Relative influence of assigned gender and androgen exposure 2
Running head: RELATIVE INFLUENCE OF ASSIGNED GENDER AND ANDROGEN EXPOSURE
The relative influence of birth-assigned gender and prenatal androgen exposure on adult gender identity in intersex and related conditions.
Relative influence of assigned gender and androgen exposure 2
Abstract
While most contemporary authors agree that both birth-assigned gender and prenatal androgen exposure play a role in adult gender identity development in individuals with intersex and related conditions, little is known about the relative influence of each of these variables. In this research, data were collected from individual cases over age 18 reported in the academic literature. The variables collected for each case were birth-assigned gender and whether this was delayed, prenatal androgen exposure (measured by degree of genital virilization), and adult gender identity. There was a no significant correlation between delayed gender assignment and later gender reassignment. Logistic regression models showed birth-assigned gender and prenatal androgen levels did not differ significantly in their prediction of adult gender identity. However, there were only data available from individuals who experienced male-typical and intermediary levels of prenatal hormone levels. Given these results, we recommend birth-assignment of gender should take into account prenatal androgen exposure. Infant genital surgery should be avoided because of the large proportion of cases that later reassigned their gender.
Relative influence of assigned gender and androgen exposure 2
The relative influence of assigned gender and prenatal androgen exposure on gender identity in intersex and related conditions.
Persons with intersex conditions have discordance between their sexual genotype (sex chromosomes) and phenotype (genital appearance) (Mazur, Colsman, & Sandberg, 2007). These individuals usually receive prenatal androgen levels that are intermediary between male-typical and female-typical levels, which lead to genitalia development that is intermediary between male and female typical formation. These individuals are given a male or female assignment at birth and sometimes given surgery to make their genitals conform to their assigned gender. However the disproportionately large number of persons with intersex conditions who undergo gender reassignment later in their life indicates that clinicians should also be aware of other influences, including prenatal androgen levels, on the gender that the intersex person will identify as in adulthood. Other individuals experience male-typical prenatal androgen levels but male typical genitalia is not formed due to cloacal exstrophy or penile agenesis, or the penis is accidently ablated (Meyer-Bahlburg, 2005). While these individuals do not fit the definition of intersex above, because many are given a female gender assignment at birth they are of interest to persons studying the effects of prenatal androgens and gender assignment.
While some authors have suggested that birth-assignment of persons with intersex conditions be sex-assigned based on the amount of prenatal androgen exposure (Diamond & Sigmundson, 1997), from reviewing more recent literature, Cohen-Kettenis (2005b) concluded that “in virtually all syndromes, sex of assignment appears to be the best predictor of adult gender identity” (p. 29). The goal of this article is to perform an analysis of data collected from literature that has followed-up persons with intersex and related conditions in adulthood and calculate the relative effects of prenatal hormones and gender assignment on adult gender identity. We hypothesize that birth-assigned gender will be a stronger predictor of adult gender identity than prenatal androgen exposure.
Method
Data Collection
Data were collected from published academic articles that included gender identity outcomes in adulthood of persons with intersex and related conditions. Cases were taken from four recent reviews (Cohen-Kettenis, 2005a; Dessens, Slijper, & Drop, 2005; Mazur, 2005; Meyer-Bahlburg, 2005), although often the original article had to be viewed to obtain sufficient information for the analysis, articles that have been published subsequently and articles that were not included in these reviews (see Table 1).
From these articles eight variables were obtained. Assigned gender was coded as 1 for female and 5 for male (5 was used here because it aligned with the male score for other variables). Puberty virilisation was coded as 0 for no and 1 for yes. Adult gender identity was coded as 1 for female, 3 if the case was reported as “gender dysphoric” but had not taken steps to reassign their gender, and 5 for male. A degree of gender change score was also calculated as the difference between assigned gender score and adult gender identity score. Time before gender assignment and year of publication were also collected. Degree of prenatal androgens was on a scale of 1 to 5, with 1 representing female typical prenatal androgens and 5 representing male typical. For cases of congenital adrenal hyperplasia (CAH), degree of prenatal androgens was coded as the Prader score (a scale of 1 to 5 ref) where this was reported, and where this was not reported it was coded as a score of 3. For cases of partial androgen insensitivity (PAIS) and mixed gonadal dysgenesis (MGD) a score of 3 was given as these individuals experience a prenatal androgen level that is intermediary between male-typical and female-typical. A score of 4 was given for patients with micropenis since the penis is fully formed in these cases it is probable that prenatal androgen levels are more towards the male-typical end. For cases of 5α-reductase deficiency (5α-RD), 17β-hydroxysteroid dehydrogenase deficiency (17β-HDD), penile agenesis, penile ablation, and cloacal and classical exstrophy (all genetic males) a score of 5 was given as they experience male-typical prenatal androgen levels.
[Insert Table 1 about here]
Cases were only included in the analysis if they were reported as over 18 and their assigned gender score differed from their prenatal androgen score. This meant that all cases with complete androgen insensitivity were excluded because they were exposed to (or at least their bodies responded to) female typical prenatal androgen levels and all of these cases were assigned female at birth. Those cases with a prenatal androgen score of 5 that were assigned male were also excluded.
Results
A total of 491 cases were included. Table 2 outlines the clinical diagnosis of the cases. Age data could be collected for 278 cases and time before gender assignment could be collected for 292 cases. Degree of gender change was correlated with reported age r = .26, p < .001, with year of publication ρ = -.36, p < .001, and not significantly correlated with time before gender assignment ρ = .02.
[Insert Table 2 about here]
Data were entered into an ordinal regression model with adult gender identity (male, gender dysphoric, or female) as the dependent variable and birth-assigned gender and prenatal androgen level as independent variables. However this model did not meet the proportional odds assumption (χ²4 = 13.41, p = .01), so those cases categorized as gender dysphoric (n = 14) had their adult gender identity recoded to that of their birth-assigned sex and a logistic regression was conducted instead. The results of this analysis (Model 1) are given in Table 3 and they show that those cases assigned female were around 86 times more likely to develop a female gender identity than those assigned male. Those cases that had a prenatal androgen score of 3 (half way between male-typical and female-typical prenatal androgen) were around 22 times more likely to develop a female gender identity than those with male-typical prenatal androgen levels (score of 5). Cases with prenatal androgen scores of 4 and 2 were also significantly more likely to develop a female gender identity than those with a score of 5, and as expected, each level of prenatal androgen score decrease resulted in an increase in the odds ratio of developing a female gender identity. However, because there were only five cases with a prenatal androgen score of 2, this variable had a large confidence interval. In this analysis, the deviance goodness of fit score was marginally significant (χ²2 = 6.08, p = .05); however, the Pearson goodness of fit score was not significant (χ²2 = 4.13, p = .13). Pseudo r² scores were Cox and Snell .45, Nagelkerke .60, and McFadden .44.
To give more information about the relative effects of birth-assigned gender and prenatal androgen exposure, these were entered into the logistic regression as continuous variables. Results of this regression analysis (Model 2) are given in Table 4. In this model the logistic regression coefficients for birth-assigned gender and prenatal androgen did not differ significantly and the deviance goodness of fit score was also no longer significant (χ²4 = 6.60, p = .16).
[Insert Tables 3 and 4 about here]
When cases with 5α-RD and 17β-HDD were excluded from the analysis the model met the proportional odds assumption (χ²4 = 5.17, p = .27), so analysis continued with an ordinal regression. The results of this analysis (Model 3) are given in Table 5 and they show that those cases assigned female were around 74 times more likely to develop a female gender identity (as opposed to a gender dysphoric or male gender identity) than those assigned male. Those cases that had a prenatal androgen score of 3 were around five times more likely to develop a female gender identity than those with male-typical prenatal androgen levels. Cases with prenatal androgen scores of 2 were also significantly more likely to develop a female gender identity than those with a score of 5, and as expected, each level of prenatal androgen score decrease resulted in an increase in the odds ratio of developing a female gender identity. In this analysis, the deviance goodness of fit score was marginally significant (χ²8 = 15.29, p = .05), however the Pearson goodness of fit score was not significant (χ²8 = 11.45, p = .18). Pseudo r² scores were Cox and Snell .52, Nagelkerke .66, and McFadden .46. Results of ordinal regression analysis with the independent variables as continuous predictors (Model 4) are given in Table 6. In this model the logistic regression coefficients for birth-assigned gender and prenatal androgen did not differ significantly and the deviance goodness of fit score was also no longer significant (χ²10 = 15.92, p = .10).
[Insert Tables 5 and 6 about here]
Discussion
From the cases reported in the academic literature following up the gender identity individuals with intersex and related conditions in adulthood, we found no evidence to reject the null hypothesis that birth-assigned gender and prenatal androgen did not differ significantly. In Model 1 there was not evidence to suggest that even those with intermediary levels of prenatal androgen exposure (prenatal androgen score 3) had significantly greater odds of developing an adult female gender identity when birth-assigned gender was held constant. Results from Model 2 show that for a one unit increase in prenatal androgen level there was a 4.61 greater likelihood of a female gender identity developing when birth-assigned gender was held constant. The corresponding odds ratio for birth-assigned gender was 2.92 and it was not significantly different from the odds ratio for prenatal androgen. Although both of these variables were measured on a scale of 1 (female) to 5 (male), birth-assigned gender wasn’t measured in one unit intervals like prenatal androgens was. Therefore, this odds ratio should be thought of as the odds of developing a female gender identity when birth-assigned female split into what it would be if there were five birth-assigned gender categories instead of two to make it comparable to the prenatal androgen variable. Care should be taken, however, in directly comparing these two odds ratios because we only have sufficient data for prenatal androgen scores 5 (male-typical), 4, and 3 and birth-assigned gender scores 1 and 5. In effect we are comparing data from different levels of the same continuum. A significant limitation of this research is that we were not able to use cases that had female-typical prenatal androgen levels. However, given the finding that individuals with complete AIS almost invariably develop a female gender identity (Mazur, 2005) we have no reason to believe that the effect of prenatal androgens on adult gender identity is not linear, and if we had cases with prenatal androgen level 1 and more with prenatal androgen level two we would expect to see a continuation of the pattern that we had seen from prenatal androgen levels 3 to 5.
We included a separate analysis that excluded cases with 5α-RD and 17β-HDD (Models 3 and 4) because the majority of these cases are from non-Western societies where there is a cultural history of others with this condition and because these individuals experience substantial virilization at puberty – the fact that these individuals start to look more like males may influence their choosing to live in the male gender role. When these cases are excluded, prenatal androgen level has substantially less predictive power in the models. In Model 6, however, the odds ratios for these two variables still do not differ significantly from each other.
The positive correlation between gender change and age was expected given older cases have more time to undergo gender reassignment. The negative correlation between degree of gender change and year of publication was due to many 5α-RD and 17β-HDD cases (who often report gender reassignment) being reported in the 1970s and 1980s. When 5α-RD and 17β-HDD cases are excluded from this calculation the correlation becomes non-significant.