madman
Super Moderator
* Despite these limitations, this study provides a large-scale, head-to-head comparison of men receiving testosterone therapy with vs without evidence of hypogonadism, with up to 10 years of follow-up, race/ethnicity stratification, rigorous propensity-score matching, and complementary modelling approaches.
Important limitations should be acknowledged. As with all observational analyses, residual and unmeasured confounding cannot be excluded. Confounding by indication remains a central concern. Although we used an active-comparator, new-user design among TT initiators, propensity-score matching, complementary adjusted Cox models, and laboratory-anchored sensitivity analyses, these approaches can only reduce, not eliminate, confounding related to the clinical context of TT initiation. Men initiating TT without documented hypogonadism data may have differed from those with evidence of hypogonadism in unmeasured ways, including symptom burden, prescribing setting, clinician decision-making, patient preference, lifestyle factors, non-medical or performance-related use, and monitoring intensity. Accordingly, the findings should be interpreted as associations according to available hypogonadism evidence and documented prescribing context, not as causal estimates of TT itself.
Testosterone prescribing was not standardised and may have varied across countries, healthcare systems, primary and specialist care, and private settings. Although exploratory region-restricted analyses showed directionally consistent findings, regional differences in diagnostic work-up, prescription capture, laboratory availability, and documentation may still have contributed to exposure misclassification and residual confounding.
Treatment duration, adherence, discontinuation, formulation-specific route, formulation switching, dose, and cumulative exposure could not be reliably reconstructed in the available structured EHR data. Therefore, the analysis should be interpreted as an index-exposure, intention-to-treat–like comparison rather than as a strictly on-treatment, formulation-specific, or dose–response analysis. Intermittent use, early discontinuation, or treatment interruptions in either group may have influenced the observed associations, but could not be assessed reliably. Documentation of hypogonadism in routine clinical practice may be incomplete, and ascertainment of hypogonadism status in structured EHR data is inherently imperfect. In contrast, the available structured EHR data did not uniformly capture timing of blood sampling, fasting status, assay characteristics, repeated confirmatory measurements, symptom severity, or clinical reasoning. Therefore, some men classified as receiving TT with evidence of hypogonadism may have been identified on the basis of a single low testosterone value or diagnostic coding alone, whereas some men classified as receiving TT without evidence of hypogonadism based on available data may have had true biochemical deficiency or symptoms that were not captured. Accordingly, absence of hypogonadism evidence in the available structured EHR data should be interpreted as a proxy for potentially non–guideline-concordant TT prescribing, rather than as a definitive measure of off-label use. Because cohort assignment relied on available structured clinical data, the absence of documented hypogonadism, testosterone measurements, hypogonadal symptoms, or prior cardiovascular events should not be equated with their definitive clinical absence. Consequently, some men classified as receiving TT “without evidence of hypogonadism” may have had true biochemical deficiency that was not captured in the available data. To reduce potential misclassification, we added a laboratory-anchored sensitivity analysis restricted to patients with documented testosterone measurements; in this analysis, the primary association remained directionally consistent and statistically significant.
Despite these limitations, this study provides a large-scale, head-to-head comparison of men receiving testosterone therapy with vs without evidence of hypogonadism, with up to 10 years of follow-up, race/ethnicity stratification, rigorous propensity-score matching, and complementary modelling approaches. Importantly, this question is unlikely to be resolved by conventional randomised designs in the near term: a long-term trial that intentionally enrols and randomises men without evidence of biochemical deficiency to testosterone exposure would face substantial ethical constraints, as guideline-based care requires confirmation of hypogonadism before treatment initiation. The central message is clear—approximately one third of men receiving testosterone therapy lacked evidence of hypogonadism, and this prescribing pattern was associated with a clinically meaningful excess risk of major adverse cardiovascular and cerebrovascular events over long-term follow-up. Together with randomised evidence in selected trial populations, our findings emphasise that cardiovascular safety may depend on prescribing context, reinforcing the importance of guideline-based diagnostic work-up before treatment initiation.
Important limitations should be acknowledged. As with all observational analyses, residual and unmeasured confounding cannot be excluded. Confounding by indication remains a central concern. Although we used an active-comparator, new-user design among TT initiators, propensity-score matching, complementary adjusted Cox models, and laboratory-anchored sensitivity analyses, these approaches can only reduce, not eliminate, confounding related to the clinical context of TT initiation. Men initiating TT without documented hypogonadism data may have differed from those with evidence of hypogonadism in unmeasured ways, including symptom burden, prescribing setting, clinician decision-making, patient preference, lifestyle factors, non-medical or performance-related use, and monitoring intensity. Accordingly, the findings should be interpreted as associations according to available hypogonadism evidence and documented prescribing context, not as causal estimates of TT itself.
Testosterone prescribing was not standardised and may have varied across countries, healthcare systems, primary and specialist care, and private settings. Although exploratory region-restricted analyses showed directionally consistent findings, regional differences in diagnostic work-up, prescription capture, laboratory availability, and documentation may still have contributed to exposure misclassification and residual confounding.
Treatment duration, adherence, discontinuation, formulation-specific route, formulation switching, dose, and cumulative exposure could not be reliably reconstructed in the available structured EHR data. Therefore, the analysis should be interpreted as an index-exposure, intention-to-treat–like comparison rather than as a strictly on-treatment, formulation-specific, or dose–response analysis. Intermittent use, early discontinuation, or treatment interruptions in either group may have influenced the observed associations, but could not be assessed reliably. Documentation of hypogonadism in routine clinical practice may be incomplete, and ascertainment of hypogonadism status in structured EHR data is inherently imperfect. In contrast, the available structured EHR data did not uniformly capture timing of blood sampling, fasting status, assay characteristics, repeated confirmatory measurements, symptom severity, or clinical reasoning. Therefore, some men classified as receiving TT with evidence of hypogonadism may have been identified on the basis of a single low testosterone value or diagnostic coding alone, whereas some men classified as receiving TT without evidence of hypogonadism based on available data may have had true biochemical deficiency or symptoms that were not captured. Accordingly, absence of hypogonadism evidence in the available structured EHR data should be interpreted as a proxy for potentially non–guideline-concordant TT prescribing, rather than as a definitive measure of off-label use. Because cohort assignment relied on available structured clinical data, the absence of documented hypogonadism, testosterone measurements, hypogonadal symptoms, or prior cardiovascular events should not be equated with their definitive clinical absence. Consequently, some men classified as receiving TT “without evidence of hypogonadism” may have had true biochemical deficiency that was not captured in the available data. To reduce potential misclassification, we added a laboratory-anchored sensitivity analysis restricted to patients with documented testosterone measurements; in this analysis, the primary association remained directionally consistent and statistically significant.
Despite these limitations, this study provides a large-scale, head-to-head comparison of men receiving testosterone therapy with vs without evidence of hypogonadism, with up to 10 years of follow-up, race/ethnicity stratification, rigorous propensity-score matching, and complementary modelling approaches. Importantly, this question is unlikely to be resolved by conventional randomised designs in the near term: a long-term trial that intentionally enrols and randomises men without evidence of biochemical deficiency to testosterone exposure would face substantial ethical constraints, as guideline-based care requires confirmation of hypogonadism before treatment initiation. The central message is clear—approximately one third of men receiving testosterone therapy lacked evidence of hypogonadism, and this prescribing pattern was associated with a clinically meaningful excess risk of major adverse cardiovascular and cerebrovascular events over long-term follow-up. Together with randomised evidence in selected trial populations, our findings emphasise that cardiovascular safety may depend on prescribing context, reinforcing the importance of guideline-based diagnostic work-up before treatment initiation.
