Article Tools
ORIGINAL REPORTS
Epidemiology
Article Tools
Androgen Deprivation Therapy and the Risk of Dementia in Patients With Prostate Cancer
Abstract
Recent observational studies have associated the use of androgen deprivation therapy (ADT) with an increased risk of dementia and Alzheimer’s disease, but these studies had limitations. The objective of this study was to determine whether the use of ADT is associated with an increased risk of dementia, including Alzheimer’s disease, in patients with prostate cancer.
Using the United Kingdom’s Clinical Practice Research Datalink, we assembled a cohort of 30,903 men newly diagnosed with nonmetastatic prostate cancer between April 1, 1988 and April 30, 2015, and observed them until April 30, 2016. Time-dependent Cox proportional hazards models were used to estimate adjusted hazard ratios with 95% CIs of dementia associated with the use of ADT compared with nonuse. ADT exposure was lagged by 1 year to account for delays associated with the diagnosis of dementia and to minimize reverse causality. Secondary analyses assessed whether the risk varied with cumulative duration of use and by ADT type.
During a mean (standard deviation) follow-up of 4.3 (3.6) years, 799 patients were newly diagnosed with dementia (incidence, 6.0; 95% CI, 5.6 to 6.4) per 1,000 person-years. Compared with nonuse, ADT use was not associated with an increased risk of dementia (incidence, 7.4 v 4.4 per 1,000 person-years, respectively; adjusted hazard ratio, 1.02; 95% CI, 0.87 to 1.19). In secondary analyses, cumulative duration of use (P for heterogeneity = .78) and no single type of ADT were associated with an increased risk of dementia.
Androgen deprivation therapy (ADT) is the mainstay treatment of patients with advanced prostate cancer and it is being used increasingly in patients with clinically localized disease.1 Although ADT has been shown to delay prostate cancer progression and to improve survival,2 this therapy has been associated with several adverse events, including fractures, type 2 diabetes, cardiovascular disease, acute kidney injury, and possibly cognitive impairment.3-7 Indeed, the latter is based on the known association between age-related decline in testosterone levels and cognitive decline.8,9 Such declines have been associated with reduced performance in cognitive domains present in both Alzheimer’s disease and other dementias.10,11 However, the association between ADT and cognitive impairment is controversial, with a recent meta-analysis finding no significant differences in cognitive domains pertinent to the diagnosis of dementia, including performance on attention/working memory, executive function, language, verbal memory, visual memory, and visuospatial ability.7
To date, three observational studies have examined the association between ADT and Alzheimer’s disease and dementia, but they have resulted in mixed findings.12-14 Such discrepancies may have been caused by important methodologic limitations, including lack of consideration of exposure lag periods,12-14 exposure misclassification,12 and immortal time bias.12-14 Given the fact that men with prostate cancer represent an older population already at increased risk of cognitive impairment and dementia, it is imperative to assess whether ADT is associated with an increased incidence of this outcome. Thus, the objective of this study was to determine whether the use of ADT is associated with an increased risk of dementia, including Alzheimer’s disease, in patients diagnosed with prostate cancer.
This study was conducted using the United Kingdom’s Clinical Practice Research Datalink (CPRD). The CPRD is the world’s largest primary care database, consisting of > 14 million individuals enrolled across 700 general practices.15 The CPRD contains data on anthropometric measures, such as body mass index (BMI), clinical diagnoses and procedures, and prescriptions written by general practitioners. Patients in the CPRD have been shown to be broadly representative of the population of the United Kingdom in terms of age, sex, ethnicity, and BMI.15 The CPRD uses the Read code classification to record medical diagnoses and procedures. Diagnoses recorded in the CPRD have been shown to have high validity, with a median positive predictive value of 88.6% for all diagnoses.16 Prescriptions coded in the CPRD are based on the United Kingdom Prescription Pricing Authority dictionary.15
The study protocol (16_128R) was approved by the independent scientific advisory committee of the CPRD and by the research ethics board of Jewish General Hospital, Montreal, Canada.
Using the CPRD, we identified a cohort of men who were at least 40 years of age and newly diagnosed with prostate cancer between January 1, 1988 and April 30, 2015. Cohort entry was the date of the first-ever diagnosis of prostate cancer. We excluded patients with < 1 year of medical history in the CPRD before cohort entry, those with a history of ADT use (to exclude patients with prevalent disease), and those diagnosed with metastatic disease. We also excluded patients with a previous diagnosis of any dementia (including Alzheimer’s disease, as defined in Appendix Table A1, online only) and those previously diagnosed with ischemic stroke or transient ischemic attack at any time before cohort entry because these conditions can lead to cognitive impairments. Finally, all patients were required to have at least 1 year of follow-up after cohort entry to minimize the inclusion of prevalent dementia, given the known diagnostic delays associated with this condition.17
Patients meeting the study inclusion criteria were observed starting 1 year after cohort entry, until they received an incident diagnosis of dementia (identified on the basis of Read codes as defined in Appendix Table A1), or were censored on death from any cause, they received a new diagnosis of stroke or transient ischemic attack, or they ended their registration with the general practice, or until the end of study period (April 30, 2016), whichever occurred first.
Exposure to ADT included receiving gonadotropin-releasing hormone (GnRH) agonists (leuprolide, buserelin, goserelin, triptoerlin), oral antiandrogens (cyproterone acetate, flutamide, bicalutamide, nilutamide), and estrogens (diethylstilbestrol, estramustine), and undergoing bilateral orchiectomy. A time-dependent exposure definition was used, which allowed patients to transition from a period of nonexposure to a period of exposure to ADT. Patients were considered exposed to ADT starting 1 year after the date of the first prescription, or surgery date for patients who underwent bilateral orchiectomy (ie, after accounting for a 1-year lag period), until the end of follow-up. Thus, when events occurred within the year after treatment initiation, the patients were considered unexposed. Lagging the exposure was necessary because of the known diagnostic delays associated with dementia17 and to minimize biases related to detection and reverse causality (a situation in which exposure might be initiated or terminated at early signs or symptoms of the outcome). This approach also imposed a minimal time period between treatment initiation and onset of the outcome (ie, a latency period). Nonuse of ADT served as the reference category in all analyses. A figure depicting the exposure definition can be found in Appendix Figure A1 (online only).
On the basis of Read codes, we identified all incident events of dementia, including Alzheimer’s disease, occurring during the follow-up period (Appendix Table A1, online only). The diagnosis of dementia and Alzheimer’s disease in the CPRD has been previously shown to be valid, with 83% to 90% of the diagnostic codes confirmed upon detailed review of the records.18-20
Descriptive statistics were used to summarize the characteristics of the study cohort, and then separately for patients exposed to ADT and unexposed to ADT in the first 6 months of follow-up. Crude incidence of dementia and 95% CIs that were based on the Poisson distribution were calculated for all exposure groups. Time-dependent Cox proportional hazards models were used to estimate adjusted hazard ratios (HRs) and 95% CIs of incident dementia associated with the use of ADT compared with nonuse. This was considered the primary analysis.
All models were adjusted for the following variables measured at cohort entry: age (modeled as a flexible variable using a restricted cubic spline with five interior knots), year of cohort entry (entered categorically), alcohol-related disorders, smoking status (ever, never, unknown), and BMI (< 25 kg/m2, 25 to 30 kg/m2, ≥ 30.0 kg/m2, unknown). The models were also adjusted for comorbidities using the Deyo version of the Charlson comorbidity index (with dementia excluded)21 and the last prostate-specific antigen value measured before cohort entry (< 4 ng/mL, 4 to 10 ng/mL, > 10 ng/mL). In addition, the models were adjusted for history of head injury and use of antidepressants, antipsychotics, benzodiazepines, and hypnotics at any time before cohort entry.
We conducted three secondary analyses. First, we assessed whether there was a duration-response relationship in terms of cumulative duration of ADT use and the incidence of dementia. Cumulative duration of use was calculated, in a time-dependent fashion, by summing the prescriptions associated with each ADT drug or from the date of surgery for bilateral orchiectomy until the date of the risk set (ie, the date of the event). Separate HRs were estimated for five predefined duration categories (< 6, 6 to 12, 13 to 18, 19 to 24, ≥ 25 months), and a test of heterogeneity was used to assess differences across these duration categories. Cumulative duration of use was also modeled using a restricted cubic spline with five interior knots to produce a smooth curve of the HR as a function of the duration of ADT use.22 Second, we assessed whether the risk of dementia varied according to the use of specific ADT types. Thus, the use of ADT was further categorized, in a time-dependent fashion, into four mutually exclusive groups: gonadotropin-releasing hormone agonists only; oral antiandrogens only; gonadotropin-releasing hormone agonists and oral antiandrogens in combination; and other treatment combinations (bilateral orchiectomy, estrogens alone, and other ADTs). Finally, we also assessed separately the association between the use of ADT and Alzheimer’s disease and other dementias.
We conducted seven sensitivity analyses to assess the robustness of the findings. First, we repeated the primary analysis by lengthening the exposure lag period to 2 and 3 years. Second, we repeated the primary analysis but excluded patients with a history of antipsychotic use at baseline and, in addition, censored patients at the date of the first antipsychotic prescription. The latter analysis was performed to minimize outcome misclassification by accounting for the possibility that such patients were already being treated for symptoms of dementia. Third, to further account for the effect of chemotherapy on cognition, we excluded patients previously treated with chemotherapy before cohort entry and censored on a new chemotherapy treatment during follow-up. Fourth, we used multiple imputation methods23 for variables with missing data (smoking and BMI). Fifth, we used the disease risk score method24 as an alternative means of controlling for confounding (described in the Appendix). Sixth, to account for potential time-dependent confounding during the follow-up period, we repeated the primary analysis using a Cox proportional hazards marginal structural model25,26 (described in the Appendix). Finally, we used the Fine and Gray model27 to account for competing risks (death from any cause, ischemic stroke, and transient ischemic attack).
A total of 30,903 patients met the inclusion criteria (Fig 1). The mean (standard deviation) age at cohort entry was 70.7(8.9) years, and the cohort was observed for a mean (standard deviation) of 4.3 (3.6) years, generating 133,180 person-years of follow-up. (The reasons for end of follow-up are listed in Appendix Table A2). During this time, 799 patients were newly diagnosed with dementia, generating a crude incidence of 6.0 (95% CI, 5.6 to 6.4) per 1,000 person-years. This included 293 of patients (36.7%) with Alzheimer’s disease and 506 patients (63.3%) with other dementias. (A detailed distribution of these events is listed in Appendix Table A3). During the follow-up period, 17,994 patients (58.2%) used ADT, of whom 15,310 (49.5%) used ADT within the first 6 months of follow-up. The median (quartile 1 to quartile 3) duration of use was 2.3 (1.2 to 4.0) years.
Fig 1. Study flow diagram displaying the assembling of the study cohort of men newly diagnosed with prostate cancer in the United Kingdom Clinical Practice Research Datalink between 1988 and 2015. ADT, androgen deprivation therapy.
Table 1 presents the baseline characteristics of the entire cohort and the characteristics stratified by ADT use in the first 6 months of follow-up. Compared with nonusers, ADT users were older and were more likely to have ever smoked. They were also more likely to have had higher prostate-specific antigen levels and had a higher prevalence of comorbidities.
|
The results of the primary and secondary analyses are listed in Table 2. Compared with nonuse, the use of ADT was not associated with an overall increased risk of dementia (7.4 v 4.4 per 1,000 person-years, respectively; adjusted HR, 1.02; 95% CI, 0.87 to 1.19). In secondary analyses, the risk did not vary with cumulative duration of use, with all HRs close to the null value (P for heterogeneity = .78). This pattern was replicated in the restricted cubic spline model (Appendix Fig A2). Similar findings were observed when assessing the risk by ADT type (Table 2). Finally, we observed similar results when assessing the association with Alzheimer’s disease (adjusted HR, 1.11; 95% CI, 0.85 to 1.44) versus other dementias (adjusted HR, 0.97; 95% CI, 0.80 to 1.18; Appendix Tables A4 and A5). These null associations were also consistent in terms of cumulative duration of use and ADT type.
|
The results of the sensitivity analyses are summarized in Figure 2 and listed in Appendix Tables A4-A8. Lengthening the exposure lag period to 2 and 3 years did not materially change the HRs (adjusted HR, 0.95; 95% CI, 0.81 to 1.12 and adjusted HR, 0.99; 95% CI, 0.82 to 1.18, respectively). Similarly, the HRs were around the null value when excluding and censoring patients using antipsychotics (adjusted HR, 1.00; 95% CI, 0.85 to 1.17) and when excluding and censoring patients treated with chemotherapy (adjusted HR, 1.02; 95% CI, 0.87 to 1.19). Consistent findings were also observed when using multiple imputation (adjusted HR, 1.02; 95% CI, 0.88 to 1.20); when stratifying the model on disease risk score (adjusted HR, 1.02; 95% CI, 0.88 to 1.18); and when using a marginal structural model to account for time-dependent confounding (marginal HR, 1.10; 95% CI, 0.90 to 1.34). Finally, similar results were obtained when accounting for competing risks (subdistribution HR, 0.95; 95% CI, 0.80 to 1.12).
Fig 2. Forest plot displaying adjusted HRs and 95% CIs from the results of the sensitivity analyses to assess the association between androgen deprivation therapy and the incidence of dementia. HR, hazard ratio.
In this large population-based study, the use of ADT was not associated with an increased risk of dementia, including Alzheimer’s disease. We observed similar findings in secondary analyses that assessed the association with duration of use and by specific ADTs. Finally, these findings remained consistent in several sensitivity analyses.
To our knowledge, three observational studies have examined the association between the use of ADT and the risk of dementia or Alzheimer’s disease,12-14 but they had conflicting findings. In the first study, compared with nonuse, ADT use was associated with an increased risk of Alzheimer’s disease (adjusted HR, 1.66; 95% CI, 1.05 to 2.64).12 In this study, the start of follow-up was different between users and nonusers of ADT, which may have introduced immortal time bias.28,29 Specifically, whereas the start of follow-up was the time of initiation of treatment of ADT users, it was the time of diagnosis plus the median time of ADT use for nonusers. Because it is possible that some ADT users may have initiated treatment years after their diagnosis, the nonusers may have had a longer average follow-up than ADT users, thereby inflating the denominator of the rate and leading to an overestimation of the HR. In addition, the exposure definition was solely based on a clinical text pipeline, an approach that may have introduced misclassification. Similarly, in two Taiwanese studies, the use of ADT was associated with nonsignificant increased risks of dementia (HR, 1.21; 95% CI, 0.82 to 1.78)13 and Alzheimer’s disease (HR, 1.76; 95% CI, 0.55 to 5.62).14 In these studies, cohort entry for ADT users was the date of the prescription, whereas cohort entry for nonusers was the date of the prostate cancer diagnosis. As with the previous study,12 immortal time bias was introduced because of the differential cohort entry points for users and nonusers. In this instance, the time between prostate cancer diagnosis and ADT initiation was excluded from the analysis, whereas this person-time should have been classified as unexposed person-time. (A graphical illustration of this bias can be found in Appendix Fig A3). Finally, both of these studies were limited by small sample sizes (1,314 and 1,335, respectively), and there was no consideration given to exposure lag periods.13,14
Our study was specifically designed to circumvent the limitations of the previous studies. First, we assembled a large population-based cohort of men newly diagnosed with prostate cancer that was statistically powered to detect an HR > 1.19 in the primary analysis and > 1.25 for the highest category of cumulative duration of use. Second, the use of the CPRD allowed us to adjust the models for important potential confounders, such as smoking and BMI, variables often missing in administrative databases. Third, we used a time-dependent exposure definition, which eliminated immortal time bias.28,29 In addition, we lagged the exposure by 1 year, an approach meant to minimize misclassifications associated with the timing of dementia, possible detection bias after treatment initiation, and reverse causality and to account for a minimal latency period between treatment initiation and the onset of dementia. Finally, our findings remained consistent in a number of sensitivity analyses that assessed the impact of different sources of bias, such as the use of different exposure lag periods, multiple imputation, marginal structural models to account for possible time-dependent confounding, and disease risk score to control for confounding.
Although previous studies have demonstrated that low testosterone levels are associated with cognitive impairment and Alzheimer’s disease,8,30 the association between ADT and cognitive impairment remains controversial. Indeed, observational studies reporting a positive association between ADT use and different cognitive domains had a number of methodologic limitations, which included small sample sizes; short durations of follow-up; cross-sectional study designs; lack of a comparator group (pre-post study design) or comparator groups consisting of patients without cancer; and lack of adjustment for important confounders including lifestyle variables.7,31-36 Consistent with our findings, a recent meta-analysis found that although patients treated with ADT performed worse on visuomotor tasks in comparison with control subjects or their own baseline assessments, there were no significant differences in cognitive domains that are pertinent to dementia including performance on attention/working memory, executive function, language, verbal memory, visual memory, and visuospatial ability.7 These results were consistent across study designs, with no significant heterogeneity across different cognitive domains.7
Our study has some limitations. First, our event definition was based on diagnoses recorded by general practitioners. However, previous studies have shown that a diagnosis of dementia in the CPRD has a high validity.18-20 Furthermore, the incidence of dementia estimated in our study (six per 1,000 person-years) is similar to that which has been reported for men older than 70 years of age.37 Because of the insidious nature of dementia, there may be misclassifications related to the exact date of onset of the disease. However, we observed consistent results when lengthening the exposure lag period to 2 and 3 years. Prescriptions in the CPRD represent those issued by general practitioners, and thus misclassification of exposure is possible if patients were treated by specialists. However, this misclassification is likely minimal, given the important role of general practitioners in the management of prostate cancer in the United Kingdom.38 This is supported by the finding that > 58% of the cohort was exposed to ADT during the follow-up period, an exposure prevalence that is consistent with that which has been reported in other studies.39 Given the observational nature of the study, residual confounding by unmeasured variables such as years of education,40 socioeconomic status,41 and physical activity42 is possible. However, these variables are modestly associated with the outcome, and it is unclear whether these variables influence the decision to prescribe ADT. Furthermore, we did not have information on tumor stage and grade, although these variables are not established risk factors for dementia. Reassuringly, we obtained consistent results in a sensitivity analysis using marginal structural models to account for time-dependent confounding, which included radical prostatectomy, radiation therapy, and chemotherapy—variables shown to be correlated with tumor grade and stage.43 Finally, on the basis of the Array approach,44 such unmeasured or unknown confounders are unlikely to have biased the observed association under most reasonable assumptions (explained in detail in Appendix Fig A4).
The findings of this large population-based study indicate that the use of ADT is not associated with an increased risk of dementia either overall, by duration of use, or by type. Additional studies in different settings are needed to confirm our findings.
Supported by a foundation grant from the Canadian Institutes of Health Research.
Conception and design: Farzin Khosrow-Khavar, Soham Rej, Armen Aprikian, Laurent Azoulay
Financial support: Laurent Azoulay
Collection and assembly of data: Farzin Khosrow-Khavar, Laurent Azoulay
Data analysis and interpretation: All authors
Manuscript writing: All authors
Final approval of manuscript: All authors
Accountable for all aspects of the work: All authors
The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/site/ifc.
No relationship to disclose
No relationship to disclose
No relationship to disclose
No relationship to disclose
No relationship to disclose
We used the disease risk score (DRS) approach as an alternative means of controlling for confounding.24 This method has been shown to have a performance comparable to that of the propensity score method (Arbogast, et al: Am J Epidemiol 174:613-620, 2011). We first fitted a Cox proportional hazards model that included the potential confounders assessed at baseline, in addition to individual components of the Charlson comorbidity index and the exposure status at baseline. The DRS was then calculated for each patient by taking the products of the regression coefficients and the individual covariate values (with exposure status set to zero) and summing these products. Hazard ratios of dementia comparing the use of androgen deprivation therapy with nonuse were estimated by stratifying the Cox proportional hazards model on DRS deciles. The c statistic for the model was 0.73, which is in line with previous dementia prediction models (range, 0.68 to 0.78; Tang, et al: PLoS One 10: e0136181, 2015).
Marginal structural models account for potential time-dependent confounding over the follow-up period.25,26 Two pooled logistic regression models were fitted to estimate the conditional probability of being exposed to androgen deprivation therapy at 30-day intervals during follow-up, one for the numerator and the other for the denominator of the stabilized inverse probability of treatment weights (IPTWs). The numerator model included baseline covariates (age; year of prostate cancer diagnosis; alcohol-related disorders; smoking status; body mass index; modified Charlson comorbidity index; prostate-specific antigen; head injury; and ever use of antidepressants, antipsychotics, benzodiazepines, and hypnotics); and follow-up time. The denominator model included all covariates included in the numerator model, with the addition of radical prostatectomy, radiation therapy, and chemotherapy measured at each time interval, and follow-up time. The follow-up time variable was modeled using a restricted cubic spline with five knots to minimize bias caused by model misspecification from linearity assumptions. We also estimated the inverse probability of censoring weights (IPCWs) in a similar manner to account for possible informative censoring. Stabilized IPTWs and IPCWs for each patient were computed using the predicted probabilities from the treatment and censoring models. The products of these stabilized IPTWs and IPCWs were used to estimate the marginal hazard ratio of dementia associated with the use of androgen deprivation therapy with 95% CIs calculated using robust variance estimators. All analyses were conducted with SAS version 9.4 (SAS Institute, Cary, NC).
Fig A1. Cohort entry, time-dependent exposure definition, and 1-year exposure lagged analysis. Dashed blue line corresponds to person-time not included in the analysis; dark gold line represents exposed person-time; light gold line represents unexposed person-time. Cohort entry corresponded to incident diagnosis of prostate cancer. However, the start of follow-up (ie, person-time at risk) was 1 year after the diagnosis of prostate cancer. This was to account for delays in diagnosis of dementia and thus to exclude prevalent events. Exposure was lagged by 1 year to account for a minimal latency period and to minimize detection bias. The time between the start of follow-up and the end of the lag period is classified as unexposed person-time, and the time after the end of the 1-year lag is classified as exposed person-time. (A and B) Exposed events (gray solid squares). (C and D) Unexposed events (red solid squares). ADT, androgen deprivation therapy.
Fig A2. Smooth restricted cubic spline curve of adjusted hazard ratio of dementia as a function of continuous duration of androgen deprivation therapy use. ADT, androgen deprivation therapy.
Fig A3. Depiction of immortal time bias. Immortal time is the span of follow-up time during which the event of interest cannot occur because of exposure definition. (A) Immortal time bias caused by exposure misclassification. Cohort entry corresponds to diagnosis of prostate cancer. The time between prostate cancer diagnosis and ADT initiation is classified as exposed person-time. (B) Immortal time bias caused by selection bias. Cohort entry corresponds to ADT initiation in the exposed cohort, whereas cohort entry in the unexposed cohort corresponds to diagnosis of prostate cancer. This differential inclusion criterion for cohort entry leads to exclusion of person-time in the exposed cohort from the analysis and leads to immortal time bias because only the exposed cohort must have survived to ADT initiation. (C) Elimination of immortal time bias caused by time-dependent definition of exposure. Time between prostate cancer diagnosis and ADT initiation is taken into account and classified correctly as unexposed person-time. ADT, androgen deprivation therapy.
Fig A4. Sensitivity analysis for residual confounding using the Array method.44 The graph was constructed on the basis of an observed HR of 1.02 and assuming a confounder prevalence of 0.2 among nonusers of ADT. The association would be strongly confounded by an unknown or unmeasured confounder only if it is strongly associated with the outcome and greatly unbalanced between the exposure groups. Thus, the observed HR of 1.02 is unlikely to have been strongly confounded under the most reasonable assumptions. All models were adjusted for age; alcohol-related disorders; smoking status; body mass index; modified Charlson comorbidity index; prostate-specific antigen; history of head injury; use of antidepressants, antipsychotics, benzodiazepines, and hypnotics; and year of prostate cancer diagnosis. In addition, the marginal structural model adjusted for radical prostatectomy, radiation therapy, and chemotherapy. ADT, androgen deprivation therapy; CD, confounder-disease association; HR, hazard ratio; PC0, confounder prevalence among other non-ADT users; PC1, confounder prevalence among ADT users; RR, risk ratio.
|
|
|
|
|
|
|
|
| 1. | Gilbert SM, Kuo YF, Shahinian VB: Prevalent and incident use of androgen deprivation therapy among men with prostate cancer in the United States. Urol Oncol 29:647-653, 2011 Crossref, Medline, Google Scholar |
| 2. | Schröder F, Crawford ED, Axcrona K, et al: Androgen deprivation therapy: Past, present and future. BJU Int 109:1-12, 2012 Crossref, Medline, Google Scholar |
| 3. | Alibhai SM, Duong-Hua M, Sutradhar R, et al: Impact of androgen deprivation therapy on cardiovascular disease and diabetes. J Clin Oncol 27:3452-3458, 2009 Link, Google Scholar |
| 4. | Lapi F, Azoulay L, Niazi MT, et al: Androgen deprivation therapy and risk of acute kidney injury in patients with prostate cancer. JAMA 310:289-296, 2013 Crossref, Medline, Google Scholar |
| 5. | Shahinian VB, Kuo YF, Freeman JL, et al: Risk of fracture after androgen deprivation for prostate cancer. N Engl J Med 352:154-164, 2005 Crossref, Medline, Google Scholar |
| 6. | Zhao J, Zhu S, Sun L, et al: Androgen deprivation therapy for prostate cancer is associated with cardiovascular morbidity and mortality: A meta-analysis of population-based observational studies. PLoS One 9:e107516, 2014 Crossref, Medline, Google Scholar |
| 7. | McGinty HL, Phillips KM, Jim HS, et al: Cognitive functioning in men receiving androgen deprivation therapy for prostate cancer: A systematic review and meta-analysis. Support Care Cancer 22:2271-2280, 2014 Crossref, Medline, Google Scholar |
| 8. | Holland J, Bandelow S, Hogervorst E: Testosterone levels and cognition in elderly men: A review. Maturitas 69:322-337, 2011 Crossref, Medline, Google Scholar |
| 9. | Hsu B, Cumming RG, Waite LM, et al: Longitudinal relationships between reproductive hormones and cognitive decline in older men: The Concord Health and Ageing in Men Project. J Clin Endocrinol Metab 100:2223-2230, 2015 Crossref, Medline, Google Scholar |
| 10. | Moffat SD, Zonderman AB, Metter EJ, et al: Longitudinal assessment of serum free testosterone concentration predicts memory performance and cognitive status in elderly men. J Clin Endocrinol Metab 87:5001-5007, 2002 Crossref, Medline, Google Scholar |
| 11. | Thilers PP, Macdonald SW, Herlitz A: The association between endogenous free testosterone and cognitive performance: A population-based study in 35 to 90 year-old men and women. Psychoneuroendocrinology 31:565-576, 2006 Crossref, Medline, Google Scholar |
| 12. | Nead KT, Gaskin G, Chester C, et al: Androgen deprivation therapy and future Alzheimer’s disease risk. J Clin Oncol 34:566-571, 2016 Link, Google Scholar |
| 13. | Kao LT, Lin HC, Chung SD, et al: No increased risk of dementia in patients receiving androgen deprivation therapy for prostate cancer: A 5-year follow-up study. Asian J Androl [epub ahead of print on May 27, 2016] Google Scholar |
| 14. | Chung SD, Lin HC, Tsai MC, et al: Androgen deprivation therapy did not increase the risk of Alzheimer’s and Parkinson’s disease in patients with prostate cancer. Andrology 4:481-485, 2016 Crossref, Medline, Google Scholar |
| 15. | Herrett E, Gallagher AM, Bhaskaran K, et al: Data resource profile: Clinical Practice Research Datalink (CPRD). Int J Epidemiol 44:827-836, 2015 Crossref, Medline, Google Scholar |
| 16. | Herrett E, Thomas SL, Schoonen WM, et al: Validation and validity of diagnoses in the General Practice Research Database: A systematic review. Br J Clin Pharmacol 69:4-14, 2010 Crossref, Medline, Google Scholar |
| 17. | Bradford A, Kunik ME, Schulz P, et al: Missed and delayed diagnosis of dementia in primary care: prevalence and contributing factors. Alzheimer Dis Assoc Disord 23:306-314, 2009 Crossref, Medline, Google Scholar |
| 18. | Dunn N, Mullee M, Perry VH, et al: Association between dementia and infectious disease: Evidence from a case-control study. Alzheimer Dis Assoc Disord 19:91-94, 2005 Crossref, Medline, Google Scholar |
| 19. | Seshadri S, Zornberg GL, Derby LE, et al: Postmenopausal estrogen replacement therapy and the risk of Alzheimer disease. Arch Neurol 58:435-440, 2001 Crossref, Medline, Google Scholar |
| 20. | Jick H, Zornberg GL, Jick SS, et al: Statins and the risk of dementia. Lancet 356:1627-1631, 2000 Crossref, Medline, Google Scholar |
| 21. | Deyo RA, Cherkin DC, Ciol MA: Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases. J Clin Epidemiol 45:613-619, 1992 Crossref, Medline, Google Scholar |
| 22. | Heinzl H, Kaider A, Zlabinger G: Assessing interactions of binary time-dependent covariates with time in Cox proportional hazards regression models using cubic spline functions. Stat Med 15:2589-2601, 1996 Crossref, Medline, Google Scholar |
| 23. | Rubin DB: Multiple Imputation for Nonresponse in Surveys. New York: John Wiley & Sons, 1987 Google Scholar |
| 24. | Glynn RJ, Gagne JJ, Schneeweiss S: Role of disease risk scores in comparative effectiveness research with emerging therapies. Pharmacoepidemiol Drug Saf 21:138-147, 2012 Crossref, Medline, Google Scholar |
| 25. | Hernán MA, Brumback B, Robins JM: Marginal structural models to estimate the causal effect of zidovudine on the survival of HIV-positive men. Epidemiology 11:561-570, 2000 Crossref, Medline, Google Scholar |
| 26. | Platt RW, Delaney JA, Suissa S: The positivity assumption and marginal structural models: The example of warfarin use and risk of bleeding. Eur J Epidemiol 27:77-83, 2012 Crossref, Medline, Google Scholar |
| 27. | Fine JP, Gray RJ: A proportional hazards model for the subdistribution of a competing risk. J Am Stat Assoc 94:496-509, 1999 Google Scholar |
| 28. | Suissa S: Immortal time bias in pharmaco-epidemiology. Am J Epidemiol 167:492-499, 2008 Crossref, Medline, Google Scholar |
| 29. | Giobbie-Hurder A, Gelber RD, Regan MM: Challenges of guarantee-time bias. J Clin Oncol 31:2963-2969, 2013 Link, Google Scholar |
| 30. | Lv W, Du N, Liu Y, et al: Low testosterone level and risk of Alzheimer’s disease in the elderly men: A systematic review and meta-analysis. Mol Neurobiol 53:2679-2684, 2016 Crossref, Medline, Google Scholar |
| 31. | Gonzalez BD, Jim HS, Booth-Jones M, et al: Course and predictors of cognitive function in patients with prostate cancer receiving androgen-deprivation therapy: A controlled comparison. J Clin Oncol 33:2021-2027, 2015 Link, Google Scholar |
| 32. | Chao HH, Uchio E, Zhang S, et al: Effects of androgen deprivation on brain function in prostate cancer patients - a prospective observational cohort analysis. BMC Cancer 12:371, 2012 Crossref, Medline, Google Scholar |
| 33. | Green HJ, Pakenham KI, Headley BC, et al: Altered cognitive function in men treated for prostate cancer with luteinizing hormone-releasing hormone analogues and cyproterone acetate: A randomized controlled trial. BJU Int 90:427-432, 2002 Crossref, Medline, Google Scholar |
| 34. | Green HJ, Pakenham KI, Headley BC, et al: Quality of life compared during pharmacological treatments and clinical monitoring for non-localized prostate cancer: A randomized controlled trial. BJU Int 93:975-979, 2004 Crossref, Medline, Google Scholar |
| 35. | Yang J, Zhong F, Qiu J, et al: Dissociation of event-based prospective memory and time-based prospective memory in patients with prostate cancer receiving androgen-deprivation therapy: A neuropsychological study. Eur J Cancer Care (Engl) 24:198-204, 2015 Crossref, Medline, Google Scholar |
| 36. | Wu LM, Tanenbaum ML, Dijkers MP, et al: Cognitive and neurobehavioral symptoms in patients with non-metastatic prostate cancer treated with androgen deprivation therapy or observation: A mixed methods study. Soc Sci Med 156:80-89, 2016 Crossref, Medline, Google Scholar |
| 37. | Satizabal CL, Beiser AS, Chouraki V, et al: Incidence of dementia over three decades in the Framingham Heart Study. N Engl J Med 374:523-532, 2016 Crossref, Medline, Google Scholar |
| 38. | Neal RD: NICE prostate cancer clinical guideline: Implications for primary care. Br J Gen Pract 58:607-608, 2008 Crossref, Medline, Google Scholar |
| 39. | Keating NL, O’Malley AJ, Smith MR: Diabetes and cardiovascular disease during androgen deprivation therapy for prostate cancer. J Clin Oncol 24:4448-4456, 2006 Link, Google Scholar |
| 40. | Xu W, Tan L, Wang HF, et al: Education and risk of dementia: Dose-response meta-analysis of prospective cohort studies. Mol Neurobiol 53:3113-3123, 2016 Crossref, Medline, Google Scholar |
| 41. | Zeki Al Hazzouri A, Haan MN, Kalbfleisch JD, et al: Life-course socioeconomic position and incidence of dementia and cognitive impairment without dementia in older Mexican Americans: Results from the Sacramento area Latino study on aging. Am J Epidemiol 173:1148-1158, 2011 Crossref, Medline, Google Scholar |
| 42. | Blondell SJ, Hammersley-Mather R, Veerman JL: Does physical activity prevent cognitive decline and dementia? A systematic review and meta-analysis of longitudinal studies. BMC Public Health 14:510, 2014 Crossref, Medline, Google Scholar |
| 43. | National Institute for Health and Care Excellence: Prostate cancer: Diagnosis and management. https://www.nice.org.uk/guidance/cg175 Google Scholar |
| 44. | Schneeweiss S: Sensitivity analysis and external adjustment for unmeasured confounders in epidemiologic database studies of therapeutics. Pharmacoepidemiol Drug Saf 15:291-303, 2006 Crossref, Medline, Google Scholar |
