Thyroid function and age-related macular degeneration: a prospective population-based cohort study - the Rotterdam Study

The Rotterdam Study is a prospective population-based cohort study that addresses determinants and occurrence of cardiovascular, neurological, ophthalmologic, psychiatric, and endocrine diseases in the elderly living in Ommoord, a suburb of Rotterdam. The aims and design of the Rotterdam study have been described in detail elsewhere [22]. For this analysis we included participants from two independent cohorts from the Rotterdam Study. The Rotterdam Study Cohort 1 (RSI) started in 1989 and included a total of 7,983 participants (response rate 78%) aged 55 years and older. Baseline data were collected from 1990 until 1993, and four follow-up examinations were performed in 1993–1995, 1997–1999, 2002–2004, and 2009–2011. The second cohort is the Rotterdam Study Cohort II (RSII), which includes a total of 3,011 participants (response rate 67%) aged 55 years and older. Baseline data were collected from 2000–2001, and follow-up examinations were performed in 2004–2005 and 2011–2012.

Participants from baseline study cohorts RSI (RSI-1) and RSII (RSII-1) were eligible for these analyses if they had TSH and/or FT4 measurements and had gradable fundus photographs at baseline and at least one follow-up eye examination. Since not all participants from RSI had thyroid measurements at baseline, additional baseline samples were drawn from RSI visit 3 (RSI-3). Participants with AMD at baseline (N = 567) were excluded from further analyses. In total, 5,573 participants from these two cohorts were eligible to be included in our analyses (Additional file 1: Figure S1). The Medical Ethics Committee of the Erasmus University approved the study protocols, and participants gave a written informed consent in accordance with the Declaration of Helsinki.

For RSI-1, serum TSH (TSH Lumitest; Henning, Berlin, Germany), anti-TPOAb (ELISA; Milenia; Diagnostic Products Corp, Los Angeles, CA, USA) and free T4 levels (FT4; Vitros, ECI Immunodiagnostic System; Ortho-Clinical Diagnostics, Amersham, UK) were determined in a random subset of the baseline serum samples (n = 1,855). Thyroid function assessment was also performed in baseline serum samples for TSH and FT4 (electrochemiluminescence immunoassay for thyroxine and thyrotropine, “ECLIA”, Roche) for RSI-3 and RSII-1. The tests’ TSH reference ranges did not differ substantially and had a good Spearman correlation coefficient (0.96 for TSH, P < 0.0001 and 0.81 for FT4, P < 0.0001). We determined the cut-off values for normal range TSH as 0.4-4.0 mIU/L according to national guidelines. The reference range for FT4 was 11–25 pmol/L, and anti-TPOAb levels greater than 60 kU/mL were regarded as positive.

All eligible participants underwent fundus photography after pharmacologic mydriasis. For visits RSI-1 to RSI-3 and RSII-1 a 35° film fundus camera was used (Topcon TRV-50VT, Topcon Optical Company, Tokyo, Japan), after which a 35° digital color fundus camera (Topcon TRC-50EX, Topcon Optical Company, Tokyo, Japan with a Sony DXC-950P digital camera; 0.44 megapixel, Sony Corporation, Tokyo, Japan) followed for visits RSI-4, RSI-5, RSII-2, and RSII-3. Fundus transparencies were graded according to the Wisconsin age-related maculopathy grading system [23] and the modified International Classification System [24] by trained graders under the supervision of senior retinal specialists (JRV, CCWK). The eyes of each participant were graded and classified separately, and the eye with the more severe grade was used to classify the person. In the analyses incident early and late AMD combined was used as the outcome variable. In the manuscript this is referred to as AMD. Besides AMD we also investigated AMD-specific lesions as a separate outcome variable. These lesions included retinal pigmentary alterations, large drusen (≥125 μm), and large drusen area (≥5,331,820 μm²) [25].

Smoking was derived from computerized baseline questionnaires, and participants were categorized as current or non-current smokers. Blood pressure, systolic and diastolic, was calculated as the average of two consecutive measurements, using random-zero mercury sphygmomanometers. Hypertension was defined as a systolic blood pressure ≥ 140 mmHg or a diastolic blood pressure ≥ 90 mmHg or participant use of anti-hypertensive medication at baseline. Cholesterol was measured at baseline by the CKCL (Centra Clinical Chemical Laboratory) of the Erasmus University Medical Center. A subgroup of measurements was carried out in the laboratory of the Department of Epidemiology and Biostatistics (Erasmus University Medical School). History of diabetes was defined by a repeated impaired fasting glucose ≥ 7 or use of anti-glycemic medication at baseline. Body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared.

Participant baseline characteristics were compared using a χ² or t-test. Due to a skewed distribution, TSH was log-transformed for the statistical analyses. We used the Cox proportional hazards model to calculate the relationship between TSH and FT4 at baseline and the risk of incident AMD, first including all participants and then including only those with normal range TSH and/or FT4 values. We performed a crude Cox model including only thyroid parameters, after which we also included quadratic and cubic terms to explore possible nonlinear relationships. We then performed additional models adjusting first for age and sex and second also adding smoking, hypertension, cholesterol, diabetes, and BMI to the model. Hypertension, cholesterol, diabetes, and BMI could act as confounders and possible mediators depending on the presumed pathway through which thyroid function is related to AMD. These variables were included in the multivariable model as possible confounders of non-vascular pathways. We looked at the association between AMD and TSH or FT4 both continuously and in quintiles, as well as overall and within the normal range of TSH. The middle quintile was used as a reference group, as biologically it is expected to represent the subgroup with the most normal thyroid function within the euthyroid group. We performed predefined stratification by sex and age categories, using a cutoff of 65 years, as this is the median of the current population and the treatment threshold for subclinical thyroid dysfunction according to the European guidelines [26]. Further interaction terms were introduced to the model to explore possible differential risk patterns. We performed a sensitivity analysis excluding those using thyroid medication at baseline (levothyroxine and anti-thyroid drugs) and those with prior self-reported thyroid disease at baseline. We also performed FT4 and TSH analyses with specific AMD lesions defined as retinal pigment alteration, large drusen, and large drusen area as separate outcome variables to examine possible early changes in underlying pathways. To address the issue of drop-out of individuals during follow-up that could possibly not be completely at random, we adjusted the model for inverse probability weights (IPWs). These were calculated using possible baseline explanatory variables for drop-out such as smoking, BMI, and medication use. The proportional hazards assumption was checked statistically using the Schoenfeld test and assessing the Schoenfeld plot. All statistical analyses were performed using SPSS version 21 (IBM SPSS, Armonk, NY, USA) except for the Schoenfeld tests and (Schoenfeld) plots which were performed in R (survival package, R-project, Institute for Statistics and Mathematics, R Core Team (2013), Vienna, Austria, version 3.0.2).

Genome-wide association studies (GWASs) have been performed for AMD [27] and thyroid function (TSH and FT4) [28,29]. These studies identified several single nucleotide polymorphisms (SNPs) associated to these two phenotypes. Some of the genome-wide significant SNPs in the AMD GWAS might also play a role in thyroid function and vice versa. Overlap between common genetic polymorphisms can provide insight into possible shared genetic pathways. It might also elucidate a mediation effect between the two phenotypes, that is, identify and explicate the process that underlies a possible observed relationship between thyroid function and AMD. To evaluate these potential genetic pathways, we conducted a bidirectional genetic look-up using the results of the above mentioned GWASs for AMD and thyroid function. We first extracted SNPs that reached genome-wide significance from the AMD GWAS performed by the AMD Gene Consortium [27]. We then checked whether these were significantly associated with TSH or FT4 in the thyroid function GWAS performed by Porcu et al. [28]. Hereafter we extracted the genome-wide significant SNPs for TSH or FT4 from the thyroid function GWAS and checked whether they were associated with AMD in the AMD GWAS. For the significance level, we applied a multiple testing correction (Bonferroni correction), using a P-value threshold of 0.05 divided by the amount of significant SNPs per GWAS. In case of a significant finding, we added the SNP to the multivariable model to evaluate a possible mediation effect.