Kidney Failure Risk Equation and Cost of Care in Patients with Chronic Kidney Disease

Abstract

Background and objectives 

Patients with CKD exhibit heterogeneity in their rates of progression to kidney failure. The kidney failure risk equation (KFRE) has been shown to accurately estimate progression to kidney failure in adults with CKD. Our objective was to determine health care utilization patterns of patients on the basis of their risk of progression.

Design, setting, participants, & measurements 

We conducted a retrospective cohort study of adults with CKD and eGFR of 15–59 ml/min per 1.73 m² enrolled in multidisciplinary CKD clinics in the province of Saskatchewan, Canada. Data were collected from January 1, 2004 to December 31, 2012 and followed for 5 years (December 31, 2017). We stratified patients by eGFR and risk of progression and compared the number and cost of hospital admissions, physician visits, and prescription drugs.

Results 

In total, 1003 adults were included in the study. Within the eGFR of 15–29 ml/min per 1.73 m² group, the costs of hospital admissions, physician visits, and drug dispensations over the 5-year study period comparing high-risk patients with low-risk patients were (Canadian dollars) $89,265 versus $48,374 (P=0.008), $23,423 versus $11,231 (P<0.001), and $21,853 versus $16,757 (P=0.01), respectively. Within the eGFR of 30–59 ml/min per 1.73 m² group, the costs of hospital admissions, physician visits, and prescription drugs were $55,944 versus $36,740 (P=0.10), $13,414 versus $10,370 (P=0.08), and $20,394 versus $14,902 (P=0.02) in high-risk patients in comparison with low-risk patients, respectively, for progression to kidney failure.

Conclusions 

In patients with CKD and eGFR of 15–59 ml/min per 1.73 m² followed in multidisciplinary clinics, the costs of hospital admissions, physician visits, and drugs were higher for patients at higher risk of progression to kidney failure by the KFRE compared with patients in the low-risk category. The high-risk group of patients with CKD and eGFR of 15–29 ml/min per 1.73 m² had stronger association with hospitalizations costs, physician visits, and drug utilizations.

Visual Abstract

Introduction

As kidney function declines, the costs of CKD care increase rapidly as patients experience higher rates of hospitalizations and emergency department visits (^123–4). CKD affects between 10% and 15% of the population, but disproportionately accounts for 20% of the health budget (⁵). Patients with CKD often have comorbid diseases, such as diabetes mellitus and congestive heart failure, which further increase the cost of care (⁶,⁷). In patients with kidney failure requiring dialysis, costs for treatment are exponentially higher compared with CKD (⁸), and patients struggle to remain employed and face potential relocation, thereby increasing the burden to society. Costs for CKD care have thus far been published using eGFR-based CKD stages (⁹), and several factors contribute to increased costs within each stage (⁹). For instance, an older patient with multiple comorbid conditions and complications, including anemia, hyperkalemia, and metabolic acidosis, and an eGFR of 28 ml/min can have markedly higher costs than a similar-aged patient with nonprogressive, nonproteinuric CKD and an eGFR of 28 ml/min.

The kidney failure risk equation (KFRE) was developed in 2011 and accurately predicts the risk of kidney failure requiring dialysis in individuals at risk of progression to KRT (¹⁰). Although KFRE has been validated in numerous populations (⁸), its ability to predict health care and resource utilization in patients with CKD and eGFR of 15–59 ml/min per 1.73 m² remains unknown. If KFRE can help identify patients with low-cost/nominal resource–intensive treatment from patients with high-cost/resource-intensive treatment, it could be useful to determine patient management and resource allocation in outpatient multidisciplinary care–based kidney clinics. Furthermore, the equations can provide a framework for multidisciplinary clinics to provide risk-based care, particularly in scenarios where clinics are responsible for costs as part of at-risk payment models. The purpose of this analysis was to examine the association between risk of progression by KFRE and resource utilization (hospitalizations, physician visits, and drug usage) and associated costs in the setting of a universal health care system.

Materials and Methods

Study Population

In this retrospective cohort study, we included adult (≥18 years) patients with CKD and eGFR of 30–59 or 15–29 ml/min per 1.73 m² followed in the provincial CKD multidisciplinary clinics in Saskatchewan, Canada, from January 1, 2004 to December 31, 2012, who had available laboratory data to calculate the eight-variable KFRE. Patients were followed for 5 years after the 2004–2012 time frame to assess health care utilization and cost. Ethical approval to conduct this research was granted by the Saskatchewan Health Authority Research Ethics Board (approval identification no. REB-15–32). The provincial CKD clinics offer nephrology services to a wide geographic area spanning 651,900 km² with a population of approximately 1.17 million. The roles of the allied health care team are highlighted in Supplemental Table 1.

Clinical and laboratory data were collected from the provincial kidney repository, the Medical Information Quality System, for all patients. The data included eight variables that were part of the KFRE (sex, age, eGFR, serum albumin, bicarbonate, phosphate, calcium, and urine albumin-creatinine ratio) at the time of enrollment in the clinic (Supplemental Table 2). Information on death and dialysis was also available through these databases. These variables were added to KFRE to generate the risk of progression (percentage) at 2 and 5 years after enrollment. The risk of progression was defined as progression to kidney failure depending on the eGFR ranges: in patients with CKD and eGFR of 30–59 ml/min per 1.73 m², low risk (<5% over 5 years), medium risk (5%–15% over 5 years), and high risk (>15% over 5 years), and in patients with CKD and eGFR of 15–29 ml/min per 1.73 m², low risk (<10% over 2 years), medium risk (10%–20% over 2 years), and high risk (>20% over 2 years) (¹⁰,¹¹).

The primary outcome was to examine costs of all hospitalizations, physician visits to family physicians and specialists, and drug costs on the basis of KFRE risk categories (low, medium, and high) across two eGFR ranges (15–29 and 30–59 ml/min per 1.73 m²). The main comparison of interest was comparison of high- versus low-risk categories for progression to kidney failure. Costs included in the analysis are taken from the perspective of the Canadian single-payer health care system and include all direct costs related to CKD and non-CKD care and management. To obtain health care utilization and cost information, clinical data of eligible patients with CKD (Figure 1) were linked with multiple administrative health databases using deidentified patient health card numbers. We censored patients at death and dialysis initiation.

Figure 1.: Patient recruitment flow diagram. KFRE, kidney failure risk equation.

The Discharge Abstract Database, Medical Services Billing Claims, and the Prescription Drug Database were accessed through the Saskatchewan Health Quality Council to obtain utilization and cost estimates for hospitalization, physician visits, and prescription drug dispensations of the CKD cohort during the study period. All costs were adjusted for inflation and presented in 2015/2016 Canadian dollars. The Person Health Registry System was used to obtain health coverage information and demographics such as age and sex.

Statistical Analyses

Categorical variables are expressed as numbers and percentages, and quantitative variables are expressed as mean ±SD. Categorical variables were compared by risk category for progression to kidney failure (low, medium, and high risk) using the chi-squared test. Rates evaluated in the study were presented as total events per patient-year and were statistically compared using negative binomial regression. For total and marginal costs associated with each risk stage, a generalized linear model (GLM) applying a γ-distribution and log link was applied. In addition to risk levels, there might be other factors that can affect the number of health care utilization and relevant costs of this cohort, such as demographic characteristics and comorbidities. To control for these factors, we utilized negative binomial regression and GLM to estimate the effect of different eGFR ranges (15–29 and 30–59 ml/min per 1.73 m²) on health care utilizations and costs, respectively. The marginal effect of different risk levels is calculated as the discrete change on the expected mean value of health service use (i.e., mean number of hospital inpatient admissions) for a change in the risk variable from zero to one (i.e., high risk versus low risk). Furthermore, models are also adjusted for the time that patients were in the study period. Costs were annualized in the log-γ model by applying an offset variable equal to the log-transformed patient-years of exposure. Likewise, for the utilization models (negative binomial), the same offset variable was applied to determine rates per patient-year of exposure. We also tested for an interaction between eGFR range, risk category of progression to kidney failure by KFRE, health care utilization, and cost. P values of 0.05 were considered statistically significant. Statistical analyses were performed with SAS version 9.1 (SAS Institute Inc., Cary, NC) and Stata version 11 (StataCorp, College Station, TX).

Results

There were 1794 patients (with eGFR ranges of 60–89, 30–59, 15–29, and <15 ml/min per 1.73 m²) referred to the CKD clinics across the province; 1526 patients were followed in the clinic between January 1, 2004 and December 31, 2012. In total, 230 of 1526 patients with eGFR>60 and <15 ml/min per 1.73 m² were excluded from the study. Among the remaining 1296 patients, 293 were excluded from the study as (1) they had missing information on key variables needed for the risk calculation or (2) they did not have continuous health coverage during the study period (1 year prior to enrollment to CKD clinics and 5 years after index date or death). The final CKD cohort included 1003 patients with eGFR of 30–59 or 15–29 ml/min per 1.73 m². The median follow-up time (interquartile range) was 5 (3.8–5) years in the group with eGFR of 30–59 ml/min per 1.73 m² and 5 (2.7–5) years in the group with eGFR of 15–29 ml/min per 1.73 m² (Supplemental Table 3).

The mean age (SD) of all cohorts was 71 (13) years, and 57% were men. The percentage of patients ≥65 years of age was 75%. In patients with eGFR of 30–59 ml/min per 1.73 m², 311 (59%), 150 (28%), and 68 (13%) were in low-, medium-, and high-risk categories by KFRE, respectively. Among the patients with eGFR of 15–29 ml/min per 1.73 m², 275 (58%), 86 (18%), and 113 (24%) were in similar categories, respectively. Table 1 and Supplemental Tables 4 and 5 show baseline characteristics.

Table 1. - Baseline characteristics of patients with eGFR of 15–59 ml/min per 1.73 m² in the provincial CKD multidisciplinary clinics in Saskatchewan, Canada, from January 1, 2004 to December 31, 2012

Variable	eGFR=30–59 ml/min per 1.73 m2, n=529, Mean±SD or n (%)	eGFR=15–29 ml/min per 1.73 m2, n=474, Mean±SD or n (%)	All Cohort: eGFR=15–59 ml/min per 1.73 m2, n=1003, Mean±SD or n (%)
Age, yr	71±13	72±13	71±13
Age, ≥65 yr	381 (72%)	370 (78%)	751 (75%)
Men	330 (62%)	240 (51%)	570 (57%)
eGFR, ml/min per 1.73 m2	40±8	23±4	32±11
UACR, mg/g	424±850	900±1640	639±1341
Calcium, mg/dl	9.4±0.8	9.5±0.6	9.4±0.7
Phosphate, mg/dl	3.9±0.7	3.8±0.7	3.6±0.7
Albumin, g/dl	3.8±0.5	3.7±0.5	3.8±0.5
Bicarbonate, mEq/L	25.3±3.4	24.7±11.3	25.0±8.2
Risk
Low risk	311 (59%)	275 (58%)	586 (58%)
Medium risk	150 (28%)	86 (18%)	236 (24%)
High risk	68 (13%)	113 (24%)	181 (18%)

UACR, urine albumin-creatinine ratio; risk, risk of CKD progression to kidney failure on the basis of the kidney failure risk equation.

Of the patients with eGFR of 30–59 ml/min per 1.73 m², 4% at low risk, 11% at medium risk, and 26% at high risk for progression to kidney failure advanced to dialysis by 5 years (P<0.001). In patients with eGFR of 15–29 ml/min per 1.73 m², 7% at low risk, 17% at medium risk, and 48% at high risk progressed to dialysis over 2 years (P<0.001). Among patients with eGFR of 30–59 ml/min per 1.73 m², 31% at low risk, 36% at medium risk, and 28% at high risk for progression to kidney failure died over 5 years. Of the patients with eGFR of 15–29 ml/min per 1.73 m², 15%, 21%, and 16% of low, medium, and high risk died by 2 years, respectively (Tables 2 and 3).

Table 2. - Risk of progression to dialysis or death on the basis of the kidney failure risk equation for eGFR=30–59 ml/min per 1.73 m²

eGFR=30–59 ml/min per 1.73 m2	Risk Group, n (%)			P Value
	Low Risk, n=311	Medium Risk, n=150	High Risk, n=68
Dialysis (2 yr)	6 (2%)	<6 (<5%)a	6 (9%)	0.02
Death (2 yr)	40 (13%)	19 (13%)	<6 (<9%)a	0.42
Dialysis (5 yr)	12 (4%)	16 (11%)	18 (26%)	<0.001
Death (5 yr)	96 (31%)	54 (36%)	19 (28%)	0.40

^aBecause of the Saskatchewan Health Quality Council privacy rules, small cells (less than six patients) were suppressed. P values correspond to the chi-squared test. Two years indicates risk of CKD progression (to dialysis or death) at 2 years after enrollment; 5 years indicates risk of CKD progression (to dialysis or death) at 5 years after enrollment.

Table 3. - Risk of progression to dialysis or death on the basis of the kidney failure risk equation for eGFR=15–29 ml/min per 1.73 m²

eGFR=15–29 ml/min per 1.73 m2	Risk Group, n (%)			P Value
	Low Risk, n=275	Medium Risk, n=86	High Risk, n=113
Dialysis (2 yr)	19 (7%)	15 (17%)	54 (48%)	<0.001
Death (2 yr)	41 (15%)	18 (21%)	18 (16%)	0.41
Dialysis (5 yr)	41 (15%)	29 (34%)	75 (66%)	<0.001
Death (5 yr)	118 (43%)	43 (50%)	54 (48%)	0.39

Two years indicates risk of CKD progression (to dialysis or death) at 2 years after enrollment; 5 years indicates risk of CKD progression (to dialysis or death) at 5 years after enrollment.

Comparisons of various health service utilizations and costs of the CKD cohort are presented in Figure 2 and Tables 4–7. After controlling for potential confounders, within the CKD group with eGFR of 30–59 ml/min per 1.73 m², patients at high risk for progression to kidney failure utilized 50% more hospital-based services (inpatient and day surgeries) compared with patients in the low-risk category over the 5-year study period (P=0.006). High-risk patients with eGFR of 30–59 ml/min per 1.73 m² had higher utilization of physician service compared with low-risk patients (52% more; P<0.001). Drug dispensations were not statistically different across different risk groups by KFRE (P=0.62). Similarly, our findings were consistent in patients with CKD and eGFR of 15–29 ml/min per 1.73 m², where high-risk patients consumed 72% more hospital-based services and 2.2 times more physician services compared with their low-risk counterparts (both P<0.001). We did not observe any difference in drug dispensations (Figure 2, Tables 4 and 5).

Figure 2.: Cost and number of hospital admissions (inpatient and outpatient), physician visits, and drug dispensations in stages G3 and G4 based on the risk of progression. Risk indicates the risk of CKD progression to kidney failure on the basis of KFRE. Line charts show numbers, and bar charts show costs. The charts in (A)–(C) show comparisons of expected health service utilizations and cost (over 5 years) in patients with eGFR of 30–59 ml/min per 1.73 m² after controlling for age, sex, and comorbidities, and the charts in (D)–(F) show comparisons of expected health service utilizations and cost (over 5 years) in patients with eGFR of 15–29 ml/min per 1.73 m² after controlling for age, sex, and comorbidities. The currency is Canadian dollars (CAD).

Table 4. - Health care utilization (hospital admissions, physician visits, and drug dispensations) on the basis of risk of progression for eGFR=30–59 ml/min per 1.73 m²

Risk Group	Hospital Admissions, n=529			Physician Visits, n=529			Drug Dispensations, n=529
	Rate Ratio (95% Confidence Interval)	Overall P Value	P Value	Rate Ratio (95% Confidence Interval)	Overall P Value	P Value	Rate Ratio (95% Confidence Interval)	Overall P Value	P Value
Low risk, n=311	Reference	0.009	—	Reference	<0.001	—	Reference	0.85	—
Medium risk, n=150	1.2 (1.0 to 1.5)		0.04	1.1 (0.9 to 1.2)		0.22	1 (0.8 to 1.2)		0.91
High risk, n=68	1.5 (1.1 to 2.0)		0.006	1.5 (1.2 to 1.8)		<0.001	1.1 (0.8 to 1.3)		0.62

The rate ratio is the ratio of health care utilizations. Risk indicates the risk of CKD progression to kidney failure on the basis of the kidney failure risk equation. Reference indicates the reference group (low risk). The results of binomial regression are shown after adjustment for age, sex, and Charlson Index (control variables=risk levels, age, sex, and Charlson Index; offset variable=patient time). Overall P value (a single P value) represents the trend between three risk groups, whereas multiple P values (in the column next to overall P values) compare individual risk categories with the reference group (medium risk versus low risk and high risk versus low risk). —, P value does not apply to the reference group.

Table 5. - Health care utilization (hospital admissions, physician visits, and drug dispensations) on the basis of risk of progression for eGFR=15–29 ml/min per 1.73 m²

Risk Group	Hospital Admissions, n=474			Physician Visits, n=474			Drug Dispensations, n=474
	Rate Ratio (95% Confidence Interval)	Overall P Value	P Value	Rate Ratio (95% Confidence Interval)	Overall P Value	P Value	Rate Ratio (95% Confidence Interval)	Overall P Value	P Value
Low risk, n=275	Reference	<0.001	—	Reference	<0.001	—	Reference	0.70	—
Medium risk, n=86	1.2 (0.9 to 1.5)		0.10	1.2 (1.05 to 1.5)		0.01	1.1 (0.9 to 1.3)		0.51
High risk, n=113	1.71 (1.4 to 2.1)		<0.001	2.2 (1.9 to 2.6)		<0.001	0.97 (0.8 to 1.2)		0.78

The rate ratio is the ratio of health care utilizations. Risk indicates the risk of CKD progression to kidney failure on the basis of the kidney failure risk equation. Reference indicates the reference group (low risk). The results of binomial regression are shown after adjustment for age, sex, and Charlson Index (control variables=risk levels, age, sex, and Charlson Index; offset variable=patient time). Overall P value (a single P value) represents the trend between three risk groups, whereas multiple P values (in the column next to overall P values) compare individual risk categories with the reference group (medium risk versus low risk and high risk versus low risk). —, P value does not apply to the reference group.

Even though among patients with eGFR of 30–59 ml/min per 1.73 m², the high-risk group (by KFRE) had higher utilization of hospital and physician services compared with the low-risk group, we did not observe any significant difference in hospital ($55,944 versus $36,740 [Canadian dollars]) and physician costs ($13,414 versus $10,370). However, cost of drug dispensations was 37% higher for high-risk patients compared with low-risk patients ($20,394 versus $14,902; P=0.02). In the group with eGFR of 15–29 ml/min per 1.73 m², high-risk patients for progression to kidney failure had 86% and 109% higher hospital ($89,265 versus $48,375; P=0.008) and physician costs ($23,423 versus $11,232; P<0.001) compared with low-risk patients, respectively. Cost of drug dispensations was also 30% higher for the high-risk group compared with the low-risk group ($21,854 versus $16,757; P=0.01) (Figure 2, Tables 6 and 7). Overall, high-risk patients not only consumed higher proportions of health services, but the total costs over 5 years in both patients with eGFR of 30–59 ml/min per 1.73 m² ($93,285 versus $65.845; P=0.04) and patients with eGFR of 15–29 ml/min per 1.73 m² ($132,908 versus $78,021; P<0.001) were higher as opposed to low risk as well (Supplemental Table 6).

Table 6. - Health care cost (hospital admissions, physician visits, and drug dispensations) on the basis of risk of progression for eGFR=30–59 ml/min per 1.73 m²

Risk Group	Hospital Admissions (Costs), n=529				Physician Visits (Costs), n=529				Drug Dispensations (Costs), n=529
	5-yr Cost (95% Confidence Interval)	Marginal Cost (95% Confidence Interval)	Overall P Value	P Value	5-yr Cost (95% Confidence Interval)	Marginal Cost (95% Confidence Interval)	Overall P Value	P Value	5-yr Cost (95% Confidence Interval)	Marginal Cost (95% Confidence Interval)	Overall P Value	P Value
Low risk, n=311	36,740 (27,514 to 45,966)	Reference	0.19	—	10,370 (8521 to 12,219)	Reference	0.07	—	14,902 (13,662 to 16,141)	Reference	0.02	—
Medium risk, n=150	40,644 (30,755 to 50,534)	3904 (–9697 to 17,505)		0.57	9936 (8679 to 11,192)	−434.0 (–2592 to 1724)		0.69	16,601 (14,188 to 19,013)	1699 (–981 to 4378)		0.21
High risk, n=68	55,944 (35,236 to 76,652)	19,204 (–4084 to 42,492)		0.10	13,414 (10,342 to 16,487)	3044 (–373 to 6462)		0.08	20,394 (16,068 to 24,720)	5492 (919 to 10,065)		0.02

Reference indicates the reference group (low risk). Five-year cost is the cost of health care over 5 years. Marginal cost is the difference of 5-year cost between each risk group and the reference group. Risk indicates risk of CKD progression to kidney failure on the basis of the kidney failure risk equation. The currency is Canadian dollars. The results of generalized liner models are shown after adjustment for age, sex, and Charlson Index (control variables=risk levels, age, sex, and Charlson Index; offset variable=patient time). Overall P value (a single P value) represents the trend between three risk groups, whereas multiple P values (in the column next to the overall P values) compare individual risk categories with reference group (medium risk versus low risk and high risk versus low risk). —, P value does not apply to the reference group.

Table 7. - Health care cost (hospital admissions, physician visits, and drug dispensations) on the basis of risk of progression for eGFR=15–29 ml/min per 1.73 m²

Risk Group	Hospital Admissions (Costs), n=474				Physician Visits (Costs), n=474				Drug Dispensations (Costs), n=474
	5-Year Cost (95% Confidence Interval)	Marginal Cost (95% Confidence Interval)	Overall P Value	P Value	5-Year Cost (95% Confidence Interval)	Marginal Cost (95% Confidence Interval)	Overall P Value	P Value	5-Year Cost (95% Confidence Interval)	Marginal Cost (95% Confidence Interval)	Overall P Value	P Value
Low risk, n=275	48,375 (36,573 to 60,176)	Reference	0.01	—	11,232 (9465 to 12,998)	Reference	<0.001	—	16,757 (14,808 to 18,706)	Reference	0.02	—
Medium risk, n=86	76,686 (37,003 to 116,368)	28,311 (−13,348 to 69,969)		0.18	13,475 (10,636 to 16,313)	2243 (–1165 to 5652)		0.20	20,629 (17,824 to 23,434)	3872 (366 to 7378)		0.03
High risk, n=113	89,265 (62,486 to 116,045)	40,890 (10,481 to 71,300)		0.008	23,423 (19,775 to 27,071)	12,192 (7921 to 16,463)		<0.001	21,854 (18,508 to 25,199)	5097 (944 to 9249)		0.01

Reference indicates the reference group (low risk). Five-year cost is the cost of health care over 5 years. Marginal cost is the difference of 5-year cost between each risk group and the reference group. Risk indicates risk of CKD progression to kidney failure on the basis of the kidney failure risk equation. The currency is Canadian dollars. The results of generalized liner models are shown after adjustment for age, sex, and Charlson Index (control variables=risk levels, age, sex, and Charlson Index; offset variable=patient time). Overall P value (a single P value) represents the trend between three risk groups, whereas multiple P values (in the column next to the overall P values) compare individual risk categories with reference group (medium risk versus low risk and high risk versus low risk). —, P value does not apply to the reference group.

We also tested for an interaction between eGFR group, risk level (by KFRE), health care utilization, and cost. We found a statistically significant interaction between eGFR group and risk level for utilization of physician services (P<0.001) and physician costs (P=0.001). No interaction was found between risk category of progression to kidney failure and total cost (P=0.08) by eGFR group. However, in patients with CKD and eGFR of 15–29 ml/min per 1.73 m², the high-risk group had stronger association with hospitalization cost, physician visits, and drug utilization (Supplemental Tables 7–10). Additional negative binomial and GLM regression results are demonstrated in Supplemental Tables 11–15.

Discussion

In this analysis, we examined differences in cost and health resource utilization patterns on the basis of the risk of progression to kidney failure for a 5-year period after being enrolled in a multidisciplinary care clinic. Our findings show that KFRE, intended to predict the risk of CKD progression, helps identify subgroups of patients with high resource utilization/health care costs compared with those with substantially lower health resource use. A significant proportion of patients with CKD and eGFR of 30–59 or 15–29 ml/min per 1.73 m² never progress to kidney failure and continue to have stable kidney function for years (⁸). The minority of patients who progress to kidney failure consume a disproportionate share of health care resources and the burden borne to the societal collective (¹²). This heterogeneity in progression is better captured by using risk prediction equations but goes unrecognized in the eGFR-based staging model. Prior to the availability and usage of KFRE, these patients were treated as a homogeneous group with similar risk of progression. In this study, we demonstrate that patients at higher risk of CKD progression had a greater number of inpatient visits, had higher drug costs, and undertook more outpatients visits to specialists. Taken together, these findings suggest that KFRE-based risk categories can be used to determine subgroups of patients with high health resource use for potential intervention.

A recent comprehensive economic analysis of patients with nondialysis CKD showed that the cost of CKD care averages $14,634 per year and is higher for patients with a greater number of comorbid conditions, and the annual cost of caring for Canadians with nondialysis CKD is approximately $32 billion per year (⁵). Similarly, a 2004 US study published soon after adoption of the National Kidney Foundation–Kidney Disease Outcomes Quality Initiative guidelines showed that patients with higher CKD stages in comparison with earlier stages had 1.9–2.5 times more prescriptions, had 1.3–1.9 times more outpatient visits, and were 1.6–2.2 times more likely to have had an inpatient stay (¹³). In a recently published study, mean total health care costs among patients with CKD without comorbidities were 31% higher than among patients without CKD ($7374 versus $5631 [US dollars], respectively) (¹⁴). Furthermore, a study from 2017 examining the economic effect of CKD on health plans showed that mean annualized Medicare incremental all-cause costs increased exponentially with advancing CKD stage, from $8091 (US dollars; no CKD) to $46,178 (CKD stages 4–5) (¹⁵). The authors’ findings showed that hospitalizations were the major driver of costs by CKD stage. The investigators showed that rising inpatient costs accounted for 80% of the cost increase with disease progression from stage 3b to 5 and kidney failure, whereas medication costs contributed to a mere 5% of the costs (¹⁵).

To our knowledge, this is the first study to examine differences in health care costs by risk of progression to kidney failure and is, therefore, unique as all previous economic studies in patients with CKD have described costs by eGFR-based CKD stage. In our analysis, in patients with eGFR of 15–29 ml/min per 1.73 m², the total cost of care per patient-year in the high-risk group was 98% higher than the low-risk group, and the cost in the medium-risk group was 29% higher than the low-risk patients. The costs were proportionately lower in patients with eGFR of 30–59 ml/min per 1.73 m² (58% higher in the high-risk group compared with the low-risk group and a 21% difference between the medium- and low-risk groups); 53%–58% of the total costs were due to hospitalization, and 15%–20% of the costs were borne by outpatient and clinic visits. In our study, although the number of medications did not really change, the cost of drug usage incrementally increased with risk of progression to kidney failure by KFRE. For the group with eGFR of 30–59 ml/min per 1.73 m², medium-risk patients had a 20% higher medication cost and high-risk patients had a 54% higher cost in contrast to the low-risk group (by KFRE). In the group with eGFR of 15–29 ml/min per 1.73 m², medium-risk patients had a 41% higher medication cost compared with low-risk patients, and medication costs per patient-year in higher-risk patients were 58% higher than in low-risk patients (by KFRE).

Although CKD is expensive, the cost of providing KRT is significantly higher than most other chronic disease management. A recent Canadian study showed that annual maintenance expenses were estimated as $64,214 for in-center facility hemodialysis, $43,816 for home hemodialysis with the NxStage System One, $39,236 for home hemodialysis with conventional dialysis machines, and $38,658 for peritoneal dialysis (¹⁶). In contrast, the cost of managing patients in our CKD multidisciplinary clinic was $3120 per year per patient, our national per capita health expenditure is $7064, and the cost of managing heart failure is $28,000 (¹⁷). In the United States, the recent executive order, Advancing American Kidney Health (AAKH), has stated three major goals: (1) a 25% decrease in kidney failure incidence by 2030, (2) an increase in the rate of home therapies to 80% for incident patients with kidney failure by 2025, and (3) a doubling of the rate of kidney transplantation by 2030 (¹⁸,¹⁹). In addition, AAKH creates funding/payment models such as the Kidney Care Choices and the Comprehensive Kidney Care Contracting, both of which provide strong financial incentives to US nephrologists, and managed care organizations/dialysis providers to lower Medicare Part A and B costs for recipients with CKD stage 4 (eGFR of 15–29 ml/min per 1.73 m²) (²⁰). To meet these targets, it is imperative to triage the higher-risk patients to receive multidisciplinary team–based care or patient management to optimize care to delay the rate of progression to kidney failure. These care models have been associated with better education (²¹), focus on self-care (²¹), dietary interventions (²¹), timely transplant referrals (²²), modality education (²²), and lower hospitalizations (²³,²⁴) and mortality (²³,²⁴). Although these models can be expensive to implement, by effectively triaging patients at the highest risk of progression and subsequently delaying the rate of progression and hospitalizations as well as by encouraging home therapies, the overall costs of patient care can decrease over time.

This study has several strengths. This is the first study to show that KFRE can help identify subgroups that require greater health resource utilization and can expand its scope/utility beyond prediction of kidney failure alone. The data included in this study represent all available patients who were enrolled in CKD multidisciplinary clinics in a province of 1.17 million, benefit from universal health care, and have low rates of attrition. The data gathered, therefore, included all inpatient visits, medication use, and outpatient consults. Limitations include its generalizability outside the Canadian health care system; however, it is important to acknowledge that gradients/trends in costs of care for chronic conditions in Canada are the same as in the United States and European countries, and there is little reason to suspect that KFRE would not discriminate costs or health care utilization in these health systems. Other variables, such as socioeconomic (i.e., education and income) and health behaviors (i.e., smoking status), may also affect health care use and subsequent costs. Unfortunately, such variables were not available within the utilized databases and may be a source of omitted variable bias in the regression setting.

Nonetheless, a validation study from a US or European system would further enhance the generalizability of these findings. In addition, the statistical power to detect the cost difference between various risk categories by KFRE may be limited due to a relatively small number of patients in the risk categories. Finally, it is possible that these patients selected for CKD multidisciplinary clinics represent a biased higher-risk sample of all patients with CKD, which can limit the generalizability of the findings. However, we believe that these findings would be applicable to referred patients to nephrology in the United States and in similar health care systems where primary care providers are responsible for most patients with earlier stages of CKD.

In conclusion, in our study of patients with CKD referred to multidisciplinary CKD clinics, KFRE, designed to predict the risk of dialysis in patients with CKD, helps identify patients with higher health resource utilization and health care costs compared with those with lower health resource use. Integration of KFRE in risk-based treatment pathways that guide the intensity of CKD care may improve health system and patient outcomes.

Disclosures

T.W. Ferguson reports consultancy agreements with Clinpredict Ltd., Quanta Dialysis Technologies Ltd., Strategic Health Resources, and Tricida Inc. and receiving honoraria from Baxter Corporation. M. Jafari reports employment with the Dr. T. Bhanu Prasad Medical Prof. Corp. S. Jin was an employee of the Saskatchewan Health Quality Council. J. Kappel reports employment with the Kidney Foundation of Canada. D. Kozakewycz reports employment with the Saskatchewan Health Authority. M. Osman was an employee of the Saskatchewan Health Quality Council. M. Osman reports employment with the Saskatchewan Medical Association. B. Prasad reports employment with the Saskatchewan Health Authority; receiving research funding from Baxter and Medtronic; receiving honoraria from AstraZeneca, Janssen, and Otsuka; and speakers bureau for AstraZeneca, Janssen, and Otsuka. N. Tangri reports consultancy agreements with Healthlogic, Mesentech Inc., PulseData Inc., Renibus, and Tricida Inc.; ownership interest in Clinpredict Ltd., Healthlogic, Klinrisk, Mesentech Inc., PulseData Inc., Renibus, and Tricida Inc.; receiving research funding from AstraZeneca, Bayer, BI-Lilly, Janssen, Otsuka, and Tricida Inc.; receiving honoraria from AstraZeneca, Bayer, BI-Lilly, Janssen, Otsuka Pharmaceuticals, and Pfizer; and other interests/relationships as the founder of Clinpredict Ltd., as the founder of Klinrisk, and with the National Kidney Foundation. The remaining author has nothing to disclose.

Funding

This study was awarded Saskatchewan Health Research Foundation grant 3910 (to B. Prasad) and in-kind support by the Saskatchewan Centre for Patient Oriented Research and the Saskatchewan Health Quality Council to perform the data linkage and analysis.