Is COVID-19 increasing the risk for neurodegeneration? A causal inference study on three real-world population cohorts

Code accompanying a research article by Jannis Guski, Sofie Theisen Honoré, Guillaume Azarias, Steven Sison, Søren Brunak, and Holger Fröhlich. If you have any questions regarding the code or paper, please feel free to get in touch (jannis.guski@scai.fraunhofer.de).

Objective

This is a pipeline to estimate the average exposure effects of SARS-CoV-2 infections (operationalized by either a positive test result or a documented diagnosis U07.1) on the risk of receiving a diagnosis of Alzheimer’s Disease (AD, G30.), Parkinson’s Disease (PD, G20), or unspecified dementia (F03.) in the years after in a whole cohort or strata of a cohort. Our own implementation of Targeted Maximum Likelihood Estimation (TMLE) from the PyTMLE package is used to get doubly robust estimates with a flexible integration of models for nuisance functions.

Instructions for mamba / conda

1. Clone the repository and move into the project directory.
2. Create with mamba env create -f environment.yml
3. Activate environment with mamba activate commute-tmle

Experiments

The configuration is set up in hydra and provides experiments that are a combination of some design choices:

• wave (first wave, second wave, third wave, Omicron wave, all)

• subset (tested positive, hospitalized, all)

• control group design (pre-pandemic control follow-up, equal control follow-up, tested positive vs. tested negative)

Bash scripts

The bash scripts call hydra multiruns for multiple experiments. Change the environment variables like INPUT_CSV or EXPERIMENTS for custom calls.

00_create_mock_input.sh: Creates some random data; only for test or development purposes.

01_set_dates.sh: Selects subsets and sets index / censoring dates for INPUT_CSV (typically from .data/a_inputs) based on the experiments in ./conf/experiment. Results are stored in ./data/b_dates_set.

02_merge_covariates.sh: Maps the covariates to the outputs of 01_set_dates.sh and stores the results in ./data/c_covariates_merged. Either static covariates or the last available value before the index for longitudinal covariates. Needs to be customized to each cohort.

03_fit_tmle.sh: Performs a nested cross validation of an initial hazards estimator, then updates the predictons using TMLE from the PyTMLE package for the datasets with merged covariates from ./data/c_covariates_merged.

The experiment-specific outputs of each script (result tables, plots) will be saved in the multirun folder.

Inputs

The input to 01_set_dates.sh is a cohort-specific csv file with the following fields:

patient_id	birth_date	date_first_tested_positive	date_first_covid_diagnosis	hospitalized_due_to_covid	date_first_tested	date_first_ad_diagnosis	date_first_pd_diagnosis	date_first_unspecified_dementia_diagnosis	death_date	censoring_global
identifier of the patient	date of birth	date, only if applicable	date, only if applicable	boolean, whether the patient was hospitalized due to her / his first COVID infection; only if applicable	date (positive or negative) only if applicable	date, AD (G30) only if applicable	date, PD (G20.*) only if applicable	date, unspecified dementia (F03.*) only if applicable	date, only if applicable	global cohort censoring date; may be the same for the whole cohort