Two lifestyle profiles emerged, but the total score did most of the work
A cross-sectional study in Yunnan, China, separated 957 ethnic minority adults with hypertension, diabetes, or dyslipidemia into two probabilistic health-promoting lifestyle groups, yet the statistical divide between them was modest. One profile contained 508 participants, or 53.08%, and the other included 449 participants, or 46.92%. Across all six domains of the Health-Promoting Lifestyle Profile-II Revised, the lower-profile group scored lower and the higher-profile group scored higher, producing a clean quantitative ordering rather than sharply different behavioral phenotypes.
The clinical signal is real, but it is not dramatic. Classification separation was limited, with entropy of 0.506, and 184 participants, or 19.23%, had a maximum posterior probability below 0.70. In practical terms, the model identified broad lifestyle strata, not rigid compartments. That matters because health-promoting behavior in chronic cardiometabolic care usually moves along a continuum: diet, exercise, stress management, and health responsibility tend to cluster, but rarely with perfect internal consistency.
A probabilistic model captured lifestyle engagement better than a rigid cutoff
Investigators compared one- through five-profile latent profile models under equal-volume, equal-shape diagonal and varying-volume, varying-shape diagonal covariance structures. The two-profile EEI solution offered the most parsimonious fit while preserving adequate class sizes. Modal assignment by the latent profile model agreed closely with a simple empirical median split of the total HPLP-II R score, with 95.61% concordance and Cohen’s κ of 0.912.
Even more striking, the continuous total score nearly perfectly separated the modal groups, with an area under the receiver operating characteristic curve of 0.995. That result means the latent profile solution added limited incremental information beyond the total score itself. In other words, the distribution looked different on paper, but the underlying message stayed the same: a higher total lifestyle score tracked closely with membership in the higher-engagement profile.
The study also ran supplementary psychometric analyses. Subscale reliability remained consistent, and no substantial measurement non-equivalence appeared across the five major ethnic groups. A careful measurement framework matters here because lifestyle research can drift quickly when a questionnaire behaves differently across subpopulations. Here, the instrument held up well enough to support the broader comparison.
Age, multimorbidity, and physical activity tracked with the higher profile
After accounting for classification uncertainty through posterior-probability multiple-imputation logistic regression using 5,000 pseudo-class draws, three factors stood out. Older age, meeting the study criteria for two cardiometabolic conditions, and higher PARS-3 physical activity level were associated with the higher lifestyle profile. The estimate for three conditions pointed in the same direction, but it did not reach statistical significance.
Physical activity deserves particular attention because it linked the lifestyle construct to a concrete behavior domain. PARS-3 total score showed a modest positive correlation with the standardized HPLP-II R physical activity dimension score, with Spearman’s ρ of 0.315, 95% CI 0.258–0.372, and P < 0.001. When the physical activity dimension was removed from the profile indicators in sensitivity analysis, high PARS-3 physical activity still predicted the higher profile, while moderate physical activity lost statistical significance. The direction was stable; the strength was not.
Older adults may have accumulated more experience navigating chronic disease routines, or they may simply have become more consistent about self-management after repeated clinical contact. The dataset cannot sort out mechanism, but the association is clinically plausible. Multimorbidity can also act as a behavioral forcing function, pushing patients toward more regular medication use, diet attention, and self-monitoring once the burden becomes difficult to ignore.
Ethnicity signal weakened once region and recruitment source entered the model
Ethnicity-related associations were attenuated after adjustment for geographic region and recruitment source with site-clustered standard errors. The investigators were careful to frame those contrasts as context-dependent differences within the sampled sites rather than independent ethnic effects. In a region as heterogeneous as Yunnan, where geography, access to services, language environment, and local recruiting patterns can all shape health behavior, that caution is justified.
This is the point where naive interpretation would fail. Ethnic category alone does not explain the observed lifestyle gradient, and the adjusted model did not support a simple story of group-level behavioral superiority or deficit. Instead, location and the practical realities of recruitment appear to account for part of the apparent difference. For clinicians and public health teams, the implication is straightforward: interventions should be built around local context, not assumed ethnic archetypes.
What the profile analysis adds, and what it does not
The selected profiles represented quantitatively ordered levels of health-promoting lifestyle, not qualitatively distinct syndromes. They probably reflect how several health behaviors cluster together in everyday chronic disease management. A person who walks more, eats more deliberately, and pays more attention to self-care often also scores higher on related lifestyle dimensions because these behaviors reinforce each other through routine, self-efficacy, and repeated contact with the health system.
Still, the study leaves the strongest practical conclusion in the hands of the total score. Median-split and continuous total-score analyses generated largely concordant association patterns, and the latent profile model supplied limited extra clinical information beyond that aggregate measure. The profile assignment should therefore be treated as an exploratory probabilistic marker of overall lifestyle engagement rather than a diagnostic label.
For chronic cardiometabolic care, the message is less about hidden subtypes and more about scalable measurement. A brief, reliable lifestyle instrument can separate broad groups, but the boundary between those groups remains fuzzy. In that setting, clinicians and public health planners may gain more from targeting the behaviors themselves than from overinterpreting latent classes.
One final point is easy to miss. Nineteen percent of participants landed in a classification gray zone, and that uncertainty is not a flaw of the analysis so much as a reminder of how smoothly lifestyle risk varies across real populations. The model identified direction, not destiny. And in chronic disease management, direction is often the more useful signal anyway.