This repository contains the code for reproducing the results in the Epidemics paper entitled Estimating nosocomial transmission of micro-organisms in hospital settings using patient records and culture data..

Authors: Jaime Cascante Vega $^{1,4}$, Rami Yaari $^{1}$, Tal Robin $^{1}$, Lingsheng Wen $^{2}$, Jason Zucker $^{2}$, Anne-Catrin Uhlemann $^{2}$, Sen Pei $^{1}$, Jeffrey Shaman $^{1,3}$

Affiliations:

1. Department of Environmental Health Sciences, Mailman School of Public Health, Columbia University, New York, NY, USA.
2. Division of Infectious Diseases, Department of Medicine, Columbia University, College of Physicians and Surgeons, New York, New York, USA.
3. Columbia Climate School, Columbia University, New York, NY, USA.
4. Current affiliation: Department of Biology, Center for Genomics and Systems Biology, New York University, New York, NY, USA

The repository tracks the entire research process in the associated Jupyter notebooks. These were used to process the data, conduct parameter inference, and produce the manuscript figures. As a result, the repository is not well-organized. Below, I list what you would need to reproduce the results. However, the data contains individual patient information and is therefore private. To gain access to the data, please email the corresponding authors: Jeffrey Shaman and Sen Pei.

We use an Iterated Ensemble Adjustment Kalman Filter (EAKF) to infer posterior parameter estimates. We developed another Python package that implements the iterated EAKF. To install the package, run the code below.

git clone -b numpy-version https://github.com/ChaosDonkey06/pompjax.git
cd pompjax
pip install .

We first present the data we used in Figure 1, the notebook associated is in figure 1. The associated Supplementary Material (SM) Figures are those presenting the data: the Ward size distribution (SM Fig S1), the Transfer matrices at the ward and building levels (SM Fig S2), the Hospital traffic at the building level (SM Fig S3), the Hospital traffic at the ward level (SM Fig S4), the Relationship between hospitalizations and admissions (SM Fig S5) and the Length of stay distributions at hospital and building level (SM Fig S6).

We then study the "parameter identifiability" of the agent based model. I use the "" because people mean different things with the term and study it from different perspectives. We conducted inferences on simulated data and investigated if the method recovered the true parameters. In the main text Figure 2 we present those inferences. The notebook to run the inferences on synthetic/simulated data are in jupyter notebook. In the associated SM Figures we present Convergence plots of inference for synthetic data (SM Fig S7), and investigate the Goodness-of-fit of posterior inference on synthetic data, Monte Carlo and statistical error analysis (SM Fig S8). Lastly we present the Hospital-level simulation of synthetic tests and calibration (SM Fig S9). In the main Figure we ran synthetic inferences on importation rates of $\gamma\in${25%, 50%}. The reviewers were concerned on the ability of the system to infer parameters on lower importation rates, so we ran more inferences on $\gamma\in${5%, 10%, 15%}. We present those posterior inferences in Identifiability, parameter estimates on simulated data (SM Fig S10), and show Convergence plots of inferences for synthetic data with low $\mathbf{\rho}$ and $\mathbf{\beta}$ (SM Fig S11). Those inferences were conducted on a cluster the associated python and slurm files scripts are inference.py and inference.sh

We then ran inferences on the real data for the 7 bacterial pathogens. In the main text we present the parameter inferences in Figure 3 (Fig 3) and Simulated with those estimates and measured calibration at the hospital level (Fig 4). The associated SM Figures include first the Marginal posterior parameter estimates for each pathogen (Fig S12) and the associated Building-level simulation of posterior parameter estimates and calibration (Fig S14). The reviewers were concerned with our parametrization of the importation rate $\gamma$. We extended inferences diminishing $\gamma$ until $f=30%$ of the highest value we found in the literature. This sensitivity analysis is presented in Marginal posterior parameter estimates across different importation rate (Fig S13). The files for reproducing the results in a cluster are python file and slurm file. Additionally, as our study period covered the first COVID-19 hospitalizations peak in NYC we conducted an additional sensitivity analysis: Schematic for sensitivity analyses ignoring weeks at the beginning of the study period (Fig S16) and Sensitivity analyses for the beginning of the time series, and the COVID-19 period (Fig S17). The files for reproducing the results in a cluster are python file and slurm file.

The additional SM Figures are associated with a theoretical model for Understanding the effective sensitivity (Fig S15) and the Total clinical cultures in each hospital and it's buildings (Fig S19).