Is Drusen Area Really So Important?

This letter is in reference to the article by Friberg et al. entitled “Is Drusen Area Really So Important?” ¹ The paper concludes that a larger drusen area does not necessarily increase an eye's risk of conversion to choroidal neovascularization (CNV) (Table 1), which disagrees with the established evidence that a larger drusen area is associated with increased risk of CNV based on large cohort studies (Table 2). ^2–4 I am concerned that many of the multivariate analyses were inappropriate because of the small number of CNV events, leading to misinterpretation of the results. 

Table 1. 

 

Summary Results on the Association of Drusen Area with CNV*

	1000 μm		3000 μm
	AREDS	PTAMD	AREDS	PTAMD
	OR (95% CI)	OR (95% CI)	OR (95% CI)	OR (95% CI)
Drusen area (1 SD increase)	1.6 (1.3–2.2)	0.8 (0.3–2.0)	1.3 (0.9–1.8)	1.1 (0.5–2.3)
≥1 Large drusen (y/n)	2.6 (0.96–7.1)	8.2 (0.3–220)	3.5 (1.1–11)	Inf (0–Inf)
≥1 Large drusen equivalent area (y/n)	5.4 (1.5–19)	Inf (0–Inf)	3.3 (0.4–26)	NA
≥5 Large drusen equivalent area (y/n)	3.3 (1.3–8.1)	0.7 (0.1–6.6)	4.9 (1.2–19)	NA

 CI, confidence interval; SD, standard deviation; y/n, yes/no; Inf, infinity.

*  Adapted from Friberg et al. Tables 1–4. ¹

Table 2. 

 

Association of Drusen Area with CNV from the AREDS Study*

Drusen Area (DA)†	No. of Large Drusen (125-μm diameter) Equivalent to Drusen Area‡	Eyes	CNV (%)
0	0	879	1 (0.1)
≥0.0017, <0.0069	(0.25, 1)	643	6 (0.9)
≥0.0069, <0.028	(1, 4)	463	7 (1.5)
≥0.028, <0.056	(4, 8)	315	11 (3.5)
≥0.05, <0.19	(8, 28)	259	12 (4.6)
≥0.19, <0.5	(28, 72)	186	16 (8.6)
≥0.5, <1.0	(72, 145)	164	20 (12.2)
≥1.0	≥145	303	39 (12.9)

 DA, disc area.

*  Adapted from Table 2 of the AREDS Report No. 17. ²

†  Based on total drusen area of all sizes and types of drusen within 3000 μm; graded using a standard grid template.

‡  The area of one large drusen (125-μ diameter) is 0.012272 mm² = 0.0069 DA.

First, the study analyzed the data from two very different clinical trial cohorts. The first cohort comprised a subset (approximately 14%) of the Age-Related Eye Disease Study (AREDS) population consisting of 444 participants (820 eyes) with drusen ranging from none in both eyes to large or extensive intermediate drusen in one or both eyes. ³ The second cohort comprised 62 participants (62 untreated eyes) in the Prophylactic Treatment of Age-Related Macular Degeneration (PTAMD) with at least five intermediate or large drusen in the central macula. ⁵ In these two cohorts, only a small number of CNV events occurred (26 CNV events in the AREDS, 6 events in the PTAMD). The evaluation of the association between drusen area (in the central 1000- or 3000-μm diameter regions of the macular) and development of CNV was made for each cohort separately and as combined, by using a multivariate logistic regression model with adjustment for age, hyperpigmentation, and the fellow-eye status of CNV, yet without adjustment by the other strong risk factors of CNV, such as current smoking status. The small number of CNV events in each cohort (particularly six CNV in the PTAMD cohort), falls far short of the minimum of 10 events per risk factor required to provide reliable results in multivariate logistic regression. ⁶ The estimated odds ratios (ORs) from the models can be biased towards either the positive or negative direction, ⁶ thus results of this study can be totally misleading or not interpretable. 

Second, the study subjects had a wide range of length of follow-up, ranging from 12 days to 8 years (mean, 3.5 years). The logistic regression models ignore the differences in time of the event and the differences in the length of follow-up. Ignoring the different lengths of follow-up can lead to very biased or inefficient results. ⁷ Survival analysis approaches account for both differing times to development of CNV and differing durations of follow-up and would have been more appropriate; however, the same problem noted above of too few events to provide reliable results from multivariate models would persist with survival analysis. 

Finally, the article did not provide any description of the baseline characteristics of study participants, such as age, smoking status, how many eyes had hyperpigmentation, how many patients had CNV in the fellow eye, what percentage of eyes had one or five large drusen, and so forth. Lack of such important baseline information in each cohort makes it very difficult to interpret the study results. 

In summary, Friberg and coauthors used a small subset of the AREDS population and reached a conclusion different from the conclusion reached when the full AREDS population was analyzed. The data analytic methods that were used provide unreliable results when several covariates are considered simultaneously. While the more precise measurement of drusen area used by Friberg may be useful in further specifying the risk of CNV, larger datasets and more careful statistical analysis are needed.