incidence rate enhancements

[jreps]

Expected behavior

Incidence rate currently only lets you calculate proportion and rate using number of people with outcome (only one count per person)

Enhancement behavior

It would be nice to also have the choice to pick number of events (e.g., a person with the outcome 3 times during TAR counts 3 times rather than just once)

[jreps]

• User can pick between event count/person count when calculating incidence proportion/rates
• Would also be good to have a tab in the incidence rate view called 'Diagram' that draws a interactive Venn diagram based on the user settings that can be copied and pasted into publications
• Would be nice to have a check for invalid time at risk settings: (e.g., if a user picks 'Time at risk starts with start date plus 7 days' and 'Time at risk ends with start date plus 0 days')

[chrisknoll]

(e.g., a person with the outcome 3 times during TAR counts 3 times rather than just once)

Question on this: How would this impact the incidence proportion calculation? Currently, it's cases / population, but if you had a situation where the was only 1 person int he analysis, and the person was counted as 3 cases, the IP would be 3/1 = 3?

Related to incidence rate: incidence rate is cases / total follow up time (so, cases per year is the rate). How would multiple cases per person be handled for follow up? Do we follow up to the last case for purposes of 'total follow up'? When the option for allowing > 1 case per person, do we not stop follow up at the event (so then the rate is total cases / total observed time after index)?

[pbr6cornell]

It would be useful to expand the Incidence Rate summary to handle the more general cases around defining time-at-risk and prior events (with the input parameters around censoring), as well as handling multiple events, in a manner consistent with how is currently handled in Estimation and Prediction.

To @chrisknoll question: Incidence Proportion is defined by the unit of the denominator. So, in our context, that denominator is normally 'persons', which means the unit for the numerator may only be 'person'. However, Incidence rates can have different units for numerator and denominator, since denominator is 'person-time', and numerator can either be count of persons or count of events. Censoring rules as input parameters would govern whether multiple events would be allowable. If the user decides to censor at time of first ever event, then a person who has an event prior to target start will not be included in any calculation and a person who has multiple events post-target start will only be included until the time of the first event. If a user decides to censor at at time of first post-index event, then events prior to target start are not considered and all persons in T will be considered at risk. If a person decides to not censor at any event, then it is possible for a person to have events prior to T start and post T start, and the full time-at-risk will be considered, and all events during the time-at-risk can be counted.

[pbr6cornell]

Expanding output within Incidence Summary to show time-to-event distributions and the impact of Censoring would be useful, but can wait until ATLAS 3.0.

[chrisknoll]

Incase this matters: we do calculate distribtion of TTE (time to event) and it is included in the export of results, but we don't have it visible anywhere in the UI. If it's something we want to persue in 2.8 (I don't think it would be hard to present boxplots for the distributions we already calcuate, but if we're thinking about extending them, it shoudl probalby wait utnil 3.0 like you said.