Systemic Dyslipidemia in Age-related Macular Degeneration

Abstract

Topic

Though lipid and cholesterol dyshomeostasis is thought to contribute to the pathogenesis of age-related macular degeneration (AMD), there is no consensus regarding which elements of systemic lipid homeostasis are perturbed in AMD. In this systematic review and meta-analysis, an update to that performed by Wang et al in 2016, we characterized serum lipoprotein profiles in patients with AMD and its various stages.

Clinical Relevance

These findings may identify novel therapeutic approaches for AMD, a leading cause of blindness among older adults in the industrialized world.

Methods

We used MEDLINE, Embase, and Web of Science to identify articles from database inception to May 2022 that reported blood/serum levels of lipid subspecies (triglycerides [TGs], total cholesterol [TC], low-density lipoprotein [LDL], and high-density lipoprotein [HDL]) in patients with AMD compared with controls. We meta-analyzed the data by generating multilevel random-effects models using restricted maximum likelihood estimation.

Results

Our updated meta-analysis included 56 studies, almost 3 times as many studies as the 2016 meta-analysis with a total of 308 188 participants. There were no significant differences in serum TG, TC, LDL, or HDL between patients with AMD and non-AMD controls. Given significant heterogeneity, we performed subanalyses specifically in patients with early to intermediate nonexudative AMD, advanced nonexudative AMD, and advanced exudative AMD. Compared with non-AMD controls, patients with early to intermediate nonexudative AMD had significantly lower serum TG (standardized mean difference [SMD]: −0.03; 95% confidence interval [95% CI]: −0.06 to −0.01) and higher serum HDL (SMD: 0.07; 95% CI: 0.04–0.11). Patients with advanced exudative AMD had significantly higher serum LDL (SMD: 0.33; 95% CI: 0.04–0.62) compared with non-AMD controls. There were no other significant differences identified.

Conclusion

We found that there is significant heterogeneity in systemic lipoproteins in patients with AMD compared with non-AMD controls. The specific pattern of lipid dyshomeostasis appeared to be distinct based on AMD stage. These findings highlight both the underlying heterogeneity of AMD as well as the presence of distinct pathophysiological mechanisms involved at different stages or subtypes of AMD and may inform the development of novel therapeutic approaches.

Financial Disclosure(s)

Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

Age-related macular degeneration (AMD) is a leading cause of blindness in the industrialized world. Over 19 million Americans > 40 years of age have AMD, and of those, approximately 1.5 million Americans have vision-threatening forms of AMD.¹ Though we have therapies for advanced exudative disease targeting VEGF and a newly approved drug targeting the complement pathway that may have some role for slowing growth of geographic atrophy lesions,² there are no treatments available to prevent progression to advanced disease. As the disease burden of AMD will likely increase with the aging population, there is a need for further research to uncover the pathophysiology underlying AMD to aid in the development of novel therapeutic approaches.³

Numerous studies in preclinical models have suggested that lipid dyshomeostasis likely contributes to the pathophysiology of AMD, though the relationship between AMD and lipid homeostasis, as studied in humans, is incompletely understood.4, 5, 6, 7, 8, 9 Furthermore, there is no consensus regarding which elements of systemic lipid homeostasis are perturbed in AMD.⁶ We recently showed that high-dose statin therapy, which is known to lower levels of systemic low-density lipoprotein (LDL), may contribute to the regression of some high-risk features of AMD.¹⁰ Additionally, a meta-analysis in 2016 found that patients with elevated systemic triglycerides (TGs), total cholesterol (TC), and LDL had a lower prevalence of AMD, whereas patients with elevated high-density lipoprotein (HDL) had a higher prevalence of AMD.¹¹ Given a large number of new studies investigating this question since 2016, the goal of this systematic review and meta-analysis was to provide an updated characterization of systemic lipid homeostasis in AMD. These findings will not only improve our understanding of the relationship between lipid dyshomeostasis and AMD pathogenesis but may also contribute to the development of novel therapeutic strategies.

Methods

Protocol and Registration

We designed and executed a systematic literature search guided by a medical librarian (D.G.), following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 guidelines.¹² The review protocol was not registered prior to publication. This meta-analysis of published group-level data did not require institutional review board approval and informed consent was not applicable. The study adhered to the tenets of the Declaration of Helsinki.

Search Strategy and Information Sources

We performed a comprehensive literature search using Ovid MEDLINE (1946-), Embase.com (1947-), and Web of Science (Core Collection 1900-). Each search utilized a combination of controlled vocabulary and keywords focused on risk factors of AMD and dyslipidemia. The full search syntax used for each database is provided in Appendix A (available at https://www.ophthalmologyscience.org). The Cochrane filter for human studies was used for MEDLINE and Embase, and the English language database filters were used for all databases. We did not apply any restrictions based on study design, date of publication, or country of origin. The final searches were executed on May 31, 2022. This search identified a total of 9349 articles after duplicates were removed. We exported references into Endnote 7.8 for deduplication and then to Covidence for study screening, selection, and data extraction.

Eligibility Criteria, Study Selection, and Screening

The titles and abstracts for each of the manuscripts identified by our comprehensive search strategy were reviewed by one of 2 authors (B.L. or J.B.L.). Articles were omitted at this stage if they were clearly not relevant to the present study or if they were review articles. The full texts of the remaining 251 manuscripts were obtained and assessed for inclusion by 2 authors (B.L. and J.B.L.). We included all studies that quantitatively described the association between serum lipid species (TG, TC, HDL, or LDL) and AMD with an odds ratio (OR) or relative risk or provided group-level data (i.e., mean or median with standard deviation or interquartile range) comparing AMD and non-AMD groups. Studies were included if they investigated ≥ 1 serum lipid species. Studies that analyzed serum lipid species data in a categorical fashion (i.e., serum TG are normal [< 150 mg/dl], borderline high [150–199 mg/dl], or high [> 200 mg/dl]) rather than in a continuous fashion were excluded due to inability to meta-analyze this data. Conflicts were resolved by discussion at the end of each stage.

Data Extraction and Quality Assessment

We created a data extraction form to extract the study characteristics and outcomes from each of the included studies (Appendix B; available at https://www.ophthalmologyscience.org). We used a modified Newcastle–Ottawa scale to assess the quality of included studies (Appendix C; available at https://www.ophthalmologyscience.org).

Summary Measures and Synthesis of Results

We performed statistical analysis with R 4.2.1 and RStudio Desktop using the meta,¹³ metafor,¹⁴ effectsize,¹⁵ compute.es,¹⁶ tidyverse,¹⁷ and esc¹⁸ packages. For studies reporting group-level serum lipid values in patients with AMD versus control, we calculated standardized mean differences (SMDs) using Hedges’ g. For studies reporting ORs to describe the association between serum lipid levels and AMD, we converted ORs to SMD using Hedges’ g using the compute.es package.¹⁶ Four studies19, 20, 21, 22 reported a median and/or interquartile range due to nonnormal data, which we converted to an estimated mean/standard deviation using established formulas.²³ With all data represented as SMD, we ran a multilevel (nested), random-effects model to allow for the inclusion of multiple SMDs from individual studies (e.g., if a study reported data separately for early/intermediate AMD, advanced dry AMD, and advanced wet AMD). For one study,²⁴ the confidence interval (CI) of the OR was estimated to be OR = 1.00, 95% CI [0.996, 1.004] to account for likely rounding error in the published OR = 1.00, 95% CI [1.00, 1.00] since the corresponding authors (K.S.W. and P.S.J.) did not respond to our request for clarification. We also performed subgroup analysis based on AMD stage. Given the heterogeneity of AMD classification criteria used, we subdivided AMD into the following categories: early to intermediate nonexudative AMD (e.g., presence of medium to large drusen, pigmentary changes), advanced nonexudative AMD (i.e., presence of geographic atrophy), and advanced exudative AMD (i.e., presence of choroidal neovascularization). Studies that reported data for patients with “advanced AMD” but without specifying further details were unable to be included in subgroup analysis.

Assessment of Heterogeneity, Outliers, and Publication Bias

We assessed between-study heterogeneity using Higgin’s and Thompson’s I². We defined outliers as those with 95% confidence intervals (95% CIs) of their SMD that did not overlap with the 95% CI of the pooled effect size. Because of the significant between-study heterogeneity in the “all AMD” models and the fact that no significant differences were found in these comparisons, we performed sensitivity analyses in the subgroup analyses only, rerunning these models after omitting outliers, and not in the overall “all AMD” models. For all these analyses, removal of outliers did not significantly change the pooled effect size (Table S1; available at https://www.ophthalmologyscience.org). Additionally, we reanalyzed data after omitting data from the study of lower quality²⁰ based on the modified Newcastle–Ottawa scale and for the study²⁴ where the OR was adjusted to account for rounding error and found that this did not significantly change the observed results. Finally, to ensure that conversion of data from ORs to SMD did not introduce bias, we performed subgroup analyses using a mixed-effects model and found that there were no significant differences between the pooled effect of studies that underwent data conversion and ones that did not.

We analyzed publication bias with a funnel plot and tested for asymmetry with the rank correlation test. Most models showed no evidence of significant publication bias. The exception was the LDL and all AMD models (τ = 0.4058, P < 0.0001), though this was unlikely to introduce problematic bias, as we would expect publication bias to bias us toward a significant result while we found a null result instead.

Results

Study Characteristics

Our initial electronic search identified 9349 unique articles. After omitting articles that were clearly not relevant based on their title and abstract, we assessed the full texts of 251 articles. Of these, 81 (41%) were excluded for being the wrong publication type (e.g., conference abstracts), 72 (37%) were excluded since they did not quantify an association between lipids and AMD, 23 (12%) were excluded for being the wrong study design (e.g., inappropriate case definition), and 19 (10%) were excluded for not being relevant to this topic (e.g., not studying AMD). In total, this process yielded 56 articles for inclusion (Figure 1).

Figure 1.

Preferred Reporting Items for Systematic Reviews and Meta-Analyses flow diagram showing citations identified, screened, included, and excluded. AMD = age-related macular degeneration.

The meta-analysis included 308 188 patients, of which 265 477 were controls and 42 711 were patients with AMD. Of the patients with AMD, 13 060 were patients with early to intermediate nonexudative AMD, 146 were patients with advanced nonexudative AMD, and 319 were patients with advanced exudative AMD. The mean proportion of females was 54%, and the mean age was 66. The Wisconsin Age-Related Maculopathy Grading System²⁵ was used in 18 studies, the International Classification and Grading System²⁶ in 11 studies, the Age-Related Eye Disease Study²⁷ scale in 5 studies, the Rotterdam classification system²⁸ in 3 studies, the Three Continent Consortium²⁹ method in 3 studies, the Beckman classification system³⁰ in 2 studies, other nonstandardized classification criteria in 6 studies, and nonspecified methods in 8 studies. The most common countries in which the included studies were performed included 7 from the United States (13%),31, 32, 33, 34, 35, 36, 37 5 from Germany (9%),^19,38, 39, 40, 41 and 4 from Australia (7%).42, 43, 44, 45 A summary of study characteristics is provided in Table 2.

Table 2.

Study Characteristics for Studies Included in This Meta-analysis

Study; Country/Countries [reference]	Study Design	Mean Age	AMD Classification	Outcomes Data Type	Lipids Investigated	Number of Participants
						Control	All AMD	Early/Int dry AMD	Adv Dry AMD	Adv Wet AMD
Abalain et al (2022); France46	Case-control	76.2	WARMGS	Means	TG, TC, LDL, HDL	62	84	-	33	-
Ambreen et al (2014); Pakistan47	Case-control	63.7	ICGS	Means	TG, TC, LDL, HDL	100	90	-	-	-
AnandBabu et al (2016); India48	Case-control	59.1	AREDS	Means	TG, TC ,LDL, HDL	30	48	-	-	-
Arifoglu et al (2016); Turkey49	Case-control	63.8	Other	Means	TG, TC, LDL, HDL	24	20	-	-	20
Baker et al (2009); US37	Cross-sectional	78.3	WARMGS	Means	TG, TC, LDL, HDL	1737	351	-	-	-
Bikbov et al (2020); Russia50	Cross-sectional	58.5	Beckman	OR	TG, TC, LDL, HDL	4412	520	-	-	-
Blumenkranz et al (1986); US36	Case-control	76.0	Not specified	Means	TG, TC, LDL, HDL	26	23	-	-	-
Brandl et al (2016); Germany38	Cross-sectional	47.5	AREDS	OR	LDL, HDL	2208	343	270	-	-
Brandl et al (2022) - KORA-FF4; Germany39	Cohort	61.9	3CC	OR	LDL, HDL	172	122	122	-	-
Brandl et al (2022) - KORA-Fit; Germany39	Cohort	44.6	3CC	OR	LDL, HDL	172	122	122	-	-
Cackett et al (2008); Singapore51	Cross-sectional	58.7	WARMGS	OR	TG, TC, LDL, HDL	3075	190	169	-	-
Cezario et al (2015); Brazil52	Case-control	64.8	Not specified	Means	TG, TC, LDL, HDL	30	30	-	-	-
Cheung et al (2007); US35	Cross-sectional	59.8	WARMGS	Means	TG, TC	9500	454	454	-	-
Cheung et al (2012); Singapore53	Cross-sectional	53.7	AREDS	Means	TC, LDL, HDL	2961	211	-	-	-
Cheung et al (2013) - Central Indians; Singapore and India54	Cross-sectional	57.8	WARMGS	OR	TC, HDL	3213	209	209	-	-
Cheung et al (2013) - Singaporeans; Singapore and India54	Cross-sectional	57.8	WARMGS	OR	TC, HDL	3149	188	188	-	-
Cheung et al (2014); Singapore55	Cross-sectional	59.7	WARMGS	OR	TC	3046	266	-	-	-
Colak et al (2011); Serbia56	Cross-sectional	69.8	AREDS	Means	TG, TC, LDL, HDL	80	82	-	-	-
Colijn et al (2019); France, Germany, Italy, Netherlands, Norway, Portugal, UK57	Cross-sectional	69.5	Rotterdam	OR	TG, TC, LDL, HDL	19 539	5374	3768	-	-
Cougnard-Gregoire et al (2014); France58	Cross-sectional	80.2	ICGS	OR	TG, TC, LDL, HDL	630	308	247	-	-
Davari et al (2013); Iran59	Case-control	75.0	Other	Means	TG, TC, LDL, HDL	32	32	-	-	-
Delcourt et al (2001); France60	Cross-sectional	70.1	ICGS	OR	TG, TC, HDL	1415	768	730	-	-
Erke et al (2014); Norway61	Cohort	72.3	ICGS	OR	TG, TC, LDL, HDL	2048	727	635	-	-
Fan et al (2017); Several; US and non-US34	Case-control	-	Other	OR	TG, LDL, HDL	23 107	18 363	-	-	-
Fauser et al (2011); Germany, Netherlands40	Case-control	74.9	Other	Means	TG, TC, HDL	521	792	-	-	-
Fernandez et al (2012); US33	Cohort	62.0	WARMGS	Means	TC, LDL	5338	895	893	-	-
Hashemi et al (2018); Iran62	Case-control	-	Not specified	Means	TG, TC, LDL, HDL	15	15	-	-	-
Javadzadeh et al (2007); Iran63	Case-control	70.0	Not specified	Means	TG, TC, LDL, HDL	60	60	-	-	60
Joachim et al (2015); Australia42	Cohort	62.2	WARMGS	Means	TG, TC, HDL	2425	534	534	-	-
Jonasson et al (2014); Iceland64	Cross-sectional	74.7	WARMGS	OR	TC, HDL	2540	328	-	-	-
Kawasaki et al (2008); Japan65	Cross-sectional	60.4	WARMGS	Means	TC, HDL	1559	66	58	-	-
Klein et al (1993); US32	Cross-sectional	61.7	WARMGS	OR	TC, HDL	1746	773	745	11	17
Klein et al (2003); US31	Cohort	78.5	WARMGS	OR	TG, TC, LDL	1995	366	366	-	-
Koller et al (2022); Germany41	Cohort	78.2	3CC	Means	TG, TC, LDL, HDL	1635	627	-	-	-
Ludtke et al (2019); Germany19	Cohort	-	Rotterdam	Means	TG, TC, LDL, HDL	1088	753	745	-	-
Mao et al (2019); China66	Cohort	51.0	WARMGS	OR	TG, TC, LDL, HDL	4823	214	214	-	-
McCarty et al (2008); Australia43	Cross-sectional	73.9	ICGS	Means	TG, TC, LDL, HDL	160	160	-	-	-
Neethu et al (2020); India20	Cross-sectional	62.0	not specified	Means	TG, TC, LDL, HDL	39	39	-	-	-
Ngai et al (2011); UK67	Cohort	71.1	ICGS	Means	TG, TC, LDL, HDL	843	91	-	-	-
Nordestgaard et al (2021); Denmark21	Cohort	58.0	Not specified	Means	TG	104 916	1787	-	-	-
Ornek et al (2016); Turkey68	Case-control	73.0	Not specified	Means	TG, TC, LDL, HDL	78	98	-	-	47
Park et al (2014); Korea24	Cross-sectional	55.1	ICGS	OR	TG, TC, HDL	12 667	1046	958	22	66
Paunksnis et al (2005); Lithuania69	Cross-sectional	-	ICGS	Means	TG, TC	84	84	84	-	-
Peiretti et al (2014); Italy70	Cross-sectional	76.2	AREDS	OR	TC, HDL	38	136	98	-	-
Pertl et al (2016); Austria22	Cross-sectional	76.7	Other	Means	TG, TC, LDL, HDL	26	29	-	-	29
Qin et al (2014); UK71	Case-control	58.0	WARMGS	Means	TG, TC, LDL, HDL	14	14	-	-	-
Raman et al (2016); India72	Cross-sectional	72.0	ICGS	OR	TC, LDL, HDL	3805	986	893	-	-
Roh et al (2008); Korea73	Cross-sectional	52.7	Rotterdam	Means	TG, TC, LDL, HDL	9082	235	-	-	-
Semba et al (2019); Iceland74	Case-control	79.7	Not specified	Means	TC, HDL	80	160	-	80	80
Smith et al (1998); Australia44	Cross-sectional	62.2	WARMGS	OR	TG, TC, HDL	3342	312	240	-	-
Song et al (2009); Korea75	Cross-sectional	57.2	ICGS	Means	TG, TC, HDL	10 551	339	318	-	-
Taniguchi et al (2013); Japan76	Case-control	71.3	ICGS	Means	TG, TC, LDL, HDL	40	88	-	-	-
Tsang et al (1992); Australia45	Case-control	73.9	Other	Means	TG, TC	86	80	-	-	-
Ulas et al (2013); Turkey77	Cross-sectional	71.0	Not specified	Means	TG, TC, LDL, HDL	141	142	-	-	142
Vingerling et al (1995); Netherlands78	Case-control	71.0	WARMGS	Means	TC, HDL	1324	96	-	-	-
Wang et al (2012); China79	Cross-sectional	60.4	WARMGS	OR	TG, TC, LDL, HDL	2865	161	-	-	-
Xue et al (2021); China80	Cross-sectional	60.5	Beckman	OR	TG, TC, LDL, HDL	6112	1607	-	-	-
Yip et al (2015); UK81	Cohort	67.4	WARMGS	Means	TG, TC, LDL, HDL	4671	673	-	-	-

3CC = Three Continent Consortium; Adv = advanced; AMD = age-related macular degeneration; AREDS = Age-Related Eye Disease Study; HDL = high-density lipoprotein; ICGS = International Classification and Grading System; Int = intermediate; LDL = low-density lipoprotein; OR = odds ratio; TC = total cholesterol; TG = triglyceride; UK = United Kingdom; US = United States; WARMGS = Wisconsin Age-Related Maculopathy Grading System.

Quality Assessment of Included Studies

We assessed the quality of included studies based on a modified Newcastle–Ottawa scale (Appendix C; available at https://www.ophthalmologyscience.org). One study²⁰ had a score of 4, due to a lack of specific case definition, no information provided regarding the method used to obtain AMD diagnosis, and no measures performed to increase comparability between case and control groups. Otherwise, all other included studies scored ≥ 7, indicating acceptable study quality (Table S3; available at https://www.ophthalmologyscience.org). The removal of this low-scoring study from the models did not significantly impact any of the pooled effect sizes.

Systemic TGs in AMD

By meta-analyzing 53 effect sizes from 43 studies,19, 20, 21, 22^,24,31,34, 35, 36, 37^,40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 56, 57, 58, 59, 60, 61, 62, 63, 66, 67, 68, 69, 71, 73, 75, 76, 77, 79, 80, 81 we found that there was no significant difference between systemic TGs in patients with AMD versus controls (SMD: −0.03; 95% CI: −0.06 to 0.01) (Figure 2). Because there was significant interstudy heterogeneity (I² = 99.3%), we also performed subgroup analysis based on AMD stage, though heterogeneity remained high in the early to intermediate dry AMD (I² = 88.8%) and advanced wet AMD groups (I² = 96.0%). With this caveat in mind, subgroup analysis revealed a statistically significantly lower level of systemic TGs in patients with early to intermediate AMD versus controls (SMD: −0.03; 95% CI: −0.06 to −0.01). No significant difference in systemic TGs was found for patients with advanced dry AMD (SMD: −0.03; 95% CI: −0.94 to 0.88) or for patients with advanced wet AMD (SMD: −0.19; 95% CI: −0.89 to 0.51).

Figure 2.

Forest plot for systemic triglycerides (TG) in patients with age-related macular degeneration (AMD) versus non-AMD controls. Blue diamonds denote the 95% confidence interval (95% CI); dashed lines denote the 95% prediction interval. SMD = standardized mean difference.

Systemic TC in AMD

By meta-analyzing 74 effect sizes from 52 studies,^19,20,22,24,31, 32, 33^,35, 36, 37^,40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 56, 57, 58, 59, 60, 61, 62, 63, 66, 67, 68, 69, 71, 73, 75, 76, 77, 79, 80, 81, 54, 55, 53, 64, 65, 70, 72, 74, 78 we found that there was no significant difference in systemic TC levels in patients with AMD versus controls (SMD: −0.01; 95% CI: −0.02 to 0.00) (Figure 3). There was moderate to high heterogeneity (I² = 75.4%) among meta-analyzed studies. Similarly, after subgroup analysis, there were no significant differences in systemic TC levels when examining patients with early to intermediate dry AMD (SMD: −0.01; 95% CI: −0.03 to 0.00), patients with advanced dry AMD (SMD: −0.01; 95% CI: −0.01 to 0.00), or patients with advanced wet AMD (SMD: 0.08; 95% CI: −0.04 to 0.20) when compared with controls.

Figure 3.

Forest plot for systemic total cholesterol (TC) in patients with age-related macular degeneration (AMD) versus non-AMD controls. Blue diamonds denote the 95% confidence interval (95% CI); dashed lines denote the 95% prediction interval. SMD = standardized mean difference.

Systemic LDL in AMD

By meta-analyzing 49 effect sizes from 38 studies,^{19,20,22,31,33,34,}36, 37, 38, 39^,41,43,46, 47, 48, 49, 50, 51, 52, 53, 56, 57, 58, 59^,61, 62, 63, 66, 67, 68^,71, 72, 73^,76, 77, 79, 80, 81 we found no significant difference in systemic LDL in patients with AMD versus controls (SMD: −0.00; 95% CI: −0.01 to 0.00) (Figure 4). Again, because of significant interstudy heterogeneity (I² = 84.9%), we also performed subgroup analysis based on AMD stage, which improved the extent of heterogeneity. This subgroup analysis revealed that there was significantly higher LDL among patients with advanced wet AMD versus control (SMD: 0.33; 95% CI: 0.04–0.62; I² = 53.3%). In contrast, patients with early to intermediate dry AMD (SMD: −0.01; 95% CI: −0.02 to 0.00; I² = 90.0%) or with advanced dry AMD (SMD: −0.10; 95% CI: −1.72 to 1.52; I² = 0.0%) did not exhibit a significant difference in systemic LDL when compared with controls.

Figure 4.

Forest plot for systemic low-density lipoprotein (LDL) in patients with age-related macular degeneration (AMD) versus non-AMD controls. Blue diamonds denote the 95% confidence interval (95% CI); dashed lines denote the 95% prediction interval. SMD = standardized mean difference.

Systemic HDL in AMD

By meta-analyzing 69 effect sizes from 49 studies,^{19,20,22,24,32,34,}36, 37, 38, 39, 40, 41, 42, 43, 44^,46, 47, 48, 49, 50, 51, 52, 53, 64, 56, 57, 58, 59, 60, 61, 62, 63, 66, 67, 68, 69, 71, 73, 75, 76^,70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81 we found no significant difference in systemic HDL in patients with AMD compared with controls (SMD: 0.19; 95% CI: −0.10 to 0.48) (Figure 5). There was high heterogeneity among meta-analyzed studies (I² = 100.0%). Even after subgroup analysis, heterogeneity remained high in the early to intermediate dry AMD (I² = 99.0%). With this caveat in mind, patients with early to intermediate dry AMD exhibited significantly higher systemic HDL compared with controls (SMD: 0.07; 95% CI: 0.04–0.11). There was no significant difference in systemic HDL when comparing patients with advanced dry AMD (SMD: 0.01; 95% CI: −0.04 to 0.07; I² = 0.0%) or with advanced wet AMD (SMD: −0.06; 95% CI: −0.18 to 0.06; I² = 23.6%) to controls.

Figure 5.

Forest plot for systemic high-density lipoprotein (HDL) in patients with age-related macular degeneration (AMD) versus non-AMD controls. Blue diamonds denote the 95% confidence interval (95% CI); dashed lines denote the 95% prediction interval. SMD = standardized mean difference.

Discussion

Here, we performed a comprehensive systematic review with guidance from an experienced medical librarian (D.G.) to update our understanding of the association between systemic lipoprotein profiles and AMD. Our meta-analysis included 58 studies and 308 188 patients, almost 3 times as many compared to the prior 2016 meta-analysis by Wang et al,¹¹ which included 19 studies and 104 270 patients. When analyzing all patients with AMD regardless of stage of disease, we found no differences in systemic TGs, TC, LDL, or HDL compared to controls (Table 4). On the other hand, when analyzing patients with early to intermediate AMD separately, there were significantly lower levels of TGs and higher levels of HDL but no differences in the levels of TC or LDL. Patients with advanced dry AMD had no significant differences in TGs, TC, LDL, or HDL. Finally, patients with advanced wet AMD had significantly higher systemic LDL but no differences in TGs, TC, or HDL. Taken together, our results highlight the significant heterogeneity of AMD and suggest that the specific aspects of lipid homeostasis that are perturbed may depend on the underlying stage or subtype of disease.

Table 4.

Summary of SMDs with 95% Confidence Intervals and Measure of Heterogeneity (I²) in Triglycerides, Total Cholesterol, LDL, and HDL between Patients With AMD or Specific Stages of AMD Compared with Controls

	SMD [95% Confidence Interval]; I2
	All AMD	Early/Intermediate Dry AMD	Advanced Dry AMD	Advanced Wet AMD
Triglycerides	−0.03 [−0.06, 0.01]; 99.3%	−0.03 [−0.06, −0.01]; 88.8%	−0.03 [−0.94, 0.88]; 19.6%	−0.19 [−0.89, 0.51]; 96.0%
Total cholesterol	−0.01 [−0.02, 0.00]; 75.4%	−0.01 [−0.03, 0.00]; 93.7%	−0.01 [−0.01, 0.00]; 0.0%	0.08 [−0.04, 0.20]; 43.3%
LDL	−0.00 [−0.01, 0.00]; 84.9%	−0.01 [−0.02, 0.00]; 90.0%	−0.10 [−1.72, 1.52]; 0.0%	0.33 [0.04, 0.62]; 53.3%
HDL	0.19 [−0.10, 0.48]; 100.0%	0.07 [0.04, 0.11]; 99.0%	0.01 [−0.04, 0.07]; 0.0%	−0.06 [−0.18, 0.06]; 23.6%

AMD= age-related macular degeneration; HDL = high-density lipoprotein; LDL = low-density lipoprotein; SMD = standardized mean difference.

These findings may explain, in part, why there is a discrepancy in the literature published to date with respect to the specific facets of lipid dyshomeostasis that are associated with the pathophysiology of AMD.⁶ We speculate that among patients with AMD, there may be facets of lipid dyshomeostasis that are specific to certain stages of the disease process. In fact, recent studies have shown that patients with soft drusen are distinct from those with subretinal drusenoid deposits (SDDs) on the basis of their systemic associations, serum profiles, and genetic risk profiles and provided evidence that high HDL was associated with classic soft drusen, whereas low HDL was associated with SDDs.^82,83 Interestingly, these authors also found an association between SDDs and the presence of coexisting systemic vascular disease (e.g., aortic stenosis, myocardial infarction) and suggest that both of these (i.e., low HDL and presence of SDDs) may be related to systematic atherosclerosis.

Further research is necessary to explore this heterogeneity within patients with AMD. It may be necessary to tailor therapy for these disparate underlying pathophysiologies. We have previously shown in a small pilot study that high-dose statin therapy may lead to reduction in progression to advanced exudative AMD in high-risk patients.¹⁰ Similarly, in a study of commercially insured patients, no patients taking very-high-dose lipophilic statins progressed to exudative AMD.⁸⁴ These findings correspond well with the findings of this meta-analysis, as statins are most known for lowering systemic LDL, and elevated LDL was found to be most prominent in patients with advanced wet AMD. Lipid-based therapies may be efficacious if used in the appropriate patient population. Future studies investigating the association between systemic dyslipidemia and AMD should include careful characterization of patients with AMD. Some factors that would be important to consider are AMD stage (i.e., early/intermediate vs. late), intermediate AMD subtype (i.e., drusen vs. SDDs), demographic factors (e.g., age, sex, ethnicity), presence of coexisting systemic vascular disease, and genetic variants in genes related to lipid pathways such as cholesteryl ester transfer protein, apolipoprotein E, and hepatic lipase.⁸⁵ These detailed studies may identify subtypes of AMD and thereby partially explain the heterogeneity found within the present study.

Our meta-analysis had several limitations. The most prominent limitation was the moderate to high heterogeneity of the meta-analyzed effect sizes (I² > 90%), which affects the precision and reliability of the pooled effect size estimate. In many cases, heterogeneity remained high even after subgrouping the patients by AMD stage. This interstudy heterogeneity was likely related to underlying differences with respect to the studies’ country of origin, the AMD classification criteria used, the types of outcome data presented, and the study design, among other factors. However, they could also reflect the significant heterogeneity of AMD as a disease process and suggest that there is a need for improved, more precise AMD classification criteria based on our improving understanding of its underlying pathophysiological mechanisms at the cellular and molecular level.

Additionally, there were much less data from patients with advanced dry AMD (N = 146) or advanced wet AMD (N = 319) compared to those from patients with early to intermediate dry AMD (N = 13 060), leading to reduced statistical power in these associated meta-analyses. Finally, because studies differed in terms of how they reported their outcomes (e.g., reporting means of various groups vs. reporting ORs), we had to convert all of these outcomes to SMDs to allow for an estimate of the pooled effect size. Though we used validated mathematical formulas for these conversions, they could have introduced bias due to their underlying assumptions.

Overall, our findings support that there is systemic dyshomeostasis in patients with AMD compared with non-AMD controls and that the specific pattern of impaired lipid homeostasis appears to depend on AMD stage. We propose that these findings reflect the underlying heterogeneity of AMD, the presence of distinct pathophysiological mechanisms involved at different stages of AMD, and may even suggest the existence of distinct AMD subtypes that should be investigated further to inform novel therapeutic approaches.

Manuscript no. XOPS-D-23-00062.