Dealing with missing values in adaptive N-of-1 trials

This dataset provides the results of our simulations on missing value imputation in adaptive N-of-1 trials. It comprises of two variants: One with an effect difference between simulated interventions (folder "10"; for $\mu_1 = 1; mu_2 = 0$) and one without an effect difference (folder "22"; for $\mu_1 = 2; mu_2 = 2$).Within each variant, we repeated simulations inserting missing values in three different manners: random (MCAR) and linearly or exponentially time-dependent (both MAR). At each of these simulation tracks, we simulated 1000 patients for 28 measurements per person. In half of those, 30 % of measurements were set to missing while the adaptive trials simulation continued. We further repeated each track with one of 8 different imputation methods: LOCF, mean (global/individual and overall/treatment-wise), clustering, kNN, and a version of a DR regressor. With this dataset, you can run the notebook missing_results.ipynb in our repository at github.com/HIAlab/missingness_imputation_adaptive_nof1 to compare the results between missingness mechanisms, effect differences, and imputation methods.