Toke Bek Professor and Chairman Department of Ophthalmology Århus University Hospital DK-8000 Århus C Denmark Tel: + 45 89 49 32 23 Fax: + 45 86 12 16 53 Email: email@example.com
Introduction: Proliferative diabetic retinopathy is treated with panretinal photocoagulation, which improves the visual prognosis in this complication considerably. The visual acuity (VA) and grade of retinopathy before treatment are known indicators of the visual prognosis after treatment, but the prognostic value of other clinical background and treatment parameters is unknown.
Methods: The study reports predictors for visual outcome identified among retrospective clinical background data and treatment parameters from 4422 panretinal photocoagulation sessions for proliferative diabetic retinopathy in 1013 eyes of 601 patients performed at the Department of Ophthalmology, Århus University Hospital between 1985 and 2002.
Results: High pretreatment VA and low age were strong positive predictors of post-treatment VA (p < 0.0001). However, diabetes type, diabetes duration and calendar year of treatment showed no influence on post-treatment VA (p = 0.7829, 0.1782, and 0.3747, respectively). The visual prognosis was inversely related to the number of treatment sessions (p = 0.0259) and the number of vitrectomies (OR = 2.66 [1.24; 5.69], p = 0.0117, for more than two operations). However, the visual prognosis was unrelated to any of the other parameters studied.
Conclusions: Pretreatment VA, age and the number of panretinal photocoagulation treatment sessions and vitrectomies necessary to halt the disease are indicators of the visual prognosis after panretinal laser photocoagulation for proliferative diabetic retinopathy.
It is generally recommended that PRP should include more than 2000 laser applications, but there is only sparse documentation available to indicate how the treatment should be applied to optimize the positive effects and minimize the adverse effects. Relevant factors might include the number of applications and the length of the time intervals between treatment sessions. Other clinical background parameters such as visual acuity (VA) and the severity of retinopathy have been shown to predict the visual prognosis after PRP (Rand et al. 1985; Kaufman et al. 1989; Kaiser et al. 2000), whereas the influence of parameters such as age, diabetes type and diabetes duration is largely unknown.
The present study reports predictors for visual outcome identified among retrospective clinical background data and treatment parameters from 4422 PRP sessions for PDR in 1013 eyes of 601 patients performed at the Department of Ophthalmology, Århus University Hospital between 1985 and 2002.
Materials and Methods
The study is based on data from the database for diabetic retinopathy at the Department of Ophthalmology, Århus University Hospital. This database contains clinical data for all patients who have been screened, examined or treated for diabetic retinopathy in the department since 1985. The department treats all the diabetes patients in Århus County (approximately 644 000 inhabitants) and until 1996 it also treated patients from two neighbouring counties. At the time of data sampling (June 1st, 2002) the database contained information on 7784 diabetes patients, of whom 1991 had type 1 diabetes (defined as those whose age at onset of diabetes was less than 30 years and who required insulin from onset), 5656 had type 2 diabetes (the remaining patients with known age of onset and treatment), and 137 had unknown diabetes type due to lack of information about age of onset and diabetes treatment.
The registered laser treatment parameters include number, size, and energy of applications as well as the initials of the treating ophthalmologist. According to the department's routine, a standard spot size of 500 µm was used and the energy increased until the laser burns produced a ‘definite whitening’ (ETDRS 1987). In cases with cataract and other media opacities where the energy could not be increased sufficiently to produce a visible burn, a smaller spot size was used. A total of 4613 sessions were registered as having been performed by 73 ophthalmologists. Of these, 48% were performed by one ophthalmologist (the first author), 66% were performed by three ophthalmologists, and 90% were performed by 25 ophthalmologists. Prior to 1994 PRP was carried out by all ophthalmologists in the department, the only guideline being to give at least 2000 laser applications in the retinal periphery when new vessels were detected by ophthalmoscopy or fluorescein angiography. No systematic quality assurance was performed. This resulted in varying treatment strategies and inconsistent registration of treatment parameters. These data were entered into the database post hoc from the patient records. Since 1994 treatment has been performed in a standardized way, preferably in three sessions per eye, by the same ophthalmologist (the first author) or under his supervision, and all patients have been followed up with VA measurement and fundus photography 3 months after treatment. These data have been entered into the database at the time of data collection. At the time of data sampling for this study the database contained information on 4690 PRP sessions in 1484 eyes. Due to incomplete registration before 1994, data were complete for only 4422 treatment sessions in 1013 eyes of 601 patients (253 males, 348 females; 297 type 1 diabetes patients, 297 type 2 diabetes patients, seven patients with unknown diabetes type). The reporting of anonymized data from the database does not require approval from the local scientific ethics committee.
The study was designed to identify predictive parameters for VA after photocoagulation treatment for PDR. However, VA values tended to be clustered in discrete values corresponding to the standard levels on generally used VA charts, which resulted in a non-even distribution unsuitable for standard regression techniques. We therefore used the proportional odds model for ordinal data (McCullagh 1980). In order to do this we introduced five VA categories corresponding to four cut-points: c1 = 0.1, c2 = 0.3, c3 = 0.5 and c4 = 0.7. If x = (x1,...,xp) denotes the set of explanatory variables, the model for the cumulative odds is:
The model has similarities with an ordinary logistic regression model. For example, the cumulative odds ratio is obtained for one of the covariates (x1) as usual: OR1= exp(βl). The odds ratio compares the odds for a result below the cut-point cj with a result above the cut-point, and each of the four cut-points will yield the same odds ratio. The change of sign of the covariates implies that a negative â-coefficient is associated with an odds ratio >1 (i.e. an increased risk of a bad prognosis). A crucial assumption in the model is that the regression coefficients β = (β1,...,βp) do not depend on the category cj, j = 1,2,3,4 (proportional odds assumption). This implies that the interpretation of β = (β1,...,βp) does not depend on the particular choice of categories and adjacent categories can be pooled. We used stata Version 8 to fit the model. As 68% of the patients had been treated in both eyes we had to take account of a possible correlation between eyes in the same patient. This was carried out using the generalized estimation equation principle (GEE) by means of the cluster option in the stata procedure ologit.
The final model was built in a stepwise manner, where the most significant variable among the variables not already in the model was added to the model and the most non-significant variable (if any) was removed. In the initial step only the intercept was included in the model and this procedure was repeated until every variable not included in the model was non-significant and every variable in the model was significant.
The following data were used for every included patient.
The effect parameter was the VA at the last examination after photocoagulation treatment. The values were assigned to each of five groups defined by the four clinically relevant cut-points 0.1, 0.3, 0.5 and 0.7.
• Visual acuity immediately before treatment: five groups with cut-points as for the effect parameter, range (n): <0.1 (130); 0.1–0.29 (116); 0.3–0.49 (170); 0.5–0.69 (158), and ≥0.7 (439).
• Age at first treatment (years): six groups, range (n): ≤30 (148); 31–40 (184); 41–50 (178); 51–60 (194); 61–70 (197), and ≥71 (112).
• Diabetes type: two groups, type (n): type 1 (297), type 2 (297). Eleven eyes from seven patients with unknown diabetes type were not included in the analysis.
• Diabetes duration at first treatment session (years): mean = 18.8; median = 18.6; SD = 11.54, and range 0–62.0.
Treatment intensity is a function of both the number of applications given at each treatment session and the time interval between individual sessions. We therefore calculated the area under the curve (AUC) of the cumulated number of applications as a function of time. This parameter was divided with the total number of applications × the time interval between the first and last treatments to obtain the area fraction (AF). An area fraction close to 0.5 would indicate that treatment had been distributed evenly during the treatment interval (a low treatment intensity), whereas an area fraction close to 0 or 1 would indicate that most of the treatment had been performed during a short period of the treatment interval (a high treatment intensity). Consequently, the following values were defined:
• Calendar year for starting treatment: three groups, range (n): up to and including 1994 (353); 1995–98 (254), and 1999 onwards (406).
• Number of treatments per eye: six groups, treatments (n): one (143); two (266); three (312); four (147); five (73), and six and over (70).
• Total number of laser applications per eye: four groups, applications (n): ≤2000 (266); 2001–2500 (219); 2501–3000 (180), and ≥3001 (348).
• The interval between each treatment session in days (dt).
• The number of laser applications at each treatment session (nt).
• The number of treatment sessions (ns).
On the basis of these data the following were calculated as measures of duration and intensity of the treatment:
• The total time (in days) from the first to last treatment (T): mean = 631; median = 96, and range 0–6682.
• The total number of laser applications given to each eye (N): mean = 2709; median = 2549, and range 86–7810.
• The product of total time (days) and total number of laser applications (TA = T × N): mean = 2 145 110; median = 202 622, and range 0−59 138 831.
• The AUC of the number of laser applications and the time interval (days) between examinations (AUC = Σ n × dt): mean = 537 459; median = 80 964, and range 0–17 504 957.
• The treatment intensity as expressed by the area fraction (AF = AUC/TA): mean = 0.45; median = 0.36, and range 0–1.
• Time of cataract operation: three groups (n): 0 = none or before baseline VA measurement (944); 1 = between last treatment and last VA measurement (33), and 2 = during treatment (36).
• Number of vitrectomies: three groups (n): none (781); one (198), and two = (64).
• Time of vitrectomy: three groups (n): 0 = none or after the last VA measurement (781); 1 = within 30 days after baseline VA (78), and 2 = more than 30 days after baseline VA (184).
• Follow-up time (days) between last VA measurement and last treatment (photocoagulation): median = 136, range 32–6380.
The change in VA after treatment with PRP is shown in Fig. 1 and Table 1. Both pretreatment VA and age were strong predictors of post-treatment VA (p < 0.0001 for both comparisons). Figure 2 shows the results of the combined model. For increasing pretreatment VA there is a decreasing chance of retaining or improving post- treatment VA with increasing age. However, diabetes type, diabetes duration and calendar year of treatment show no influence on post-treatment VA (p = 0.7829, 0.1782 and 0.3747, respectively).
Table 1. The number of treated eyes according to visual acuity before treatment and at follow-up.
In the statistical analysis a correction was made for the interactions found between the treatment variables. Consequently, the visual prognosis was inversely related to the number of treatment sessions (p = 0.0259) and the number of vitrectomies (OR = 2.66 [1.24; 5.69], p = 0.0117, for more than two operations). However, the visual prognosis was unrelated to the total number of treatment sessions (p = 0.7505), laser applications (p = 0.2818), the intensity of the laser treatment (p = 0.1824), or any of the other parameters studied.
The present study is the first to report the influence of treatment parameters and clinical background data other than VA and retinopathy grade on the visual prognosis after panretinal laser photocoagulation for proliferative diabetic retinopathy (Rand et al. 1985; Kaufman et al. 1989; Kaiser et al. 2000). The weaknesses of the study lie in its retrospective design and the fact that many ophthalmologists were involved in the treatment. However, all ophthalmologists followed the generally recommended guidelines for the treatment of PDR. Furthermore, there was no difference between the results of patients treated before 1994 or after this time, when treatment was almost exclusively performed by one person. This confirms recent data showing that the pattern of visual loss secondary to diabetic retinopathy in Århus County did not change when the treatment pattern was standardized around 1994 (Jeppesen & Bek 2004). Finally, although the clinical data and treatment parameters were incomplete for the first years of the studied period, and the excluded patients (n = 471) were on average slightly (2.2 years) older (p = 0.064) than those included, there was no difference between the two groups with regard to diabetes duration (p = 0.1421). Therefore, the shortcomings of the retrospective design can be assumed to be of little importance to the interpretation of the results, and are greatly outbalanced by the large patient material, which included all patients treated in the county over a long time period.
Visual acuity only measures limited aspects of visual function, and a number of visual modalities known to be affected by retinal laser photocoagulation, such as contrast vision (Lovestam-Adrian et al. 2000), visual fields (Pahor 1998) and adaptation (Mantyjarvi 1989), were not assessed by this parameter. Therefore, the visual prognosis reported in the present study may not fully cover the complaints experienced by the patients. However, VA is generally a good indicator of reading capability and other central visual functions and is measured routinely in clinical practice, which makes it suitable as an effect parameter by which to evaluate therapeutic intervention in diabetic retinopathy.
The study confirmed existing evidence that pretreatment VA is an indicator of the visual prognosis after PRP for PDR (Rand et al. 1985; Kaiser et al. 2000). However, the treatment parameters were found not to have affected VA at the time of follow-up, implying that the visual blurring and other acute adverse effects known to be prevalent immediately after treatment had disappeared (Doft & Blankenship 1982). These results are encouraging but also point to the necessity of treating PDR before retinal damage has occurred. As has been shown previously, the grade of PDR is also an important clinical predictor of visual outcome after treatment (Rand et al. 1985; Kaufman et al. 1989; Kaiser et al. 2000). In the present study this parameter could not be systematically and consistently evaluated as data were not available for the period before 1992 when the department's screening clinic for diabetic retinopathy was launched.
The data confirm the clinical practice of informing patients that treatment will not improve vision, but that the VA may at best be preserved at the level it had reached before treatment. In addition to this it was found that the risk of experiencing a loss in VA secondary to panretinal laser photocoagulation increases significantly with age, and this increased risk is most pronounced when pretreatment VA was high. This emphasizes the need for special information about the risks of PRP in older age groups.
The results demonstrate that neither the total number of laser applications, the time period over which the treatment was given, nor the treatment intensity were related to the visual prognosis after panretinal laser photocoagulation for PRP. This extends the findings of other authors who found no difference in visual prognosis after single versus multiple treatment sessions (Doft & Blankenship 1982). The findings also argue against assessing the progression and effect of the treatment on the basis of the number of given laser applications and indicate that the visual blurring that can be experienced after extensive treatment mostly disappears after some period of recovery. It has been clinically routine to augment the treatment until a sufficient clinical response has been reached, which is probably the cause of the poor visual prognosis in patients who were subjected to many photocoagulation sessions and many vitrectomies. These patients also had low pretreatment VA and the increased number of treatments therefore indicates that their disease was of a severe grade which needed extensive treatment (Kaiser et al. 2000). Therefore, it may be concluded that the major determining factor for amount and intensity of treatment should be the observed clinical response (Rogell 1983).
In conclusion, the study has shown that, in addition to retinopathy stage, pretreatment VA and age are the most significant predictors of visual prognosis after panretinal photocoagulation for proliferative diabetic retinopathy. The effect of the treatment should be adjusted after the clinical response and is independent of the number of laser applications, the time interval and the intensity of the treatment.