Integrating Clinical Decision Support Systems in CKD Management

Introduction

Chronic kidney disease (CKD) affects 10% to 14% of the population, contributing to significant morbidity and mortality worldwide.^1,2 Although there are therapies that reduce CKD-associated morbidity and mortality, CKD care is mired by critical gaps.³ Albuminuria testing is underutilized; and key therapies proven to slow CKD progression, such as sodium-glucose cotransporter 2 inhibitors (SGLT2i), mineralocorticoid receptor antagonists (MRAs), glucagon-like peptide-1 receptor agonists (GLP1-RAs), alongside renin-angiotensin-aldosterone system inhibitors (RAASi), remain underprescribed.^3,4 For example, RAASi are used in only 55% of patients with an indication and SGLT2i are prescribed in just 11.9%.^4,5

Integrating clinical decision support (CDS) systems within electronic health records offers one strategy to close these gaps. However, many CDS systems are poorly received because of alert fatigue and workflow disruption. To address these challenges, the Kidney Precision Medicine Centre of Excellence at Johns Hopkins University deployed Patient Insight (PI), a noninterruptive, patient-specific flow sheet into the ambulatory electronic health record view of nephrology providers in January 2022.

PI displays key CKD-relevant information including the following: 2-year and 5-year kidney failure risk (estimated using the Kidney Failure Risk Equation⁶ for patients with estimated glomerular filtration rate [eGFR] < 60 ml/min per 1.73 m²), 2-year cardiovascular event risk (for patients with eGFR < 30 ml/min per 1.73 m²), all eGFR values (calculated using CKD-Epidemiology Collaboration 2021 equation), albuminuria values, kidney-related events (biopsies, emergency department visits, and hospitalizations), and medication use over time (RAASi, SGLT2i, MRAs, and GLP1-RAs; Supplementary Figure S1). Unlike traditional alerts, PI was designed to reduce workflow disruption by integrating into existing clinical practice patterns. Methodological details are provided in Supplementary Methods S1–S7.

Results

In the nephrology group, the mean age was 65 years, 54% were male, 47% had diabetes, and 46% had advanced CKD. In the endocrinology comparison group, the mean age was 68 years, 33% to 35% were male, 54% had diabetes, and 9% had advanced CKD (Table 1).

Table 1.

Baseline characteristics of participants before Patient Insight

Characteristics	Nephrology		Endocrinology
	Before PI	After PI	Before PI	After PI
Total number of patients	4526	4918	2325	2823
Number of encounters	12893	14570	7741	9345
Patient characteristics
Age, yrs (mean [SD])	64.6 (14.7)	65.0 (14.9)	67.8 (12.2)	68.7 (11.9)
Male sex	2427 (53.6%)	2597 (52.8%)	823 (35.4%)	930 (32.9%)
Race
White	2083 (46.0%)	2287 (46.5%)	1320 (56.8%)	1676 (59.4%)
Black	2065 (45.6%)	2183 (44.4%)	839 (36.1%)	938 (33.2%)
Other	378 (8.4%)	448 (9.1%)	166 (7.1%)	209 (7.4%)
eGFRa, ml/min/m2	32 (14)	33 (14)	46 (13)	47 (12)
UACRb, mg/g	73 (14, 487)	67 (14, 423)	16 (12, 67)	19 (12, 67)
Diabetes (Yes)	2109 (46.6%)	2148 (43.7%)	1265 (54.4%)	1537 (54.4%)
Advanced CKDc	2091 (46.2%)	2167 (44.1%)	237 (10.2%)	256 (9.1%)

eGFR, estimated glomerular filtration rate; KFRE, Kidney Failure Risk Equation; PI, Patient Insight; UACR, urine albumin-to-creatinine ratio.

UACR: to convert to mg/mmol, multiply by 0.113. UACR measurements were not available in ∼10% of the cohort (Supplementarty Table S1).

^aMean (SD).

^bMedian (Q1, Q3).

^cDefined as eGFR < 30 ml/min or with a 2-year KFRE > 10% within 1 yr before visit (or at the time of the visit).

On average, 29 out of 40 nephrology providers accessed the PI longitudinal viewer ≥1/yr. In 2022, 472 unique patients (22%) were viewed, and in 2023, 502 (22%) were accessed. The proportion of outpatient nephrology notes that included the Kidney Failure Risk Equation smart phrase increased from <10% in January 2022 to approximately 25% in December 2023 (Supplementary Figure S2).

Following PI deployment, prescription rates for kidney-protective therapies showed differential changes between nephrology and endocrinology clinics. The most pronounced difference was observed with SGLT2i. In nephrology, SGLT2i prescriptions increased from 12.9% to 26.2% (13.3% absolute increase); whereas in endocrinology, rates increased from 13.3% to 20.7% (7.4% absolute increase). This represents a 5.9% greater improvement in SGLT2i prescribing associated with PI (crude P < 0.001, adjusted P < 0.001; Figure 1).

Figure 1.

Prescription rates before (2020–2021) and after (2022–2023) Patient Insight (PI) among nephrology and endocrinology clinics. Note that the Y-axis is on a different scale for each medication group. This was done intentionally to highlight the change in prescription rates within each group. GLP1-RA, glucagon-like peptide-1 receptor agonist; MRA, mineralocorticoid receptor antagonist; PI, Patient Insight; SGLT2, sodium-glucose cotransporter-2 inhibitor.

For other medication classes, differences between groups were smaller. RAASi prescriptions increased from 67.5% to 69.5% in nephrology and from 60.0% to 61.3% in endocrinology (0.7% difference, crude P = 0.64, adjusted P = 0.40). MRA prescriptions increased from 10.2% to 13.8% in nephrology and from 11.5% to 14.8% in endocrinology (0.3% difference, crude P = 0.85, adjusted P = 0.50). GLP1-RA prescriptions increased from 9.7% to 14.6% in nephrology and from 19.5% to 25.7% in endocrinology (1.3% smaller increase in nephrology, crude P = 0.29, adjusted P = 0.83).

Clinical care practice measurements changed across multiple domains. Patients had more frequent nephrology encounters, with 45.7% of patients with advanced CKD having ≥4 visits annually compared with 41.8% before PI. Laboratory monitoring improved, with eGFR measurements increasing from 86.0% to 89.7% in patients with advanced CKD and from 64.8% to 70.5% in patients with nonadvanced CKD. Urine albumin-to-creatinine ratio testing improved, with the proportion of patients lacking urine albumin-to-creatinine ratio measurements decreasing from 11.5% to 9.4% among those with advanced CKD and from 13.3% to 7.9% among those with nonadvanced CKD (Supplementary Table S1).

Discussion

This study demonstrates that integrating a noninterruptive, patient-specific CDS tool was associated with meaningful improvements in evidence-based prescribing patterns, particularly for SGLT2i. The pronounced effect on SGLT2i prescribing likely reflects several factors. Although the 2022 Kidney Disease: Improving Global Outcomes guideline recommending SGLT2i for patients with diabetes and CKD coincided with our postdeployment period, the substantially larger increase in nephrology (13.3 percentage points) compared with endocrinology (7.4 percentage points) may reflect differences in how guideline recommendations were implemented in practice.⁷ During this period, PI was introduced with focused education, primarily to increase awareness of its availability. This tool may have supported the identification of appropriate candidates and facilitated the application of guideline recommendations at the point of care. However, the observational design precludes attribution of these changes to PI alone.

RAASi prescribing increased modestly, likely because this medication class was already standard practice in nephrology with baseline rates of 67.5 %, leaving limited room for improvement. MRA prescribing showed minimal change, likely reflecting the complexity of prescribing decisions requiring careful consideration of potassium levels and kidney function. Potassium values were not displayed within the PI tool, which may have limited its utility for MRA-related decision making and represents a potential area for future refinement. In addition, GLP1-RA prescribing rates were higher in endocrinology (1.3% difference; crude P = 0.29, adjusted P = 0.83), reflecting several factors. Endocrinologists favor GLP1-RAs for their superior glycated hemoglobin reduction and diabetes management benefits, which is further supported by their dedicated GLP1-RA clinic at our institution that may have facilitated higher prescribing rates.⁸ In addition, the timing of kidney-specific evidence on GLP1-RA, the FLOW trial, was published in 2024, after our observation period.⁹

The study limitations include an observational design that prevents causal inference; a single institution setting with robust nephrology infrastructure and extensive electronic health record integration that limits generalizability; inability to evaluate medication adherence or patient outcomes; and potential confounding from unmeasured factors, including provider preferences and formulary changes. PI use was modest, and we were unable to link provider- or patient-level tool use to prescribing decisions, limiting our ability to assess the direct effect of PI or compare users and nonusers. Our reliance on electronic health record data introduces potential information bias because of the incomplete capture of International Classification of Diseases codes, laboratory values, or medication history. In addition, the predeployment period (2020–2021) coincided with COVID-19 pandemic disruptions, though our endocrinology comparison group experienced similar effects. We could not determine which provider specialty ordered tests or prescribed medications; however, the differential SGLT2i increase (5.9 percentage points, crude P < 0.001, adjusted P < 0.001) suggests clinic-specific effects. The comparison of prescription rates between nephrology and endocrinology included all patients and was not limited to patients with an indication. In addition, the endocrinology cohort had less severe kidney disease, including higher eGFR and lower albuminuria, which may have reduced recognition of kidney-specific indications for SGLT2i during the study period. At the same time, providers may be less likely to initiate SGLT2i in patients with more advanced CKD because of concerns regarding low eGFR. Therefore, the overall effect of baseline CKD severity on prescribing patterns is difficult to predict.

Our findings demonstrate that well-designed, nonintrusive CDS systems might improve clinical practice patterns. Future studies should evaluate whether CDS-driven practice changes improve patient outcomes, including kidney function decline, cardiovascular events, and mortality. With continued refinement and broader applications, these tools have the potential to transform kidney disease management across nephrology, primary care, and specialty settings.

Disclosure

All the authors declared no competing interests.

Supplementary Material

Supplementary File (PDF)

Supplementary Methods. Supplementary References. Figure S1. Patient Insight CDS tool. Figure S2. Utilization of the KFRE smart phrase in outpatient nephrology clinic notes. Table S1. Clinical care practice measurements in the nephrology cohort before and after PI deployment. STROBE checklist.