Hormonal Target Validation based on Biochemical Shifts in Gender-Affirming Hormone Therapy

Hormonal Target Validation based on Biochemical Shifts in Gender-Affirming Hormone Therapy | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Hormonal Target Validation based on Biochemical Shifts in Gender-Affirming Hormone Therapy Jeroen Vervalcke, Lorenzo Marinelli, Dorte Glintborg, Louise Lehmann Christensen, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8473261/v1 This work is licensed under a CC BY 4.0 License Status: Under Revision Version 1 posted 5 You are reading this latest preprint version Abstract Background: Hormonal targets for gender-affirming hormone therapy (GAHT) remain largely unvalidated in gender-diverse individuals, with no clinical markers to assess dosing adequacy. We hypothesize that a biochemical shift towards the hormonal profile of the identified sex correlates with appropriate dosing. Methods: A data-subset from the National Health and Nutrition Examination Survey (NHANES), comprising 5433 assumed-cisgender individuals (47.2% male) aged <52 years, was used to train and validate a Random Forest model to predict sex-assigned-at-birth, incorporating platelet count, HDL%, SHBG, creatinine, and hematocrit. Being assigned-female-at-birth was weighted at 0.54. The model was subsequently run in 171 transgender men (TM) and 119 transgender women (TW) from the European Network for the Investigation of Gender Incongruence (ENIGI) Ghent cohort. Blood samples were collected at 0, 3, 12, 18, 24 and 36 months after GAHT-initiation. Hormonal profiles were compared based on predicted sex-assigned-at-birth. Results: The model achieved 91% accuracy in NHANES, with baseline accuracy of 81% in ENIGI, declining to 11% at 36 months. Median E2 levels in TW predicted male-assigned-at-birth (62.5pg/mL, IQR: 40.9-77.1) were significantly lower (p<0.0001) than in TW predicted female-assigned-at-birth (81.2pg/mL, IQR: 64.9-102.0). Difference in T-levels in TM classified female-assigned-at-birth and predicted male-assigned-at-birth did not reach significance. Classification according to sex-at-birth seemed less likely to occur at higher E2 levels in TW. Conclusion: The innate sexual dimorphism of biochemical parameters might provide a framework for GAHT target validation. Significantly different hormonal profiles can be seen in people depending on the biochemical shift experienced during GAHT. Transgender hormone therapy GAHT random forest biochemistry sexual dimorphism Figures Figure 1 Figure 2 Figure 3 Figure 4 KEY POINTS 1. Hormonal target values for people receiving gender-affirming hormone therapy (GAHT) are often based on data from cisgender individuals, and lack validation. 2. Sexually dimorphic biochemical markers enable a simple random forest model to predict sex-assigned-at-birth with high accuracy. 3. GAHT-induced biochemical shifts can be monitored using this model, providing an objective measure of the completeness of biochemical transitioning. 4. The degree of biochemical transitioning shows a relationship with underlying hormonal thresholds, which could potentially help inform treatment decisions. INTRODUCTION The Endocrine Society guideline ( 1 ) is a universally recognized and relatively straightforward tool to help guide gender-affirming hormone therapy by providing simple target hormonal values. For feminizing hormone therapy, maintaining estradiol levels within a physiological range whilst not exceeding 200 pg/mL is recommended ( 1 ). These targets are designed to reflect hormonal levels observed in cisgender individuals and align with the estradiol range of 60–150 pg/mL established for menopausal hormone replacement therapy ( 1 , 2 ). However, extrapolating reference values from one population to another with distinct physiological needs presents limitations. Identifying whether specific hormone levels correlate with adverse health outcomes has been prioritized as a critical research area in gender-affirming care ( 1 , 3 ). Determining both lower and upper target hormone levels presents unique challenges. The lower bound should balance maintaining physiological homeostasis and long-term health outcomes, such as bone mass preservation in feminizing hormone therapy users. Additionally, patient satisfaction with physical and emotional changes induced by gender-affirming hormone therapy (GAHT) is a crucial consideration. The upper bound represents a toxic threshold, with inappropriately high E2 levels potentially leading to health-threatening occurrences such as thromboembolic events ( 1 ). As these adverse events are relatively rare, robust epidemiological studies with extensive follow-up are required to establish safe upper limits ( 5 ). Acknowledging the multifaceted nature of transitioning may help establish appropriate lower bound hormonal targets. Transitioning encompasses social, professional, emotional, sexual, and physical dimensions ( 6 ). It is reasonable to hypothesize that individuals experiencing the highest satisfaction with physical changes may have achieved optimal hormone levels. Patient-reported outcome measures (PROMs) related to physical appearance could thus inform target ranges. However, satisfaction is inherently subjective and influenced by gender identity and individualized care goals. As a result, patient satisfaction with physical changes alone represents an unreliable metric for guiding hormonal target recommendations ( 7 ). An alternative approach involves using objective physiological markers such as body composition parameters in lieu of subjective physical markers. Unfortunately, a significant correlation between continuous estradiol levels and most clinical outcomes often appears to be absent, as shown in a narrative review by a group of Australian researchers ( 8 ). A potential explanation is the existence of a threshold value, beyond which clinical outcomes are optimized despite a lack of direct correlation with hormone levels. Identifying such a ‘breakpoint’ could help refine reference ranges (Fig. 1 ). While inducing physical changes with GAHT is a primary treatment goal, using secondary sex characteristics alone as a framework for hormone target validation is likely insufficient. Extending beyond physical changes, transitioning entails biochemical shifts, such as changes in hematocrit levels observed in both feminizing and masculinizing hormone therapy users ( 9 – 12 ). Individuals whose biochemical parameters shift toward reference values of their identified gender may have achieved adequate hormone levels. This hypothesis paper will explore alternative methods of hormonal target value generation by developing a predictive model for sex-assigned-at-birth using common, sexually dimorphic biochemical markers from cisgender individuals in the National Health and Nutrition Examination Survey (NHANES) ( 13 ). The model will then be applied to hormone-naïve, gender-diverse individuals in the European Network for the Investigation of Gender Incongruence (ENIGI) study ( 14 ) to assess its external validity. Hormone levels in ENIGI participants whose sex-assigned-at-birth is accurately or inaccurately predicted after three years of GAHT will be compared. Identifying such thresholds could inform lower target recommendations for feminizing hormone therapy, addressing existing gaps in the 2017 Endocrine Society guidelines ( 1 ). By identifying biochemical patterns associated with these outcomes, this study seeks to generate hypotheses regarding lower hormonal target ranges for feminizing hormone therapy and to address gaps in the 2017 Endocrine Society clinical practice guidelines. Rather than optimizing predictive performance, this exploratory work introduces a deliberately non-optimal modeling framework to offer a novel conceptual perspective on hormone target validation and to stimulate further academic discussion. METHODOLOGY A total of 5433 individuals aged 18 to 52 years were randomly selected from the publicly available NHANES database, including data from the 2013–2014, 2015–2016, and 2017–2020 cycles ( 13 ). These individuals were presumed to be cisgender, or gender-diverse people not undergoing GAHT. The age range was selected to include adults within the likely premenopausal range for cisgender women ( 15 ), though menopausal status was not explicitly verified. Extracted variables included sex (presumed sex-at-birth), age, sex-hormone binding globulin (SHBG), platelet count (Tc), hematocrit (Hct), high-density lipoprotein (HDL) cholesterol, total cholesterol (TotChol), and serum creatinine (SCreat). These biochemical parameters were chosen based on their availability in the ENIGI dataset, their known sexual dimorphic distribution, and their routine measurement in clinical practice. The 95% confidence intervals for biochemical markers were calculated separately for each sex and compared to internal reference ranges from Ghent University Hospital (GhUH) as provided in Supplementary Materials A (ESM A) . Differences between NHANES and GhUH reference ranges, likely attributable to interlaboratory measurement variability, were addressed by converting NHANES values to Z-scores and recalibrating them based on GhUH reference ranges. Biochemical techniques for quantifying hormone and biochemical values are listed in ESM B. The dataset was split into training and validation subsets (n = 2716, and n = 2717, 47.2% assigned male at birth (AMAB). A random forest model was trained to predict ‘sex-at-birth’ and validated using RStudio software version 2024.04.01, incorporating Hct, Tc, SCreat, SHBG, and HDL cholesterol/TotChol ratio as predictors. Briefly, a chronological age correction factor was considered for SCreat. Age-related decline in renal function might cause SCreat-levels to rise. Elevated SCreat-levels might then inaccurately be interpreted as an indicator of higher lean body mass. This could potentially predispose older adults to be identified as AMAB by the model. However, age-corrected SCreat did not outperform uncorrected SCreat. Model performance was assessed using receiver operating characteristic (ROC) curves and area under the curve (AUC) values. The RStudio packages randomForest, ROCR, readxl, and ggplot2 were utilized for analysis. Predictor suitability was evaluated using mean decrease in accuracy and mean decrease in Gini index, indicating overall predictive power and data-splitting performance, respectively. Class weights were optimized iteratively, with final values set at 0.54 for female and 0.46 for male classifications. A random forest approach was preferred over logistic regression due to its ability to handle complex interactions and prioritize predictive accuracy over interpretability ( 16 ). Logistic regression was also avoided to prevent misinterpretation of probability scores as a "femininity index," which could exacerbate gender dysphoria. Variables are reported as median (1st quartile – 3rd quartile) unless otherwise specified. Based on prior empirical analyses of tabular clinical datasets, tree-based algorithms such as random forests generally require ample sample sizes to reach stable discriminative performance. Specifically, median sample sizes of ~ 3400 were sufficient for random forests to achieve near-maximal AUCs, whereas simpler methods like logistic regression required fewer samples and neural networks required substantially more (median > 12000). Accordingly, our split of 2716 participants for training and 2717 for validation falls within the empirically supported range for random forests, balancing predictive performance with efficient use of available data ( 17 ). The validated model was applied to ENIGI data to predict sex-assigned-at-birth in hormone-naïve individuals at baseline and at fixed intervals during GAHT. The ENIGI study protocol has been extensively described in prior publications ( 14 ). At 36 months, individuals were categorized based on whether their sex-assigned-at-birth was accurately or inaccurately predicted by the model. Total testosterone (TT) and estradiol (E2) levels were analyzed by birth sex and prediction status. Participants using feminizing hormone therapy were excluded if TT > 50pg/mL after GAHT initiation to minimize data noise introduced by varying anti-androgenic effect sizes. To account for baseline model inaccuracies, a post hoc correction factor was applied. For participants AMAB, 5% of individuals with the lowest E2 values were reclassified from the inaccurate to the accurate prediction group, where accurate denotes successful identification of sex-assigned-at-birth. Given that AMAB individuals under androgen-deprivation therapy (ADT) show a slight tendency to be classified as AFAB, this correction penalized the lowest E2 values. Conversely, for AFAB participants, 5% with the highest TT values were reclassified from the accurate to the inaccurate prediction group, as standard dosing regimens that produce high TT values yet remain labeled as incomplete biochemical shift are more likely to reflect model inaccuracies than true underdosing. RESULTS The median age of the NHANES training subset was 35 years (27–43), whilst the mean age in the validation subset was 35 years (27–44). The Hct and SCreat were to most important contributors to model accuracy performance, followed by SHBG. Both Tc and HDLratio contributed significantly to the model accuracy, but markedly less so than aforementioned predictors (Fig. 2 ). The ROC of the random forest in the validation-subset had an AUC of 0.966, with accuracy being 91.3%. The model was then introduced in hormone-naïve gender-diverse people from the ENIGI Ghent subcohort. Concretely, this sample existed of 290 individuals, 58.9% were assigned female at birth (AFAB). The median age was 22.0 years (19.6–27.8) for participants AFAB and 28.6 years (21.9–43.3) for people AMAB. Among participants initiating masculinizing GAHT (mGAHT) at baseline, the majority (97.7%) received intramuscular testosterone undecanoate (Nebido®, typically 1 g/4 mL every 12 weeks). A smaller proportion (1.8%) started with short-acting testosterone esters (Sustanon®, typically 250 mg/0.75 mL every 2 weeks), and 0.6% with transdermal testosterone gel. Doses were individualized, so not all participants received the standard regimen. For feminizing GAHT (fGAHT), the most commonly prescribed regimen (72.3%) was oral estradiol valerate (Progynova®, typically 2 mg twice daily) combined with cyproterone acetate, generally at high doses (≥ 25 mg once daily), which was discontinued upon gonadectomy. The second most common fGAHT was transdermal estrogen patches (22.7%), with transdermal gel being the least frequently used (5.0%). By the end of follow-up, most participants on mGAHT continued testosterone undecanoate (74.9%), while 25.1% were receiving short-acting esters. Among those on fGAHT, treatment distribution was oral tablets 66.4%, patches 19.3%, and gel 14.3%. The ROC analysis of the random forest showed strong initial performance, with an AUC of 0.874 and an accuracy of 81.3%. However, model performance declined sharply over time. At the three-month timepoint, the AUC dropped to 0.407, with accuracy falling to 39.9%. By twelve months, accuracy had decreased further to 23.4%, accompanied by an AUC of 0.106. At twenty-four months, the decline slowed, with accuracy at 19.5% and an AUC of 0.110. At the study conclusion at thirty-six months, the model reached its lowest performance, with accuracy of 10.5% and an AUC of 0.0459. At the 36 months timepoint, 160 (93.6%) of the AFAB individuals receiving mGAHT were identified as AMAB. In the group of AMAB receiving fGAHT 87 (73%) were identified as AFAB. So, in both groups people were more frequently assigned to their identified gender group, rather than their sex-assigned at birth group. The random forest initially predicted sex-assigned-at-birth reliably. Performance quickly declined, with a moment of maximal ambiguity around three months, after which predictions increasingly aligned with gender identity rather than sex-assigned-at-birth. Over time, the model could still distinguish between profiles but predominantly assigned them according to gender identity, reflecting the full biochemical shift under GAHT. A composite figure of the ROC curves can be seen in Fig. 3 . In the mGAHT group, E2 levels were significantly lower among participants classified according to their sex-assigned-at-birth. This difference was observed both in the average E2 during the final year of GAHT (average of 24m and 36m data, p < 0.001) and in the average E2 across the entire follow-up period excluding baseline (p 0.05) between groups, whether assessed as final-year averages or as averages across the full follow-up. Interpretation of TT was limited by missing information on administration timing, which precluded adjustment for peak, mid, or trough values. In the fGAHT group, TT levels were similarly non-significant between groups. By contrast, E2 levels were significantly higher in participants classified according to gender identity rather than sex-assigned-at-birth, both for the average E2 during the final year (p < 0.001) and for the average E2 across the entire follow-up (p < 0.001). The hormonal and biochemical values can be consulted in Table 1 . The correction factor caused one AFAB individual to move from the accurate to inaccurate group, whilst it shifted five AMAB individuals from the inaccurate to the accurate group. If no correction factor was applied, hormonal results remained comparable. In the fGAHT group, median E2 levels were significantly (p < 0.001) different at 81.2pg/mL (64.9–102.0) versus 62.5pg/mL (40.9–77.1) in the inaccurate (n = 92) and accurate (n = 27) group respectively. In the mGAHT group, median TT values were not-significantly (p > 0.05) different between groups, at 613ng/dL (478–761) and 571ng/dL (503–783) in the inaccurate (n = 159) and accurate (n = 12) group respectively. Post hoc power-testing showed that power to detect differences in lab variables varied across groups due to unbalanced sample sizes. Large effects, such as for Hct, SHBG, and E2, had high power (> 80–95%) and are reliably detected. Moderate effects (HDL ratio, TT, SCreat) had lower power (50–75%), and small differences (Tc) were underpowered (< 20%), so nonsignificant results should be interpreted cautiously. A visual representation for the median hormone levels over the complete study follow-up duration is given in Fig. 4 . Table 1 Median hormonal values between model-assigned groups at timepoint 36 months mGAHT - AFAB fGAHT - AMAB Label assigned by model AFAB (n = 11) AMAB (n = 160) AFAB (n = 87) AMAB (n = 32) Final year TT (ng/dL) 606 (539–797) 638 (478–869) 15.1 (11.7–21.7) 12.4 (10.6–16.7) Total TT (ng/dL) 571 (503–783) 613 (478–761) 15.1 (12.0-19.2) 12.6 (11.0-17.5) Final year E2 (pg/mL) 25.0 (25.0-30.3) 32.7 (26.0-42.2)*** 81.4 (63.4–110.0) 38.8 (31.9–62.8)**** Total E2 (pg/mL) 27.8 (27.2–31.3) 36.1 (29.4–42.7)** 81.2 (64.9–102.0) 62.5 (40.9–77.1)**** Hct (%) 40.5 (39.0–41.0) 46.9 (45.3–49.2)**** 40.2 (38.8–41.5) 41.6 (40.3–43.6)** HDL ratio (%) 32.2 (29.2–35.8) 26.1 (21.7–31.4)* 34.3 (27.3–40.5) 29.3 (22.7–36.3) SCreat (mg/dL) 0.82 (0.77–0.87) 0.88 (0.81–0.96) 0.79 (0.74–0.85) 0.89 (0.76–0.94)* SHBG (nmol/L) 47.8 (38.2–62.6) 28.7 (21.5–36.2)** 70.8 (53.6–108.0) 47.9 (33.9–61.9)**** Tc (10^3/mcL) 264 (240–293) 242 (215–276) 252 (212–288) 250 (216–282) AFAB: assigned female at birth, AMAB: assigned male at birth, E2: estradiol, fGAHT: feminizing gender-affirming hormone therapy, Hct: hematocrit, HDLratio: high-density-lipoprotein cholesterol as percentage of total cholesterol, mGAHT: masculinizing gender-affirming hormone therapy, SCreat: serum creatinine, SHBG, sex-hormone binding globulin, Tc: platelet, TT: total testosterone. *: p ≤ 0.05, **: p ≤ 0.01, ***: p ≤ 0.001, ****: p ≤ 0.0001. Significance only reported within GAHT type groups. We were unable to identify which individuals were at risk of incomplete biochemical shifts due to heterogeneity in treatment regimens. Among individuals receiving fGAHT who remained classified within their sex-assigned-at-birth, 40.6% (n = 13/32) used a transdermal agent, a proportion slightly higher than that observed in the overall cohort (33.6%). In the mGAHT group, one individual who remained classified within their sex-assigned-at-birth was prescribed Sustanon at one ampoule every three weeks, deviating from the standard in-hospital regimen of every two weeks. Another individual had received their last Nebido® injection six months earlier, as confirmed by review of the electronic patient file, yet still demonstrated low–normal male-range TT levels of 243 ng/dL at the study visit. A third individual exhibited elevated SHBG values while on standard Nebido® treatment. However, a concomitant diagnosis of Graves’ hyperthyroidism may have contributed to the SHBG elevation. DISCUSSION The random forest model, trained on cisgender NHANES participants, initially predicted sex-assigned-at-birth with high accuracy in hormone-naïve ENIGI participants. Model performance declined over the course of GAHT. By three months, accuracy was near chance, and after twelve months, the model increasingly classified participants according to gender identity rather than sex-assigned-at-birth. This decline corresponds to biochemical changes induced by GAHT. At thirty-six months, AMAB participants identified within their feminine gender identity had median E2 levels approximately 20 pg/mL higher than those still classified according to birth sex. AMAB individuals in which the model could no longer identify the sex-assigned-at-birth, only infrequently had E2 levels 150pg/mL. In AFAB participants, testosterone levels did not differ between prediction groups, but E2 concentrations were lower in those still classified by birth sex. This might reflect lower TT exposure and subsequent TT aromatization in this group. There is historical precedent for applying random forest models to establish treatment goals. For example, researchers have developed a machine learning–based approach to optimize vancomycin dosing by selecting the most appropriate pharmacokinetic models. This model-informed guidance yielded more stable therapeutic drug monitoring outcomes, suggesting that integrating such tools can enhance individualized drug monitoring and dosing ( 18 ). Beyond pharmacotherapy, machine learning can also help identify at-risk patient subgroups who may require additional or tailored care. An illustrative case is the development of a random forest model designed to predict the severity of postoperative pain in orthopedic surgical patients. The study demonstrated that random forest algorithms can effectively stratify patients at high risk of severe postoperative pain, thereby supporting early and targeted interventions ( 19 ). We noted an important discrepancy between our AUC-values and the accuracy of the generated model. The high AUC reflects an excellent ‘ranking’ ability of the model, whilst the poorer accuracy could reflect issues when assigning subjects to their groups. Whilst this could reflect data imbalance within the ENIGI cohort, with an overrepresentation of one gender identity group at a given timepoint, it also seemed to be present within the model validation subset. This could reflect problems at the level of the categorization threshold. The R package that was used generally defines this threshold at 0.50, which might have been inappropriate. We did not reconfigure this threshold seeing we did not intend to build the most optimal model possible, rather, we intended to introduce this as a theoretical framework for hormonal target validation research to build upon. Future models could rethink which predictors can be included to strengthen model performance. Many additional sexual dimorphic parameters exist, such as creatinine kinase, high sensitivity CRP, markers of bone turn-over or B-type natriuric peptide (BNP) ( 20 – 22 ). Newer iterations of the model could be strengthened by incorporating non-biochemical variables, integrating laboratory parameters with anthropometric measurements, such as fat distribution ( 23 ). However, the inclusion of more predictors must be approached with caution. Increasing model complexity raises the risk of overfitting, which in turn necessitates larger sample sizes for both training and validation. This, in turn, may push the analysis toward more complex machine learning approaches, potentially making the modeling process less practical and more difficult to interpret ( 17 ). The present research is set within the ENIGI protocol, which restricts the range of predictors available for model expansion. Nevertheless, focusing exclusively on routinely measured biochemical parameters enhances the feasibility of implementing such tools within standard gender-affirming care. In addition to incorporating a broader set of predictors, future research should examine whether biochemical recognition can be linked to functional outcomes such as metabolic health, bone density, or patient-reported measures, including satisfaction and hormonal symptom burden. Establishing such associations would enable subsequent validation of hormonal targets generated by the random forest model through their relationship with clinically meaningful outcomes in long-term follow-up ( 24 ). While large-scale, longitudinal studies within stakeholder cohorts will ultimately be required to define evidence-based hormonal thresholds, the availability of theoretically derived targets may allow for more efficient validation against real-world outcomes. This might reduce sample size requirements compared to establishing thresholds de novo. Previous research from the Netherlands has identified E2 levels of 50pg/mL (183pmol/L) as a protective threshold for bone health ( 25 ). This finding aligns well with our data, showing that AMAB individuals with a female-signature biochemical profile seldomly had E2 levels lower than 50pg/mL. This cohort of ENIGI participants generally stayed within the proposed E2 target range as stipulated by the Endocrine Society ( 1 ). However, other centres might see individuals with markedly higher E2 concentrations, especially when intramuscular or sublingual administration routes are used ( 26 – 28 ). Dosing regimens and E2 level range within the ENIGI Ghent subcohort might be insufficiently diverse to execute an optimal dose finding study. Simultaneously, whilst a random forest approach can help assist in establishing a lower therapy threshold, it is not suited for defining upper treatment limits. Fixing upper limits relates to balancing efficacy with safety concerns, and should be inform by GAHT-related adverse health outcomes ( 1 , 29 ). It was not possible to determine which individuals were most likely to remain biochemically aligned with their sex-assigned-at-birth after three years of GAHT. This limitation highlights the complexity and variability of treatment trajectories in this population. Notably, we observed one case in which an individual receiving mGAHT presented with a total testosterone level of 243 ng/dL, a value just below the lower limit of the male reference range used at the GhUH internal laboratory (253ng/dL) ( 30 ). Closer inspection revealed that this individual had discontinued testosterone therapy six months earlier. This observation highlights the limitations of relying solely on TT as an outcome measure, as TT may lack the nuance required to capture treatment dynamics. Free testosterone could have been a viable alternative ( 31 ). However, because SHBG was already included as a predictor in the random forest model, adding free testosterone as an outcome might have introduced interference. This example also shows that biochemical changes may signal treatment issues before TT levels begin to decline. Dose-finding in mGAHT might therefore be better guided by exposure parameters, such as testosterone dose per kg per day, rather than relying only on blood measures. Moreover, we observed that individuals AMAB who remained classified as such by the model exhibited higher Hct and SCreat and lower SHBG. This could relate to obesity-driven changes in SHBG and muscle mass, or to hypoxemia from sleep apnea, which can raise Hct ( 32 ). However, we could not establish such a pattern within our data. Importantly, this model is limited in applicability to nonbinary individuals. It was trained on binary cisgender reference data and cannot guide hormone optimization for nonbinary participants. Nonbinary identity does not imply a unique hormonal profile, and many nonbinary individuals may pursue hormone regimens that overlap with binary targets. Additional limitations exist within this research. Due to the long existence and the living protocol of ENIGI, biochemical analysis methods changed over follow-up. Even though manufacturer-guided corrective factors were applied to the measurements, changes to the analysis method introduced noise in the data. Generating and validating a model in a US cohort, and applying it to a Belgian setting is far-from ideal. Notably, differences in anthropometric parameters such as BMI could have introduced interpopulation changes with regard to SHBG or SCreat levels, which can be influenced by body size and obesity status ( 33 ). Body weight was not included as a predictor variable. Cisgender men and transgender women are on average taller than cisgender women, and assuming a comparable healthy BMI, absolute body weight would therefore be systematically higher. Including body weight as a predictor could inadvertently reveal sex-assigned-at-birth. The use of BMI-adjusted values for SHBG and SCreat could be explored in future studies. However, incorporating such adjustments would add complexity to an already assumption-heavy exploratory framework and detract from the conceptual focus of this hypothesis-generating work. This study did not aim to optimize predictive performance. Rather, it sought to introduce a novel biochemical perspective on hormone target validation. Within this context, model performance as assessed by ROC analysis was considered adequate. If a biochemical-based validation approach were to be adopted for clinical application or guideline development, substantially greater methodological rigor would be required. Additionally, it is unclear to which extent the estrogenic component of fGAHT influences model performance. In cisgender men receiving ADT notable biochemical and anthropomorphic shifts are documented mimicking those seen in fGAHT. Examples being reduced erythropoiesis, reduced lean mass, and increased HDL cholesterol levels ( 34 – 36 ). As a result, cisgender men receiving ADT might also be prone to being labeled AFAB by the model. If the androgen-lowering effect is too domineering over the estrogenic component the model might be inherently flawed as a dose finding tool. Whilst we tried to compensate for this by including predictors such as SHBG with a clear link to E2 levels, the two most impactful predictors within the model, SCreat and Hct, are clearly androgen linked. Including a group of cisgender men on ADT to help generate and validate the model, could perhaps remedy this. Additionally, whilst the black box nature of random forest models does not allow to easily deconstruct the decision-making process of the model, one AFAB individual receiving mGAHT appeared to have been assigned the AFAB label due to increased SHBG levels. A concomitant diagnosis of hyperthyroidism was found, which is known to cause elevated SHBG levels ( 32 ). Participant selection should perhaps have been more stringent, by omitting people with thyroidal illnesses. Finally, whilst the post-hoc corrective factor was an attempt to correct for baseline differences in model performance between NHANES and ENIGI, it was a mere exploratory adjustment to solidify hormonal difference between groups ( 37 ). CONCLUSION This hypothesis paper explores a theoretical framework for validating hormonal targets in GAHT by analyzing biochemical shifts in transgender individuals. Current GAHT guidelines are often extrapolated from other populations, such as hormone replacement therapy data in cisgender postmenopausal women. Developing population-specific guidelines based on stakeholder data is a research priority. To address this, we trained a random forest model on data from cisgender individuals to predict sex-assigned-at-birth using routinely measured sexually dimorphic biomarkers. We then tracked the model’s accuracy over three years of GAHT, observing a rapid decline as participants’ biochemical profiles shifted toward their identified gender. While no significant differences in TT were observed among mGAHT users, E2 levels were markedly higher in fGAHT users recognized as “female” by the model; approximately 20 pg/mL (73 pmol/L) higher than those identified as “male.” Although these findings are not yet sufficient to guide clinical practice, they may provide useful context for future dose-finding initiative. Declarations Funding: This work was supported by The Fund for Innovation and Clinical Research of Ghent University Hospital, Belgium [Grant number FIKO21/TYPE2/025]. Ethical Committee: The protocol of the ENIGI study has been approved by the ethical committee of the Ghent University Hospital (EC/2009/622). Consent to Participate: Written consent provided by participants. Data Sharing Agreement: Upon reasonable request Code Availability: Upon reasonable request Conflict of Interest: None to declare References Hembree WC, Cohen-Kettenis PT, Gooren L, Hannema SE, Meyer WJ, Murad MH et al (2017) Endocrine Treatment of Gender-Dysphoric/Gender-Incongruent Persons: An Endocrine Society Clinical Practice Guideline. J Clin Endocrinol Metab 102(11):3869–3903 Glynne S, Reisel D, Kamal A, Neville A, McColl L, Lewis R et al (2024) The range and variation in serum estradiol concentration in perimenopausal and postmenopausal women treated with transdermal estradiol in a real-world setting: a cross-sectional study. Menopause Feldman J, Brown GR, Deutsch MB, Hembree W, Meyer W, Meyer-Bahlburg HF et al (2016) Priorities for transgender medical and healthcare research. Curr Opin Endocrinol Diabetes Obes 23(2):180–187 Institute of Medicine Committee on Standards for Developing Trustworthy Clinical Practice (2011) G. In: Graham R, Mancher M, Miller Wolman D, Greenfield S, Steinberg E, editors. Clinical Practice Guidelines We Can Trust. Washington (DC): National Academies Press (US) Copyright 2011 by the National Academy of Sciences. All rights reserved Abou-Ismail MY, Citla Sridhar D, Nayak L (2020) Estrogen and thrombosis: A bench to bedside review. Thromb Res 192:40–51 Magalhães M, Aparicio-García ME, García-Nieto I (2020) Transition Trajectories: Contexts, Difficulties and Consequences Reported by Young Transgender and Non-Binary Spaniards. Int J Environ Res Public Health. ;17(18) van de Grift TC, Cohen-Kettenis PT, Steensma TD, De Cuypere G, Richter-Appelt H, Haraldsen IR et al (2016) Body Satisfaction and Physical Appearance in Gender Dysphoria. Arch Sex Behav 45(3):575–585 Nolan BJ, Cheung AS (2021) Relationship Between Serum Estradiol Concentrations and Clinical Outcomes in Transgender Individuals Undergoing Feminizing Hormone Therapy: A Narrative Review. Transgend Health 6(3):125–131 Cheung AS, Lim HY, Cook T, Zwickl S, Ginger A, Chiang C et al (2021) Approach to Interpreting Common Laboratory Pathology Tests in Transgender Individuals. J Clin Endocrinol Metab 106(3):893–901 Defreyne J, Vantomme B, Van Caenegem E, Wierckx K, De Blok CJM, Klaver M et al (2018) Prospective evaluation of hematocrit in gender-affirming hormone treatment: results from European Network for the Investigation of Gender Incongruence. Andrology 6(3):446–454 Krupka E, Curtis S, Ferguson T, Whitlock R, Askin N, Millar AC et al (2022) The Effect of Gender-Affirming Hormone Therapy on Measures of Kidney Function: A Systematic Review and Meta-Analysis. Clin J Am Soc Nephrol 17(9):1305–1315 Vita R, Settineri S, Liotta M, Benvenga S, Trimarchi F (2018) Changes in hormonal and metabolic parameters in transgender subjects on cross-sex hormone therapy: A cohort study. Maturitas 107:92–96 Centers for Disease Control and Prevention (CDC) NCfHSN (2025) National Health and Nutrition Examination Survey Data. US Department of Health and Human Services, Centers for Disease Control and Prevention, Hyattsville, MD Cocchetti C, Romani A, Collet S, Greenman Y, Schreiner T, Wiepjes C et al (2022) The ENIGI (European Network for the Investigation of Gender Incongruence) Study: Overview of Acquired Endocrine Knowledge and Future Perspectives. J Clin Med. ;11(7) Gold EB (2011) The timing of the age at which natural menopause occurs. Obstet Gynecol Clin North Am 38(3):425–440 Couronné R, Probst P, Boulesteix AL (2018) Random forest versus logistic regression: a large-scale benchmark experiment. BMC Bioinformatics 19(1):270 Silvey S, Liu J (2024) Sample Size Requirements for Popular Classification Algorithms in Tabular Clinical Data: Empirical Study. J Med Internet Res 26:e60231 Lee S, Song M, Han J, Lee D, Kim BH (2022) Application of Machine Learning Classification to Improve the Performance of Vancomycin Therapeutic Drug Monitoring. Pharmaceutics. ;14(5) Shi G, Liu G, Gao Q, Zhang S, Wang Q, Wu L et al (2023) A random forest algorithm-based prediction model for moderate to severe acute postoperative pain after orthopedic surgery under general anesthesia. BMC Anesthesiol 23(1):361 Redfield MM, Rodeheffer RJ, Jacobsen SJ, Mahoney DW, Bailey KR, Burnett JC (2002) Jr. Plasma brain natriuretic peptide concentration: impact of age and gender. J Am Coll Cardiol 40(5):976–982 Bafei SEC, Yang S, Chen C, Gu X, Mu J, Liu F et al (2023) Sex and age differences in the association between high sensitivity C-reactive protein and all-cause mortality: A 12-year prospective cohort study. Mech Ageing Dev 211:111804 Wang D, Ma C, Zou Y, Yu S, Li H, Cheng X et al (2020) Gender and age-specific reference intervals of common biochemical analytes in Chinese population: Derivation using real laboratory data. J Med Biochem 39(3):384–391 Wells JC (2007) Sexual dimorphism of body composition. Best Pract Res Clin Endocrinol Metab 21(3):415–430 Ou FS, Michiels S, Shyr Y, Adjei AA, Oberg AL (2021) Biomarker Discovery and Validation: Statistical Considerations. J Thorac Oncol 16(4):537–545 Wiepjes CM, de Jongh RT, de Blok CJ, Vlot MC, Lips P, Twisk JW et al (2019) Bone Safety During the First Ten Years of Gender-Affirming Hormonal Treatment in Transwomen and Transmen. J Bone Min Res 34(3):447–454 Doll E, Gunsolus I, Thorgerson A, Tangpricha V, Lamberton N, Sarvaideo JL (2022) Pharmacokinetics of Sublingual Versus Oral Estradiol in Transgender Women. Endocr Pract 28(3):237–242 Rothman MS, Ariel D, Kelley C, Hamnvik OR, Abramowitz J, Irwig MS et al (2024) The Use of Injectable Estradiol in Transgender and Gender Diverse Adults: A Scoping Review of Dose and Serum Estradiol Levels. Endocr Pract 30(9):870–878 Iwamoto SJ, Defreyne J, Kaoutzanis C, Davies RD, Moreau KL, Rothman MS (2023) Gender-affirming hormone therapy, mental health, and surgical considerations for aging transgender and gender diverse adults. Ther Adv Endocrinol Metab 14:20420188231166494 Combined hormonal contraception (2017) and the risk of venous thromboembolism: a guideline. Fertil Steril 107(1):43–51 Laboratoriumgids Klinische Biologie UZGent (2025) Consulted 2025-08-14 via URL: https://labgids.uzgent.be/ Keevil BG, Adaway J (2019) Assessment of free testosterone concentration. J Steroid Biochem Mol Biol 190:207–211 Thaler MA, Seifert-Klauss V, Luppa PB (2015) The biomarker sex hormone-binding globulin - from established applications to emerging trends in clinical medicine. Best Pract Res Clin Endocrinol Metab 29(5):749–760 Vasunilashorn S, Kim JK, Crimmins EM (2013) International differences in the links between obesity and physiological dysregulation: the United States, England, and Taiwan. J Obes 2013:618056 Smith MR, Saad F, Egerdie B, Sieber PR, Tammela TL, Ke C et al (2012) Sarcopenia during androgen-deprivation therapy for prostate cancer. J Clin Oncol 30(26):3271–3276 Albuquerque CP, Freitas FR, Martinelli AEM, Lima JH, Coelho RF, Serrano CV Jr. et al (2020) Androgen deprivation therapy improves the in vitro capacity of high-density lipoprotein (HDL) to receive cholesterol and other lipids in patients with prostate carcinoma. Lipids Health Dis 19(1):133 Gagliano-Jucá T, Pencina KM, Ganz T, Travison TG, Kantoff PW, Nguyen PL et al (2018) Mechanisms responsible for reduced erythropoiesis during androgen deprivation therapy in men with prostate cancer. Am J Physiol Endocrinol Metab 315(6):E1185–e93 Dankowski T, Ziegler A (2016) Calibrating random forests for probability estimation. Stat Med 35(22):3949–3960 Supplementary Files ESMrandom.docx Cite Share Download PDF Status: Under Revision Version 1 posted Editorial decision: Major revisions 24 Mar, 2026 Reviewers agreed at journal 20 Feb, 2026 Reviewers invited by journal 20 Feb, 2026 Editor assigned by journal 30 Dec, 2025 First submitted to journal 29 Dec, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8473261","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":594435950,"identity":"53130309-2734-48e0-bc02-c235d268fd46","order_by":0,"name":"Jeroen Vervalcke","email":"data:image/png;base64,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","orcid":"https://orcid.org/0009-0004-3909-892X","institution":"Universitair Ziekenhuis Gent","correspondingAuthor":true,"prefix":"","firstName":"Jeroen","middleName":"","lastName":"Vervalcke","suffix":""},{"id":594435951,"identity":"dd467a0a-3fca-4343-ad33-12c8b92a78bc","order_by":1,"name":"Lorenzo Marinelli","email":"","orcid":"","institution":"University of Turin: Universita degli Studi di Torino","correspondingAuthor":false,"prefix":"","firstName":"Lorenzo","middleName":"","lastName":"Marinelli","suffix":""},{"id":594435952,"identity":"0c1b155c-4f4b-4f5e-b463-fc3f04abcb10","order_by":2,"name":"Dorte Glintborg","email":"","orcid":"","institution":"Odense University Hospital: Odense Universitetshospital","correspondingAuthor":false,"prefix":"","firstName":"Dorte","middleName":"","lastName":"Glintborg","suffix":""},{"id":594435953,"identity":"8f57d8ae-af9c-4e45-b81d-4ef077038d10","order_by":3,"name":"Louise Lehmann Christensen","email":"","orcid":"","institution":"Odense Universitetshospital","correspondingAuthor":false,"prefix":"","firstName":"Louise","middleName":"Lehmann","lastName":"Christensen","suffix":""},{"id":594435954,"identity":"e5146219-0959-4bc4-8a21-eab757e9485a","order_by":4,"name":"Martin den Heijer","email":"","orcid":"","institution":"Amsterdam UMC - Locatie AMC: Amsterdam UMC Locatie AMC","correspondingAuthor":false,"prefix":"","firstName":"Martin","middleName":"den","lastName":"Heijer","suffix":""},{"id":594435955,"identity":"2ec24c8a-4f7a-4993-9d36-a1e980ab0474","order_by":5,"name":"Alessandra Fisher","email":"","orcid":"","institution":"University of Florence: Universita degli Studi di Firenze","correspondingAuthor":false,"prefix":"","firstName":"Alessandra","middleName":"","lastName":"Fisher","suffix":""},{"id":594435956,"identity":"fb3a4b46-4e72-4e84-bc1b-1f67811cee88","order_by":6,"name":"Konstantina Barouti","email":"","orcid":"","institution":"Helena Venizelou general and Maternity District Hospital: Geniko Nosokomeio Elena Benizelou","correspondingAuthor":false,"prefix":"","firstName":"Konstantina","middleName":"","lastName":"Barouti","suffix":""},{"id":594435957,"identity":"01ac5f85-6881-480b-9a59-7ce2d59cf7e3","order_by":7,"name":"Guy T Sjoen","email":"","orcid":"","institution":"Universitair Ziekenhuis Gent","correspondingAuthor":false,"prefix":"","firstName":"Guy","middleName":"T","lastName":"Sjoen","suffix":""}],"badges":[],"createdAt":"2025-12-29 13:02:18","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8473261/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8473261/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":103351535,"identity":"84e01ee1-1c60-44fe-99b4-5d698c46519c","added_by":"auto","created_at":"2026-02-24 17:14:40","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":16096,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eTheoretical distribution of datapoints. Data sparsity in the low value range in the presence of random scatter in the higher range can explain lack of meaningful correlation (depicted by the red trendline) \u0026nbsp;when analyzing hormone values as a continuous variable. In this example, a breakpoint value for hormone levels can be determined (blue dotted line). Below this breakpoint, participants perform much more poorly for a given outcome variable.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8473261/v1/b71f84461f8dce03e7b691b4.png"},{"id":103351534,"identity":"9ebadb5b-294c-4a06-8abb-6c0c9638e58b","added_by":"auto","created_at":"2026-02-24 17:14:40","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":8518,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eAccuracy decrease for predictor variables when building the model in the NHANES dataset.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8473261/v1/c32aac9953f41984f411d1c5.png"},{"id":103351533,"identity":"7eccc38b-febc-4b4f-9e75-5c8cc55296a3","added_by":"auto","created_at":"2026-02-24 17:14:40","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":38267,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eReceiver operating curves showing model performance within the NHANES cohort (reference) and during the first three years of gender-affirming hormone therapy.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8473261/v1/6185856359e1c88df121f777.png"},{"id":103351537,"identity":"59ca4455-8494-4dee-95d7-6e4b5c404c43","added_by":"auto","created_at":"2026-02-24 17:14:40","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":66880,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eMedian hormone values for the entire study duration\u003c/em\u003e\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8473261/v1/abc0986383ca044dfa11ad34.png"},{"id":103509990,"identity":"9ba1cef5-5ec6-4c31-a700-e1cba8df8148","added_by":"auto","created_at":"2026-02-26 14:02:39","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":551340,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8473261/v1/e231b112-7fd7-44fc-b1db-6d2c3ee3fd34.pdf"},{"id":103507262,"identity":"4c3939bf-6a27-4767-ae8b-2a09f98bfd58","added_by":"auto","created_at":"2026-02-26 13:40:49","extension":"docx","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":20383,"visible":true,"origin":"","legend":"","description":"","filename":"ESMrandom.docx","url":"https://assets-eu.researchsquare.com/files/rs-8473261/v1/f5705d6316d9181255073365.docx"}],"financialInterests":"","formattedTitle":"Hormonal Target Validation based on Biochemical Shifts in Gender-Affirming Hormone Therapy","fulltext":[{"header":"KEY POINTS","content":"\u003cp\u003e1. Hormonal target values for people receiving gender-affirming hormone therapy (GAHT) are often based on data from cisgender individuals, and lack validation.\u003c/p\u003e\u003cp\u003e2. Sexually dimorphic biochemical markers enable a simple random forest model to predict sex-assigned-at-birth with high accuracy.\u003c/p\u003e\u003cp\u003e3. GAHT-induced biochemical shifts can be monitored using this model, providing an objective measure of the completeness of biochemical transitioning.\u003c/p\u003e\u003cp\u003e4. The degree of biochemical transitioning shows a relationship with underlying hormonal thresholds, which could potentially help inform treatment decisions.\u003c/p\u003e"},{"header":"INTRODUCTION","content":"\u003cp\u003eThe Endocrine Society guideline (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e) is a universally recognized and relatively straightforward tool to help guide gender-affirming hormone therapy by providing simple target hormonal values. For feminizing hormone therapy, maintaining estradiol levels within a physiological range whilst not exceeding 200 pg/mL is recommended (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e). These targets are designed to reflect hormonal levels observed in cisgender individuals and align with the estradiol range of 60\u0026ndash;150 pg/mL established for menopausal hormone replacement therapy (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e). However, extrapolating reference values from one population to another with distinct physiological needs presents limitations. Identifying whether specific hormone levels correlate with adverse health outcomes has been prioritized as a critical research area in gender-affirming care (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eDetermining both lower and upper target hormone levels presents unique challenges. The lower bound should balance maintaining physiological homeostasis and long-term health outcomes, such as bone mass preservation in feminizing hormone therapy users. Additionally, patient satisfaction with physical and emotional changes induced by gender-affirming hormone therapy (GAHT) is a crucial consideration. The upper bound represents a toxic threshold, with inappropriately high E2 levels potentially leading to health-threatening occurrences such as thromboembolic events (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e). As these adverse events are relatively rare, robust epidemiological studies with extensive follow-up are required to establish safe upper limits (\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAcknowledging the multifaceted nature of transitioning may help establish appropriate lower bound hormonal targets. Transitioning encompasses social, professional, emotional, sexual, and physical dimensions (\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e). It is reasonable to hypothesize that individuals experiencing the highest satisfaction with physical changes may have achieved optimal hormone levels. Patient-reported outcome measures (PROMs) related to physical appearance could thus inform target ranges. However, satisfaction is inherently subjective and influenced by gender identity and individualized care goals. As a result, patient satisfaction with physical changes alone represents an unreliable metric for guiding hormonal target recommendations (\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e). An alternative approach involves using objective physiological markers such as body composition parameters in lieu of subjective physical markers. Unfortunately, a significant correlation between continuous estradiol levels and most clinical outcomes often appears to be absent, as shown in a narrative review by a group of Australian researchers (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e). A potential explanation is the existence of a threshold value, beyond which clinical outcomes are optimized despite a lack of direct correlation with hormone levels. Identifying such a \u0026lsquo;breakpoint\u0026rsquo; could help refine reference ranges (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). While inducing physical changes with GAHT is a primary treatment goal, using secondary sex characteristics alone as a framework for hormone target validation is likely insufficient. Extending beyond physical changes, transitioning entails biochemical shifts, such as changes in hematocrit levels observed in both feminizing and masculinizing hormone therapy users (\u003cspan additionalcitationids=\"CR10 CR11\" citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e). Individuals whose biochemical parameters shift toward reference values of their identified gender may have achieved adequate hormone levels.\u003c/p\u003e \u003cp\u003eThis hypothesis paper will explore alternative methods of hormonal target value generation by developing a predictive model for sex-assigned-at-birth using common, sexually dimorphic biochemical markers from cisgender individuals in the National Health and Nutrition Examination Survey (NHANES) (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e). The model will then be applied to hormone-na\u0026iuml;ve, gender-diverse individuals in the European Network for the Investigation of Gender Incongruence (ENIGI) study (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e) to assess its external validity. Hormone levels in ENIGI participants whose sex-assigned-at-birth is accurately or inaccurately predicted after three years of GAHT will be compared. Identifying such thresholds could inform lower target recommendations for feminizing hormone therapy, addressing existing gaps in the 2017 Endocrine Society guidelines (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e). By identifying biochemical patterns associated with these outcomes, this study seeks to generate hypotheses regarding lower hormonal target ranges for feminizing hormone therapy and to address gaps in the 2017 Endocrine Society clinical practice guidelines. Rather than optimizing predictive performance, this exploratory work introduces a deliberately non-optimal modeling framework to offer a novel conceptual perspective on hormone target validation and to stimulate further academic discussion.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"METHODOLOGY","content":"\u003cp\u003eA total of 5433 individuals aged 18 to 52 years were randomly selected from the publicly available NHANES database, including data from the 2013\u0026ndash;2014, 2015\u0026ndash;2016, and 2017\u0026ndash;2020 cycles (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e). These individuals were presumed to be cisgender, or gender-diverse people not undergoing GAHT. The age range was selected to include adults within the likely premenopausal range for cisgender women (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e), though menopausal status was not explicitly verified. Extracted variables included sex (presumed sex-at-birth), age, sex-hormone binding globulin (SHBG), platelet count (Tc), hematocrit (Hct), high-density lipoprotein (HDL) cholesterol, total cholesterol (TotChol), and serum creatinine (SCreat). These biochemical parameters were chosen based on their availability in the ENIGI dataset, their known sexual dimorphic distribution, and their routine measurement in clinical practice.\u003c/p\u003e \u003cp\u003eThe 95% confidence intervals for biochemical markers were calculated separately for each sex and compared to internal reference ranges from Ghent University Hospital (GhUH) as provided in \u003cb\u003eSupplementary Materials A (ESM A)\u003c/b\u003e. Differences between NHANES and GhUH reference ranges, likely attributable to interlaboratory measurement variability, were addressed by converting NHANES values to Z-scores and recalibrating them based on GhUH reference ranges. Biochemical techniques for quantifying hormone and biochemical values are listed in \u003cb\u003eESM B.\u003c/b\u003e\u003c/p\u003e \u003cp\u003eThe dataset was split into training and validation subsets (n\u0026thinsp;=\u0026thinsp;2716, and n\u0026thinsp;=\u0026thinsp;2717, 47.2% assigned male at birth (AMAB). A random forest model was trained to predict \u0026lsquo;sex-at-birth\u0026rsquo; and validated using RStudio software version 2024.04.01, incorporating Hct, Tc, SCreat, SHBG, and HDL cholesterol/TotChol ratio as predictors. Briefly, a chronological age correction factor was considered for SCreat. Age-related decline in renal function might cause SCreat-levels to rise. Elevated SCreat-levels might then inaccurately be interpreted as an indicator of higher lean body mass. This could potentially predispose older adults to be identified as AMAB by the model. However, age-corrected SCreat did not outperform uncorrected SCreat. Model performance was assessed using receiver operating characteristic (ROC) curves and area under the curve (AUC) values. The RStudio packages randomForest, ROCR, readxl, and ggplot2 were utilized for analysis. Predictor suitability was evaluated using mean decrease in accuracy and mean decrease in Gini index, indicating overall predictive power and data-splitting performance, respectively. Class weights were optimized iteratively, with final values set at 0.54 for female and 0.46 for male classifications. A random forest approach was preferred over logistic regression due to its ability to handle complex interactions and prioritize predictive accuracy over interpretability (\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e). Logistic regression was also avoided to prevent misinterpretation of probability scores as a \"femininity index,\" which could exacerbate gender dysphoria. Variables are reported as median (1st quartile \u0026ndash; 3rd quartile) unless otherwise specified.\u003c/p\u003e \u003cp\u003eBased on prior empirical analyses of tabular clinical datasets, tree-based algorithms such as random forests generally require ample sample sizes to reach stable discriminative performance. Specifically, median sample sizes of ~\u0026thinsp;3400 were sufficient for random forests to achieve near-maximal AUCs, whereas simpler methods like logistic regression required fewer samples and neural networks required substantially more (median\u0026thinsp;\u0026gt;\u0026thinsp;12000). Accordingly, our split of 2716 participants for training and 2717 for validation falls within the empirically supported range for random forests, balancing predictive performance with efficient use of available data (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe validated model was applied to ENIGI data to predict sex-assigned-at-birth in hormone-na\u0026iuml;ve individuals at baseline and at fixed intervals during GAHT. The ENIGI study protocol has been extensively described in prior publications (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e). At 36 months, individuals were categorized based on whether their sex-assigned-at-birth was accurately or inaccurately predicted by the model. Total testosterone (TT) and estradiol (E2) levels were analyzed by birth sex and prediction status. Participants using feminizing hormone therapy were excluded if TT\u0026thinsp;\u0026gt;\u0026thinsp;50pg/mL after GAHT initiation to minimize data noise introduced by varying anti-androgenic effect sizes. To account for baseline model inaccuracies, a post hoc correction factor was applied. For participants AMAB, 5% of individuals with the lowest E2 values were reclassified from the inaccurate to the accurate prediction group, where accurate denotes successful identification of sex-assigned-at-birth. Given that AMAB individuals under androgen-deprivation therapy (ADT) show a slight tendency to be classified as AFAB, this correction penalized the lowest E2 values. Conversely, for AFAB participants, 5% with the highest TT values were reclassified from the accurate to the inaccurate prediction group, as standard dosing regimens that produce high TT values yet remain labeled as incomplete biochemical shift are more likely to reflect model inaccuracies than true underdosing.\u003c/p\u003e"},{"header":"RESULTS","content":"\u003cp\u003eThe median age of the NHANES training subset was 35 years (27\u0026ndash;43), whilst the mean age in the validation subset was 35 years (27\u0026ndash;44). The Hct and SCreat were to most important contributors to model accuracy performance, followed by SHBG. Both Tc and HDLratio contributed significantly to the model accuracy, but markedly less so than aforementioned predictors (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). The ROC of the random forest in the validation-subset had an AUC of 0.966, with accuracy being 91.3%.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe model was then introduced in hormone-na\u0026iuml;ve gender-diverse people from the ENIGI Ghent subcohort. Concretely, this sample existed of 290 individuals, 58.9% were assigned female at birth (AFAB). The median age was 22.0 years (19.6\u0026ndash;27.8) for participants AFAB and 28.6 years (21.9\u0026ndash;43.3) for people AMAB. Among participants initiating masculinizing GAHT (mGAHT) at baseline, the majority (97.7%) received intramuscular testosterone undecanoate (Nebido\u0026reg;, typically 1 g/4 mL every 12 weeks). A smaller proportion (1.8%) started with short-acting testosterone esters (Sustanon\u0026reg;, typically 250 mg/0.75 mL every 2 weeks), and 0.6% with transdermal testosterone gel. Doses were individualized, so not all participants received the standard regimen. For feminizing GAHT (fGAHT), the most commonly prescribed regimen (72.3%) was oral estradiol valerate (Progynova\u0026reg;, typically 2 mg twice daily) combined with cyproterone acetate, generally at high doses (\u0026ge;\u0026thinsp;25 mg once daily), which was discontinued upon gonadectomy. The second most common fGAHT was transdermal estrogen patches (22.7%), with transdermal gel being the least frequently used (5.0%). By the end of follow-up, most participants on mGAHT continued testosterone undecanoate (74.9%), while 25.1% were receiving short-acting esters. Among those on fGAHT, treatment distribution was oral tablets 66.4%, patches 19.3%, and gel 14.3%.\u003c/p\u003e \u003cp\u003eThe ROC analysis of the random forest showed strong initial performance, with an AUC of 0.874 and an accuracy of 81.3%. However, model performance declined sharply over time. At the three-month timepoint, the AUC dropped to 0.407, with accuracy falling to 39.9%. By twelve months, accuracy had decreased further to 23.4%, accompanied by an AUC of 0.106. At twenty-four months, the decline slowed, with accuracy at 19.5% and an AUC of 0.110. At the study conclusion at thirty-six months, the model reached its lowest performance, with accuracy of 10.5% and an AUC of 0.0459. At the 36 months timepoint, 160 (93.6%) of the AFAB individuals receiving mGAHT were identified as AMAB. In the group of AMAB receiving fGAHT 87 (73%) were identified as AFAB. So, in both groups people were more frequently assigned to their identified gender group, rather than their sex-assigned at birth group. The random forest initially predicted sex-assigned-at-birth reliably. Performance quickly declined, with a moment of maximal ambiguity around three months, after which predictions increasingly aligned with gender identity rather than sex-assigned-at-birth. Over time, the model could still distinguish between profiles but predominantly assigned them according to gender identity, reflecting the full biochemical shift under GAHT. A composite figure of the ROC curves can be seen in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eIn the mGAHT group, E2 levels were significantly lower among participants classified according to their sex-assigned-at-birth. This difference was observed both in the average E2 during the final year of GAHT (average of 24m and 36m data, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) and in the average E2 across the entire follow-up period excluding baseline (p\u0026thinsp;\u0026lt;\u0026thinsp;0.01). TT levels did not differ significantly (p\u0026thinsp;\u0026gt;\u0026thinsp;0.05) between groups, whether assessed as final-year averages or as averages across the full follow-up. Interpretation of TT was limited by missing information on administration timing, which precluded adjustment for peak, mid, or trough values. In the fGAHT group, TT levels were similarly non-significant between groups. By contrast, E2 levels were significantly higher in participants classified according to gender identity rather than sex-assigned-at-birth, both for the average E2 during the final year (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) and for the average E2 across the entire follow-up (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). The hormonal and biochemical values can be consulted in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eThe correction factor caused one AFAB individual to move from the accurate to inaccurate group, whilst it shifted five AMAB individuals from the inaccurate to the accurate group. If no correction factor was applied, hormonal results remained comparable. In the fGAHT group, median E2 levels were significantly (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) different at 81.2pg/mL (64.9\u0026ndash;102.0) versus 62.5pg/mL (40.9\u0026ndash;77.1) in the inaccurate (n\u0026thinsp;=\u0026thinsp;92) and accurate (n\u0026thinsp;=\u0026thinsp;27) group respectively. In the mGAHT group, median TT values were not-significantly (p\u0026thinsp;\u0026gt;\u0026thinsp;0.05) different between groups, at 613ng/dL (478\u0026ndash;761) and 571ng/dL (503\u0026ndash;783) in the inaccurate (n\u0026thinsp;=\u0026thinsp;159) and accurate (n\u0026thinsp;=\u0026thinsp;12) group respectively. Post hoc power-testing showed that power to detect differences in lab variables varied across groups due to unbalanced sample sizes. Large effects, such as for Hct, SHBG, and E2, had high power (\u0026gt;\u0026thinsp;80\u0026ndash;95%) and are reliably detected. Moderate effects (HDL ratio, TT, SCreat) had lower power (50\u0026ndash;75%), and small differences (Tc) were underpowered (\u0026lt;\u0026thinsp;20%), so nonsignificant results should be interpreted cautiously. A visual representation for the median hormone levels over the complete study follow-up duration is given in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eMedian hormonal values between model-assigned groups at timepoint 36 months\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003emGAHT - AFAB\u003c/span\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003efGAHT - AMAB\u003c/span\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eLabel assigned by model\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAFAB (n\u0026thinsp;=\u0026thinsp;11)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAMAB (n\u0026thinsp;=\u0026thinsp;160)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAFAB (n\u0026thinsp;=\u0026thinsp;87)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAMAB (n\u0026thinsp;=\u0026thinsp;32)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFinal year TT (ng/dL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e606 (539\u0026ndash;797)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e638 (478\u0026ndash;869)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e15.1 (11.7\u0026ndash;21.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e12.4 (10.6\u0026ndash;16.7)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTotal TT (ng/dL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e571 (503\u0026ndash;783)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e613 (478\u0026ndash;761)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e15.1 (12.0-19.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e12.6 (11.0-17.5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFinal year E2 (pg/mL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e25.0 (25.0-30.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e32.7 (26.0-42.2)***\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e81.4 (63.4\u0026ndash;110.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e38.8 (31.9\u0026ndash;62.8)****\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTotal E2 (pg/mL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e27.8 (27.2\u0026ndash;31.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e36.1 (29.4\u0026ndash;42.7)**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e81.2 (64.9\u0026ndash;102.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e62.5 (40.9\u0026ndash;77.1)****\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHct (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e40.5 (39.0\u0026ndash;41.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e46.9 (45.3\u0026ndash;49.2)****\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e40.2 (38.8\u0026ndash;41.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e41.6 (40.3\u0026ndash;43.6)**\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHDL ratio (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e32.2 (29.2\u0026ndash;35.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e26.1 (21.7\u0026ndash;31.4)*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e34.3 (27.3\u0026ndash;40.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e29.3 (22.7\u0026ndash;36.3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSCreat (mg/dL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.82 (0.77\u0026ndash;0.87)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.88 (0.81\u0026ndash;0.96)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.79 (0.74\u0026ndash;0.85)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.89 (0.76\u0026ndash;0.94)*\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSHBG (nmol/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e47.8 (38.2\u0026ndash;62.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e28.7 (21.5\u0026ndash;36.2)**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e70.8 (53.6\u0026ndash;108.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e47.9 (33.9\u0026ndash;61.9)****\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTc (10^3/mcL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e264 (240\u0026ndash;293)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e242 (215\u0026ndash;276)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e252 (212\u0026ndash;288)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e250 (216\u0026ndash;282)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003eAFAB: assigned female at birth, AMAB: assigned male at birth, E2: estradiol, fGAHT: feminizing gender-affirming hormone therapy, Hct: hematocrit, HDLratio: high-density-lipoprotein cholesterol as percentage of total cholesterol, mGAHT: masculinizing gender-affirming hormone therapy, SCreat: serum creatinine, SHBG, sex-hormone binding globulin, Tc: platelet, TT: total testosterone. *: p\u0026thinsp;\u0026le;\u0026thinsp;0.05, **: p\u0026thinsp;\u0026le;\u0026thinsp;0.01, ***: p\u0026thinsp;\u0026le;\u0026thinsp;0.001, ****: p\u0026thinsp;\u0026le;\u0026thinsp;0.0001. Significance only reported within GAHT type groups.\u003c/em\u003e \u003c/p\u003e \u003cp\u003eWe were unable to identify which individuals were at risk of incomplete biochemical shifts due to heterogeneity in treatment regimens. Among individuals receiving fGAHT who remained classified within their sex-assigned-at-birth, 40.6% (n\u0026thinsp;=\u0026thinsp;13/32) used a transdermal agent, a proportion slightly higher than that observed in the overall cohort (33.6%). In the mGAHT group, one individual who remained classified within their sex-assigned-at-birth was prescribed Sustanon at one ampoule every three weeks, deviating from the standard in-hospital regimen of every two weeks. Another individual had received their last Nebido\u0026reg; injection six months earlier, as confirmed by review of the electronic patient file, yet still demonstrated low\u0026ndash;normal male-range TT levels of 243 ng/dL at the study visit. A third individual exhibited elevated SHBG values while on standard Nebido\u0026reg; treatment. However, a concomitant diagnosis of Graves\u0026rsquo; hyperthyroidism may have contributed to the SHBG elevation.\u003c/p\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eThe random forest model, trained on cisgender NHANES participants, initially predicted sex-assigned-at-birth with high accuracy in hormone-na\u0026iuml;ve ENIGI participants. Model performance declined over the course of GAHT. By three months, accuracy was near chance, and after twelve months, the model increasingly classified participants according to gender identity rather than sex-assigned-at-birth. This decline corresponds to biochemical changes induced by GAHT. At thirty-six months, AMAB participants identified within their feminine gender identity had median E2 levels approximately 20 pg/mL higher than those still classified according to birth sex. AMAB individuals in which the model could no longer identify the sex-assigned-at-birth, only infrequently had E2 levels\u0026thinsp;\u0026lt;\u0026thinsp;50pg/mL. The model was likewise unable to identify birth sex \u0026ldquo;\u003cem\u003eaccurately\u003c/em\u003e\u0026rdquo; in AMAB individuals with E2 levels\u0026thinsp;\u0026gt;\u0026thinsp;150pg/mL. In AFAB participants, testosterone levels did not differ between prediction groups, but E2 concentrations were lower in those still classified by birth sex. This might reflect lower TT exposure and subsequent TT aromatization in this group.\u003c/p\u003e \u003cp\u003eThere is historical precedent for applying random forest models to establish treatment goals. For example, researchers have developed a machine learning\u0026ndash;based approach to optimize vancomycin dosing by selecting the most appropriate pharmacokinetic models. This model-informed guidance yielded more stable therapeutic drug monitoring outcomes, suggesting that integrating such tools can enhance individualized drug monitoring and dosing (\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e). Beyond pharmacotherapy, machine learning can also help identify at-risk patient subgroups who may require additional or tailored care. An illustrative case is the development of a random forest model designed to predict the severity of postoperative pain in orthopedic surgical patients. The study demonstrated that random forest algorithms can effectively stratify patients at high risk of severe postoperative pain, thereby supporting early and targeted interventions (\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eWe noted an important discrepancy between our AUC-values and the accuracy of the generated model. The high AUC reflects an excellent \u0026lsquo;ranking\u0026rsquo; ability of the model, whilst the poorer accuracy could reflect issues when assigning subjects to their groups. Whilst this could reflect data imbalance within the ENIGI cohort, with an overrepresentation of one gender identity group at a given timepoint, it also seemed to be present within the model validation subset. This could reflect problems at the level of the categorization threshold. The R package that was used generally defines this threshold at 0.50, which might have been inappropriate. We did not reconfigure this threshold seeing we did not intend to build the most optimal model possible, rather, we intended to introduce this as a theoretical framework for hormonal target validation research to build upon. Future models could rethink which predictors can be included to strengthen model performance. Many additional sexual dimorphic parameters exist, such as creatinine kinase, high sensitivity CRP, markers of bone turn-over or B-type natriuric peptide (BNP) (\u003cspan additionalcitationids=\"CR21\" citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e). Newer iterations of the model could be strengthened by incorporating non-biochemical variables, integrating laboratory parameters with anthropometric measurements, such as fat distribution (\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e). However, the inclusion of more predictors must be approached with caution. Increasing model complexity raises the risk of overfitting, which in turn necessitates larger sample sizes for both training and validation. This, in turn, may push the analysis toward more complex machine learning approaches, potentially making the modeling process less practical and more difficult to interpret (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e). The present research is set within the ENIGI protocol, which restricts the range of predictors available for model expansion. Nevertheless, focusing exclusively on routinely measured biochemical parameters enhances the feasibility of implementing such tools within standard gender-affirming care.\u003c/p\u003e \u003cp\u003eIn addition to incorporating a broader set of predictors, future research should examine whether biochemical recognition can be linked to functional outcomes such as metabolic health, bone density, or patient-reported measures, including satisfaction and hormonal symptom burden. Establishing such associations would enable subsequent validation of hormonal targets generated by the random forest model through their relationship with clinically meaningful outcomes in long-term follow-up (\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e). While large-scale, longitudinal studies within stakeholder cohorts will ultimately be required to define evidence-based hormonal thresholds, the availability of theoretically derived targets may allow for more efficient validation against real-world outcomes. This might reduce sample size requirements compared to establishing thresholds de novo. Previous research from the Netherlands has identified E2 levels of 50pg/mL (183pmol/L) as a protective threshold for bone health (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e). This finding aligns well with our data, showing that AMAB individuals with a female-signature biochemical profile seldomly had E2 levels lower than 50pg/mL. This cohort of ENIGI participants generally stayed within the proposed E2 target range as stipulated by the Endocrine Society (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e). However, other centres might see individuals with markedly higher E2 concentrations, especially when intramuscular or sublingual administration routes are used (\u003cspan additionalcitationids=\"CR27\" citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e). Dosing regimens and E2 level range within the ENIGI Ghent subcohort might be insufficiently diverse to execute an optimal dose finding study. Simultaneously, whilst a random forest approach can help assist in establishing a lower therapy threshold, it is not suited for defining upper treatment limits. Fixing upper limits relates to balancing efficacy with safety concerns, and should be inform by GAHT-related adverse health outcomes (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIt was not possible to determine which individuals were most likely to remain biochemically aligned with their sex-assigned-at-birth after three years of GAHT. This limitation highlights the complexity and variability of treatment trajectories in this population. Notably, we observed one case in which an individual receiving mGAHT presented with a total testosterone level of 243 ng/dL, a value just below the lower limit of the male reference range used at the GhUH internal laboratory (253ng/dL) (\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e). Closer inspection revealed that this individual had discontinued testosterone therapy six months earlier. This observation highlights the limitations of relying solely on TT as an outcome measure, as TT may lack the nuance required to capture treatment dynamics. Free testosterone could have been a viable alternative (\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e). However, because SHBG was already included as a predictor in the random forest model, adding free testosterone as an outcome might have introduced interference. This example also shows that biochemical changes may signal treatment issues before TT levels begin to decline. Dose-finding in mGAHT might therefore be better guided by exposure parameters, such as testosterone dose per kg per day, rather than relying only on blood measures. Moreover, we observed that individuals AMAB who remained classified as such by the model exhibited higher Hct and SCreat and lower SHBG. This could relate to obesity-driven changes in SHBG and muscle mass, or to hypoxemia from sleep apnea, which can raise Hct (\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e). However, we could not establish such a pattern within our data.\u003c/p\u003e \u003cp\u003eImportantly, this model is limited in applicability to nonbinary individuals. It was trained on binary cisgender reference data and cannot guide hormone optimization for nonbinary participants. Nonbinary identity does not imply a unique hormonal profile, and many nonbinary individuals may pursue hormone regimens that overlap with binary targets. Additional limitations exist within this research. Due to the long existence and the living protocol of ENIGI, biochemical analysis methods changed over follow-up. Even though manufacturer-guided corrective factors were applied to the measurements, changes to the analysis method introduced noise in the data. Generating and validating a model in a US cohort, and applying it to a Belgian setting is far-from ideal. Notably, differences in anthropometric parameters such as BMI could have introduced interpopulation changes with regard to SHBG or SCreat levels, which can be influenced by body size and obesity status (\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e). Body weight was not included as a predictor variable. Cisgender men and transgender women are on average taller than cisgender women, and assuming a comparable healthy BMI, absolute body weight would therefore be systematically higher. Including body weight as a predictor could inadvertently reveal sex-assigned-at-birth. The use of BMI-adjusted values for SHBG and SCreat could be explored in future studies. However, incorporating such adjustments would add complexity to an already assumption-heavy exploratory framework and detract from the conceptual focus of this hypothesis-generating work. This study did not aim to optimize predictive performance. Rather, it sought to introduce a novel biochemical perspective on hormone target validation. Within this context, model performance as assessed by ROC analysis was considered adequate. If a biochemical-based validation approach were to be adopted for clinical application or guideline development, substantially greater methodological rigor would be required.\u003c/p\u003e \u003cp\u003eAdditionally, it is unclear to which extent the estrogenic component of fGAHT influences model performance. In cisgender men receiving ADT notable biochemical and anthropomorphic shifts are documented mimicking those seen in fGAHT. Examples being reduced erythropoiesis, reduced lean mass, and increased HDL cholesterol levels (\u003cspan additionalcitationids=\"CR35\" citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e). As a result, cisgender men receiving ADT might also be prone to being labeled AFAB by the model. If the androgen-lowering effect is too domineering over the estrogenic component the model might be inherently flawed as a dose finding tool. Whilst we tried to compensate for this by including predictors such as SHBG with a clear link to E2 levels, the two most impactful predictors within the model, SCreat and Hct, are clearly androgen linked. Including a group of cisgender men on ADT to help generate and validate the model, could perhaps remedy this. Additionally, whilst the black box nature of random forest models does not allow to easily deconstruct the decision-making process of the model, one AFAB individual receiving mGAHT appeared to have been assigned the AFAB label due to increased SHBG levels. A concomitant diagnosis of hyperthyroidism was found, which is known to cause elevated SHBG levels (\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e). Participant selection should perhaps have been more stringent, by omitting people with thyroidal illnesses. Finally, whilst the post-hoc corrective factor was an attempt to correct for baseline differences in model performance between NHANES and ENIGI, it was a mere exploratory adjustment to solidify hormonal difference between groups (\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e).\u003c/p\u003e"},{"header":"CONCLUSION","content":"\u003cp\u003eThis hypothesis paper explores a theoretical framework for validating hormonal targets in GAHT by analyzing biochemical shifts in transgender individuals. Current GAHT guidelines are often extrapolated from other populations, such as hormone replacement therapy data in cisgender postmenopausal women. Developing population-specific guidelines based on stakeholder data is a research priority. To address this, we trained a random forest model on data from cisgender individuals to predict sex-assigned-at-birth using routinely measured sexually dimorphic biomarkers. We then tracked the model\u0026rsquo;s accuracy over three years of GAHT, observing a rapid decline as participants\u0026rsquo; biochemical profiles shifted toward their identified gender. While no significant differences in TT were observed among mGAHT users, E2 levels were markedly higher in fGAHT users recognized as \u0026ldquo;female\u0026rdquo; by the model; approximately 20 pg/mL (73 pmol/L) higher than those identified as \u0026ldquo;male.\u0026rdquo; Although these findings are not yet sufficient to guide clinical practice, they may provide useful context for future dose-finding initiative.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cem\u003e\u003cu\u003eFunding:\u003c/u\u003e\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by The Fund for Innovation and Clinical Research of Ghent\u003c/p\u003e\n\u003cp\u003eUniversity Hospital, Belgium [Grant number FIKO21/TYPE2/025].\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u003cu\u003eEthical Committee:\u003c/u\u003e\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe protocol of the ENIGI study has been approved by the ethical committee of the Ghent University Hospital (EC/2009/622).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u003cu\u003eConsent to Participate:\u003c/u\u003e\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eWritten consent provided by participants.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u003cu\u003eData Sharing Agreement:\u003c/u\u003e\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eUpon reasonable request\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u003cu\u003eCode Availability:\u003c/u\u003e\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eUpon reasonable request\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u003cu\u003eConflict of Interest:\u003c/u\u003e\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eNone to declare\u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eHembree WC, Cohen-Kettenis PT, Gooren L, Hannema SE, Meyer WJ, Murad MH et al (2017) Endocrine Treatment of Gender-Dysphoric/Gender-Incongruent Persons: An Endocrine Society Clinical Practice Guideline. J Clin Endocrinol Metab 102(11):3869\u0026ndash;3903\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGlynne S, Reisel D, Kamal A, Neville A, McColl L, Lewis R et al (2024) The range and variation in serum estradiol concentration in perimenopausal and postmenopausal women treated with transdermal estradiol in a real-world setting: a cross-sectional study. Menopause\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFeldman J, Brown GR, Deutsch MB, Hembree W, Meyer W, Meyer-Bahlburg HF et al (2016) Priorities for transgender medical and healthcare research. Curr Opin Endocrinol Diabetes Obes 23(2):180\u0026ndash;187\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eInstitute of Medicine Committee on Standards for Developing Trustworthy Clinical Practice (2011) G. In: Graham R, Mancher M, Miller Wolman D, Greenfield S, Steinberg E, editors. Clinical Practice Guidelines We Can Trust. Washington (DC): National Academies Press (US) Copyright 2011 by the National Academy of Sciences. All rights reserved\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAbou-Ismail MY, Citla Sridhar D, Nayak L (2020) Estrogen and thrombosis: A bench to bedside review. Thromb Res 192:40\u0026ndash;51\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMagalh\u0026atilde;es M, Aparicio-Garc\u0026iacute;a ME, Garc\u0026iacute;a-Nieto I (2020) Transition Trajectories: Contexts, Difficulties and Consequences Reported by Young Transgender and Non-Binary Spaniards. Int J Environ Res Public Health. ;17(18)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evan de Grift TC, Cohen-Kettenis PT, Steensma TD, De Cuypere G, Richter-Appelt H, Haraldsen IR et al (2016) Body Satisfaction and Physical Appearance in Gender Dysphoria. Arch Sex Behav 45(3):575\u0026ndash;585\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNolan BJ, Cheung AS (2021) Relationship Between Serum Estradiol Concentrations and Clinical Outcomes in Transgender Individuals Undergoing Feminizing Hormone Therapy: A Narrative Review. Transgend Health 6(3):125\u0026ndash;131\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCheung AS, Lim HY, Cook T, Zwickl S, Ginger A, Chiang C et al (2021) Approach to Interpreting Common Laboratory Pathology Tests in Transgender Individuals. J Clin Endocrinol Metab 106(3):893\u0026ndash;901\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDefreyne J, Vantomme B, Van Caenegem E, Wierckx K, De Blok CJM, Klaver M et al (2018) Prospective evaluation of hematocrit in gender-affirming hormone treatment: results from European Network for the Investigation of Gender Incongruence. Andrology 6(3):446\u0026ndash;454\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKrupka E, Curtis S, Ferguson T, Whitlock R, Askin N, Millar AC et al (2022) The Effect of Gender-Affirming Hormone Therapy on Measures of Kidney Function: A Systematic Review and Meta-Analysis. Clin J Am Soc Nephrol 17(9):1305\u0026ndash;1315\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVita R, Settineri S, Liotta M, Benvenga S, Trimarchi F (2018) Changes in hormonal and metabolic parameters in transgender subjects on cross-sex hormone therapy: A cohort study. Maturitas 107:92\u0026ndash;96\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCenters for Disease Control and Prevention (CDC) NCfHSN (2025) National Health and Nutrition Examination Survey Data. US Department of Health and Human Services, Centers for Disease Control and Prevention, Hyattsville, MD\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCocchetti C, Romani A, Collet S, Greenman Y, Schreiner T, Wiepjes C et al (2022) The ENIGI (European Network for the Investigation of Gender Incongruence) Study: Overview of Acquired Endocrine Knowledge and Future Perspectives. J Clin Med. ;11(7)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGold EB (2011) The timing of the age at which natural menopause occurs. Obstet Gynecol Clin North Am 38(3):425\u0026ndash;440\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCouronn\u0026eacute; R, Probst P, Boulesteix AL (2018) Random forest versus logistic regression: a large-scale benchmark experiment. BMC Bioinformatics 19(1):270\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSilvey S, Liu J (2024) Sample Size Requirements for Popular Classification Algorithms in Tabular Clinical Data: Empirical Study. J Med Internet Res 26:e60231\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee S, Song M, Han J, Lee D, Kim BH (2022) Application of Machine Learning Classification to Improve the Performance of Vancomycin Therapeutic Drug Monitoring. Pharmaceutics. ;14(5)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShi G, Liu G, Gao Q, Zhang S, Wang Q, Wu L et al (2023) A random forest algorithm-based prediction model for moderate to severe acute postoperative pain after orthopedic surgery under general anesthesia. BMC Anesthesiol 23(1):361\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRedfield MM, Rodeheffer RJ, Jacobsen SJ, Mahoney DW, Bailey KR, Burnett JC (2002) Jr. Plasma brain natriuretic peptide concentration: impact of age and gender. J Am Coll Cardiol 40(5):976\u0026ndash;982\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBafei SEC, Yang S, Chen C, Gu X, Mu J, Liu F et al (2023) Sex and age differences in the association between high sensitivity C-reactive protein and all-cause mortality: A 12-year prospective cohort study. Mech Ageing Dev 211:111804\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang D, Ma C, Zou Y, Yu S, Li H, Cheng X et al (2020) Gender and age-specific reference intervals of common biochemical analytes in Chinese population: Derivation using real laboratory data. J Med Biochem 39(3):384\u0026ndash;391\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWells JC (2007) Sexual dimorphism of body composition. Best Pract Res Clin Endocrinol Metab 21(3):415\u0026ndash;430\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOu FS, Michiels S, Shyr Y, Adjei AA, Oberg AL (2021) Biomarker Discovery and Validation: Statistical Considerations. J Thorac Oncol 16(4):537\u0026ndash;545\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWiepjes CM, de Jongh RT, de Blok CJ, Vlot MC, Lips P, Twisk JW et al (2019) Bone Safety During the First Ten Years of Gender-Affirming Hormonal Treatment in Transwomen and Transmen. J Bone Min Res 34(3):447\u0026ndash;454\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDoll E, Gunsolus I, Thorgerson A, Tangpricha V, Lamberton N, Sarvaideo JL (2022) Pharmacokinetics of Sublingual Versus Oral Estradiol in Transgender Women. Endocr Pract 28(3):237\u0026ndash;242\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRothman MS, Ariel D, Kelley C, Hamnvik OR, Abramowitz J, Irwig MS et al (2024) The Use of Injectable Estradiol in Transgender and Gender Diverse Adults: A Scoping Review of Dose and Serum Estradiol Levels. Endocr Pract 30(9):870\u0026ndash;878\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIwamoto SJ, Defreyne J, Kaoutzanis C, Davies RD, Moreau KL, Rothman MS (2023) Gender-affirming hormone therapy, mental health, and surgical considerations for aging transgender and gender diverse adults. Ther Adv Endocrinol Metab 14:20420188231166494\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCombined hormonal contraception (2017) and the risk of venous thromboembolism: a guideline. Fertil Steril 107(1):43\u0026ndash;51\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLaboratoriumgids Klinische Biologie UZGent (2025) Consulted 2025-08-14 via URL: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://labgids.uzgent.be/\u003c/span\u003e\u003cspan address=\"https://labgids.uzgent.be/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKeevil BG, Adaway J (2019) Assessment of free testosterone concentration. J Steroid Biochem Mol Biol 190:207\u0026ndash;211\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eThaler MA, Seifert-Klauss V, Luppa PB (2015) The biomarker sex hormone-binding globulin - from established applications to emerging trends in clinical medicine. Best Pract Res Clin Endocrinol Metab 29(5):749\u0026ndash;760\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVasunilashorn S, Kim JK, Crimmins EM (2013) International differences in the links between obesity and physiological dysregulation: the United States, England, and Taiwan. J Obes 2013:618056\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSmith MR, Saad F, Egerdie B, Sieber PR, Tammela TL, Ke C et al (2012) Sarcopenia during androgen-deprivation therapy for prostate cancer. J Clin Oncol 30(26):3271\u0026ndash;3276\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlbuquerque CP, Freitas FR, Martinelli AEM, Lima JH, Coelho RF, Serrano CV Jr. et al (2020) Androgen deprivation therapy improves the in vitro capacity of high-density lipoprotein (HDL) to receive cholesterol and other lipids in patients with prostate carcinoma. Lipids Health Dis 19(1):133\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGagliano-Juc\u0026aacute; T, Pencina KM, Ganz T, Travison TG, Kantoff PW, Nguyen PL et al (2018) Mechanisms responsible for reduced erythropoiesis during androgen deprivation therapy in men with prostate cancer. Am J Physiol Endocrinol Metab 315(6):E1185\u0026ndash;e93\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDankowski T, Ziegler A (2016) Calibrating random forests for probability estimation. Stat Med 35(22):3949\u0026ndash;3960\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"hormones","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"HORM","sideBox":"Learn more about [Hormones](https://www.springer.com/journal/42000)","snPcode":"42000","submissionUrl":"https://www.editorialmanager.com/horm/default2.aspx","title":"Hormones","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Transgender, hormone therapy, GAHT, random forest, biochemistry, sexual dimorphism","lastPublishedDoi":"10.21203/rs.3.rs-8473261/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8473261/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground: \u003c/strong\u003eHormonal targets for gender-affirming hormone therapy (GAHT) remain largely unvalidated in gender-diverse individuals, with no clinical markers to assess dosing adequacy. We hypothesize that a biochemical shift towards the hormonal profile of the identified sex correlates with appropriate dosing.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods: \u003c/strong\u003eA data-subset from the National Health and Nutrition Examination Survey (NHANES), comprising 5433 assumed-cisgender individuals (47.2% male) aged \u0026lt;52 years, was used to train and validate a Random Forest model to predict sex-assigned-at-birth, incorporating platelet count, HDL%, SHBG, creatinine, and hematocrit. Being assigned-female-at-birth was weighted at 0.54. The model was subsequently run in 171 transgender men (TM) and 119 transgender women (TW) from the European Network for the Investigation of Gender Incongruence (ENIGI) Ghent cohort. Blood samples were collected at 0, 3, 12, 18, 24 and 36 months after GAHT-initiation. Hormonal profiles were compared based on predicted sex-assigned-at-birth.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults: \u003c/strong\u003eThe model achieved 91% accuracy in NHANES, with baseline accuracy of 81% in ENIGI, declining to 11% at 36 months. Median E2 levels in TW predicted male-assigned-at-birth (62.5pg/mL, IQR: 40.9-77.1) were significantly lower (p\u0026lt;0.0001) than in TW predicted female-assigned-at-birth (81.2pg/mL, IQR: 64.9-102.0). Difference in T-levels in TM classified female-assigned-at-birth and predicted male-assigned-at-birth did not reach significance. Classification according to sex-at-birth seemed less likely to occur at higher E2 levels in TW.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion: \u003c/strong\u003eThe innate sexual dimorphism of biochemical parameters might provide a framework for GAHT target validation. Significantly different hormonal profiles can be seen in people depending on the biochemical shift experienced during GAHT.\u003c/p\u003e","manuscriptTitle":"Hormonal Target Validation based on Biochemical Shifts in Gender-Affirming Hormone Therapy","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-24 17:14:35","doi":"10.21203/rs.3.rs-8473261/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Major revisions","date":"2026-03-24T05:48:58+00:00","index":"","fulltext":""},{"type":"reviewerAgreed","content":"","date":"2026-02-20T12:57:02+00:00","index":0,"fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-02-20T12:09:10+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-12-30T12:56:50+00:00","index":"","fulltext":""},{"type":"submitted","content":"Hormones","date":"2025-12-29T08:02:06+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"hormones","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"HORM","sideBox":"Learn more about [Hormones](https://www.springer.com/journal/42000)","snPcode":"42000","submissionUrl":"https://www.editorialmanager.com/horm/default2.aspx","title":"Hormones","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"45e12d13-f6b5-471c-b582-1e59047b50ca","owner":[],"postedDate":"February 24th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"in-revision","subjectAreas":[],"tags":[],"updatedAt":"2026-03-24T09:49:38+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-24 17:14:35","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8473261","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8473261","identity":"rs-8473261","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}