The Association Between Anion Gap and Diabetic Retinopathy: Based on Hospital Registry Data in Liaoning Diabetic Microvascular Complications Study


 Background: Current evidences supported a highly relevant between acid-load and diabetes. The rise of serum anion gap (AG) has performed the significances in metabolic acidosis. Some factors correlated to diabetes might affect its complications such as diabetic retinopathy (DR).Methods: A total of 3,411 Chinese adults with type 2 diabetes were selected form the hospital registry data on DR Sub-study of Liaoning Diabetic Microvascular Complications Study (LD-MCS), included 1,137 DR as the cases and 2,274 non-DR as the matching controls. Fasting venous blood test was applied to ascertain the ion levels, and serum AG (mmol/L) was computed by: AG = (sodium + potassium) – (chlorine + bicarbonate). According to Early Treatment for Diabetic Retinopathy Study standards, DR was diagnosed by two-field fundus photographs and classified as mild non-proliferative DR (NPDR), moderate NPDR, and vision-threatening DR (VTDR). Logistic regression models and linear regression models were used to analyze the relationships.Results: In setting of the non-acidosis, higher AG was significantly associated with DR (P = 0.001), increased with aggravating retinopathy but decreased in VTDR level (P < 0.001). Multivariable-adjusted models showed that AG quartiles were independently linked with higher odds to occurrence (P for trend < 0.001) and severity (P for trend < 0.05) of DR, but with attention to the fluctuation of 75th AG quartiles. Linear logistic regression by stepwise method suggested the growth of age (P = 0.014), glycated hemoglobin (P = 0.018), and the homeostasis model assessment of insulin resistance index (P < 0.001) played an intimate role in the association between AG and DR.Conclusions: Higher AG was independently related to the occurrence and progression of DR. Our findings suggested that serum AG might alter the risk of DR by affecting glucose metabolism and insulin sensitivity in patients with type 2 diabetes.

The Association Between Anion Gap  Original investigation was also suggested to worsen insulin resistance [3] Other indicators of acidosis included blood ketone body [7] and related urine tests [8,9] were suggested to alter the status of diabetes.
In view of the causality between diabetes and its complications, some status correlated to diabetic metabolism would be treated as essential conditions in the trigger or progression of retinopathy. Serum AG, as a frequently-used testing index in clinics, is served to evaluate the acid-base balance of human body, and its rise has performed the signi cant utility in the diagnosis of metabolic acidosis [10]. To date, the relationship between serum AG and DR has not yet been evaluated. Thus, this study designed a nested case-control model to analyze the relationship between serum AG and the presence as well as severity of DR based on the hospital-based population with type 2 diabetes. shown in Additional le (see Additional le 1). In current DR sub-study, designed as a nested case-control research, selected 3,411 subjects from LD-MCS. The selection method of the subjects was in accordance with Figure S1 (see Additional le 2). Firstly, 1,073 of 5,066 (21.2%) subjects were excluded for the following reasons: 1) diabetic ketoacidosis, hyperosmotic nonketotic diabetic coma and other diabetic acute metabolic disorders (n = 86, 1.7%); 2) the level of estimated glomerular ltration rate (eGFR) ≤ 60 ml/min, albuminuria or history of chronic kidney disease (n = 656, 12.9%); 3) pregnancy or lactation (n = 114, 2.3%); 4) major mental or physical illness for more than 3 years (n = 159, 3.1%), including myocardial infarction, cerebral infarction, malignant tumor, severe organ dysfunction and mental disorder; 5) incomplete information (n = 58, 1.1%). Then, after reviewing the pathological and clinical situations of the remaining 3,993 subjects, 1,137 (28.5%) patients with type 2 diabetes were diagnosed with DR, and they were de ned as the case group. Two controls for each case (n = 2,274) were randomly sampled from the same hospitalized register database, and they were alive and no history of DR. All controls and cases were matched according to gender and age, and used the quotas performed on the diagnosis date. Finally, the complete medical records of 3,411 con rmed subjects was eligible for analyzing.

Measurements
Each participant from DR Sub-study of LD-MCS carried out a comprehensive clinical assessment and her/his basic characteristics were recorded. This study extracted the following information, included  [11]. The level of eGFR was estimated by Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) creatinine equation [12].
De nition Type 2 diabetes was de ned following the standards of American Diabetes Association (ADA) [13]. Fundus assessments depended on two-eld colorful retinal images, and the diagnosis of DR was con rmed from the worse eye abiding by the criterion of Early Treatment for Diabetic Retinopathy Study (ETDRS) [14]. DR was rated as 3 severity scales, included mild non-proliferative DR (NPDR), moderate NPDR, and vision-threaten DR (VTDR) [14,15]. The whole process of fundus photographic and diagnosis was completed by trained ophthalmologists in a double-blind condition. Serum AG was de ned as the concentration difference between undetermined anions (UA) and undetermined cations (UC), which was generally calculated as Na + -(Cl − + HCO 3 − ). Because of the concentration of K + in blood was much smaller compared to Na + , Cl − or HCO 3 − , K + was often omitted in clinical calculation of AG. However, this research computed AG (mmol/L) using the following equation: AG = (Na + + K + ) -(Cl − + HCO 3 − ) [16].

Statistical analysis
All epidemiological analysis was completed by Statistical Product and Service Solutions (IBM, version 24), and EmpowerStats (www.empowerstats.com, X&Y Solutions, inc. Boston MA). The subjects were described as mean ± SD (measurement data), or quantity and percentage (counting data). (OR) and 95% con dence intervals (CI). Linear regression models were performed to explore the correlations between the level of AG and potential DR-related factors, using entering and stepwise methods to select variables, respectively. In addition, a smooth curve-tting adjusted by confounding factors was used to visualize the dose-response between DR and serum AG. All statistical analysis regarded non-DR of the control group or Q1 of AG quartiles as the references. P < 0.05 indicated that statistical differences were signi cant.   Table 2 showed the variations of each index based on AG quartiles. There were positive relationships between HbA1c, FPG, insulin, HOMA-IR, Ca 2+ , P and AG quartiles, and the negative ones were between Cl − , HCO 3 − and AG quartiles (all P < 0.001). In addition, the serum levels of Na + (P = 0.008) and Scr (P = 0.027)

Results
were also signi cantly correlated with AG quartiles. Moreover, the rising of AG quartiles was signi cantly associated with the proportion of DR (P = 0.001) and its severe cases (P < 0.001), respectively.  inversely U-shape with the aggravation of retinopathy. In addition, multivariate smoothing spline plots was used to investigate the correlations between serum AG and the cases of each severity of DR (exposure) (see Figure S2 in Additional le 3). With the aggravation of retinopathy, the dose-responses between serum AG and the exposure changed from a nearly straight liner to a curve (P = 0.0005 for anyseverity of DR; P = 0.0115 for mild NPDR; P = 0.0151 for moderate NPDR; P = 0.0205 for VTDR).
Furthermore, an optimal cut-off for AG level at 16.00 was identi ed as the best value for increasing the risk for presence of DR.   Abbreviation: AG, anion gap; DR, diabetic retinopathy; eGFR, estimated glomerular ltration rate; HbA1c, hemoglobin A1c; HOMA-IR, the homeostasis model assessment of insulin resistance index; Mg 2+ , magnesium; PP, pulse pressure; SBP, systolic blood pressure. a Using stepwise method to select variables. Stepwise method ordinally included the most signi cant variable (P < 0.05) and then excluded the most indistinctive variable (P > 0.10), which was aimed to obtain an optimal combine of variables based on Forward method and Backward method.

Discussion
We conducted a large sample of nested case-control study to explore the relationship between DR and serum AG, in which all participants were hospitalized patients with comprehensive data. A survey exceeding 6,000 sets of inpatients' subjects showed a 37% increase in their levels of serum AG, which meant that hospitalized patients, more comorbidities compare to general public probably, had a relatively raised serum AG in most cases [17]. In the setting of non-acidosis, the results of this study considered that the level of AG was associated with the presence and severity of DR. With the aggravation of DR severity, the level of AG remained rising but decreased in VTDR level. Multi-factor logistic regression models showed that AG quartiles were independently linked with higher odds to the incidence and severity of DR, but with premise of the uctuation on Q3 of AG quartiles. The adjusted smooth curvetting visualized that serum AG formed a nearly positive linear correlation with the severity of DR. Linear regression models were applied between the level of AG and DR-related factors in order to investigate the potential mechanism of this correlation. When using stepwise method to search the most optimal combination of variables, the growth of age, HbA1c and HOMA-IR played an intimate role in the increase of serum AG. Therefore, we surmised that serum AG might alter the risk of retinopathy by affecting glucose metabolism and insulin sensitivity in patients with type 2 diabetes. However, it should be noted that HOMA-IR was generally at a higher level in DR patients in this study, although the differences were not statistical.
Recent research suggested that participants with insulin resistance or type 2 diabetes were considered to relate to rising serum ketone bodies and declining urine PH [7,9], which were relevant indexes of metabolic acidosis. In order to controlled the complicated in uences of renal excretion and insulin status imposing on serum AG, this study excluded participants with the level of eGFR ≤ 60 ml/min or renal disfunction, and still obtained the statistical correlations of AG and DR. Further, a more signi cant association was developed after adding two variables as HOMA-IR and eGFR in model 3. They all meant that there were interlaced contacts among insulin resistance, acid load and renal function in DR or diabetic patients. Patients suffering from type 2 diabetes would develop a higher level of renal acid excretion and organic acid production [6]. Owing to the direct consequence of insulin resistance or longterm dysglycemia metabolism, incomplete renal function impeded ammonia production in proximal convoluted and potassium ion excretion in distal convoluted tubules [18][19][20][21]. Further, the functional defect of diabetic nephron was more prominent when suffering type 4 renal tubular acidosis, and the confusion of acid-base would be exacerbated [21]. But it is worthy to note that non-diabetic people with greater insulin resistance were also statistically accompanied with hypocapnia and low levels of citrate in the urine [8].
As the dysfunction of glycometabolism, diabetes causes multiple organ system diseases, and gradually forms a series of vascular or nonvascular complications. Insulin resistance is the main representation of type 2diabetes, although its pathogenesis of insulin has been widely concerned, it is not totally clear.
Previous studies have shown that in ammatory reactions, the disturbance of fatty acid metabolism and mitochondrion function are all implicated in insulin resistance [22]. Metabolic acidosis obstructs the binding of insulin to receptor [23,24], or decreases the sensitivity of tissue to insulin [25], which has been observed in human studies and rats models. Cell culture study revealed that excessive acid blocked receptor phosphorylation and insulin signal transduction [26]. Clinical tests of acid-base metabolism such as decreasing plasma HCO 3 − and increasing AG, have been proposed to induce insulin resistance in non-diabetic population, even these indicators were still at normal levels [6]. While in female patients with type 2 diabetes, the similar negative correlation between HCO 3 − and the risk of diabetes remained signi cant after controlling for C-reactive protein [3]. In addition, multiple studies have pointed that organic anion as one of the sources of endogenous acids, the level of its generation rate was signi cantly rising in patients with diabetes [27]. However, higher level of AG is also a result of the accumulation of endogenous acids. Population-based epidemiologic investigations among African American [4] and Swedish male [5] demonstrated a positive link between serum lactate (a component of AG) and the prevalence of type 2 diabetes, probably acting as the signal of impaired glucose tolerance and hyperinsulinism. The ndings in this study suggested that increasing serum AG was independently related to the occurrence and progression of DR. Therefore, the higher serum AG was not only one of the feedbacks of acid overload in clinic, its monitoring might also act as a potential warning signal for type 2 diabetes and diabetic complications including DR.
Moreover, heavier acid loading would promote the rising of blood pressure [27,28], as well as stimulate cortisol [29][30][31] and growth hormone [32] performing insulin anti-regulatory, then further triggered in ammatory response and protein consumption. Insulin resistance added the risks of type 2 diabetes and its complications, as well as poor blood pressure control and in ammation were also important mechanisms of DR [1,33]. The high level of AG probably affected the trigger and progression of retinopathy through various channels. In this study, the level of AG increased with the worsen of DR levels. According to the adjusted smoothing spline plots, the ndings suggested that serum AG should be controlled at a low-normal level to prevent diabetic patients from developing DR. Interestingly, among diabetic patients with serum AG levels between 16.00 to 18.00 may have a high risk for presence of moderate-DR and VTDR, and further prospective clinical studies are needed to validate our ndings.
This study was one of the series projects of DR Sub-study of LD-MCS, which rstly investigated the association between serum AG, and presence as well as severity of DR in the setting of non-acidosis.
However, the following limitations need to be mentioned: First, the proportion of mild NPDR, moderate NPDR and VTDR in this study was 67.9% (n = 772), 10.9% (n = 124), and 21.2% (n = 241) respectively. A non-linear uctuation between AG and DR levels might be induced by the inclusion of less moderate NPDR or more VTDR. Second, the research results stemmed from cross-sectional subjects, since the basic characteristics of participants were recorded only once, it was still unclear how the level of AG changes the presence of DR and its severity over time, as well as the causal link of this relativity. Third, the serum AG is de ned as the concentration of major cation (Na + ) minus the major anion (Cl − and HCO 3 − ), and its existence is mostly implicated the net anionic valence of plasma proteins (primarily albumin) [34]. The worries in this study when using AG was that the level of albumin had not been corrected. Fourth, we did not have records of blood PH or plasma ketone body, which limited the procedures of detailed reviewing participants' acid-base metabolism. But patients with acidosis have been excluded according to the historical diagnosis, and the average values of all ions and acid-base indexes were at normal ranges, so we believed that there was unlikely a highly bias in the results.
Similarly, the diet habits and nutrition of study participants was not traced, which played a partial role in acid-base physiology [35]. Further studies are still warranted to complement the above limitations.