- Email: [email protected]
Re: Yeung et al.: β-blockers and neovascular age-related macular degeneration (Ophthalmology. 2017;124:409-411)
Ophthalmology Volume 124, Number 9, September 2017 Available online: June 15, 2017. Correspondence: Joshua D. Stein, MD, MS, Associate Professor, Ophthalmology and Visual Sciences, University of Michigan, Department of Ophthalmology and Visual Sciences, Kellogg Eye Center, 1000 Wall Street, Ann Arbor, MI 48105. E-mail: [email protected].
References 1. Wang SY, Andrews CA, Herman WH, et al. Incidence and risk factors for developing diabetic retinopathy among youths with type 1 or type 2 diabetes throughout the United States. Ophthalmology. 2017;124:424-430. 2. Klein R, Klein BE, Moss SE, et al. The Wisconsin epidemiologic study of diabetic retinopathy. II. Prevalence and risk of diabetic retinopathy when age at diagnosis is less than 30 years. Arch Ophthalmol. 1984;102:520-526.
Re: Yeung et al.: b-blockers and neovascular age-related macular degeneration (Ophthalmology. 2017;124:409-411) We read with great interest the report by Yeung et al1 regarding the increased hazard among b-blocker (BB) users for neovascular age-related macular degeneration (nAMD).1 Certainly, the frequency with which BB are used makes the results of their study concerning. However, given the potential ramifications, we hope for further clarity on numerous issues regarding the study. First, several questions arise based on the authors’ use of their propensity score. “The propensity score is the probability of treatment assignment (in this case initiation of a BB vs. not) conditional on observed baseline characteristics.”2 It is curious then why the authors chose to make the date of hypertension (HTN) diagnosis as the index date and not the date of BB initiation. With this seemingly innocuous switch, they are introducing a concerning potential source of bias since it is unclear how eligibility of patients would change as covariates change with time. Additionally, propensity scores balance covariates over the entire matched set, which may not continue to hold true once the authors subdivided cohort 1 into 1A and 1B. Last, it is unclear what the authors intended as the alternative to a propensity of initiating BB therapy. Cohort 2 was defined as “usage of 1 non-b-blocker antihypertensive medications for 270 days per year”1; however, is this someone who is receiving HTN care at baseline without initiation of new anti-HTN therapy? Ideally, the propensity score would have been used to select a patient who had a specific propensity to receive a BB, yet, instead received another class of anti-HTN medication that was selected a priori (such as calciumchannel blockers). This step alone may have improved the distribution of baseline covariates and better equalized the cohorts.3 Furthermore, to create the best comparison, because cohort 1 patients were required to take a BB for a specified amount of time before inclusion, the alternative HTN class of medication should have also been mandated to occur over a similar time frame. This would reduce the “healthy user effect,” because it is well-known that patients who are adherent to medications have different underlying disease risk rates than those who are not compliant.4 TO THE EDITOR:
e70
Next, although the use of initial HTN diagnosis for the index date allows for a convenient point of comparison across cohorts, this choice introduces other concerns beyond covariate balance. First, it may not accurately reflect the severity of HTN at the time of BB initiation owing to the potentially large gap between the start of study observation time and initiation of BB therapy. How this time in each cohort was accounted for is unclear. It is conceivable that a large gap in time between HTN diagnosis and BB initiation could represent a period of poorly controlled HTN with multiple medication trials. Conversely, it could indicate relatively stable HTN course during which an additional diagnosis (e.g., atrial fibrillation) was made for which BB therapy was indicated in conjunction with maintaining HTN control. Given these uncertainties, it is surprising that the authors felt confident enough to unequivocally state, “Thus, hypertension itself is no longer a confounding factor.”1 Although the authors did attempt to match the cohorts by “hypertensionrelated end-organ diseases and the number of antihypertensive medications used,”1 no data are reported to support their conclusion that the severity of HTN as a source of bias has been negated. Last, the number and baseline characteristics of cohort 1 patients who were split into 1A and 1B were never specified, nor do the authors state how many patients (total or by each cohort) developed nAMD. Also of note is how the authors dealt with dry AMD in their cohorts. It is possible that one of the underlying indications for BB use had an independent association with dry AMD, which in turn may have led to an imbalance in the prevalence of dry AMD between the 2 cohorts. If large enough, an unequal distribution of dry AMD between comparators could easily have contributed to the higher hazard ratio for the development of nAMD seen in BB users. Ideally, the authors would have restricted this study to only patients with known dry AMD status because it is a necessary intermediary step to developing nAMD.
ANTON M. KOLOMEYER, MD, PHD1 ANASTASIA TRABAND, MD1 BRIAN L. VANDERBEEK, MD, MPH1,2,3 1 Scheie Eye Institute, Department of Ophthalmology, University of Pennsylvania Perelman School of Medicine, Philadelphia, Pennsylvania; 2Center for Clinical Epidemiology and Biostatistics, Department of Biostatistics & Epidemiology, University of Pennsylvania Perelman School of Medicine, Philadelphia, Pennsylvania; 3Leonard Davis Institute, University of Pennsylvania Perelman School of Medicine, Philadelphia, Pennsylvania
Financial Disclosures: The authors have no proprietary or commercial interest in any materials discussed in this article. Funded by the National Institutes of Health K23 Award (1K23EY025729 - 01). The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. Additional funding was provided by Research to Prevent Blindness and the Paul and Evanina Mackall Foundation. Available online: June 15, 2017. Correspondence: Brian L. VanderBeek, MD, MPH, Scheie Eye Institute, 51 North 39th Street, Philadelphia, PA 19104. E-mail: [email protected].
References 1. Yeung L, Huang TS, Lin YH, et al. b-blockers and neovascular agerelated macular degeneration. Ophthalmology. 2017;124:409-411.
Correspondence 2. Austin PC. An introduction to propensity score methods for reducing the effects of confounding in observational studies. Multivariate Behav Res. 2011;46:399-424. 3. Traband A, Shaffer JA, VanderBeek BL. Systemic beta-blockers in neovascular age-related macular degeneration. Retina. 2017;37: 41-46. 4. Dormuth CR, Patrick AR, Shrank WH, et al. Statin adherence and risk of accidents: a cautionary tale. Circulation. 2009;119: 2051-2057. REPLY: We thank Kolomeyer et al for their interest in our study, and for providing the opportunity to further clarify our methods and results. The purpose of our study was to evaluate the effect of the continuous use of b-blockers on the cumulative incidence of new-onset neovascular age-related macular degeneration (nAMD) in hypertensive patients. The date of the first diagnosis of hypertension was used as the index date for enrollment, which helped to ensure the duration of hypertension in each patient. The patients were then classified into b-blocker users and noneb-blocker users according to their initial antihypertensive medications. The patients diagnosed with hypertension but who did not receive any anti-hypertensive medications were excluded from the study according to the exclusion criterion: “(5) inadequate exposure time of antihypertensive medications.” Therefore, the date of the diagnosis of hypertension was actually similar to the date of initiating antihypertensive medications. Furthermore, in our statistical analysis we excluded anyone who suffered from nAMD events within 180 days after the diagnosis of hypertension in both groups (see Figure 11). Therefore, our analysis approximated an incident user design to avoid immortal time bias.2 Patients switching from a noneb-blocker regimen to a b-blocker regimen (or vice versa) during follow-up were excluded according to the exclusion criterion: “(3) switch to different regimens during follow-up.” In addition, the number of anti-hypertensive medications used in each cohort was well balanced through propensity score matching. Thus, it is unlikely that 1 cohort had a higher likelihood of initiating a new anti-hypertensive regimen than the other. To better compare the b-blocker users and noneb-blocker users, we balanced the patients in cohorts 1 and 2 using propensity score matching. To demonstrate the effects of different durations of exposure to b-blockers, we further divided cohort 1 into 1A (180e269 days per year) and 1B (270 days per year) subgroups. We agree with Kolomeyer et al that dividing cohort 1 into 2 subgroups may have resulted in an imbalanced distribution. To address this concern, we reanalyzed the data comparing cohorts 1 and 2. The Cox regression model showed an increased hazard ratio of nAMD in cohort 1 (adjusted hazard ratio, 1.94; 95% confidence interval, 1.54e2.43; P < 0.001) compared with cohort 2. In addition, we also reconducted propensity score matching in comparing cohorts 1A and 1B with cohort 2, and we obtained similar results to the original study. The patients in cohort 1A had a higher risk of nAMD (adjusted hazard ratio, 1.54; 95% confidence interval, 1.15e2.05; P ¼ 0.003) compared with cohort 2, and the patients in cohort 1B (270 days per year) had a further increased risk of nAMD (adjusted hazard ratio, 2.18; 95% confidence interval, 1.52e3.12; P < 0.001) compared with cohort 2. Proportional assumptions of these models were all valid. As stated in the limitations section of the study, the severity of hypertension is not available in the National Health Insurance
Research Database in Taiwan. However, it is well-known that blood pressure has a linear relationship with both cardiovascular disease and stroke.3 For individuals aged 40 to 89 years, for every 20mmHg increase in systolic blood pressure or 10-mmHg increase in diastolic blood pressure, the risk of mortality doubles from both cardiovascular disease and stroke.3 Therefore, we believe that it is reasonable to use these end-organ diseases as a surrogate for the severity of hypertension. We did not evaluate the incidence of dry age-related macular degeneration (AMD) in this study. This is because most cases of dry AMD are asymptomatic and progress slowly, and these patients may not seek medical help until they are at an advanced stage. Therefore, the incidence of dry AMD would be difficult to evaluate using data from a claims database. Furthermore, dry AMD and nAMD may involve different pathologic pathways and could be associated with different systemic diseases. For example, a recent study showed that Alzheimer’s disease and the development of senile dementia were associated with nonexudative AMD but not exudative AMD.4 Nevertheless, a causal association between the drugs of interest and the risk of nAMD could not be inferred from our study results due to the use of a claims database. To elucidate this issue, we encourage experts in the field to conduct further well-designed observational studies using causal diagrams, controlling for timedependent variables, adjusting for time-varying confounders, and considering informative censoring for competing events and death.5
LING YEUNG, MD1,2 TING-SHUO HUANG, MD, PHD3,4,5,6 CHI-CHIN SUN, MD, PHD1,4,5 1 Department of Ophthalmology, Chang Gung Memorial Hospital, Keelung, Taiwan; 2College of Medicine, Chang Gung University, Taoyuan, Taiwan; 3Department of General Surgery, Chang Gung Memorial Hospital, Keelung, Taiwan; 4Department of Chinese Medicine, College of Medicine, Chang Gung University, Taoyuan, Taiwan; 5Osteoporosis Prevention and Treatment Center, Chang Gung Memorial Hospital, Keelung, Taiwan; 6Community Medicine Research Center, Chang Gung Memorial Hospital, Keelung, Taiwan
Financial Disclosures: The authors have no proprietary or commercial interest in any materials discussed in this article. Available online: June 15, 2017. Correspondence: Chi-Chin Sun, MD, PhD, Department of Ophthalmology, Chang Gung Memorial Hospital, 222 Mai Chin Road, An Leh District, Keelung, Taiwan. E-mail: [email protected].
References 1. Yeung L, Huang TS, Lin YH, et al. b-blockers and neovascular agerelated macular degeneration. Ophthalmology. 2017;124(3):409-411. 2. Suissa S. Immortal time bias in observational studies of drug effects. Pharmacoepidemiol Drug Saf. 2007;16(3):241-249. 3. Lewington S, Clarke R, Qizilbash N, et al. Age-specific relevance of usual blood pressure to vascular mortality: a meta-analysis of individual data for one million adults in 61 prospective studies. Prospective Studies Collaboration. Lancet. 2002;360:1903-1913. 4. Tsai DC, Chen SJ, Huang CC, et al. Age-related macular degeneration and risk of degenerative dementia among the
e71
References 1. Wang SY, Andrews CA, Herman WH, et al. Incidence and risk factors for developing diabetic retinopathy among youths with type 1 or type 2 diabetes throughout the United States. Ophthalmology. 2017;124:424-430. 2. Klein R, Klein BE, Moss SE, et al. The Wisconsin epidemiologic study of diabetic retinopathy. II. Prevalence and risk of diabetic retinopathy when age at diagnosis is less than 30 years. Arch Ophthalmol. 1984;102:520-526.
Re: Yeung et al.: b-blockers and neovascular age-related macular degeneration (Ophthalmology. 2017;124:409-411) We read with great interest the report by Yeung et al1 regarding the increased hazard among b-blocker (BB) users for neovascular age-related macular degeneration (nAMD).1 Certainly, the frequency with which BB are used makes the results of their study concerning. However, given the potential ramifications, we hope for further clarity on numerous issues regarding the study. First, several questions arise based on the authors’ use of their propensity score. “The propensity score is the probability of treatment assignment (in this case initiation of a BB vs. not) conditional on observed baseline characteristics.”2 It is curious then why the authors chose to make the date of hypertension (HTN) diagnosis as the index date and not the date of BB initiation. With this seemingly innocuous switch, they are introducing a concerning potential source of bias since it is unclear how eligibility of patients would change as covariates change with time. Additionally, propensity scores balance covariates over the entire matched set, which may not continue to hold true once the authors subdivided cohort 1 into 1A and 1B. Last, it is unclear what the authors intended as the alternative to a propensity of initiating BB therapy. Cohort 2 was defined as “usage of 1 non-b-blocker antihypertensive medications for 270 days per year”1; however, is this someone who is receiving HTN care at baseline without initiation of new anti-HTN therapy? Ideally, the propensity score would have been used to select a patient who had a specific propensity to receive a BB, yet, instead received another class of anti-HTN medication that was selected a priori (such as calciumchannel blockers). This step alone may have improved the distribution of baseline covariates and better equalized the cohorts.3 Furthermore, to create the best comparison, because cohort 1 patients were required to take a BB for a specified amount of time before inclusion, the alternative HTN class of medication should have also been mandated to occur over a similar time frame. This would reduce the “healthy user effect,” because it is well-known that patients who are adherent to medications have different underlying disease risk rates than those who are not compliant.4 TO THE EDITOR:
e70
Next, although the use of initial HTN diagnosis for the index date allows for a convenient point of comparison across cohorts, this choice introduces other concerns beyond covariate balance. First, it may not accurately reflect the severity of HTN at the time of BB initiation owing to the potentially large gap between the start of study observation time and initiation of BB therapy. How this time in each cohort was accounted for is unclear. It is conceivable that a large gap in time between HTN diagnosis and BB initiation could represent a period of poorly controlled HTN with multiple medication trials. Conversely, it could indicate relatively stable HTN course during which an additional diagnosis (e.g., atrial fibrillation) was made for which BB therapy was indicated in conjunction with maintaining HTN control. Given these uncertainties, it is surprising that the authors felt confident enough to unequivocally state, “Thus, hypertension itself is no longer a confounding factor.”1 Although the authors did attempt to match the cohorts by “hypertensionrelated end-organ diseases and the number of antihypertensive medications used,”1 no data are reported to support their conclusion that the severity of HTN as a source of bias has been negated. Last, the number and baseline characteristics of cohort 1 patients who were split into 1A and 1B were never specified, nor do the authors state how many patients (total or by each cohort) developed nAMD. Also of note is how the authors dealt with dry AMD in their cohorts. It is possible that one of the underlying indications for BB use had an independent association with dry AMD, which in turn may have led to an imbalance in the prevalence of dry AMD between the 2 cohorts. If large enough, an unequal distribution of dry AMD between comparators could easily have contributed to the higher hazard ratio for the development of nAMD seen in BB users. Ideally, the authors would have restricted this study to only patients with known dry AMD status because it is a necessary intermediary step to developing nAMD.
ANTON M. KOLOMEYER, MD, PHD1 ANASTASIA TRABAND, MD1 BRIAN L. VANDERBEEK, MD, MPH1,2,3 1 Scheie Eye Institute, Department of Ophthalmology, University of Pennsylvania Perelman School of Medicine, Philadelphia, Pennsylvania; 2Center for Clinical Epidemiology and Biostatistics, Department of Biostatistics & Epidemiology, University of Pennsylvania Perelman School of Medicine, Philadelphia, Pennsylvania; 3Leonard Davis Institute, University of Pennsylvania Perelman School of Medicine, Philadelphia, Pennsylvania
Financial Disclosures: The authors have no proprietary or commercial interest in any materials discussed in this article. Funded by the National Institutes of Health K23 Award (1K23EY025729 - 01). The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. Additional funding was provided by Research to Prevent Blindness and the Paul and Evanina Mackall Foundation. Available online: June 15, 2017. Correspondence: Brian L. VanderBeek, MD, MPH, Scheie Eye Institute, 51 North 39th Street, Philadelphia, PA 19104. E-mail: [email protected].
References 1. Yeung L, Huang TS, Lin YH, et al. b-blockers and neovascular agerelated macular degeneration. Ophthalmology. 2017;124:409-411.
Correspondence 2. Austin PC. An introduction to propensity score methods for reducing the effects of confounding in observational studies. Multivariate Behav Res. 2011;46:399-424. 3. Traband A, Shaffer JA, VanderBeek BL. Systemic beta-blockers in neovascular age-related macular degeneration. Retina. 2017;37: 41-46. 4. Dormuth CR, Patrick AR, Shrank WH, et al. Statin adherence and risk of accidents: a cautionary tale. Circulation. 2009;119: 2051-2057. REPLY: We thank Kolomeyer et al for their interest in our study, and for providing the opportunity to further clarify our methods and results. The purpose of our study was to evaluate the effect of the continuous use of b-blockers on the cumulative incidence of new-onset neovascular age-related macular degeneration (nAMD) in hypertensive patients. The date of the first diagnosis of hypertension was used as the index date for enrollment, which helped to ensure the duration of hypertension in each patient. The patients were then classified into b-blocker users and noneb-blocker users according to their initial antihypertensive medications. The patients diagnosed with hypertension but who did not receive any anti-hypertensive medications were excluded from the study according to the exclusion criterion: “(5) inadequate exposure time of antihypertensive medications.” Therefore, the date of the diagnosis of hypertension was actually similar to the date of initiating antihypertensive medications. Furthermore, in our statistical analysis we excluded anyone who suffered from nAMD events within 180 days after the diagnosis of hypertension in both groups (see Figure 11). Therefore, our analysis approximated an incident user design to avoid immortal time bias.2 Patients switching from a noneb-blocker regimen to a b-blocker regimen (or vice versa) during follow-up were excluded according to the exclusion criterion: “(3) switch to different regimens during follow-up.” In addition, the number of anti-hypertensive medications used in each cohort was well balanced through propensity score matching. Thus, it is unlikely that 1 cohort had a higher likelihood of initiating a new anti-hypertensive regimen than the other. To better compare the b-blocker users and noneb-blocker users, we balanced the patients in cohorts 1 and 2 using propensity score matching. To demonstrate the effects of different durations of exposure to b-blockers, we further divided cohort 1 into 1A (180e269 days per year) and 1B (270 days per year) subgroups. We agree with Kolomeyer et al that dividing cohort 1 into 2 subgroups may have resulted in an imbalanced distribution. To address this concern, we reanalyzed the data comparing cohorts 1 and 2. The Cox regression model showed an increased hazard ratio of nAMD in cohort 1 (adjusted hazard ratio, 1.94; 95% confidence interval, 1.54e2.43; P < 0.001) compared with cohort 2. In addition, we also reconducted propensity score matching in comparing cohorts 1A and 1B with cohort 2, and we obtained similar results to the original study. The patients in cohort 1A had a higher risk of nAMD (adjusted hazard ratio, 1.54; 95% confidence interval, 1.15e2.05; P ¼ 0.003) compared with cohort 2, and the patients in cohort 1B (270 days per year) had a further increased risk of nAMD (adjusted hazard ratio, 2.18; 95% confidence interval, 1.52e3.12; P < 0.001) compared with cohort 2. Proportional assumptions of these models were all valid. As stated in the limitations section of the study, the severity of hypertension is not available in the National Health Insurance
Research Database in Taiwan. However, it is well-known that blood pressure has a linear relationship with both cardiovascular disease and stroke.3 For individuals aged 40 to 89 years, for every 20mmHg increase in systolic blood pressure or 10-mmHg increase in diastolic blood pressure, the risk of mortality doubles from both cardiovascular disease and stroke.3 Therefore, we believe that it is reasonable to use these end-organ diseases as a surrogate for the severity of hypertension. We did not evaluate the incidence of dry age-related macular degeneration (AMD) in this study. This is because most cases of dry AMD are asymptomatic and progress slowly, and these patients may not seek medical help until they are at an advanced stage. Therefore, the incidence of dry AMD would be difficult to evaluate using data from a claims database. Furthermore, dry AMD and nAMD may involve different pathologic pathways and could be associated with different systemic diseases. For example, a recent study showed that Alzheimer’s disease and the development of senile dementia were associated with nonexudative AMD but not exudative AMD.4 Nevertheless, a causal association between the drugs of interest and the risk of nAMD could not be inferred from our study results due to the use of a claims database. To elucidate this issue, we encourage experts in the field to conduct further well-designed observational studies using causal diagrams, controlling for timedependent variables, adjusting for time-varying confounders, and considering informative censoring for competing events and death.5
LING YEUNG, MD1,2 TING-SHUO HUANG, MD, PHD3,4,5,6 CHI-CHIN SUN, MD, PHD1,4,5 1 Department of Ophthalmology, Chang Gung Memorial Hospital, Keelung, Taiwan; 2College of Medicine, Chang Gung University, Taoyuan, Taiwan; 3Department of General Surgery, Chang Gung Memorial Hospital, Keelung, Taiwan; 4Department of Chinese Medicine, College of Medicine, Chang Gung University, Taoyuan, Taiwan; 5Osteoporosis Prevention and Treatment Center, Chang Gung Memorial Hospital, Keelung, Taiwan; 6Community Medicine Research Center, Chang Gung Memorial Hospital, Keelung, Taiwan
Financial Disclosures: The authors have no proprietary or commercial interest in any materials discussed in this article. Available online: June 15, 2017. Correspondence: Chi-Chin Sun, MD, PhD, Department of Ophthalmology, Chang Gung Memorial Hospital, 222 Mai Chin Road, An Leh District, Keelung, Taiwan. E-mail: [email protected].
References 1. Yeung L, Huang TS, Lin YH, et al. b-blockers and neovascular agerelated macular degeneration. Ophthalmology. 2017;124(3):409-411. 2. Suissa S. Immortal time bias in observational studies of drug effects. Pharmacoepidemiol Drug Saf. 2007;16(3):241-249. 3. Lewington S, Clarke R, Qizilbash N, et al. Age-specific relevance of usual blood pressure to vascular mortality: a meta-analysis of individual data for one million adults in 61 prospective studies. Prospective Studies Collaboration. Lancet. 2002;360:1903-1913. 4. Tsai DC, Chen SJ, Huang CC, et al. Age-related macular degeneration and risk of degenerative dementia among the
e71