- Short report
- Open access
- Published:
Mapping of the PROMIS global health measure to the PROPr in the United States
Journal of Patient-Reported Outcomes volume 8, Article number: 5 (2024)
Abstract
Background
The Patient-Reported Outcomes Measurement and Information System (PROMIS®) global health items (global-10) yield physical and mental health scale scores and the PROMIS-Preference (PROPr) scoring system estimated from PROMIS domain scores (e.g., PROMIS-29 + 2) produces a single score anchored by 0 (dead or as bad as being dead) to 1 (full health). A link between the PROMIS global-10 and the PROPr is needed.
Methods
The PROMIS-29 + 2 and the PROMIS global-10 were administered to 4102 adults in the Ipsos KnowledgePanel in 2022. The median age was 52 (range 18–94), 50% were female, 70% were non-Hispanic White, and 64% were married or living with a partner. The highest level of education completed for 26% of the sample was a high school degree or general education diploma and 44% worked full-time. We estimated correlations of the PROPr with the PROMIS global health items and the global physical and mental health scales. We examined the adjusted R2 and estimated correlations between predicted and observed PROPr scores.
Results
Product-moment correlations between the PROMIS global health items and the PROPr ranged from 0.50 to 0.63. The PROMIS global physical health and mental health scale scores correlated 0.74 and 0.60, respectively, with the PROPr. The adjusted R2 in the regression of the PROPr on the PROMIS global health items was 64%. The equated PROPr preference scores correlated (product-moment) 0.80 (n = 4043; p < 0.0001) with the observed PROPr preference scores, and the intra-class correlation (two-way random effects model) was 0.80. The normalized mean absolute error (NMAE) was 0.45 (SD = 0.43). The adjusted R2 in the OLS regression of the PROPr on the PROMIS global health scales was 59%. The equated PROPr preference scores correlated (product-moment) was 0.77 (n = 4046; p < 0.0001) with the observed PROPr preference scores, and the intra-class correlation was 0.77. The NMAE was 0.49 (SD = 0.45).
Conclusions
Regression equations provide a reasonably accurate estimate of the PROPr preference-based score from the PROMIS global health items or scales for group-level comparisons. These estimates facilitate cost-effectiveness research and meta-analyses. The estimated PROPr scores are not accurate enough for individual-level applications. Future evaluations of the prediction equations are needed.
Introduction
The Patient-Reported Outcomes Measurement Information System (PROMIS®) was developed as part of the National Institutes of Health Roadmap initiative to develop, evaluate, and standardize item banks to assess health-related quality of life across different medical conditions and in the general population [1]. The 10 PROMIS global health items (PROMIS-10) include ratings of five primary domains (physical function, fatigue, pain, emotional distress, and social health) and perceptions of general health that cut across domains. Four of the items are used to create the global physical health scale and four others to produce the global mental health scale [2]. These scales can also be estimated using two-item short forms. Internal consistency reliability coefficients for the four and two-item global physical health scales were 0.81 and 0.73, respectively, and 0.86 and 0.81 for the respective global mental health scales [3].
The PROMIS-Preference (PROPr) scoring system is based on seven PROMIS multi-item domains: physical function, pain interference, depression, fatigue, ability to participate in social roles and activities, sleep disturbance, and cognitive function. The PROMIS domain scores can be estimated from items in the domain banks, short forms (e.g., PROMIS-29 + 2), or via computer-adaptive testing [4]. All the items are administered with five polytomous response options and use a last seven-day recall period except for physical function and ability to participate in social roles and activities (which do not have an explicit recall interval).
Including a preference-based measure directly is the preferred option for obtaining a single summary score, but preference-based measures are often not administered in research studies. Many PROMIS investigators in the interest of parsimony elect to administer only the PROMIS-10 [5,6,7]. Being able to estimate the PROPr from the PROMIS global health items and scales provides an option for estimating a single summary score when a separate preference-based measure has not been administered.
Previous studies have used the PROMIS-10 to estimate the EQ-5D-3L [8–9] and the Health Utilities Index [10] but an estimate of the PROPr from the PROMIS-10 has not yet been published. This study derives regression equations to estimate the PROPr from the PROMIS-10.
Methods
We administered a general health survey in English to members of KnowledgePanel®, an online panel that relies on probability-based sampling methods for recruitment and provides a representative sample of non-institutionalized adults 18 and older residing in the U.S. [11].
The survey vendor (Ipsos) sent an email invitation to 7,224 KnowledgePanel members on September 22, 2022, and gave them 10 days to complete the general health survey. Email reminders were sent to non-responders on Day 3 of the field period. Additional reminders were sent to the remaining non-responders every 3 days for up to 10 days. Upon completion, respondents received an entry into the KnowledgePanel sweepstakes. 57% (n = 4,121) completed the survey and we excluded 19 who reported having one or two of the fake health conditions [12] included in the survey to identify careless respondents, resulting in a baseline sample of 4,102.
Measures
Demographic characteristics
We measured age in years, gender (female vs. male), race/ethnicity, and education: No high school diploma or general education diploma (GED); High school graduate (high school diploma or the equivalent GED); Some college or associate degree; Bachelor’s degree; Master’s degree or higher.
Health conditions
Thirteen health conditions were assessed by asking: “Have you ever been told by a doctor or other health professional that you had”: (1) hypertension; (2) high cholesterol; (3) heart disease; (4) angina; (5) heart attack; (6) stroke; (7) asthma; (8) cancer; (9) diabetes; (10) chronic obstructive pulmonary disease (COPD); 11) arthritis; 12) anxiety disorder; and 13) depression. In addition, the survey asked respondents if they were ever told they had “Syndomitis” (a fake condition). Further, participants were asked “Do you currently have…” 9 other conditions: (1) allergies or sinus trouble; (2) back pain; (3) sciatica; (4) neck pain; (5) trouble seeing; (6) dermatitis; (7) stomach trouble; (8) trouble hearing; and (9) trouble sleeping. Respondents were also asked if they currently had “Chekalism” (a fake condition).
PROMIS measures
The PROMIS global-10 includes the most widely used self-rated health item (“In general, would you say your health is…”; global01) and an item that provides a pure rating of physical health (global03), a rating of overall quality of life item (global02), and rating of mental health (global04). The remaining items provide global ratings of physical function (global06), fatigue (global08), pain (global07), emotional distress (global10), and social health (global05 and global09). All the items except the 0–10 rating of pain on average (global07) are administered using five-category response scales. Seven of the 10 items use a general non-specific time frame and three are prefaced with “In the past 7 days…” We scored the 10 individual PROMIS items so a higher score represents better health, and derived the global physical health (global03, global06, global07, global08) and mental health (global02, global04, global05, global10) scale scores using existing item response theory item parameters.
We used the PROPr scoring function obtained from the U.S. standard gamble valuations that yield possible scores ranging from − 0.022 to 1 [4].
Subjects
Those who completed the survey had a median age of 52 (range 18–94), 50% female, 70% non-Hispanic White, 64% were married or living with a spouse, the highest level of education completed for 26% of the sample was a high school degree or GED, and 44% were working full-time (Table 1). The most common health condition reported was allergies (45% of the sample), followed by hypertension and high cholesterol (38% each).
Analysis plan
We estimate product-moment correlations of the PROMIS global health items and physical and mental health scale scores with the PROPr. Then, we regress the PROPr on the PROMIS global health items and then on the global physical and mental health scale scores. We used linear equating to address the problem of over-prediction of low scores and under-prediction of high scores due to regression to the mean [13]. That is, we transformed predicted scores from each of the regression models linearly to have the same mean and SD as the observed PROPr preference-based scores. We recoded scores outside of the observed range to the nearest minimum or maximum observed scores. Ordinary least squares (OLS) models were evaluated in terms of adjusted R2 and estimated product-moment and intraclass correlations between the predicted and observed PROPr scores. In addition, we provide Bland-Altman plots [14] with the mean of the PROPr and the equated (predicted) PROPr preference scores on the x-axis and the difference between them (PROPr – equated PROPr) on the y-axis. The 95% upper and lower limits of agreement (bias) are estimated using: mean difference +/- 1.96*SDdifference. Scatter bias is present when the amount of disagreement between the PROPr and equated PROPr varies by the mean. We also report the normalized mean absolute error (NMAE): average deviations between observed and predicted scores divided by the standard deviation of the observed score. Lower values of the NMAE indicate better prediction. We evaluated differences in NMAE by key demographic variables: gender, age, and race/ethnicity. The magnitude of NMAE was interpreted as small or trivial for correlations of it with gender and age less than 0.243 (i.e., small correlation) and if effect size differences by race/ethnicity were less than 0.5 SD (i.e., small effect sizes or less).
Results
Correlations between PROMIS global health items and scale scores with the PROPr
Table 2 shows that the product-moment correlations between the PROMIS global health items and the PROPr ranged from 0.47 (global05) to 0.63 (global08). The items in the two-item global physical health short-form correlated 0.54 (global03) and 0.59 (global06) and the items in the global mental health short-form correlated 0.52 (global4) and 0.47 (global5) with the PROPr. The PROMIS four-item global physical health and mental health scale scores correlated 0.74 and 0.60, respectively, with the PROPr.
Predicting PROPr from PROMIS global health items
The observed PROPr mean was 0.539 (SD = 0.249, observed score range: -0.018 to 0.954). The adjusted R2 in the OLS regression of the PROPr on the PROMIS global health items was 64%. The equated PROPr scores had a mean of 0.542 and an SD of 0.239. The equated PROPr preference scores correlated (product-moment) 0.80 (n=4043; p < 0.0001) with the observed PROPr preference scores, and the intra-class correlation (two-way random effects model) between observed and equated PROPr preference scores was 0.80. The NMAE was 0.45 (SD = 0.43). Although differences between predictions and PROPr scores generally were within the 95% confidence interval, the Bland-Altman plot revealed scatter bias, with the predicted values overestimating the observed PROPr scores in the middle of the distribution (Fig. 1). The equations to predict the PROPr are shown in Table 3.
The NMAE was significantly negatively correlated with age (r = -0.03, p = 0.0334) and female gender (r = -0.04, p = 0.0396), indicating that the accuracy of prediction of PROPr from the global health items was slightly lower among younger adults and males. The Tukey-Kramer multiple range test indicated that NMAE was significantly higher for non-Hispanic Blacks (mean = 0.53) and Hispanics (mean = 0.51) than for multi-racial respondents (mean = 0.43). All these differences are small in magnitude.
Predicting PROPr from PROMIS global physical and mental health scale scores
The adjusted R2 in the OLS regression of the PROPr on the PROMIS global physical and mental health scale scores was 59%. The equated PROPr scores had a mean of 0.538 and an SD of 0.238 compared with the observed PROPr mean of 0.538 and SD of 0.249. The equated PROPr preference scores correlated (product-moment) 0.77 (n = 4046; p < 0.0001) with the observed PROPr preference scores, and the intra-class correlation (two-way random effects model) between observed and equated PROPr preference scores was 0.77. The NMAE was 0.49 (SD = 0.45). The Bland-Altman plot shows scatter bias, with the predicted values overestimating the observed PROPr scores in the middle of the distribution (Fig. 2). The equations to predict the PROPr are shown in Table 3.
The NMAE was significantly negatively correlated with age (r = -0.06, p = 0.0002) and female gender (r = -0.07, p = 0.0001), indicating that the accuracy of prediction of PROPr from the global physical and mental health scale scores was slightly lower among younger adults and males. The Tukey-Kramer multiple range test indicated that the NMAE was significantly higher for non-Hispanic Blacks (mean = 0.58) and Hispanics (mean = 0.55) than for non-Hispanic Whites (mean = 0.46) and multi-racial respondents (mean = 0.43). All these differences are small in magnitude.
Discussion
The OLS regression model indicated substantial (64% and 59%) shared variance between the PROPr and the PROMIS global health items and scale scores, respectively. The variance explained is comparable to the 65% of the variance in the EQ-5D-5L predicted by PROMIS-10 items in a previous study [8] and higher than the 40–48% of the variance shared between the PROMIS-10 and the Veterans RAND-12 physical and mental health summary scores in another study [15]. The NMAE of 0.45 (PROMIS global health items) and 0.49 (PROMIS global health scales) indicate that on average the predicted values were less than a half-standard deviation of the observed scores. Both OLS regression models were slightly less accurate in predicting PROPr scores for males, younger adults, non-Hispanic Blacks, and Hispanics.
Only two of the global health items were not significantly and uniquely related to the PROPr: the global self-rating of health (global01) and overall quality of life (global02). This is due to the previously noted [2] local dependence between global01 and the global rating of physical health (global03), and the fact that global02 is highly correlated with global ratings of mental health (global04). Both PROMIS global physical health and mental health scale scores were significantly uniquely associated with the PROPr.
While the 57% response rate exceeds the 44% average response rate found in a meta-analysis of online surveys [16], nonresponse can affect the generalizability of the results. The use of a well-known probability-based panel representative of the U.S. population [11] is a strength of the study. The unweighted sample was similar in gender and education, slightly older (52 versus 48), and had fewer Hispanics (12% versus 17%) than general population estimates from the U.S. Current Population Survey [17]. The underrepresentation of Hispanics was in part due to the limitation of the study to English-language respondents. Multivariate analyses have yielded similar results for weighted and unweighted data [18].
Methods other than OLS have been used such as Tobit and Censored Least Absolute Deviation, mixture models, and adjusted limited dependent variable mixture models to map scores from one measure to another [19]. For example, beta-binomial regression was found to perform better than OLS for several fit criteria (root mean squared error, mean absolute error, normal root mean squared error, normalized mean absolute error, and correlation between predicted and observed values) in a prior study, but the fit was similar to two decimal places (e.g., root mean squared error of 0.1218 versus 0.1191 for OLS and beta-binomial, respectively) [20]. Moreover, we used linear equating to address the problem of OLS models leading to over-predicting at the lower end and underpredicting at the upper end. Finally, the estimated scores should be limited to group-level applications because of the lack of accuracy of individual-level estimates.
Future studies are needed to further examine the accuracy of the prediction equations, but the OLS regression equations derived here can facilitate cost-effectiveness research and meta-analyses. These equations make it possible to provide a reasonable estimate of a bottom-line preference-based summary score when only the PROMIS global health items have been administered. Further exploration of the less accurate predictions for younger age, males, and among Black and Hispanic respondents is needed.
Data Availability
The data set analyzed for this study is not publicly available yet because the project is still in progress, but the data are available from the first author on reasonable request.
Abbreviations
- PROMIS® :
-
Patient-Reported Outcomes Measurement Information System
- PROPr:
-
PROMIS-Preference
- OLS:
-
Ordinary least squares
- NMAE:
-
Normalized mean absolute error
References
Cella D, Choi SW, Condon DM, Schalet B, Hays RD, Rothrock NE et al (2019) PROMIS® adult health profiles: efficient short-form measures of seven health domains. Value in Health 22(5):537–544
Hays RD, Bjorner J, Revicki DA, Spritzer KL, Cella D (2009) Development of physical and mental health summary scores from the patient-reported outcomes Measurement Information System (PROMIS) global items. Qual Life Res 18:873–880
Hays RD, Schalet BD, Spritzer KL, Cella D (2017) Two-item PROMIS global physical and mental health scales. JPRO 1:2
Dewitt B, Feeny D, Fischoff B, Cella D, Hays RD, Hess R et al (2018) Estimation of a preference-based summary score for the patient-reported outcomes Measurement Information System: the PROMIS®-Preference (PROPr) Scoring System. Med Decis Making 38(6):683–698
Vartanian K, Fish D, Kenton N, Gronowski B, Wright B, Robicsek A (2023) Integrating patient-reported physical, mental, and social impacts to classify long COVID experiences. Sci Rep 13(1):16288
Woznica DN, Milligan M, Krymis H, Peters KC, O’Connor MI, Grant RA (2023) Telemedical Interdisciplinary Care Team Evaluation and treatment of people with Low Back Pain: a retrospective observational study. Arch Rehabil Res Clin Transl 5(3):100269
Williams GR, Fowler M, Giri S, Dai C, Harmon C, Al-Obaidi M et al (2023) Association of unmet basic resource needs with frailty and quality of life among older adults with cancer-results from the CARE registry. Cancer Med 12(12):13846–13855
Revicki DA, Kawata AK, Harnam N, Chen WH, Hays RD, Cella D (2009) Predicting EuroQol (EQ-5D) scores from the patient-reported outcomes measurement information system (PROMIS) global items and domain item banks in a United States sample. Qual Life Res 18(6):783–791
Thompson NR, Lapin BR, Katzan IL (2017) Mapping PROMIS global health items to EuroQol (EQ-5D) utility scores using linear and equipercentile equating. PharmacoEconomics 35(11):1167–1176
Hays RD, Revicki DA, Feeny D, Fayers P, Spritzer KL, Cella D (2016) Using Linear Equating to Map PROMIS(®) Global Health Items and the PROMIS-29 V2.0 Profile Measure to the Health Utilities Index Mark 3. Pharmacoeconomics 34(10):1015-22
Ipsos (2019) Be Sure with KnowledgePanel New York: Ipsos Available via https://www.ipsos.com/en-us/solution/knowledgepanel. Accessed 1 August 2023
Hays RD, Qureshi N, Herman PM, Rodriguez A, Kapteyn A, Edelen MO (2023) Effects of excluding those who report having Syndomitis or Chekalism on data quality: longitudinal health survey of a sample from Amazon’s mechanical Turk. J Med Internet Res 25:e46421
Fayers PM, Hays RD (2014) Should linking replace regression when mapping from profile to preference-based measures? Value in Health 17:261–265
Bland JM, Altman DG (1986) Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1(8476):307–310
Schalet BD, Rothrock NE, Hays RD, Kazis LE, Cook KF, Rutsohn JP, Cella D (2015) Linking physical and mental health summary scores from the veterans RAND 12-Item Health Survey (VR-12) to the PROMIS(®) Global Health Scale. J Gen Intern Med 30(10):1524–1530
Wu M-J, Zhao K, Fils-Aime F (2022) Response rates of online surveys in published research: a metaanalysis. Computers in Human Behavior Reports 7:100206
Current Population Survey Annual Social and Economic Supplement (2022) https://www.census.gov/data/datasets/time-series/demo/cps/cps-asec.2022.html#list-tab-165711867 Accessed 15 July 2023
Cornesse C, Blom AG, Dutwin D, Krosnick JA, De Leeuw ED, Legleye S et al (2020) A review of conceptual approaches and empirical evidence on probability and nonprobability sample survey research. J Surv Stat Methodol 8(1):4–36
Mukuria C, Rowen D, Harnan S, Rawdin A, Wong R, Ara R et al (2019) An updated systematic review of studies mapping (or cross-walking) measures of health-related quality of life to generic preference-based measures to generate utility values. Appl Health Econ Health Policy 17(3):295–313
Lamu AN, Olsen JA (2018) Testing alternative regression models to predict utilities: mapping the QLQ-C30 onto the EQ-5D-5L and the SF-6D. Qual Life Res 27(11):2823–2839
Acknowledgements
We thank the study participants, colleagues who worked on the project, and Victor Gonzalez for his administrative assistance.
Funding
This work was supported by the National Center for Complementary and Integrative Health (1R01AT010402-01A1).
Author information
Authors and Affiliations
Contributions
All authors participated in the design of the study. Hays conceived the manuscript, wrote the first draft, and revised it based on the coauthor’s input. All others provided comments on the drafts.
Corresponding author
Ethics declarations
Ethics approval and consent to participate
This study was approved by the RAND Human Subjects Protection Committee (2019-0651-AM02). All study participants provided consent.
Consent for publication
Not applicable.
Competing interests
None.
Additional information
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Hays, R.D., Herman, P.M., Qureshi, N. et al. Mapping of the PROMIS global health measure to the PROPr in the United States. J Patient Rep Outcomes 8, 5 (2024). https://doi.org/10.1186/s41687-023-00677-6
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s41687-023-00677-6