Identification of independent risk factors for the development of diabetic retinopathy requiring treatment
2009; Wiley; Volume: 89; Issue: 6 Linguagem: Inglês
10.1111/j.1755-3768.2009.01742.x
ISSN1755-3768
AutoresJesper Mehlsen, Mogens Erlandsen, Per Løgstrup Poulsen, Toke Bek,
Tópico(s)Diabetes, Cardiovascular Risks, and Lipoproteins
ResumoIntroduction: Diabetic retinopathy is screened by fundus photography and screening intervals are defined according to general rules to ensure that vision threatening complications are detected even if the progression of the disease is fast. The resulting superfluous examinations of patients with slow disease progression can be reduced by a more exact decision model that allows an adjustment of the screening interval to each patient's individual risk profile. This requires an identification of independent risk factors for reaching treatment end points for diabetic retinopathy. Methods: Clinical data from 5365 patients who had undergone 23 324 examinations at the Department of Ophthalmology, Århus University Hospital between Jan 1st 1994 and Dec 31st 2007 were used to identify independent risk factors for progression of treatment requiring retinopathy. Results: The risk of reaching a treatment end point was in both diabetes types independently affected by retinopathy grade and HbA1c. Furthermore, in type 1 diabetic patients the risk of reaching a treatment end point was independently affected by disease duration and by a recommended control interval of less than 3 months, in spite of correction for retinopathy grade and other studied confounders, whereas in type 2 diabetes this risk was affected by increasing age of diagnosis of the disease. Conclusions: Only a subset of known risk factors for development and progression of diabetic retinopathy should be used to construct a decision model for optimizing screening intervals for diabetic retinopathy. Diabetic retinopathy is a frequent cause of visual impairment in the Western world (Munier et al. 1998; WHO 2008). The risk of developing the sight-threatening consequences of the disease can be reduced if the lesions are detected (Younis et al. 2003a, 2003b) and can be reduced if retinal lesions are detected by ophthalmoscopy or fundus photography before irreversible damage to retinal function has developed (1978; Diabetic Retinopathy Study (DRS) Group 1981; Ferris et al. 1987; Kaufman et al. 1989). On the basis of parameters such as diabetes type, duration of diabetes and grade of retinopathy, the interval to the next screening examination is determined so that vision-threatening complications are detected even if the progression of the disease is fast (Stefansson et al. 2000; American Academy of Ophthalmology 2003). This implies that patients with average or slow disease progression are examined a number of times without this resulting in a shortening of the screening interval or referral for treatment. A reduction in the number of these superfluous examinations requires a more exact decision model that includes sufficient risk factors to allow an adjustment of the screening interval to each patient's individual risk profile. The aim of the present study was to identify relevant variables to be entered into a decision model for individualizing the control interval for patients participating in a screening programme for diabetic retinopathy. Based on clinical and laboratory data from the diabetic retinopathy database at the Department of Ophthalmology, Aarhus University Hospital, we evaluated the significance and possible interdependence of diabetes type, age of onset of diabetes, duration of diabetes, severity of retinopathy, gender, haemoglobin A1c (HbA1c) and blood pressure as risk factors for the progression of retinopathy to a treatment-requiring stage. This study was based on the database for diabetic retinopathy at the Department of Ophthalmology, Aarhus University Hospital, which contained clinical data from all patients examined in the department's screening clinic for diabetic retinopathy since 1992. The examined patients had been referred for screening for diabetic retinopathy from four regional departments of endocrinology, from the regional department for the control of diabetes during pregnancy and the regional department treating renal insufficiency. Additionally, patients had been examined as part of epidemiological research projects into type 2 diabetes, and had been referred from private practitioning ophthalmologists or departments of ophthalmology in other counties for a second opinion or treatment. The neighbouring counties started photocoagulation treatment for diabetic retinopathy in 1996, and consequently the data on treatment for diabetic retinopathy are only complete for patients from Aarhus County from that time. Additionally, the database contained information about all patients in Aarhus County who had become members of the Danish Society for the Blind because of vision loss secondary to diabetic retinopathy since 1992. All examinations performed since 1999 included a measurement of blood pressure using a standardized method. Finally, the database contained all measurements of HbA1c initiated by endocrinologists and general practitioners in Aarhus County since 1994. These measurements had largely been performed asynchronously with the eye examinations. According to Danish law, the reporting of anonymized data from this database did not require ethical approval. We noted the name, date of birth, personal identification number and previous treatment for diabetic retinopathy or other eye diseases in all patients. The patient was asked about the date of diagnosis, age of onset and treatment of diabetes mellitus, as well as family history of the disease; the patient's weight and height were also measured. The patient was classified as having type 1 diabetes mellitus (T1D) if the age of onset was below 30 years or if the age of onset was between 30 and 40 years, insulin treatment was commenced within 1 year of diagnosis of diabetes and the body mass index was below 25. The remaining patients were classified as having type 2 diabetes mellitus (T2D). LogMAR visual acuity was determined in accordance with the principles used in the Early Treatment of Diabetic Retinopathy Study (ETDRS) visual acuity charts (Ferris et al. 1982). Mydriasis was induced by tropicamide 1% and phenylephrine 2.5% eyedrops. After approximately 30 min, fundus photography was performed using a Canon-60UV fundus camera (Canon, Tokyo, Japan). Two 60° images were taken, one centred on the fovea and one on the optic disc. Resting blood pressure (BP) was measured using an electronic sphygmomanometer (Omron M4, HEM-722c1-E Omron, Kyoto, Japan). All fundus photographs were graded by a certified nurse grader who counted the number of each type of retinal lesion within the macula-centred photograph, and determined whether that patient should follow the standardized control intervals recommended in the screening programme, should attend the screening programme more frequently or should be referred for treatment. In cases of doubt about the grading, the photographs were re-evaluated by a retina specialist. Finally, the interval to the next screening examination was recommended on the basis of the guidelines shown in Table 1. The patients referred for treatment underwent slit-lamp biomicroscopy and fluorescein angiography in case there was doubt about whether diabetic maculopathy contained an ischaemic component (Norton & Gutman 1965). In cases of proliferative diabetic retinopathy – defined as the new vessels from the larger retinal venules, the optic disc or rubeosis iridis – panretinal photocoagulation was performed according to the guidelines described in Bek & Kjaergaard (2004) and Bek & Erlandsen (2006). In cases of clinically significant macular oedema, macular laser photocoagulation was performed in the areas with retinal oedema and other retinopathy lesions according to the guidelines described in Bek et al. (2000). All data obtained between 1 January 1994 and 31 December 2007 were sampled for the analysis. This included information from 39 559 examinations performed on 11 970 diabetic patients (2590 with T1D, 9375 with T2D and five with unknown diabetes type), 4061 treatment sessions for diabetic retinopathy or its consequences (retinal photocoagulation or surgery for vitreous haemorrhage, neovascular glaucoma or tractional retinal detachment) performed on 1784 patients, 263 710 HbA1C measurements from 11 654 patients and 23 548 BP values from 7315 patients. At the time of data sampling 8602 patients in the database were alive, of whom 1830 (21.3%) had T1D, 6768 (78.7%) had T2D and four had an undefined diabetes type. The referral rate declined during the first years after 1994, which reflects the gradual inclusion of the population from the referring endocrinological departments. Significantly more men than women with T2D were referred, whereas there was no gender difference in the referral pattern for T1D patients. At the first examination 3217 patients had no retinopathy (648 with T1D, 2569 with T2D), 2084 had retinopathy not requiring treatment (718 with T1D, 1366 with T2D), 69 had retinopathy requiring treatment or had already been treated (26 with T1D, 43 with T2D) and 37 had undefined retinopathy grade (two with T1D, 35 with T2D). The further data processing is illustrated in Table 2, in which exclusion refers to removal of a patient including all examinations performed on this patient, whereas censoring refers to removal of some, but not all, examinations performed on a patient. All personal identification numbers were checked in the Central Office of Civil Registration in order to obtain information about migration in and out of the county, which resulted in the exclusion of 69 patients representing 88 examinations because of non-existing migrational information (row 1). 631 examinations were excluded because they had been performed after the patient had resided outside the county for a period of more than 6 months during which screening data were unavailable (row 2). 14 persons representing 33 examinations were excluded because they had reached a treatment end-point after having resided outside the county for a period of more than 6 months (row 3). 964 examinations performed during pregnancies in 244 women were censored because the control interval during pregnancy was 3 months or shorter, irrespective of the retinopathy grade (row 4). 536 examinations from 536 patients were censored out because they had been treated in connection with the first screening visit and therefore had no follow-up time (row 5). 445 patients representing 2601 examinations were excluded because they had received the first treatment before the beginning of the sampling period (row 6). 6128 examinations were censored out because they occurred after a treatment end-point was reached (row 7). 5254 examinations representing 5254 patients were censored out because they had had one examination only, and therefore had no follow-up time (row 8). Finally 98 patients representing 650 examinations were excluded because the determination of the screening intervals were registered to be to the result of conditions other than diabetic retinopathy (row 9) This cleansing resulted in 22 674 examinations from 5311 patients, comprising 1385 patients with T1D and 3926 with T2D. Among these patients there were a total of 13 508 BP measurements from 4519 patients. HbA1C values were adjusted by linear interpolation to the time of the eye examinations, which resulted in 20 983 values from 5262 patients. Among the patients with T1D who reached a treatment end-point, 69 out of 223 patients (30.9%) had developed diabetic maculopathy and the remaining 154 patients (69.1%) had developed proliferative diabetic retinopathy or sequelae to this complication. Among T2D patients, 395 (73%) and 146 out of 541 (27%) had reached these treatment end-points, respectively. In T1D patients, diastolic BP decreased by 0.16 mmHg/year (linear regression, P = 0.005) from an initial level of 77.4 [95% confidence interval (CI) 76.1–78.6] mmHg, whereas systolic BP increased by 0.47 mmHg/year (linear regression, P < 0.001) from an initial level of 128.4 (95% CI 126.2–130.5) mmHg and HbA1c decreased significantly by 0.05%/year (linear regression, P < 0.001) from an initial value of 9.00 (95% CI 8.9–9.1)%. In T2D patients diastolic BP was initially 84.7 (95% CI 84.2–85.3) mmHg, systolic BP initially 142.9 (95% CI 141.7–144.1) mmHg and HbA1c initially 8.8 (95% CI 8.8–8.9) %. All three parameters declined significantly over time (−0.76 mmHg/year, P < 0.001; −0.64 mmHg/year, P <0.001; and −0.12%/year, P < 0.001, respectively). A treatment end-point was defined as the time of the first laser treatment for diabetic maculopathy or treatment for proliferative diabetic retinopathy by laser or complications of this, such as vitreous haemorrhage, retinal detachment or neovascular glaucoma. However, the true but unobserved treatment end-point would have occurred some time during the time interval between the previous and the actual examination, implying that the occurrence of events was interval-censored. Because the length of the next allocated screening interval depended on the disease status at each examination (so-called time-dependent confounding), the time to treatment could not be analysed from survival analysis by ordinary methods. Therefore, in order to obtain a crude impression of the time-to-treatment survival curves for men and women with T1D and T2D, survival curves were constructed according to Turnbull's method (Turnbull 1976). In order to explain the observed differences in reaching a treatment end-point among men and women with T1D and T2D, two-way anova was used to test baseline differences among these 2 × 2 groups with respect to the following parameters: age at diagnosis, known disease duration, number of retinal haemorrhages, hard exudates and HbA1c at baseline (from 1994), and BP from the time at which these measurements became standardized (from 1999). If an interaction between diabetes type and gender was found, Student's t-test was used to determine which variable was significant. Multilevel mixed-effects linear regression was used to test whether the modifiable risk factors BP and HbA1c changed during the observation period. Finally, multiple logistic regression was used to identify independent variables for reaching a treatment end-point during the interval between two examinations. In short, for each examination, the repeated series of examinations for each patient was used to calculate the probability of an event in the next interval using information from the current examination, including the length of the allocated screening interval. Because the recommended screening interval depended on diabetes type, calculations were made separately on patients with T1D and T2D diabetes. Risk factors were quantified as odds ratios. The logistic regression was modified by stabilized weights calculated by the principle known as inverse probability of treatment weighting (IPTW) to correct for the fact that screening intervals were not assigned at random but shortened as retinopathy progressed (Robins et al. 2000; Diggle et al. 2002), i.e. the screening intervals should be considered as a time-dependent confounder. The calculation was repeated for all data collected since 1999 in order to include BPs in the analysis. Figure 1 shows the proportion of men and women with T1D and T2D who had reached a treatment end-point as a function of time from first examination of diabetes mellitus. Patients with T2D reached the treatment end-point after a shorter known diabetes duration than T1D patients. Men with T1D mellitus reached a treatment end-point after a shorter disease duration than women (P < 0.05). The proportion of men and women with type 1 and type 2 diabetes who had not reached a treatment end-point. Survival curves were constructed using Turnbull's method for interval-censored data. Table 3 shows the comparison at baseline of individual risk factors for reaching a treatment end-point among men and women with T1D and T2D. It appears that all the recorded risk factors differed significantly between patients with the two diabetes types. Additionally, compared with women, men with T2D had a significantly lower age at disease diagnosis; among patients with T1D, men also had a significantly higher number of retinal haemorrhages and hard exudates, and significantly higher baseline and average HbA1c. Table 4 show the odds ratios for reaching a treatment end-point of individual risk factors calculated by the weighted multiple logistic regression. It appears that the risk of reaching a treatment end-point in both diabetes types was increased by increasing number of retinal haemorrhages and increasing HbA1c. Furthermore, in patients with T1D, the risk of reaching a treatment end-point increased by increasing disease duration and by a recommended control interval of less than 3 months, whereas in T2D this risk was increased by increasing age at diagnosis of the disease. The incorporation of known risk factors into an evidence-based model for optimizing the screening interval in diabetic retinopathy could improve diabetes care considerably, both by reducing the average costs per patient in the screening programme and by creating capacity for patients who have not yet been enrolled into a screening programme. The development of such a model requires a set of standardized clinical data that are representative of the target population to be collected with a long enough follow-up period for a significant number of end-points to have been reached, and are based on a reproducible and exhaustive recording of these end-points. The data analysed in the present study meet these requirements to a large extent. The data have been collected in a standardized manner during a period of more than 14 years, the number of patients is large, and on the basis of information from the Central Office of Civil Registration, corrections could be made for deaths and migration in and out of the studied geographical area. Furthermore, whereas most previous studies that have studied risk factors for reaching pseudo end-points such as prevalence and incidence of diabetic retinopathy (Klein et al. 1984a, 1984b; DCCT 1995; UKPDS34 1998; Estacio et al. 2000; Gaede et al. 2003), the present study has assessed risk factors for reaching hard treatment end-points. The incomplete registration of patients who had not reached a treatment end-point, especially those with T2D, might decrease the precision of the predictions about the progression of retinopathy in the total population of patients with diabetes. However, previous studies have shown that the patients not participating in the screening programmes from which the data were collected are predominantly those with minimal or no retinopathy (Hove et al. 2004). Additionally, the number of T2D patients who develop sight-threatening maculopathy without being in a screening programme is small, and many of these patients do not enter a screening programme even if encouraged to do so (Hove et al. 2004). The study showed that high HbA1c is a significant independent risk factor for reaching a treatment end-point in diabetic retinopathy. This confirms the findings of previous studies showing that hyperglycaemia is associated with progression of retinopathy (Marshall et al. 1993; Agardh et al. 1997; Stratton et al. 2000) and that lowering of blood glucose can reduce the progression rate of retinopathy in both T1D and T2D significantly (DCCT 1995; Stratton et al. 2001). Several previous studies have shown that both systolic and diastolic BPs are risk factors for the progression of diabetic retinopathy (Klein et al. 1984a, 1984b; Agardh et al. 1997; Adler et al. 2000; Gallego et al. 2008), and lowering of BP has been shown to reduce the progression of diabetic retinopathy in both T1D (Larsen et al. 1990; Chaturvedi et al. 1998; Knudsen et al. 2003) and T2D (Adler et al. 2000; Stratton et al. 2001). In other studies either systolic or diastolic BP have been found to be related to the progression of diabetic retinopathy in both diabetes types (Klein et al. 1988; Chase et al. 1990; Marshall et al. 1993; Mehler et al. 1997; Adler et al. 2000). In the present study only diastolic BP in T2D patients was found to be an independent risk factor for progression to a treatment end-point, however statistically insignificant. This finding might be because of a better control of BP among the included patients compared with studies where systolic and mean BP were found to be risk factors (Mehler et al. 1997; Matthews et al. 2004; Patel et al. 2007). However, in the ADVANCE study on T2D patients (Patel et al. 2007), for instance, where BP was found to be a risk factor for developing treatment requiring retinopathy, the mean BP (140/80) was lower than in the present study (143/85). However, it is possible that significant risk factors for reaching a treatment end-point may differ from risk factors for the development and progression of any retinopathy, and it should be considered that BP to a large extent covaries with the grade of retinopathy, which may even be a much stronger predictor for reaching a treatment end-point. The finding that patients with T2D seem to reach a treatment end-point after a significantly shorter diabetes duration than patients with T1D may be caused by the fact that the time of onset of the disease in T2D patients is often defined less precisely. Previous studies have shown that diabetes mellitus is diagnosed on average 7–8 years after the onset of the disease (Thompson et al. 1996; Kuo et al. 1999; Harris & Eastman 2000), which may explain the observed differences between the two diabetes types. Thus, it can be hypothesized that although the risk of developing retinopathy as a function of diabetes duration is different among patients with early and late onset of diabetes mellitus (Klein et al. 1984a, 1984b), the risk of developing a vision-threatening diabetes complication may be similar. Several epidemiological studies have shown that the incidence of newly diagnosed diabetes mellitus is higher in men than in women in Denmark and elsewhere (Jorgensen et al. 2003; Carstensen et al. 2008; Economics CIoH 2008). However, the incidence of screen-detected diabetes mellitus was found to be similar in a Danish population of men and women with T2D (Lauritzen et al. 2000), which suggests that more men than women develop symptoms that encourage them to enter the medical care system. This is in accordance with results from the UKPDS, which showed that the incidence of retinopathy in T2D patients was higher among men than women (Stratton et al. 2001); it also concurs with the results of the present study, in which men were found to have higher retinopathy grade, BP and HbA1c than women. Additionally, men had received the diagnosis of diabetes mellitus at a younger age and had reached a treatment end-point after a shorter diabetes duration than women. This finding may be related to differences in lifestyle among men and women. The fact that diabetes duration was a significant risk factor for T1D patients whereas the age of onset of disease was significant for T2D patients, but not the reverse, may be because the age of onset of diabetes varies less for T1D than for T2D patients, and because the progression of retinopathy declines with age. The fact that HbA1c was a risk factor independent of retinopathy grade may indicate that this factor can contribute to retinopathy reaching a treatment-requiring stage in a manner that is not evidenced in retinopathy grade as assessed by the number of microaneurysms and haemorrhages or hard exudates used in this study. Therefore, it is possible that other retinal features could be included in the analysis to reflect the exposition to hyperglycaemia, such as cotton-wool spots or vascular abnormalities; alternatively, the exposure to hyperglycaemia might not be reflected in the visible fundus lesions. The fact that the recommendation of a control interval of 3 months was an individual risk factor independent of retinopathy grade indicates that the grader had used factors in the fundus photographs other than retinopathy grade (as reflected in the number of microaneurysms/haemorrhages) to recommend this screening interval. It is possible that these unknown factors may have included the distribution and the dynamics of retinal lesions (Dobree 1970; Bek & Helgesen 2001; Hove et al. 2006). Therefore, in order to improve prediction of the control interval, studies should be conducted to investigate the influence of the spatial distribution and dynamics of retinopathy lesions. In conclusion, we have described that diabetes type, age at diagnosis of diabetes, diabetes duration, retinopathy grade and HbA1c are independent risk factors for progression of diabetic retinopathy to treatment-requiring end-points based on a large population of patients with diabetes followed in a standardized screening programme for more than 14 years. These data extend previous studies of risk factors for the development and progression of retinopathy not requiring treatment, and can be used to construct a decision model for individualizing the control interval for diabetic retinopathy screening. This study was supported by the VELUX Foundation and the Jochum and Marie Jensen Foundation.
Referência(s)