Three widely used health questionnaires can agree strongly overall while still measuring different parts of people’s health. This study shows why a single summary score can hide important differences. Health-related quality of life information is used in clinical decision-making and also supports the allocation of funds in the health care system.

There are many different questionnaires used to measure health-related quality of life, but the most appropriate questionnaire in different groups of people is not known. This study compares three popular ways of measuring health-related quality of life to understand the relationship between them.

The results suggest that the concepts measured have both similarities and differences, and the scoring systems have different characteristics that could lead to different inputs into clinical decision-making and about how to fund the healthcare system. The PROMIS item banks and PROMIS-29 are becoming established in many health settings in the United States and are being promoted internationally.

The PROMIS-29 is the most widely used PROMIS profile measure, while the EQ-5D is in established use in many countries. The proliferation of instruments within the same measurement space, but developed using different approaches, means that the measurement characteristics of each measure require assessment.

That assessment helps users understand the usefulness of each instrument in different populations and the consistency between evidence produced by each. The aim was to assess the measurement relationship between the EQ-5D-5L, PROMIS-29, and PROPr using self-reported data from Australia.

Tests of agreement and construct validity were used to generate evidence about the use of EQ-5D-5L and PROMIS-29 as alternative measures of patient-reported outcomes. The data came from an online Australian survey administering health-related quality-of-life measures to patients with common conditions and the general population.

The common conditions targeted included diabetes, depression, pain, and arthritis, and respondents were invited via email and online advertisements. The instruments were completed in a random order, so approximately half of the sample completed the EQ-5D-5L before the PROMIS-29 and vice versa.

Demographic and self-reported health questions were also administered. PROPr allows preference-based scores to be estimated from health states described by seven PROMIS domains and can be estimated from the PROMIS-29. The PROMIS-29 and PROPr share six domains: PF, D, F, PA, SL, and SOC.

The final domain, CF, is not included in PROMIS-29, so it had to be imputed. Because the data predated the currently recommended approach, the imputation approach developed by Dewitt and colleagues, using linear regression approaches to estimate CF, was used.

EQ-5D-5L and PROPr utilities, and PROMIS-29 domain T-scores, were compared descriptively. The internal consistency of PROMIS-29 domain scores was estimated using Cronbach’s Alpha, with a range of zero point seventy to zero point ninety-five indicating a positive rating.

The overall distribution of utility values was displayed using histograms, and Pearson’s moment coefficient of skewness was used to compare distributions. The frequencies of commonly reported EQ-5D-5L health states and PROMIS-29 T-score patterns were also examined.

In total, 794 respondents, or 87.5 percent of those accessing the survey, fully completed it. Among those who accessed the survey but did not fully complete it, 40 dropped out before completing any question, leaving 73 answering at least one survey question.

Twenty-four respondents provided incomplete EQ-5D-5L or PROMIS-29 data, so sensitivity analysis including this small number was not conducted. Overall, 500 respondents, or 63 percent of the sample, reported at least one long-term health condition, with 52 percent reporting comorbid health conditions.

Table two describes the seven hundred ninety-four respondents included in the analysis. The sample was fifty-two point one percent female, seventy-eight point nine percent born in Australia, and included substantial health burdens: two hundred ninety-two people reported zero conditions, while two hundred eighty-eight reported three or more.

It also shows that thirty-five point three percent had six or more GP visits in the last year, helping the authors characterize the population underlying the PROMIS and EQ-5D-5L comparisons. The utilities for PROPr differ from those for the EQ-5D-5L: PROPr has a substantially lower mean value and a smaller range of reported values.

The EQ-5D-5L scores also differ significantly across value sets, particularly in the range of values for the same health states. The best health-state value on PROPr is zero point nine zero five, which does not equate to full health. Mean PROMIS-29 domain T-scores are between 47 and 53, indicating a level of equivalence with the United States population, and internal consistency ranges from zero point eighty-six to zero point ninety-five.

Table 4 reports correlations between dimensions and value sets as an indicator of convergence for the overall sample and for people with physical and mental health conditions. Correlations greater than zero point five are defined as strong, and strong correlations appear between dimensions where the measured concepts were hypothesized to be similar.

These results demonstrate overlap between the health-related quality-of-life constructs measured. Lower correlations with PROMIS fatigue and sleep suggest that fatigue and sleep problems are indirectly, but not explicitly, measured by EQ-5D-5L.

Table four reports correlations between PROMIS-29 domains, EQ-5D-5L dimensions, and Australian and United States value sets, for the whole sample and for physical- and mental-health condition groups. The hypothesised links are often strong: for example, PROMIS pain correlates with EQ-5D-5L pain/discomfort at zero point seven one in the whole sample, while PROMIS physical functioning correlates with mobility at negative zero point seven six.

The authors use these results as evidence of convergent validity, while noting that longitudinal data are still needed to assess changes over time. At the dimension and value-set level, PROMIS dimensions are strongly correlated with EQ-5D-5L value sets at a generally higher level than EQ-5D-5L dimensions are with PROPr.

There is a low correlation between EQ self-care and PROPr, while the correlations between EQ-5D-5L value sets and PROPr are strong. This demonstrates a strong relationship between the values at the utility level. However, some differences in the measurement relationship between dimensions are not detected at the utility level.

Both the EQ-5D value sets and PROPr can distinguish between the majority of groups, with effect sizes in the high range. PROPr distinguished between levels of self-reported health at a higher level than EQ-5D-5L, indicating that PROPr might be more sensitive in the general population and in patients with mild problems.

EQ-5D-5L and PROPr distinguish between groups defined by the presence or absence of a health condition, with an EQ effect-size range of zero point seventy-eight to zero point eighty-three. All three value sets are sensitive to pain, depression, and anxiety differences at a strong level, but less sensitive to the impacts of hypertension.

Table 5 presents known-group validity results for the Australian EQ-5D, the United States EQ-5D-5L, and PROPr. All reported group comparisons are significant at less than zero point zero zero one, with effect sizes generally in the high range; for example, the condition comparison has effect sizes of zero point eight three, zero point eight zero, and zero point seven eight, respectively.

The smallest discrimination is reported for GP visits, with effect sizes ranging from zero point one two to zero point five three, while PROPr shows an effect size of zero point nine five for self-rated health, compared with zero point six four for EQ-5D-5L. At the descriptive-system level, both measures exhibit acceptable properties, particularly in detecting differences between respondents’ self-reported health levels.

There is consistency between dimensions measuring similar constructs, but each measure also includes constructs not explicitly assessed by the other. More respondents are likely to report no problems on EQ-5D-5L dimensions than on PROMIS domains.

Each PROMIS-29 domain includes four items rather than one and asks questions using a combination of frequency and severity. Lower correlations between EQ self-care and PROMIS domains may suggest that self-care is not clearly captured by PROMIS-29. That result is confounded by the low variation in self-care scores displayed by the sample, and further research could examine it in a patient population with a higher level of self-care-related issues.

Similarly, EQ-5D-5L could be limited in populations where sleep problems and fatigue are important concerns. PROMIS physical activity is highly correlated with mobility, usual activities, and pain or discomfort, but not self-care. All of the value sets exhibit strong known-group validity, but different value-set characteristics affect the distributions of data in this sample.

The strong overall correlation masks measurement differences at the utility level. Value-set characteristics also differ because of the methods used to develop the value sets. Time trade-off and standard gamble differ in their approach to eliciting values, and the states selected for valuation also affect the models produced.

The strength of the measurement relationship between EQ-5D-5L, PROMIS-29, and PROPr differs depending on which validity indicator is used and whether items and dimensions or value sets are compared. This has implications for using each instrument in health assessment and subsequent decision-making.

The development of the PROMIS system and the availability of PROPr have potential implications for the international use of EQ-5D. Further work is warranted to understand the advantages and disadvantages of each instrument in different populations. EQ-5D-5L, PROMIS-29, and PROPr overlap, but they are not interchangeable.

The right choice depends on which health concepts and distinctions a study needs to capture.