A four item measure offered as a screen for “panic disorder”
Details #
The PADIS one of a number of what I call “1 + n” or “1 `+ (k-1)” measures. These are measures where the answer to the first question determines whether the respondent is required to answer the other questions (hence “k-1″ as k is the conventional symbol for the number of items in a measure).
The PADIS comes out of Batterham PJ, Mackinnon AJ, Christensen H. (2015). The Panic Disorder Screener (PADIS): Development of an accurate and brief population screening tool. Psychiatry Research, 228, 72-76, I confess I haven’t read that. It starts with this:
“This scale is designed to screen individuals in the community for presence of panic disorder and severity of panic disorder symptoms.
A sudden feeling of anxiety, fear, discomfort or uneasiness may indicate a panic episode. A panic episode occurs unexpectedly, peaks within 10 minutes and includes four or more of the following sensations at the same time:
Hot flushes or chills
Skipping, racing or pounding heart
Sweating or clammy hands
Trembling or shaking
Shortness of breath or difficulty breathing
A choking feeling or lump in your throat
Chest pain, pressure or discomfort
Nausea or stomach problems
Feeling dizzy, unsteady, lightheaded or faint
Feeling strange, unreal, detached or unfamiliar
Fear that you are losing control or going crazy
Fear that you are dying
Feeling tingling or numbness
Based on this description, about how many panic episodes have you experienced in the past month?”
Which is answered on a five level scale:
“None”, “one”, “2-5”, “6-10”, and “11 or more”
That is scored 0 to 4. Anyone scoring zero has finished and scores zero on the PADIS but everyone else is asked these questions:
“In the past month, how often have you been worried about having another panic episode?”
“In the past month, how often have you been worried about the consequences of having another panic episode?”
and
“In the past month, how often did you avoid places, situations or activities because you were afraid of having or triggering a panic episode? (For example, not going to public places alone, not leaving your house, avoiding physical exertion or avoiding stressful activities.)”
Those are answered on a four level scale: “Never”, “Occasionally”, “Often” and “All the time” each scored 0 to 3. For these respondents their PADIS score is whatever they scored for the first item and the sum of the scores of their answers to items 2,3 and 4. That gives a total possible score from zero to 13.
There seem to be some positive psychometric explorations of the measures, see https://nceph.anu.edu.au/research/tools-resources/panic-disorder-screener–padis.
It’s not a major criticism but I have some reservations about simply applying analyses like internal reliability and factor analyses to such “1+(k-1)” as the psychometric theory behind these analyses rather assumes no structural dependence between items making their findings a clear reflection of the presence or absence of inter-item correlations across responses. When responses on some items must be zero if the answer to another item is zero we have “structural dependence” and internal reliability must be the same or higher than it would be if everyone simply answered all the k items. I can see the logic of using this “skip the rest if item 1 scores zero” rule and particularly when the first item requires quite a lot of reading but I do think we need some more thinking about how best to interpret findings, particularly of factor analyses of such data.
My much more serious reservation about the PADIS, but this is one I have about many short measures, conventional or “1+(k-1)” measures, is that describing them as screening measures is going to lead to very poor screening parameters for the measure unless used with a group of people with very high prevalence of whatever you are screening for, here “panic disorder”. (And yes, I am always a bit leary of diagnostic models!)
Try also #
Chapters #
Not covered in the OMbook.
Online resources #
None of mine though at some point I will play around more with simulations of how these “1+(k-1)” measures are likely to behave when their data is crunched with analyses assuming structural independence of items.
Dates #
First created 26.v.26.
