Nephrology Referral Based on Laboratory Values, Kidney Failure Risk, or Both: A Study Using Veterans Affairs Health System Data

Abstract

Rationale & Objective:

Current guidelines for nephrology referral are based on laboratory criteria. We sought to evaluate whether nephrology referral patterns reflect current clinical practice guidelines and to estimate the change in referral volume if they were based on the estimated risk of kidney failure.

Study Design:

Observational cohort.

Setting & Participants:

Retrospective study of 399,644 veterans with chronic kidney disease (October 1, 2015 through September 30, 2016).

Exposure:

Laboratory referral criteria based on Veterans Affairs/Department of Defense guidelines, categories of predicted risk for kidney failure using the Kidney Failure Risk Equation, and the combination of laboratory referral criteria and predicted risk.

Outcome:

Number of patients identified for referral.

Analytical Approach:

We evaluated the number of patients who were referred and their predicted 2-year risk for kidney failure. For each exposure, we estimated the number of patients who would be identified for referral.

Results:

There were 66,276 patients who met laboratory indications for referral. Among these patients, 11,752 (17.7%) were referred to nephrology in the following year. The median 2-year predicted risk of kidney failure was 1.5% (interquartile range, 0.3%−4.7%) among all patients meeting the laboratory referral criteria. If referrals were restricted to patients with a predicted risk of ≥1% in addition to laboratory indications, the potential referral volume would be reduced from 66,276 to 38,229 patients. If referrals were based on predicted risk alone, a 2-year risk threshold of 1% or higher would identify a similar number of patients (72,948) as laboratory-based criteria with median predicted risk of 2.3% (interquartile range, 1.4%−4.6%).

Limitations:

Missing proteinuria measurements.

Conclusions:

The current laboratory-based guidelines for nephrology referral identify patients who are, on average, at low risk for progression, most of whom are not referred. As an alternative, referral based on a 2-year kidney failure risk exceeding 1% would identify a similar number of patients but with a higher median risk of kidney failure.

National and international guidelines identify patients who may benefit from nephrology referral to delay progression of chronic kidney disease (CKD), manage complications, and prepare for kidney failure.^1–4 Timely recognition of CKD is a necessary requisite for nephrology referral, and it is facilitated by the automated reporting of estimated glomerular filtration rate (eGFR) by laboratories.² In addition to level of eGFR, guidelines from KDIGO (Kidney Disease: Improving Global Outcomes) and other organizations outline additional indications for referral that can be identified through laboratory testing, such as level of albuminuria and rapid disease progression. Despite the availability of guidelines to facilitate recognition of CKD, there has been little progress in reducing the burden of CKD or improving preparation of those who progress to kidney failure.^5,6

These observations have spurred interest in electronic clinical decision support tools that can disseminate relevant guidelines to improve identification of CKD and patients who may benefit from referral.⁷ However, large increases in referral volume may be problematic without appropriate prioritization for patients who need nephrology care the most. A study of a single health care system has shown that complete adherence to referral guidelines could result in 67% higher referrals to nephrology, resulting in 38% higher nephrology patient volume.⁷ Accommodating such an increase in workload would require significant changes in care delivery and resources, and some have argued that the additional nephrology referrals would be unjustified without clear evidence of benefit to patients. One proposed method to efficiently utilize nephrology resources involves targeting high-risk patients for nephrology care using kidney failure risk prediction calculators,^8,9 an approach already in use at certain centers.¹⁰ Indeed, this method of risk-based triage is recommended in the KDIGO guidelines to plan for kidney replacement therapy.² Implementation directly at the point of care with clinical decision support tools would allow physicians to make decisions with direct estimates of risk, rather than requiring them to review a checklist of indications. However, thresholds of risk at which to refer to a nephrologist have not been standardized.

The Veterans Health Administration of the Department of Veterans Affairs (VA) is a national, integrated health care system and the largest provider of CKD care in the United States. The VA and Department of Defense (DoD) have developed clinical practice guidelines for CKD management, including suggested “potential indications” for nephrology referral based on laboratory values.¹ Recently, these guidelines were updated to reflect a diminished emphasis on laboratory-based indications and an increased emphasis on assessing the risk of progression. The VA is planning to implement clinical decision support for CKD evaluation and management. In preparation for the pilot implementation of these decision support tools, we sought to estimate the effect on nephrology referral volume if providers were to refer all patients who met a laboratory-based indication for nephrology referral. We also sought to estimate how the incorporation of different thresholds of risk, both alone and supplementing potential indications for referral, would modify referral volumes.

Methods

Patient Population

We conducted a retrospective cohort study of patients with CKD who sought care within the VA. We used laboratory and administrative data from the VA, Medicare claims, and the US Renal Data System, a national registry of patients receiving dialysis therapy for kidney failure. We used outpatient visit data from the VA Corporate Data Warehouse to identify patients who had 2 or more visits to a VA primary care physician between October 1, 2013, and September 30, 2015. We used the CKD-EPI equation to calculate eGFR from serum creatinine measurements and demographic data.¹¹ We defined CKD as 2 or more eGFR assessments persistently between 15 and 59 mL/min/1.73 m² over a period of 90 or more days between October 1, 2015, and September 30, 2016. Patients with diagnostic codes for hospice care were excluded, as this denotes a valid reason for non-referral. After excluding patients with an eGFR value <15 mL/min/1.73 m², kidney transplantation, or receipt of dialysis during this interval up to the index date (defined below), the analytic cohort consisted of 399,644 patients.

This study was approved by the Human Subjects Panel and Institutional Review Board at Stanford University and Veterans Affairs Palo Alto R&D Committee. The Stanford University institutional review board waived the requirement for informed consent because the study used data that had already been collected; therefore, the study was considered to have minimal risk to individuals. We followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline. A timeline of data collection is shown in Figure S1.

Referral Indications

Referral indications were captured from October 1, 2015, to September 30, 2016. The date of first meeting a referral indication was the index date. We extracted laboratory-based potential indications for nephrology referral based on VA/DoD guidelines. These guidelines provided recommendations, but not requirements, for referral. These included eGFR < 30 mL/min/1.73 m²; heavy proteinuria, defined as a urinary protein-creatinine ratio (UPCR) > 500 mg/g or urinary albumin-creatinine ratio (UACR) > 300 mg/g in those without diabetes, or a UPCR > 3,000 mg/g in those with diabetes; and decline of eGFR of ≥5 mL/min/1.73 m² per year (Table S1). If the UACR value was missing, UPCR or urinalysis protein measurements were converted to UACR.^12,13 Because the guidelines do not specify a minimum time frame or number of assessments to determine eGFR decline, we operationally defined this as an absolute decrease in eGFR of ≥5 mL/min/1.73 m² from any previous eGFR value measured 90–365 days preceding the latest assessment. In a sensitivity analysis, we used a more stringent definition of an absolute decrease in eGFR of ≥10 mL/min/1.73 m² from any previous eGFR value measured 365–730 days preceding the latest assessment.

Assessment of Patient Characteristics and Kidney Failure Risk

We recorded patient demographic characteristics from electronic health record data on the index date. We used patient zip code data to estimate median income from the American Community Survey.¹⁴ We ascertained comorbid conditions in the 2 years preceding the index date with VA medical SAS data sets and Medicare claims using International Classification of Disease, Ninth Revision, and Current Procedural Terminology codes. We calculated the Charlson comorbidity index values using the Deyo modification.¹⁵ We used the Kidney Failure Risk Equation (KFRE), which incorporates age, eGFR, sex, and UACR, to estimate kidney failure risk⁸ at the index date. If the UACR values on the index date were not available, we used the latest measurement in the 2 years preceding the index date. Among all patients with CKD not already receiving nephrology care, a total of 150,442 (41.5%) had a urine albumin or protein measurement and therefore had computable kidney failure risk.

Ascertainment of Referrals and Progression to Kidney Failure

We defined nephrology referral as either the placement of a consultation order or a completed visit with a nephrologist between October 1, 2015, and September 30, 2016. We ascertained kidney failure, defined as the initiation of maintenance dialysis or receipt of kidney transplantation, with a linkage to the US Renal Data System. We followed patients for 2 years after the index date to ascertain kidney failure.

Statistical Analysis

We compared characteristics of referred versus unreferred patients, using median and interquartile range (IQR) for continuous variables and proportions for categorical variables. Among the patients who met each referral indication, we evaluated the number of patients who were referred, the predicted 2-year risk for kidney failure, and the frequency of progression to kidney failure over 2 years.

Next, among all patients with CKD who had not received nephrology care in the prior year, we estimated the maximum referral volume that would be generated if VA providers referred all patients in each of 3 scenarios: (1) patients who met laboratory-based indications for nephrology referral, (2) patients who met a specified kidney failure risk threshold in addition to laboratory-based indications, and (3) patients who met a specified kidney failure risk threshold alone. We evaluated the 2-year kidney failure risk thresholds of 1%, 2%, 3%, 5%, and 10%.

Due to the large number of patients with missing urine albumin and urine protein measurements, a requirement for the KFRE, we extrapolated a base-case estimate by multiplying the proportions of patients meeting the predicted risk thresholds from the subset of patients with a computable KFRE by the total number of patients in the study cohort. We also estimated the number of patients meeting the risk thresholds when missing UACRs were replaced with values of 5 mg/g and 300 mg/g, to represent a plausible range of values in the general CKD population. Data collection and analysis were performed using R, version 3.6.1 (R Foundation for Statistical Computing) and SAS Enterprise Guide, version 7.1 (SAS Institute).

Results

Laboratory-Based Indications and Nephrology Referrals

Of 399,644 patients with CKD under VA primary care, 37,560 patients had visited a nephrologist in the previous year, while 362,084 patients had not (Fig 1). Among those who had not previously seen a nephrologist, 66,276 (18.3%) met an indication for referral, and 11,752 (17.7%) were referred. A total of 295,808 patients did not meet a referral indication, and 10,015 (3.4%) were referred.

Figure 1.

Cohort flow diagram. Comprises 399,644 veterans with CKD stages 3 or 4 (by eGFR) and regular visits to a VA primary care physician who were not under hospice care. The pathway on the left describes cohort selection for all patients with laboratory-based indications for nephrology referral. The pathway on the right describes cohort selection for all patients with CKD, regardless of whether they met an indication for referral. Abbreviations: CKD, chronic kidney disease; eGFR, estimated glomerular filtration rate; VA, Veterans Affairs.

The characteristics of patients meeting potential referral indications who were and were not referred to nephrology are detailed in Table 1. The median 2-year predicted risk of kidney failure was 1.5% (IQR, 0.3%−4.7%). The majority of patients who met an indication for referral did so based on the eGFR < 30 mL/min/1.73 m² criterion (Table 2). Referred patients were more likely to meet multiple potential indications for referral, especially eGFR < 30 mL/min/1.73 m² and heavy proteinuria. Those meeting the heavy proteinuria with diabetes indication or the eGFR < 30 mL/min/1.73 m² indication had the highest predicted risk of kidney failure at 2 years, 10.0% and 7.1%, respectively (Table S2). Among the referral indications, rapid eGFR decline was associated with the lowest predicted 2-year risk for kidney failure. For each potential referral indication, the actual frequency of kidney failure over 2 years closely mirrored predicted risk.

Table 1.

Characteristics of Patients Meeting Potential Indications for Nephrology Referral in the Veterans Affairs Health Care System in 2014, Stratified by Referral Status

	Not Referred (n = 54,524)	Referred (n = 11,752)
Age, y	78.2 ± 9.6	72.8 ± 9.5
Male sex	97%	97%
Race
White	88%	77%
Black	12%	23%
American Indian or Alaska Native	1%	1%
Asian	1%	1%
Native Hawaiian or Pacific Islander	1%	1%
Hispanic	4%	5%
Diabetes mellitus	58%	69%
Myocardial infarction	13%	13%
Congestive heart failure	35%	38%
Peripheral vascular disease	28%	29%
Cerebrovascular disease	22%	23%
Liver disease	7%	11%
Cancer	29%	26%
Dementia	9%	7%
Depression	28%	34%
Posttraumatic stress disorder	11%	16%
Charlson Comorbidity Index	4.2 ± 2.9	5.1 ± 2.7
Zip Code median income	$46,398 [$38,381-$57,833]	$45,000 [$36,986-$56,612]
2-year risk of kidney failure
Median	1.2% [0.3%−3.9%]	2.9% [0.9%−8.2%]
Mean	3.7% ± 7.1%	7.1% ± 11.1%

Values for continuous variables given as mean ± SD or median [interquartile range]. Kidney failure risk could not be computed in 40,519 patients due to missing urine albumin and urine protein measurements.

Table 2.

Proportion of Patients Meeting Each Referral Indication

Indication for Referral	Patients With Referral Indication (n = 66,276)	Patients With Referral Indication Who Were Referred (n = 11,752)
eGFR <30 mL/min/1.73 m2	37,357 (56.4%)	9,040 (24.2%)
Rate of decline of eGFR ≥ 5 mL/min/1.73 m2 per year	35,155 (53.0%)	5,469 (15.6%)
Heavy proteinuria
Patients without diabetes	1,157 (1.7%)	376 (32.5%)
Patients with diabetes	2,852 (4.3%)	941 (33.0%)
Met >1 referral indication	9,724 (14.7%)	3,739 (38.5%)

Number (percentage) of patients meeting each indication for referral to nephrology. Percentages of patients with referral indication who were referred are calculated by row from the number of patients meeting that indication. As many patients met multiple indications on the same date, proportions sum to greater than 100%. Heavy proteinuria is defined as a urinary protein-creatinine ratio > 500 mg/g or urinary albumin-creatinine ratio > 300 mg/g in those without diabetes, and a urinary protein-creatinine ratio > 3,000 mg/g in those with diabetes. Abbreviation: eGFR, estimated glomerular filtration rate.

When applying a more stringent definition for rapid eGFR decline, 50,690 patients met an indication for referral, including 30.9% who met the rapid eGFR decline indication (Table S3). The median predicted risk of kidney failure was 0.6% (IQR, 0.2%−2.1%) in those meeting an indication for referral using the stringent rapid eGFR decline definition, and the rapid eGFR decline indication had a lowest predicted risk of kidney failure (Table S4), similar to the main analysis.

Referral Volumes Using Both Laboratory-Based Indications and Risk of Kidney Failure

Approximately 57.7% of patients with computable kidney failure risk had a value exceeding 1%. If this proportion were extrapolated to all patients with a referral indication, an estimated 38,229 of 66,276 patients would be referred to nephrology. The number of patients meeting risk thresholds ranging from 1% to 10% under the base case alone or in addition to laboratory-based indications for referral are shown in Figure 2. Estimated referral volumes with high and low imputed values for missing urine albumin and urine protein measurements are shown in Figure S2.

Figure 2.

Projected annual referral volumes using laboratory indications and kidney failure risk. The dotted line indicates the number of patients meeting Veterans Affairs/Department of Defense laboratory indications for referral. The dark circles indicate the projected number of patients meeting 2-year predicted risk of kidney failure thresholds ranging from 1% to 10%. The lighter circles indicate the projected number of patients meeting both laboratory indications for referral and kidney failure risk thresholds. Projections are extrapolated from patients with available urine albumin or urine protein measurements. Abbreviation: KFRE, Kidney Failure Risk Equation.

Referral Volumes Using Risk of Kidney Failure Alone

We estimated that 72,948 (20.1%) patients would meet a 1% 2-year risk threshold, comparable to the number of patients meeting laboratory-based indications. The median 2-year predicted risk of kidney failure among all patients meeting this threshold was 2.3% (IQR, 1.4%−4.6%). If the threshold for referral were increased to 2%, an estimated 41,101 patients would be identified. The estimated referral volumes with high and low imputed values for missing urine albumin and urine protein measurements are shown in Figure S3.

Discussion

In this study of CKD care in the VA health care system, we found that the volume of new outpatient nephrology referrals would more than double if all patients meeting potential indications for nephrology referral in VA/DoD CKD guidelines had been referred. If a minimum kidney failure risk of 1% over 2 years were applied to all new patients meeting laboratory-based potential referral indications, the number of patients targeted for nephrology referral would be reduced from 66,276 to an estimated 38,229 patients, a 42.3% reduction. Alternatively, referral based only on predicted risk would result in a similar number of patients identified for referral at a predicted risk threshold of 2% or higher. The application of more stringent risk thresholds would identify progressively fewer patients. The findings illustrate that referral of all patients meeting potential indications in VA/DoD CKD guidelines based solely on laboratory data could result in impractically large referral volumes, which would likely increase wait times without an expansion of nephrology resources.

We also found that, on average, patients who were referred for nephrology subspecialty care had a higher risk of kidney failure than those who were not referred. These findings confirm that primary care providers are generally able to identify patients at higher risk of progression to kidney failure, though the large overlap in risk profiles of the referred and unreferred populations suggests that a targeted approach could be used to improve consistency.

These findings also suggest that it may not currently be feasible to care for all patients who meet the guideline-suggested laboratory-based indications for nephrology referral. If a 5-year risk threshold of 3% alone were adopted, as is currently used in Manitoba, Canada,¹⁰ with perfect adherence an estimated 74,714 patients per year could overwhelm nephrology services. Indeed, referring all patients who meet the criteria based solely on laboratory thresholds may not represent best care. VA/DoD guidelines recommend shared decision-making to determine whether nephrology care is consistent with the values and preferences of individual patients,¹ recognizing that referral may not be appropriate in some clinical scenarios. For example, older adults with advanced chronic illnesses, in exploring their own goals of care, may decide that they would like to limit some referrals and interventions. Similarly, patients with severe dementia may be managed more conservatively than detailed in current guidelines.

Implementing decision support tools may increase referrals by alerting primary care providers of referral indications at the point of care. Although it is unlikely that providers would refer all patients who meet a potential indication, it is important for policymakers to understand the possible impact of disseminating guidelines through decision support tools, as well as the effects on wait times and volume of referrals. Prioritizing referrals based on the risk of adverse outcomes has been proposed by guideline organizations and evaluated in smaller studies in non-US health care systems,^2,16,17 but evidence that this approach improves outcomes is not yet available.

Health systems must make decisions on the appropriate allocation of nephrology resources based on their unique patient populations, weighing the potential benefits of trying to efficiently target referrals against the possible drawbacks. Barriers to referral may result in delayed medical assessment and treatment for patients with less-common presentations or CKD complications that are not reflected in risk of kidney failure. As our findings illustrate, targeting referrals based on the KFRE could result in fewer referrals of patients with rapid eGFR decline. The identification of individuals with rapid eGFR in electronic health records and the value of nephrology care in this population merit further study.

Although current VA/DoD guidelines endorse the use of risk prediction tools in the management of patients with CKD, policymakers felt there was insufficient evidence to recommend for or against the use of a specific equation or threshold for action. Using the data generated in this study, health care system leaders can estimate the effect of specific risk thresholds on referral volume. For example, a threshold of 1% 2-year kidney failure risk in addition to meeting one or more laboratory-based indications would reduce the number of potential new referrals by half. A similar threshold system has been instituted in one Canadian province, with a resultant reduction in patient wait times.¹⁰ Additional options for population management include triaging lower-risk patients to electronic consultation for initial visits or to advance-practice providers for follow-up evaluation. Higher thresholds could be used to identify patients for more intensive interdisciplinary care and dialysis access creation referral. Risk-based population management using the KFRE would only be possible with increased efforts to implement routine UACR measurements in patients with CKD.

Our study has several strengths. VA is a national, integrated health care system that cares for a large population of patients with CKD. Our study simulated referral decisions with laboratory data available at the point of care. We were able to capture new nephrology referrals in addition to nephrology visits. The study applied a validated risk prediction tool, the KFRE, which can be adapted for use into electronic clinical decision support tools. Although we were not able to account for all care outside of the VA, sensitivity analyses limited to Medicare-eligible patients suggested similar results.

Our study also has limitations. We were unable to determine the reason(s) for nephrology referral. Although preventing kidney disease progression is a key objective of nephrology care, other important referral indications include management of CKD complications such as anemia and mineral bone disease, medication management, and cardiovascular risk reduction. These indications can be difficult to assess at a large scale in electronic health record data. We could not determine the extent to which increased referral volumes would affect wait times or clinical outcomes. Our estimates of referral volume using predicted risk are based on extrapolation from patients who had urine albumin or urine protein measurements, and missingness is likely not random. Lack of urine albumin and urine protein measurement is common in at-risk populations and is an important hurdle that decision support implementers will need to overcome. To operationalize risk-based triage, efforts to increase urine albumin or protein measurement will be necessary. Finally, these analyses may not be generalizable to other health care systems; our cohort contained few women, and equivalent inputs into the KFRE predict lower risk of kidney failure in women than in men.

In conclusion, a significant proportion of patients identified by laboratory-based indications for nephrology referral have a predicted risk of kidney failure less than 1%. A referral system based on a 2-year kidney failure risk exceeding 1% would identify a similar number of patients while targeting those with higher risk for kidney failure. These findings may inform clinical decision support development to target nephrology referrals to patients most likely to benefit.

Supplementary Material

Supplement

Figure S1: Timeline of data acquisition.

Figure S2: Projected referral-eligible volumes using combination of potential referral indications and predicted risk thresholds with high and low estimates.

Figure S3: Projected referral-eligible volumes using predicted risk thresholds with high and low estimates.

Table S1: Laboratory-based indications for referral to nephrology.

Table S2: Proportion of patients with computable kidney failure risk meeting each referral indication, predicted kidney failure risk, and frequency of progression to kidney failure.

Table S3: Proportion of patients meeting each referral indication using the more stringent definition for decline in eGFR.

Table S4: Proportion of patients with computable kidney failure risk meeting each referral indication, predicted kidney failure risk, and frequency of progression to kidney failure using the more stringent definition for decline in eGFR.

PLAIN-LANGUAGE SUMMARY.

Kidney disease guidelines for nephrology referral are transitioning from recommending referral based on laboratory value thresholds to using risk of kidney failure to guide referrals, but the impact on health care systems is not known. Using data from the Veterans Health Administration, we estimated the potential volume of nephrology referrals based on laboratory data, the estimated risk of kidney failure, or a combination of both. We report that referral based on a 2-year risk of kidney failure of 1% or higher would result in a similar number of referrals as current laboratory-based guidelines but would identify patients at higher median risk for kidney failure.