The QRP Reporting Tool outputs an attrition table for all study designs conducted in the QRP, whether descriptive or inferential. The attrition table output for descriptive studies provides a stepwise explanation of how each study cohort was created using aggregated data. Each step in the attrition table describes the reason for patient (or episode) exclusion, as well as the number excluded and remaining at that step. For inferential studies, additional rows are appended to the descriptive attrition table that are specific to comparative analyses. Users should note that attrition tables output by the QRP Reporting Tool are summarized versions of the detailed attrition information available as an output dataset from the QRP. Attrition tables are not created for user-specified subgroups.
The basic format of the QRP Reporting Tool attrition table is similar across all study designs. A single row output by the QRP Reporting Tool in an attrition table may be a summation of multiple steps of the QRP cohort design. The QRP Reporting Tool will only output rows in the attrition table that are relevant to the type of study performed. For detailed information about a particular row shown in this documentation, users should reference the QRP documentation for that feature.
All cohorts requested to be output in the [Groups-File] or analysis groups requested in [L2-Comparison-File] will be output in one attrition table, receiving a separate header to differentiate cohorts or analysis groups.
Descriptive studies executed in the QRP may have been performed entirely on the patient-level, entirely on the episode-level, or using a mixture of both units. Attrition tables output for these studies reflect the design in row labels. For study designs that create exposure episodes, the QRP Reporting Tool will output a similar attrition table as the patient-level table, with slightly different row labels and headings to describe episode instead of patient exclusion.
All study designs (whether patient- or episode-level) begin with a section describing “Members meeting enrollment and demographic requirements.” This section summarizes patients excluded in the first several exclusion steps of the QRP.
The next section in all attrition tables is “Members with a valid index event.” Where applicable, the first row describes the number of patients who had a cohort-defining claim during the query period. An example of a cohort-defining claim in a drug-based cohort would be a pharmacy dispensing with a National Drug Code corresponding to the drug of interest. For studies of medical product use in pregnancy, a cohort-defining claim is always a pregnancy outcome, and the label for this row reflects that definition. A later row in this section describes the number of patients who had an episode-defining index claim during the query period. Episode-defining in this context means that the cohort-defining index claim occurs during the query period after QRP has constructed treatment episodes and applied stockpiling as per the query specifications.
For study designs that construct episodes, the “Members with a valid index event” section is split into two sections, with the first row remaining under the previously specified header and the rest of the rows falling under a section headed “Cohort episodes with a valid index date.” The first row in this section states the number of episodes from among the members remaining at the prior step. As such, the number of episodes excluded at that row is listed as N/A since this is the first step where the QRP has switched to the episode as the unit of analysis. For studies of medical product use during pregnancy, “Cohort episodes” is replaced with “Pregnancy outcome” to reflect the specific meaning of episodes in these studies.
Cohorts are formed by first evaluating enrollment and demographic requirements as well as index events among members, then evaluating index dates, pre-index history, and post-index follow-up among pregnancy episodes. Because of this, the number remaining often increases from the member- to episode-level steps. A footnote will appear underneath the table to remind users of this difference.
The next section in all attrition tables describes patients or episodes with required pre-index history. Here, the number of patients who are excluded for not meeting the collective inclusion/exclusion criteria is displayed within a single row labeled “Met inclusion and exclusion criteria”. The subsequent italicized rows provide a breakdown of the individual inclusion/exclusion reasons. Inclusion criterion are listed as “No Evidence of [CONDLEVEL]” while exclusion criteria are listed as “Evidence of [CONDLEVEL]” where CONDLEVEL is user-defined in the
qrp.InclusionCodesFile. For example, if hypertension was an exclusion criterion, the row would read as “Evidence of Hypertension.”
It should be noted that the number of patients excluded for each individual criterion is calculated irrespective of the other criteria. If some patients meet multiple inclusion/exclusion criteria, the sum of the patients excluded in the criterion-specific rows would be larger than the number excluded in the “Met inclusion and exclusion criteria” row. Therefore, the criterion-specific counts are not mutually exclusive. In analyses where more than one inclusion/exclusion criterion is specified, a footnote will appear underneath the table to remind users of the lack of mutual exclusivity in the criterion-specific attrition rows.
The penultimate section in attrition tables for descriptive studies describes members or episodes “with required post-index follow-up.” This section details members or episodes lost to user-specified post-index enrollment requirements or other parameters that were applied after the patient indexed or episode began. Episode-level attrition tables contain an additional row in this section compared to patient-level attrition tables: “Episode occurred after first event.” This row is applicable to analyses which consider all eligible episodes from a given patient prior to the occurrence of the outcome of interest, since in these study designs any episodes occurring after the outcome of interest are excluded.
The final section in attrition tables for descriptive studies is headed “Final Cohort” and describes the number of patients (and episodes if applicable) in the final aggregated cohort after QRP has executed. Because these rows only provide information and don’t describe patient or episode loss, all “Excluded” columns appear as “N/A.” Studies of medical product use during pregnancy additionally describe the number of non-pregnant matched comparator episodes when this group was requested in the QRP.
Attrition tables for inferential analyses are quite similar to those output for descriptive studies. One distinct feature is that the headers in the top row of the table display labels for Analysis Groups instead of Cohort Groups. This allows the QRP Reporting Tool to output separate attrition columns for different analytical techniques. Because there are always at least two groups in each analysis group of an inferential study, there are two “Remaining” and “Excluded” columns for each analysis group, whereas descriptive studies only display one “Remaining” and “Excluded” column for each cohort group. For instance, if the QRP had executed 1:1 propensity score matching as well as inverse probability of treatment weighting (IPTW), the attrition table would include “Remaining” and “Excluded” columns for the Exposure of Interest and Comparator Groups for the first analysis, followed by the second analysis in the subsequent columns.
Because inferential analyses often apply additional cohort restriction criteria after the original cohorts were created during QRP execution, attrition tables for these studies append additional rows to describe these steps.
The first additional section output for all inferential analyses describes all steps that are applied to create comparable cohorts. This section includes exclusion of patients who initiated the exposure of interest and comparator medical product on the same day, had previously initiated the comparator, or were excluded due to propensity score trimming. As in attrition tables for descriptive studies, only the rows relevant to the analysis are output. The final row in this section provides information on the final number of patients included in the analytical cohort.
The final section for attrition tables of inferential analyses provides information on the final cohort, including the number of events identified and the number of patients with a truncated inverse probability of treatment weight, if applicable. Because the row describing “Number of events in comparative analysis” includes uninformative events, it may not match the number of informative events shown in a requested effect estimate table.
As mentioned above, users may specify custom header labels for groups of study cohorts in descriptive attrition tables. Otherwise, attrition table rows and columns are automatically generated to match the query specifications.
All Cohort Groups requested in Groups File are output in descriptive attrition tables and all analysis groups requested in [L2 Comparison File] are output in attrition tables for inferential analyses.