Factors Associated with Visual Outcome after Photocoagulation for Diabetic Retinopathy

Size: px

Start display at page:

Download "Factors Associated with Visual Outcome after Photocoagulation for Diabetic Retinopathy"

Angel Turner
5 years ago
Views:

1 Investigative Ophthalmology & Visual Science, Vol. 30, No. 1, January 1989 Copyright Association for Research in Vision and Ophthalmology Factors Associated with Visual Outcome after Photocoagulation for Diabetic Retinopathy Diabetic Retinopathy Study Report #13 Steven C. Kaufman,* Frederick L. Ferris, III,* Daniel G. Seigel,* Matthew D. Davis, t David L. DeMers4 and the DR5 Research Group Six risk factors for severe visual loss despite panretinal (scatter) photocoagulation were identified by analyzing data collected during the first 5 years after randomization in the Diabetic Retinopathy Study. Proportional hazards regression revealed NVD (neovascularization on/around the optic disc) to be the most important risk factor. The risk of severe visual loss rose with increasing NVD, hemorrhages/microaneurysms, retinal elevation, proteinuria, and hyperglycemia and fell with increasing "treatment density." These results are similar to previous DRS findings on untreated eyes. The importance of "treatment density" as an independent predictor of visual outcome is a new finding and lends support to the common clinical practice of repeating photocoagulation if initial treatment does not reduce or stabilize retinal neovascularization. Invest Ophthalmol Vis Sci 30:23-28,1989 The Diabetic Retinopathy Study (DRS) demonstrated that panretinal (scatter) photocoagulation reduces the risk of "severe visual loss" (SVL, denned below) in eyes with proliferative or severe nonproliferative diabetic retinopathy by 50 to 60%. 1>2 In addition to testing the effect of photocoagulation, data collected by this study are useful for examining associations between baseline characteristics and visual outcome. For example, the third DRS report described the association of four ocular factors with the development of severe visual loss in the untreated eyes: preretinal/vitreous hemorrhage, retinal neovascularization, neovascularization on or within one disc diameter of the optic disc (NVD), and severity of new vessels. 3 The 2-year incidence of severe visual loss increased as the number of these risk factors increased, ranging from 4% in the untreated eyes lacking any of these four factors to 37% in those with all four risk factors. From the 'Biometry and Epidemiology Program of the National Eye Institute, Bethesda, Maryland, and the fdepartment of Ophthalmology and the JBiostatistics Center of the University of Wisconsin-Madison, Madison, Wisconsin. A complete list of the D.R.S. Research Group appeared in Investigative Ophthalmology & Visual Science 21:3, Reported in part at the 1986 Annual Meeting of the Association for Research in Vision and Ophthalmology. Submitted for publication: June 21, 1988; accepted July 20, Reprint requests: F. L. Ferris, Biometry and Epidemiology Program, Bldg. 31, Room 6A10, National Eye Institute, Bethesda, MD Another analysis used stepwise linear regression with the DRS data set to assess 200 baseline characteristics for association with severe visual loss in untreated eyes within the first 2 years of follow-up. 4 NVD was the single most important predictor. Other factors shown to be important in that report include initial visual acuity, retinal hemorrhages/microaneurysms, arteriolar abnormalities, venous caliber abnormalities, preretinal/vitreous hemorrhage, perivenous exudates, proteinuria and amputation. This paper reports the results of an analogous analysis of the DRS treated eyes and identifies six factors associated with severe visual loss following photocoagulation. Since photocoagulation is now the usual treatment for eyes with proliferative or severe nonproliferative diabetic retinopathy, these baseline and treatment characteristics may be considered to be the factors influencing the current prognosis of patients with advanced diabetic retinopathy. Materials and Methods The data were collected during the 7-year course of the Diabetic Retinopathy Study ( ), a multicenter randomized controlled clinical trial. 5 ' 6 Eligibility requirements included: (1) best-corrected visual acuity of 20/100 or better in each eye; (2) proliferative diabetic retinopathy in at least one eye or severe nonproliferative diabetic retinopathy in both eyes; (3) age less than 70 years; (4) no previous photocoagulation or pituitary ablation treatment; and (5) no progressive ocular disease unrelated to diabetes. 20

24 INVESTIGATIVE OPHTHALMOLOGY 6 VISUAL SCIENCE / January 1989 Vol. 30 Table 1. Explanatory (independent) variables evaluated in Phase I of the analysis I. Ocular characteristics A.

2 24 INVESTIGATIVE OPHTHALMOLOGY 6 VISUAL SCIENCE / January 1989 Vol. 30 Table 1. Explanatory (independent) variables evaluated in Phase I of the analysis I. Ocular characteristics A. Determined by clinical eye exam Initial visual acuity Spherical equivalent B. Determined by fundus photography 1. Field 1 Lesions (photo centered on disc) NVD (neovascularization on/around the disc) Plane of proliferation Fibrous proliferation of disc 2. Highest score found in Fields 3-7 (peripheral retina) Venous caliber abnormalities Perivenous exudates Retinal elevation NVE (neovascularization elsewhere) 3. Highest score found in Fields 2-7 (macula & peripheral retina) Soft exudates IRMA (intraretinal microvascular abnormalities) H/Ma (hemorrhages and/or microaneurysms) C. Derived by combining independent measurements made by clinical eye exam and fundus photography Preretinal/vitreous hemorrhage Macular edema II. Nonocular characteristics Age Sex Duration of diabetes Diabetes class (classic juvenile-onset, other) Serum creatinine Proteinuria Systolic blood pressure After eligibility was confirmed and informed consent obtained, the eye to be treated was selected at random as was the type of treatment (argon laser or xenon arc photocoagulation). Over a period of 39 months, 1742 patients were recruited at 15 centers. This resulted in 887 right and 855 left eyes being treated, 867 with the argon laser and 875 with the xenon arc photocoagulator. 2 The presence and severity of each of the several lesions comprising diabetic retinopathy were determined at baseline with a clinical eye examination and stereoscopic fundus photographs covering seven standard fields. 6 The fundus photographs were assessed independently by two different graders at one central location, using a modification of the Airlie House classification of diabetic retinopathy. 7 ' 8 In this analysis, the mean of the two gradings was used for those lesions graded in only one field. For each lesion graded in multiple fields, the mean of the first and second graders' highest scores in any of the fields graded for that lesion was used. Macular edema scores were derived by combining the information obtained from the clinical exam and the photos, as has been described elsewhere. 9 A similar approach was taken with preretinal/vitreous hemorrhage and vitreous detachment. Two variables were used to estimate the extent to which each eye was treated. The first, "area treated" was computed by multiplying the number of burns applied by the area of each burn (using the "spot size" as the diameter of a circle). 10 The second, "treatment density," refers to the percentage of the superonasal photographic field of the fundus (DRS standard field six 6 ) covered by burns in the post-treatment photographs taken within 3 days of the initial treatment. Field six was graded for treatment density in the DRS because it was one of the two in which scatter treatment was expected to cover the entire photographic field. Further methodologic details concerning the DRS have been published previously. 5 ' 6 Explanatory Variables Forty baseline characteristics were chosen for analysis. Each was selected either because it was found in DRS Report #10 to be an important predictor of visual outcome in untreated eyes or because it has been shown to have (or is suspected by DRS ophthalmologists to have) a clinically significant influence on the effect of photocoagulation. Our initial analysis, referred to below as Phase I of the model-building process, was limited to the 21 of these 40 variables which previous clinical experience and/or research findings suggested were most likely to be important. (The number 21 was used because 209 of the 1742 eyes suffered severe visual loss; stepwise regression techniques may give unreliable results if one begins with more than one-tenth as many explanatory variables as the number of uncensored observations." These 21 variables are listed in Table 1. In Phase II, we assessed the abilities of four treatment characteristics to contribute additional useful information to the "best" model found in Phase I. They are: treatment type (argon or xenon), number of initial treatment episodes, area treated, and treatment density. The protocol allowed the treating ophthalmologists to tailor initial treatment, within limits, to each patient. For example, eyes randomized to treatment with the argon laser could be given anywhere from 800 to fim burns or 500 to uta burns. This flexibility raised the possibility that eyes with more advanced retinopathy might be treated more extensively. In fact, evidence suggests that this occurred. For this reason, we examined the four treatment descriptors after the 21 primary baseline characteristics, thereby adjusting for the severity of retinopathy when testing the effects of the treatment variables. Finally, in Phase III, the remaining 19 baseline characteristics were allowed to be added to the model,

3 No. 1 DRS REPORT #13/ Kaufman er al 25 in case any of them could contribute additional useful information. They are listed in Table 2. In exploratory analyses of a large number of variables, there is a risk of finding associations of spuriously high statistical significance. This needs to be kept in mind when interpreting the results, particularly with regard to the variables examined in Phase III. Outcome Variable The outcome variable studied in this report was the time, in days, from randomization to the occurrence of severe visual loss (SVL). SVL is defined as the occurrence of visual acuity worse than 5/200 at any two consecutive visits. Visual acuity was measured at each study visit. Except for the 6-week post-treatment visit, follow-up visits were scheduled every 4 months after randomization. This analysis is limited to the data collected during the first 5 years following randomization. Missing values for visual acuity were replaced by averages of the values recorded at the two visits "bracketing" them, when available. (For example, if visual acuity data were missing at follow-up visits six and seven, they were replaced by the average of the values found at visits five and eight.) Statistical Techniques Proportional hazards regression 1213 was used to determine which baseline and treatment covariates were most strongly associated with time to SVL. Backward elimination as well as stepwise selection were used to build the models, each using a significance level (alpha) of 0.01 at each step in the modelbuilding process. The validity of the proportional hazards model assumptions was assessed graphically by comparing separate predicted and "observed" (Kaplan-Meier) survival curves for each tertile of each variable in the model. 14 (Tertiles are analogous to quartiles; they differ in that quartiles split a distribution into four equal parts while tertiles refer to the upper, middle, and lower thirds of the distribution). A standardized regression coefficient was computed for each variable. 15 This was exponentiated to yield a standardized hazard ratio, which is the fractional change in the hazard function for each standard deviation change in the corresponding explanatory variable. The hazard function indicates the probability of suffering severe visual loss at any point in time given that it had not yet occurred. Results The six variables found to be significantly associated with the risk of severe visual loss are listed in Table 3, in order of selection by the stepwise analysis. Table 2. Explanatory (independent) variables evaluated in Phase III of the analysis I. Ocular characteristics A. Determined by fundus photography 1. Field 1 lesions Dilated tips 2. Field 2 lesions Cystoid changes Retinal distortion from tension lines Fibrous proliferation of macula 3. Highest score found in all fields graded FPE (fibrous proliferation elsewhere) Plane of proliferation Arteriolar abnormalities 4. Highest score found in all fields graded Hard exudates B. Derived by combining independent measurements made by clinical eye exam and fundus photography 1. Vitreous detachment II. Nonocular characteristics Diastolic blood pressure Race Insulin usage at entry Height Weight Relative body weight Serum cholesterol Plasma glucose (The same set of variables was obtained by backward elimination.) The first four of these were chosen in Phase I, ie, from the initial set of 21 candidate variables. The next variable was selected during Phase II, ie it was the most significant of the four treatment descriptors. The last one was chosen during Phase III. Standardized hazard ratios are reported in Table 3, rather than the more conventional unadjusted hazard ratios, to facilitate comparing different explanatory variables' effects. For example, to the extent that this model is "true," the standardized hazard ratio for NVD of 1.74 means that for each standard deviation difference between two eyes' NVD scores, the eye Table 3. Final proportional hazards model (censoring at time of death or withdrawal, variables listed in order of selection) Standardized Variables hazard ratio Variables chosen in Phase I* Neovascularization on/around the disc (NVD) Proteinuria Hemorrhages/microaneurysms (H/Ma) Retinal elevation Variable chosen in Phase II Density of treatment Variable chosen in Phase III Plasma glucose ("random") P-value < < < <0.000l * The three phases of the model-building process are described above under "Explanatory Variables."

4 26 INVESTIGATIVE OPHTHALMOLOGY 6 VISUAL SCIENCE / January 1989 Vol. 30 Neovascularization of the Disc Hemorrhages/Microaneurysms o..o Years to Severe Visual lo»/l>ithdra> a! Proteinuria Treatment Density il Lois/Withdrawal Retinal Elevation Plasma Glucose Viiual loss/withdrawal o Severe Visual Loss/Wit Fig. 1. Thisfigureshows survival curves predicted by the six-variable model given in Table 3 separately for the first, second and third tertiles of each of the variables in the model. These six plots show how the predicted probability of severe visual loss increases over time and with increasing degrees of thefivebaselineriskfactors. In contrast, predicted risk decreases with increasing degrees of treatment density. The curves are labelled as follows: NVD: moderate/ severe = worse than or equal to standard photo 10A, mild = present but better than photo 10A; Hemorrhages/Microaneurysms: severe = worse than photo 2A, moderate = worse than photo 1 but not worse than photo 2A, mild = better than photo 1; treatment density = percent of superonasal field covered by burns; plasma glucose = mg/dl, "random." o.,o a toss/wllhdrawal o Sarera Visual Lou/Withdrawal

No. 1 DRS REPORT #10 / Kaufman er al 27 with worse NVD would have a 74% greater hazard, ie, it would be 1.74 times as likely to suffer severe visual loss within any given time period.

5 No. 1 DRS REPORT #10 / Kaufman er al 27 with worse NVD would have a 74% greater hazard, ie, it would be 1.74 times as likely to suffer severe visual loss within any given time period. In contrast, a one standard deviation increase in proteinuria would increase the hazard rate by only half as much, 37%. Eyes at opposite ends of the scale with respect to NVD, ie, one with no NVD compared to one with NVD worse than standard photo IOC, would differ in hazard by nearly a factor of 5 since this range encompasses 2.8 standard deviations. Figure 1 shows survival curves predicted by this six-variable model, separately for the first, second, and third tertiles of each of the variables in the model. These six plots display graphically how the risk of severe visual loss increases with increasing degrees of the five baseline risk factors and decreases with increasing treatment density. They reveal, for example, that the model predicts nearly 19% of the eyes in the highest tertile for NVD would suffer SVL within 5 years of treatment in contrast to only 7% of those in the lowest tertile. Almost 7% of the eyes in the highest tertile for treatment density (51 to 92% coverage of Field 6) are predicted to have SVL in comparison to 15% of those in the lowest tertile (below 35% coverage). Discussion The analyses described above were done to determine which of 44 baseline and treatment characteristics were most strongly associated with the development of severe visual loss during the first 5 years following photocoagulation in the Diabetic Retinopathy Study. These results underscore the importance of the degree of retinopathy with respect to prognosis. Three of the six variables comprising the final model are retinal lesions; the greater the severity of each, the worse the risk of subsequent severe visual loss. Disc neovascularization (NVD) seems most important since it was chosen first. The statistical significance of the remaining factors was not surprising. Retinal elevation and H/Ma, like NVD, have been reported previously to be significantly associated with severe visual loss in the untreated eyes of the DRS. 4 Proteinuria's selection is probably attributable to an association between retinal and renal microangiopathy. The significant association of plasma glucose, despite the variation inherent in the use of nonfasting samples, probably reflects the influence of hyperglycemia in the development of diabetic sequelae. These results are similar to those found by other analyses of DRS data. For example, linear and logistic regression have been used to determine which baseline and treatment characteristics were most strongly associated with "occurrence of SVL" and best predictive of "last-known visual acuity" in the treated eyes of the DRS. 16 Each of the four models developed in that analysis include four of the variables in Table 3 (NVD, H/Ma, treatment density, and plasma glucose). In addition, these results are consistent with the findings of DRS Report #3, 3 which described the association of neovascularization, particularly NVD, with severe visual loss within the first 2 years of follow-up. Finally, the results described above are consistent with those of the analysis of untreated eyes described in DRS Report #10, 4 despite the fact that the present analysis concerned treated eyes, employed survival analysis instead of linear regression, and used a longer follow-up period. The risk factors for severe visual loss in both treated and untreated eyes appear to be very similar. In recent years, it has become increasingly common for clinicians to perform additional photocoagulation if the initial treatment does not reduce or stabilize retinal neovascularization. 17 While the DRS was not designed to evaluate this clinical practice, the high degree of statistical significance for treatment density does lend it some support. It should be noted that its effect is independent of the other variables in the model, ie, higher treatment density was associated with a lower risk of severe visual loss regardless of the severity of NVD, proteinuria, etc. To put into perspective the magnitude of its effect, this model predicts that eyes with treatment density one standard deviation above the mean, ie, those in which 61% of Field 6 is covered by burns, would be at half the hazard of severe visual loss expected for eyes with treatment density one standard deviation below the mean, ie, those with only 27% of Field 6 covered by burns. Each of the two estimates of the extent of treatment (treatment density and area treated) was significant when evaluated individually. When competing against one another and the other two treatment descriptors in Phase II of this analysis, treatment density was the most important. This indicates that while area treated (derived from the number and sizes of burns applied) does contribute useful information with respect to predicting visual outcome, this information is not independent of that provided by treatment density (percent of Field 6 covered by burns). Both support the clinical impression that the effect of photocoagulation is, to some extent, dose-related. This analysis identified six risk factors associated with severe visual loss in eyes with proliferative or severe nonproliferative diabetic retinopathy treated by photocoagulation. Among these six, the most important predictor of time until severe visual loss was the extent of NVD. It also found evidence linking

6 28 INVESTIGATIVE OPHTHALMOLOGY & VISUAL SCIENCE / January 1989 Vol. 30 more extensive photocoagulation with decreased risk of severe visual loss, thus providing support for the clinical practice of performing additional photocoagulation when neovascularization is not reduced or stabilized by initial treatment. Key words: diabetic retinopathy, panretinal photocoagulation, xenon arc photocoagulation, argon laser photocoagulation, severe visual loss, proportional hazards regression model References 1. Diabetic Retinopathy Study Research Group: Preliminary report on effects of photocoagulation therapy. Am J Ophthalmol 81:383, Diabetic Retinopathy Study Research Group: Photocoagulation treatment of proliferative diabetic retinopathy: Clinical applications of Diabetic Retinopathy Study findings (DRS Report #8). Ophthalmology 88:583, Diabetic Retinopathy Study Research Group: Four risk factors for severe visual loss in diabetic retinopathy (DRS Report #3). Arch Ophthalmol 97:654, Rand LI, Ederer F, Prud'homme G, Canner P, and the Diabetic Retinopathy Study Research Group: Factors influencing the development of visual loss in advanced diabetic retinopathy (DRS Report No. 10). Invest Ophthalmol Vis Sci 26:983, Diabetic Retinopathy Study Research Group: Manual of Operations. Baltimore, Diabetic Retinopathy Study Coordinating Center, Diabetic Retinopathy Study Research Group: Design, methods, and baseline results (DRS Report #6). Invest Ophthalmol Vis Sci 21:149, Davis MD, Norton EWD, and Myers FL: The Airlie classification of diabetic retinopathy. In Symposium on Treatment of Diabetic Retinopathy, Goldberg MF and Fine ST, editors. Washington, D.C., U.S. Govt. Printing Office, Publication #1890, 1969, pp Diabetic Retinopathy Study Research Group: A modification of the Airlie House classification of diabetic retinopathy (DRS Report #7). Invest Ophthalmol Vis Sci 21:210, Ferris FL III, Podgor MJ, Davis MD, and the DRS Research Group: Macular edema in Diabetic Retinopathy Study patients (DRS Report #12). Ophthalmology 94:754, Okun E, Johnston GP, Boniuk I, Arribas NP, Escoffery RF, and Grand G: Xenon arc photocoagulation of proliferative diabetic retinopathy. Trans Am Ophthalmol Soc 81:229, Harrell FE Jr, Lee KL, Matchar DB, and Reichert TA: Regression models for prognostic prediction: Advantages, problems, and suggested solutions. Cancer Treatment Reports 69:1071, Cox DR: Regression modes and life tables (with discussion). J Royal Stat Soc B 34:187, Harrell FE Jr: The PHGLM Procedure. In SUGI Supplemental Library User's Guide. Cary, North Carolina, SAS Institute, 1986, pp Harrell FE Jr: Verifying assumptions of the Cox proportional hazards model. In Proceedings of the Eleventh Annual SAS Users Group International Conference. Cary, North Carolina, SAS Institute, 1986, pp Schlesselman JJ: Case-control studies. New York, Oxford University Press, 1982, p Kaufman SC: Factors influencing visual outcome after photocoagulation in the Diabetic Retinopathy Study. Unpublished master's thesis, Memorial Library, University of Wisconsin at Madison, Rogell GD: Incremental panretinal photocoagulation: Results in treating proliferative diabetic retinopathy. Retina 3:308, 1983.

Factors Influencing the Development of Visual Loss in Advanced Diabetic Retinopathy

Factors Influencing the Development of Visual Loss in Advanced Diabetic Retinopathy Diabetic Retinopathy Study (DRS) Report No. 10 Lawrence I. Rand, Gerard J. Prud'homme, Fred Ederer, Paul L. Conner, and