Assessing the Structural Validity and Measurement Invariance of the Whole Person Health Index

Abstract

Background

The emergence of “whole person health” presents valuable opportunities for understanding and treating individuals as their complete selves. At the same time, holistic approaches to health introduce measurement challenges, highlighting the need for expansive yet manageable instruments to capture a broader concept of health. The Whole Person Health Index (WPHI) is a nine-item question set that asks respondents to rate their overall general health as well as their quality of life, social and family connections, diet, physical activity, stress, sleep, existential wellbeing, and ability to manage their health. It has been suggested for use in clinical settings and population surveys, and it has shown good promise in an initial evaluation study.

Objective

The present study offers an exploratory look at the psychometric properties of the WPHI.

Methods

The WPHI items were included on the 11th round of the National Center for Health Statistics Research and Development Survey (RANDS 11), conducted January–March 2025, which used samples of adults from commercially available survey panels. Descriptive and psychometric analyses were applied to explore the factor structure and functioning of the WPHI overall and across different population subgroups.

Results

The WPHI demonstrated strong internal consistency and generally favorable measurement properties. Exploratory and confirmatory factor and bifactor analyses with split-half samples provided evidence that responses are largely driven by a dominant general factor, although model fit met some but not all conventional benchmarks for a strict single-factor model. Differential item functioning was detected for self-rated health by age, but not for other items or subgroups.

Conclusion

Overall, the WPHI’s subjective ratings appear to reflect a common underlying construct, supporting its utility as a broad measure of health and well-being consistent with the whole person health concept. Differences in item interpretation may suggest caution making direct comparisons of index scores across age groups.

Introduction

There is a growing consensus that health is best understood as a holistic construct that involves the complete person. Expanding beyond a strictly biomedical model of health to include psychosocial and other domains, the concept of whole person health has emerged as a guiding framework for advancing integrative and interdisciplinary health research.^1,2 As described by the National Center for Complementary and Integrative Health (NCCIH), whole person health (WPH) emphasizes both the interconnected functioning across body systems and the processes that support resilience and health restoration.^2-4 Biological, behavioral, social, and environmental areas may all be considered whole person health factors that can promote health or contribute to disease.

With the shift towards this integrated view of health, recent efforts have sought to apply these concepts in practice. For example, the National Institutes of Health, the Department of Veterans Affairs, the Centers for Medicare and Medicaid Services, and numerous academic centers for integrative health have launched initiatives aimed at advancing research on and implementation of whole person health care. ⁵ This interprofessional team-based care emphasizes the inclusion of the patient as an active member of the care team and seeks to support their physical, behavioral, spiritual, and socioeconomic well-being. ¹

As these conceptual and practical advances have unfolded, there has been a growing need for tools capable of capturing health and assessing well-being at the level of the whole person.^6,7 In this context, NCCIH and the Centers for Disease Control and Prevention’s (CDC) National Center for Health Statistics (NCHS) developed a set of self-report questions designed to measure WPH.^8,9 Known as the Whole Person Health Index (WPHI), the nine-item scale asks respondents to rate their overall general health as well as their quality of life, social and family connections, diet, physical activity, ability to manage stress, sleep, existential wellbeing, and ability to manage their health (see Table 1 for question wordings). The WPHI questions all use the same response scale: Excellent, Very Good, Good, Fair, or Poor.

Table 1.

The Whole Person Health Index

Domain	Question text
Self-rated health	Would you say your health in general is excellent, very good, good, fair, or poor?
Quality of life	How would you rate your quality of life, focusing on what matters most to you?
Social and family connections	How would you rate your social and family connections?
Diet	In general, how healthy is your overall diet?
Physical activity	How would you rate your physical activity, compared with people in your age group?
Ability to manage stress	How would you rate your ability to manage stress?
Sleep	How would you rate your sleep?
Meaning and purpose	How would you rate your ability to find meaning and purpose in your daily life?
Health management	How would you rate your ability to manage your health, focusing on aspects of your health that matter most to you?

Note. All items use the same response scale: Excellent, Very Good, Good, Fair, or Poor.

A mixed-methods evaluation study of the 9 items included cognitive and pilot testing and validated each item against existing questionnaires in each of the covered domains. Results of this study indicated that the WPHI items largely function as desired, capturing the intended health constructs. ¹⁰ The existing evaluation research shows good promise for the WPHI as a composite measure with utility in academic and health care settings and in population-level surveys.

The WPHI is included in the 2025 National Health Interview Survey (NHIS) to provide nationally representative benchmarks, and it will be a required clinical outcome measure in a forthcoming NCCIH funding opportunity on whole person health restoration. ¹¹ As data from the WPHI becomes available, it is important for researchers and data users to understand the measurement properties of the scale, its potential strengths and possible limitations. The current study applies descriptive and psychometric analyses to explore the factor structure and its functioning across different population subgroups.

The present study is premised on the assumption that a singular underlying construct of “whole person health” drives responses to the WPHI items. The varied domains of the WPHI capture the multifaceted nature of health, and there is reasonable debate whether summary scores of general health status are best assessed as just a weighted combination of heterogeneous indicators.^12,13 We hypothesize that WPH can be understood and measured in a more cohesive way, that it is not merely an aggregation of diverse components. Rather, the WPHI may embody a unified construct that reflects the essence of WPH, and this underlying latent trait potentially influences individuals’ responses across all domains. This hypothesis is consistent with recent work finding that a general health factor could account for covariation among 12 dimensions of self-reported health. ¹⁴ Similarly, we also use traditional psychometric methods to evaluate the WPHI’s measurement properties and this theoretical foundation.

Methods

Data

The Research and Development Survey (RANDS) is NCHS’s methodological research survey system, which uses samples of adults age 18 and older from commercially available survey panels. RANDS 11, conducted January – March 2025, was used to evaluate the WPHI items. Topics in each round of RANDS vary and generally cover a wide range of health behaviors and conditions. On RANDS 11, the WPHI items were preceded by questions on specific health conditions, cognitive decline, and knowledge and experiences with psychoactive and non-psychoactive marijuana products.

RANDS 11 was fielded for NCHS by NORC at the University of Chicago using their statistically sampled AmeriSpeak panel as the primary sample source. Sampling strata were based on age, race and Hispanic ethnicity, education, income, and sex (96 sampling strata in total). The size of the selected sample per stratum was determined such that the distribution of the completed surveys across the strata matched that of the U.S. adult population. AmeriSpeak panelists were offered points equivalent to $5 for completing the survey online and $10 for completing it over the phone.

In addition to the AmeriSpeak sample, RANDS 11 also included non-probability supplementary samples from the Cint and Prodege opt-in online panels. Quota buckets were defined by the same 96 demographic strata. The non-probability samples included an oversample of adults age 45 and older to ensure adequate sample size of that group for analysis. Incentive information is not available for the non-probability samples.

The RANDS 11 AmeriSpeak sample included 13,757 panelists, 9092 of whom completed the survey. Of these complete cases, 8487 were completed via the web, while 605 were conducted using telephone interviews. The non-probability Cint and Prodege samples included 10,724 cases completed online.

Statistical Analysis

The 19,816 participants were randomly assigned to one of four groups for a 2 × 2 experiment that tested alternate versions of the quality of life and social connectedness questions. The current analysis focuses only on the data from the group (n = 4967) that received the final version selected for the NHIS. We excluded participants with missing responses to any WPHI item (n = 189), resulting in a final study population of 4778 participants. Given the inclusion of non-probability samples and the goal of assessing measurement properties rather than population estimates, sampling weights were not used for analyses.

Descriptive analysis includes the response distributions and polychoric intercorrelations, used to estimate relationships when variables are measured on an ordinal scale. Internal consistency for the scale was calculated using McDonald’s omega and Guttman’s split-half coefficient.^15,16 Then the sample was randomly split in half, using one half for exploratory analyses and the other half for confirmatory analyses and testing differential item functioning (DIF). All analyses were conducted in R, ¹⁷ using the polycor, ¹⁸ psych, ¹⁹ lavaan, ²⁰ and lordif ²¹ packages.

Exploratory factor analysis (EFA) was used to determine how many factors may underlie the 9 WPHI items. One-, two-, and three-factor solutions were evaluated; Promax rotation was used for the two- and three-factor solutions to allow for correlations between the factors. ²² Percent of variance accounted for and item loadings were used to assess model fit.

After EFA was conducted, confirmatory factor analyses (CFA) were used to assess the goodness-of-fit of the best-fitting measurement model obtained from the EFA. To account for the ordinal response scale, the polychoric correlation matrix was examined using the WLSMV (weighted least squares mean and variance adjusted) estimation method. ²³ Overall model fit was assessed via robust comparative fit index (CFI) (>0.90), robust root mean square error of approximation (RMSEA) (<0.06), and standardized root mean square residual (SRMR) values (<0.08). ²⁴ Item loadings were also considered, and modification indices were used to test alternative models. ²⁰

At the suggestion of a reviewer, exploratory and confirmatory bifactor analyses were also conducted to further evaluate the hypothesis that a general underlying latent trait influences individuals’ responses across all domains. ²⁵ Bifactor analyses test whether a general factor is sufficiently strong that the scale can be treated as essentially unidimensional, even if a multidimensional structure is assumed. A strong general factor would be confirmed if explained common variance (ECV) > 0.70, omega hierarchical (ωh) > 0.70, H index > 0.80, and factor determinacy (FD) > 0.90. ²⁶

Invariance Assessment

Using the confirmatory data subset, DIF was examined using item response theory (IRT) combined with ordinal logistic regression. WPHI items were analyzed across sex, age, race and Hispanic ethnicity, and educational attainment. Using the lordif function, likelihood ratio chi-squared tests were used to evaluate whether item parameters differed significantly across groups (P < 0.01), while using the graded response model for IRT trait estimates. ²¹ McFadden’s pseudo-R² acted as the measure of effect size for total DIF, including both uniform DIF (where bias is consistent across all levels of the latent trait) and nonuniform DIF (where bias varies with the level of the latent trait). We used ΔR² > 0.02 as the threshold for DIF, consistent with the criteria in lordif for flagging items with potential practical significance. This cutoff is more sensitive than the 0.035 threshold suggested by others for identifying items with at least moderate DIF. ²⁷

Results

Descriptive Statistics

As shown in Figure 1, the response distribution for each of the WPHI items exhibits an approximately normal pattern, with a slight negative skew observed for quality of life, social and family connections, diet, meaning and purpose, and health management. Table 2 provides the cross-item polychoric correlations for the nine WPHI items. Overall, the intercorrelations range between 0.41 and 0.67, suggesting moderate associations among most of the WPHI items. The highest intercorrelations can be seen for quality of life (with meaning and purpose, health management, and social and family connections). The lowest intercorrelations were observed between sleep and other items, particularly with social and family connections (0.41) and physical activity (0.42).

Figure 1.

Whole Person Health Index item response distributions. Note. Frequencies of missing values were: quality of life (n = 10), social and family connections (n = 7), diet (n = 48), physical activity (n = 32), ability to manage stress (n = 28), sleep (n = 41), meaning and purpose (n = 25), and health management (n = 12). SOURCE: NCHS, Research and Development Survey (RANDS) 11, 2025

Table 2.

Cross-Item Polychoric Correlations

​	Self-Rated Health	Quality of Life	Social and Family Connections	Diet	Physical Activity	Ability to Manage Stress	Sleep	Meaning and Purpose	Health Management
Self-rated health	1.00	0.65	0.45	0.59	0.63	0.46	0.46	0.45	0.63
Quality of life	0.65	1.00	0.67	0.55	0.54	0.58	0.51	0.66	0.67
Social and family connections	0.45	0.67	1.00	0.48	0.43	0.51	0.41	0.62	0.57
Diet	0.59	0.55	0.48	1.00	0.64	0.51	0.47	0.49	0.63
Physical activity	0.63	0.54	0.43	0.64	1.00	0.48	0.42	0.45	0.61
Ability to manage stress	0.46	0.58	0.51	0.51	0.48	1.00	0.56	0.62	0.62
Sleep	0.46	0.51	0.41	0.47	0.42	0.56	1.00	0.50	0.53
Meaning and purpose	0.45	0.66	0.62	0.49	0.45	0.62	0.50	1.00	0.65
Health management	0.63	0.67	0.57	0.63	0.61	0.62	0.53	0.65	1.00

Note. Total number of complete cases: 4778.

SOURCE: NCHS, Research and Development Survey (RANDS) 11, 2025.

Internal consistency for the scale was good, as indicated by values above 0.85: McDonald’s omega = 0.93 and Guttman’s split-half coefficient = 0.89.

Exploratory Factor Analysis

The one-factor solution demonstrated the best fit. All items loaded above 0.60 onto the factor and the items explain 53% of the shared variance. See Table 2 for item loadings for the one-factor solution. The two-factor solution resulted in three items cross-loading (quality of life, sleep, and health management), with only one (quality of life) loading above 0.45 on either factor. The factors were correlated at r = 0.77 and the two-factor model only explained an additional 6% of the variance. The three-factor solution resulted in four items cross-loading (quality of life, diet, meaning and purpose, and health management), with meaning and purpose and health management being almost evenly split between two factors and having relatively low loadings. The three-factor model explained only 3% more variance than the two-factor model.

Confirmatory Factor Analysis

The one-factor CFA showed mixed model fit, with a robust CFI = 0.90 and SRMR = 0.056, but RMSEA exceeded conventional thresholds (robust RMSEA = 0.151; 90% CI: 0.145, 0.158). Modification indices were used to examine the potential for local dependence; for one set of items with elevated values, there was an easy justification for why they may covary beyond the latent trait. Correlated residuals for diet and physical activity improved the fit of the one-factor CFA: robust CFI = 0.92, SRMR = 0.050, and robust RMSEA = 0.136 (0.130, 0.143). Additional CFA specifications, including models excluding individual items and models allowing additional correlated residuals, are reported as Supplemental Material (Table S1).

See Table 3 for full item-level results. All but one item (sleep) loaded at or above 0.70. Likewise, the variance in each item explained by the factor (R²) was above 0.50 for all items but sleep and physical activity. The uniformly strong loadings across most items indicate that a substantial portion of shared variance is attributable to a common underlying factor.

Table 3.

Exploratory and Confirmatory Factor Analyses (One-Factor Solution)

	EFA	CFA
	Item loading	Item loading	R2	Item residual
Self-rated health	0.69	0.76	0.58	0.43
Quality of life	0.80	0.84	0.71	0.29
Social and family connections	0.67	0.73	0.53	0.47
Diet	0.74	0.74	0.55	0.45
Physical activity	0.70	0.70	0.49	0.52
Ability to manage stress	0.73	0.75	0.56	0.43
Sleep	0.65	0.65	0.42	0.57
Meaning and purpose	0.75	0.80	0.64	0.36
Health management	0.82	0.85	0.72	0.27

Note. R² and item residuals values do not always sum to 1.00 due to rounding. Total number of complete cases: 2393 (exploratory) and 2385 (confirmatory).

SOURCE: NCHS, Research and Development Survey (RANDS) 11, 2025.

Bifactor Analysis

Exploratory bifactor modeling suggested three group factors, with the existential well-being item cross-loading onto two different group factors. This structure was subsequently evaluated using confirmatory bifactor analysis. The bifactor model showed improved fit relative to the one-factor CFA (robust RMSEA = 0.091, SRMR = 0.021), and global fit was excellent (robust CFI = 0.976). Bifactor indices for the confirmatory model suggest that responses are dominated by a general factor (ωh = 0.88, ECV = 0.78). Reliability estimates for the group factors were low (ω = 0.15 for each), indicating that they contribute minimal unique variance beyond the general factor. Construct replicability (H = 0.92) and factor determinacy (FD = .95) lend further support that the general factor is well-defined, would be expected to replicate in new samples, and can be estimated with good accuracy. See the Supplemental Material (Tables S2-S4) for full bifactor model results.

Differential Item Functioning

We investigated DIF by sociodemographic group to understand where item differences may exist. Items flagged with statistically significant DIF by sociodemographic groups are identified in Table 4; Table 4 also includes the magnitude of the total DIF effect (ΔR²). IRT threshold parameters and slopes are included as Supplemental Material (Tables S5-S7) for items with statistically significant DIF.

Table 4.

Change in McFadden’s Pseudo-R² (ΔR²) for Whole Person Health Index Items by Sociodemographic Group

​	Sex	Age	Race and Hispanic ethnicity
Self-rated health	<0.001	*0.0323	*0.0038
Quality of life	<0.001	*0.0062	*0.0035
Social and family connections	*0.0016	<0.001	<0.001
Diet	<0.001	*0.0037	0.0018
Physical activity	*0.0040	0.0012	*0.0018
Ability to manage stress	*0.0101	*0.0147	<0.001
Sleep	<0.001	*0.0049	0.0017
Meaning and purpose	<0.001	0.0013	<0.001
Health management	<0.001	<0.001	<0.001

Note. Statistically significant DIF (P < .01) is indicated by an asterisk. Results for educational attainment are not shown as statistically significant DIF was not found for any of the items. Practically significant DIF is detected if the DIF is statistically significant and McFadden pseudo-R² change is greater than 0.02, indicated in bold. Progressively complex logistic regression models were used to test for both uniform and non-uniform DIF. This analysis looks at total DIF by comparing the model incorporating sociodemographic group membership and the interaction between group and trait level (Model 3 in lordif) with the model that includes only the trait as a predictor of response (Model 1). Total number of complete cases: 4778.

SOURCE: NCHS, Research and Development Survey (RANDS) 11, 2025.

Three of the nine items exhibited statistically significant DIF between males and females. Five items exhibited statistically significant DIF across age groups. Three items exhibited significant DIF across racial and ethnic groups. No items were flagged for significant DIF across educational attainment groups. Based on ΔR² values, only the self-rated health item, by age group, demonstrated the potential for practical impact (ΔR² = 0.0323). The effect sizes for all other statistically significant DIF can be considered negligible.

Threshold parameters indicate when one response option becomes more likely than the previous one along the normal distribution of the latent continuum. For self-rated health, the values for each threshold (Table S2) increased linearly from the youngest age group (18-29) to the oldest (60+). That is, on this item, younger age groups consistently provided more positive ratings at lower levels of the latent trait compared to older age groups. This was most noticeable for excellent responses: Adults 18-29 were likely to respond excellent when their z-score exceeded 0.91, whereas adults 60+ were likely to respond excellent only when their z-score exceeded 1.83.

Discussion

We conducted descriptive and psychometric analyses to explore the measurement properties of the WPHI, which was included on a recent high-quality survey using a statistically sampled panel supplemented by an additional non-representative opt-in panel. The descriptive analysis of the WPHI items demonstrates generally favorable basic measurement properties, with strong internal consistency, good variability, reasonably symmetric distributions, utilization of all response options, and neither clustering at extreme categories nor restrictions due to floor or ceiling effects.

Our EFA, CFA, and bifactor analyses provided evidence supporting a single underlying factor across the nine items, although the RMSEA suggested possible misfit. Though RMSEA is a traditional index for assessing fit, it has limitations, especially for establishing unidimensionality. ²⁸ A single-factor model can lead to higher RMSEA values if the model does not capture all aspects of the data structure, as RMSEA penalizes models with fewer degrees of freedom more heavily for any misfit. ²⁹ So even when there is a strong common trait underlying responses and a scale is “unidimensional enough” for modeling index values with limited bias, seemingly unacceptable RMSEA values (0.10 to 0.20) can often occur. ³⁰

We acknowledge that the one-factor CFA does not fully reproduce the observed covariance structure among the WPHI items. However, considering the strong factor loadings, favorable CFI and SRMR values, and bifactor indices confirming that a large proportion of reliable variance is attributable to a general factor, we conclude that the WPHI items largely reflect a common underlying construct. These findings support interpreting the WPHI as reflecting a shared underlying dimension of health for purposes such as constructing an index score.

This evidence of unidimensionality should be considered in the context of the emerging literature on whole person health measurement. Recent reviews indicate that relatively few self-report measures of whole person health have undergone formal psychometric evaluation, and comparative evidence on structural validity across instruments remains limited. ⁶ Future methodological work could further evaluate the interpretation of model fit indices for measures of this type. For example, simulation studies examining expected fit statistics under conditions similar to the WPHI (eg, small numbers of items representing broad constructs) could help clarify appropriate benchmarks for evaluating dimensionality in this emerging measurement area.

It may be challenging for any reasonably sized multi-item scale to achieve optimal fit for a concept as complex as whole person health. Existing measures of WPH emphasize various domains, and there is no consensus among experts about the domains essential to the WPH concept. ⁶ Moreover, it has been suggested that self-report measures of WPH should separate causal indicators — determinants like diet, exercise, and sleep that can be the subject of interventions to improve WPH — from reflective indicators of overall functioning and well-being such as self-rated health and quality of life ³¹ ; the correlation observed between diet and physical activity after controlling for the latent trait lends some credence to this view. However, this correlated residual was specified post-hoc, so conclusions based on this finding should be confirmed in future studies. Future work could also examine whether other behavioral determinants of health might exhibit similar relationships beyond the general factor.

We do not wish to suggest that the evidence of a strong general factor presented here will resolve any debate about causal vs reflective indicators. However, we believe that our results support the hypothesis that all 9 items of the WPHI largely reflect a shared underlying dimension of health. Perhaps this is because the WPHI questions on diet, physical activity, and sleep seek subjective rather than objective assessments of these determinants and therefore may function more like reflections of overall well-being.

Regarding Sleep

Looking at the item factor loadings, the sleep item had the weakest relationship with the underlying factor, relative to the other items, which were all above 0.70. However, the sleep item loading (0.65) remains well above the commonly used thresholds (eg, 0.4-0.5) for retaining items in a factor model. ³² Sensitivity analyses (see Supplemental Material, Table S1) demonstrated that removal of the sleep item did not meaningfully improve model fit. There are both theoretical and empirical justifications for retaining sleep as a component of whole person health measurement. Moreover, it was the number one answer among those who responded to NCCIH’s public request for information (RFI) to help identify factors to consider as part of a framework for measuring whole person health. ³³ The sleep item was therefore retained in the model.

Further investigation could explore possible wording changes that might improve the functioning of the sleep item as part of the WPHI. As currently worded (How would you rate your sleep?), cognitive interviewing results suggest that some respondents think differently than others when responding to this item. Many considered the number of hours of sleep they got, whereas others considered how often they woke up during the night or their lack of deep sleep. ¹⁰ This differential interpretation could lead to the lower item factor loadings observed. If separate specific questions on “quantity of sleep” and “quality of sleep” (each with the excellent-to-poor rating scale) were fielded with the other WPHI items, one could examine if either wording change leads to stronger loadings than did the current sleep item.

Differential Item Functioning

The DIF observed in item thresholds for some items indicates that individuals in different groups with the same underlying level of WPH did not have the same probability of selecting response categories. However, effect size measures suggest that only the self-rated health item met the criteria for potentially problematic DIF, and only for age-related comparisons. When younger and older age groups have equivalent levels of WPH, younger age groups tend to rate their general health more positively. This implies age-related differences in item interpretation or use of response categories. Prior research suggests that interpretations of self-rated health vary across the life course, with respondents applying different internal standards and reference frames when selecting response categories.^34,35 For example, older adults, in particular, perhaps due to more complex health experiences and a greater awareness of health decline, may apply a stricter standard requiring a higher level of underlying health before describing their health as “excellent.” Future studies could see whether this DIF pattern replicates.

In practical terms, the presence of DIF suggests that direct comparisons of WPHI scores across groups defined by age should be interpreted cautiously. But it is worth noting that, while the magnitude of DIF for self-rated health by age (ΔR² = 0.0323) exceeded our predefined threshold to flag potential DIF (ΔR² > 0.02), it remained below other commonly cited thresholds for meaningful DIF. For example, Jodoin and Gierl suggest that values of ΔR² < 0.035 may be considered negligible DIF. ²⁷ So given the negligible effect of DIF for all other items (regardless of threshold) and the possible lack of meaningful DIF by age (depending on threshold), the WPHI generally shows promise for cross-group comparisons.

Measurement Considerations in Clinical Settings

Valid measurement is a necessary foundation for science, but the evidentiary requirements for measurement depend on how an instrument is intended to be used. While a robust single-factor measurement model (or a robust general factor in a bifactor model) may be important when the WPHI is used to construct an index score for research or population health applications, it is not a necessary precondition in clinical contexts where the goal is to support reflection or discussion about different aspects of health.

Although the analyses here are oriented towards research and population health applications, the WPHI was also developed to support conversations between patients and providers. In such contexts, the instrument may function primarily as a structured profile of responses rather than as a strict measurement scale. Regardless of the factor structure, the instrument may still help individuals reflect on their health, motivate healthier behaviors, and track their own changes over time. When the WPHI is used to form a profile of WPH rather than as a reflection of an underlying dimension of health, model fit is less central to its usefulness. However, if the WPHI is to be used in this way, additional qualitative research with target populations may be needed to provide evidence that the most important domains are included and appropriately represented.

Use in clinical settings is therefore an important area for future research. Whole person health emerged in the context of patient care, and several frameworks and instruments have been developed specifically for clinical use.^13,36 In a recent systematic review of whole person health instruments designed to support patient–clinician interaction, Thomas and colleagues found that no existing approach was ready for routine implementation in general practice, although several instruments showed promise if further adapted and evaluated. ³⁷ They emphasize the need for continued work translating the theoretical foundations of whole person health into clinical settings, highlighting that such approaches must remain flexible, adaptable and grounded in the therapeutic relationship.

Scoring and Limitations

For public health purposes, there is generally an interest in population-based estimates, comparing groups, and looking at estimates over time. Pitcher and Langevin^8,9 have proposed that an index value for the WPHI can be calculated as the sum of individual item scores such that an increase in WPHI over time represents an improvement in health (see also their accompanying Viewpoint article). The convergence of evidence with both the single-factor measurement model and the bifactor model supports creating an index value in this way. The RANDS data reveal a strong correlation (r = 0.99) between the underlying factor score from the CFA and this index value. Given these findings, the WPHI may be confidently used to interpret scale scores (index values) or assess patterns of response nationally, but direct comparisons of scores and means across subgroups defined by age should be made with caution due to the identified DIF.

The analyses here do not directly answer the question of whether one can reliably compare national or subgroup estimates from one time point to another. Test-retest analyses would be necessary to assess the stability of the WPHI over time and its sensitivity to detect changes following major life events (eg, trauma) or lifestyle modifications, health and wellness coaching, or other interventions. For other subjective global health measures, test-retest reliability is generally good (correlations > 0.80) over 3-6 months in clinical settings with stable patients.^38,39 WPHI responses are expected to be similarly stable over short intervals in the absence of real health change.

DIF by age, however, suggests that how people use the scale may change as they age. With only cross-sectional data we could not rule out that this DIF might reflect generational differences rather than the process of aging. Still, we expect that population means can be compared reliably over time provided that the age distribution of the population has not changed or can be standardized. ⁴⁰

The analyses also do not address the possibility of differential item functioning across people with different illnesses or impairments, or between people likely to be seen in clinical practice and those who are not. Researchers focusing on WPH among specific clinical populations may wish to do such studies.

Finally, we note that the inclusion of non-representative opt-in panel samples alongside the statistically sampled panel could have introduced selection bias that might limit the generalizability of the present results. The opt-in panel was included to ensure sufficient sample sizes for examining DIF. Sensitivity analyses that limited the EFA, CFA, and bifactor analyses to just the statistically sampled panel did not reveal any meaningful differences with the results reported here (see Supplemental Material, Tables S8-S10).

Conclusion

In conclusion, our psychometric analysis of the WPHI supports its potential as a broad measure of WPH. The scale demonstrated strong internal consistency, and results provide evidence that the items reflect a common underlying construct, even though model fit missed conventional benchmarks for a strict one-factor model. The findings highlight the complexity of measuring a multifaceted construct like WPH, and the observed DIF across age groups underscores the importance of considering contextual influences. The WPHI holds promise as a robust tool for enhancing health assessments, research, and surveillance. Moving forward, further validation efforts, including test-retest reliability and determining how well the common underlying factor represents the WPH concept, can strengthen the WPHI’s applicability.

ORCID iDs

Catherine Lamoreaux https://orcid.org/0009-0002-3373-4935

Sarah E. Forrest https://orcid.org/0009-0000-3873-9992

Stephen J. Blumberg https://orcid.org/0000-0002-2086-4768

Ethical Considerations

This secondary data analysis was reviewed by CDC, deemed not research with human subjects, and was conducted consistent with applicable federal law and CDC policy.

Data Availability Statement

De-identified RANDS data files are available from NCHS (https://www.cdc.gov/nchs/rands/index.html).