Choose Language Hide Translation Bar
Staff (Retired)
Understanding shift plots in JMP Clinical

JMP Clinical has several features to summarize records from SDTM Findings domains, data that result from “planned evaluations to address specific tests or questions.” In other words, this includes any data from the myriad of tests or procedures that are performed as part of the doctor’s examination: laboratory tests, ECG results, responses to questionnaires, vital signs etc. One available analysis within JMP Clinical generates shift plots for all available tests from within a selected domain. So naturally, a question we frequently hear when presenting the Findings Analysis Menu is “What is a shift plot anyway?"

Shift tables

Before we discuss shift plots, however, we’ll take a step back and describe shift tables. Section of the International Conference on Harmonisation (ICH) guidance E3: Structure and Content of Clinical Study Reports suggests the use of shift tables as a means to summarize individual patient changes in laboratory parameters. See Figure 1 for an example shift table of our favorite blood test used to indicate liver health: alanine aminotransferase (ALT).

In Figure 1, columns are used to indicate frequency and percent (overall percentage) for important categories defined using the baseline measurement (the last measurement prior to dosing) compared with the upper limit of normal (ULN). Rows depict categories using the average of ALT results after a subject begins dosing. From here, we can understand important “shifts” in laboratory results. For example, 31 subjects had ALT < 2 x ULN at baseline with an average on-study elevation in ALT between 5 and 10 x ULN.

Row and column categories should reflect important thresholds for the particular lab test in question, though the categories Low (result  < LLN, Lower Limit of Normal), Normal ( LLN ≤ result ≤ ULN) and High (result > ULN) could alternatively be used. Additionally, shift tables can be presented by treatment to further understand any potential risks due to treatment, presented using alternate summary measures such as median, minimum or maximum values, or generated for each and every visit of scheduled labs after subjects begin dosing to identify when such elevations occur.

Of course, shift tables bombard the reviewer with several columns of numbers – gleaning any meaning from the table will take time, particularly since they’ll potentially need to review a shift table for each treatment arm under study. Further, there are numerous lab tests to examine for important elevations or reductions.

Alternatively, ICH guidance suggests the use of shift plots like the one above in Figure 2. Here, the x-axis represents the baseline lab result normalized by the ULN, taking log2 to temper the skewness of the findings. The y-axis represents the log2 mean of on-study normalized results. A unit change for either axis can be considered a doubling of the result in the original units. The diagonal reference line illustrates no change between baseline and the on-study average (that is, y = x). Because most symbols lie north of the reference line, we can easily conclude that ALT tended to increase on-study. Since the figures are interactive, this allows us to select individual points to easily identify subjects with particularly worrisome findings.

A few final thoughts

The histograms summarize the marginal distribution of the baseline and on-study results. Bars within the histogram can be selected to highlight these individuals within the plot and see the corresponding distribution of on-study measurements (Figure 3) from the other axis. Further,  regression lines can be added to better understand trends in the data (Figure 4, using the Group By and Fit Line options under the Red Triangle menu).

Finally, while we often talk about the interactivity of graphics within JMP Clinical, the same applies to newly available summary tables such as the one in Figure 1. The table can be subset to various combinations of treatment, age, gender and race. Further, the Tabulate control panel allows the user to change overall percentages to column percentages to understand risk relative to each particular baseline category, or remove subjects with missing records at either baseline or on-study and update percentages accordingly.

Though the guidance discusses shift tables and plots to better understand individual changes in laboratory values, these plots are useful to identify outliers from any Findings domain.

1 Comment
Community Member

Big Question:  If one has access only to SAS but not JMP, how does one create these good charts?  Thanks.