Clustering psychopathology in male anabolic–androgenic steroid users and nonusing weightlifters

Abstract Introduction Prior research has demonstrated that personality disorders and clinical psychiatric syndromes are common among users of anabolic–androgenic steroids (AAS). However, the prevalence, expression, and severity of psychopathology differ among AAS users and remain poorly understood. In this study, we examine the existence of potential clinically coherent psychopathology subgroups, using cluster procedures. Methods A sample of 118 male AAS users and 97 weightlifting nonusers was assessed using the Millon Clinical Multiaxial Inventory‐III (MCMI‐III), measuring personality disorders and clinical syndromes. Group differences in MCMI‐III scales were assessed using Wilcoxon‐Mann–Whitney tests and Fisher's exact test. Agglomerative hierarchical clustering was used to identify clusters based on MCMI‐III scale scores from the whole sample. Results AAS users displayed significantly higher scores on all personality disorder (except narcissistic) and clinical syndrome scales compared to nonusing weightlifters. The clustering analysis found four separate clusters with different levels and patterns of psychopathology. The “no psychopathology” cluster was most common among nonusing weightlifters, while the three other clusters were more common among AAS users: “severe multipathology,” “low multipathology,” and “mild externalizing.” The “severe multipathology” cluster was found almost exclusively among AAS users. AAS users also displayed the highest scores on drug and alcohol dependence syndromes. Conclusions AAS users in our sample demonstrated greater psychopathology than the nonusing weightlifters, with many exhibiting multipathology. This may pose a significant challenge to clinical care for AAS users, particularly as there appears to be significant variation in psychopathology in this population. Individual psychiatric profiles should be taken into consideration when providing treatment to this group. Significant Outcomes As a group, AAS users displayed markedly greater psychopathology than nonusing weightlifters. Multipathology was common among AAS users. Four different subgroups of personality profiles were identified with distinct patterns of pathology and severity. Limitations The cross‐sectional nature of the study precludes inferences about causality. The study is limited by possible selection bias, as participants choosing to be involved in research may not be entirely representative for the group as a whole. The study is vulnerable to information bias, as the results are based on self‐report measures and interviews.

• Four different subgroups of personality profiles were identified with distinct patterns of pathology and severity.

Limitations:
• The cross-sectional nature of the study precludes inferences about causality.
• The study is limited by possible selection bias, as participants choosing to be involved in research may not be entirely representative for the group as a whole.
• The study is vulnerable to information bias, as the results are based on self-report measures and interviews.

K E Y W O R D S
anabolic steroids, doping, hierarchical clustering, personality pathology, psychopathology INTRODUCTION Anabolic-androgenic steroids (AAS) are a family of drugs that includes the male hormone testosterone, together with hundreds of synthetic analogues of testosterone (Kicman, 2008;Pope, Wood, et al., 2014).
In addition, most studies have looked at single psychiatric diagnoses instead of also focusing on comorbidity, which may not fully capture the complex and overlapping nature of psychopathology among users of AAS. It is now well recognized that psychiatric diagnoses tend to occur co-morbidly, and that looking at dimensions or subtypes of psychopathology might be a better alternative than looking at individual diagnoses alone (Caspi et al., 2014). Thus, to better capture potential associations between AAS use and psychopathology, we sought to look for possible psychopathological profiles based on response patterns, as has previously been performed in patients with other types of substance-use disorders using cluster analyses (McMahon, 2008).
The purpose of the current study was to investigate personality pathology and specific psychiatric syndromes in a group of male AAS users and a comparison group of weightlifters reporting no use of AAS, using the MCMI-III. In particular, we sought to identify potential clinically coherent psychopathology subgroups within the dataset, using cluster procedures. We hypothesized that (1) AAS users as a group would display more severe psychopathology than weightlifting controls, and (2) cluster analysis would reveal different categories of psychopathology within the AAS using group. In particular, we hypothesized that we would find one cluster characterized by multipathology, one cluster with low or no psychopathology, and distinct clusters with elevated scores on either externalizing or internalizing traits. A group of 118 current or past male AAS users and a comparison   group of 97 nonusing male weightlifters completed the Millon Clinical Multiaxial Inventory-III (MCMI-III) as a part of a longitudinal study investigating brain, medical, and psychiatric effects associated with long-term use of AAS at Oslo University Hospital. The sample is partly overlapping with the one described in Bjørnebekk et al. (2021). The data collection was performed at two different time points from 2013 to 2015 and from 2017 to 2019. At both time points, the participants were recruited through posters and flyers distributed at selected gyms in Oslo, and through online forums and websites addressing heavy resistance training and/or AAS use. Prior to enrollment, all participants received an information brochure with a complete description of the study, and written informed consent was collected from all subjects.

Participants
The participants were compensated with 1000 Norwegian kroner at time point 1 and with 500 Norwegian kroner at time point 2. AAS use or nonuse was assessed among the entire sample from which our group of participants was drawn. Testing for AAS drug use was conducted at the WADA-accredited laboratory at Oslo University Hospital as described in detail previously . The criteria used to determine external androgen use were: (1) urine samples positive for synthetic testosterone compounds, and (2) a testosterone to epitestosterone ratio (T/E) greater than 15. In previous publications , our group found excellent agreement between these laboratory results and participants' self-report, with no controls testing positive for AAS. One AAS user was excluded based on the MCMI-III's validity exclusion criteria (Jankowski, 2002), leaving 117 AAS users and 97 nonusers for analysis. Seven of the AAS users were unable to come to Oslo for testing, and thus, performed only the assessment inventories. Consequently, information regarding the nature of their AAS use and training regimens is missing.

Instruments
The Millon Clinical Multiaxial Inventory-III (MCMI-III) is a 175-item, true-false, self-reported inventory for measuring personality disorders and identifying existing syndromes (Millon, 2006 , 1994). Each scale is scored using base rate (BR) scores, measuring the prevalence and severity of a psychological characteristic on a continuum, and is representative of the actual prevalence of the particular attribute in the psychiatric population. The scales range from BR score 1 to 115, where a score of 60 is the median score obtained by psychiatric patients (Jankowski, 2002).
A score greater than or equal to 75 on a scale indicates the pres-ence of a trait or syndrome (Millon, 2006). The disclosure index scale (scale x), one of the scales to detect response bias, measures the participant's response style, and is based on raw scores. Scores less than 34 on the disclosure index scale indicate excessive secretiveness, whereas scores greater than 178 on the same scale indicate that the participant is too self-revealing, and when MCMI-III is used in clinical samples, such scores would invalidate the test results (Jankowski, 2002). Seven AAS users and 17 nonusing weightlifters scored higher than 34 on the disclosure index scale, while two AAS users scored lower than 178.
However, as this was an exploratory study on a non-clinical sample, these participants were not excluded from our analyses.
We also administered a semistructured interview, specifically designed for this study, covering demographic data, weight training history, health-related information, and history of prescription drug use.
Also, the history and characteristics of AAS use were assessed with questions assessing age at onset of use, weekly AAS dose estimates (mg), years of use, and whether the AAS use was current or previous.

Statistical analyses
Group differences in demographic data were evaluated with independent-sample t-tests on continuous variables and chi-square tests for independence on categorical data using SPSS Statistics 26.
The remaining statistical analyses were performed using R version 4.0.3 (R Core Team, 2018). Comparisons between AAS users and nonusers on MCMI-III scales were assessed using Wilcoxon-Mann-Whitney tests to account for the non-normal distributions on these measures. Scores were dichotomized using the clinical cut-off of BR greater than or equal to 75 and compared between groups using Fisher's exact test to account for the small number of expected elevated scores, particularly in the group of nonusing weightlifters.
p-Values were adjusted using the Benjamini-Hochberg procedure to control for the false discovery rate. The same analyses were repeated to compare previous and current AAS users.

Agglomerative hierarchical clustering with Euclidean distances and
Ward's D2 method was used to identify clusters based on BR scores on all MCMI-III scales from the whole sample. The optimal number of clusters was determined using a combination of the NbClust package (Charrad et al., 2014), which calculates 25 indices of optimal cluster number, in addition to a within-cluster sum of squares scree plot.
Chi-square tests were used to compare the prevalence of cluster membership between AAS users and nonusing weightlifters, and effect sizes were calculated using the phi coefficient.

Sample characteristics
The groups did not significantly differ in age, but the nonusing weigtlifters had on average about 2 years more education than the AAS users (Table 1). Although the nonuser group reported significantly

Characteristics of AAS use
In the AAS user group, the mean age of onset of AAS use was 22.6 years ( Table 1). The mean cumulative lifetime duration of AAS use was 10.6 years. Using criteria proposed by Pope and Katz (1994), we categorized the users weekly doses into "low" (<300 mg of testosterone equivalent per week), "medium" (300-1000 mg/week), and "high" (>1000 mg/week). Among the 100 users for whom these data were available, we classified 11% as using low, 52% as using medium, and 37% as using high weekly doses.

MCMI-III-scales
The AAS group displayed significantly higher scores than the nonuser group on all MCMI-III-scales except for the narcissistic personality disorder scale ( Table 2). The antisocial, aggressive, passive-aggressive, self-defeating, avoidant, depressive, and schizotypal personality disorder traits, in that order according to the z-values, were the scales on which the AAS users differed most from nonusers, based on the median score. Among the clinical syndrome scales, the drug and alcohol dependency scales differed most strongly between AAS users and nonusers. Similar findings emerged when assessing the number of individuals scoring above the clinical cut-off of BR greater than or equal to 75 on the various scales (Supporting Information Table S1).

MCMI-III clusters
When performing the cluster analysis, we found that eight indices from the NbClust package suggested that two clusters were optimal, while seven indices suggested that four clusters were optimal (Charrad et al., Figure S1). The within-cluster sum of squares scree plot indicated that four clusters yielded significantly less within-cluster variation than the two-cluster solution (Supporting Information Figure S2). The decision to select four clusters was also retrospectively supported by the distinctive and clinically relevant characteristics of the clusters (Supporting Information Figure S3). The results of the agglomerative hierarchical clustering are presented in Table 3 and Figure 1, with an overview of groups of participants in each cluster presented in Table 4.

2014) (Supporting Information
The first cluster (no psychopathology) was characterized by low scores on all scales except for the narcissistic personality pattern.
The second cluster (mild externalizing) was characterized by moderately high scores on the narcissistic, antisocial, and sadistic personality pattern scales, together with the drug dependence syndrome  in this cluster also showed the highest mean score on the drug dependence and alcohol dependence syndrome scales. The fourth cluster (mild multipathology) was characterized by a high mean score on the depressive personality pattern scale, and moderately high scores on the schizoid, narcissistic, dependent, and antisocial personality pattern scales, as well as the somatoform and anxiety syndrome scales. About two thirds of the nonusing weightlifters, but relatively few AAS users, fell into the "no psychopathology" clus-ter. "Severe multipathology" was seen almost exclusively in the AAS users.

Characteristics of MCMI-III personality subgroups
An analysis comparing the characteristics of the participants in the previously described clusters (Supporting Information Table S2) found that participants in cluster 1 (no psychopathology) had significantly more education than the remaining groups. Additionally, participants in cluster 3 (severe multipathology) differed significantly in age at onset of AAS use as compared to participants in cluster 1 (no psychopathology) and 2 (mild externalizing), with a mean age at onset of AAS use of 18.18 in the former group. Apart from this difference, no other statistically TA B L E 3 Cluster analysis, mean base rate scores, and number of scales above the clinical cut-off (≥75) within each cluster  AAS, anabolic androgenic steroids F I G U R E 1 Cluster analysis heat map. Dark blue means a higher score on the scales. PTSD, post-traumatic stress disorder. significant differences were found between the groups with regard to age, total years used, or weekly dose estimate.

Current and previous AAS use
We possessed data on 104 of the 117 AAS users with regard to current versus previous AAS use. A total of 72 individuals reported that they were current users, and 32 reported only previous use. When comparing these two subgroups on the various MCMI-III measures, we found no significant differences on median scores (Supporting Information   Table S3) or on the proportion of individuals exhibiting elevated scores (BR ≥ 75) (Supporting Information Table S4) on any measure.

DISCUSSION
The findings from the present study using the MCMI-III demonstrated marked differences in psychopathology between the groups of 117 AAS-using and 97 nonusing weightlifters. AAS users showed significantly higher BR scores on all personality and syndrome scales, except for the narcissistic scale, with a much higher number of scale scores reaching the clinical cut-off (BR ≥ 75). The cluster analysis revealed four different subgroups of psychopathology profiles with distinct patterns of personality and syndrome scale scores. A comparison between individuals reporting current AAS use with those reporting previous AAS use showed no significant differences between these groups. This suggests that the MCMI-III was detecting stable traits in the AASusing population, rather than state-dependent effects associated with current AAS use.
Nearly one-third of the AAS sample, as opposed to about onetenth of the weightlifting controls, exhibited elevated scores on the depressive personality and/or anxiety syndrome scales, with the highest scores in the two multipathology clusters. This finding is consistent with previous studies documenting anxiety and mood disorders among AAS users (Amaral et al., 2022;Gestsdottir et al., 2021;Ip et al., 2011;Pope & Katz, 1994;. In particular, it is likely that hypogonadism precipitated by AAS withdrawal may cause depressive syndromes with attendant anxiety in some men Malone et al., 1995;Pope & Katz, 1988;Rasmussen et al., 2016).
Of note, a 30-year follow-up study of former elite male athletes in power sports suggests possible long-term effects of AAS use on mental health, in that former AAS users more often sought professional help for mental problems, such as depression and anxiety, compared to athletes with no history of AAS exposure (Lindqvist et al., 2013). Also of note, it has been reported that among AAS users who sought help to quit AAS use, most did so as a result of mental health problems, with depression and anxiety being the most prominent effects described (Havnes et al., 2019). It has also been reported that AAS users are more prone to suicide or suicidal ideation than nonusers (Gestsdottir et al., 2021;Patel et al., 2021;Petersson et al., 2006;Thiblin et al., 1999), and that this risk increases if personality disorders are present (Borjesson et al., 2020).
In our study, the most pronounced difference in personality pathology between AAS users and nonusers was the antisocial personality pattern. The findings of high scores on antisocial personality disorder have been presented on a part of this sample previously , where our research group found that antisocial personality traits could be an important mediator in the relationship between AAS use, aggression, and violence. In addition to antisocial personality disorder, sadistic, and borderline personality disorder were scales that differed markedly between AAS users and controls. These findings are consistent with previous studies of male AAS users (Borjesson et al., 2020;Cooper et al., 1996;Hallgren et al., 2015;Perry et al., 2003;Yates et al., 1990), and one study of female users (Scarth, Havnes et al., 2022).
Similar profiles have been found in individuals with substance-use disorders (Verheul, 2001), suggesting an overlap in personality factors between AAS users and users of other illicit substances.
A striking finding of our study was a large number of AAS users exhibiting multiple areas of psychopathology. About half of the AAS users had psychopathology profiles fitting into two multipathology clusters. This observation appears congruent with the description of a "general psychopathology" p-factor as described by Caspi et al. (2014), and the fact that severe disturbances tend to be comorbid (Caspi et al., 2014;Rosenstrom et al., 2019). Accumulating evidence suggests that the combination of externalizing and internalizing pathology poses a greater risk of life impairment, subsequent psychiatric disorders, criminal offenses, and other self-reported problems (Caspi et al., 2014;Sourander et al., 2007). These findings accord well with the results of the present study, where individuals in the severe multipathology cluster showed not only high scores on all scales, but also high-risk behaviors such as high drug dependence scores and low age at onset of AAS use.
Polypharmacy is common among AAS users (DiClemente, 2014; Dodge & Hoagland, 2011;DuRant et al., 1995;Pope et al., 2012;Pope & Katz, 1994;Sagoe et al., 2015;Skarberg et al., 2009), as was also the case in our sample, where the median BR score on "drug dependence" was 60 for the AAS users. The majority of AAS users scored below the cut-off, but 14.4% displayed elevated scores, with the most problematic substance use seen in the severe multipathology cluster. This finding resembles the findings of another recent study where ongoing use of narcotic agents and alcohol was more common in AAS users diagnosed with a personality disorder (Borjesson et al., 2020). It has also been reported that patients with substance-use disorders who also exhibited AAS dependence showed more severe personality pathology than substance-use disorder patients not using AAS (Scarth, Havnes et al., 2022). In studies of inmates  and of patients in treatment for substance-use disorders (Havnes, Jørstad, et al., 2020), individuals using AAS were found to exhibit more severe overall substance use than AAS nonusers. The direction of causality underlying these associations is complex, likely reflecting a combination of genetic liability, environmental, and drug exposure effects.
Last, as has also been found in other studies , it is worth noting that 22.2% of the AAS users had a personality profile fitting into the "no psychopathology" cluster, meaning that they had low scores overall, except for scores on the narcissistic personality pathology scale. Narcissism has previously been linked to appearanceand fitness-enhancing behaviors in general (Martinovic et al., 2022;Miller & Mesagno, 2014). The finding of no psychopathology in a group of the AAS users in our sample is expected, as it resembles several other studies finding that the majority of AAS users reported minimal psychiatric effects, or finding few significant psychopathological differences between weightlifting controls and AAS users as a whole (Bahrke et al., 1996;Moss et al., 1992;Tricker et al., 1996;Windfeld-Mathiasen et al., 2022). The differences that we observed in psychiatric morbidity both between AAS users and weightlifting controls, and within the group of AAS users, may be affected by many different factors linked to AAS use. For instance, AAS users experience more physical health problems compared to controls, which may contribute to increased psychopathology (Barnett et al., 2012;Horwitz et al., 2019). Interestingly, we found no significant differences among AAS users in the four clusters with respect to lifetime duration of AAS use, with even individuals in the "no psychopathology" cluster reporting a mean of more than 10 years of cumulative AAS exposure. A potential confounding variable in such comparisons may be level of education, as this, in turn, is often linked to socioeconomic status, and as lower education level is associated with an increased risk for psychotic and mood disorders (Kivimäki et al., 2020). In our study, the highest level of education was found in the weightlifting control group, and in the "no psychopathology" cluster, both of which had low psychopathology scores.

Limitations
The cross-sectional nature of the study precludes inferences about causality, and longitudinal studies will be required to better delineate the causal pathways involved in the psychopathology observed. The present study is also limited by possible selection bias, in that the AAS users and nonusers who chose to participate in the study may not have been entirely representative of their respective source populations in the community. The findings are also vulnerable to information bias, in that the results are based on interviews and self-report measures that could not be confirmed with objective measures. However, given the magnitude of the differences observed among the groups on the various measures, and the relatively large number of participating AAS users, it seems unlikely that the differences observed could be attributable purely to these sources of bias. It is also important to note that the MCMI-III was designed and standardized for use in clinical populations, but has also proven to be useful in research settings (Millon, 2006). Nevertheless, our findings of reported elevated scales in this research population should not be considered as diagnoses, but rather as general indicators of pathology.

CONCLUSIONS
In a study comparing long-term AAS users and nonusing weightlifters on the Millon Clinical Multiaxial Inventory, we found markedly greater psychopathology among AAS users than among weightlifting nonusers.
The cluster analysis revealed four different subgroups of personality profiles with distinct patterns of personality pathology and severity. These findings may help to inform health care providers about potential personality profiles of AAS users, to guide interventions for individual users. In particular, health care providers should be aware that a proportion of AAS users show complex personality pathology, characterized by severe internalizing and externalizing pathology and consequently a substantial need for mental health care services.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are not publicly available due to their sensitive nature, where our ethical approval prevents us from sharing data beyond named collaborators. However, upon reasonable request, we will allow necessary insight into the material.
Further inquiries can be directed to the corresponding author.

CONFLICT OF INTEREST STATEMENT
All authors declare that they have no conflict of interest.

PEER REVIEW
The peer review history for this article is available at https://publons. com/publon/10.1002/brb3.3040