develops a machine learning framework that balances the speed-accuracy tradeoff of outbreak detection via dedicated feature engineering and GRF (c.f. athey2019generalized), and apply it to detect county-level COVID-19 outbreaks. Specifically, the algorithm chooses an adaptive fitting window size for each county based on a rich set of features that affect the disease spread, such as face mask mandates, social distancing policies, the CDC’s Social Vulnerability Index, changes in tests performed and rate of positive tests. Furthermore, for counties with insufficient data to capture the recent policy changes, the algorithm pools together all relevant incident case growth trends across U.S. counties and throughout the COVID-19 pandemic history to adjust for these policy changes.
During an epidemic outbreak, incident case number at county $c\in C\subset \mathbb{N}$ on day $t\in T \subset \mathbb{N}$, i,e. $I_{t,c}$, is governed by an exponential growth model,
ma2014estimating.\footnote{See Appendix (ref) for the epidemiological definition of incident case number and how we compute $I_t$.} Here $r$ is the incident cases exponential growth rate of outbreaks for all counties $c\in C$, while $I_{0,c}$ captures the initial incident case numbers at the beginning of the exponential case growth at county $c$. To obtain the most recent exponential growth rate of county-level COVID-19 incident case number, we estimate the instantaneous counterpart of ((ref)),
where the dependent variable is the log-linearized incident case number $ln(I_{t,c})$; independent variable is day $t$. The parameter of interest is the COVID-19 incident case exponential growth rate of county $c$ at day $t$, $r_{t,c}$. The intercept term, $\alpha_{t,c}$, is a parameter capturing the log-linearized initial incident case number of county $c$ at the beginning of the outbreak. $\varepsilon_{t,c}$ is the error term.
Notably, the exponential growth rate $r_{t,c}$ in ((ref)) varies in both day $t$ and county $c$. In other words, we are interested in estimating the instantaneous county-level exponential growth rate of COVID-19 incident cases. Specifically, since COVID-19 related regulations are changing every day in the U.S. (c.f. CUSP), the county-level exponential growth rates, affected by these policies, also change from day to day. Therefore, in order to detect recent outbreaks, we need to estimate the most recent incident cases exponential growth rate. This instantaneous exponential growth rate $r_{t,c}$, similar to the instantaneous reproduction number commonly used in the epidemiology literature (c.f. fraser2007estimating,cori2013new), captures the projected exponential growth rate of incident cases in county $c$ should the future COVID-19 regulations remain the same as those in day $t$.
The exponential growth rate of COVID-19 incident cases can be affected by many factors ranging from day-to-day changes in COVID-19 regulations to difference in population density and healthcare resources between counties. Hence, in order to estimate the instantaneous county-level exponential growth rate $r_{t,c}$ defined in ((ref)), we need to control for these day-level and county-level heterogeneity. Provided that we have relevant features $X_{t,c}\in \mathbb{R}^m$ that captures these aforementioned factors affecting the instantaneous county-level exponential growth rate $r_{t,c}$,\footnote{Refer to Appendix (ref) for the list of relevant features used in this study.} we can identify the conditional average partial treatment effect $r_{t,c}(X_{t,c}):=\mathbb{E}[r_{t,c}|X_{t,c}]$ as defined by wooldridge2010econometric. When data of these relevant features affecting COVID-19 disease spread is available, we can rewrite ((ref)) as
following the common “redundant" assumption (c.f. wooldridge2010econometric), i.e.
This section discusses how we estimate the exponential growth rate of COVID-19 incident cases $r_{t,c}(X_{t,c})$ defined in ((ref)). Specifically, we first formulate the estimation problem in \S(ref) and then present our estimation algorithm in \S(ref).
First, we need the following “unconfoudness" assumption (c.f. rosenbaum1983central):
That is, day $t$ is independent of all unobservable heterogeneity conditional on the feature vector $X_{t,c}$. Under Assumption (ref), we can derive the following moment equations for ((ref)): $\forall X_{t,c}\in \mathbb{R}^m$,
through the instrumental variable $\Gamma(t):=[\mathbb{1}_{0}(t);\mathbb{1}_{1}(t);\mathbb{1}_{2}(t);\dots]$, where $\mathbb{1}_{s}(t)$ is a Boolean function equal 1 if $t=s$.
However, we note that the moment equations ((ref)) cannot identify the exponential growth rate $r_{t,c}(X_{t,c})$. Specifically, the moment equation system ((ref)) is underidentified as the number of unknown parameters $\{\alpha_{t,c}(X_{t,c}),r_{t,c}(X_{t,c})\}_{t\in T, c\in C}$ is exactly two times the number of moment equations. To address this problem, we assume that the exponential growth rate of day $t^*$ equal the local average causal response (c.f. abadie2003semiparametric,angrist1995two) from day $t^*-1$ to day $t^*$, i.e.
Parameters $\{\alpha_{t,c}(X_{t,c}),r_{t,c}(X_{t,c})\}_{t\in T, c\in C}$ are exactly identified by the moment equations ((ref)) under Assumption (ref).
Given the above assumptions, the exponential growth function of COVID-19 incident cases ((ref)) can be estimated using the GRF algorithm by athey2019generalized. Specifically, Assumption (ref) implies that the exponential growth rate of a county $c$ on day $t^*$ is identified by a targeted data “block" $\{(I_{t^*,c},X_{t^*,c}), (I_{t^*-1,c},X_{t^*-1,c})\}$. Hence, when estimating $r_{t^*,c}(X_{t^*,c})$, we can partition the panel dataset $\{(I_{t,c},X_{t,c})\}_{t\in T, c\in C}$ into $|C|\times \left \lfloor{\frac{t^*}{2}}\right \rfloor $ data blocks, and pool those data blocks “similar" to the targeted block to construct an adaptive window size for this estimation. Particularly, the similarity measure is provided by the GRF algorithm. In Appendix (ref), Algorithm (ref) provides the pseudocode to construct these data blocks. Algorithm (ref) explains how we feed these data blocks into the GRF algorithm to obtain the estimates of interest.
To benchmark our method against non-adaptive window size choices, we compare the Mean Absolute Percentage Error (MAPE) of these methods' 7-day ahead predictions. Specifically, we use the NYTimes COVID-19 Dataset (c.f. NYTimes2020) as the source of daily reported cases per county. Additional results for performance evaluation are available at Appendix (ref).
As shown in Figure (ref) and Table (ref), our method outperforms methods with fixed 2-, 4-, 8- or 16-day fitting window sizes. Specifically, Figure (ref) shows that our method provides a uniformly better performance roughly 100 days after the first recorded COVID-19 case in the NYTimes COVID-19 Dataset, when there were enough historical data for the GRF algorithm to conduct meaningful partition. Furthermore, Table (ref) demonstrates that even if the early-day MAPEs are included in the comparison, our method still has the best median MAPE among the 4 methods. Last but not least, we note that when only comparing the performance of non-adaptive window size choices, there are no obvious best choice. While choosing shorter fitting window sizes could in general lead to lower median MAPEs (c.f. Table (ref)), it was still frequently outperformed by longer fitting window size choices (c.f. Figure (ref)).
In this work, we developed a novel framework that allows GRF to adequately balance the the speed-accuracy tradeoff in COVID-19 outbreak detection. This estimation framework can be readily extended to other epidemiology problems where fitting window size impacts model performance.