1.8 Outputting Tables of Censoring Data in the QRP Reporting Tool

In studies of background rates, exposures with follow-up time, and medical product utilization, users can request that the QRP Reporting Tool output tables summarizing aggregated data on patient or episode censoring in the study. This data can be viewed in three ways:

  1. number of episodes censored overall, by episode length
  2. number of episodes censored for each reason as requested by the user, and
  3. number of episodes censored for a user-specified reason, by episode length.

In studies of manufacturer-level product utilization and switching patterns, users can request that the QRP Reporting Tool output tables summarizing aggregated data on episode censoring in the study. Descriptive statistics are presented for each censoring reason.

1.8.1 Output

Patient censoring data is summarized using the measure(s) of study time requested in the QRP. In the context of these tables, “censoring” refers to any reason that leads to the end of observed time.

“Observed time” is the total time a patient spends in the study post-index until they are no longer present in the data. Because this measure of time does not consider all reasons a patient might have been censored, the following footnote will appear below any censoring table representing observable time: “Time to end of observable data is for characterization purposes only. It does not necessarily represent at-risk time, and does not consider episode end, outcome occurrence, blackout period, or delay risk period start.”

“Treatment time” is similar to observed time, but additionally ends when the exposure episode ends or when a user-specified censoring criterion occurs.

“At-risk time” is similar to treatment time, with the additional restriction that it ends if a user-specified outcome of interest occurs and doesn’t begin until after any user-specified blackout periods end or at the start of a delayed risk window. In a manufacturer-level product utilization and switching patterns analysis, at-risk time ends when a switch occurs. A full description of each reason for censoring and their impact of measures of time can be seen in Table 1.13. These descriptions are output as footnotes in each censoring table that portrays the respective criteria.

Table 1.13: Descriptions of Reasons for Censoring
Censoring Reason Definition Occurrence of censoring criteria ends:18
Disenrollment Represents episodes censored due to disenrollment from health plan. Data Partners often artificially assign a “disenrollment” date equal to data end date for members still enrolled on that date. Therefore, a patient may have dual reasons for censoring as “disenrollment” and “end of data” on the same day - this can be interpreted as right-censoring in most cases.
  • Observable time
  • At-risk time
  • Treatment time
Evidence of Death Represents episodes censored due to evidence of death. Death data source and completeness varies by Data Partner.
  • Observable time
  • At-risk time
  • Treatment time
End of Data Represents episodes censored due to Data Partner data end date. This end date represents the last day of the most recent year-month in which all of a Data Partner’s data tables in the Sentinel Common Data Model have at least 80% of the record count relative to the prior month.
  • Observable time
  • At-risk time
  • Treatment time
End of Study Period Represents episodes censored due to user-specified study end date.
  • Observable time
  • At-risk time
  • Treatment time
Any User-Defined Censoring Criteria Represents episodes censored due to occurrence of additional user-defined criteria using drug, procedure, diagnosis, and/or laboratory codes.
  • At-risk time
  • Treatment time
End of Exposure Episode Represents episodes censored due to end of the exposure episode. In as-treated analyses, exposure episodes are defined using days supplied as recorded in outpatient pharmacy dispensing records, and episodes end after days supplied are exhausted or a pre-determined maximum episode duration is met. In point exposure analyses, exposure episodes end when a pre-determined maximum episode duration is met.
  • At-risk time
  • Treatment time
Outcome of Interest Represents episodes censored due to occurrence of request-defined event.
  • At-risk time
First switch Represents episodes censored due to occurrence of first switch.
  • At-risk time
Second switch Represents episodes censored due to occurrence of second switch.
  • At-risk time

The first censoring table that can be output in studies of background rates, exposures with follow-up time, and medical product utilization describes the length of time an episode persisted in the study, as shown in Table 2. When no stratifications are requested, these tables portray the number of patients or episodes in each category of study time in addition to descriptive statistics of that measure of time. To protect patient privacy, no descriptive statistics are presented in stratified tables.

The second censoring table that can be requested for studies of background rates and exposures with follow-up time—and the first that can be requested for studies of medical product utilization and manufacturer-level product utilization and switching patterns—is a table summarizing the number of patients or episodes censored for multiple censor reasons. Note that in medical product utilization and manufacturer—level product utilization and switching pattern analyses, this censoring table will provide descriptive statistics on the time to censoring. To output a table describing the number of patients or episodes censored for one particular reason by length of time before censoring, users can output an additional table for studies of background rates, exposures with follow-up time, and medical product utilization. Because censoring reasons are evaluated independently in the QRP (see QRP documentation), it is possible for a patient to meet multiple censoring criteria in the same day. Thus, when multiple censoring reasons are presented in one table, a footnote will read: “An episode may be censored due to more than one reason if they occur on the same date. Therefore, the sum of the reasons for censoring may be greater than the total number of episodes.”

When requested, stratified tables are assigned the same table number as the corresponding non-stratified—i.e., “overall”—table, with the addition of a letter. If multiple tables describing patients or episodes censored for one particular censoring reason are requested, each table will receive its own number in the following order: end of exposure episode, occurrence of outcome of interest, occurrence of any user-defined censoring criteria, evidence of death, disenrollment, end of data, and end of study period. Stratified censoring tables are not available for manufacturer-level product utilization and switching patterns.

Table 1.14: Requestable Censoring Tables
Study Design Measure of Time Unit of Analysis Table Order Table Title, followed by “for [ReportTitle] in the [Database Name] from [Query Start] to [Query End]”19
Background rates Observable time Patients 1 Summary of Time to End of Observable Data
Observable time Patients 2 Summary of Reasons for End of Observable Data
Observable time Patients 3 Summary of Time to End of Observable Data due to [CensorReason]20
Exposures with follow-up time Observable time Episode 1 Summary of Time to End of Observable Data
Observable time Episode 2 Summary of Reasons for End of Observable Data”
Observable time Episode 3 Summary of Time to End of Observable Data due to [CensorReason]21
At-risk time Episode 1 Summary of Time to End of At-Risk Period
At-risk time Episode 2 Summary of Reasons for End of At-Risk Period
At-risk time Episode 3 Summary of Time to End of At-Risk Period due to [CensorReason]22
Medical product utilization Treatment time First treatment episode 1 Summary of Reasons First Treatment Episodes Ended
Treatment time First treatment episode 2 Summary of Episode Duration for First Treatment Episodes Ended due to [CensorReason]23
Treatment time Treatment episode 3 Summary of Reasons Treatment Episodes Ended
Treatment time Treatment episode 4 Summary of Episode Duration for Treatment Episodes Ended due to [CensorReason]
Manufacturer-level product utilization and switching patterns Treatment time Treatment episode 1 Summary of Episode Duration by Reason Episodes Ended
At-risk time Treatment Episode 2 Summary of Time to First Switch or Episode End
At-risk time Treatment episode 3 Summary of Time to Second Switch or Episode End

1.8.2 User Options

For studies of medical product utilization, users must specify integer-based divisions of treatment time that should cover the full spectrum of possible values usingthe QRP Report Table File. In each division, the upper bound is binding and the final division should end with the + symbol. In other words, the time value will fall within a given division when the value of the measure is less than or equal to the upper bound of that division. The divisions of continuous time output for studies of background rates and exposure with follow-up time are user-specified in qrp.Type1File and qrp.Type2File respectively and are carried over into the report.

For censoring tables that show specific reasons for censoring—i.e., all tables for studies of medical product utilization and manufacturer-level product utilization and switching patterns, and tables 2 and 3 otherwise—users may specify which censoring reason should be output inthe QRP Report Table File. If no censoring reasons are specified for said tables, all applicable reasons will be output.

Users may also choose to re-label any censoring reason using the QRP Report Label File. Of note, censoring reason labels will apply to all requested cohorts.

If desired, users can choose to output multiple cohort groups under a combined header by specifying the appropriate parameter in the QRP Report Label File.

Users may request that censoring tables for studies of background rates and exposures with follow-up time be stratified using the appropriate structure inthe QRP Report Table File. All studies for which censoring tables are output can also be stratified by Data Partner upon request in the QRP Report Create Report File.

Table titles are shown in 1.14, and allow users to specify a custom “Report Title” for each censoring table. When not specified, the default label “Exposures of Interest” will be used.

1.8.3 Calculations

Before the measures of time can be presented in the QRP Report, they must first be aggregated. To do so, the QRP Reporting Tool identifies unique values of each measure and sums the relevant analysis variable from each data partner for each unique combination of QRP run, cohort group, stratification level, and censoring reason (if applicable).

Given:

  • \(x_i\) is the non-missing value of the utilization measure for observation \(i\);
  • \(f_i\) is the frequency that is associated with \(x_i\);
  • \(n\) is the number of non-missing values of \(x_i, \sum f_i\);
  • \(n_c\) is the number of non-missing values of \(x_i, \sum f_i\) contained within a given category;
  • \(p\) is \(\frac{t}{100}\) for the \(t^{th}\) percentile;
  • \(j\) is the integer component of \(np\) where \(np=j+g\); and
  • \(g\) is the fractional component of \(np\) where \(np=j+g\);

Then, each metric reported in censoring tables is calculated as described in Table 1.15 below.

Table 1.15: Calculations for each metrics reported in censoring tables
Metric Reported Calculation
Total number of [count variable] \(n\)
Number of [count variable] in category \(n_c\)
Percent of all [count variable] \(\left(\frac{n_c}{n}\right)\times100\)
Minimum The minimum value of \(x_i\)
Q1, Median, Q3 \(y=\frac{1}{2}\left(x_j+x_{j+1}\right)\) if \(g=0;y=x_{j+1}\) if \(g\gt0\)
Maximum The maximum value of \(x_i\)
Mean The arithmetic mean, \(\frac{\sum x_i}{\sum f_i}\)
Standard deviation The square root of the variance, \(\sqrt{\frac{1}{n-1}\sum f_i\left[x_i-\bar{x}\right]^2}\)

  1. Unless otherwise specified in the QRP↩︎

  2. [ReportTitle] is user-specified in the QRP Report Label File, [Database Name] is user-specified in the QRP Report DP Info File, [Query Start] is user-specified in qrp.MonitoringFile, and [Query End] is the maximum date of all the included data partner end dates (unless an explicit follow-up end date is specified in qrp.MonitoringFile and is earlier than the maximum data partner end date, in which case the end date is the follow-up end date).↩︎

  3. [CensorReason] is the user-specified censoring reason shown in the corresponding table↩︎

  4. [CensorReason] is the user-specified censoring reason shown in the corresponding table↩︎

  5. [CensorReason] is the user-specified censoring reason shown in the corresponding table↩︎

  6. [CensorReason] is the user-specified censoring reason shown in the corresponding table↩︎