Feature request: faceted K-M plots and tabular summary for CohortMethod results

[jmb01]

Background

Currently, the "Meta" tab provides a good summary of relative incidence across databases, but it lacks information on absolute incidence, which is also important for interpretation. If the user wants to find target- or comparator-specific incidence rates or survival probabilities at specific points during follow-up, they need to view database-specific results and cannot easily compare these across databases, comparators, outcomes, or TARs.

Faceted K-M plots

One solution would be to add faceted (i.e., multi-panel) K-M plots somewhere, likely without the "Number at risk" table to save space. Some ideas for user inputs:

• Select target
• Select up to two faceting variables from: database, comparator, outcome, TAR. If one faceting variable is selected, you can use facet_wrap(), and if two are selected you can use facet_grid().
• Select values for remaining variables besides target and faceting variables
• Toggle confidence limits on/off (as overplotting could be an issue)
• Toggle y-axis between standard K-M survival probability and it's complement, cumulative incidence
• Toggle y-axis limits between fixed, free, or user specified range
• Toggle x-axis limits between fixed, free, or user-specified range

Tabular summary

Another helpful output would be estimates of the survival function at specific timepoints, by database, target, comparator, outcome, and TAR (essentially a prettier version of the cm_kaplan_meier_dist table). Some ideas for user inputs:

• Select database(s), target(s), comparator(s), outcome(s), TAR(s)
• Toggle confidence limits on/of
• Toggle printed estimate between standard K-M survival probability and it's complement, cumulative incidence
• Set decimal places of printed estimate
• Set timepoint(s) of interest

In terms of the last item, I am not sure if CohortMethod is currently outputting the full survival function (i.e., whether the cm_kaplan_meier_dist table contains estimates for every value of time_day at which the function changes, or whether it picks a pre-specified number of timepoints). In the former case, approx() can be used to get estimates for any timepoint requested by the user, but in the latter case I guess the user has to choose from the unique time_day values in the table.