Time-to-event analyses for adverse events and medical queries

[#1]

For clinical trials in medical product development, the default analyses for adverse events (AEs) tend to feature incidence analyses for each and every preferred term (PT), organized by system organ class (SOC). For example, if a patient experiences one or more headaches, they are considered a “success” in that the event occurred (= 1). Otherwise, patients who do not experience a headache are considered failures (= 0). The number of patients who experience one or more headaches are tabulated for each treatment arm, and the proportion (or odds ratio or risk ratio) of patients experiencing a headache are compared between the treatment arms. This analysis has an important limitation in that the timing at which these events occur is completely ignored within the analysis.

To create a more informative analysis, non-parametric Kaplan-Meier (KM) analyses can be used to consider the timing up until the first event occurs, censoring patients who do not experience an event at the end of an appropriately defined follow up. These time-to-event (TTE) analyses are rather straightforward, but likely not performed often in practice due to the sheer number of data visualizations required for interpretation. I believe this data visualization burden is a major reason recurrence analyses are also not performed regularly (as I mention here and here).

In JMP Clinical 19.2, the Adverse Events Time to Event (TTE) report has been revised. Instead of an analysis that considers the time to study discontinuation due to an AE, it now includes an analysis comparable to the Adverse Events Risk Report or Adverse Events Recurrence that analyzes each PT individually. A complementary report for medical queries, including Standardized MedDRA Queries (SMQs) and Office of New Drug Custom Medical Queries (OCMQs) has also been produced: Medical Query Time to Event.

In the examples that follow, we focus on the treatment emergent adverse events (TEAEs) that occurred in a clinical trial of patients with probable mild-to-moderate Alzheimer’s disease. For background, the trial duration was 26 weeks with 254 patients randomized to one of three treatment arms, high or low dose xanomeline or placebo. A total of 242 different preferred terms (PTs) were observed during the study, with 230 PTs observed among the TEAEs. A total of 1,126 TEAEs were experienced, indicating some recrudescence of PTs within patient across time.

Our principal focus is on pruritis (itching) TEAEs. Let’s begin with Adverse Events Time to Event. Figure 1 presents events plots by patient (left) and arm (right). The markers indicate the timing of the first pruritis event that occurs for each patient. Any follow up occurring after the pruritis event in the patient event plot is presented with a dotted line, which emphasizes the duration of time after the first event to help the analyst understand whether patients continue to stay in the trial after this event or drop out. Both event plots communicate that the majority of first events occur within the first 100 days, with several patients in the high dose having itching events almost immediately.

Figure 1. Event plots by patient and treatment for pruritis TEAEs

Figure 2. Kaplan-Meier plot of pruritis TEAEs

Figure 2 presents a failure plot version of the KM plot. The y-axis starts at 0, implying no patients start with pruritis TEAEs. By day 200, nearly 40%, 29%, and 11% of high dose, low dose, and placebo patients have had at least one pruritis event, respectively. The time at risk table (adjustable) communicates the number of patients still at risk for having a pruritis event or censoring outcome. The KM plot (Figure 3) can be modified by the user to include confidence intervals (pointwise and/or simultaneous), details on the outcome of the log-Rank and Wilcoxon tests, or to change the Number at Risk table to summarize Cumulative Events, the number of patients experiencing an event up to and including the summarized day (26, 21, and 8 patients for high dose, low dose, and placebo patients, respectively).

Figure 3. Kaplan-Meier plot of pruritis with intervals, tests, and cumulative events

Of course, the big challenge is to screen across all events. A double-dot plot (Figure 4) of the cumulative proportions of events and the differences in those proportions as well as a max difference plot (Figure 5) are produced. The double-dot plot is produced at the minimum treatment follow-up time observed across all events so that a single time can be used to consistently compare all PTs. For this example, this follow-up time is 197 days, meaning that the cumulative proportions presented in Figure 4 are computed at that time, with pruritis showing the greatest differences with placebo. So why a double-dot plot and not the more traditional dot-forest plot that is computed for Adverse Events Risk Report or Adverse Events Recurrence? It’s due to how the confidence intervals (CIs) are produced for each KM curve, based on a transformation of the curve to ensure that the CIs are strictly non-negative (as proportions tend to be). There simply is not a straightforward computation to compute this interval based on the KM estimate, and rather than use these proportions in a CI formula based on normal approximations (which would not align with the KM plot), I left out CIs entirely in the right panel. What is presented are the estimates of the differences in cumulative probabilities between the active doses with placebo (the user may change the control group).

Figure 4. Double-dot plot of the cumulative proportion of events at minimum treatment follow-up

Figure 5. Max difference plot of the cumulative proportion of events

However, similar to the downsides in recurrence of using a single time point to determine the importance of an event, Figure 5 communicates the maximum difference in KM curves between the available treatments. The plot is subset to only include curves where the maximum difference between KM curves was at least 10% at least once. This limits the screening effort to seven different TEAEs, with pruritis showing the greatest risk over time (though application site pruritis also exhibits notable differences through the first 100 days, only losing to pruritis by the end).

While analyzing individual events using PTs is useful, additional analyses can be performed to consider groups of events based on specific characteristics by using the Collapse Events option and the Adverse Events Filter. Clicking Collapse Events will produce a TTE analysis to the first TEAE of any kind (Figure 6). Given the large preponderance of dotted follow-up lines, most patients who experience their first AE tend to stay in the trial. The right panel shows that many AEs occur for high and low dose patients at the start of the trial, possibly signaling tolerability issues.

Figure 6. Event plots by patient and treatment for all TEAEs

Figure 7 presents a KM plot of all TEAEs. By Day 60, 93%, 85%, and 62% of patients have experienced at least one TEAE. This grouped analysis is extremely informative, and can be applied to produce informative TTE analyses in other ways:

• Moderate and Severe TEAEs.
• TEAEs with notable relationships to study drug.
• Serious adverse events (SAEs).

Figure 7. Kaplan-Meier plot of all TEAEs

Medical queries (SMQs and OCMQs) are an additional type of “grouped events” that seeks to collapse the large number of preferred terms into a more manageable set of terms that represent medical concepts. As seen in Figures 4 and 5, for example, “pruritis” and “application site pruritis" are analyzed as separate events. Two other terms with “pruritis” exist among TEAEs, “eye pruritis” and “pruritis generalised” Medical queries (MQ) group related PTs into MQ terms. Analyses of TTE for MQs can be performed using Medical Query Time to Event (Figure 8). Users can specify the query type, version, scope, and the PTs to consider when producing the analysis.

Figure 8. Medical Query Time to Event dialog

Perhaps unsurprisingly, an analysis of narrow-scoped OCMQs identifies the pruritis MQ as having the greatest differences between the active doses and placebo, with nearly 75% of high dose patients experiencing at least one pruritis MQ by the end of follow up. This is nearly double the 40% rate for the TTE analysis of the pruritis PT in Figure 3!

Figure 9. Kaplan-Meier plot of OCMQ pruritis (narrow scope)

So, to review, JMP Clinical has the following major analyses:

• Incidence (ignores time and multiple events).
• Time to event (considers time, but ignores subsequent events beyond the first).
• Recurrence (considers time and all events).

These analyses will be complete for both AEs and MQs with the release of JMP Clinical 20 in 2027 – which is when Medical Query Recurrence becomes available. Similar to AEs, there is no reason to limit the analysis of MQs to the first occurrence. This will be a topic for a future blog!