Introducing the Subgroup role in the Response Screening and Fit Model platforms ...

Richard_Zink

Response Screening is a workhorse of a platform available in JMP Pro. It automates the process of conducting tests across a large number of responses, with the test results and summary statistics presented directly in the report or in data tables to enable further data exploration. Further, testing many responses for the effects of factors can be challenging, if not misleading, without using appropriate statistical methodology to protect against false-positive findings. A false-discovery rate (FDR) approach guards against incorrect declarations of significance. Plots of adjusted p-values make the results easily interpretable. Due to the flexibility and performance of Response Screening, it is an often-used platform for the development of new features within JMP Clinical for analyzing many endpoints, time points, or groups, which are common considerations for clinical trials. In JMP Pro 19 and JMP Clinical 19, the new Subgroup role makes it easier than ever before to explore endpoints within subgroups of observations.

First, what’s a subgroup? A subgroup is a subset of a full set of observations, which is often represented by all rows of a data table. Often, subgroups are selected due to their similarity across one or more characteristics that they share, and there may be interest in performing analyses within these subgroups to understand how results may vary from the analysis of the full set of observations.

For the purposes of our discussion, the term factor refers to a variable that is used either alone or in conjunction with other factors to determine subgroups, which we define as the individual levels of a factor or a combination of two or more factor levels. For example, the variable Sex would be considered a factor, while Male and Female would be considered subgroups. Perhaps a second factor, Age, contains two subgroups, < 65 years and ≥ 65 years. Subgroups based on Sex and Age could produce the following four subgroups:

Male, < 65 Years
Male, ≥ 65 Years
Female, < 65 Years
Female, ≥ 65 Years

Throughout this blog post, we illustrate the Subgroup role using the Nicardipine data available within JMP’s Sample Data Index. This data contains multiple rows per patient, one for each adverse event. Select the Unique Subject Identifier column, go to Rows > Rows Selection > Select Duplicate Rows. This highlights all duplicate rows determining duplication using Unique Subject Identifier only. For the purposes of this analysis, using Hide and Exclude on these rows leaves 882 rows available for analysis.

Suppose we wanted to compared Patient Death (Y/N) (using the Patient Died Flag, setting the Target Level to Y) between Nicardipine and Placebo (Description of Planned Arm, setting Placebo as the Control arm) within subgroups based on the following factors:

Sex: M, F
Age (years)
Pooled Race Group 1
Medication/Intervention Flags (Y/N)
- Anticonvulsants Flag
- Antiemetics or Phenothiazines Flag
- Antifibrinolytics Flag
- Antihypertensives Flag
- Blood Transfusion Flag
- Central Venous Pressure Monitoring Flag
- Induced Hypertension Flag
- Intentional Hypervolemia Flag
- Intentional Hemodilution Flag
- Low Molecular Weight Dextran Flag
- Mannitol Flag
- Steroids Flag
- Swan Ganz Monitoring Flag
- Vasopressors Flag

The dialog in Figure 1 enables us to perform the analysis.

Figure 1. Response Screening analysis with Subgroup role

Note the presence of the continuous variable age in the Subgroup role. While it is always possible to create our own categories of continuous variables, Response Screening does this automatically using tertiles. Finally, to get pairwise combinations of the levels of each factor (e.g., males with blood transfusions), select the Subgroup Twoway option.

The output is presented in Figure 2.

Figure 2. Response Screening of death within subgroups

With Subgroup Twoway selected, these 17 factors produce 612 different subgroups (easily visible from the Pvalues data table as there are 612 rows, Save Tables > Save PValues). Of these, 605 subgroups produce p-values (Figure 2). All potential subgroups based on individual and pairwise combinations of factor levels are observed. “Empty” subgroups would be presented in data tables with Count equal to 0 (which is best for transparency and completeness of the subgroup analysis). Subgroups range in size from one patient up to 852 patients.

When a single factor is used to produce a subgroup, as in the first row of the table of Figure 2, Subgroup2 = All. When two factors are used, both Subgroup and Subgroup2 have non-All values listed. The analysis result in the entire population of patients (here, 882) has Subgroup and Subgroup2 both equal to All. Accounting for all of the multiple tests performed with FDR adjustment, there are no differences observed in the rates of death between the treatment arms.

Forest plots of naïve 95% confidence intervals for the relative risk, risk difference, and odds ratio of death are presented on the 2 by M Results tab. These plots confirm few notable differences between treatments, though these analyses are unadjusted for the number of tests performed.

Figure 3. Forest plots of death within subgroups

The Response Screening personality in Fit Model makes it possible to fit more complex models for continuous endpoints to allow for the inclusion of multiple covariates into the analysis of subgroups.

Response Screening in JMP Pro 19 and JMP Clinical 19 makes it easier than ever before to explore results within subgroups. Soon, we’ll discuss the Subgroup Screening report within JMP Clinical that uses the power of Response Screening to gain insight into the primary and key secondary endpoints of clinical trials.