Ophthalmology
Volume 99, Number 12, December 1992
bacterium acnes endophthalmitis even lacking of struc tures (residual lens or capsular materials) in which the bacterium could be sequestered within and the favorable prognosis despite the chronic course with periodic severe relapses. Furthermore we would like to emphasize the efficacy of a noninvasive, orally administered antibiotic therapy (chloramphenicol 1 gjday) for the treatment of culture-proven P. acnes endophthalmitis and the good results, after a complete resolution of the uveitis, of an argon lasertrabeculoplasty in controlling lOP in such cases in which a physical barrier (fibrin, inflammatory cells) to the normal aqueous outflow may be proposed as the caus ative agent of the arised lOP. PAOLA PIVETTI-PEZZI, MD MASSIMO ACCORINTI, MD Rome, Italy References
I. Chien AM, Raber IM, Fisher DH, eta!. Propionibacterium acnes endophthalmitis after intracapsular cataract extrac tion. Ophthalmology 1992;99:487-90. 2. Meisler OM, Mandelbaum S. Propionibacterium-associated endophthalmitis after extracapsular cataract extraction: re view of reported cases. Ophthalmology 1989;96:54-61. 3. Zambrano W, Flynn HW Jr, Ptlugfelder SC, eta!. Man agement options for Propionibacterium acnes endophthal mitis. Ophthalmology 1989;96:1100-5.
Author's reply
Dear Editor: We appreciate the comments of Dr. Pivetti-Pezzi regard ing the successful treatment of a case of culture-proven Propionibacterium acnes endophthalmitis with oral chloramphenicol. However, in the United States, given the small but finite risk of irreversible aplastic anemia associated with chloramphenicol and the large number of antibiotic options available, chloramphenicol would not be the first agent of choice. The cephalosporins, semisyn thetic penicillins, vancomycin, clindamycin, and eryth romycin have been shown to be effective against P. acnes. There have been at least two cases in the literature in which oral cephalosporins were used in conjunction with topical cephalosporins to successfully treat recurrences of culture-positive cases of P. acnes endophthalmitis. 1•2 Dr. Pivetti-Pezzi does report an initial trial of oral cephalo sporins in his case, but the length of treatment is not spec ified. ANN M. CHIEN, MD IRVING M. RABER, MD DAVID H. FISCHER, MD RALPH C. EAGLE, JR., MD MICHAEL A. NAIDOFF, MD Philadelphia, Pennsylvania References
I. Brady SE, Cohen EJ, Fischer DH. Diagnosis and treatment of chronic postoperative bacterial endophthalmitis. Ophthalmic Surg 1988;19:580-4. 2. Zambrano W, Flynn HW Jr, Pflugfelder SC, et a!. Man agement options for Propionibacterium acnes endophthal mitis. Ophthalmology 1989;96:1100-5.
1754
Case-Control Study of Retinal Vein Occlusion Dear Editor: We read with interest the recent article by Rath and co workers entitled "Risk Factors for Retinal Vein Occlu sion" (Ophthalmology 1992;99:509-514) in which they investigate possible risk factors for retinal vein occlusion by means of a case-control study. The potential associ ations between retinal vein occlusions and systemic vas culopathies have been widely reported, but case-control studies in this area are few. 1. 2 Dr. Rath and co-workers compare the relative frequencies of a number of systemic and ocular conditions among a series of 87 subjects with retinal vein occlusions ("cases") drawn from their pho tography files with the frequencies among a group of 85 subjects without retinal vein occlusion ("controls") drawn from the patient files of two "general ophthalmologists" in their department. They conclude that systemic hyper tension, chronic open-angle glaucoma, and male sex are significant "risk factors" for retinal vein occlusion, but that race, diabetes, and coronary artery disease are not. We believe there are design flaws in the study that render some of these conclusions open to question. The major weakness of their construct relates to their choice ofcontrols. First, it is altogether unclear what were the diagnoses ofthe subjects in their control group? With out limiting inclusion criteria to "nonorganic" complaints (for example refractive needs), the contention that an ophthalmology practice can serve as the source of age matched controls is suspect at best, and is fraught with confounders. 1 For example, 64% of their controls were females, leading to the conclusion that male sex is a risk factor for vein occlusion. However, a significantly in creased prevalence of diabetic retinopathy, cataract, and possibly macular degeneration, has been reported among females even when controlling for age. 3•4 A bias in favor of one gender in the control group can thereby confound the role of sex as a risk factor among cases. The role of race is similarly obscured. Despite an in creased prevalence of blacks among cases, black race was not shown to be significantly associated with vein occlu sion. This could represent a selection bias in favor of the black race among the controls where an unknown number could have had glaucoma or diabetic retinopathy. Fur thermore, the authors' assertion that the known associa tion between black race and hypertension on the one hand, and hypertension and vein occlusion on the other, pre cludes black race acting as an "independent risk factor" for vein occlusion is false. For example, black race and hypertension are independent risk factors for open-angle glaucoma5 despite a significant unrelated interaction be tween black race and hypertension. In addition to sex and race, diabetic status also can be a serious confounder in a general ophthalmic practice. 1•2 This effect is potentially accentuated by a preponderance of diabetes6 and diabetic retinopathy, 3 particularly among older females who also accounted for the majority of the control subjects used. Most likely, it was exactly because of this bias in favor of diabetes among the control group that a previous studyi failed to show a significant asso
Letters to the Editor ciation between diabetes and retinal vein occlusion when using a series of cataract patients as controls, but reached the opposite conclusion when comparing the cases to sta tistics from a series of subjects from the general popula tion. The utility oflogistic analysis in excluding interactions among multiple confounders, particularly in a small series, is limited-a need that may be obviated by the matching of certain variables. 7 This is why logistic regression ac cording to the authors' own contention failed to show any "significant interactions among [the] variables" tested, despite all epidemiologic evidence to the contrary; as for example between race and hypertension, or race and glaucoma. 5·6 Matching alone clearly does not offer the perfect so lution to such problems, as it would preclude the assess ment of the matched variables' effects on the risk of dis ease. Short of a population-based prevalence study, risk factor analysis can still be conducted effectively when the case-control selection is free of obvious concurrent con founders. However, when many confounding variables are left unmatched, and the randomization process and control definition are poorly defined, particularly in a set ting of small sample sizes, caution is urged in the inter pretation of such data especially related to the exclusion and ranking of potential risk factors. MOHAMAD R. DANA, MD, MPH KAY ROSHEIM, MPH Chicago, Illinois References 1. Elman MJ, Bhatt AK, Quinlan PM, Enger C. The risk for systemic vascular diseases and mortality in patients with central retinal vein occlusion. Ophthalmology 1990;97: 1543-8. 2. Johnston RL, Brucker AJ, Steinmann W, eta!. Risk factors of branch retinal vein occlusion. Arch Ophthalmol 1985; 103: 1831-2. 3. Kahn HA, Moorhead HB. Statistics on Blindness in the Model Reporting Area 1969-70. Washington, DC: National Institutes of Health, 1973. (DHEW pub!; no (NIH) 73-427.) 4. Leibowitz HM, Krueger DE, Maunder LR, et a!. The Fra mingham Eye Study Monograph. Surv Ophthalmol 1980;24(Suppl):335-61 0. 5. Tielsch JM, Sommer A, Katz J, et a!. Racial variations in the prevalence of primary open-angle glaucoma. The Bal timore Eye Survey. JAMA 1991;266:369-74. 6. Whelton PK, Russell RP. Systemic hypertension. In: Harvey AM, Johns RJ, McKusick VA, et a!, eds. The Principles and Practice of Medicine, 21st ed. Norwalk, CT: Appleton Century-Crofts, 1984; chap. 25. 7. Schlesselman JJ. Case-Control Studies: Design, Conduct, Analysis. New York: Oxford University Press, 1982;115 23.
Authors' reply
Dear Editor: Dana and Rosheim raise a number of critical issues re garding our article. Their major criticism deals with our choice of a control group, which they claim was done with the "randomization process and control definition
poorly defined," and was "fraught with confounders." The choice of appropriate control groups for case-control studies is difficult, as we pointed out in our article (p. 512), and as others have also noted at length. 1.2 We chose our control group from patients of two general ophthal mologists in our department, largely for the sake of con venience: the medical records of these individuals, con taining the data we believed were relevant to our study, were readily available to us. Because they were patients of general ophthalmologists, we believed that they would not have an unusual prevalence of ophthalmic (e.g., glau coma) or nonophthalmic (diabetes, hypertension, athero sclerosis) conditions that might bias our results. We believe that our description of the control group and its method of selection (p. 510) is sufficiently detailed that the term "poorly defined" is inappropriate. We did not exclude from our control group individuals with any ophthalmic diagnosis (other than retinal vein occlusion), nor did we specifically choose subjects with "nonorganic" ophthalmic complaints (e.g., refractive errors). {Indeed, to limit our selection to subjects who came to our institute solely for refractive complaints might under-represent glaucoma or diabetes in the control group.) That our method had at least some validity would seem to be demonstrated by the highly significant differences we observed between subjects with vein occlusions and controls in the preva lence ofglaucoma and of systemic hypertension. We agree that the most nearly ideal choice of a control group for such a study would have been individuals randomly cho sen from the general population, matched with the vein occlusion subjects only by, perhaps, age and postal zip code. However, given the resources we had available, such a study would have been difficult to carry out. Other points raised by Dana and Rosheim strike us as curious. For example, they state, "Without limiting in clusion criteria to 'nonorganic' complaints . . . the con tention that an ophthalmology practice can serve as the source of age-matched controls is suspect at best, and is fraught with confounders. For example, 64% of their con trols were females. . . " This strikes us as something of a non sequitur, unless Dana and Rosheim wish to suggest that most ofthe patients who attend ophthalmology prac tices are female. (They may be right!) We were ourselves surprised by the female preponderance in our control group, and the even larger female preponderance in the control group of Johnston et aV but this was not the case in the study of Elman et al. 4 Although a female predom inance of diabetic retinopathy, cataract, and macular de generation has been reported in some studies, these ocular conditions were not variables we evaluated in our own investigation, and hence they seem irrelevant to the ar gument. However, the National Diabetes Data Group re ported in 1985 that the prevalence of noninsulin-depen dent diabetes was substantially higher-in some age groups as much as twice as high-for black females older than 45 years of age than for black males or for whites of either sex. 5 We are unaware of any reported increased prevalence ofglaucoma or systemic hypertension in males. Dana and Rosheim also state, "Despite an increased prevalence of blacks among cases, black race was not
1755
Volume 99, Number 12, December 1992
bacterium acnes endophthalmitis even lacking of struc tures (residual lens or capsular materials) in which the bacterium could be sequestered within and the favorable prognosis despite the chronic course with periodic severe relapses. Furthermore we would like to emphasize the efficacy of a noninvasive, orally administered antibiotic therapy (chloramphenicol 1 gjday) for the treatment of culture-proven P. acnes endophthalmitis and the good results, after a complete resolution of the uveitis, of an argon lasertrabeculoplasty in controlling lOP in such cases in which a physical barrier (fibrin, inflammatory cells) to the normal aqueous outflow may be proposed as the caus ative agent of the arised lOP. PAOLA PIVETTI-PEZZI, MD MASSIMO ACCORINTI, MD Rome, Italy References
I. Chien AM, Raber IM, Fisher DH, eta!. Propionibacterium acnes endophthalmitis after intracapsular cataract extrac tion. Ophthalmology 1992;99:487-90. 2. Meisler OM, Mandelbaum S. Propionibacterium-associated endophthalmitis after extracapsular cataract extraction: re view of reported cases. Ophthalmology 1989;96:54-61. 3. Zambrano W, Flynn HW Jr, Ptlugfelder SC, eta!. Man agement options for Propionibacterium acnes endophthal mitis. Ophthalmology 1989;96:1100-5.
Author's reply
Dear Editor: We appreciate the comments of Dr. Pivetti-Pezzi regard ing the successful treatment of a case of culture-proven Propionibacterium acnes endophthalmitis with oral chloramphenicol. However, in the United States, given the small but finite risk of irreversible aplastic anemia associated with chloramphenicol and the large number of antibiotic options available, chloramphenicol would not be the first agent of choice. The cephalosporins, semisyn thetic penicillins, vancomycin, clindamycin, and eryth romycin have been shown to be effective against P. acnes. There have been at least two cases in the literature in which oral cephalosporins were used in conjunction with topical cephalosporins to successfully treat recurrences of culture-positive cases of P. acnes endophthalmitis. 1•2 Dr. Pivetti-Pezzi does report an initial trial of oral cephalo sporins in his case, but the length of treatment is not spec ified. ANN M. CHIEN, MD IRVING M. RABER, MD DAVID H. FISCHER, MD RALPH C. EAGLE, JR., MD MICHAEL A. NAIDOFF, MD Philadelphia, Pennsylvania References
I. Brady SE, Cohen EJ, Fischer DH. Diagnosis and treatment of chronic postoperative bacterial endophthalmitis. Ophthalmic Surg 1988;19:580-4. 2. Zambrano W, Flynn HW Jr, Pflugfelder SC, et a!. Man agement options for Propionibacterium acnes endophthal mitis. Ophthalmology 1989;96:1100-5.
1754
Case-Control Study of Retinal Vein Occlusion Dear Editor: We read with interest the recent article by Rath and co workers entitled "Risk Factors for Retinal Vein Occlu sion" (Ophthalmology 1992;99:509-514) in which they investigate possible risk factors for retinal vein occlusion by means of a case-control study. The potential associ ations between retinal vein occlusions and systemic vas culopathies have been widely reported, but case-control studies in this area are few. 1. 2 Dr. Rath and co-workers compare the relative frequencies of a number of systemic and ocular conditions among a series of 87 subjects with retinal vein occlusions ("cases") drawn from their pho tography files with the frequencies among a group of 85 subjects without retinal vein occlusion ("controls") drawn from the patient files of two "general ophthalmologists" in their department. They conclude that systemic hyper tension, chronic open-angle glaucoma, and male sex are significant "risk factors" for retinal vein occlusion, but that race, diabetes, and coronary artery disease are not. We believe there are design flaws in the study that render some of these conclusions open to question. The major weakness of their construct relates to their choice ofcontrols. First, it is altogether unclear what were the diagnoses ofthe subjects in their control group? With out limiting inclusion criteria to "nonorganic" complaints (for example refractive needs), the contention that an ophthalmology practice can serve as the source of age matched controls is suspect at best, and is fraught with confounders. 1 For example, 64% of their controls were females, leading to the conclusion that male sex is a risk factor for vein occlusion. However, a significantly in creased prevalence of diabetic retinopathy, cataract, and possibly macular degeneration, has been reported among females even when controlling for age. 3•4 A bias in favor of one gender in the control group can thereby confound the role of sex as a risk factor among cases. The role of race is similarly obscured. Despite an in creased prevalence of blacks among cases, black race was not shown to be significantly associated with vein occlu sion. This could represent a selection bias in favor of the black race among the controls where an unknown number could have had glaucoma or diabetic retinopathy. Fur thermore, the authors' assertion that the known associa tion between black race and hypertension on the one hand, and hypertension and vein occlusion on the other, pre cludes black race acting as an "independent risk factor" for vein occlusion is false. For example, black race and hypertension are independent risk factors for open-angle glaucoma5 despite a significant unrelated interaction be tween black race and hypertension. In addition to sex and race, diabetic status also can be a serious confounder in a general ophthalmic practice. 1•2 This effect is potentially accentuated by a preponderance of diabetes6 and diabetic retinopathy, 3 particularly among older females who also accounted for the majority of the control subjects used. Most likely, it was exactly because of this bias in favor of diabetes among the control group that a previous studyi failed to show a significant asso
Letters to the Editor ciation between diabetes and retinal vein occlusion when using a series of cataract patients as controls, but reached the opposite conclusion when comparing the cases to sta tistics from a series of subjects from the general popula tion. The utility oflogistic analysis in excluding interactions among multiple confounders, particularly in a small series, is limited-a need that may be obviated by the matching of certain variables. 7 This is why logistic regression ac cording to the authors' own contention failed to show any "significant interactions among [the] variables" tested, despite all epidemiologic evidence to the contrary; as for example between race and hypertension, or race and glaucoma. 5·6 Matching alone clearly does not offer the perfect so lution to such problems, as it would preclude the assess ment of the matched variables' effects on the risk of dis ease. Short of a population-based prevalence study, risk factor analysis can still be conducted effectively when the case-control selection is free of obvious concurrent con founders. However, when many confounding variables are left unmatched, and the randomization process and control definition are poorly defined, particularly in a set ting of small sample sizes, caution is urged in the inter pretation of such data especially related to the exclusion and ranking of potential risk factors. MOHAMAD R. DANA, MD, MPH KAY ROSHEIM, MPH Chicago, Illinois References 1. Elman MJ, Bhatt AK, Quinlan PM, Enger C. The risk for systemic vascular diseases and mortality in patients with central retinal vein occlusion. Ophthalmology 1990;97: 1543-8. 2. Johnston RL, Brucker AJ, Steinmann W, eta!. Risk factors of branch retinal vein occlusion. Arch Ophthalmol 1985; 103: 1831-2. 3. Kahn HA, Moorhead HB. Statistics on Blindness in the Model Reporting Area 1969-70. Washington, DC: National Institutes of Health, 1973. (DHEW pub!; no (NIH) 73-427.) 4. Leibowitz HM, Krueger DE, Maunder LR, et a!. The Fra mingham Eye Study Monograph. Surv Ophthalmol 1980;24(Suppl):335-61 0. 5. Tielsch JM, Sommer A, Katz J, et a!. Racial variations in the prevalence of primary open-angle glaucoma. The Bal timore Eye Survey. JAMA 1991;266:369-74. 6. Whelton PK, Russell RP. Systemic hypertension. In: Harvey AM, Johns RJ, McKusick VA, et a!, eds. The Principles and Practice of Medicine, 21st ed. Norwalk, CT: Appleton Century-Crofts, 1984; chap. 25. 7. Schlesselman JJ. Case-Control Studies: Design, Conduct, Analysis. New York: Oxford University Press, 1982;115 23.
Authors' reply
Dear Editor: Dana and Rosheim raise a number of critical issues re garding our article. Their major criticism deals with our choice of a control group, which they claim was done with the "randomization process and control definition
poorly defined," and was "fraught with confounders." The choice of appropriate control groups for case-control studies is difficult, as we pointed out in our article (p. 512), and as others have also noted at length. 1.2 We chose our control group from patients of two general ophthal mologists in our department, largely for the sake of con venience: the medical records of these individuals, con taining the data we believed were relevant to our study, were readily available to us. Because they were patients of general ophthalmologists, we believed that they would not have an unusual prevalence of ophthalmic (e.g., glau coma) or nonophthalmic (diabetes, hypertension, athero sclerosis) conditions that might bias our results. We believe that our description of the control group and its method of selection (p. 510) is sufficiently detailed that the term "poorly defined" is inappropriate. We did not exclude from our control group individuals with any ophthalmic diagnosis (other than retinal vein occlusion), nor did we specifically choose subjects with "nonorganic" ophthalmic complaints (e.g., refractive errors). {Indeed, to limit our selection to subjects who came to our institute solely for refractive complaints might under-represent glaucoma or diabetes in the control group.) That our method had at least some validity would seem to be demonstrated by the highly significant differences we observed between subjects with vein occlusions and controls in the preva lence ofglaucoma and of systemic hypertension. We agree that the most nearly ideal choice of a control group for such a study would have been individuals randomly cho sen from the general population, matched with the vein occlusion subjects only by, perhaps, age and postal zip code. However, given the resources we had available, such a study would have been difficult to carry out. Other points raised by Dana and Rosheim strike us as curious. For example, they state, "Without limiting in clusion criteria to 'nonorganic' complaints . . . the con tention that an ophthalmology practice can serve as the source of age-matched controls is suspect at best, and is fraught with confounders. For example, 64% of their con trols were females. . . " This strikes us as something of a non sequitur, unless Dana and Rosheim wish to suggest that most ofthe patients who attend ophthalmology prac tices are female. (They may be right!) We were ourselves surprised by the female preponderance in our control group, and the even larger female preponderance in the control group of Johnston et aV but this was not the case in the study of Elman et al. 4 Although a female predom inance of diabetic retinopathy, cataract, and macular de generation has been reported in some studies, these ocular conditions were not variables we evaluated in our own investigation, and hence they seem irrelevant to the ar gument. However, the National Diabetes Data Group re ported in 1985 that the prevalence of noninsulin-depen dent diabetes was substantially higher-in some age groups as much as twice as high-for black females older than 45 years of age than for black males or for whites of either sex. 5 We are unaware of any reported increased prevalence ofglaucoma or systemic hypertension in males. Dana and Rosheim also state, "Despite an increased prevalence of blacks among cases, black race was not
1755
