Abstract
Rationale & Objective
Metabolic acidosis is a risk factor for progression of chronic kidney disease (CKD), but little is known about its effect on health care costs and resource utilization. We describe the associations between metabolic acidosis, adverse kidney outcomes, and health care costs in patients with CKD stages G3-G5 and not receiving dialysis.
Study Design
Retrospective cohort study.
Setting & Participants
An integrated claims-clinical data set of US patients with CKD stages G3-G5, with serum bicarbonate values of 12 to <22 mEq/L (metabolic acidosis group) or 22 to 29 mEq/L (normal serum bicarbonate level group).
Predictor
The primary exposure variable was the baseline serum bicarbonate level.
Outcomes
The primary clinical outcome was the composite of all-cause mortality, maintenance dialysis, kidney transplant, or a decline in the estimated glomerular filtration rate of ≥40% (DD40). The primary cost outcome was all-cause predicted per-patient per-year cost, assessed over a 2-year outcome period.
Analytical Approach
Logistic and generalized linear regression models, adjusted for key covariates such as age, sex, race, kidney function, comorbidities, and pharmacy insurance coverage, were used to assess serum bicarbonate levels as a predictor of DD40 and health care costs, respectively.
Results
51,558 patients qualified. The metabolic acidosis group experienced higher rates of DD40 (48.3% vs. 16.7%, P < 0.001) and higher all-cause yearly costs ($65,172 vs. $24,681, P < 0.001). Two-year adjusted odds ratio of DD40 per 1-mEq/L increase in serum bicarbonate levels was 0.873 (95% CI, 0.866-0.879); the parameter estimate (±SE) for costs was −0.070 ± 0.0075 (P < 0.001).
Limitations
Possible residual confounding.
Conclusions
Patients with CKD and metabolic acidosis had higher costs and rates of adverse kidney outcomes compared with patients with normal serum bicarbonate levels. Each 1-mEq/L increase in serum bicarbonate levels was associated with a 13% decrease in 2-year DD40 events and a 7% decrease in per-patient per-year cost.
Index Words: Chronic kidney disease, metabolic acidosis, serum bicarbonate level, health care costs, observational
Plain Language Summary.
Metabolic acidosis is a risk factor for progression of chronic kidney disease (CKD), but little is known about its effect on health care costs and resource utilization. The serum bicarbonate level is a common laboratory test used to identify metabolic acidosis (<22 mEq/L). We used logistic and linear regression in a large cohort of US patients with advanced CKD to examine the associations of serum bicarbonate levels with adverse health outcomes and health care costs and resource utilization, adjusted for other patient characteristics. We found that patients with CKD and metabolic acidosis had higher costs and more frequent adverse kidney outcomes than patients with normal serum bicarbonate levels. The management of metabolic acidosis in patients with CKD could help reduce the system costs of CKD in the United States.
Chronic kidney disease (CKD) is a major public health problem: individuals with CKD account for 14% of Medicare fee-for-service beneficiaries aged 65 or older and generate 25% of the total expenditures.1 Metabolic acidosis is a common complication of advanced CKD and is associated with higher mortality and several adverse clinical outcomes that are associated with increased costs, including CKD progression, cardiovascular events, and muscle wasting.2, 3, 4, 5, 6
All-cause costs in patients with CKD increase dramatically as the disease progresses. Among Medicare fee-for-service beneficiaries aged 65 years and older, per-patient per-year (PPPY) nondialysis medical costs for patients with CKD stages G4-G5 were 55% higher than those with CKD stages G1-G2 ($30,641 vs $19,799, in 2017 USD, respectively).1 Similarly, a cross-sectional study drawn from claims of Medicare and commercial insurance plans (exclusive of dialysis costs) found costs were much higher for patients with CKD stages G4-G5 than those for patients with CKD stage G2 ($76,969 vs $16,770, respectively, for patients younger than 65 years; $46,128 vs $14,493, respectively, for patients aged 65 years and older).7 Accordingly, there are important economic and clinical reasons to consider complications and other risk factors that might modify the progression of CKD or drive differences in cost.
Despite reported associations between metabolic acidosis and CKD progression and between CKD progression and higher costs, there is scant real-world information about the effect of metabolic acidosis on the costs associated with CKD. We used electronic health records (EHRs) and medical claims data from a large community cohort of patients with nondialysis-dependent stage G3-G5 CKD to examine the association between metabolic acidosis, mortality, adverse kidney outcomes, and all-cause costs over time.
Methods
Study Design and Data Sources
We conducted an observational, retrospective cohort study of US patients with advanced nondialysis-dependent CKD. De-identified patient data from January 2007, through March 2017, were extracted from Optum’s Integrated Claims-Clinical data set. The Optum database is a longitudinal repository of EHR data that included >81 million patients, as of 2017 from all insurance types/statuses, and a subset of patients linked by unique patient identifiers to a health care claims database of commercial and Medicare Advantage plans.8 Data extracted from inpatient and outpatient EHRs and administrative systems included laboratory results, prescribed medications, coded diagnoses and procedures, and provider notes extracted by natural language processing. Data cleaning is described in Table S1.
Study Cohort
Patients included in the database extract had at least 1 year of EHR activity with at least 3 estimated glomerular filtration rate (eGFR) results of <60 mL/min/1.73 m2 and at least 3 serum bicarbonate level results, with at least 1 result between 12 and 29 mEq/L. To qualify for the study cohort, patients were required to have 2 consecutive serum bicarbonate level results 28 to 365 days apart, both in the range of 12 to <22 mEq/L (metabolic acidosis group) or 22 to 29 mEq/L (normal serum bicarbonate level group). The first value in each pair established the baseline serum bicarbonate level, and its test date established the index date. Inclusion also required a baseline eGFR value between >10 and <60 mL/min/1.73 m2, calculated as the mean of eGFR values for 90 days preceding the last eGFR test ≤ the index date. Serum bicarbonate levels and eGFR values collected during hospital inpatient admissions or emergency department visits with diagnosed acute kidney injury were excluded as acute events. Inclusion also required EHR activity for ≥ 1 year pre-index plus ≥ 2 years postindex dates unless the patient died within this 2-year period. Patients with pre-index evidence of maintenance dialysis or kidney transplantation (by diagnosis or procedural code or outpatient eGFR of ≤10 mL/min/1.73 m2) were excluded. Patients were classified according to the baseline CKD stage for subgroup analyses. This methodology has been reported elsewhere.5
A cost analysis cohort was created for patients with linked medical claims concurrent with EHR data. Patient selection for both cohorts is depicted in Fig S1, which also details minor differences in criteria to ensure a sufficient sample size in the cost analysis cohort, and in Tables S2 and S3.
To ensure a sufficient representation at low serum bicarbonate levels, an iterative selection algorithm was used to oversample patients with metabolic acidosis. This methodology is further explained in a previous publication, along with sensitivity analyses showing that primary clinical findings are sustained in the absence of oversampling.5
Variables
The primary exposure variable was the baseline serum bicarbonate level. Demographic and clinical covariates known or hypothesized to be associated with CKD progression and higher costs2, 3, 4, 5, 6 were assessed in the EHR data as of the index date and included age, sex, race, diabetes, hypertension, heart failure, comorbidity burden measured by weighted Charlson Comorbidity Index,9 baseline eGFR, and urine albumin-to-creatinine ratio (ACR).
Health care cost in the data extract was provided from the payer perspective (ie, health plan payments rather than provider billed charges) as Optum-defined “standard cost.” an approximated cost per service or prescription based on private insurance payment levels in 2015 US dollars. This methodology reduces the variation in payment amounts to similar providers for the same service.
Missing data were not imputed. In statistical models on clinical outcomes, exclusions were applied to patients with missing ACR data (n=3,781 [22%] patients with metabolic acidosis and n=15,767 [46%] patients with normal serum bicarbonate levels in the overall study cohort; n=1,267 [41%] in the cost analysis cohort). Exclusions were also applied to patients with missing cost data (n=49 [2%]).
A complete list of variable definitions, including date and validity parameters, data sources, and conversions, is provided in Table S1.
Outcomes
The primary clinical outcome was the composite of all-cause mortality, maintenance dialysis, kidney transplant, or a decline in eGFR of ≥40% (DD40). Death was identified by linkage to Social Security data before data de-identification. The initiation of maintenance dialysis was identified by a diagnostic or procedural code, or a nonemergency outpatient eGFR test result of ≤9 mL/min/1.73 m2. This definition (Table S1) was established by an internal validation study of patients with concurrent data from the EHR and insurance claims.5 Kidney transplant was identified by a diagnostic code, procedural code, or inpatient diagnosis-related group. A decline in the eGFR was assessed by comparing the baseline eGFR with the mean of eGFR values during 90 days before the last eGFR result, before the initiation of maintenance dialysis or kidney transplant or at end of the 2-year outcome period.
The primary cost outcome was the all-cause PPPY cost, assessed over the 2-year outcome period. For descriptive analyses in the full study cohort (EHR data), a patient-level predicted cost was established using general linear regression models in the cost analysis cohort. Models predicted log costs per month in patients with stage G3 and stages G4-G5 CKD, adjusted for the age group (<65 years or ≥65 years), sex, diabetes, hypertension, pharmacy coverage, baseline metabolic acidosis, whether the patient experienced a DD40 outcome, and the interaction between metabolic acidosis and DD40 outcome. To address typical skew in health care costs, 2-year cost data were adjusted with a 98% winsorization.10 The predicted all-cause cost per month was calculated from these models and applied to each patient in the EHR study population based on the patient’s characteristics and months of EHR activity (24 months or until death). Model details are shown in Item S1 and Table S4. The predicted all-cause PPPY cost was calculated for each patient as cost per month × number of months/2 years and was averaged between the metabolic acidosis and normal serum bicarbonate level groups for comparison in unadjusted analyses. For adjusted analyses, the actual PPPY cost was assessed for the patients in the cost analysis cohort.
Secondary outcomes included utilization endpoints, consisting of the frequency of inpatient admissions and emergency department visits and the percentage of patients having such services at least once, evaluated using EHR data in the full study cohort.
Outcomes were evaluated over the outcome period of 2 years beginning on the index date. Costs were reported in 2018 USD, normalized at 3% per annum.
Statistical Analysis
Characteristics of the study population were compared between the metabolic acidosis and normal serum bicarbonate level groups using the χ2 test, t test, or Wilcoxon rank sum test as appropriate. The unadjusted rates of DD40 events and mean all-cause predicted PPPY costs by CKD stage and in total, and utilization outcomes in total were compared between the 2 groups over the 2-year outcome period using the χ2 test, t test, and Mood median test, respectively.
The adjusted effect of serum bicarbonate levels on PPPY costs (primary cost outcome) was assessed in the cost analysis cohort using a generalized linear regression model on actual log all-cause PPPY costs, adjusted for demographics, eGFR, log ACR, comorbidities, and pharmacy insurance coverage. The model was weighted so that patients aged 65 years or older and younger than 65 years contributed 60% and 40% to the results, respectively, to ensure that the results were representative of the age distribution of the patient population with CKD in the United States.11,12 Logistic regression models examined the percentage of patients with ≥1 inpatient admission and the percentage of patients with ≥1 emergency department visit using the same covariates.
Sensitivity analyses using logistic regression were performed to assess the effect of minimizing survival bias by establishing the baseline serum bicarbonate level on the second of 2 confirming serum bicarbonate level tests required for study inclusion and without ACR as a covariate to assess the effect of missing data. Additional analyses examined the interaction of serum bicarbonate levels with DD40 outcome on log all-cause cost PPPY, characterizing metabolic acidosis as a dichotomous variable on DD40, and in an outpatient-testing cohort that excluded patients for which 1 or both serum bicarbonate level tests establishing index serum bicarbonate level and metabolic acidosis or normal serum bicarbonate level status was performed during inpatient or hospital emergency department care.
All statistical analyses were performed using SAS/STAT software, version 9.2. P values of <0.05 were considered statistically significant.
Results
A total of 319,126 patients met the specified criteria for inclusion in the database extract, and 51,558 patients met inclusion criteria for the study cohort: 17,350 patients were classified as having baseline metabolic acidosis and 34,208 patients were classified as having normal serum bicarbonate level at baseline (Table 1). Compared with individuals with normal serum bicarbonate levels at the baseline, patients with baseline metabolic acidosis (metabolic acidosis group) were younger (mean age, 70.3 vs 74.3 years), more often African American (14.9% vs 7.4%), had more advanced CKD (baseline eGFR, 37.2 vs 43.2 mL/min/1.73 m2), and had a higher comorbidity burden (cardiovascular comorbidities, proteinuria by ACR, and Charlson Comorbidity Index score) (all differences, P < 0.001). The mean (±standard deviation) follow-up during the 2-year outcome period was 21.8 ± 5.5 months (19.8 ±7 .2 months in the metabolic acidosis group and 22.8 ± 4.0 months in the normal serum bicarbonate level group, because of differences in mortality) (Table 1). The characteristics of the cost analysis cohort are profiled in Table S3.
Table 1.
Baseline Characteristics
| Characteristic | Total Study Cohort N=51,558 |
Metabolic Acidosis Group N=17,350 |
Normal Serum Bicarbonate Group N=34,208 |
P |
|---|---|---|---|---|
| Sex, n (%) | ||||
| Female | 27,094 (53) | 9,011 (52) | 18,083 (53) | 0.05 |
| Male | 24,464 (47) | 8,339 (48) | 16,125 (47) | 0.05 |
| Age, y, mean ± SD | 72.9 ± 11.5 | 70.3 ± 13.3 | 74.3 ± 10.3 | <0.001 |
| Race, n (%) | ||||
| African American | 5,128 (10) | 2,585 (15) | 2,543 (7) | <0.001 |
| Asian | 996 (2) | 398 (2) | 598 (2) | <0.001 |
| White | 42,055 (82) | 12,866 (74) | 29,189 (85) | <0.001 |
| Other/unknown | 3,379 (7) | 1,501 (9) | 1,878 (5) | <0.001 |
| Baseline laboratory values, mean ± SD | ||||
| Serum bicarbonate, mEq/L | 24.0 ± 3.6 | 19.7 ± 1.1 | 26.1 ± 2.0 | <0.001 |
| eGFR, mL/min/1.73 m2 | 41.2 ± 12.1 | 37.2 ± 13.3 | 43.2 ± 10.9 | <0.001 |
| ACR, urinary, mg/g | 190 ± 554 | 277 ± 692 | 127 ± 414 | <0.001 |
| CKD stage, n (%) | ||||
| Stage G3a | 22,431 (44) | 5,719 (33) | 16,712 (49) | <0.001 |
| Stage G3b | 19,081 (37) | 5,987 (35) | 13,094 (38) | <0.001 |
| Stage G4 | 8,736 (17) | 4,747 (27) | 3,989 (12) | <0.001 |
| Nondialysis stage G5 | 1,310 (3) | 897 (5) | 413 (1) | <0.001 |
| Comorbidities/conditions, n (%) | ||||
| Hypertension | 31,761 (62) | 12,879 (74) | 18,882 (55) | <0.001 |
| Diabetes | 16,168 (31) | 7,391 (43) | 8,777 (26) | <0.001 |
| Coronary artery disease | 14,329 (28) | 6,249 (36) | 8,080 (24) | <0.001 |
| Peripheral vascular disease | 10,052 (19) | 5,038 (29) | 5,014 (15) | <0.001 |
| Heart failure | 10,029 (19) | 5,119 (30) | 4,910 (14) | <0.001 |
| Charlson Comorbidity Index, weighted, mean ± SD | 2.3 ± 2.7 | 3.5 ± 3.1 | 1.7 ± 2.3 | <0.001 |
| Additional baseline laboratory values, mean ± SD | ||||
| Serum albumin, g/dL | 3.7 ± 0.6 | 3.5 ± 0.7 | 3.9 ± 0.5 | <0.001 |
| Serum calcium, corrected, mg/dL | 9.3 ± 0.6 | 9.3 ± 0.7 | 9.4 ± 0.5 | <0.001 |
| Hemoglobin, g/dL | 12.2 ± 2.0 | 11.3 ± 2.1 | 12.6 ± 1.8 | <0.001 |
| Serum potassium, mEq/L | 4.4 ± 0.6 | 4.5 ± 0.7 | 4.4 ± 0.5 | <0.001 |
Note: 32,007 patients (13,569 in the metabolic acidosis group and 18,438 in the normal serum bicarbonate group) had values for ACR and were included in the statistical model of clinical outcomes. Counts of patients contributing laboratory values are presented in Table S2. P values: difference between the metabolic acidosis and normal serum bicarbonate groups.
Abbreviations: ACR, albumin-creatinine ratio; CKD, chronic kidney disease; SD, standard deviation.
The DD40 outcome was experienced by 27.3% of the study cohort within the 2-year outcome period. DD40 outcomes occurred in a significantly higher percentage of patients with metabolic acidosis (48.3%) than those in patients with normal serum bicarbonate levels (16.7%) in the total study population (all CKD stages combined) and for each CKD stage (Table 2). Within the individual components of the composite DD40 end point, patients with metabolic acidosis fared worse compared with patients with normal serum bicarbonate level at the baseline (death: 30.9% vs 10.2%, P < 0.001; kidney replacement therapy: 19.6% vs 5.5%, P < 0.001; ≥40% eGFR decline: 9.7% vs 3.6%, P < 0.001, respectively).
Table 2.
Comparison of DD40, Predicted Costs, and Health Care Utilization: Baseline Metabolic Acidosis vs Normal Serum Bicarbonate Level (Unadjusted)
| Outcome | Metabolic Acidosis Group N=17,350 |
Normal Serum Bicarbonate Group N=34,208 |
P |
|---|---|---|---|
| DD40, N with outcome (%) | |||
| Total study population | 8,377/17,350 (48) | 5,698/34,208 (17) | <0.001 |
| CKD stage G3a | 2,204/5,719 (39) | 1,970/16,712 (12) | <0.001 |
| CKD stage G3b | 2,595/5,987 (43) | 2,061/13,094 (16) | <0.001 |
| CKD stage G4 | 2,799/4,747 (59) | 1,327/3,989 (33) | <0.001 |
| CKD stage G5 | 779/897 (87) | 340/413 (82) | 0.03 |
| Predicted cost PPPY, mean (SD) | |||
| Total study population | $65,172 (43,353) | $24,681 (20,384) | <0.001 |
| CKD stage G3a | $58,830 (36,252) | $21,810 (14,119) | <0.001 |
| CKD stage G3b | $63,588 (41,664) | $22,615 (15,497) | <0.001 |
| CKD stage G4 | $68,838 (48,922) | $37,898 (35,380) | <0.001 |
| CKD stage G5 | $96,778 (48,851) | $78,691 (39,719) | <0.001 |
| Utilization PPPY, in the total study population, median (IQR) | |||
| IP admissions PPPY | 1.0 (0.5-2.0) | 0.5 (0.0-1.0) | <0.001 |
| ED visit utilization PPPY | 0.5 (0.0-1.5) | 0.0 (0.0-0.5) | <0.001 |
| Hospital OP encounters PPPY | 9.5 (2.5-25.5) | 4.5 (0.5-12.5) | <0.001 |
| Office/clinic encounters PPPY | 8.0(2.5-19.5) | 7.0 (2.5-15.5) | <0.001 |
| Other OP encounters PPPY | 7.5 (0.5-22.0) | 2.0 (0.0-11.0) | <0.001 |
Note: The denominators for each outcome by the CKD stage represent the total number of patients within each CKD stage in the metabolic acidosis or normal serum bicarbonate group at baseline. P values: difference in outcome between the metabolic acidosis and normal serum bicarbonate groups.
Abbreviations and definitions: CKD, chronic kidney disease; DD40, composite end point of death, dialysis/kidney transplant, or ≥ 40% decline in eGFR; SD, standard deviation; ED, emergency department; IQR, intraquartile range; IP, inpatient; OP, outpatient; PPPY, per-patient per-year; SD, standard deviation.
The mean all-cause unadjusted predicted PPPY costs in the cohort were $65,172 in the metabolic acidosis group and $24,681 in the normal serum bicarbonate level group (P < 0.001) (Table 2 and Fig 1). PPPY costs were also significantly higher among patients with metabolic acidosis than those among patients with normal serum bicarbonate level in each CKD stage (all P < 0.001), with the greatest differences observed among patients with CKD stage G3a (approximately 2.5-fold higher in the metabolic acidosis group) and stage G3b (approximately 3-fold higher in the metabolic acidosis group).
Figure 1.
Predicted all-cause PPPY costs (unadjusted). Abbreviations: CKD, chronic kidney disease; PPPY, per-patient per-year. ∗P < 0.001 (metabolic acidosis vs normal serum bicarbonate).
Patients with a DD40 outcome had higher predicted all-cause PPPY costs compared with those who did not experience these outcomes. Baseline metabolic acidosis was associated with higher predicted all-cause PPPY costs regardless of whether a DD40 outcome occurred. These findings were consistent within each CKD stage. Having both baseline metabolic acidosis and a DD40 outcome were associated with the highest PPPY cost at each CKD stage (Fig 2).
Figure 2.
Predicted all-cause PPPY costs for patients with CKD or without metabolic acidosis, by DD40 status (unadjusted). Abbreviations and definitions: CKD, chronic kidney disease; DD40, death, dialysis, kidney transplantation, or a ≥40% decline in eGFR; PPPY, per-patient per-year. ∗P < 0.001 (metabolic acidosis vs normal serum bicarbonate).
The serum bicarbonate level had a strong independent association with DD40 outcomes after controlling for covariates, with an odds ratio (OR) of 0.873 (95% CI, 0.866-0.879) (Table 3). The serum bicarbonate level also had a strong independent association with PPPY costs in generalized linear regression modeling, after controlling for covariates. Each 1-mEq/L increase in serum bicarbonate levels was associated with a 7% decrease in all-cause PPPY health care costs (parameter estimate ± standard error for log PPPY costs: −0.070 ± 0.0075; P < 0.001), equivalent to $2,700 PPPY per 1-mEq/L increase when evaluated at the study cohort means for costs and serum bicarbonate levels. Higher age and hypertension were also associated with lower costs, whereas the African American race and a Charlson Comorbidity Index of ≥3 were associated with higher costs (Table 3).
Table 3.
Effect of Serum Bicarbonate Levels on 2-Year Adverse Kidney Outcomes/Mortality and on Log Cost PPPY, Adjusted for Covariates
| Regression Parameter | Logistic Regression Model on DD40 Within 2 Years |
Generalized Linear Model on Log Cost PPPY Assessed Over 2 Years |
||
|---|---|---|---|---|
| OR (95% CI) | P | Parameter Estimates ± SE | P | |
| Age, per 1-y increase | 1.000 (0.998-1.002) | 0.8460 | −0.013 ± 0.0025 | <0.001 |
| Male (vs female) | 1.197 (1.135-1.263) | <0.0001 | −0.150 ± 0.0628 | 0.02 |
| Race (vs White) | ||||
| African American | 1.436 (1.321-1.560) | <0.0001 | 0.302 ± 0.1100 | 0.006 |
| Asian | 0.845 (0.704-1.013) | 0.0683 | −0.690 ± 0.2368 | 0.004 |
| Other/unknown race | 1.094 (0.989-1.210) | 0.0794 | 0.356 ± 0.1494 | 0.02 |
| Baseline laboratory values, continuous | ||||
| eGFR, per 1-mL/min/1.73 m2 increase | 0.967 (0.965-0.969) | <0.0001 | −0.003 ± 0.0029 | 0.25 |
| Serum bicarbonate, per 1-mEq/L increase | 0.873 (0.866-0.879) | <0.0001 | −0.070 ± 0.0075 | <0.001 |
| ACR, log, per 1-unit increase | 1.209 (1.190-1.229) | <0.0001 | 0.008 ± 0.0194 | 0.67 |
| Baseline comorbidities (yes vs no) | ||||
| Diabetes | 0.924 (0.868-0.983) | 0.0128 | 0.044 ± 0.0723 | 0.54 |
| Hypertension | 0.938 (0.871-1.011) | 0.0938 | −0.234 ± 0.0874 | 0.007 |
| Heart failure | 1.828 (1.711-1.953) | <0.0001 | 0.432 ± 0.0793 | <0.001 |
| Charlson Comorbidity Index score (vs 0) | ||||
| 1 | 1.157 (1.045-1.281) | 0.0049 | 0.173 ± 0.1145 | 0.13 |
| 2 | 1.212 (1.100-1.334) | <0.0001 | −0.026 ± 0.1150 | 0.82 |
| ≥3 | 1.739 (1.597-1.895) | <0.0001 | 0.524 ± 0.1032 | <0.001 |
| Other | ||||
| Pharmacy insurance coverage (yes vs no) | NA | 0.503 ± 0.0653 | <0.001 | |
Note: Covariates are age, sex, race, serum bicarbonate, eGFR, log ACR, diabetes, heart failure, hypertension, Charlson Comorbidity Index score, and pharmacy insurance coverage. P values: difference between the metabolic acidosis and normal serum bicarbonate groups.
Abbreviations and definitions: ACR, albumin-creatinine ratio (urinary); CI, confidence interval; DD40, death, dialysis, kidney transplantation, or a ≥40% decline in eGFR; eGFR, estimated glomerular filtration rate; GLM, generalized linear regression model; OR, odds ratio; PPPY, per-patient, per-year; SE, standard error.
Patients with CKD and metabolic acidosis used significantly more health care services (inpatient, emergency department, and outpatient services) compared with patients with CKD and normal serum bicarbonate levels (P < 0.001 for all comparisons) (Table 2). In generalized linear regression models, the serum bicarbonate level was independently associated with the frequency of inpatient admissions and emergency department visits after adjusting for potential confounders (age, sex, race, eGFR, ACR, diabetes, hypertension, heart failure, and Charlson Comorbidity Index). Parameter estimates per 1-mEq/L increase in serum bicarbonate levels were −0.082 (P < 0.001) for inpatient admissions PPPY and −0.039 (P < 0.001) for emergency department visits PPPY. Evaluated at the study cohort means, a 1-mEq/L increase in serum bicarbonate level was associated with reductions in projected PPPY utilization of 8.6% in inpatient admissions and 4.5% in emergency department visits (Tables S5-S6) or 8.2 inpatient admissions and 3.3 emergency department visits per 100 patient-years.
Sensitivity analysis evaluating the association of baseline serum bicarbonate level and DD40 using the second of 2 qualifying serum bicarbonate level values confirmed the study findings (Table S7). In addition, omitting ACR to avoid excluding patients for missing data made little change in the OR for serum bicarbonate level on DD40 (Table S8). Similarly, a reanalysis in an outpatient-testing cohort confirmed the study findings (adjusted OR per 1-mEq/L increase in the serum bicarbonate level on DD40, 0.876; 95% CI, 0.866-0.887). A sensitivity analysis assessing log all-cause PPPY cost adding DD40 as a covariate confirmed a significant independent association between higher serum bicarbonate levels and lower costs, even after adjustment for postindex kidney failure, CKD progression, and death (Table S9). Finally, a logistic model on the DD40 structuring serum bicarbonate level as a categorical variable showed that the odds of developing a DD40 outcome within 2 years were 3 times higher with metabolic acidosis than that with the normal serum bicarbonate level (OR, 2.998; 95% CI, 2.836-3.170). All sensitivity analyses and supplemental models were adjusted for the same covariates displayed in Table 3, except as noted.
Discussion
In this longitudinal cohort study of more than 50,000 community-based patients with nondialysis-dependent CKD, we found that higher baseline serum bicarbonate levels were independently associated with a lower risk of CKD progression, dialysis, and all-cause mortality and lower health care utilization and all-cause costs. After controlling for covariates, each 1-mEq/L increase in the baseline serum bicarbonate level was associated with a 13% reduction in the 2-year odds of DD40, a 7% reduction in PPPY all-cause costs and 8.6% and 4.5% reductions in PPPY admissions to inpatient facilities and emergency departments, respectively.
Previous studies have examined the association between declining kidney function and increased costs.1,7,13,14 In a study of health maintenance organization insurance enrollees that included 14,000 patients with CKD, Smith et al14 found that costs and utilization over a mean of 38-51 months were consistently higher among patients with stage G4 CKD compared those in patients with stage G2 or stage G3 CKD. Golestaneh et al7 studied 105,000 patients with stage G2-G5 CKD or end-stage kidney disease who were prescribed renin-angiotensin-aldosterone inhibitors and reported that mean annualized cost per-patient (excluding costs of kidney dialysis) increased with each CKD stage: at least 57% among patients aged younger than 65 years and 45%-68% among patients aged 65 years or older. Among patients with diabetes and CKD, Lage et al15 found that total medical costs were significantly higher for patients with an eGFR <30 mL/min/1.73 m2 compared with those with stage G1-G2 CKD, with each 1-mL/min/1.73 m2 reduction in eGFR associated with an increase of $1,845 in all-cause total medical costs. Our findings are consistent with these 3 studies and are the first to focus on metabolic acidosis, which may be a disease-modifying complication of CKD, as a contributor to these cost differences. Notably, each of the models in this study showed that the serum bicarbonate level to be a more powerful predictor of costs and health care utilization than the baseline eGFR. Furthermore, the association of serum bicarbonate levels with health care costs was also shown to be independent of age and comorbidities known to be associated with CKD and the overall comorbidity burden.
Our findings have important clinical and health policy implications. The association between metabolic acidosis and adverse kidney outcomes confirms findings from multiple, large observational studies and supports current recommendations to treat metabolic acidosis to delay CKD progression.16 The association of metabolic acidosis with increased costs and higher health care utilization is a novel finding and deserves further study. Clinical trials of treatments for metabolic acidosis should add outcomes such as inpatient visits and emergency department utilization, in addition to related economic analyses. The multiorgan adverse effects of metabolic acidosis (eg., on bone, muscle, kidney, and cardiovascular system) suggest that successful treatment may affect a range of outcomes and related costs that extend beyond the kidney system.
This study has several strengths. We used conservative definitions to increase the specificity of our outcome measures, such as measuring the eGFR decline by averaging results over a 90-day look back to avoid defining outcomes based on a single, potentially spurious value. In addition, we conservatively required 2 serum bicarbonate level tests for study inclusion, reducing the bias inherent in using a single test result as the baseline value. Finally, by including DD40 in the sensitivity analysis of costs, we were able to show that the effect of metabolic acidosis was not only independent of baseline kidney function but was also independent of kidney-related clinical outcomes.
This study also has several limitations. Because this was a retrospective and observational study, we are unable to exclude the existence of residual confounding, which was not adjusted for in our analyses. Our conservative requirement of 2 serum bicarbonate level tests for study inclusion introduced a small risk of survival bias, although a sensitivity analysis addressing this point sustained our findings. Although we excluded consideration of serum bicarbonate level tests collected during inpatient admissions in which acute kidney injury was reported by the diagnostic code, we recognize that diagnostic codes are not always reliably reported in EHRs; the analysis in an outpatient-testing cohort confirmed primary findings. The identification of dialysis initiation in the US EHR data are limited by the lack of data from the specialized providers of the most dialysis care. We addressed this limitation by defining dialysis based on internal validation that cross-referenced EHR data and medical insurance claims; however, this definition has not been externally validated. Owing to gaps in information about medication use in the database, we were unable to include the use of medications, including oral alkali, in the study. Finally, the need to develop a predictive model of costs to examine unadjusted cost differences introduces unknown sources of bias; however, the adjusted analyses were conducted among patients with claims data using actual rather than predicted costs and therefore were less subjected to possible confounding.
In conclusion, we found that in a large cohort of patients with nondialysis CKD stages G3-G5, each 1-mEq/L higher baseline serum bicarbonate level was independently related to reductions in the composite outcome DD40 (death, dialysis, or 40% reduction in eGFR), in all-cause health care costs, and inpatient and emergency department utilization.
Article Information
Authors’ Full Names and Academic Degrees
Nancy L. Reaven, MA, Susan E. Funk, MBA, Vandana Mathur, MD, and Navdeep Tangri, MD, PhD
Authors’ Contributions
Research idea and study design: NLR, SEF, VM, NT; data acquisition: NLR, SEF; data analysis/interpretation: NLR, SEF, VM, NT; statistical analysis: NLR, SEF; supervision or mentorship: N/A. Each author contributed important intellectual content during manuscript drafting or revision and accepts accountability for the overall work by ensuring that questions pertaining to the accuracy or integrity of any portion of the work are appropriately investigated and resolved.
Support
This study was funded by Tricida, Inc, which was involved in the decision to submit for publication.
Financial Disclosure
NLR, SEF, VM, and NT were paid consultants to Tricida, Inc. in connection with the development of this manuscript. NLR, VM, NT report consultancy, personal fees, and equity ownership from Tricida, Inc., related to the submitted work. VM and NT are members of the advisory boards at Tricida. VM is listed on patents related to work for Tricida. VM reports additional consulting fees from Tricida, Equillium, Myovant, Rigel, Corvidia, Acuta, Frazier, Intarcia, PTC Bio, Escient, Galderma, and Sanifit outside the submitted work. SEF report consultancy and personal fees from Tricia, Inc.
Acknowledgments
The authors would like to thank Jun Shao (Tricida Inc) and Kathryn Boorer for editorial assistance in the preparation of this manuscript.
Data Sharing
The data that support the findings of this study are available from Optum, but restrictions apply to the availability of these data, which were used under license for this study and so are not publicly available.
Peer Review
Received July 12, 2022. Evaluated by 2 external peer reviewers, with direct editorial input from the Statistical Editor, an Associate Editor, and the Editor-in-Chief. Accepted in revised form January 16, 2023.
Footnotes
Complete author and article information provided before references.
Figure S1: Patient selection flow chart.
Item S1: Cost prediction model methodology.
Table S1: Sources, Measurements, and Definitions.
Table S2: Patients Contributing Laboratory Values.
Table S3: Characteristics of the Cost Analysis Cohort.
Table S4: Cost Prediction GLM Regression on Log Cost Per Month: Regression Parameters.
Table S5: Contribution of a 1-mEq/L Increase in Serum Bicarbonate level to Mean Inpatient Utilization PPPY, Evaluated at Cohort Means in a Generalized Linear Model.
Table S6: Contribution of a 1-mEq/L Increase in Serum Bicarbonate level to Mean Utilization of Emergency Department Visits PPPY, Evaluated at Cohort Means in a Generalized Linear Model.
Table S7: Logistic regression on DD40 in the full study cohort (N=51558): sensitivity analysis replacing baseline serum bicarbonate level with the confirming consecutive serum bicarbonate level value 28-365 days later (ie, eliminating survival bias).
Table S8: Logistic regression on DD40 in the full study cohort (N=51,558): sensitivity analysis removing ACR (assessing effect of missing data).
Table S9: Sensitivity Analysis Using the DD40 Outcome as an Additional Predictor of Total Cost PPPY, Weighted to 60% Age ≥65 Years and 40% Age <65 Years.
Supplementary Material
Figure S1; Item S1; Table S1-S9
References
- 1.Saran R., Robinson B., Abbott K.C., et al. US Renal Data System. 2019 Annual Data Report: epidemiology of kidney disease in the United States. Am J Kidney Dis. 2020;75(1 suppl 1):S1–S64. doi: 10.1053/j.ajkd.2019.09.003. [DOI] [PubMed] [Google Scholar]
- 2.Chen W., Abramowitz M.K. Metabolic acidosis and the progression of chronic kidney disease. BMC Nephrol. 2014;15:55. doi: 10.1186/1471-2369-15-55. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Dobre M., Yang W., Chen J., et al. Association of serum bicarbonate with risk of renal and cardiovascular outcomes in CKD: A report from the Chronic Renal Insufficiency Cohort (CRIC) study. Am J Kidney Dis. 2013;62(4):670–678. doi: 10.1053/j.ajkd.2013.01.017. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Wang X.H., Mitch W.E. Mechanisms of muscle wasting in chronic kidney disease. Nat Rev Nephrol. 2014;10(9):504–516. doi: 10.1038/nrneph.2014.112. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Tangri N., Reaven N.L., Funk S.E., Ferguson T.W., Collister D., Mathur V. Metabolic acidosis is associated with increased risk of adverse kidney outcomes and mortality in patients with non-dialysis dependent chronic kidney disease: An observational cohort study. BMC Nephrol. 2021;22(1):185. doi: 10.1186/s12882-021-02385-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Collister D., Ferguson T.W., Funk S.E., Reaven N.L., Mathur V., Tangri N. Metabolic acidosis and cardiovascular disease in CKD. Kidney Med. 2021;3(5):753–761. doi: 10.1016/j.xkme.2021.04.011. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Golestaneh L., Alvarez P.J., Reaven N.L., et al. All-cause costs increase exponentially with increased chronic kidney disease stage. Am J Manag Care. 2017;23(10 Suppl):S163–S172. [PubMed] [Google Scholar]
- 8.Wallace P.J., Shah N.D., Dennen T., Bleicher P.A., Crown W.H. Optum Labs: building a novel node in the learning health care system. Health Aff (Millwood) 2014;33(7):1187–1194. doi: 10.1377/hlthaff.2014.0038. [DOI] [PubMed] [Google Scholar]
- 9.Beddhu S., Bruns F.J., Saul M., Seddon P., Zeidel M.L. A simple comorbidity scale predicts clinical outcomes and costs in dialysis patients. Am J Med. 2000;108(8):609–613. doi: 10.1016/s0002-9343(00)00371-5. [DOI] [PubMed] [Google Scholar]
- 10.Ruppert D. In: Encyclopedia of Statistical Sciences. 2nd ed. Kotz S., Read C.B., Balakrishnan N., Vidakovic B., Johnson N.L., editors. Wiley; 2006. Trimming and winsorization; p. 5. [Google Scholar]
- 11.Saran R., Robinson B., Abbott K.C., et al. US Renal Data System. 2018 Annual Data Report: Epidemiology of kidney disease in the United States. Am J Kidney Dis. 2019;73(3 suppl 1):A7–A8. doi: 10.1053/j.ajkd.2019.01.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Inker L.A., Coresh J., Levey A.S., Tonelli M., Muntner P. Estimated GFR, albuminuria, and complications of chronic kidney disease. J Am Soc Nephrol. 2011;22(12):2322–2331. doi: 10.1681/ASN.2010111181. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Golestaneh L., Alvarez P.J., Reaven N.L., et al. Longitudinal study of chronic kidney disease progression and associated costs. J Clin Pathways. 2017;3(9):40–49. S1–S10. [Google Scholar]
- 14.Smith D.H., Gullion C.M., Nichols G., Keith D.S., Brown J.B. Cost of medical care for chronic kidney disease and comorbidity among enrollees in a large HMO population. J Am Soc Nephrol. 2004;15(5):1300–1306. doi: 10.1097/01.asn.0000125670.64996.bb. [DOI] [PubMed] [Google Scholar]
- 15.Lage M.J., Boye K.S., Bae J.P., Wu J., Mody R., Botros F.T. The association between the severity of chronic kidney disease and medical costs among patients with type 2 diabetes. J Med Econ. 2019;22(5):447–454. doi: 10.1080/13696998.2019.1581208. [DOI] [PubMed] [Google Scholar]
- 16.Stevens P.E., Levin A. Kidney Disease: Improving Global Outcomes Chronic Kidney Disease Guideline Development Work Group Members. Evaluation and management of chronic kidney disease: synopsis of the kidney disease: improving global outcomes 2012 clinical practice guideline. Ann Intern Med. 2013;158(11):825–830. doi: 10.7326/0003-4819-158-11-201306040-00007. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Figure S1; Item S1; Table S1-S9