Occurrence and clinical characteristics of Compulsive Sexual Behavior Disorder (CSBD): A cluster analysis in two independent community samples (2020)

CSBD diagnostic framework and criteria

Even when CSBD has been included in the ICD-11, the appropriate diagnostic framework and criteria for this clinical condition are still under discussion (Kraus et al., 2018; Walton, Cantor, Bhullar, & Lykins, 2017). Concerning current nosological status, a myriad of theoretical positions about how CSBD should be classified has been proposed and this clinical condition has been conceptualized as an addictive disorder (Potenza, Gola, Voon, Kor, & Kraus, 2017), a sexual disorder (Kafka, 2010; Walton et al., 2017), an impulse control disorder (Reid, Berlin, & Kingston, 2015), or not considered a disorder at all (Moser, 2013). Each theoretical approach proposes different criteria for the diagnosis of this condition, further emphasizing the conceptual chaos and hindering the identification of an unique profile of patients displaying symptoms of this clinical condition (Karila et al., 2014; Wéry & Billieux, 2017).

Current evidence derived from studies in clinical populations suggests that CSBD satisfies the majority of the core criteria proposed for the operational definition of behavioral addictions (Billieux et al., 2017; Kardefelt-Winther et al., 2017): (a) excessive time/effort spent on sexual behavior; (b) impaired self-control; (c) systematic failure to fulfill family, social, or work responsibilities; and (d) persistence in the sexual behavior despite its consequences. These criteria coincide with those proposed for the inclusion of CSBD in the ICD-11 (World Health Organization, 2018) and with some of the criteria proposed by Kafka (2010) for the recognition of Hypersexual Disorder (HD) in the DSM-5. Additionally, Kafka’s proposal included an important criterion not considered by the ICD-11: i.e., repetitively engaging in sexual fantasies, urges, or behaviors in response to dysphoric mood states (e.g., anxiety or depression) or in response to stressful life events (work problems, bereavement, etc.). Different studies support the relevance of the use of sex as a maladaptive coping mechanism aimed to compensate for unpleasant affective states or stressful life events in people with CSBD (Reid, Carpenter, Spackman, & Willes, 2008; Schultz, Hook, Davis, Penberthy, & Reid, 2014).

Furthermore, there are other symptoms not directly included neither in the DSM-5 nor the ICD-11 but relevant in the manifestation of CSBD: i.e., preoccupation with sex, salience, and self-perceived sexual problems. These symptoms constitute common cognitive manifestations of CSBD. Seminal models such as the “component model of addiction” (Griffiths, 2005) or recent network analysis have highlighted the important role of cognitive symptoms in cybersex addiction (Baggio et al., 2018) or HD (Werner, Štulhofer, Waldorp, & Jurin, 2018). As defined by Griffiths (2005, p. 193), salience refers to “when the particular activity [sex] becomes the most important activity in the person’s life and dominates their thinking (preoccupations and cognitive distortions), feelings (cravings) and behavior (deterioration of socialized behavior)”. Similarly, different studies highlight the crucial role of self-perceived sexual problems in the identification of patients displaying CSBD (Grubbs, Perry, Wilt, & Reid, 2019c).

Main approaches in the identification and classification of people with CSBD

Clinicians and researchers should be very cautious when diagnosing CSBD (Humphreys, 2018). One of the issues that hinders the reliability of many studies in the field is the way in that these researches identify and classify participants with CSBD. Different criteria have been employed to address this aim. Some studies have identified individuals with CSBD based on their scores on different self-report measures (Parsons, Grov, & Golub, 2012). Unfortunately, the majority of CSBD assessment scales do not provide reliable cutoff scores derived from clinical samples (Miner, Raymond, Coleman, & Swinburne Romine, 2017), so proposed thresholds are often arbitrary and/or based on statistical (not clinical) criteria. The study conducted by Bőthe et al. (2019) constitutes an illustrative example: after analyzing psychometric properties of the Hypersexual Behavior Inventory in a large nonclinical sample, these authors were unable to find a sensitive and specific cutoff score for the diagnosis of CSBD. Furthermore, positive predictive value for the cutoff typically used for the diagnosis of hypersexuality (raw score >53) was 14% (meaning that among participants scoring above 53 in the HBI, only 14% really qualified for this diagnosis). Thus, they recommended the use of alternative indicators and measures for the diagnosis of this condition.

Alternatively, other researchers have considered self-identification as having problems controlling sexual behavior (Smith et al., 2014) or seeking treatment for CSBD (Scanavino et al., 2013) as reliable indicators of CSBD. As an example, recently Grubbs et al. (Grubbs, Grant, & Engelman, 2019a; Grubbs, Kraus, & Perry, 2019b) conducted two studies in which problematic pornography use was measured through single items such as “I am addicted to pornography” or “I would call myself an internet pornography addict”. However, some individuals recognizing themselves as having CSBD problems may not actually exhibit either the clinical characteristics or the severity of this disorder, but only moral disapproval of their own sexual behavior (Grubbs, Perry, et al., 2019c; Grubbs, Wilt, Exline, Pargament, & Kraus, 2018; Kraus & Sweeney, 2019).

Finally, other studies identified CSBD participants through structured or semi-structured clinical interviews (Reid et al., 2012). Even when this approach is considered as a “golden rule” when assessing the presence and severity of CSBD (Hook, Hook, Davis, Worthington, & Penberthy, 2010; Womack, Hook, Ramos, Davis, & Penberthy, 2013), the quality of this assessment often relies on the particular diagnostic criteria guiding this semi-structured interview. Further, assessment through structured clinical interview is time-consuming, so applicability of this procedure in research (i.e., studies comprising large samples) is often limited.

In the absence of an accurate diagnostic framework for CSBD (Kraus & Sweeney, 2019), an alternative approach is to identify individuals with CSBD through data-driven approaches (e.g., cluster analyses). This procedure is particularly advised in research contexts, where a large number of participants should be assessed in a limited time frame and classification as sexually compulsive or not occurs post hoc. A recent study by Efrati & Gola (2018b) satisfactorily identified adolescents with CSBD (12 and 14% of two independent samples) through a data-driven approach (Latent Profile Analyses, LPA). Internal and external validity of this cluster approach was demonstrated by analyzing psychosexual profile of adolescents in the CSBD cluster (characterized by an external locus of control, anxious attachment, greater loneliness, higher frequency of pornography use, and more online sexual activities). Similarly, Bőthe et al. (2019) identified adults with high risk of serious hypersexuality (around 1% of the sample) using LPA. Therefore, in the absence of an appropriate diagnostic framework as well as brief and sound screening tools (Montgomery-Graham, 2017), data-driven approaches constitute a reliable method to explore CSBD in research contexts comprising large samples.

The present study

The purpose of the present study was to explore the occurrence and sociodemographic, sexual, and clinical characteristics of CSBD in two independent community samples. However, we tackled two limitations of previous research before addressing this aim: (1) the lack of standardized screening tools for assessing the whole range of cognitive, behavioral, and emotional symptoms of CSBD and (2) the low accuracy of different approaches usually applied in research contexts to identify CSBD patients. Therefore, we followed a three-step process to address the study aim.

First, we developed a new composite index to assess the whole range of CSBD symptoms. This index relied on three previously validated scales for the assessment of CSBD: the Hypersexual Behavior Inventory (HBI, Reid, Garos, & Carpenter, 2011b), the Sexual Compulsivity Scale (SCS, Kalichman & Rompa, 1995), and the Sexual Addiction Screening Test (SAST, Carnes, 1983). Independently, these measures tend to be excessively narrow in the assessment of CSBD, not covering the wide range of symptoms that should be explored to accurately assess this clinical condition (Womack et al., 2013); however, altogether these scales offer a very comprehensive assessment of CSBD symptoms and severity. To deal with the problem of using these scales independently, we performed a comprehensive review of their content, linking their items with different CSBD symptoms and creating a composite index assessing the following criteria: (a) loss of control, (b) neglect, (c) unable to stop, (d) continued engagement despite interference, (e) coping, and (f) Preoccupation, salience, and self-perceived sexual problems (for a comprehensive description of each symptom, see Table A1 in the Appendix). The theoretical frameworks for linking scale items with each specific symptom were the ICD-11 CSBD criteria (World Health Organization, 2018), the DSM-5 proposal for the diagnosis of hypersexuality (Kafka, 2010), and the component model of addiction (Griffiths, 2005). The procedure was equivalent to that followed by Womack et al. (2013) in their review of hypersexuality measures: two independent coders linked each item with a diagnostic criterion, and a third independent coder resolved any discrepancies. For the sake of clarity, items assessing more than one CSBD symptom or not clearly assessing any symptom were excluded from the new composite index.

Based on this composite index, we subsequently identified individuals with CSBD through a cluster analytic approach. Cluster analysis lets to uncover homogeneous groups of individuals according to the magnitude and the pattern of scores in different indicators, and has been increasingly used for the identification of people with different mental health issues (such as problematic use of mobile dating apps [Rochat, Bianchi-Demicheli, Aboujaoude, & Khazaal, 2019] or excessive engagement in videogames [Musetti et al., 2019]). Through this method, we classified 2,899 participants derived from two independent samples into two clusters (non-CSBD and CSBD participants). Considering the preliminary nature of proposed CSBD criteria and the precarious development of cutoff scores, this data-driven approach presents advantages in the identification of this clinical population, such as avoiding the use of arbitrary cutoff scores or relying on self-perception of sexual problems. Furthermore, cluster analysis is useful for understanding intraindividual dynamics, instead of interindividual differences (such in the case of variable-oriented approaches) (Bergman & Magnusson, 1997). Finally, compared to more complex data-driven approaches that require the use of advanced statistical software for their calculation (e.g., LPA), cluster analysis could be easily implemented through popular software (e.g., SPSS), with a high degree of overlap between the results of both statistical procedures (DiStefano & Kamphaus, 2006; Eshghi, Haughton, Legrand, Skaletsky, & Woolford, 2011).

Finally, we employed clusters derived from the previous analyses to explore the occurrence and characteristics of participants qualifying as sexually compulsives. Different a priori hypotheses were tested. Because current evidence points out that the prevalence of CSBD ranges between 1 and 6% (Bőthe et al., 2019; Walton et al., 2017), it was hypothesized that occurrence of CSBD in our samples will fall into this range, with males comprising a large proportion (∼80%) of participants in this group. As for offline and online sexual behavior, we expect to find a greater frequency, variety, and severity of sexual behaviors among CSBD participants (Klein et al., 2014; Odlaug et al., 2013; Winters, Christoff, & Gorzalka, 2010). Linked to this increased sexual activity, we expect that CSBD participants will score higher in sexual dispositional traits such as sexual sensation seeking (Kalichman & Rompa, 1995; Klein et al., 2014) or erotophilia (Rettenberger, Klein, & Briken, 2015). Finally, to the extent that CSBD patients tend to use sex as a coping mechanism, we also hypothesized that scores on scales assessing depression (Schultz et al., 2014), anxiety (Carvalho, Guerra, Neves, & Nobre, 2014; Reid, Bramen, Anderson, & Cohen, 2014; Voon et al., 2014), and self-esteem (Chaney & Burns, 2015; Reid, Carpenter, Gilliland, & Karim, 2011a) would be increased in CSBD participants.

Participants and procedure

Participants in this research were recruited from two independent studies on CSBD. Data acquisition for the first sample was conducted between 2012 and 2015. During this period, we used a cross-sectional, street intercept survey method to collect data on a large convenience sample of Spanish college students. In particular, the research team set an information table in the main entrance of different higher education centers and a member of the team actively approached potential participants. Students were asked to voluntarily collaborate with a research on sexual behavior. Those who accepted, completed an individual in-office assessment where an experienced clinical psychologist administered various self-reports. The average time to complete the study was around 1 hour and 45 minutes and participants received 10€ as a compensation for their participation.

Data acquisition for the second sample was conducted between 2016 and 2018. Sampling objective was to assess CSBD in a large sample of Spanish-speaking community members. The research was conducted online through a secured online platform aimed to provide information and assessment about CSBD (https://adiccionalsexo.uji.es/). Participants were enrolled utilizing a combination of active and passive recruitment strategies. Active recruitment included: (1) email blast through different institutions’ listservs (universities, organizations, etc.); (2) dissemination of the study on radios and newspapers websites; (3) posting banners on Facebook through the «suggested publications» marketing service and; (4) posting tear-off flyers in high-density spots (shopping centers, supermarkets, etc.). The study survey was also accessible through any search engine by using terms such as “sexual addictions” and/or “sex addiction assessment” (in Spanish) (passive recruitment). During the time the study was accessible, 3,025 participants accessed the survey. Initial data derived from the online platform were screened to avoid duplicitous, inconsistent, and/or fake responses (e.g., participants reporting >100 years old). Given that one of the CSBD scales that we used for participants clustering (the Hypersexual Behavior Inventory, HBI) was placed at the end of the online survey, only those participants who completed 100% of the survey were included in the study. After removals, a total of 1,318 participants were included in the final dataset. The average time to complete the study was 27.82 minutes (SD = 13.83) and participants did not receive compensation for participating.

Consequently, a total of 2,899 from two independent samples participated in the study. The first dataset included a convenience sample of 1,581 Spanish university students (56.9% females) ranging between 18 and 27 years old (M = 20.58; SD = 2.17). The second dataset included a more heterogeneous sample of 1,318 community members (43.6% females) aged from 18 to 75 years old (M = 32.37; SD = 13.42). Table 1 shows participants’ characteristics in both samples.

Measures

Data analysis

We undertook analyses in four steps. First, descriptive analyses were conducted to characterize participants in terms of sociodemographic data using the SPSS statistic package (version 25.0). To compare participants’ characteristics in sample 1 and 2, we performed t tests (continuous variables) and chi-square tests (categorical variables). Two effect size indices (Cohen’s d and Cramer’s V) were computed by using G*Power (version 3.1). For Cohen’s d, effect sizes of about .20 were considered small, close to .50 moderate and greater than .80 large (Cohen, 1988); for Cramer’s V, these sizes corresponded to values of .10, .30 and .50 (Ellis, 2010).

Second, a Confirmatory Factor Analysis (CFA) was conducted to test the psychometric suitability of our theoretically driven classification of CSBD symptoms. EQS software (version 6.2) was used to perform the CFA. Due to the non-normal distribution of the data, robust estimation methods were used. The CFA’s goodness of fit was analyzed with the following indices: Satorra-Bentler chi-square (χ²), relative chi-square (χ²/df), general model significance (P), root mean square error of approximation (RMSEA), comparative and incremental fit indexes (CFI and IFI), and the Standardized Root Mean Square Residual (SRMR). An appropriate fit was considered when χ² was not significant (P > .05), χ²/df was between 1 and 2, CFI and IFI were ≥.95, and RMSEA and SRMR was ≤.05 (Bagozzi & Yi, 2011). According to less restrictive criteria, values between 2 and 3 for χ²/df, ≥ .90 for CFI and IFI, ≤ .08 for RMSEA, and ≤.10 for the SRMR were considered acceptable (Hooper, Coughlan, & Mullen, 2008). Two reliability indices were calculated for each CSBD symptoms’ subscale: Cronbach’s alpha (α) and McDonald’s Omega (ω). The «userfriendlyscience» R package (Peters, 2014) was used to estimate these indices.

Third, we employed data clustering techniques to identify subgroups of participants with similar CSBD profiles. The six CSBD symptoms’ subscales confirmed during the previous analytic stage were used to estimate the presence of different CSBD profiles. As recommended (Hair, Black, & Babin, 2010; Henry, Tolan, & Gorman-Smith, 2005), this aim was addressed by combining hierarchical and non-hierarchical clustering strategies and confirming the accuracy of the resulting clusters through different strategies. At a first step, a hierarchical cluster analysis was conducted (Ward’s method, Euclidian distance measurement) to propose a tentative estimation of the number of homogeneous clusters in the dataset on the basis of the agglomeration schedule and the dendogram. Then, the optimal number of CSBD profiles and the cluster membership were determined using a two-step cluster classification method. Two indices were used to assess the goodness of fit of the proposed cluster solution in comparison with competing models ranging from 1 to 10 clusters: the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC). Despite its simplicity, this “auto-cluster” procedure has demonstrated its superiority to other more complex estimation methods in determining the optimal number of clusters to be retained (Eshghi et al., 2011; Gelbard, Goldman, & Spiegler, 2007). To confirm the accuracy of this cluster solution, we applied the following strategies: (a) we re-analyzed the data from dataset 1 through k-means (specifying the number of clusters derived from previous analyses) and estimated the convergence between both methods (Fisher & Ransom, 1995); (2) we randomly split the sample from dataset 1 into two equal subsamples, analyzed each half separately and compared the solution (Michaud & Proulx, 2009); (3) we applied the same cluster solution in a completely independent database (sample 2); and (4) we tested the criterion-related validity of the cluster solution (i.e., if the resulting clusters differ in variables of interest in ways consistent with theory). The criterion-validity of proposed clusters was assessed by comparing scores on the six CSBD subscales (internal validity); additionally, external validity was explored by comparing clusters in relation to sociodemographic, sexual, and clinical indicators (SSS scores, time online for sexual purposes, etc.).

Ethics

The study procedures were carried out in accordance with the Declaration of Helsinki. The Institutional Review Board of the Jaume I University approved the study. Volunteer participants in the research were informed about the study aim and they provided informed consent.

Confirmatory Factor Analysis (CFA) of CSBD symptoms

In order to verify the psychometric goodness of fit of our theoretically-driven classification of CSBD symptoms (Table 1), a CFA was performed in both sample 1 and 2. Goodness of fit of two possible models was tested: a model where the six first-order factors (i.e., CSBD symptoms) were correlated (M1) and a model where these factors were grouped under a second order factor (M2). This second approach was in line with models proposing a unidimensional expression of CSBD symptoms (Graham, Walters, Harris, & Knight, 2016) and has received support by recent works on the factorial structure of a CSBD assessment scale (Castro-Calvo et al., 2018). As Table 2 shows, M1 obtained the best model fit in both sample 1 and 2. Factor loadings derived from the CFA are included as an additional content in appendices (Table A2 in the appendix).

Regarding internal consistency (Table 3), ordinal Cronbach’s α and McDonald’s ω for the majority of the CSBD subscales indicated an appropriate internal consistency (α and ω between .67–.89 in sample 1 and .68–.91 in sample 2).

Cluster formation

To identify subgroups of participants with similar CSBD profiles, we conducted a hierarchical cluster analysis in sample 1. The six CSBD subscales confirmed during the previous step were employed as clustering variables in this analysis. To ensure that these variables share a common metric, their scores were z-transformed. The hierarchical cluster analysis was performed using Ward’s method with Squared Euclidian distance measurement, revealing that the appropriate number of clusters to be considered was two. The subsequent two-step method as well as the analysis of the BIC and AIC values confirmed the same cluster solution. Cluster 1 (labeled “non-CSBD”) consisted of 1,421 participants (89.88%) displaying a low-CSBD risk profile; cluster 2 (“CSBD”) included 160 participants (10.12%) with a high-CSBD risk profile.

To confirm the accuracy of this two-cluster solution, we conducted three confirmation analyses. First, data from sample 1 were re-analyzed by using an alternative, non-hierarchical, cluster approach: k-means. Once performed, we compared cluster membership convergence between both solutions, finding that 100% of those participants originally included in the non-CSBD cluster and 86.3% of those assigned to the CSBD were categorized in the same cluster through this alternative approach. The second confirmation approach consisted in randomly split the sample from dataset 1 into two equal subsamples, analyze each half separately through the two-step method, and compare the accuracy of cluster membership assignment. The convergence through this method was even higher, with 98.4 and 100% of participants assigned to non-CSBD and CSBD clusters categorized in the original profiles. Finally, we replicated the initial clustering method in a totally independent sample (sample 2), obtaining once again the same advised two-cluster solution. In this case, non-CSBD cluster comprised 92.19% of the sample (n = 1,215) whereas the CSBD cluster included the other 7.81% (n = 103).

Analyses of the resulting clusters

The criterion-related validity of the two-cluster solution was tested by comparing participants on direct CSBD indicators (internal validity) as well as by analyzing the sociodemographic, sexual, and clinical profile of CSBD participants (external validity). As displayed in Table 4, participants in the CSBD cluster significant differs from non-CSBD participants in their scores on the six CSBD subscales, both in sample 1 and 2 (all the differences significant at P < 0.001 and large effect sizes). CSBD symptoms that better discriminated between both clusters were loss of control (d = 2.46 [sample 1]; d = 2.75 [sample 2]), neglect (d = 2.42; d = 2.07), and preoccupation (d = 2.32; d = 2.65). The proportion of participants scoring above the HBI, SCS, and SAST cut-offs ranged between 30.1 and 63.1% in the CSBD cluster, compared to 0.1–2.6% in the non-CSBD group.

Concerning external correlates (Table 5), CSBD participants were mostly males (69.4 and 72.8% in sample 1 and 2) and included a higher prevalence of heterosexual participants (82.5 and 66%). In sample 2, CSBD participants were younger than non-CSBD participants (d = 0.22) whereas in sample 1, the prevalence reporting having a steady partner was lower (V = 0.10). CSBD participants were more sexual sensation seekers (d = 1.02 [sample 1]; d = 0.90 [sample 2]), showed slightly increased erotophilic tendencies (d = 0.26 in sample 1), and displayed an increased online sexual activity. In particular, CSBD participants spent twice as long on the Internet for sexual purposes (d = 0.59; d = 0.45), scored significantly higher in a scale assessing excessive and problematic engagement in this behavior (ISST, d = 0.98; d = 1.32), and an important proportion answered affirmatively to questions related to severity perception (50% of respondents in sample 2 considered they spent too much time online for sexual purposes and 60% was worried about this behavior). Offline sexual behavior of CSBD participants in sample 1 was characterized by a higher number of sexual partners (d = 0.37), a higher frequency of sexual intercourse (V = 0.11), and an increased prevalence of different sexual behaviors. Offline sexual behavior of CSBD participants in sample 2 only differed from non-CSBD participants in the frequency of sexual intercourse (V = 0.10) and the prevalence of same-sex sexual intercourse (V = 0.07). Finally, CSBD participants in both samples showed greater levels of depression and anxiety than non-CSBD participants, as expressed by their increased scores in the BDI-II and STAI-state (d of 0.68 and 0.33 respectively) and the HADS-Depression and HADS-Anxiety (d of 0.78 and 0.85 respectively). On the contrary, CSBD participants displayed lower levels of self-esteem (d of 0.35 in sample 1 and 0.55 in sample 2).

Limitations and future directions

Despite a number of interesting and novel findings, this study was limited in different ways. First, this research is correlational and therefore, do not address whether CSBD determines the emergence of the sexual and clinical profile typically observed in this condition or, on the contrary, the presence of certain previous psychological configurations (e.g., high erotophilia, sexual sensation seeking, or emotional problems) increases the vulnerability to develop CSBD. Second, the occurrence of CSBD reported in the study may be biased (inflated) due to our sampling approach. The first study was advertised as a sexuality survey; therefore, people with especial interest in sex (more prone to suffer from CSBD) may be overrepresented. Similarly, participants in the second study were recruited through the Internet, advertising the study as a sexuality survey. Additionally, the survey was accessible under search terms such as “sexual addiction”, thus increasing the probability that people experiencing CSBD symptoms accessed the survey.

Furthermore, CSBD profile was determined through a novel composite index derived from well-stablished self-report measures. This index was designed according to the most relevant and reliable criteria to identify CSBD (Kafka, 2010; Kraus et al., 2018; Wéry & Billieux, 2017). However, even when self-reports are considered as a well-meaning first approach for the screening of CSBD, its diagnosis actually requires a more in-depth assessment of the nature and context of individual’s sexual problems. For that reason, instead of (or in combination with) self-report measures, the use of structured or semi-structured clinical interviews focused on excessive and uncontrolled sexual behavior (e.g., the HD Diagnostic Clinical Interview [HD-DCI]) are usually advised for the appropriate diagnosis of CSBD (Womack et al., 2013). Thus, future research should consider the inclusion of a more in-depth exploration of the presence and severity of CSBD through more reliable assessment procedures (e.g., that followed in the DSM-5 field trial for hypersexual disorder) (Reid et al., 2012).