Low serum total CO2 and its association with mortality in patients being followed up in the nephrology outpatients clinic

Large-scale studies have not been conducted to assess whether serum hypobicarbonatemia increases the risk for kidney function deterioration and mortality among East-Asians. We aimed to determine the association between serum total CO2 (TCO2) concentrations measured at the first outpatient visit and clinical outcomes. In this multicenter cohort study, a total of 42,231 adult nephrology outpatients from 2001 to 2016 were included. End-stage renal disease (ESRD) patients on dialysis within 3 months of the first visit were excluded. Instrumental variable (IV) was used to define regions based on the proportion of patients with serum TCO2 < 22 mEq/L. The crude mortality rate was 12.2% during a median 77.0-month follow-up period. The Cox-proportional hazard regression model adjusted for initial kidney function, alkali supplementation, and the use of diuretics demonstrated that low TCO2 concentration was not associated with progression to ESRD, but significantly increased the risk of death. The IV analysis also confirmed a significant association between initial TCO2 concentration and mortality (HR 0.56; 95% CI 0.49–0.64). This result was consistently significant regardless of the underlying renal function. In conclusion, low TCO2 levels are significantly associated with mortality but not with progression to ESRD in patients with ambulatory care.

www.nature.com/scientificreports/ Notably, it is difficult to evaluate acid-base status in an outpatient setting because of the difficulty associated with performing arterial blood gas analysis in this setting. In this study, we investigated the long-term clinical impact of hypobicarbonatemia on the development of incident ESRD and overall mortality in East-Asians, using the level of TCO 2 on venous sampling. We attempted to determine the effects of hypobicarbonatemia in a retrospective cohort from the patients with various regions of South Korea, using instrumental variable (IV) analysis.

Methods
Study subjects. This was a retrospective observational cohort study of patients tracked in the outpatient nephrology clinic of two tertiary hospitals in South Korea (Seoul National University Hospital and Seoul National University Boramae Hospital). Hospital administrative data were extracted, and the 10 th International Code of Disease Classification (ICD-10) was used. Data were collected using electronic health records following the methods also described in previous studies 16 . In detail, the date of birth, sex, age, body mass index (kg/m 2 ), and data related to other comorbidities (using the diagnostic code by ICD-10) and medication were recorded.
A total of 50,703 patients were examined from January 1, 2001, to December 31, 2016. Of the total 50,703 patients, we excluded 1,838 patients who had already been on dialysis with ESRD and 1,237 patients with acute kidney injury requiring renal replacement therapy who had been on dialysis within three months of hospital visit. Patients without record of serum creatinine levels or total CO 2 (TCO 2 ) (n = 5,298), patients with no address (n = 27), and those with extreme values of TCO 2 (n = 72) were also excluded. The final analysis was performed including 42,231 patients (Fig. 1).
Data collection. The Low TCO 2 group was defined as TCO 2 < 22 mEq/L. Among the drugs used, medications potentially affecting the status of TCO 2 level, data of patients who had a prescription of sodium bicarbonate, renin-angiotensin-aldosterone system (RAAS) blockades, including angiotensin-converting-enzyme inhibitors (ACE inhibitors, ACEi) and angiotensin II receptor blockers (ARBs), and diuretics were identified. Diuretics group included acetazolamide, spironolactone, amiloride, triamterene, furosemide, torasemide, thiazide, indapamide and metolazone.
We included only those patients who had undergone treatment for more than 30 days in the treatment group. The CKD grade was determined by calculating eGFR from creatinine level using Modification of Diet in Renal Clinical outcomes. The primary outcome was to clarify the all-cause mortality according to the TCO 2 level, and the secondary outcome was to find a new development of ESRD that required dialysis and progression to mortality from various causes. We obtained mortality information from the national statistics office 17 , and ESRD incidence was defined as follows: patients recently undergoing hemodialysis were extracted from the electronic medical records using the clinical data warehouse in our study centers. Patients who had started hemodialysis therapy were identified by the combination of operational definitions as follows: the occurrence of a new claim for the payment code of hemodialysis by screening ICD-10 codes, and vascular access operations; incident dialysis was also identified by the first date of when dialysis was suggested.
Statistical analysis. We used the chi-squared test for categorical variables and the Student's t-test for continuous variables to compare the demographics. Categorical variables were reported as the percentage of all patients, and continuous variables as the mean ± standard deviation. The multivariate Cox proportional hazards model was used to calculate the hazard ratio (HR) with a 95% CI for all-cause mortality and renal outcomes. Kaplan-Meier curves and the multivariate Cox proportional hazard models were used to compare the outcomes between the Low TCO 2 group and control groups. We included covariates, such as TCO 2 values, age, sex, history of hypertension, diabetes mellitus, MDRD-GFR, and SB usage, in the final model using multivariate Cox regression analysis.
To minimize the bias of this retrospective observational study, we conducted an IV analysis [18][19][20][21] . The IV in this study was regional classification of proportion of patients with TCO 2 < 22 mEq/L. We used the IV to independently associate the effects of low TCO 2 level and clinical outcomes. All statistical analyses were performed using R version 3.5.0 22 (Comprehensive R Archive Network: http://cran.r-proje ct.org) and SPSS version 22.0 (IBM Corp., Armonk, NY, USA). The maps in supplementary figures was generated that outlines of administrative districts were obtained from Statistical Geographic Information Service of Statistics Korea (URL, https :// sgis.kosta t.go.kr/jsp/engli sh/index .jsp). Maps were produced with R in conjunction with the software packages ggmap, maptools, rgdal, abd ggplot2 [23][24][25][26] . In all analyses, P < 0.05 was considered statistically significant.

Results
Baseline characteristics and demographics. The baseline characteristics and demographics of the study population divided according to the CKD stage are shown in Table 1. The mean age of the study subjects was 54 years, and 48.9% were male. Hypertension and diabetes were found in 29.2% and 20.9% of patients, respectively. The mean eGFR was 70.5 mL/min/1.73 m 2 and the TCO 2 was 26.2 mEq/L. CKD stage 2 accounted for 44.8% of all patients, and about 10% of patients were in advanced stage with CKD stage 4-5. As CKD progressed, a decrease in TCO 2 levels was observed, and the use of sodium bicarbonate, RAAS blockade, or diuretics was more frequent.
Meanwhile, the serum TCO 2 level was < 22 mEq/L in 10.1% of all patients, and we graphically depicted the proportion of patients with a TCO 2 < 22 mEq/L by the patient's residential location (Supplementary Information Fig. S1). When we divided patients into four groups based on the proportion of serum TCO 2 concentration < 22 mEq/L by residential area, we found that hypertension and smoking history were the highest in the 4 th group (Supplementary Information Table S1). The first group comprised patients who were older, had a higher prevalence of diabetes mellitus, and had the best baseline renal function. The fourth group had the highest rate of low TCO 2 level (P < 0.001).
Association between low TCO 2 and patient mortality. During the median 77.0-month followup period, the crude mortality rate was 12.2%. According to the Kaplan-Meier curve (Fig. 2), patients with TCO 2 < 22 mEq/L had a higher risk of mortality (HR 0.899; 95% CI 0.893-0.905; P < 0.001) than those in the other groups. The Cox-proportional hazard regression model adjusted for sex, age, hypertension, diabetes mellitus, and alkali supplementation showed that low TCO 2 concentration increased the risk for mortality (model 3, Table 2). After adding of initial eGFR levels for adjustment, the association between low TCO 2 concentration and mortality remained consistently significant (model 5). These results were consistently significant even when the use of RAAS blockades and diuretics were additionally adjusted (model 6). In the IV analysis, low TCO 2 concentration also increased the mortality risk. When adjusted for baseline renal function, alkali supplementation, RAAS blockade, and diuretics, the mortality risk reduced by approximately 44% with each 1 mEq/L increase in serum TCO 2 concentration (model 6; 95% CI 0.485-0.644; P < 0.001). Next, the TCO 2 level was set as a continuous variable and the correlation with the risk of death was visually shown (Fig. 3). As a result, it was confirmed that the lower the TCO 2 level, the higher the risk of death (Fig. 3A). Even after adjusting for several variables including age, sex, hypertension, diabetes, baseline kidney function, usage of sodium bicarbonate, RAAS blockades and diuretics, a consistent pattern was shown (Fig. 3B).
It was also verified that these results were consistent in Table 3, which was expressed as a categorical variable. Based on the serum TCO 2 concentration, patients were divided into three groups as follows: low, < 22 mEq/L; normal, 22-29 mEq/L; high, ≥ 30 mEq/L. Compared to the normal group, mortality risk was significantly higher in patients with serum TCO 2 concentration below 22 mEq/L; however, the risk decreased when TCO 2 concentrations were above 30 mEq/L. These results were significant even after adjusting for other variables.    www.nature.com/scientificreports/ In the subgroup analysis by initial eGFR, mortality risk increased with lower serum TCO 2 concentrations, regardless of initial GFR values (Table 4). These results were verified both in the Cox-proportional hazard regression model and on IV analysis. This also showed significant results when adjusting for underlying renal function, underlying disease, and medications taken, and the effect was more pronounced in patients with relatively good renal function with a baseline GFR of 60 mL/min/1.73 m 2 or higher.

Association between low TCO 2 and the risk of progression to ESRD. The risk of progression to
ESRD steadily increased over time and occurred in 8.5% of patients during the follow-up period. Patients with TCO 2 < 22 mEq/L showed higher risks of progression to ESRD (HR 0.844; 95% CI 0.838-0.850; P < 0.001) than the other groups in univariate analysis ( Supplementary Information Fig. S2, Table S2). However, the protective effect was the lowest in final model 6, which was adjusted for baseline renal function and medications. These results are consistent with the IV analysis. After adjusting for baseline renal function and medications taken, there was no association between serum TCO 2 concentration and progression to ESRD (model 6; HR 1.057, 95% CI 0.892-1.251, P = 0.5234).
The subgroup analysis of the relationship between the TCO 2 level and ESRD progression risk by TCO 2 level groups is also presented in Supplementary Information Fig. S3 and Table S3. In subgroup IV analysis by initial eGFR, serum TCO 2 concentration did not affect ESRD progression, regardless of eGFR (Supplementary Information Table S4).

Discussion
This study is the first population-scale study to show that low TCO 2 level affected clinical outcomes in an East-Asian population. The main objective of this study was to determine whether the single measurement of TCO 2 level affected the clinical outcomes in patients with ambulatory care in out-patients clinic, whose diet patterns were thought to be less acidic than that of their Western counterparts. According to the Korea National Health and Nutrition Examination Survey data 27,28 , regional differences were noted in meat intake between urban and rural areas. These differences in animal-based acidic diet pattern were likely related to the rate of dietary acid loading, causing eubicarbonatic H + retention without overt metabolic acidosis [13][14][15] . We used the fraction of low TCO 2 levels (< 22 mEq/L) by the residence area of participants as the IV to determine the independent relationship between TCO 2 level and clinical outcomes. To be selected as an IV, several important assumptions must be met 19,20 . First, the exposure values should have definite predictive associations with the causal inference for IV. This IV should not be directly related to the outcome to be predicted. The four groups in Supplementary   www.nature.com/scientificreports/ Information Table S1 did not differ significantly in terms of renal function by eGFR or other baseline characteristics, but only in terms of the proportion of patients with low TCO 2 level. Final Cox regression models and IV analysis including covariates of initial eGFR and information on the use of medications such as SB, RAAS blockade, and diuretics, showed that higher TCO 2 concentrations were protective in patients at risk of all-cause mortality, but not for preventing progression to ESRD (Tables 2 and  Supplementary Information Table S2). Serum bicarbonate concentrations were not very low in most of the patients in this study; only 10% of the cohort (4,242/42,331) had a serum bicarbonate level of < 22 mEq/L. Our results are consistent with those of previous studies which showed that long-term prognosis improved in patients with preserved renal function when serum bicarbonate levels were at the upper limit of the normal range 6 . Notably, in this study, the authors noted that hypobicarbonatemia (identified by low TCO 2 levels) may not indicate metabolic acidosis, and this is a possible limitation of the study. The identification of acid-base status based on TCO 2 as bicarbonate level may have led to misclassification because low TCO 2 might be due to chronic respiratory or mixed acid-base disorders. These misclassifications could be seen in patients with ESRD receiving hemodialysis 29 and in patients with acute renal failure requiring continuous renal replacement therapy 30 . However, in patients with preserved renal function in an outpatient setting, TCO 2 is considered effective for predicting serum bicarbonate level 31 . Thus, we believe that the low TCO 2 group in this study mainly reflects hypobicarbonatemia, irrespective of pH status. A recent study by Hirai et al. showed that serum TCO 2 correlated with bicarbonate concentration in Japanese patients with pre-dialysis CKD (sensitivity 91.7%, specificity 73.4%) 32 , supporting our results.
In terms of ESRD development, there was significance in model 3 without eGFR as a covariate; however, the protective effect of high TCO 2 disappeared when baseline eGFR was considered (Supplementary Information Table S2, model 4 to 6). Recent studies on early CKD (CKD2-3a) patients reveal that H + retention is first seen, even if low TCO 2 is not present yet as eubicarbonatemia. This H + retention itself affects CKD progression, which can be diagnosed by hypocitrauria 13 and urine ammonium 15 . In addition, low TCO 2 itself is a highly independent risk factor, which is more pronounced in early CKD and above (Table S3), and in advanced CKD (GFR 30 mL/min/1.73 m 2 and below), the influence on the ESRD of various factors other than TCO 2 level itself is much greater (e.g., vascular calcification, cardiovascular disease, and commodities). In the results of our study, the final Cox model and IV analysis were inconclusive in terms of ESRD incidence as there were limitations due to the retrospective study design, and heterogeneity of the operational definition of ESRD was also limitation. Another possible rationale for hypobicarbonatemia being associated only with mortality but not with ESRD incidence in our study involves systemic compensation capacity. A recent study in patients with CKD found that hypobicarbonatemia was involved in CKD progression only when pH reached acidemia (pH < 7.32) 33 . The acid-base status (using gas analysis of blood pH) on the clinical outcome of hypobicarbonatemia was not considered in existing studies 4,6,11 . According to researchers at the University of Osaka, in a study of 1,058 patients with CKD (eGFR < 60 mL/min), TCO 2 level and pH were measured simultaneously through venous blood gas analysis (VBGA). This study showed that the strength of this association is significantly changed by acidemia. Hypobicarbonatemia is associated with an increased risk of ESRD only when pH is low (pH < 7.32) but not when pH is normal. This association suggests that the compensatory reservoir might be important in terms of CKD prognosis 15 . In our study, patients with CKD 1,2 (eGFR > 60 mL/min) were 67% of the total. These patients are likely to have a normal pH. Results from the Osaka researchers also found that 40% of the group with bicarbonate levels under 21.5 mEq/L did not present acidemia as a result of systemic compensation 33 . In addition, in a prospective cohort study of an elderly population without CKD, similar to our results, 60% of the group with low bicarbonate levels did not have acidemia 34 . This study showed a correlation between low bicarbonatemia levels and high mortality, regardless of pH, and the modification of the effects by pH did not affect this association, which supports our findings. These findings suggest that combining information on pH with bicarbonate levels, especially to improve renal outcome, helps identify patients at high risk of ESRD. In addition, needless alkali treatments could be reduced in non-acidic hypobicarbonatemic patients 15,33 . Further studies may identify the need for clinicians to further improve non-acidicidc hypobicarbonatemic or eubicarbonatic H + retension in early CKD patients for slowing CKD progression. Treatment and consequent improvement of acidosis in CKD patients are expected to improve the clinical outcomes; however, the optimal interventional methods and population remain inconclusive, and The cautions is needen for the replacement of alkali using SB which is associated with complications, such as deterioration of sodium retention, edema, and heart failure aggravation.
In conclusion, low TCO 2 levels were significantly associated with mortality, but not with progression to ESRD in IV analysis using regional classification by proportion of low TCO 2 as an IV. Further studies, for a prospective interventional study on a nation-wide scale for the effects of alkali therapy on CKD progression in patients with eubicarbonatemia with early CKD and explanations for possible biologic mechanisms, in Korean CKD patients are needed.

Data availability
All data associated with this study are available in the main text or the supplementary materials following reasonable request. www.nature.com/scientificreports/ Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.