Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice J Biopharm Stat . Author manuscript; available in PMC: 2026 Apr 15. Published in final edited form as: J Biopharm Stat. 2023 May 14;35(5):888–901. doi: 10.1080/10543406.2023.2210684 Search in PMC Search in PubMed View in NLM Catalog Add to search Applying latent profile analysis to identify adolescents and young adults with chronic conditions at risk for poor health-related quality of life Suwei Wanga Suwei Wanga a Center for Health Measurement, Department of Population Health Sciences, Duke University School of Medicine, Durham, North Carolina, USA Find articles by Suwei Wanga a , Cara J Arizmendi Cara J Arizmendi a Center for Health Measurement, Department of Population Health Sciences, Duke University School of Medicine, Durham, North Carolina, USA Find articles by Cara J Arizmendi a , Dandan Chen Dandan Chen a Center for Health Measurement, Department of Population Health Sciences, Duke University School of Medicine, Durham, North Carolina, USA Find articles by Dandan Chen a , Li Lin Li Lin a Center for Health Measurement, Department of Population Health Sciences, Duke University School of Medicine, Durham, North Carolina, USA Find articles by Li Lin a , Dan V Blalock Dan V Blalock b Department of Psychiatry and Behavioral Sciences, Duke University School of Medicine, Durham, North Carolina, USA c Health Services Research & Development, Durham Veterans Affairs Medical Center, Durham, North Carolina, USA Find articles by Dan V Blalock b, c , I-Chan Huang I-Chan Huang d Department of Epidemiology and Cancer Control, St. Jude Children’s Research Hospital, Memphis, Tennessee, USA Find articles by I-Chan Huang d , David Thissen David Thissen e Department of Psychology and Neuroscience, University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, USA Find articles by David Thissen e , Darren A DeWalt Darren A DeWalt f Department of Medicine, University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, USA Find articles by Darren A DeWalt f , Wei Pan Wei Pan a Center for Health Measurement, Department of Population Health Sciences, Duke University School of Medicine, Durham, North Carolina, USA g Division of Clinical Health Systems and Analytics in the School of Nursing, Duke University, Durham, North Carolina, USA Find articles by Wei Pan a, g , Bryce B Reeve Bryce B Reeve a Center for Health Measurement, Department of Population Health Sciences, Duke University School of Medicine, Durham, North Carolina, USA h Department of Pediatrics, Duke University School of Medicine, Durham, North Carolina, USA Find articles by Bryce B Reeve a, h Author information Article notes Copyright and License information a Center for Health Measurement, Department of Population Health Sciences, Duke University School of Medicine, Durham, North Carolina, USA b Department of Psychiatry and Behavioral Sciences, Duke University School of Medicine, Durham, North Carolina, USA c Health Services Research & Development, Durham Veterans Affairs Medical Center, Durham, North Carolina, USA d Department of Epidemiology and Cancer Control, St. Jude Children’s Research Hospital, Memphis, Tennessee, USA e Department of Psychology and Neuroscience, University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, USA f Department of Medicine, University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, USA g Division of Clinical Health Systems and Analytics in the School of Nursing, Duke University, Durham, North Carolina, USA h Department of Pediatrics, Duke University School of Medicine, Durham, North Carolina, USA Issue date 2025 Aug. PMC Copyright notice PMCID: PMC13078171 NIHMSID: NIHMS2151316 PMID: 37183393 The publisher's version of this article is available at J Biopharm Stat Abstract The impact of chronic diseases on health-related quality of life (HRQOL) in adolescents and young adults (AYAs) is understudied. Latent profile analysis (LPA) can identify profiles of AYAs based on their HRQOL scores reflecting physical, mental, and social well-being. This paper will (1) demonstrate how to use LPA to identify profiles of AYAs based on their scores on multiple HRQOL indicators; (2) explore associations of demographic and clinical factors with LPA-identified HRQOL profiles of AYAs; and (3) provide guidance on the selection of adult or pediatric versions of Patient-Reported Outcomes Measurement Information System ® (PROMIS ® ) in AYAs. A total of 872 AYAs with chronic conditions completed the adult and pediatric versions of PROMIS measures of anger, anxiety, depression, fatigue, pain interference, social health, and physical function. The optimal number of LPA profiles was determined by model fit statistics and clinical interpretability. Multinomial regression models examined clinical and demographic factors associated with profile membership. As a result of the LPA, AYAs were categorized into 3 profiles: Minimal, Moderate, and Severe HRQOL Impact profiles. Comparing LPA results using either the pediatric or adult PROMIS T-scores found approximately 71% of patients were placed in the same HRQOL profiles. AYAs who were female, had hypertension, mental health conditions, chronic pain, and those on medication were more likely to be placed in the Severe HRQOL Impact Profile. Our findings may facilitate clinicians to screen AYAs who may have low HRQOL due to diseases or treatments with the identified risk factors without implementing the HRQOL assessment. Keywords: LPA, AYA, HRQOL, chronic conditions, PROMIS 1. Introduction Health-related quality of life (HRQOL) is an important outcome to examine the impact of diseases and/or treatments on the lives of patients. It is a multi-dimensional construct that includes aspects of physical, mental, and social well-being ( CDC 2021 ; Cella and Tulsky 1990 ; Revicki et al. 2014 ; Siegrist and Junge 1989 ; Sprangers 2002 ). Ideally, HRQOL is self-reported by the patient using a patient-reported outcome (PRO) measure such as the NIH’s Patient-Reported Outcomes Measurement Information SystemR (PROMISR), which includes both pediatric and adult versions of measures to capture HRQOL in participants with a broad range of diseases or health conditions ( Cella et al. 2007 ). After patients completing the HRQOL assessments including multiple domains, traditional approaches to assessing HRQOL in a population often look at one outcome at a time (e.g., pain intensity), and repeat the analysis for each HRQOL domain ( Colver et al. 2015 ; Devinsky et al. 1999 ; Hallstrand et al. 2003 ; Jennes-Coussens et al. 2006 ; Kamp-Becker et al. 2010 ; Parkinson et al. 2015 ; Pemberger et al. 2005 ; Uzark et al. 2008 ; Varni et al. 2007 ). This granular approach doesn’t capture the totality of HRQOL impact in a single analysis. Additionally, it is difficult to interpret the disease and treatment impacts when they are inconsistent across multiple domains. Although factor analysis as a traditional technique assesses multiple HRQOL domains simultaneously, it is a variable-centered approach to suggest a structure that explains the relations among variables but not to identify groups of patients ( Laursen and Hoff 2006 ; Stewart 1981 ). Latent profile analysis (LPA) is a person-centered approach ( Laursen and Hoff 2006 ) and it assumes that the population is heterogeneous that often reflects reality. LPA is based on a statistical model which analyzes interval- or ratio-level continuous scores from multiple measures to place patients into profiles, where patients in the same profile have similar HRQOL impacts ( Barsevick and Aktas 2013 ; Buckner et al. 2014 ; Davis et al. 2003 ; Hinds et al. 2021 ; Miaskowski et al. 2006 ; Wang and Lanza 2010 ; Wang and Wang 2012 ). The number of profiles is based on model fit, sample size in each subgroup, and clinical interpretability. K -means and hierarchical clustering are related but distinct approaches that can be utilized as a person-centered technique to divide patients into sub-groups based on the similarity of scores from multiple measures. However, these approaches are not based on a statistical model and do not provide the probability that a given person has a more severe HRQOL impact. Additionally, determining the final number of subgroups is subjective in cluster analysis ( Kim et al. 2013 ). LPA allows researchers to examine the probability of each patient being placed in different profiles and the uncertainty of the classification. For example, if the probabilities for a given patient to be placed in three HRQOL impact profiles are 0.90, 0.04, and 0.06, researchers could be confident about the LPA grouping results. Other advantages of LPA over factor analysis, k -means, and/or hierarchical clustering are that variables can be continuous, ordinal, counts or any combination of these and that demographics and other covariates can be used for profile description ( Spurk et al. 2020 ). Latent Classification Analysis (LCA) or LPA can be used depending on whether the indicators are categorical or continuous ( Collins and Lanza 2009 ). AYAs with chronic conditions are uniquely affected as they struggle both with transitioning to adulthood and dealing with the long-term effects of their diseases and treatments ( Reeve et al. 2016 ). Moreover, the ability to identify AYAs at risk for poor HRQOL will enable healthcare providers and services to direct resources to help this vulnerable population. The pediatric versions of PROMIS measures are applicable for 8-17 years old and the adult versions of PROMIS measures are applicable for 18 years or older. The AYA population spans this age threshold and includes adolescents as young as 13 years and continues to young adulthood (Aubin et al. 2011; Bonnie et al. 2014 ; NCI 2022 ; Closing the gap 2011 ). AYA researchers may struggle with which version to use, and clinicians may have doubts about whether it is appropriate to give adult version measures to pediatric patients who transition to adult care settings. A previous study by Blalock et al. (2020) showed that there was no differential item functioning (DIF) between adolescents and young adults on using either the pediatric or adult PROMIS measures ( Blalock et al. 2020 ). We provide guidance on the use of PROMIS measures in AYA populations. This present study also explores whether different results are obtained from the LPA using either version in an AYA population who experience chronic conditions. LPA is used over LCA because the PROMIS T-scores are continuous indicators of HRQOL. The goal of this study is to demonstrate how to use LPA to cluster AYAs into statistically distinct and clinically meaningful profiles based on their HRQOL scores and how to identify demographic and clinical factors associated with HRQOL profile membership. 2. Materials and methods This study is a secondary analysis of data collected as part of NIH-funded research to provide the linking of domain scores between pediatric and adult versions of the Patient-Reported Outcome Measurement Information SystemR (PROMISR) measures in AYAs ( Reeve et al. 2016 ; Tulsky et al. 2019 ). The current study was ruled exempt from IRB review by the Duke University Health System. This study included 874 AYAs aged 14-20 years with “special healthcare needs”, that is AYAs who have or are at an increased risk of a chronic physical, developmental, behavioral, or emotional condition and who also require health care or related services beyond what patients generally require ( Reeve et al. 2016 ). Details about participant recruitment are provided elsewhere ( Blalock et al. 2020 ; Reeve et al. 2016 ; Tulsky et al. 2019 ; Wang et al. 2022 ). Each participant completed a demographic form including age, gender, race and ethnicity, a checklist of chronic conditions, and reported blindness, deafness, and whether assistance was required to get around. Each AYA completed the adult version of PROMIS measures including the PROMIS SF v1.0 - Physical Function 10a, PROMIS SF v1.0-Pain Interference 8a, PROMIS SF v1.0-Fatigue 8a, PROMIS SF v2.0-Social Health: Emotional Support, PROMIS SF v1.0-Depression 8b, PROMIS SF v1.0-Anxiety 8b, PROMIS SF v1.0-Anger 8a. Each participant also completed the pediatric version of PROMIS measures including the PROMIS Ped SF v1.0-Mobility 8a, PROMIS Ped SF v1.0-Upper Extremity 8a, PROMIS Ped SF v1.0-Pain Interference 8a, PROMIS Ped SF v1.0-Fatigue 10a, PROMIS Ped SF v1.0-Peer Relations 8a, PROMIS Ped SF v1.0-Depressive Symptoms 8a, PROMIS Ped SF v1.0-Anxiety 8a, and PROMIS Ped SF v1.0-Anger 6a ( Reeve et al. 2016 ). Almost all measures (except the adult PROMIS measure of Physical Function) use a 7-day reference period and include 5-ordinal level response options ( Reeve et al. 2016 ). Scores on the PROMIS measures are on a T-score metric, with a mean of 50 and a standard deviation of 10 in the original calibration samples. Higher PROMIS T-scores for symptom domains represent worse symptom burden, and higher PROMIS T-scores for functional domains represent better functioning. To compare results across the pediatric and adult versions of PROMIS, we transformed the scores into a common metric. Pediatric PROMIS profile mean scores in the anger, anxiety, depression, fatigue, pain interference, mobility, and upper extremity functions were converted into adult PROMIS profile mean scores using the linking coefficients provided by Reeve et al. (2016) and Tulsky et al. (2019) ( Reeve et al. 2016 ; Tulsky et al. 2019 ). PROMIS Pediatric peer relation profile mean score was not transformed to adult PROMIS social health-emotional support because they measure different facets of social health and do not have linking coefficients. 3. Statistical analysis A commonly used naive three-step LPA approach ( Bakk and Kuha 2021 ) was carried out on the pediatric and adult versions of the PROMIS measures T-scores independently. The first step of the naive three-step LPA was to determine the optimal number of HRQOL profiles based on three criteria: 1) statistical model fit, 2) clinical interpretability, and 3) profile sample size. The statistical evaluation was conducted by generating a hierarchically-nested series of LPA models with an increasing number of latent profiles and iteratively comparing the fit of each successive model k with the previous ( k -1) model using Akaike, Bayesian, and sample-size adjusted Bayesian information criterion indices (AIC, BIC, and SABIC, respectively), the Lo-Mendell-Rubin likelihood ratio (LMR LR) test, and the Vuong-Lo-Mendell-Rubin likelihood ratio (VLMR LR) test ( Lo 2001 ; McCutcheon 1987 ; Oberski 2016 ; Spurk et al. 2020 ). Smaller AIC, BIC, and SBIC values indicate better model fit. A significant p-value from the LMR LR or ALMR LR test indicates the model with the larger number of profiles ( k ) should be preferred over the model with the smaller number of profiles ( k -1 classes) ( Nylund et al. 2007 ). We also examined the entropy for each profile solution, which evaluates models with respect to confidence with which individuals have been classified as belonging to one group or another ( Oberski 2016 ; Spurk et al. 2020 ). Entropy values over 0.8 indicate a good separation of the latent profiles, and values approaching 1 indicate a clear delineation of profiles ( Celeux and Soromenho 1996 ). For clinical interpretability, our multi-disciplinary research team reviewed each of the profile solutions. Keeping statistical rigor and parsimony in mind, our team considered how the identification of each profile may help stakeholders (e.g., clinicians, patients, and caregivers) understand the unique HRQOL impact on AYAs to make informed-decisions in research and clinical care. In addition, the team wanted to see meaningful differences in most PROMIS T-scores between the profiles and used a half standard deviation (5 points in T-scores) to represent a meaningful difference which is deemed a large effect size ( Gignac and Szodorai 2016 ). In terms of sample size of each profile, Spurk et al. suggested each profile should include>1% of the total sample size ( Spurk et al. 2020 ). However, our recommendation is that the sample size criteria should be context specific based on the total sample size and intended use of the findings. In our study sample ( N = 872), 9 patients represent 1% that could be in a profile. Our team was not confident that a 9-patient profile would represent a distinctive HRQOL impact profile in the population because it may be an artifact of the measure or study design. In terms of intended use, if these profiles were to be used subsequently to generate prediction models to triage who may get more resources for clinical care, then larger sample sizes may be necessary as opposed to another research study that was using the LPA results for more descriptive purposes. For this study, we specified that no profile should have less than 10% of the sample to feel confident about interpreting the results. But we suggest other studies select the threshold taking into consideration the available sample size and preferred confidence in the generalization of findings. When meeting all the above criteria is difficult in determining the number of profiles, we suggest prioritizing by model fit statistics and clinical interpretability. In the second step, we classified AYAs into their most likely HRQOL profiles using the highest estimated posterior probabilities of being categorized into each profile. This resulted in a categorical variable where each AYA was assigned to one HRQOL profile. Mplus version 8.7 was used for LPA modeling. The “mclust” (2022) and “tidyLPA” package in R were also available for the LPA ( Wardenaar 2021 ). The process of identifying HRQOL profiles and assigning AYAs to one of the profiles was performed independently for the pediatric and adult PROMIS measures. We assessed the agreement in profile classifications (4-profile solution) based on the adult versus the pediatric PROMIS by examining the proportion of agreement and kappa. Criteria for kappa agreement were poor (κ<0.20), fair (0.20≤κ<0.40), moderate (0.40≤κ<0.60), good (0.60≤κ<0.80), and very good (κ≥0.80) ( Landis and Koch 1977 ). In the third step, we used a multinomial logit model including demographic and clinical factors (or covariates, e.g., hypertension) to look for factors associated with the most-likely profile membership. We chose to use multinomial over multi-ordinal logistic regression because the multinomial model does not impose the proportional odds assumption. We also tested the model prediction accuracy in Huberty’s I index, with 0.35 as a threshold for good effect ( Granado 2015 ; Huberty and Lowman 2000 ). In the naive three-step LPA approach described above, the uncertainty in HRQOL Impact profiles was neglected when the HRQOL Impact profiles were treated as fixed in the third step in the logit modeling. The two-step LPA approach ( Bakk and Kuha 2021 ) and the bias-corrected three-step LPA approach ( Bakk et al. 2016 ; Bolck et al. 2004 ; Vermunt 2010 ) are alternative approaches to address the bias of the naive three-step approach for handling uncertainty. In the two step approach, the measurement model is estimated first, and without assigning patients to any latent profile, the parameters of the measurement model were held fixed when the structure model is estimated next ( Bakk and Kuha 2018 , 2021 ). As a sensitivity analysis, we performed LPA comparing the naive three-step, bias-corrected three-step, and two-step LPA approaches with the same AYA data and yielded the same HRQOL membership profiles, suggesting the effect of different approaches was minimal ( Wang et al. 2022 ). The naive three-step LPA method can be performed using mainstream software packages such as R and Mplus, while the two-step or bias-corrected three-step approach is not implemented by default in standard software and requires manual syntax inputs from researchers ( Bakk and Kuha 2021 ). Therefore, we implemented the naive three-step approach for the current study, balancing the above considerations. 3.1. Pediatric vs. adult T-score results comparison We further explored the consistency of results between the pediatric and adult PROMIS measures by looking at factors associated with inconsistent or consistent results. If patients were grouped into the same HRQOL profiles by both the pediatric and adult measures, they were placed in a Consistent group. If patients were grouped into a poorer HRQOL profile using pediatric scores vs. adult scores (reference), patients were further placed into the “Over-estimate” group. We used LPA profiles from adult T-score as a reference because a total of 459 (53%) AYAs aged 18 and above in this study, and adult version PROMIS T-scores were calibrated using the US general population. If patients were grouped into a higher HRQOL profile using pediatric vs. adult T-scores, they were placed into “Underestimate” group. Using the “consistent”, “Over-estimate” and “Under-estimate” categories for each participant as an outcome, we built multinomial logistic regression models including patient demographics and chronic disease conditions, on medication status, and their parents’ education, relationship status, and language spoke at home as covariates. 4. Results The initial sample size was 874 AYA participants. One participant was excluded due to missing PROMIS scores. Another participant was ineligible due to their age (<14 years). A total of 872 AYA patients were included in the analysis ( Table 1 ). Participant demographics and clinical characteristics are presented in Table 1 and described in detail elsewhere ( Wang et al. 2022 ). Table 1. Demographics and clinical characteristics of AYA participants ( N = 872). N (%) * AYA Characteristics Gender: male 406 (46.6%) Age (years): Mean (SD) 17.4 (1.92) Race and ethnicity Non-Hispanic White 299 (34.3%) Non-Hispanic Asian 61 (7.0%) Non-Hispanic Black 137 (15.7%) Non-Hispanic Other 42 (4.8%) Hispanic 333 (38.2%) Language spoken at home: English 768 (88.1%) On medications 671 (77.1%) AYA’s Health Condition ADHD or ADD 244 (28.0%) Mental health condition 199 (22.8%) Hypertension 198 (22.7%) Asthma 197 (22.6%) Self-reported chronic pain 190 (21.8%) Allergies 178 (20.4%) Overweight 151 (17.3%) Diabetes 88 (10.1%) Born prematurely 37 (4.2%) Intestinal disease 36 (4.1%) Thyroid disease 32 (3.7%) Epilepsy or other seizure disorders 32 (3.7%) Kidney disease 31 (3.6%) Rheumatic disease 25 (2.9%) Cancer 25 (2.9%) Deaf or hard of hearing 17 (1.9%) Congenital heart disease 16 (1.8%) Sickle cell disease 15 (1.7%) Requires assistance to get around 14 (1.6%) Blind 12 (1.4%) Cerebral palsy 10 (1.1%) None of the conditions above 65 (7.5%) Parent Characteristics Parent relationship status Married or living together 560 (64.4%) Single or living as single † 309 (35.6%) Parent highest education level Less than high school 83 (9.5%) High school degree 185 (21.3%) College and above 581 (66.8%) Unknown 21 (2.4%) Open in a new tab * % based on available data only. N = 1 participant had missing information in gender, N = 2 participants had miss-ing information on “On medication”, N = 3 participants had missing information on parent relationship status, N = 2 participants had missing information on parent highest education level. Parent relationship status: † Other = never together, or separated, or divorced, or widowed. N, number; SD, standard deviation; ADHD, attention deficit hyperactivity disorder; ADD: attention deficit disorder. As noted previously, we determined the number of acceptable HRQOL profiles for both pediatric and adult versions of PROMIS measures based on three criteria. In terms of model fit indices (see Table 2 ), while the AIC, BIC, and sample-adjusted BIC were lowest for the 5-profile solution, the two likelihood ratio tests (VLMR LRT, LMR LRT) favored the 4-profile solution. Differences in entropy values from the 2-, 3-, or 4-profile solution were negligible. The criteria of sample size in each profile (>10% of AYAs) was acceptable for the 2-, 3-, 4- or 5-profile solutions for the Pediatric version of PROMIS. Using the adult PROMIS measures, sample sizes for each profile was acceptable for the 2-, 3- or 4-profile solutions, but not for the 5-profile solution (one profile had 9.6%, N = 84). Our multidisciplinary team considered 3 and 4 HRQOL profiles to have a similar level of clinical meaningfulness. We settled on the 3-profile solution for both the pediatric and adult PROMIS measures as a matter of parsimony for making comparisons across versions. We recognize that other researchers may determine that the 4-profile solution is optimal for their intended research study goal. Table 2. Comparison of different LPA model fit indices using information criterion indices and likelihood ratio tests. PROMIS version LPA Model AIC BIC SABIC VLMR LRT (p-value) LMR LRT (p-value) Entropy Pediatric 2-profile 51377.4 51496.7 51417.3 2955.1 (<0.001) 2907.4 (<0.001) 0.92 3-profile 50400.0 50562.2 50454.3 995.398 (<0.001) 979.3 (<0.001) 0.91 4-profile 49966.1 50171.3 50034.7 451.9 (0.032) 444.6 (0.034) 0.90 5-profile 49658.5 49906.6 49741.4 325.6 (0.0564) 320.4 (0.0585) 0.88 Adult 2-profile 43560.7 43665.6 43595.8 2510.9 (<0.001) 2465.4 (<0.001) 0.90 3-profile 42811.4 42954.6 42859.3 765.2 (0.001) 751.4(<0.001) 0.87 4-profile 42372.0 42553.3 42432.6 455.5 (<0.001) 447.2 (<0.001) 0.89 5-profile 42157.1 42376.5 42230.4 230.9 (0.21) 226.7 (0.22) 0.88 Open in a new tab PROMIS, Patient-Reported Outcome Measurement Information System; LPA, latent profile analysis; AIC, Akaike information criterion; BIC, Bayesian information criterion; SABIC, sample size adjusted BIC; VLMR LRT, Vuong-Lo-Mendell-Rubin likelihood ratio test; LMR LRT, Lo-Mendell-Rubin adjusted likelihood ratio test. Both the VLMR-LR and the LMR-LR tests compare one nested LPA model to another model with one additional profile with statistically significant improvement (p < .05) suggesting the model with more profiles reflects a better fit. Entropy is a measure of uncertainty in the posterior classifications of the model with higher entropy values reflecting less uncertainty. Participants were categorized into three distinctive profiles (Minimal, Moderate, and Severe HRQOL Impact Profiles) by using either pediatric PROMIS untransformed scores ( Figure 1 ) or adult PROMIS scores ( Figure 2 ). When using pediatric PROMIS T-scores (untransformed to adult scores), the group mean T-score differences across the HRQOL domains was 12.1 points (range 1.6–18.4) between Moderate and Minimal HRQOL Impact profiles, and 8.4 points (range 2.9-22.5) between Severe and Moderate HRQOL Impact profiles. When using adult PROMIS T-scores, the group mean T-score differences across the HRQOL domains was 10.6 points (range 6.6-14.4) between Moderate and Minimal HRQOL Impact profiles, and 9.4 points (range 2.1-11.2) between Severe and Moderate HRQOL Impact profiles. The group mean T-score differences were all greater than 5 points, a half standard deviation which can be viewed as a threshold of meaningful difference ( Gignac and Szodorai 2016 ; Norman et al. 2003 ). Figure 3 presents the 3-profile solutions from analyzing the PROMIS Pediatric T-scores (transformed for comparison ( Reeve et al. 2016 ; Tulsky et al. 2019 )) and adult PROMIS T-scores. Overall, the T-scores in each domain were similar with the largest difference observed in the upper extremity function domain in the Minimal HRQOL Impact Profile (56.6 in adult T-score vs. 50.5 in pediatric transformed T-score) in Figure 3 . Figure 1. Open in a new tab Mean Pediatric version of PROMIS HRQOL domain T-scores (95% confidence interval) of three latent profiles estimated by LPA among 872 patients. PROMIS scores are on a T-score metric, normed in a reference group to have a mean of 50 and a standard deviation of 10. PROMIS, Patient-Reported Outcomes Measurement Information System; LPA, latent profile analysis. For symptom domains including anger, anxiety, depression, fatigue, and pain interference, higher scores indicate worse symptom burden; for function domains of social health, upper extremity function, and mobility, higher scores indicate better functioning. The dotted vertical line separates symptom severity scores (on the left, clear background color) and functioning scores (on the right, pale yellow background color). The y-axis scales for symptom severity and function are reversed for interpretation. N = 254, 300, and 318 in Minimal, Moderate, and Severe HRQOL Impact profiles, respectively. Figure 2. Open in a new tab Mean Adult version of PROMIS HRQOL domain T-scores (95% confidence interval) of three latent profiles estimated by LPA among 872 AYAs. PROMIS scores are on a T-score metric, normed in a reference group to have a mean of 50 and a standard deviation of 10. PROMIS, Patient-Reported Outcomes Measurement Information System; LPA, latent profile analysis. For symptom domains including anger, anxiety, depression, fatigue, and pain interference, higher scores indicate worse symptom burden; for function domains of social health and physical functioning, higher scores indicate better functioning. The dotted vertical line separates symptom severity scores (on the left, clear background color) and functioning scores (on the right, pale yellow background color). The y-axis scales for symptom severity and function are reversed for interpretation. N = 179, 337, and 356 in Minimal, Moderate, and Severe HRQOL Impact profiles, respectively. Figure 3. Open in a new tab Mean Adult version of PROMIS HRQOL domain T-scores and pediatric transformed to adult version T-scores of three latent profiles estimated by LPA among 872 AYAs. PROMIS scores are on a T-score metric, normed in a reference group to have a mean of 50 and a standard deviation of 10. PROMIS, Patient-Reported Outcomes Measurement Information System; LPA, latent profile analysis. For symptom domains including anger, anxiety, depression, fatigue, and pain interference, higher scores indicate worse symptom burden; for function domains of social health and physical functioning, higher scores indicate better functioning. The dotted vertical line separates symptom severity scores (on the left, clear background color) and functioning scores (on the right, pale yellow background color). The y-axis scales for symptom severity and function are reversed for interpretation. N = 179, 337, and 356 in Minimal, Moderate, and Severe HRQOL Impact profiles using adult versions of PROMIS T-scores, respectively. A total of 620 out of 872 (71%) participants were assigned to the same HRQOL Impact profiles by using Adult or Pediatric version PROMIS measure T-scores ( Table 3 , Figure 4 ). Only one patient was assigned to the Severe HRQOL Impact profile with pediatric T-scores but Minimal profile with adult T-scores, and only two patients were in the vice versa situation. Unweighted Kappa was 0.561, suggesting moderate agreement ( McHugh 2012 ). With the HRQOL membership profile as the outcome in multinomial logistic regressions, we found that AYAs who were female, had hypertension, mental health conditions, self-reported chronic pain, and those on medication were statistically significantly more likely to be included in the Severe HRQOL Impact Profile compared to the Minimal HRQOL Impact Profile using either pediatric or adult PROMIS T-scores. AYAs who had thyroid disease or spoke English at home had a higher likelihood of being placed in the Severe HRQOL Impact Profile compared to the Minimal profile using pediatric T-scores. See Table 4 for a complete tabulation of these results. The variables included in the pediatric multinomial regression model correctly predicted the LPA HRQOL Impact Profile membership in 523 (60.5%) AYAs with a Huberty’s I index of 0.41, higher than the 0.35 threshold of good effect. Similarly, the adult model correctly predicted the membership in 488 (56.5%) AYAs with a Huberty’s I index of 0.37, also above the 0.35 threshold of good effect. Table 3. Number of patients assigned to minimal, moderate, and severe HRQOL impact profiles by using the pediatric and adult PROMIS measure scores Pediatric Adult Minimal Moderate Severe Total Minimal 165 13 1 179 Moderate 87 194 56 337 Severe 2 93 261 356 Total 254 300 318 872 Open in a new tab Figure 4. Open in a new tab HRQOL profile membership agreement between profiles assigned by the LPA using Adult PROMIS scores (on left) and using Pediatric PROMIS scores (on right). N = 620 out of864 (71%) patients were assigned to the same profile by using the two sets of PROMIS scores. Blue color represents Minimal HRQOL Impact Profile, orange color represents Moderate HRQOL Impact Profile, and red color represents Severe HRQOL Impact Profile. The flow from left to right represents how AYAs in each HRQOL Impact profile from LPA using adult version T-scores change to the HRQOL Impact profiles from LPA using pediatric version T-scores. Table 4. Characteristics statistically associated with HRQOL Impact profiles using multinomial logistic regression model ( n = 872) Predictors Comparator (HRQOL Impact) Reference group (HRQOL Impact) Pediatric OR (95% CI) Adult OR (95% CI) Gender: male * Moderate Minimal 0.51 (0.34-0.76) 0.74 (0.49-1.13) Severe Minimal 0.53 (0.35-0.83) 0.5 (0.32-0.79) Severe Moderate 1.01 (0.68-1.51) 0.64 (0.45-0.91) Hypertension Moderate Minimal 3.34 (1.84-6.07) 2.67 (1.39-5.11) Severe Minimal 7.45 (4.14-13.40) 4.79 (2.51-9.13) Severe Moderate 2.31 (1.51-3.53) 1.82 (1.23-2.68) Mental health condition Moderate Minimal 3.39 (2.07-5.54) 3.76 (2.01-7.02) Severe Minimal 2.5 (1.45-4.30) 5.04 (2.65-9.59) Severe Moderate 0.79 (0.52-1.21) 1.44 (0.98-2.12) Kidney disease Moderate Minimal 0.43 (0.09-2.11) 1.46 (0.26-8.21) Severe Minimal 1.75 (0.43-7.21) 2.42 (0.44-13.49) Severe Moderate 5.15 (1.43-18.54) 1.76 (0.65-4.79) Chronic pain Moderate Minimal 2.85 (1.47-5.53) 2.79 (1.38-5.65) Severe Minimal 12.13 (6.46-22.77) 7.38 (3.69-14.74) Severe Moderate 4.54 (2.86-7.22) 2.69 (1.78-4.05) Asthma Moderate Minimal 1.39 (0.88-2.20) 1.02 (0.62-1.66) Severe Minimal 0.67 (0.38-1.15) 0.75 (0.43-1.28) Severe Moderate 0.51 (0.31-0.83) 0.75 (0.49-1.16) Thyroid disease Moderate Minimal 2.57 (0.64-10.39) 2.09 (0.42-10.50) Severe Minimal 5.33 (1.30-21.87) 4.21 (0.86-20.66) Severe Moderate 2 (0.73-5.52) 1.93 (0.75-4.93) Intestinal disease Moderate Minimal 2.61 (0.87-7.85) 3.77 (0.82-17.36) Severe Minimal 0.86 (0.24-3.12) 2.1 (0.42-10.57) Severe Moderate 0.3 (0.12-0.78) 0.58 (0.25-1.32) On medication Moderate Minimal 1.44 (0.92-2.25) 1.61 (1.01-2.57) Severe Minimal 4.46 (2.47-8.05) 3.97 (2.27-6.93) Severe Moderate 3.02 (1.69-5.38) 2.52 (1.54-4.10) Language at home: non-English ** Moderate Minimal 0.74 (0.41-1.36) 1.47 (0.77-2.78) Severe Minimal 0.36 (0.16-0.79) 0.93 (0.44-1.97) Severe Moderate 0.48 (0.22-1.04) 0.65 (0.35-1.23) Open in a new tab The full regression model adjusted for AYA patient’s gender, age (year), race/ethnicity, different health conditions (hypertension, ADHD or ADD, mental health condition, kidney disease, self-reported chronic pain, asthma, thyroid disease, overweight, rheumatic disease, diabetes, intestinal disease, congenital heart disease, epilepsy or other seizure disorders, allergy, on medication or not), parent relationship status (married or living together, other), parent highest education level (less than high school, high school, college and above, unknown), and language spoken at home (English, non-English). Only variables statistically associated (p < .05) with the outcome response (HRQOL Impact profile) are presented in this table. *Gender reference is female; **language at home: reference is English. HRQOL, health-related quality of life; OR, odds ratio; CI, confidence interval. Results interpretation example, the odds of being in the severe vs. minimal HRQOL impact profile in AYAs with hypertension is 7.45 times the odds of being in the severe vs. minimal HRQOL impact profiles in AYAs without hypertension. The sample size was 864 (out of 872) due to missing values for some covariates. In logit models, when the outcome was categorized as “consistent”, “over-estimation”, “under-estimation” in HRQOL Impact Profiles by pediatric vs. adult T-scores, patients who had hypertension were associated with higher odds of being placed in a more severe HRQOL membership profile by using pediatric T-scores instead of using adult T-scores; patients who were older, had mental health conditions, epilepsy, non-English language spoken at home, and without chronic pain were associated with higher odds of being placed into a better/less severe HRQOL membership profile by using pediatric T-scores instead of using adult T-scores ( Table 5 ). Table 5. Characteristics statistically associated with the estimation consistency of HRQOL Impact profiles using adult vs. pediatric t-scores using multinomial logistic regression Model ( n = 872) Predictors Response vs. consistent estimation OR (95%CI) Age (years) Over-estimation 1.01 (0.88-1.17) Under-estimation 1.14 (1.03-1.26) Race and Ethnicity * Non-Hispanic Asian Over-estimation 1.56 (0.65-3.76) Under-estimation 0.72 (0.33-1.56) Non-Hispanic Black Over-estimation 0.64 (0.27-1.52) Under-estimation 0.51 (0.29-0.91) Non-Hispanic other Over-estimation 0.79 (0.21-2.91) Under-estimation 0.32 (0.12-0.88) Hispanic Over-estimation 0.77 (0.39-1.54) Under-estimation 0.48 (0.30-0.77) Hypertension Over-estimation 2.54 (1.41-4.60) Under-estimation 0.73 (0.45-1.16) Mental health condition Over-estimation 0.59 (0.28-1.28) Under-estimation 1.57 (1.05-2.36) Self-reported chronic pain Over-estimation 1.52 (0.84-2.76) Under-estimation 0.42 (0.25-0.72) Epilepsy or other seizure disorders Over-estimation 3.07 (0.82-11.50) Under-estimation 2.69 (1.14-6.31) Language at home: non-English * Over-estimation 0.27 (0.05-1.31) Under-estimation 2.23 (1.24-4.01) Open in a new tab The full regression model adjusted for AYA patient’s gender, age (year), race/ethnicity, different health conditions (hypertension, ADHD or ADD, mental health condition, kidney disease, self-reported chronic pain, asthma, thyroid disease, overweight, rheumatic disease, diabetes, intestinal disease, congenital heart disease, epilepsy or other seizure disorders, allergy, on medication or not), parent relationship status (married or living together, other), parent highest education level (less than high school, high school, college and above, unknown), and language spoken at home (English, non-English). Only variables statistically associated (p < .05) with the outcome response (HRQOL Impact profile) are presented in this table. **Language at home: reference is English. HRQOL, health-related quality of life; OR, odds ratio; CI, confidence interval. Consistent estimation is AYA is placed in the same HRQOL impact profile by using pediatric and adult T-scores. Over-estimation is defined as AYA is placed in a more severe pediatric HRQOL impact profile than the adult profile; under-estimation is defined as AYAs is placed in a less severe pediatric HRQOL impact profile than the adult profile. Results interpretation example, the odds of being over-estimated in HRQOL impact profile among AYAs who had hypertension is 2.54 times the odds of being over-estimated in HRQOL impact profile among AYAs who did not have hypertension. The sample size was 864 (out of 872) due to missing values for some covariates. 5. Discussion In this article we demonstrated how LPA can be used to create a profile of HRQOL by taking a composite of multiple measure scores in a group of AYAs with a broad range of chronic conditions. We demonstrated how LPA model statistics together with clinical interpretability can be used to determine the optimal number of subgroups. LPA is a useful methodology to look at multiple HRQOL domains and identify vulnerable populations at risk for poor outcomes. Severe HRQOL Impact Profile membership can be used as a clinically meaningful threshold for a vulnerable patient population instead of percentile cutoff values which are arbitrary and may vary across samples. AYAs who were in the Severe HRQOL Impact profiles may be provided with additional care or monitored more closely than those in moderate or mild HRQOL impact profiles. Potential applications of this type of work are to develop risk prediction models to estimate patients’ HRQOL, using the significantly associated demographic and clinical factors to triage patients into mild, moderate, or severely impacted profiles. This prediction model could be implemented into an electronic-based medical record system to flag at-risk patients. With such an application, the models could include more clinical factors than were available in this secondary analysis data set. Pediatric and adult versions of PROMIS T-scores produced similar results for the AYA sample. The pediatric versions included two physical functioning domains (upper extremity function and mobility) while the adult versions only include one physical functioning domain, which could contribute to the difference in HRQOL impact profiles. We found consistent demographic factors and chronic condition prevalence associated with more severe HRQOL impact profile membership in AYAs by using either version of T-scores. Compatible with previous results showing that there was no differential item functioning (DIF) in any item in either pediatric or adult versions of PROMIS measures ( Blalock et al. 2020 ), the results here showed similar results in HRQOL profile membership by using multi-domain scores in either pediatric or adult versions of PROMIS measures. Either version may be used by clinicians to assess HRQOL among AYAs who were 14-20 years old. Thus, the selection of the right version of PROMIS measures is at the discretion of the investigator based on the age range and characteristics of the population (e.g., developmental delays). One can also consider using both pediatric and adult PROMIS measures for adolescents and young adults, respectively, using the established linking coefficients to connect results ( Reeve et al. 2016 ; Tulsky et al. 2019 ). The data collected here were cross-sectional. Future studies may be more interested in assessing HRQOL impact among AYAs over time, in particular as the adolescent with a chronic condition transitions from pediatric to adult care, or transition to living independently outside their family’s house. An extension of the LPA for longitudinal data (i.e., latent transition analysis (LTA) ( Lanza et al. 2010 )) can be used for this purpose. For example, one can examine if an adolescent transitions from a mild HRQOL impact group at one time to a moderate HRQOL impact group at a later time. Every transition from one profile to another over time can be modeled with LTA. Additionally, multinomial logistic regression can be used to examine which factors associated with the profile transitions. LPA analysis was performed with commercial Mplus software. As a sensitivity analysis, LPA was also performed using the “tidyLPA” package in R, producing equivalent results using pediatric T-scores (99.9% AYA patient membership agreement) or adult T-scores (98.6% AYA patient membership agreement) compared to the Mplus results. Thus, researchers can also consider using free packages such as “tidyLPA” and “mclust” in R for LPA analysis. Limitations in the study included study sample restricted to AYAs who can read and respond in English and low prevalence (<5%) in some chronic conditions which may lead to underestimation of their association with HRQOL impact profile results in regression models. 6. Conclusion This study demonstrates how LPA can be used to place a sample into subgroups (profiles) based on their scores across multiple HRQOL indicators. We also demonstrated how one can test for demographic and chronic condition factors associated with profile membership that could be used in future applications for developing risk prediction models. Important for AYA research, we demonstrated that one can get comparable results using either pediatric or adult versions of PROMIS measures. Funding The work was supported by the AstraZeneca [MaRS Post-Doc Education Agreement]; United States Department of Veterans Affairs Health Services Research and Development (HSR&D) Service [Career Development Award 19-035 (IK2HX003085-01A]; Takeda Pharmaceuticals U.S.A. [MaRS Post-Doc Education Agreement] Footnotes Ethics approval This study was performed in line with the principles of the Declaration of Helsinki. The current study was ruled exempt from IRB review by the Duke University Health System. Disclosure statement No potential conflict of interest was reported by the author(s). Data availability statement The authors have no relevant financial or non-financial interests to disclose. References Aubin S. 2011, What should the age range be for AYA oncology? Mary Ann Liebert, Inc 140 Huguenot Street, 3rd Floor, New Rochelle, NY: 10801 USA. p. 3–10. [ Google Scholar ] Bakk Z, and Kuha J. 2018. Two-step estimation of models between latent classes and external variables. Psychometrika 83 (4):871–892. doi: 10.1007/s11336-017-9592-7. [ DOI ] [ PubMed ] [ Google Scholar ] Bakk Z, and Kuha J. 2021. Relating latent class membership to external variables: An overview. The British Journal of Mathematical and Statistical Psychology 74 (2):340–362. doi: 10.1111/bmsp.12227. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Bakk Z, Oberski DL, and Vermunt JK. 2016. Relating latent class membership to continuous distal outcomes: Improving the LTB approach and a modified three-step implementation structural equation modeling. Structural Equation Modeling: A Multidisciplinary Journal 23 (2):278–289. doi: 10.1080/10705511.2015.1049698. [ DOI ] [ Google Scholar ] Barsevick AM, and Aktas A. 2013. Cancer symptom cluster research: New perspectives and tools. Current Opinion in Supportive and Palliative Care 7 (1):36–37. doi: 10.1097/SPC.0b013e32835defac. [ DOI ] [ PubMed ] [ Google Scholar ] Blalock DV, Lin L, Wang M, Thissen D, DeWalt DA, Huang IC, Reeve BB. 2020. Analysis of differential item functioning in PROMISR pediatric and adult measures between adolescents and young adults with special health care needs. Psychological Test and Assessment Modeling 62 (4):417–428. https://www.psychologieaktuell.com/fileadmin/Redaktion/Journale/ptam-2020-4/PTAM_4-2020_21900493_ebook_blalock-1.pdf [ Google Scholar ] Bolck A, Croon M, and Hagenaars J. 2004. Estimating latent structure models with categorical variables: One-step versus three-step estimators. Political Analysis 12 (1):3–27. doi: 10.1093/pan/mph001. [ DOI ] [ Google Scholar ] Bonnie R, Stroud C, Breiner H 2014. Committee on improving the health, Safety, and well-being of young adults. Investing in the health and well being of young adults. [ Google Scholar ] Buckner TW, Wang J, DeWalt DA, Jacobs S, Reeve BB, and Hinds PS. 2014. Patterns of symptoms and functional impairments in children with cancer. Pediatric Blood & Cancer 61 (7):1282–1288. doi: 10.1002/pbc.25029. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] CDC. Health-Related Quality of Life (HRQOL). 2021. [cited 2022 March 14]; Available from: https://www.cdc.gov/hrqol/index.htm . Celeux G, and Soromenho G. 1996. An entropy criterion for assessing the number of clusters in a mixture model. Journal of Classification 13 (2):195–212. doi: 10.1007/BF01246098. [ DOI ] [ Google Scholar ] Cella D, Riley W, Stone A, Rothrock N, Reeve B, Yount S, Amtmann D, Bode R, Buysse D, Choi S, Cook K 2007. The Patient-Reported Outcomes Measurement Information System (PROMIS). Medical Care 45 (5):S3–11. [ Google Scholar ] Cella DF, and Tulsky DS. 1990. Measuring quality of life today: Methodological aspects. Oncology (Williston Park) 4 (5):29–38. discussion 69. [ Google Scholar ] Closing the gap: Research and cancer care imperatives for adolescents and young adults with cancer. 2011. Collins LM and Lanza ST 2009. Latent class and latent transition analysis: With applications in the social, behavioral, and health sciences. Vol. 718. John Wiley & Sons. [ Google Scholar ] Colver A, Rapp M, Eisemann N, Ehlinger V, Thyen U, Dickinson HO, Parkes J, Parkinson K, Nystrand M, Fauconnier J, et al. 2015. Self-reported quality of life of adolescents with cerebral palsy: A cross-sectional and longitudinal analysis. The Lancet. 385(9969):705–716. doi: 10.1016/S0140-6736(14)61229-0. [ DOI ] [ Google Scholar ] Davis PJ, Reeves JL, Graff-Radford SB, Hastie BA, and Naliboff BD. 2003. Multidimensional subgroups in migraine: Differential treatment outcome to a pain medicine program. Pain Medicine 4 (3):215–222. doi: 10.1046/j.1526-4637.2003.03027.x. [ DOI ] [ PubMed ] [ Google Scholar ] Devinsky O, Westbrook L, Cramer J, Glassman M, Perrine K, and Camfield C. 1999. Risk Factors for Poor Health-Related Quality of Life in Adolescents with Epilepsy. Epilepsia 40 (12):1715–1720. doi: 10.1111/j.1528-1157.1999.tb01588.x. [ DOI ] [ PubMed ] [ Google Scholar ] Gignac GE, and Szodorai ET. 2016. Effect size guidelines for individual differences researchers. Personality and Individual Differences 102:74–78. doi: 10.1016/j.paid.2016.06.069. [ DOI ] [ Google Scholar ] Granado EA 2015. Comparing three effect sizes for latent class analysis (Doctoral dissertation). TX: University of North Texas. [ Google Scholar ] Hallstrand TS, Curtis JR, Aitken ML, and Sullivan SD. 2003. Quality of life in adolescents with mild asthma. Pediatric Pulmonology 36 (6):536–543. doi: 10.1002/ppul.10395. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Hinds PS, Weaver MS, Withycombe JS, Baker JN, Jacobs SS, Mack JW, Maurer SH, McFatrich M, Pinheiro LC, Reeve BB, et al. 2021. Subjective Toxicity Profiles of Children in Treatment for Cancer: A New Guide to Supportive Care? Journal of Pain and Symptom Management. 61(6):1188–1195 e2. doi: 10.1016/j.jpainsymman.2020.10.017. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Huberty CJ, and Lowman LL. 2000. Group Overlap as a Basis for Effect Size. Educational and Psychological Measurement 60 (4):543–563. doi: 10.1177/0013164400604004. [ DOI ] [ Google Scholar ] Jennes-Coussens M, Magill-Evans J, and Koning C. 2006. The quality of life of young men with Asperger syndrome: A brief report. Autism 10 (4):403–414. doi: 10.1177/1362361306064432. [ DOI ] [ PubMed ] [ Google Scholar ] Kamp-Becker I, Schroder J, Remschmidt H, and Bachmann CJ. 2010. Health-related quality of life in adolescents and young adults with high functioning autism-spectrum disorder. Psycho-Social Medicine 7. doi: 10.3205/psm000065. [ DOI ] [ Google Scholar ] Kim H-J, Abraham I, and Malone PS. 2013. Analytical methods and issues for symptom cluster research in oncology. Current Opinion in Supportive and Palliative Care 7 (1):45–53. doi: 10.1097/SPC.0b013e32835bf28b. [ DOI ] [ PubMed ] [ Google Scholar ] Landis JR, and Koch GG. 1977. The measurement of observer agreement for categorical data. Biometrics 33 (1):159–174. doi: 10.2307/2529310. [ DOI ] [ PubMed ] [ Google Scholar ] Lanza ST, Patrick ME, and Maggs JL. 2010. Latent transition analysis: Benefits of a latent variable approach to modeling transitions in substance use. Journal of Drug Issues 40 (1):93–120. doi: 10.1177/002204261004000106. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Laursen B, and Hoff E. 2006. Person-centered and variable-centered approaches to longitudinal data. Merrill-Palmer Quarterly (1982) 52 (3):377–389. doi: 10.1353/mpq.2006.0029. [ DOI ] [ Google Scholar ] Lo Y. 2001. Testing the number of components in a normal mixture. Biometrika 88 (3):767–778. doi: 10.1093/biomet/88.3.767. [ DOI ] [ Google Scholar ] McCutcheon AL 1987. Latent class analysis. No. 64. Sage. [ Google Scholar ] McHugh ML 2012. Interrater reliability: The kappa statistic. Biochemia Medica 22 (3):276–282. doi: 10.11613/BM.2012.031. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Mclust: Gaussian mixture modelling for model-based clustering, classification, and density estimation. [cited 2022 July 29]; Available from: https://cran.r-project.org/web/packages/mclust/index.html . Miaskowski C. 2006. Online Exclusive-Subgroups of Patients with Cancer with Different Symptom Experiences and Quality-of-Life Outcomes: A Cluster Analysis. In Oncology Nursing Forum-Oncology Nursing Society, [Pittsburgh PA, etc.], Oncology Nursing Society. https://www.ons.org/onf/33/5/subgroups-patients-cancer-different-symptomexperiences-andquality-life-outcomes-cluster [ Google Scholar ] NCI. Adolescents and Young Adults with Cancer. [cited 2022 September 13]; Available from: https://www.cancer.gov/types/aya#1 . Norman GR, Sloan JA, and Wyrwich KW. 2003. Interpretation of changes in health-related quality of life: The remarkable universality of half a standard deviation. Medical Care 41 (5):582–592. doi: 10.1097/01.MLR.0000062554.74615.4C. [ DOI ] [ PubMed ] [ Google Scholar ] Nylund KL, Asparouhov T, and Muthen BO. 2007. Deciding on the Number of Classes in Latent Class Analysis and Growth Mixture Modeling: A Monte Carlo Simulation Study. Structural Equation Modeling. Structural Equation Modeling: A Multidisciplinary Journal 14 (4):535–569. doi: 10.1080/10705510701575396. [ DOI ] [ Google Scholar ] Oberski D. 2016. Mixture models: Latent profile and latent class analysis. In Modern statistical methods for HCI, pp. 275–287. Springer. https://link.springer.com/book/10.1007/978-3-319-26633-6 . [ Google Scholar ] Parkinson KN, Adamson AJ, Basterfield L, Reilly JK, Le Couteur A, and Reilly JJ. 2015. Influence of adiposity on health-related quality of life in the Gateshead Millennium Study cohort: Longitudinal study at 12 years. Archives of Disease in Childhood 100 (8):779–783. doi: 10.1136/archdischild-2014-307498. [ DOI ] [ PubMed ] [ Google Scholar ] Pemberger S, Jagsch R, Frey E, Felder-Puig R, Gadner H, Kryspin-Exner I, and Topf R. 2005. Quality of life in long-term childhood cancer survivors and the relation of late effects and subjective well-being. Supportive Care in Cancer 13 (1):49–56. doi: 10.1007/s00520-004-0724-0. [ DOI ] [ PubMed ] [ Google Scholar ] Reeve BB, Thissen D, DeWalt DA, Huang I-C, Liu Y, Magnus B, Quinn H, Gross HE, Kisala PA, Ni P, et al. 2016. Linkage between the PROMISR pediatric and adult emotional distress measures. Quality of Life Research. 25 (4):823–833. doi: 10.1007/s11136-015-1143-z. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Revicki Dennis A., Kleinman Leah & Cella David. 2014. A history of health-related quality of life outcomes in psychiatry, Dialogues in Clinical Neuroscience 16 (2):127–135. doi: 10.31887/DCNS.2014.16.2/drevicki. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Siegrist J, and Junge A. 1989. Conceptual and methodological problems in research on the quality of life in clinical medicine. Social Science & Medicine 29 (3):463–468. doi: 10.1016/0277-9536(89)90295-5. [ DOI ] [ PubMed ] [ Google Scholar ] Sprangers MAG 2002. Quality-of-Life Assessment in Oncology. Acta oncologica 41 (3):229–237. doi: 10.1080/02841860260088764. [ DOI ] [ PubMed ] [ Google Scholar ] Spurk D, Hirschi A, Wang M, Valero D, and Kauffeld S. 2020. Latent profile analysis: A review and “how to” guide of its application within vocational behavior research. Journal of Vocational Behavior 120:103445. doi: 10.1016/j.jvb.2020.103445. [ DOI ] [ Google Scholar ] Stewart DW 1981. The application and misapplication of factor analysis in marketing research. Journal of Marketing Research 18 (1):51–62. doi: 10.1177/002224378101800105. [ DOI ] [ Google Scholar ] Tulsky DS, Kisala PA, Boulton AJ, Jette AM, Thissen D, Ni P, DeWalt DA, Huang I-C, Liu Y, Mulcahey MJ, et al. 2019. Determining a transitional scoring link between PROMISR pediatric and adult physical health measures. Quality of Life Research. 28(5):1217–1229. doi: 10.1007/s11136-018-2073-3. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Uzark K, Jones K, Slusher J, Limbers CA, Burwinkle TM, and Varni JW. 2008. Quality of Life in Children with Heart Disease as Perceived by Children and Parents. Pediatrics 121 (5):e1060–1067. doi: 10.1542/peds.2006-3778. [ DOI ] [ PubMed ] [ Google Scholar ] Varni JW, Limbers CA, and Burwinkle TM. 2007. Impaired health-related quality of life in children and adolescents with chronic conditions: A comparative analysis of 10 disease clusters and 33 disease categories/severities utilizing the PedsQL ™ 4.0 Generic Core Scales. Health and Quality of Life Outcomes 5 (1):43. doi: 10.1186/1477-7525-5-43. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Vermunt JK 2010. Latent class modeling with covariates: Two improved three-step approaches. Political Analysis 18 (4):450–469. doi: 10.1093/pan/mpq025. [ DOI ] [ Google Scholar ] Wang S, Arizmendi C, Blalock D, Chen D, Lin L, Thissen D, Huang I, Dewalt D, Reeve B 2022. Health-Related Quality of Life Profiles in Adolescents and Young Adults with Chronic Conditions (under review). Quality of Life Research. https://www.springer.com/journal/11136 . [ Google Scholar ] Wang J, and Lanza ST. 2010. Preface of methods and applications of mixture models, special journal issue of advances and applications of statistical sciences. Advances and Applications in Statistical Sciences 3:1–6. [ Google Scholar ] Wang J, and Wang X. 2012. Structural equation modeling with Mplus: Methods and applications. Wiley. [ Google Scholar ] Wardenaar KJ 2021. Latent Profile Analysis in R: A tutorial and comparison to Mplus. doi: 10.31234/osf.io/wzftr. [ DOI ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Data Availability Statement The authors have no relevant financial or non-financial interests to disclose. ACTIONS View on publisher site PDF (1.0 MB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top