Assessing the duration of overlap of medications and adverse events

Richard_Zink

In the development of new medical products, and more generally in clinical trials and clinical research, it is important to understand the overlap of outcomes that arise within individual patients. Often times, this problem is oversimplified to ascertain whether pairs of adverse events (AEs) or medications may happen simultaneously within-patient. For example, if a patient takes both dexamethasone and phenobarbital, the traditional approach has been to assume that these medications “occurred together” since they were both taken by the patient during the trial. However, closer examination may show that there is no true overlap of these medications in time – one could be taken at the start of the trial, while the other is taken at the end of the study. This is an approach taken in the analysis of spontaneous safety reports in medical product surveillance, where medications and AEs noted on the same report are assumed to “occur together” without any explicit understanding if these events truly occurred at the same time. However, due to the limitations of the data, this is generally the best we can do.

The recent introduction of FDA Medical Queries (FMQs) and some literature on Standardized MedDRA Queries (SMQs), both of which are approaches to group AEs for analyses of “medical concepts” in clinical trials and beyond, has invoked some discussion of the overlap of events (see this article for some discussion). For example, SMQs that utilize algorithms to assess whether multiple AE types occur within patients made no specific mention of assessing the overlap of these events in time. Signals required review to determine whether overlap, in fact, occurred. Several algorithmic FMQs, on the other hand, have explicit overlap as a requirement for the signal to occur. This situation is more challenging to assess, but the Algorithmic FDA Medical Query Risk Report in JMP Clinical evaluates this required overlap.

Assessing this overlap for algorithmic SMQs is a good first step, but can we do better in practice more generally with occurrences in clinical trials? With the forthcoming Co-occurrence report in JMP Clinical 19, we certainly can do better.

First, what is an occurrence? Per the CDISC documentation on Occurrence Data:

“Occurrence analysis is the counting of subjects with a given record or term and often includes a structured hierarchy of dictionary coding categories. Examples of data that fit into this structure include those used for typical analysis of adverse events, concomitant medications, and medical history.”

The Co-occurrence report enables users to explore all pairs of occurrences that happen simultaneously within a patient to obtain estimates of the time overlap as the number of days the paired occurrences happen together. Further, users can compare the incidence of paired occurrences to the marginal frequencies where occurrences happen individually. Finally, occurrence pairs are presented by treatment to assess if certain pairs occur more frequently than others within specific treatments or to compare the amount of time pairs occur together within treatment. Co-occurrence currently considers occurrences such as adverse events (AE), concomitant medications (CM), and deaths.

The dialog for Co-occurrence is shown in Figure 1. Users can refer to AEs or medications according to their individual coded names as specified by MedDRA or WHODrug Global, or they can refer to terms according to their reported text or higher level classification (e.g., High Level Term or Original Category for Medication). These designations become the “Terms of Interest” that are assessed for overlap by Co-occurrence. Users can opt to explore all AEs or medications or subset to terms that occur during specific periods of the study (i.e., Pre-Treatment, On Treatment, Off Treatment Follow-up, Treatment Emergent).

The option Truncate data to the start of the study is a way to limit data, particularly for medications that have been used chronically for many years, to the days after each patient has begun the study. The presence of numerous medications that have been taken for many years can affect performance, but it also assesses some overlap for a period that may be less relevant to the current trial.

Figure 1. Co-occurrence dialog

Running the report for the Nicardipine study provides the following output in Figure 2.

Figure 2. Nicardipine example

The primary output consists of an UpSet plot which has several important areas of focus. UpSet plots are used to communicate the frequency of subgroups produced according to one or more factors compared to the marginal frequencies of considering each factor alone. The vertical and horizontal axes are ordered according to these frequencies so that the further right (or down) in the plot, the smaller the frequency.

The line and dots area of the figure illustrate the connections between Terms of Interest along the Occurrence vertical axis and the Occurrence Group horizontal axis.
The bar chart in the Overall Frequency of Patients in Occurrence Group vertical axis and Occurrence Group horizontal axis presents the frequency of each Occurrence Group, which for now is limited to pairs of occurrences. This bar chart is presented in decreasing frequency, left to right.
The bar chart in the Occurrence vertical axis and the Overall Frequency of Patients with Occurrences horizontal axis presents the frequency of each Occurrence individually. This bar chart is presented in decreasing frequency, top to bottom.
The top area of the figure, within the Proportion of Treatment vertical axis and the Occurrence horizontal axis is a stacked bar chart, which shows the proportion of Occurrence Group pairs that are attributable to a particular treatment. This part of the graph has numerous additional options that are discussed further below.

For example, phenobarbital and dexamethasone are the left-most pair of terms, indicating that they occur with the greatest frequency (n=470) among all possible pairs of terms. This overlap in Terms of Interest can be compared to the individual marginal frequencies of phenobarbital (n=695) and dexamethasone (n=599). In other words, 695-470=225 patients had phenobarbital with no overlapping dexamethasone, while 599-470=129 patients had dexamethasone with no overlapping phenobarbital. The underlined text highlights a very important distinction about the Co-occurrence report, which specifically looks for pairs of Terms of Interest overlapping in time within patients. A patient could take both phenobarbital and dexamethasone, but if there is no overlap on the same study days, it is not counted toward the 470 patients with phenobarbital and dexamethasone.

So, among the 225 patients with phenobarbital with no overlapping dexamethasone:

A subset will not have dexamethasone at all.
The remaining patients will have taken some dexamethasone, but not on the same days as phenobarbital.

Similarly, of the 129 patients that had dexamethasone with no overlapping phenobarbital:

A subset will not have phenobarbital at all.
The remaining patients will have taken some phenobarbital, but not on the same days as dexamethasone.

Of these 470 pairs, 48% and 52% of them occur within patients on Nicardipine or Placebo, respectively. Marginal counts for each treatment are visible in the data table, or the user can click Frequencies by Treatment in the Display Options area to change the bar charts to side-by-side bar charts with sample sizes for each treatment (Figure 3). Although filtering may be needed to view treatment-specific sample sizes, we can observe that 230 and 240 patients on Nicardipine or Placebo, respectively, had the “Dexamethasone, Phenobarbital” pair.

Figure 3. Frequencies by treatment

As mentioned above, the upper part of the UpSet plot (the stacked bar chart summarizing the proportion of treatments) can be modified for other details. As seen in Figures 2 and 3, the UpSet Format is selected as Stacked Bar, which is the default view for the report. Three other views are possible:

Box Plot: Produces side-by-side box plots summarizing the Number of Days of Overlap for Occurrence Groups (Figure 4).
Range Bar: Produces a range bar plot, formatted to appear like a confidence interval, that summarizes the median value as a symbol with low and high interval values represented by the minimum and maximum observed values, respectively (Figure 5).
Range Bar: Produces a range bar plot, formatted to appear like a confidence interval, that summarizes the median value as a symbol with low and high interval values represented by the first quartile and third quartile values, respectively (Figure 6).

Figure 4. UpSet with box plot

Figure 5. UpSet with range bar

Figure 6. UpSet with interquartile range bar

Note that tabular summaries by treatment and overall for the Number of Days of Overlap is available below the UpSet plot.

Figure 7. Summary statistics for the Number of Days of Overlap

The Term of Interest list box in the Display Options makes it straightforward to select all pairs in the data filter that contain a particular term. However, it is important to clear the data filter of current selections since selecting a Term of Interest will respect current filter selections.

For example, clear the Section Filter, and select Cerebral ischaemia in the Term of Interest list box (Figure 8). All pairs that include this Term of Interest are now presented in the UpSet plot.

Figure 8. Occurrence groups with cerebral ischaemia

Unlike the Nearby Occurrences report that expects users to identify reference occurrences of interest to explore other nearby occurrences that appear within a user-defined time window, Co-occurrence looks for the overlap of all possible pairs of occurrences according to a “Term of Interest” and determines the length of time these occurrence pairs take place within each patient. Nearby Occurrences is more appropriate to examine specific terms of interest, while Co-occurrence examines overlap for all terms to see which relationships might arise. Nearby Occurrences is more specific, while Co-occurrence may be more hypothesis generating. In other words, think of Nearby Occurrences as a scalpel, and Co-occurrence as a shotgun.

JMP Clinical 19 makes it easier than ever to understand how important outcomes occur together!