PROMIS-29

PROMIS-29 in Trial Outcomes, Utilities and Mapping

Because its short forms are drawn from calibrated PROMIS item banks, the PROMIS-29 reports each domain as a T-score whose mean of 50 and standard deviation of 10 come, for many domains, from a sample matched to the US general population, so a trial result reads as a distance from that mean. Its own scores carry no preference weights, which means a study that collects only this profile reaches QALYs through the PROMIS-Preference (PROPr) scoring system or through a mapping to EQ-5D-5L. Klapproth and colleagues reported in 2022 that, in general population samples from the UK, France and Germany, mean PROPr scores were 0.32 to 0.35 lower than the EQ-5D-5L values the same people reported directly. The choice of utility source can therefore move an ICER even when the underlying health-related quality of life responses are identical.

Domains, items and recall periods of the PROMIS-29

Cella and colleagues described the development of three PROMIS adult profiles, ranging from 29 to 57 items. Each profile covers the same seven domains with short forms of four, six or eight items per domain, and each adds an average pain intensity item on a 0 to 10 numeric rating scale. The PROMIS-29 uses the four-item short forms, so seven domains of four items plus the single pain rating give 29 items.

Cella and colleagues note that the physical function and social role items ask about current abilities and not about experiences over the previous 7 days. The pain intensity item runs from 0 for no pain to 10 for the worst pain imaginable. HealthMeasures, the distribution site for PROMIS, lists the profile as version 2.1.

Item banks, item response theory and the T-score metric

The short forms are drawn from calibrated item banks. Cella and colleagues reported in 2010 that the first wave of PROMIS, run from 2005 to 2008, tested 14 item pools in the US general population and clinical groups and used item response theory, through a graded response model, to calibrate 11 item banks on a sample of 21,133 people. Because every item in a bank sits on one latent scale, a four-item short form, a longer short form and a computer adaptive test can all report scores on the same metric.

In the 2019 profile paper, Cella and colleagues reported that, with few exceptions, the domain short forms were highly reliable across at least three standard deviations, or 30 T-score units, and correlated strongly with full bank scores. Construct validity was supported against ratings of general health and quality of life.

Domain scores are reported as T-scores. HealthMeasures states that 50 is the mean of a relevant reference population and 10 is that population's standard deviation, and that for many PROMIS measures the reference population was a sample matched to the 2000 US general census.

$$T = 50 + 10z$$

where $T$ is the reported domain score and $z$ is the respondent's distance from the reference population mean, in standard deviation units, estimated from the item responses. A higher T-score always means more of the concept being measured. For physical function and social roles a higher score is better health, while for anxiety, depression, fatigue, sleep disturbance and pain interference a higher score is worse health. The pain intensity item stays on its own 0 to 10 scale.

Physical and mental health summary scores

Profile scores answer clinical questions domain by domain, but a trial often needs one or two summary outcomes. Hays and colleagues derived physical health and mental health summary scores from PROMIS-29 version 2.0 data and reported reliabilities of 0.98 and 0.97 respectively. They described the evidence as preliminary support for reliability and validity and called for further evaluation.

These summary scores sit closer in spirit to the component summaries of the SF-12 than to a utility. They are standardised descriptions of health, built without any preference weights, so they cannot be multiplied by time to give QALYs.

Converting PROMIS-29 profile scores to PROPr utilities

PROPr is the preference-based scoring system for PROMIS. Dewitt and colleagues elicited preferences for PROMIS health states with the standard gamble in an online survey of a nationally representative US sample of 983 people. They estimated a single-attribute scoring function for each of seven domains and combined them in a multiplicative multi-attribute utility function. The seven PROPr domains are cognitive function abilities, depression, fatigue, pain interference, physical function, sleep disturbance and ability to participate in social roles and activities.

The PROMIS-29 and PROPr overlap only in part. Dewitt, Jalal and Hanmer set out that the profile and PROPr share six domains, that cognitive function abilities is unique to PROPr, and that anxiety and pain intensity are unique to the profile. Anxiety and pain intensity responses therefore do not enter a PROPr utility at all. The PROPr scoring algorithm runs from minus 0.022 to 1, but Dewitt, Jalal and Hanmer give the theoretical range of PROPr scored from the PROMIS-29+2 profile as minus 0.022 to 0.954, since a value of 1 can be reached only with other PROMIS measures.

Three practical routes fill the cognition gap. HealthMeasures lists a PROMIS-29+2 profile, version 2.1, made up of the PROMIS-29 and two cognitive function abilities items, created to support PROPr scoring. Where only the PROMIS-29 was collected, Dewitt, Jalal and Hanmer used data from 3,982 members of US online panels with scores on all nine PROMIS domains to compare five ways of predicting the missing cognition score. A Bayesian neural network had the lowest root mean squared error, but linear regression performed almost as well and is easier to use. HealthMeasures also notes that the seven PROPr domains can be measured with computer adaptive tests or short forms instead of a profile.

Once each respondent has a PROPr value, the usual cost-utility analysis steps follow. Utilities at each trial visit are combined with time, by the area under the curve, to give QALYs per patient, and incremental QALYs feed the cost per QALY.

Mapping PROMIS-29 responses to EQ-5D-5L

Where an HTA body's reference case asks for EQ-5D values, analysts holding only PROMIS-29 data turn to a mapping function estimated in a separate mapping study. Klapproth and colleagues used general population samples from the UK (1,509 people), France (1,501) and Germany (1,502) and applied stepwise regression with backward selection to all seven PROMIS-29 domains to predict EQ-5D-5L crosswalk values for each country. Normalised root mean squared errors were 0.076 in the UK, 0.075 in France and 0.079 in Germany, and adding higher-order terms reduced overprediction of low utilities.

Aghdaee and colleagues surveyed 3,013 adults from the Australian general population who completed both instruments. They compared direct and indirect methods, including linear regression, Tobit, beta regression mixtures, generalised ordered logit and the adjusted limited dependent variable mixture model, and preferred a direct adjusted limited dependent variable mixture model. Their stated motive was that the PROMIS-29 is gaining popularity with funders seeking value-based care but is limited in its ability to estimate utilities.

A mapped value inherits the country value set that the mapping study used, so an algorithm built on crosswalk values or on an Australian set does not give the UK EQ-5D-5L value set without further work. Predicted values also carry error that a probabilistic analysis should reflect.

PROPr and EQ-5D-5L values in the same people

PROPr values and directly observed EQ-5D-5L values do not agree on level. Hanmer and colleagues tested PROPr in two US online samples of 983 and 3,000 people and found age and gender stratified mean PROPr scores lower than EQ-5D and Health Utilities Index scores, with fewer respondents at scores corresponding to perfect health. Klapproth and colleagues, using the same UK, French and German samples as their mapping work and predicting cognition by the recommended linear regression, reported mean PROPr scores of 0.48, 0.53 and 0.48 against mean EQ-5D-5L index values of 0.82, 0.85 and 0.83 from the EQ-5D-5L questionnaire completed by the same respondents.

A lower level does not by itself say whether QALY gains will be larger or smaller, because a QALY gain depends on the difference between arms. The gap does show that PROPr values and EQ-5D-5L values are not interchangeable health state utility values and should not be mixed within one model.

Illustrative worked example with PROPr and mapped utilities

All numbers in this example are illustrative. A trial collects the PROMIS-29+2 in 200 adults with chronic pain. At baseline a patient has a pain interference T-score of 65.

$$z = \frac{65-50}{10} = 1.5$$

where 65 is the patient's T-score, 50 is the reference mean and 10 is the reference standard deviation. The patient reports pain interference 1.5 standard deviations above the reference population mean, which on this domain means worse health. Patients starting around this level of pain interference are the trial population behind the utility gains in the steps that follow.

Step 1. Suppose the new treatment raises mean PROPr utility by 0.06 relative to usual care, and the gain lasts 2 years without discounting. The QALY gain is 0.06 × 2 = 0.12.

Step 2. The incremental cost is GBP 2,400 per patient.

$$\text{ICER} = \frac{2400}{0.06 \times 2} = 20000$$

where 2,400 is the incremental cost in GBP, 0.06 is the mean utility gain and 2 is the duration in years. The ICER is GBP 20,000 per QALY.

Step 3. Now suppose the same PROMIS-29 responses are mapped to EQ-5D-5L and give a mean utility gain of 0.04. The QALY gain is 0.04 × 2 = 0.08.

$$\text{ICER} = \frac{2400}{0.04 \times 2} = 30000$$

where 0.04 is the mapped mean utility gain and the other symbols are as before. The ICER rises to GBP 30,000 per QALY. At an illustrative threshold of GBP 25,000 per QALY the treatment looks cost-effective with PROPr and not cost-effective with mapped EQ-5D-5L values, although both rest on the same patient responses. The direction of the difference in a real trial cannot be assumed from this example, so the utility source belongs in the base case justification and in a scenario analysis.

How the PROMIS-29 differs from nearby measures

The PROMIS-29 is a patient-reported outcome measure, a generic instrument and a profile instrument. It is not a preference-based measure, since its own scores carry no preference weights, and PROPr is a separate scoring system applied to PROMIS domain scores. The distinction matters when a study reports that it "used the PROMIS-29 to calculate QALYs", because the reader needs to know whether PROPr, a mapping or some other method produced the utilities.

Compared with the SF-12 and SF-36, the PROMIS-29 gives fixed-length short forms linked to item banks, so a domain can be lengthened or replaced with a computer adaptive test without changing the score metric. Compared with EQ-5D-5L, it gives more detail on anxiety, fatigue, sleep and social participation, while EQ-5D-5L is scored directly with country value sets.

Limitations and common misreadings

A T-score of 50 is the mean of the centring sample, which for many measures was matched to the 2000 US census, so 50 is not a threshold for good health. Higher scores mean worse health on five of the seven domains, which is easy to misread when domains are plotted together.

PROPr values rest on US preferences elicited by standard gamble, and a PROPr value derived from the PROMIS-29 alone includes a predicted cognition score with its own error. Anxiety and pain intensity information collected by the PROMIS-29 is lost in the conversion. Mapped EQ-5D-5L values carry prediction error and the preferences of whichever country's value set the mapping study used. The physical and mental health summary scores look like single indices but are not utilities.

Sources

• Aghdaee M, Gu Y, Sinha K, Parkinson B, Sharma R, Cutler H. Mapping the Patient-Reported Outcomes Measurement Information System (PROMIS-29) to EQ-5D-5L. PharmacoEconomics. 2023;41(2):187-198. https://doi.org/10.1007/s40273-022-01157-3
• Cella D, et al. The Patient-Reported Outcomes Measurement Information System (PROMIS) developed and tested its first wave of adult self-reported health outcome item banks: 2005-2008. Journal of Clinical Epidemiology. 2010;63(11):1179-1194. https://doi.org/10.1016/j.jclinepi.2010.04.011
• Cella D, et al. PROMIS adult health profiles: efficient short-form measures of seven health domains. Value in Health. 2019;22(5):537-544. https://doi.org/10.1016/j.jval.2019.02.004
• Dewitt B, Feeny D, Fischhoff B, et al. Estimation of a preference-based summary score for the Patient-Reported Outcomes Measurement Information System: the PROMIS-Preference (PROPr) scoring system. Medical Decision Making. 2018;38(6):683-698. https://doi.org/10.1177/0272989X18776637
• Dewitt B, Jalal H, Hanmer J. Computing PROPr utility scores for PROMIS profile instruments. Value in Health. 2020;23(3):370-378. https://doi.org/10.1016/j.jval.2019.09.2752
• Hanmer J, Dewitt B, et al. Cross-sectional validation of the PROMIS-Preference scoring system. PLoS One. 2018;13(7):e0201093. https://doi.org/10.1371/journal.pone.0201093
• Hays RD, Spritzer KL, Schalet BD, Cella D. PROMIS-29 v2.0 profile physical and mental health summary scores. Quality of Life Research. 2018;27(7):1885-1891. https://doi.org/10.1007/s11136-018-1842-3
• HealthMeasures. PROMIS adult domains, list of adult measures (PROMIS-29+2 Profile v2.1 entry). Northwestern University; accessed 17 October 2026. https://www.healthmeasures.net/explore-measurement-systems/promis/intro-to-promis/list-of-adult-measures
• HealthMeasures. Interpret scores, PROMIS, and PROMIS reference populations. Northwestern University; accessed 17 October 2026. https://www.healthmeasures.net/score-and-interpret/interpret-scores/promis/reference-populations/279-reference-populations-promis
• Klapproth CP, et al. Predicting EQ-5D-5L crosswalk from the PROMIS-29 profile for the United Kingdom, France, and Germany. Health and Quality of Life Outcomes. 2020;18:389. https://doi.org/10.1186/s12955-020-01629-0
• Klapproth CP, Sidey-Gibbons CJ, Valderas JM, Rose M, Fischer F. Comparison of the PROMIS Preference Score (PROPr) and EQ-5D-5L index value in general population samples in the United Kingdom, France, and Germany. Value in Health. 2022;25(5):824-834. https://doi.org/10.1016/j.jval.2021.10.012