Form Design: Setup Considerations for Repeating Row Tables

Sometimes a form that looks straightforward turns out to have several reasonable ways it could be built.

Here’s an example we recently worked through.

The form records abnormal physical examination findings. For each finding, we need three pieces of information:

• Body System
• Abnormality: clinically significant or non-clinically significant
• Finding Description

The paper form gives us four rows for entering findings. At the bottom, though, it lists 11 possible body systems, including General Appearance, Musculoskeletal, Central Nervous System, Cardiovascular, and others.

That immediately raises a question: What happens if five body systems have abnormal findings?

There’s nothing about the data itself that limits us to four. In fact, all 11 body systems could potentially have findings. The four rows are really a limitation of the paper form, not necessarily a limitation we want to carry into the electronic version.

So how should we build it?

As Thomas Sowell put it, “There are no solutions, only trade-offs.”

Option A: Add Rows as You Need Them

The first approach is probably the most natural translation of the paper form.

Start with one row containing three fields:

Body System | Abnormality | Finding Description

Body System is a coded field with 11 possible values. Abnormality has two choices: clinically significant or non-clinically significant. Finding Description is a text field.

If another abnormal body system needs to be recorded, the user clicks New and another row appears.

On the back end, each finding becomes its own record with the same three variables:

Body System | Abnormality | Finding Description

So if a participant has three abnormal body systems at a visit, there are three records.

That’s the big advantage of this approach. The dataset stays simple. Whether a participant has one finding or eight, the statistician is still working with the same three variables. Counting findings, looking at specific body systems, or summarizing abnormalities is relatively straightforward.

Option B: Repeat the Fields Behind the Scenes

Form B can look and behave exactly like Form A to the person entering the data. They still add a new row whenever another abnormal body system needs to be recorded.

The difference is entirely behind the scenes.

Instead of creating a new record for each finding, we define enough fields in advance to accommodate all 11 possible findings:

Body System 1 | Abnormality 1 | Finding Description 1
Body System 2 | Abnormality 2 | Finding Description 2
Body System 3 | Abnormality 3 | Finding Description 3

And so on through Body System 11.

That’s 33 variables altogether.

Now the visit remains a single record in the exported dataset, with the findings spread across columns rather than rows.

From a data-entry perspective, Forms A and B may be indistinguishable. From the statistician’s perspective, they’re very different.

With Form B, the statistician has to keep track of as many as 33 variables. The numbered positions also don’t have consistent clinical meaning. Body System 1 could be Cardiovascular for one participant and Musculoskeletal for another.

The advantage is that the data-entry person can add findings in whatever order they encounter them while all the information remains on a single visit record. That can be useful when there may be a relatively high number of abnormalities and you don’t know which systems will be involved or the order in which the information will be gathered.

The trade-off is considerably more work with the resulting dataset.

Option C: Assess Every Body System

Form C takes a different approach. Instead of asking the user to select a body system and add rows as needed, put all 11 body systems directly on the form.

Because the body system is built into each row, we no longer need a separate Body System variable. Each body system gets two variables of its own: Abnormality and Finding Description.

That gives us 22 variables rather than 33.

We’d also change the Abnormality field. Instead of only recording the two types of abnormal findings, each body system would have four possible responses:

• Normal
• Abnormal, clinically significant
• Abnormal, non-clinically significant
• Not documented

That’s really the main reason to consider Form C. The user makes an explicit determination for every body system.

With Forms A and B, we’re recording the body systems where an abnormality was found. With Form C, we get a value for every body system. We know that General Appearance was normal, abnormal, or not documented. The same is true for HEENT, Neck, Cardiovascular, and every other system.

The addition of Not documented is important. It keeps us from treating the absence of information as evidence that a body system was normal.

The trade-off comes on the analysis side. The statistician now has separate variables for General Appearance Abnormality, HEENT Abnormality, Neck Abnormality, and so forth. Something as simple as counting the number of abnormal systems requires working across those different variables.

So Which One?

There are a few things I’d consider when deciding among them.

How common are abnormal findings? When abnormalities are relatively uncommon, Forms A and B have an advantage. The person entering the data only deals with the body systems where something was found. If there are two abnormal systems, they enter two rows rather than working through all 11. That minimizes the number of clicks and keeps data entry efficient.

As abnormalities become more common, Form C becomes more attractive. Rather than adding body systems one at a time, all 11 are already there. More importantly, you’re documenting that each body system was considered and a determination was recorded: normal, abnormal and clinically significant, abnormal and non-clinically significant, or not documented.

What information is actually available? This may be just as important as how common the abnormalities are. If someone is abstracting information from a medical record, the source may document abnormal findings without explicitly documenting normal findings for every body system.

In that situation, Forms A and B may be a natural fit. The data-entry person captures the abnormal findings that are documented without having to search for and account for every body system.

Form C asks for more. The data-entry person works through all 11 systems. If the source supports a determination, they record it. If the information isn’t there, they select Not documented. That preserves the distinction between a body system that was normal and one for which the necessary information simply wasn’t available.

The clinical setting can therefore matter. A specialty record may contain excellent information about the systems relevant to that specialty while saying little about the others. A more comprehensive examination may reasonably provide enough information to work through the entire list.

What do we want the dataset to look like? Form A gives the statistician a small number of variables and a separate record for each finding. Forms B and C keep the information together on the visit record but create considerably more variables to manage. Which structure is easier depends in part on what analyses and summaries need to be produced from the data.

Do we need flexibility in the order of entry? Form B can be useful when there may be a relatively high number of abnormalities, but you don’t know which systems will be involved or the order in which the information will be gathered. The data-entry person can add findings as they go without having to follow a predefined sequence. The trade-off is the 33-variable structure the statistician has to deal with later.

What have we done before? I almost always ask this. The research team may already have datasets, analysis programs, reports, or other routines built around a particular structure. If existing code expects the data in a certain format, there can be real value in staying consistent rather than redesigning everything from scratch.

These decisions are usually best made with input from the people entering the data, the statistician who will work with it, and whoever is designing and setting up the study. Each brings a different perspective, and someone who regularly builds studies can also bring experience with how similar design choices have worked in the past.

In practice, Forms A and C are probably where I’d land most of the time. Form A works particularly well when you’re capturing documented findings and want a compact, easy-to-analyze dataset. Form C becomes attractive when it’s reasonable to work through every body system and you want to explicitly distinguish normal, abnormal, and not documented. Form B has its place, but I’d generally want a particular reason for taking on the additional complexity.