The U-Shaped Association between Bilirubin and Diabetic Retinopathy Risk: A Five-Year Cohort Based on 5323 Male Diabetic Patients

Aims This study aimed at assessing the impact of baseline bilirubin (TBiL) on the incidence of diabetic retinopathy (DR) based on a five-year cohort study which consisted of 5323 Chinese male diabetic patients. Methods A cohort study based on 5323 male diabetic patients was conducted in Beijing, from 2009 to 2013. Both baseline TBiL and follow-up changes were measured. Cox proportional risk model was used to calculate the hazard ratio (HR) of TBiL for DR risk. Results During the follow-up period, there were 269 new DR cases. The incidence of five-year follow-up was 5.1% (95% CI: 4.5%~5.6%). The TBiL level of those who had diabetic retinopathy was lower than that of those without (12.51+ 1.20 mol/L and 13.11+ 1.32 μmol/L, P = 0.033). And more interestingly, along with the quintiles of baseline TBiL, there showed a U-shaped curve with DR incidence. And the RRs were 0.928 (95% CI: 0.646–1.331), 0.544 (95% CI: 0.365–0.811), 0.913 (95% CI: 0.629–1.324), and 1.035 (95% CI: 0.725–1.479) for the second, third, fourth, and fifth quintiles of baseline TBiL levels, respectively, compared with the first quintile. For follow-up TBiL changes, after being adjusted for related covariables and baseline TBiL levels (as continuous variable) in the model, the RRs for DR were 1.411 (95% CI: 1.081–1.842) for those who had decreased TBiL level and 0.858 (95% CI: 0.770–0.947) for those who had increased TBiL level during follow-up. And this association was more prominent among those with lower baseline TBiL level. Conclusions Serum TBiL had a U-shaped relationship with DR incidence, which was independent of control status of diabetes and other related covariates.


Introduction
Diabetic retinopathy (DR) is one of the important vascular complications of diabetes. Data show that this is the main cause of blindness among working age population in developing countries [1]. Therefore, exploring the pathogenesis of DR is of most importance [2]. Previous basic and clinical research data indicated that oxidative stress played an important role in the development of DR.
Total bilirubin (TBIL) has been considered as a powerful endogenous antioxidant in recent years. A number of studies have shown that elevated TBiL levels were negatively correlated with cardiovascular disease and diabetes mellitus [3][4][5]. In addition, a number of studies have reported a protective relationship between TBiL levels and diabetic vascular complications. There was also a meta-analysis of the association between TBiL and DR published in 2016 [6]. However, these studies were mostly cross-sectional or casecontrol studies with small sample sizes and inconsistent findings. Some studies show that there was no association between TBiL and DR risk [7,8]. In addition, there was little evidence of dose-response effects, which was of great value in determining appropriate clinical thresholds of TBiL levels among diabetic patients. Therefore, based on this five-year cohort study with a large sample (more than 5000 elderly diabetic patients), our study assessed the relationship between TBIL and its changes and the incidence of DR.

Subjects.
This cohort study consisted of elderly diabetic patients. We conducted the baseline survey in 2009, and a total of 6861 elderly were recruited, and the follow-up survey was done in 2013. The details were described in a previous article [9]. Considering the possible effects of related diseases on TBiL or diabetes, we excluded those patients who had hepatobiliary diseases (n = 290), malignant tumors (n = 53), acute diabetic complications such as ketoacidosis (n = 12), previous ocular diseases including cataract, glaucoma, and other ocular diseases (n = 340). Considering female just accounted for 13.7% (843) of the total participants, only 5323 male participants were left for further analysis. Figure 1 shows the flow chart of participants' inclusion.

Data Collection.
Baseline anthropometric and physical examination information was collected according to the standard process. Fasting blood was collected, and the related biochemical indexes were detected. The details were described in a previous article [9].
2.3. Definitions. The duration of diabetes was calculated by age minus the age of first diagnosis on diabetes. Quintiles of baseline TBiL level were defined as follows: Q1: ≤9.20 μmol/ L, Q2: 9.20-12.60 μmol/L, Q3: 12.60-13.80 μmol/L, Q4: 13.80-16.50 μmol/L, and Q5: ≥16.50 μmol/L. Follow-up changes of TBiL were defined as follow-up TBiL levels minus baseline TBiL levels and were divided into three categories: ≤−2 μmol/L; −2 to 2μmol/L; and ≥2 μmol/L. Diabetes, hypertension, and dyslipidemia were defined according to the corresponding guidelines. DR was defined according to the Chinese version of guidelines for the prevention and treatment of type 2 diabetes: the presence of mild or moderate or proliferative retinopathy in either eye [10], and was independently diagnosed by two senior ophthalmologists.

Statistical Analysis.
For continuous data, mean ± SD was used for description, and analysis of variance was used for comparisons. For categorical variables, chi-square test is used for comparison. Multivariate Cox proportional hazard model was used to estimate the hazard ratio (HR) and 95% confidence intervals (CIs) of DR based on baseline TBiL levels and follow-up TBiL changes. Restricted cubic spline functions with 5 knots were used to test the potential nonlinear association and display the data graphically. SPSS software was used for data analysis. P < 0 05 was statistically significant.

Results
3.1. Baseline Characteristics according to DR Incidence.  baseline TBiL level (p < 0 05). Baseline characteristics according to quintiles of baseline TBiL levels were presented in Appendix Table 1. Along with the increase of baseline TBiL levels, it showed shorter duration; higher Hb, HDL-C, FPG, and 2hPG levels; and lower percentage of diabetes control status.

Incidence of DR according to Baseline TBiL Quintiles and Follow-Up TBiL
Changes. There were a total of 269 DR cases during the 21,586 person-years. The total five years' incidence was 5.1% (95% CI: 4.5%-5.6%). As we can see from Table 2, the incidence of the quintiles of baseline TBiL for DR had fluctuations; the third quintile had the lowest incidence while the first and the fifth had the highest incidence. The incidence density showed a similar trend. The incidence of follow-up TBiL changes for DR was lowest among those who had an increase of ≥2 μmol/L while highest among those who had a decrease of ≤−2 μmol/L ( Table 2).
3.3. HRs and 95% CI of DR Incidence according to Baseline TBiL Levels. Table 3 showed the HRs of baseline TBiL levels for DR incidence. We adjusted related covariables according to results of univariate analysis (Appendix Table 1 Table 4, compared with those who had relatively stable TBiL levels (−2 μmol/ L < follow-up TBiL changes < 2 μmol/L), the HR for DR incidence was higher (HR = 1.411, 95% CI: 1.081-1.842) among those who had follow-up TBiL changes ≥2 μmol/L, and the HR was lower (HR = 0.858, 95% CI: 0.770-0.947) among those with follow-up TBiL changes ≤−2 μmol/L. Besides, the decreasing trend of follow-up TBiL changes for DR incidence was more obvious among those with lower baseline TBiL level (≤12.5 μmol/L, n = 2669). However, among those with relatively higher baseline TBiL level (>12.5 μmol/L, n = 2654), the HRs of follow-up TBiL changes showed no significant results.
In the sensitivity analysis, when participants who had DR that happened within less than 2 years were excluded (n = 37) or divided by age groups (≤80 yrs and >80 yrs), the trend of adjusted HRs was similar with the results from Tables 3  and 4 (Appendix Tables 3, 4, 5, and 6).

Discussion
In this study, we did a deep study and evaluated the relationship between baseline TBiL and DR incidence using a large cohort of more than 5000 male elderly. The results showed that baseline TBiL had a U-shaped relationship with DR incidence, rather than a simple linear relationship. And this association was independent of control status of diabetes and other related covariates. And more interestingly, increased follow-up TBiL changes had a higher DR risk, and this association was more prominent among those with lower baseline TBiL level. There were several researches focused on relationships between TBiL level and DR risk [11,12], even one metaanalysis published in 2016 [6]. Most of the previous studies showed that there was a negative relationship between TBiL and DR. However, the result of our deep study showed that the association between TBiL and DR risk was not a simple linear association but a U-shaped curve. From the metaanalysis of TBiL and DR, we could see that most studies were cross-sectional ones or case-control studies with a small sample. Our study was based on a five-year cohort which consisted of more than 5000 elderly diabetic patients. And this U-shaped curve was consistent with previous ones about relationships about TBiL with cardiovascular diseases. A cohort study based on 7685 middle-aged British men firstly revealed that there was a U-shaped relationship between TBiL and risk of ischemic heart disease [13]. And in 2012, results from one of the biggest cohorts based on more than 130,000 patients who received statin treatment showed that a U-shaped association appeared between TBiL before statin prescription and coronary heart diseases [14]. Our result was consistent with these prospective studies [15]. And this implies that the DR risk would no longer decrease when baseline TBiL increased to a relatively higher level. Meanwhile, there were a series of evidences showing that a higher TBiL level (beyond normal range) indicated hepatocellular injuries, and the latter one was proved to be associated with increased risk of diabetes and cardiovascular disease [7]. Besides, the negative relationship might be part of the U-shaped curve, since most of the crosssectional studies were of small sample size. And the actual U-shaped association might be the combination of antioxidation and liver toxicity effects.
The pathogenesis of DR has not been fully studied. Oxidative stress caused by high glucose is one of the hotspots of the current research [16]. TBiL, not only a metabolite of hemoglobin, is also considered to be an important endogenous antioxidant. Past studies have shown that TBiL has important protective effects on cardiovascular disease, diabetes, and diabetic macrovascular complications, mainly through anti-inflammatory and antioxidant effects [17,18]. For DR, studies have revealed that TBiL has significant protective effects [6]. And results from the first known human case of heme oxygenase-1 deficiency showed that TBiL at physiological concentration has strong antioxidant activity, which can prevent low-density lipoprotein lipid peroxidation, and results based on mice showed that 1 mol/L direct bilirubin can remove about 2 mol/L free radicals [19]. All these evidences support the protective effect of TBiL on DR. However, the premise is in the normal range. If the TBiL was beyond the normal range, it is most probably an indicator of liver damage, which was an important risk factor for diabetes and its complications. There were several prospective studies with large sample size including our study which have showed the protective effect of TBiL in the normal range and the harmful effect in the relative higher range for cardiovascular disease and diabetes (including its complications) [14,15,20], which confirmed the U-shaped curve of TBiL for DR in our study. This implies that in clinical practice, the previous "higher TBiL indicates lower risk of complications" was not always right. Doctors should also pay enough attention to the abnormal range and whether there were liver toxicity effects.
Follow-up changes of TBiL also had protective effects on DR incidence, which indicated that medical workers should also pay attention to both the baseline and the fluctuation of TBiL levels among diabetic patients, and it will play an important role in predicting DR incidence. Most of the previous studies only assessed the relationship between baseline TBiL level and DR; however, only a few studies focused on the follow-up volatility. In our study, for those who had increased TBiL changes (≥2 μmol/L), the risk of DR incidence had increased about 40%. And the more obvious decreasing trend among those with relatively lower baseline TBiL levels verified previous results about the U-shaped curve of the association between TBiL and DR risk.
As far as we know, this was the first study to analyze the relationship between TBiL and DR based on a large sample cohort in fully adjusted models. In addition, besides baseline TBiL levels, TBiL changes also had an independent and inverse impact on DR incidence. The cohort had rigorous investigation process, strict training for all the staff in the field survey, and high response rate because of the wellcontrolled follow-up system. However, this study had the following limitations. First, TBiL was measured only once at either baseline or followup, so it could not reflect the actual fluctuations. Second, only TBiL was collected; there was no specific value for direct and indirect bilirubin levels, so it was impossible to distinguish the exact role of these two types. Third, all participants were retired male, whose economic and medical security was relatively good, and the representation was limited for the general population.
In summary, this large-sample cohort study had shown that TBiL and its changes in diabetic patients were independently associated with DR incidence, and this association was independent of both classical risk factors and diabetes control status. Clinical medical staff should pay attention to the monitoring of TBiL levels in order to early detect and better control DR.