ePRO reporting should provide data filtering functionality that allows users to view specific subsets of PRO data and exclude other data in a specific ePRO report. Data filtering functionality accommodates choosing a smaller part of the ePRO data set and using that subset for viewing or analysis. The resulting report (referred to as a view) is generally (but not always) temporary meaning the data is kept, but the report image is temporary. The goal is to support ways to extract only what is relevant to the purpose at hand while delivering an engaging user interface.
Provide data filtering as a form of data analytics to return report views showing subsets of PRO data that support subgroup analysis
As a data analytic technique, filtering functionality facilitates patient comparisons (for detailed information on comparative PRO data see the Comparative Information guideline), decision support in consideration of PRO data, and sometimes setting the stage for predictive analytics. For example, when working in a clinical context, filtering ePRO data for outcomes based upon patient characteristics (e.g., age, comorbidities, smoking status) may bring insight into decisions about whether to initiate an intervention. Figure 4L illustrates some examples of filters identified through provider interviews and literature to produce various types of ePRO reports.
One of the challenges when filtering ePRO data to make comparisons among subpopulations is to resist the tendency to jump to predictive conclusions and causality when sufficient data does not exist. While patient characteristics (see examples in Figure 4L) can be identified and used for filtering to identify ePRO subpopulation data and compare filtered data sets, true predictive analytics are not always possible. The data in the ePRO system may not meet the data set size required for some statistical analyses for predictive analytics. For filtered comparison, the subpopulation size included in the filtered report should be disclosed on the ePRO report in order to avoid the danger of making associations that may not be statistically supported. Discussions among clinical and data science communities should inform recommendations about the size of data sets required to provide desired analytics to prevent inappropriate interpretations of data.
Another challenge for providing filtering functionality in a quick, efficient way (e.g., shown as selections from a pick list of patient demographics and characteristics on ePRO reporting systems) is determining what patient characteristics or demographics to offer as filter options(s). Relevant filter options are frequently driven by specialty and practice contexts.
Designers should balance the size of the pick list of filtering characteristic choices (i.e., the parameters available for selection) with the need to conserve both screen real estate and the effort required on the user’s part in selecting filter parameters. Figure 4M provides an example of an ePRO report with a pick list of filter options. To rightsize the number of filtering options, consult with users and other stakeholders regarding what to include (and what filtering defaults to provide) for various ePRO reports.
It is of note that filtering data is only one aspect of data analytics. Aggregating various types of PRO data and coupling PRO data with other data sources (e.g., social determinants of health and EHR elements) can be used more extensively to support analytics and predictions that inform individual patient and population health.