Data analysis
The latent class model is an individual-centred method to analyze the
characteristics of different groups of people and the differences in
various indicators between different categories, identify high-risk
groups, and provide a basis for targeted intervention measures(Howard &
Hoffman, 2018). LPA was conducted in Mplus version 8.3 to identify
mental health latent profiles in women with PCOS based on the
dual-factor approach. Given the differences between PHQ-9, GAD-7, BISS
and IWB scales, we converted the total scores of these four scales into
Z scores for latent profile analysis. We start from a zero model,
gradually increase the number of potential categories for model fitting,
and select the most suitable model based on the parameters of fitting
indicators such as AIC, BIC, entropy index, BLRT, and LMR while
considering the interpretability and practical significance of each
category. The optimally appropriate model is selected based on the
following model-fit indices (Table 2): (1) model fit was determined
using the Akaike information criterion (AIC),
Bayesian information criterion
(BIC) and sample-size adjusted Bayesian information criterion (aBIC),
with smaller values indicating a better fit; (2) the Lo-Mendell-Rubin
likelihood ratio test (LMR LRT) and bootstrap likelihood ratio test
(BLRT), in which p<0.05 indicates that the tested model fits
better(Toker & Green, 2021); (3) Entropy ranges from 0 to 1, whereby a
higher value indicates higher classification utility, and a value of
> 0.80 indicates a highly differentiated latent profile.
Once the optimal mental health latent class was determined, we used SPSS
26.0 for subsequent analysis after the determination of the latent
profiles. We conducted a series of analysis of ANOVAs to examine the
differences in anxiety, depression, body image satisfaction and
subjective well-being among different latent profiles. Differences in
demographic and anthropometric, cognitive reappraisal, expressive
suppression, and social support across mental health profiles were
examined through multinomial logistic regression.