Participants

The study took place at the out-patient cervical pathology and colposcopy clinic of a tertiary university hospital. All procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2013. In addition, the study protocol was approved by the local ethics research committee (no. 839/26.10.2021). A written statement of agreement to participate in this study was provided by each participant before entering the study. On the basis of the international literature, approximately 150 women aged 18–45 years were invited to participate in this study after their first visit to the out-patient cervical pathology and colposcopy clinic. Inclusion criteria for the study included cytological examination suspicious for HPV or for cellular atypia, ability to understand and read the Greek language, absence of history of depression, and absence of organic or functional pathology. Women positive for HPV infection and those with obvious cervical lesions that were scheduled for immediate medical or surgical treatment were also excluded from the present study.

Procedures

This study followed a cross-sectional design. Women that met the inclusion criteria were identified by their physician and invited by one of the researchers to participate in the study. Participation was voluntary, and data confidentiality was guaranteed. After signing the informed consent form, women answered the questionnaires on the day of their medical appointment in the initial period (1–6 weeks) after the HPV diagnosis.

Measurement tools

Demographic and clinical data were obtained including age, marital status, educational level, children, number of sexual partners and history of warts. In addition, participants completed the following tools.

1. (a) Life Satisfaction Inventory (LSI) to measure QoL. This is a 13-item tool that assesses patients’ satisfaction with physical and mental health, work, financial status, relationship with partner, sexual life, family life, role within family, friends, hobbies, appearance and general QoL during the past week. ^{Reference Muthny, Koch and Stump15} The scale shows good internal consistency with Cronbach’s α = 0,82. The tool has been translated and validated in a Greek healthy population. ^{Reference Fountoulakis, Iakovides, Christoforides and Ierodiakonou16}

2. (b) Female Sexual Function Index to assess sexual function. A version of the original English tool translated into Greek and evaluated for reliability was used. ^{Reference Zachariou, Filiponi and Kirana17} This tool was developed to assess a woman’s sexual activity in the previous 4 weeks. ^{Reference Rosen, Brown, Heiman and Leib18} It allows assessment, on a scale from 0 (no sexual activity) to 5 (always/very high sexual activity), of sexual desire, arousal, hydration, orgasm, satisfaction and dyspareunia. The score has a range of 2–35, with values ≤26.55 considered to be acceptable for diagnosis of sexual dysfunction among women of a wide age range. ^{Reference Rosen, Brown, Heiman and Leib18} In the Greek version, all six subdomains of the questionnaire demonstrated good internal validity (Cronbach’s α = 0.92, P < 0.01) and excellent reliability (intraclass correlation coefficient: 0.91, P < 0.01). ^{Reference Zachariou, Filiponi and Kirana17} The authors categorised women with and without sexual dysfunction with a cut-off value of 26 (sensitivity: 71.4%; specificity: 92.2%).

3. (c) HPV Impact Profile (HIP) questionnaire to estimate HPV-related psychosocial burden. ^{Reference Mast, Zhu, Demuro-Mercon, Cummings, Sings and Ferris19} This questionnaire contains 29 items that are summarised in seven domains. Answers are given on a 0–10 Likert scale, and total HIP scores range from 0 (no impact) to 100 (worst impact). Its Cronbach’s α has been reported to range from 0.64 to 0.90, with values ≥0.7 reported for five of the seven domains. The original version of the HIP tool exhibits acceptable discriminant construct validity and reliability and can be used to estimate the importance of the HPV-related burden. ^{Reference Mast, Zhu, Demuro-Mercon, Cummings, Sings and Ferris19}

4. (d) General Health Questionnaire-28 to measure general psychological health. This questionnaire is a self-report screening measure used to assess possible psychological disorders and aims to examine inability to carry out normal functions, as well as the appearance of new and distressing phenomena. It comprises four seven-item subscales (somatic symptoms, anxiety/insomnia, social dysfunction and severe depression) and is used to identify the presence of symptoms compared with what is normal for the individual. ^{Reference Goldberg and Hillier20} A validated Greek version has been published. ^{Reference Garyfallos, Karastergiou, Adamopoulou, Moutzoukis, Alagiozidou and Mala21}

Statistical analysis

For descriptive purposes, observed central tendencies and dispersions are reported using means and standard deviations for continuous variables and proportions or frequencies for categorical data as appropriate.

A Bayesian framework was chosen instead of a frequentist approach for all analyses as this allowed results to be interpreted intuitively through reporting of subjective probabilities. ^{Reference Kruschke22} Briefly, in frequentist linear models, results are interpreted on the basis of P-values and confidence intervals. These reflect how likely the observed data are, assuming there is no effect (the null hypothesis). For example, a 95% confidence interval means that if the experiment is repeated many times, 95% of the intervals would contain the true value. ^{Reference Kruschke22} By contrast, Bayesian linear models combine prior knowledge (or assumptions) with the data to directly produce a posterior distribution of the parameter estimates. ^{Reference Kruschke22,Reference Gelman, Carlin, Stern and Rubin23}

Consistent with the aim of our analysis, the response variables were QoL and sexual function, with HPV-related psychological burden and general psychological health serving as potential explanatory variables. In addition, baseline demographic and clinical data (age, marital status, educational background, having children, number of sexual partners and history of warts) were used as covariates to control for potential confounding. We tested moderation models both with and without adjustment for baseline covariates to assess whether the inclusion of these variables improved model fit or altered the substantive interpretation of the moderation effect. This approach enabled us to evaluate the robustness of the effect while accounting for potential confounding and/or overfitting.

For the purpose of the moderation analysis, we compared eight competing models: model 1 had HPV-related burden, model 2 had general psychological health, model 3 had both HPV-related burden and general psychological health, and model 4 had HPV-related burden and general psychological health and their interaction as predictors of QoL or sexual function. In addition, models 1–4 were adjusted for age, marital status, children, educational background, number of partners and history of warts. Models 5–8 corresponded to models 1–4 but without adjustment for baseline demographic and clinical data. A substantial interaction (model 4, with covariate adjustment; model 8, without covariate adjustment) would indicate that the effect of HPV-related psychological burden on QoL or sexual function was moderated by general psychological health. In that case, general psychological health would be the moderator of HPV-related psychological burden effect on QoL or sexual function. Identifying a moderator of an effect helps to establish the boundary conditions of that effect or the circumstances, stimuli or type of people for which the effect is large versus small, or present versus absent. ^{Reference Hayes24} Models were compared and weighted using information criteria based on out-of-sample predictive accuracy. Specifically, we used leave-one-out cross-validation to assess model fit and generalisability. For each model, the expected log predictive density was estimated, and model weights were calculated using the leave-one-out model weighting approach. ^{Reference Hayes24} This allows probabilistic model comparison and accounts for model uncertainty by estimating how likely each model is to be the best predictor of new data. This approach is particularly suited to Bayesian models, for which conventional likelihood-based criteria (such as the Akaike information criterion or Bayesian information criterion) may not be appropriate or sufficient for comparison of models with different structures or priors. In addition, the Bayesian version of R ² was used to quantify the amount of ‘explained’ outcome variance for each model. ^{Reference Gelman, Carlin, Stern and Rubin23}

The variables used to measure QoL and sexual function (LSI and Female Sexual Function Index, respectively) were log-transformed before analyses; back-transformation of the effects yielded their magnitude in percentage units. In addition, to augment interpretation of the intercept (reference category) and slope (difference between the reference and other categories), HPV-related psychosocial burden (estimated by HIP) was standardised using Z-score transformation (thus, ZHIP = 0 refers to the average HIP of the sample). Finally, general psychological health (quantified by the General Health Questionnaire-28) was entered as a categorical predictor (‘no distress’ versus ‘probable distress’) on the basis of a cut-off value ≥5. Inferences from all analyses were performed on posterior samples generated using the Hamiltonian Markov chain Monte Carlo method and through use of 95% credible intervals (CrIs). For all models, inferences were made as one-sided hypotheses, and a posterior probability that exceeded 95% was considered to suggest the existence of evidence for substantial difference. ^{Reference Kruschke22} Weakly informative priors (Student’s t and half Student’s t with three degrees of freedom for intercepts and variance parameters, respectively) were used in all models; for model effects (slopes or differences between factor levels and reference category), we used weakly informative normal priors (N ∼ 0, 10). ^{Reference Kruschke22} Data were analysed using R (version 4.3.2; R Core Team, R Foundation for Statistical Computing, Vienna, Austria) on a Windows 11 platform, employing the brms wrapper package interfaced with the probabilistic programming language Stan for Bayasian sampling (https://www.R-project.org). ^{Reference Bürkner25}