PROMIS^® Reference Populations

PROMIS measures use scores that have meaning. A PROMIS score of 50 is the average (or mean) score for a specific, relevant group of people (e.g., the U.S. general population, kids with a painful condition). That group is the reference population.

PROMIS measures use scores that are more than just numbers: PROMIS scores have meaning. For example, a PROMIS measure score of 50 is the mean score for that measure. But average for whom? This is where “score meaning” comes in. Each PROMIS measure has its own reference population, that is, its own specific group of people (e.g., the U.S. general population, kids with a painful condition) who have been sampled and then thoroughly assessed with the PROMIS measure. The average score of 50 for that PROMIS measure is their average score. Thus, when someone is newly assessed with a PROMIS measure, and we observe that her score is 50, we have a score value, and we can say that her score is the same score as the average score for the PROMIS measure’s reference population – the population group to which we turn to or refer for score meaning.

Reference Populations

A T-score is a standardized score, like z-scores and IQ scores. All standardized scores have a “middle” score; it is zero for z-scores, 100 for IQ scores, and 50 for T-scores. This middle score is the mean of a large sample that is representative of a relevant population—a reference population. The large sample used to represent the reference population is called the Centering Sample.

For some PROMIS measures the reference population (and the centering sample) was a clinical population. This is the case for PROMIS Smoking measures, for which the reference populations was daily smokers; the centering sample was a sample of daily smokers. For many PROMIS measures, the reference population was the 2000 General US Census. The centering sample was a large sample of individuals who represented the 2000 US General Census.

Centering Sample and Calibration Sample

It is helpful to remember that the middle score of a standard score range has to be defined. For measures that use a T-score metric, 50 is the mean and 10 is the standard deviation, but they do not start out that way. The scores are first estimated using an item response model and the IRT-calibrated scores are transformed to a T-score metric using a linear transformation. But first you have to decide which score on the IRT metric is going to be the middle score—a score of 50. This is done by collecting scores from a large sample that represents the reference population and then calculating the mean for that sample. That sample is the centering sample.  That score becomes the middle score (e.g., 50 for T-scores). A linear transformation spaces all other scores along the continuum so they have the correct values relative to the middle score (mean of the centering sample) used to represent the middle score.

IMPORTANT: The Centering Sample and the Calibration Sample may not be the same sample.

The purpose of a calibration sample is to estimate item parameters (item characteristics such as difficulty and discrimination) using an item response theory model. Here’s where it can get confusing. Sometimes a single sample was used as both the calibration sample AND the centering sample. Other times one sample was used as the calibration sample and another was used as the centering sample.  Sometimes, a subset of the calibration sample served as the centering sample.

The Reference Population tables show the calibration and the centering samples for PROMIS. Most users will be particularly interested in the last column (Centering Sample). If you want to know what a score in the middle is (e.g., 50 for those scored on a T-score metric), go to the Centering Sample column. For example, if you go to the row for PROMIS-Cancer-Anxiety you will see that the item parameters were estimated (calibrated) using a hybrid of individuals with cancer and individuals from the general population. BUT, the centering sample was the general population. A score of 50 on this measure is comparable to the general population average level of anxiety.

What does the middle score mean?

When developing a measure with standard scores, an important consideration is what the middle score means. The scores of such measure are purposefully “centered” at the mean of a specific sample or subsample. PROMIS uses T-score, so the middle score is always 50. Centering scores in this way allows quick interpretation of where an individual is on a symptom or outcome compared to others in the reference population. A score of 50 on PROMIS Fatigue, for example, is comparable to the U.S. “average”. T-scores have a standard deviation of 10, so a score of 60 would indicate fatigue that is a standard deviation higher than the U.S. average.

TIP : Failure to be specific about the reference population invites confusion.

This can all get very confusing because sometimes the calibration sample (the sample used to estimate item response theory parameters) and centering sample (the sample used to define the middle of the score range) were the same. But sometimes they were different. For example, a measure may be calibrated in a clinical sample but then centered in the general population. The mean of T=50 for that measure reflects the average in the general population, not the clinical sample.

Norms

A unique aspect of PROMIS measures is their use of standardized scores that are centered on a relevant reference population. Such scores are called “normative” because their value represents how close or far away they are from a normative population. The word “norm” has different meanings for different contexts. Here, we are not talking about social “norms,” the behaviors we expect from others and ourselves in society. We are also not talking about “normal” per se, even though the term originated from its reference to a standard, bell-shaped distribution curve that labels everything in the vast middle as normal. We use the word norm without applying judgment as to the “normality” of any given score relative to the distribution of scores seen on the same measure in a large group of people. Sometimes we refer to the group as a “reference group” and similarly to “norms” as reference values , because they are points of reference from which to understand a given single score.

For example, Jensen et al published reference values for eight PROMIS domains for individuals with cancer. The mean Pain Interference score for people with cancer was 52. Learn more>>

For some PROMIS measures there are subpopulation norms.

Subpopulation Norms

Norms are based on the distribution of scores on a measure for a well-characterized and relevant population. For example, I am 5’6” tall. I might be interested in comparing my height to other people in the world, or maybe just the United States since that’s where I live. On the other hand, I am a woman and so average height without considering gender doesn’t really matter to me. I would be interested in knowing my height relative to other women in the United States. So I might be interested in comparing my height to the mean height of other persons in the United States. According to Wikipedia, the average height for women in the United States is 5’4” and therefore I’m feeling pretty tall right now. I would not feel quite so tall if I were comparing myself to men in the United States (5’9”); and, I would only be eye-to-eye with the average woman in the Netherlands (5’6”), which brings us, finally to normative score comparisons for health outcomes measures.

Percentiles

A percentile can be used to reflect how an individual’s score compares to a reference population. Carle and colleagues (2021) estimated percentiles for many PROMIS Pediatric and Parent Proxy v2.0 (Anger, Anxiety, Depressive Symptoms, Fatigue, Mobility, Pain Behavior, Pain Interference, Peer Relationships, Upper Extremity Function) and v1.0 (Family Relationships, Global Health, Life Satisfaction, Meaning and Purpose, Physical Activity, Physical Stress Experiences, Positive Affect, Psychological Stress Experiences, Sleep Disturbance, Sleep Impairment) measures.

Last updated on 4/29/2024