Abstract
Hormonal factors are important in multifactorial theories of sexual offending. The relationship between hormones and aggression in nonhumans is well established, but the putative effect in humans is more complex, and the direction of the effect is usually unclear. In this study, a large sample (N = 771) of adult male sex offenders was assessed between 1982 and 1996. Gonadotrophic (follicle-stimulating hormone and luteinizing hormone) and androgen hormone (total and free testosterone; T) levels were assessed at Time 1, along with indicators of sex drive and hostility. Individuals were observed up to 20 years in the community, with an average time at risk of 10.9 years (SD 4.6). Gonadotrophic hormones correlated positively with self-reported hostility and were better predictors of recidivism than was T (area under the curve (AUC), 0.58–0.63). Self-reported hostility emerged as a partial mediator of this relationship between gonadotrophic hormones and recidivism. These results point to a potentially new area of investigation for hormones and sexual aggression.
Neurobiological factors are considered important in multifactorial theories of sexual offending, and androgenic hormones have been specifically implicated in the elicitation of both sexual and aggressive behaviors.1,2 Testosterone (T) is the primary circulating androgen in men. Its production is initiated when gonadotrophin-releasing hormone is secreted from the hypothalamus. T is mostly bound to plasma proteins in the blood, with approximately two percent being in an unbound (free) form. It has been suggested that the free form of T is physiologically the most active,3 but the bound portion may also have effects.4 In males, T levels rise in utero and again dramatically at puberty. T has both organizational and activational effects on physical development, mood, and behavior in males.5
Androgens and Aggression
Androgens such as T are related to male aggression across vertebrate species.6,–,8 Animal research studies have consistently shown that aggression appears at the onset of gonadal puberty,7,9 is reduced by castration, and can be restored by treatment with exogenous androgens.10,11 The ethological evidence suggests that androgens influence aggression in humans.
The relationship between androgens and sexual activity and aggression in humans is more complex than that demonstrated in vertebrate species, because of cognitive, emotional, social, and contextual factors.7 In a meta-analysis,12 the authors examined this question by reviewing 45 studies of human participants (total n = 9, 760); the results demonstrated a correlation of 0.14 between T and aggression. A reanalysis of these data after correcting for methodological problems, such as the inclusion of overlapping studies, resulted in a lesser correlation,13 but a further analysis by the original authors, correcting for new errors in the reanalysis found that the effect was retained (r = 0.13).14
Androgens and Sexual Behavior
T is a major hormone associated with sexual motivation and behavior.5 Low levels of T can result from hypogonadism (primary or secondary) and are associated with decreased sexual interest and function. T replacement therapy can result in demonstrable improvement in the symptoms of hypogonadism.15 Experimental research with humans and other mammalian species shows that surgical castration or antiandrogen drug treatment reduces sexual interest and behavior, but does not necessarily negate the ability to engage in sexual activities.16
Androgens and Sexual Offending
Given the relationships between androgens and both aggressive and sexual behavior, many comprehensive theories of sexual offending have incorporated hormonal factors.17 However, there is surprisingly little evidence of this putative causal role.18 The results of a castration and antiandrogen study suggest that reducing androgens can decrease sexual recidivism among identified offenders, but it was not a randomized clinical trial.16 Several studies, despite their methodological limitations, have shown an effect of antiandrogen treatment on libido and sexual behavior.19,–,22
Several studies have reported that violent sex offenders have higher levels of androgens than do nonviolent comparison groups23,–,25 and that levels correlate positively with both prior violence and the severity of sexual aggression in sex offenders.25,–,27 A limitation of these studies, however, is that the direction of any effect is not clear (e.g., whether testosterone preceded or followed sexual aggression). An exception is a recent study, in which Studer et al.4 found a modest but statistically significant association between serum T levels and subsequent sexual offending in a sample of 501 adult male sex offenders with a mean follow-up of 8.9 years.
It is unclear why testosterone levels can predict sexual offending years in the future. One possibility is that it increases aggression, including sexual aggression. Another possibility is that it increases the frequency of sexual behavior, which then raises the risk of sexual aggression (as a result of more opportunity, for example). One way of determining which explanation is more apt is to identify mediators of the association between testosterone and sexual offending. For example, finding that the relationship is better explained by hostility than by sex drive suggests that the testosterone-aggression link is more relevant, whereas findings emphasizing sex drive as a mediator suggest that the testosterone-sexual behavior link is more relevant.
Gonadotrophins
Testosterone blood levels have a circadian fluctuation, with the highest levels occurring in the early morning. In healthy men, low T results in an increase in hypothalamic gonadotropin-releasing hormone (GnRH) that in turn stimulates an increase in two gonadotropic hormones: follicle-stimulating hormone (FSH) and luteinizing hormone (LH). FSH and LH are peptide hormones that are produced in the anterior pituitary: FSH stimulates the Sertoli cells in the testes and is therefore associated with spermatogenesis and fertility, whereas LH stimulates the Leydig cells in the testes to produce T.
Although there has been considerable research on individual sex hormones in isolation, there have been comparatively fewer studies examining interactions between LH, FSH, and T. One study28 reported a spike in LH following a GnRH injection in men with pedophilia, but not in a group of men who did not have pedophilia. The finding was replicated in a study of men with pedophilia compared with a control group of men with no known paraphilic interests. Compared with the control group, the men with pedophilia showed greater increases in LH (but not FSH) than did the controls after injection of GnRH.29
There have been relatively few studies examining the relationships of FSH and LH with subsequent aggressive and sexual behavior. In an early investigation using a small sample of adult males (N = 28), Mendelson et al.30 found that violent men had higher LH levels than nonviolent men had, as shown in postmortem blood samples. The relationships found for androgens in previous research may have masked a more central role for LH or FSH in androgen regulation.
Present Study
The present study was designed to extend previous research concerning the relationship between gonadotrophic and androgen hormone levels and subsequent long-term sexual and violent recidivism in a large sample of adult male sex offenders. This study was based on a retrospective chart review of assessments conducted by the Sexual Behaviors Clinic. Distinctive features of this study include a large sample size; a long follow-up period; measurement of both total and free T, in addition to both primary pituitary gonadotrophins (LH and FSH); and examination of hostility and sex drive as potential mediators of any association between hormones and sexual or violent recidivism.
On the basis of previous findings, we hypothesized the following: T, FSH, and LH would correlate positively with the violence of the index offense and the intrusiveness of the index sexual offense, self-reported hostility, and sex drive; T, FSH, and LH would correlate positively with sexual reoffending, even many years in the future; and both hostility and sex drive would partially mediate the relationship between hormones and recidivism.
Method
Participants
The present sample was based on a retrospective chart review of adult males (18 years of age or older at the time of the index offense) who had been convicted of a sexual offense. Only participants who had hormone test results in their medical records were included in the present study.
Participants were 771 adult men who had been assessed between 1982 and 1996 at a university-affiliated forensic sexual behaviors clinic. The majority of the participants were referred via the criminal justice system (e.g., judge or defense, 66%), followed by their treating physician (18%); five percent were self-referred. In the total sample, 278 (36.1%) were classified as intrafamilial offenders against children, 190 (24.6%) were classified as extrafamilial offenders against children, 172 (22.3%) were exhibitionists, 72 (9.3%) were rapists of adult women, and 59 (7.7%) had heterogeneous victim types.
The average age of the sample was 36.7 years (SD 11.9; range, 18–78) and approximately 41 percent of the participants reported having ever been married or having lived in a common-law relationship. Twenty-six percent had charges or convictions for previous sexual offenses, 39 percent had previous violent (including sexual) offenses, and 50 percent had prior criminal offenses of any kind. The majority of the sample was assessed at pretrial (36%), followed by those assessed at presentencing (13%). Approximately 11 percent were on probation at the time of assessment. Approximately 93 percent of all participants assessed at the clinic consented to their results being used for research purposes. This research study was approved by the Research Ethics Board of the Royal Ottawa Health Care Group.
Measures
Biochemical Assay Methods
Fasting blood samples were generally drawn to assay hormone levels between 8 and 11 a.m., to limit the effect of diurnal variations in circulating T. Total or free T, FSH, and LH were measured using commercially available radioimmunoassay kits at an independent laboratory. All hormone levels were provided by the same laboratory, which has modified its method of estimating hormone levels over the years, according to standard laboratory protocols. In 1994, the clinic transitioned to measuring free T, rather than total T. Thirty-seven participants received tests measuring both free and total T; the remainder of the participants had only either free T (post-1994) or total T (pre-1994).
Hostility
The Buss-Durkee Hostility Inventory31 was developed to measure various forms of aggression and hostility and is composed of 75 true-false statements. The measure includes seven subscales (assault, indirect hostility, irritability, negativism, resentment, suspicion, and verbal hostility) and an overall measure of hostility. The measure has been shown to have adequate test-retest reliability for each subscale and overall score and good internal consistency, discriminant validity, and convergent validity.32,33 Among sex offenders, BDHI scores for rapists have been significantly higher than those for nonoffending control subjects.25,34
Index Offense Features
Offense features included two behaviorally anchored scales measuring the violence of the index offense and the intrusiveness of the index sexual offense. The violence of the index offense was rated on a 10-point scale: no force or violence, 0; threat of assault with no weapon, 1; threat of assault with a weapon, 2; minor injury with no weapon, 3; minor injury with a weapon, 4; severe beating with no weapon, 5; severe beating with a weapon, 6; potential homicide (the injuries could have caused death without successful medical intervention), 7; homicide, 8; and homicide with postmortem mutilation, 9. Intrusiveness was rated on a six-point scale: no sexual intrusiveness, 0; verbal threat, 1; attempt, 2; touching, 3; penetration, 4; and sexual assault with excessive violence, 5. These behaviorally anchored scales have shown adequate discriminant validity and predictive validity in previous investigations of sex offenders.35,36 Inter-rater reliability for the behaviorally anchored scales was not available, as offense features were rated only by the evaluating psychiatrist.
Sex Drive
Sex drive was assessed with the Derogatis Sexual Functioning Inventory (DSFI) and a quantifiable index of sexual frequency (total sexual outlet (TSO), described later). The DSFI37 is a self-report measure with 10 subscales assessing different dimensions of sexual functioning. Among these scales, sex drive is a composite summary of sexual interest that is expressed across five behavioral domains: sexual intercourse, masturbation, kissing and petting, sexual fantasy, and ideal frequency of sexual intercourse. The scale has shown adequate internal consistency and test-retest reliability.37,38 Tang et al.39 reported marginal internal consistency coefficients for the sex drive subscale (alpha = 0.61).
TSO is defined as the number of orgasms achieved through any combination of methods (e.g., intercourse, masturbation) during a specific week.40 Several large-scale epidemiological studies have utilized TSO as a behavioral indicator of sex drive.41,42
Recidivism
Recidivism data were obtained in 2002 from a national database (CPIC; Canadian Police Information Centre) of criminal arrests and convictions maintained by the Royal Canadian Mounted Police. Recidivism was defined as a new offense (charge or conviction) that occurred after the index conviction, regardless of when these offenses occurred during the follow-up period. In contrast, analyses involving fixed follow-up periods (see results below) included only the first incident of recidivism, rather than all incidents of recidivism.
The follow-up period and opportunity to re-offend began at the latest of three possible dates pertaining to the index offense: date of conviction, date of assessment, or date of release if incarcerated. Eleven participants died during the follow-up period; their opportunity to reoffend ended on the day of death, if this date was reported in the CPIC records.
Recidivism outcomes included sexual recidivism, defined as any charge or conviction for a sexual offense, whether it involved physical contact with a victim (sexual assault) or not (indecent exposure, i.e., exhibitionism), and violent (including sexual) recidivism, defined as any charge or conviction for a non-sexually violent or sexual offense. The categories are not mutually exclusive and were statistically dependent (ϕ = 0.75), such that all participants who sexually reoffended were coded as violent recidivists, and most (63%) individuals who reoffended violently were coded as sexual recidivists.
This method of coding recidivism is consistent with some prior recidivism studies.32,43 Sexual recidivism includes noncontact offenses that may have been a prelude to contact sexual offending (e.g., the Canadian Criminal Code includes an offense (invitation to sexual touching) that results from approaching a child sexually, even if no physical contact has taken place); violent recidivism captures all violent offenses, including sexual offenses that were pleaded down to nonsexually violent charges (e.g., an attempted rape that resulted in a conviction for assault); or sexually motivated offenses that resulted in nonsexually violent charges (e.g., sexually motivated homicide that was charged as first-degree murder44).
Statistical Analyses
Several effect size indicators were reported in the present investigation. The simplest indicator representing the magnitude of effect in the present study was a correlation coefficient (Pearson's r). It is generally considered that correlation values of 0.1, 0.3, and 0.5, represent small, medium, and large effects, respectively. Post hoc comparisons on the continuous variables utilized Cohen's d; corresponding values for small, medium, and large effects are 0.2, 0.5, and 0.8.45
The area under the curve (AUC) of the receiver operating characteristic (ROC) was used to examine the predictive accuracy of the hormonal variables, as this statistic is less affected by recidivism base rates or selection ratios.46 AUC values, which can range from 0 to 1, can be interpreted as the probability that a randomly selected recidivist has higher hormonal levels than a randomly selected nonrecidivist. A value of 1 represents perfect prediction, while a value of 0.5 indicates chance prediction. For descriptive purposes, minimum AUC values of 0.56, 0.64, and 0.71 are described as small, medium, and large, respectively.47 To detect significant differences between two AUC values, we determined the critical ratio z, the ratio of a difference score to the standard error of the difference score, using formulas provided by Hanley and McNeil.48
Several mediation analyses were conducted with LH and FSH as independent variables; sexual recidivism as well as violent (including sexual) recidivism served as the dependent variables. Initially, three potential mediators were included, represented by the measures for sex drive, total sexual outlet, and self-reported hostility. The procedure used for the mediation analyses is based on the method described by Baron and Kenny.49 However, while the regression equations used are based on Baron and Kenny, modifications and additional equations were used to account for multiple mediators, and transformations were used, since recidivism was treated as a dichotomous variable. The use of transformations is discussed in MacKinnon and Dwyer50 and Winship and Mare.51 An example of SAS (Statistical Analysis Software) code for a three-mediator model given by Taborga et al.52 was studied before the writing of the code used here, and all regressions were performed in SAS. The presence or absence of mediation was assessed by application of the criteria given by Baron and Kenny.49 Point estimates of the indirect effect were obtained by using the modified product-of-coefficients method discussed in MacKinnon and Dwyer,50 and corresponding asymmetric 95 percent confidence intervals were obtained using the program PRODCLIN.53 This procedure and the resulting estimates are described as the M-test of indirect effects by Williams and MacKinnon.54
Results
The follow-up period ranged from 1 to 20 years, with a mean of 10.9 years (SD 4.6). The percentage of men who reoffended sexually and violently over the duration of the follow-up period was 17.8 and 28 percent, respectively.
The mean levels of total T and free T in the sample were 21.08 nmol/L (SD 7.17) and 50.48 pmol/L (SD 17.38), respectively. The mean levels of FSH and LH were 8.19 IU/L (SD 6.95) and 9.65 IU/L (SD 6.99). All of these values are within normal limits.52 Total T, free T, and FSH levels did not distinguish any type of sex offender. However, there was a significant main effect for LH level and type of sex offender (F(4,714) = 2.58, p < .05). Tukey's least significant difference post hoc analyses revealed that intrafamilial child molesters had lower LH levels than did extrafamilial child molesters (d = 0.21; 95% CI, (0.02 to 0.40) and exhibitionists (d = 0.28; 95% CI, 0.08 to 0.48).
Pearson correlation coefficients were calculated to examine the relationships among hormone levels, self-reported hostility and aggression, the intrusiveness of the index sexual offense, the violence of the index offense, and indicators of sex drive. Correlation coefficients are displayed in Table 1; 95 percent confidence limits for r were determined using Fisher's r–z transformation and are shown in the text in parentheses. Only correlations significant at the p < .01 level are reported in the text to correct for Type I errors resulting from numerous associations being examined in the analysis.
Intercorrelation Matrix for Hormone Levels, Hostility/Aggression, and Hypersexuality
Total T correlated significantly and positively with LH (0.18, 0.32), and FSH correlated significantly and positively with LH (0.62, 0.70) and self-reported hostility (0.05, 0.19). LH associated positively with self-reported hostility (0.13, 0.27), the violence of the index offense (0.05, 0.21), and the intrusiveness of the index sexual offense (0.05, 0.21). The violence of the index offense associated positively and significantly with the intrusiveness of the index sexual offense (0.30, 0.44). Self-reported hostility associated positively with the violence of the index offense (0.09, 0.24). Finally, with regard to hypersexuality, the DSFI positively and significantly correlated with TSO (0.33, 0.46), self-reported hostility (0.16, 0.30), and the violence of the index offense (0.06, 0.22). TSO was also significantly associated with self-reported hostility (0.12, 0.26).
Predictive Validity
Table 2 presents the AUC values and 95 percent confidence intervals for the hormone levels in relation to sexual and violent (including sexual) recidivism. Results indicated that LH and FSH were significantly associated with recidivism outcomes, whereas total T and free T were unrelated to recidivism. T levels were unrelated to recidivism, even after we categorized individuals into comparative degrees of low versus high androgen levels.
Predictive Accuracy of Gonadotrophic and Androgen Hormone Levels
When comparing ROC indices using the formula provided by Hanley and McNeil,48 LH was significantly more accurate than total T in predicting sexual recidivism (n = 622; z = 3.47) and violent recidivism (n = 622; z = 3.46). LH was also more accurate than FSH in predicting violent recidivism (n = 681; z = 2.87). Finally, FSH was more accurate in predicting sexual recidivism when compared with total T (n = 607; z = 2.20).
It has been suggested that clinicians may be particularly concerned with assessing risk within relatively short, set periods.55,56 Predictive accuracy can also be evaluated by using fixed follow-up periods; this method creates an equal length of follow-up for each offender. We analyzed recidivism data as it pertains to two- and five-year fixed follow-up periods. We considered only offenders who had at least two years or five years of opportunity to reoffend (depending on the specific analysis), with recidivists being those individuals who reoffended within that specified period. Individuals who reoffended after the relevant interval were counted as nonrecidivists.
When a fixed two-year follow-up period was used, the AUC value for LH in predicting violent (including sexual) recidivism was slightly smaller, but remained statistically significant (n = 662; ROC = 0.61; 95% CI, 0.53–0.69). No other significant associations were evident. When a fixed five-year follow-up period was used, the AUC for LH in predicting sexual recidivism was slightly higher than the rate produced when an open-ended time frame was used and it was statistically significant (n = 663; ROC = 0.66; 95% CI, 0.59–0.73).
Putative Mediators Between Hormones and Recidivism
We evaluated whether sex drive, TSO, and self-reported hostility mediated the effects of LH and FSH on long-term recidivism. With regard to violent (including sexual) recidivism, we found that the relationship of LH with recidivism was mediated by self-reported hostility. Results from a logistic regression showed that LH was a significant predictor of violent recidivism (p < .0001). Next, in OLS (ordinary least squares) regressions, LH was significant in predicting total sexual outlet (p = .04) and self-reported hostility (p < .0001), but not sex drive (p = .62). Finally, in a logistic regression with the explanatory and all mediating variables, the only significant potential mediator was self-reported hostility (p < .0001); also, as required with potential mediators in the model, the absolute value of the estimated effect of LH decreased, from 0.047 to 0.038. Concluding that self-reported hostility was the only mediator, regression analyses were conducted again with only self-reported hostility included as a mediator. Using these results yielded a point estimate of 0.0065 and a 95 percent confidence interval of 0.0033 to 0.0104 for the indirect effect of LH.
In a similar analysis for FSH, results from a logistic regression showed that FSH approached significance in the prediction of violent recidivism (p = .0507). In OLS regressions, FSH was not significant in predicting sex drive (p = .48) or total sexual outlet (p = .24), but it was significant in predicting self-reported hostility (p = .003). In the final logistic regression, self-reported hostility was the only potential mediator that was significant (p < .0001), and the FSH coefficient estimate for this regression was .016, compared with 0.022 for the regression with FSH only. Again, we concluded that self-reported hostility was the only mediator, and an appropriate second set of regression equations was obtained. From the results, the point estimate of the indirect effect size was 0.0043 and the 95 percent confidence interval produced by PRODCLIN was 0.0016 to 0.0077. FSH was no longer significant in the model that included self-reported hostility (p = .151). The terms perfect and complete mediation are used in the literature49 to refer to situations in which controlling for the mediator results in loss of significance of the effect of the explanatory variables. However, given the borderline p value for FSH in the initial regression of violent recidivism, the interpretation of the change observed here is less clear.
Finally, we were interested in examining the potential mediators of sexual recidivism. We found that the relationship of LH with sexual recidivism was also mediated by self-reported hostility. Results of a logistic regression showed that LH was a significant predictor of sexual recidivism (p = .0002). Next, in OLS regression, LH was significant in predicting the potential mediator (p < .0001). Finally, in a logistic regression of sexual recidivism against both LH and self-reported hostility, the latter measure was significant (p = .0052); also, LH remained significant (p = .0025) and, as required, the absolute value of the coefficient estimate for LH decreased, from 0.0453 to 0.039. Thus, self-reported hostility, as captured by the BDHI, is a partial mediator of the effect of LH on sexual recidivism. The point estimate of the indirect effect size was 0.0029, and the 95 percent confidence interval obtained with PRODCLIN was 0.0008 to 0.0053. Application of the M-test indicates that the mediated effect was significant.
In similar analysis with FSH as the independent variable, results of a logistic regression showed that this hormone was a significant predictor of sexual recidivism (p = .0324). Also, in an OLS regression of self-reported hostility on FSH, FSH was significant in predicting the potential mediator (p = .0011). Finally, in a logistic regression of sexual recidivism against both FSH and the potential mediator, self-reported hostility was significant (p = .0014); FSH was no longer significant (p = .0693), and the absolute value of the FSH coefficient estimate decreased from 0.0259 to 0.0232, with adjustment for the mediator. These results indicate the presence of complete mediation by self-reported hostility of the effect of FSH on sexual recidivism. The point estimate of the indirect effect size was 0.0031, and the 95 percent confidence interval produced by PRODCLIN was 0.0009 to 0.0061. Again, application of the M-test indicates that the mediated effect is significant.
Discussion
In the present study, we were primarily interested in the associations of androgen and gonadotrophin levels with self-reported hostility and sex drive, offense features, and long-term recidivism in a sample of male sex offenders. Our study was unique, in that we assessed several hormones and putative mediators and observed the individuals for up to 20 years after release into the community. This study represents, to our knowledge, one of the longest follow-up studies of sex offenders in the hormones literature.
We first hypothesized that androgens, particularly testosterone, would correlate with sex drive, hostility, the violence of the index offense, and the intrusiveness of the index sexual offense. Of the sex hormones assayed in this study, only free T exhibited significant associations (with the violence of the index offense and with self-reported sex drive) but these effect sizes failed to meet our more conservative approach to account for Type I error. Nevertheless, these findings suggest that total T is not sufficiently sensitive, which is consistent with the view that most total T is biologically inactive. LH and FSH, in contrast, correlated significantly and positively with self-reported hostility, and LH was associated with both the violence of the index offense and the intrusiveness of the index sexual offense.
Given the fact that androgens correlate with subsequent recidivism in sex offenders,4 we also hypothesized that androgen and gonadotrophin levels would correlate positively with long-term sexual and violent reoffending in our sample. This hypothesis was partially supported: only FSH and LH, but not T, were associated with long-term recidivism. Although some studies have demonstrated a positive association between peripheral hormones and sexual aggression,30 most studies4 have shown that T is a more sensitive marker for subsequent aggression.
It was surprising that the gonadotrophins assayed in the present sample were better predictors of long-term sexual or violent recidivism than were the androgens. It is possible that, because LH has a narrower normal range and is a precursor hormone, it may be a more sensitive marker of neuroendocrine functioning than T. This does not explain why T is usually found to be a more sensitive marker for subsequent aggression. Another possibility is that some men have a breakdown in the mechanisms that regulate LH levels. This phenomenon most likely results in a failure of the hypothalamic pituitary axis to downregulate production of LH when T levels rise, and it may cause sex offenders to show elevated LH and T levels or to show elevated LH levels, independent of their T levels. The results of this study are consistent with the second interpretation. Studies to replicate this finding and to further investigate the reasons for the phenomenon are recommended.
We also demonstrated that self-reported hostility was an important mediator between LH (and possibly FSH) and sexual and violent recidivism, suggesting that the link between hormones and aggression is more germane than the link between hormones and sexual behavior. Comprehensive theories of sexual offending2,57 include numerous distal and proximal factors that interact in both a cumulative and synergistic manner to increase the likelihood of sexual aggression. The confluence model57 highlights the role of hostility and other negative aspects of masculinity as partial contributors to sexual aggression against peers or adults.57 Indeed, the fact that the sex hormone levels (total T, free T, LH, and FSH) in the present study were all within normal limits supports the multifactorial nature of sexual aggression. Our results do not support the idea that abnormal hormone levels explain sexual offending; instead, relatively high levels of LH and FSH must interact with other factors to increase the likelihood of sexual aggression.
Limitations
A major limitation of this study is its retrospective design. This drawback is typical of sex offender studies in which long-term recidivism is one of the primary outcome measures. We could not control for extraneous factors, such as the effects of psychological or pharmacological treatment during the follow-up period as well as we might have been able to in a prospective design. Another problem with our retrospective design is that we were unable to account for hormone levels or our putative mediators after the initial assessment (i.e., at any point during the follow-up period). A second limitation pertains to the lack of a control group with which to compare our samples' hormone levels. We were therefore unable to compare the levels obtained in this sample with clinically meaningful groups, such as community participants and other offender groups (e.g., nonsexually violent offenders). The laboratory results were reported relative to test norms, however. Finally, there was a substantial portion of missing data across the variables that primarily resulted from the lengthy period over which assessments were conducted, the transition of the clinic to measuring free T during the study period, and the retrospective nature of the study.
Future Directions
The results of this study suggest that T (total or free) is not related to risk of sexual reoffense. In contrast, LH correlated significantly with sexual and violent recidivism. This result requires replication, however, as the relationship was relatively small, and this is the first study we are aware of to show that LH rather than T can predict long-term recidivism. If confirmed, this finding suggests a new area of investigation in the neurobiology of sexual offending. One possibility is that sexual offending, at least among some individuals, is related to disruption in the modulation of LH or the interactions between LH and T. This hypothesis could be further explored through clinical laboratory studies in which variation in both LH and T is tracked prospectively. It would be valuable for our theoretical explanations of sexual offending if this work were integrated with animal and human work on the effects of LH on aggressive and sexual behavior. The animal literature is much less clear about the role of LH in these behavioral domains, compared with what is known about T.
Our main finding may also be relevant to understanding the effects of different drugs on sexual offending. Cyproterone acetate (Androcur) blocks intracellular testosterone throughout the body's receptors and thereby decreases testosterone's effects (but with feedback effects on LH secretion), as well as reducing total testosterone. Leuprolide acetate (Lupron) is a GnRH agonist that paradoxically inhibits LH and FSH release and thereby reduces T. Perhaps the key target is LH production rather than T production, and thus drugs that selectively target LH may produce larger effects than those that selectively target T.
An interesting question to examine is whether hormone levels can add to the moderate predictive accuracy provided by existing actuarial risk measures for sex offenders. Seto58 suggested that only information from new risk factors or methods different from traditional reviews of personal history can add to the prediction provided by established measures. Although the correlation between LH and recidivism is small, it may still represent unique predictive variance because it is a novel risk factor obtained through a novel method (laboratory assay). We were not able to test this idea in the current study because we did not have actuarial risk scores for all the participants. This may be an interesting avenue for future research.
Acknowledgments
We thank Don Grubin and Steve Hucker for helpful comments on an earlier version of the manuscript, Tim Ramsay and William Petrcich at the Ottawa Hospital Research Institute for statistical consultation, and Susan Curry for assistance with data collection.
Footnotes
Disclosures of financial or other potential conflicts of interest: None.
- © 2012 American Academy of Psychiatry and the Law
