Gary E. Rosenthal, MD

 Optimizing care for patients with advanced kidney disease requires close collaboration between primary care physicians (PCPs) and nephrologists. Factors associated with PCP referral to nephrology were assessed in patients with estimated glomerular filtration rates (eGFRs) less than 30 mL/min/1.73 m

 Electronic health record review at an integrated health care network.

 Factors associated with referral status were identified using Fisher’s exact tests, 

 tests, and multivariable logistic regression.

 Of 133,913 patients regularly seeing PCPs between October 2017 and September 2019, 1119 had a final eGFR less than 30 mL/min/1.73 m

 and were not on renal replacement therapy. Care was provided by 185 PCPs (61 practices). Analyses were restricted to the 97.1% (n = 1087) of patients who were African American or European American. Of these, 54.6% had not been referred to nephrology. Nonreferred patients had higher numbers of PCP visits (

 = .004). In contrast, referred patients were younger, were more often African American, and had PCPs at the academic medical center (all 

 < .0001). Referred patients had more complex medical histories with higher Charlson Comorbidity Index scores, more hospitalizations, and greater numbers of inpatient days (all 

 < .0001). Analyses restricted to patients with serum creatinine concentration of at least 2 mg/dL yielded similar results. Age, number of hospitalizations, ancestry, academic physician, diabetic end-organ damage, peripheral vascular disease, and tumor status were independent predictors of nephrology referral.

 Impediments to appropriately timed nephrology referrals persist in patients with high likelihoods of progression to end-stage kidney disease. Improved access to nephrology care should be rapidly addressed to meet targets in the 2019 Executive Order on Advancing American Kidney Health.

Timely nephrology referral improves outcomes in chronic kidney disease and assists with meeting targets set in the 2019 Executive Order on Advancing American Kidney Health. Factors associated with specialty referral were assessed in patients with advanced nephropathy at a large integrated health system.

The majority of patients were not referred to nephrology.

Referred patients had higher Charlson Comorbidity Index scores and more hospitalizations and inpatient days.

Healthier patients, potentially more suitable for kidney transplant or home dialysis, were less likely to be referred.

Impediments to appropriately timed nephrology referrals persist in patients likely to progress to end-stage kidney disease.

Optimizing outcomes for patients with advanced chronic kidney disease (CKD) can best be achieved with appropriately timed nephrology consultation.

 Nephrologists attempt to slow the progression of CKD, educate patients regarding the optimal choice of renal replacement therapy (including preemptive kidney transplantation), assess medication dosage and safety, and assist primary care providers (PCPs) with management of hypertension, cardiovascular disease risk factors, anemia, and bone and mineral health. Patients receiving regular nephrology care prior to renal replacement therapy for treatment of end-stage kidney disease (ESKD) have slower rates of CKD progression,

 higher rates of functioning arteriovenous fistulae as initial hemodialysis vascular access, and fewer tunneled central vein catheters.

 Effects of earlier referral on mortality are less clear.

Barriers to recognition of advanced CKD likely include absence of symptoms and minimally elevated serum creatinine (SCr) concentrations, particularly in elderly and frail patients. More precise equations for computing the estimated glomerular filtration rate (eGFR) are widely available and simplify the detection of kidney disease.

 However, many physicians focus on the SCr concentration in lieu of eGFR assessment and presence or severity of proteinuria.

National efforts have been undertaken to ensure that PCPs recognize the presence of CKD at early stages in an attempt to slow disease progression.

 The 2019 Executive Order on Advancing American Kidney Health targets slowing of nephropathy progression and higher rates of kidney transplant and home-based dialysis therapy.

 As such, appropriately timed referral to nephrology remains critical for improving outcomes, increasing access to home dialysis, and performing preemptive and early kidney transplant.

 The frequencies with which PCPs affiliated with a Southeastern US academic health care network referred patients with advanced kidney disease to a nephrologist were assessed. Factors associated with the likelihood of referral were also analyzed in this older population.

The Wake Forest Baptist Health (WFBH) Network (WFBHN) is an integrated network. Review of the electronic health record (EHR) identified patients with a most recent eGFR less than 30 mL/min/1.73 m

 who had been seen by a PCP in the 2-year study period between October 1, 2017, and September 30, 2019. PCPs specializing in internal medicine, family medicine, geriatric medicine, and pediatrics were included and grouped as either academic or community-based according to the location of their primary department. Only actively treated patients, defined as those who had at least 2 visits with their PCP during the 2-year study period, were captured. Nephrology referrals were captured based on having seen a nephrologist or an electronic order for referral; waiting periods until visits did not affect outcomes. Main analyses were limited to African American and European American patients, who composed the vast majority of the sample, and included all patients with at least 1 eGFR during the study period and the data fields necessary to compute a Charlson Comorbidity Index (CCI) score.

 The Wake Forest School of Medicine (WFSM) Institutional Review Board approved this study.

Clinical and demographic information was extracted from the WFBH EHR (Epic), a single enterprise-wide platform supporting integrated clinical, billing, and ancillary applications. CCI scores were computed by the WFSM Clinical and Translational Science Institute using EHR data.

 All variables examined aside from the CCI score were extracted using the Epic EHR across the health system using medication, laboratory, and diagnosis definitions using standard ontologies (SNOMED; LOINC; 

International Classification of Diseases, Tenth Revision

; National Drug Code; RxNorm) for structured data available in the EHR at the time of the outcome visit. We did not use natural processing to mine/examine free-text entered data from the EHR. The SCr was measured using the modified Jaffe method, kinetic, calibrated to the isotope dilution mass spectrometry reference measurement (Beckman Coulter AU5822). The eGFR was computed using the CKD-Epidemiology Collaboration equation.

 Patient identifiers were removed from the data set prior to analysis.

Descriptive statistics, including means, SDs, medians, and interquartile ranges (IQRs) for continuous measures and frequencies and proportions for categorical data, were calculated for all study variables. Referred and nonreferred patient groups were compared using independent 

 tests for continuous data and Fisher’s exact tests for categorical data. To assess which factors might be associated with which subjects had (or had not) been referred, a logistic regression model to assess independent measures with referral status was created. This employed a mixed models analysis of variance approach with a random intercept component for each physician. This model included the covariates that were significantly different between referred and nonreferred groups in the overall analysis. A secondary analysis included only those patients whose most recent SCr was at least 2 mg/dL. SAS version 9.4 (SAS Institute) was used for all analyses, and 

 values less than .05 were considered statistically significant.

A total of 133,913 patients in the WFBHN had at least 2 visits with their PCP during the study period. Of these, 1434 had a most recent eGFR less than 30 mL/min/1.73 m

, at least 1 eGFR during the study period, and the data necessary to compute a CCI score. Among these, 315 patients were receiving renal replacement therapy and were excluded from analyses. The remaining 1119 patients with nephropathy had primary care provided by 185 unique PCPs in 61 practices. Because 97.1% (n = 1087) of the sample was African American or European American, analyses were restricted to these patients. Of these 1087 patients, 494 (45.4%) had been referred to nephrology and 593 (54.6%) had not been referred.

 displays demographic, clinical, and laboratory data in the referred and nonreferred patient groups. Patients in both groups saw their PCPs frequently during the 2-year study period, with a median of 6 PCP visits per patient. The number of PCP visits was statistically higher in the nonreferred group (

 = .0036), likely of minor clinical relevance, and the median age of the nonreferred patients was 5 years older than that of patients in the referred group (

 < .0001). The median (IQR) eGFR at baseline was similar in the nonreferred (30 [24-36] mL/min/1.73 m

 = .81). Baseline SCr was also similar in nonreferred (2.0 [1.7-2.5] mg/dL) and referred (1.9 [1.5-2.3] mg/dL) patients (

 = .59). In contrast, significant differences were seen in the final eGFR and SCr based on referral group. Referred patients ended the study period with 2 mL/min/1.73 m

 lower median eGFR and 0.2 mg/dL higher median SCr than those not referred (both 

 <.005). The median (IQR) time between the first PCP visit after October 1, 2017, and the outcome visit was 1.6 (1.3-1.8) years overall. There was little change in covariates during this interval; the median percentage change in all conditions assessed was 2.2% (data not shown).

Hospitalizations, numbers of inpatient days, and median CCI score were significantly higher in the referred patient group (all 

 < .0001). Hospital discharge to a skilled nursing facility was also higher in referred patients (

 = .03). Relative to European Americans, African Americans were more likely to have been referred (

 < .0001). PCPs grouped as academic were significantly more likely to refer their patients to nephrology than were community-based physicians (

 < .0001). Gender and Medicare/Medicaid insurance status were not significantly different irrespective of referral to nephrology. The prevalence of congestive heart failure, chronic obstructive pulmonary disease, cerebrovascular disease, diabetes with end-organ damage, myocardial infarction, tumors, and peripheral vascular disease was significantly higher among referred patients. In contrast, no significant differences in nephrology referral were seen based on prescription of renin-angiotensin-aldosterone system antagonists or among patients with dementia, AIDS, hemiplegia, rheumatic disease, or peptic ulcer disease. The median fall in eGFR among referred patients was 6 mL/min/1.73 m

 during the 2-year study period, compared with 5 mL/min/1.73 m

A laboratory-based dipstick urinalysis was performed in 30.7% (182 of 593) of nonreferred and 69.2% (342 of 494) of referred patients (

 < .001). At least 1 quantitative spot urine protein measurement (either urine albumin-creatinine ratio [uACR] or urine protein-creatinine ratio [uPCR]) was performed in 25.6% (152 of 593) of nonreferred patients and 32.6% (160 of 494) of referred patients (

 = .015). The most recent median (IQR) uACR was 104 (32-536) mg/g in nonreferred patients (22.9% [n = 136] had this measure) vs 209 (40-848) mg/g in referred patients (29.8% [n = 147] had this measure) (

 = .084 between groups). Only 45 patients had a uPCR (25 nonreferred and 20 referred), without significant between-group differences (

 presents results of a multivariable mixed effects logistic regression model analyzing referral status as the outcome. Relationships between covariates that differed significantly between groups in Table 1 were assessed, based on whether or not patients had been referred. A random intercept model, with a clustering factor for PCP, adjusted for the effect of each PCP. SCr was the only variable excluded from the model, as it is a component of the eGFR and had a significant correlation with that measure (

 < .0001). Only initial eGFR was included as a measure of baseline kidney function. In the fully adjusted model, age (

 = .0003; those with increasing age less likely to be referred), number of hospitalizations (

 = .0066; those with increasing numbers more likely to be referred), ancestry (

 < .0001; African Americans more likely to be referred), academic center PCP (

 < .0001; academic group physicians more likely to refer), and presence of diabetic end-organ damage (

 = .0041), peripheral vascular disease (

 = .016) increased the odds of referral.

The baseline SCr was assessed in all 1087 patients with an eGFR less than 30 mL/min/1.73 m

. In all, 1 (0.1%) had a baseline SCr less than 1.5 mg/dL; 31 (2.8%), between 1.5 and less than 1.6 mg/dL; 98 (9.0%), between 1.6 and less than 1.7 mg/dL; 173 (15.9%), between 1.7 and less than 1.8 mg/dL; 255 (23.5%), between 1.8 and less than 1.9 mg/dL; 320 (29.4%), between 1.9 and less than 2.0 mg/dL; and 209 (19.2%), greater than or equal to 2.0 mg/dL. 

 displays the likelihood of nephrology referral in the 767 (70.6%) patients with a final SCr of at least 2 mg/dL. In this group, 51.2% of patients (393 of 767) were referred and 48.8% (374 of 767) were not referred. As in the full sample, younger patients, African Americans, those cared for by PCPs grouped as academic, those with greater numbers of hospitalizations and inpatient days, and those with higher median CCI scores were significantly more likely to have been referred (all 

 < .001). Although final kidney function was significantly lower in patients in the referred group, between-group differences were only 2 mL/min/1.73 m

Although the Kidney Disease: Improving Global Outcomes (KDIGO) CKD working group recommends that patients with an eGFR less than 30 mL/min/1.73 m

 this study’s results from an academic health care system reveal that only 45.4% of such patients who are actively seeing their PCPs are referred to nephrology. These data are contemporary and reflect the current era of hospitalist services. Men and women were equally likely to have been referred, as were patients with and without primary Medicaid/Medicare insurance. Relative to referred patients with advanced nephropathy, nonreferred patients were significantly older and had fewer hospitalizations, fewer inpatient days, and lower CCI scores. Thus, nonreferred patients appear to have less severe nonkidney organ system disease. Relative to European Americans, African Americans were more likely to have been referred, as were patients treated by PCPs grouped as academic vs in the community. It is possible that some older patients may be more resistant to seeing specialists; hence, a subset of nonreferrals could relate to patient choice.

Results were somewhat surprising. We considered that PCPs might have been less likely to refer patients with advanced CKD who had more severe comorbidities or diseases making them less suitable for renal replacement therapy or precluding candidacy for kidney transplantation. Access to health care appears to have been excellent in both the referred and nonreferred patient groups, with a median 6 PCP visits during the 2-year study period. Both groups had similar frequencies of receipt of renin-angiotensin-aldosterone system antagonists (

 = .45), and significant differences in the prevalence of dementia or AIDS were not observed. Although lack of nephrologists in a region could reduce referrals, it does not appear to have contributed to our findings. The WFBHN consists of 20 nephrologists, 6 advanced practitioners, and 6 nephrology fellows; 16 other nephrologists also practice in the region.

 < .0001), discharge from hospital to skilled nursing facility (

 = .0059), congestive heart failure (

 < .0001), chronic obstructive pulmonary disease (

 < .0001), diabetes with end-organ damage (

 = .017), peripheral vascular disease (

 < .0001) were present in referred patients with advanced nephropathy. We could not determine whether nephrologists were initially consulted during inpatient hospitalizations or as outpatients. Specialty care was significantly more likely to have been requested for individuals with greater numbers of comorbidities and who spent more time as inpatients.

To address the possibility that PCPs might place more weight on the SCr than the eGFR, a subsequent analysis was restricted to the 767 patients with a final SCr of at least 2 mg/dL. Results revealed that only 51.2% of these patients had been referred to nephrology. Regardless of a very low eGFR or markedly elevated SCr, patients with advanced CKD are often not being referred to nephrology. Physicians struggle to determine which patients with a low eGFR are most likely to progress rapidly to ESKD. Risk estimators have been developed that include the degree of proteinuria and plasma biomarkers of kidney injury.

 However, we note that a routine urinalysis was performed in only 30.7% of nonreferred patients who had an eGFR less than 30 mL/min/1.73 m

, and fewer PCP-managed patients had a uACR or uPCR to quantify proteinuria.

Prior studies have assessed the likelihood and timing of nephrology referral in patients with CKD. In 2018, the US Renal Data System reported that 35.4% of incident patients with ESKD received little or no predialysis nephrology care.

 Koraishy et al assessed 2170 African American and European American patients in St Louis, Missouri, between 2008 and 2015.

 The majority of patients with CKD were not referred to nephrology; however, those with faster rates of decline in eGFR and African Americans were more likely to be referred. Winkelmayer et al evaluated 3014 patients and defined “early referral” to nephrology as more than 90 days before the initiation of renal replacement therapy.

 Patients with comorbidities such as coronary artery disease, malignancy, hypertension, and diabetes were more likely to be referred early, similar to the higher numbers of comorbidities in referred patients from the present report. Earlier referral appears to slow the rate of decline in eGFR

 and increase rates of arteriovenous fistula placement for hemodialysis vascular access

 and may reduce postdialysis mortality, although results are controversial.

Other studies have reached opposing conclusions. A retrospective report assessed reasons for delayed nephrology referral.

 As in the present study, nonreferred patients were typically older. However, women, non-White patients, and those with a CCI score greater than 4 were less likely to be referred. A follow-up questionnaire revealed that 62% of PCPs were unfamiliar with Kidney Disease Outcomes Quality Initiative recommendations for referral; 87% used the SCr and 31.8% used stage of CKD for referral. PCPs reported that limited life expectancy, age older than 75 years, patient nonadherence, and dialysis refusal were major factors influencing their decisions on specialty referral.

 A national cohort study in patients with eGFR less than 30 mL/min/1.73 m

 from the Department of Veterans Affairs found that older age, heart failure, dementia, and depression were associated with a lower likelihood of nephrology referral.

 Surveys cite difficulty with PCPs establishing working partnerships with nephrologists, lack of timely and adequate information exchange, unclear roles in management, and lack of access as barriers to specialty referral.

 Finally, a qualitative study assessing when PCPs were likely to refer patients to nephrology found fewer referrals in older patients and those with moderate dementia and perceived shorter life expectancy. Factors that led to increased nephrology referral included younger age, female sex, non-White race, expectations that nephrologists would have conversations regarding goals of care, and when a PCP experienced anxiety about uncertain diagnosis.

 Based on these studies’ findings, we hypothesized that patients with more complex medical histories and shorter life expectancies might have been less likely to be referred for nephrology care. However, we found that healthier and older patients were significantly less likely to have been referred. This suggests that PCPs either underestimate severe CKD in patients with fewer symptoms or prefer to manage less symptomatic patients on their own.

Our study has many strengths, as well as limitations. Results are contemporary, from the era of hospital medicine, and the number of patients and providers was large. This study encompassed all PCPs treating adult patients in an integrated health care system with a diversity of patients and providers in academic and community practices. Data were uniformly collected using a single EHR. However, data were retrospective, observational (subject to selection bias and residual confounding), limited to 2 ancestral populations, EHR based, and from a single center. As such, they require replication and may not be generalizable to American Indians, Asians, Hispanics, and Pacific Islanders. We captured referrals to nephrology, regardless of whether or not patients were seen. We were unable to determine whether differences in physician age or years in practice were associated with likelihood of referral. It is possible that our methods missed a small number of patients referred to nephrology outside the EHR and that some PCPs may have been in the process of rechecking the eGFR prior to requesting referral. It is also possible that some patients could have had acute kidney injury (AKI). However, 280 and 290 patients in the nonreferred and referred groups, respectively, had at least 2 eGFRs at least 90 days apart that were 30 mL/min/1.73 m

 or less, and baseline eGFR in the full sample was 29 mL/min/1.73 m

. This suggests that AKI was infrequent. Finally, KDIGO guidelines include 9 potential indications for referral to nephrology

 and Tangri et al developed a kidney failure risk equation for likelihood of progression to ESKD

; KDIGO and the Tangri equation both include proteinuria and other factors, whereas our report focused solely on eGFR and did not assess proteinuria because of the low frequency of testing.

Although we found that medically complex patients were more often referred, we were unable to determine when specialty referrals were made. We speculate that referrals may have been made when patients were hospitalized, perhaps by hospitalists or non-PCP providers. We do not know how many patients were referred by their PCP as an outpatient. PCPs at the academic medical center were significantly more likely to refer their patients than community-based PCPs. In fact, community-based PCPs referred only 10% of their patients with advanced nephropathy to nephrologists. Finally, African Americans were more likely to be referred to nephrology than European Americans; this supports relatively good access to health care for this population.

Patients with lower CCI scores were significantly less likely to be referred to a nephrologist. Earlier referrals could maximize their likelihood of preemptive kidney transplant, performance of home dialysis, and placement of an arteriovenous fistula for hemodialysis vascular access. Although they were older, many nonreferred patients are likely equivalent or better candidates for these interventions than referred patients. Conservative management, care coordination, and understanding the underlying disease process are also important in our older population. Despite ongoing educational efforts, PCPs may not consider the eGFR on chemistry profiles. However, similar results were seen in an analysis restricted to patients with SCr at least 2 mg/dL (70.6% of the sample). Lower rates of timely referral among healthier patients with CKD may cause patients who were initially candidates for kidney transplant or home dialysis to develop complications that reduce their independence and could make them more likely to require in-center hemodialysis.

To improve outcomes in patients with CKD, slow nephropathy progression, maximize the likelihood of kidney transplant, and increase use of home dialysis modalities, we propose considering clinical decision support interventions in the EHR that allow PCPs to plan earlier referral to nephrology based on structured data elements similar to those in this report. Risk prediction algorithms based on plasma biomarker concentrations and EHR data are also likely to be useful in stratifying patients with CKD who are at greatest risk for nephropathy progression.

 These will help inform PCPs when to refer patients to nephrology.

 To meet targets in the 2019 Advancing American Kidney Health Executive Order, PCPs should consider early nephrology consultation. Despite advocacy efforts by the American Society of Nephrology and National Kidney Foundation, the present results suggest that existing efforts to educate primary care providers about CKD referral have had limited success. Novel efforts are required.

The authors acknowledge data-extraction services provided by the Center for Biomedical Informatics Data Team of the Wake Forest Clinical and Translational Science Institute supported by the National Center for Advancing Translational Sciences and the National Institutes of Health (UL1TR001420).

 Department of Internal Medicine, Section on General Internal Medicine (AD, RB, WYR, GER), Division of Public Health Sciences, Department of Implementation Science (AD), Division of Public Health Sciences, Department on Biostatistics and Data Science (GBR), and Department of Internal Medicine, Section on Nephrology (CEG, BIF), Wake Forest School of Medicine, Winston-Salem, NC.

 Department of Internal Medicine, Wake Forest School of Medicine, and National Center for Advancing Translational Sciences and the National Institutes of Health (UL1TR001420).

 Dr Freedman is a member of the Wake Forest University Health Sciences Dialysis Board of Directors and is employed by Wake Forest School of Medicine. The remaining authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

 Concept and design (AD, WYR, GER, BIF); acquisition of data (AD, RB, BIF); analysis and interpretation of data (AD, RB, GBR, WYR, CEG, GER, BIF); drafting of the manuscript (AD, RB, GBR, CEG, BIF); critical revision of the manuscript for important intellectual content (AD, RB, GBR, WYR, GER, CEG); statistical analysis (GBR); provision of patients or study materials (WYR); administrative, technical, or logistic support (AD, BIF); and supervision (AD, BIF).

 Barry I. Freedman, MD, Section on Nephrology, Wake Forest School of Medicine, Medical Center Blvd, Winston-Salem, NC 27157-1053. Email: bfreedma@wakehealth.edu.

1. Levin A, Stevens PE, Bilous RW, et al. Kidney Disease: Improving Global Outcomes (KDIGO) CKD Work Group: KDIGO 2012 clinical practice guideline for the evaluation and management of chronic kidney disease

2. Chen SC, Chang JM, Chou MC, et al. Slowing renal function decline in chronic kidney disease patients after nephrology referral. 

. 2008;13(8):730-736. doi:10.1111/j.1440-1797.2008.01023.x

3. Smart NA, Dieberg G, Ladhani M, Titus T. Early referral to specialist nephrology services for preventing the progression to end-stage kidney disease. 

. 2014;(6):CD007333. doi:10.1002/14651858.CD007333.pub2

4. Liu P, Quinn RR, Oliver MJ, et al. Association between duration of predialysis care and mortality after dialysis start. 

. 2018;13(6):893-899. doi:10.2215/CJN.11951017

5. Levey AS, Becker C, Inker LA. Glomerular filtration rate and albuminuria for detection and staging of acute and chronic kidney disease in adults: a systematic review. 

. 2015;313(8):837-846. doi:10.1001/jama.2015.0602

6. Primary care practitioners. National Kidney Foundation. Accessed February 25, 2020. https://www.kidney.org/professionals/physicians/pcp

7. Executive order on advancing American kidney health. White House. July 10, 2019. Accessed February 25, 2020. https://www.whitehouse.gov/presidential-actions/executive-order-advancing-american-kidney-health/

8. Gillespie BW, Morgenstern H, Hedgeman E, et al. Nephrology care prior to end-stage renal disease and outcomes among new ESRD patients in the USA. 

. 2015;8(6):772-780. doi:10.1093/ckj/sfv103

9. Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. 

. 1987;40(5):373-383. doi:10.1016/0021-9681(87)90171-8

10. Quan H, Li B, Couris CM, et al. Updating and validating the Charlson Comorbidity Index and score for risk adjustment in hospital discharge abstracts using data from 6 countries. 

. 2011;173(6):676-682. doi:10.1093/aje/kwq433

11. Levey AS, Stevens LA, Schmid CH, et al; CKD-EPI (Chronic Kidney Disease Epidemiology Collaboration). A new equation to estimate glomerular filtration rate. 

. 2009;150(9):604-612. doi:10.7326/0003-4819-150-9-200905050-00006

12. Nadkarni GN, Chauhan K, Verghese DA, et al. Plasma biomarkers are associated with renal outcomes in individuals with APOL1 risk variants. 

. 2018;93(6):1409-1416. doi:10.1016/j.kint.2018.01.026

13. Saran R, Robinson B, Abbott KC, et al. US Renal Data System 2018 annual data report: epidemiology of kidney disease in the United States. 

. 2019;73(3)(suppl 1):A7-A8. doi:10.1053/j.ajkd.2019.01.001

14. Koraishy FM, Hooks-Anderson D, Salas J, Scherrer JF. Rate of renal function decline, race and referral to nephrology in a large cohort of primary care patients. 

. 2017;34(4):416-422. doi:10.1093/fampra/cmx012

15. Winkelmayer WC, Glynn RJ, Leven R, Owen WF Jr, Avorn J. Determinants of delayed nephrologist referral in patients with chronic kidney disease. 

. 2001;38(6):1178-1184. doi:10.1053/ajkd.2001.29207

16. Astor BC, Eustace JA, Powe NR, et al. Timing of nephrologist referral and arteriovenous access use: the CHOICE Study. 

. 2001;38(3):494-501. doi:10.1053/ajkd.2001.26833

17. Stack AG. Impact of timing of nephrology referral and pre-ESRD care on mortality risk among new ESRD patients in the United States. 

. 2003;41(2):310-318. doi:10.1053/ajkd.2003.50038

18. Navaneethan SD, Kandula P, Jeevanantham V, Nally JV Jr, Liebman SE. Referral patterns of primary care physicians for chronic kidney disease in general population and geriatric patients. 

. 2010;73(4):260-267. doi:10.5414/cnp73260

19. Fung E, Chang TI, Chertow GM, Thomas IC, Asch SM, Tamura MK. Receipt of nephrology care and clinical outcomes among veterans with advanced CKD. 

. 2017;70(5):705-714. doi:10.1053/j.ajkd.2017.06.025

20. Greer RC, Liu Y, Cavanaugh K, et al; National Kidney Foundation Education Committee. Primary care physicians’ perceived barriers to nephrology referral and co-management of patients with CKD: a qualitative study. 

. 2019;34(7):1228-1235. doi:10.1007/s11606-019-04975-y

21. Campbell KH, Smith SG, Hemmerich J, et al. Patient and provider determinants of nephrology referral in older adults with severe chronic kidney disease: a survey of provider decision making. 

. 2011;12:47. doi:10.1186/1471-2369-12-47

22. Inker LA, Astor BC, Fox CH, et al. KDOQI US commentary on the 2012 KDIGO clinical practice guideline for the evaluation and management of CKD. 

. 2014;63(5):713-735. doi:10.1053/j.ajkd.2014.01.416

23. Tangri N, Stevens LA, Griffith J, et al. A predictive model for progression of chronic kidney disease to kidney failure. 

. 2011;305(15):1553-1559. doi:10.1001/jama.2011.451

24. Nelson RG, Grams ME, Ballew SH, et al; CKD Prognosis Consortium. Development of risk prediction equations for incident chronic kidney disease. 

. 2019;322(21):2104-2114. doi:10.1001/jama.2019.17379

Articles

Primary care physicians did not refer the majority of patients with severe nephropathy to specialists; nonreferred patients had fewer comorbidities and might be better kidney transplant candidates.

Primary Care Referrals to Nephrology in Patients With Advanced Kidney Disease

The learning health system (LHS) has gained traction as a powerful framework for improving the cost and quality of healthcare. The goal of an LHS is to systematically integrate internal data and experience with external evidence so patients receive higher-quality, safer, and more efficient care. However, if the goal of an LHS is to improve health, as well as healthcare, it must account for and mitigate the negative impact of social and economic factors on health, known as the social determinants of health. In this paper, we discuss the critical role the LHS can play in addressing patients’ social risk factors. We also discuss how integrating data on the social determinants and activities to reduce patients’ social risk factors could advance the mission of the LHS to enhance patient engagement, improve the delivery of personalized care, and more accurately evaluate the effectiveness of care. Without the collection and integration of data on the social determinants of health, the LHS may fail to reach its full potential to improve health and healthcare.

The learning health system has emerged as a framework for improving healthcare delivery, but it may fail to reach its full potential if it does not incorporate data on social determinants of health or activities to reduce social risk factors.

If a central goal of the learning health system is to improve health, it must mitigate the negative effects of social determinants.

Incorporating social determinants could enhance patient engagement, allow for tailoring of care, and more accurately evaluate the care experience.

A learning health system that is able to rapidly evaluate and improve care could be an ideal setting to pilot interventions to reduce patients’ social risk factors.

The learning health system (LHS) has emerged as a powerful framework for improving healthcare delivery.

 The Institute of Medicine (IOM) defined an LHS as “one in which science, informatics, incentives, and culture are aligned for continuous improvement and innovation, with best practices seamlessly embedded in the care process, patients and families active participants in all elements, and new knowledge captured as an integral by-product of the care experience.”

 An LHS can improve care at multiple points along a patient’s clinical trajectory by providing clinicians real-time access to the best available evidence, partnering with patients and their families, incentivizing high-value care, and creating a culture of continuous learning.

 Ultimately, this could lead to a patient-centered LHS on a national scale that allows data sharing across health systems and informs decisions that improve population health and reduce health disparities.

The LHS framework is beginning to take shape, and we believe that realizing its full potential is linked to a health system’s ability to effectively account for and mitigate the social and economic factors that affect health. Social determinants of health (SDH)—the conditions in which people are born, grow, live, and age&mdash;have a profound impact on morbidity and mortality. SDH can lead to social risk factors (eg, food insecurity, housing instability) that can negatively affect health.

 Historically, reducing social risk factors has been the focus of public policy and public health, not healthcare. Increasing attention has focused on health systems’ potential to reduce patients’ social risk factors.

 The National Academy of Medicine (NAM; formerly the IOM) has recommended a core set of social domains that health systems should document in the electronic health record (EHR).

 Increasing research has also shown that addressing patients’ unmet social needs in clinical settings can improve patients’ access to resources, which, in turn, may improve health outcomes and decrease costs.

 With this research as a backdrop, CMS implemented the Accountable Health Communities model to test whether addressing unmet social needs reduces healthcare utilization.

Eliminating the health disparities that emerge from SDH will require broad social, cultural, and policy changes.

 Nonetheless, health systems have an important role in improving health equity, and the LHS could fail to reach its mission of higher-quality, safer, and more efficient care if such systems do not integrate data on SDH or activities to reduce social risk factors.

In this paper, we discuss the role the LHS should play in addressing social risk factors. We further discuss 3 areas—patient engagement, personalized care, and effectiveness of care&mdash;in which the LHS is unlikely to reach its full potential without integrating SDH data.

Role of the LHS in Reducing the Impact of Social Risk Factors

If the goals of the LHS are to improve health and provide more efficient healthcare, it must mitigate the negative effects of social risk factors. Several lines of reasoning support this. First, a large body of evidence indicates that SDH have a greater impact on health than healthcare does. Even if health systems provide optimal healthcare for every patient, only a small improvement in population health would likely be seen.

 Second, although the United States spends more on healthcare, it ranks last among other developed nations for a number of important health outcomes, potentially due to lower levels of spending on social programs.

 Because the United States lacks broad bipartisan support for expansion of public health and social programs, health systems must build new partnerships with social services if they are to improve population health. Third, the drive to value-based purchasing, the emergence of accountable care organizations, and the creation of financial incentives for controlling expenditures have established a strong business case for the LHS to address the social factors that lead to increased utilization.

Although research on the effect of reducing social risk factors in clinical settings on healthcare utilization and health outcomes is limited, early results have shown promise.

 One example is Hennepin Health, which used care coordinators, social workers, and community health workers to connect patients to social services and found a 9% decrease in emergency department visits.

 Another study found that identifying and addressing patients’ unmet social needs using Health Leads, a community-based healthcare organization that connects patients with resources, led to small but significant improvements in blood pressure and cholesterol.

How Incorporating Data on SDH Could Advance the Mission of the LHS

A major goal of the LHS is to improve patient engagement so that patients have a more active role in the shared decision-making process. Understanding the social circumstances in which patients live is central to this goal. For example, advising a patient to increase physical activity when the patient does not have a safe space in which to exercise is unlikely to be effective. In one study that screened patients for social needs, clinicians reported that this information changed care delivery for almost a quarter of patients with an identified need and changed their interactions with more than half.

 However, few clinicians screen patients for unmet social needs, and there are limited data on how practitioners can incorporate interventions in clinical settings to reduce social risks.

 The LHS offers an opportunity to address these gaps by partnering with patients and the community. For example, to improve educational materials about cancer screening for patients with low health literacy and facilitate physician-patient communication, this social risk—informed care could utilize an LHS community model in which community members are involved in the development process.

In 2015, the National Institutes of Health launched the Precision Medicine Initiative (PMI) to understand the drivers of health and develop personalized treatments. A central element of the PMI is the All of Us Research Program, which plans to enroll 1 million adults to investigate how differences in genomics, environment, and lifestyle can contribute to the development of precision medicine treatments.

 Although not in the original LHS framework, the NAM published a 2015 workshop summary describing the inclusion of genomic data in the LHS.

 The LHS offers an opportunity to understand how integration of genomic data affects healthcare delivery.

 As an example, Geisinger Health is integrating individual genomic data within the EHR to inform patient care about disease risk.

 However, social risk factors often have a greater influence on disease occurrence than genetic risk factors. For the genomics-enabled LHS to provide precision medicine, it must account for these social factors that affect health and healthcare.

 Further, an LHS that combines data on individuals’ genomics and social risk factors can better elucidate causal pathways leading to health disparities and tailor the delivery of care.

A further goal of the LHS is to close the gap between evidence generation and clinical care, particularly the lack of generalizability of randomized controlled trials.

 One method to accomplish this is through pragmatic trials that evaluate the effectiveness of treatments in real-world settings. The LHS, with its EHR and robust data collection systems, is an ideal setting for pragmatic trials.

 However, without accounting for patients’ social factors, evaluating the effectiveness of treatments in the LHS could be incomplete. For example, one study evaluated the effect of neighborhood poverty on the impact of a primary care practice—based trial for depression. The authors found that the treatment effect weakened after the initial 4 months among participants living in high-poverty neighborhoods but not among participants in other neighborhoods.

 Thus, the failure to account for social risk factors could lead to effective treatments and novel healthcare delivery strategies showing minimal benefit as a result of unmeasured SDH. Similarly, collecting SDH data could shed light on ways in which social factors mediate the effect of treatments on health.

 Intermountain Healthcare is an LHS that is using an area-level deprivation index to evaluate heterogeneity in medication adherence among patients with hypertension.

The LHS is a powerful framework for improving the cost and quality of healthcare. However, realizing the full potential of the LHS will require moving beyond the traditional boundaries of healthcare and developing innovative approaches to reduce social risk factors. Although the evidence base for the effectiveness of such approaches on improving health is currently limited, early results suggest that addressing patients’ unmet social needs can decrease costs and improve outcomes. We believe it is imperative that the LHS add to this evidence. An LHS that is able to rapidly evaluate and improve care could be an ideal setting in which to pilot interventions reducing social risk factors. This process should begin with the routine collection of SDH data. Making these data a part of the LHS should enhance patient engagement, allow for tailoring of care, and more accurately evaluate the care experience.

 Department of Internal Medicine (DP, DPM, GER), Department of Pediatrics (DP), and Division of Public Health Sciences (DP, DPM), Wake Forest School of Medicine, Winston-Salem, NC.

 The authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

 Concept and design (DP, DPM, GER); drafting of the manuscript (DP, DPM, GER); and critical revision of the manuscript for important intellectual content (DP, DPM, GER).

Without integrating data on the social determinants of health, the learning health system could fail to reach its mission of higher-quality, safer, and more efficient care.

Gary E. Rosenthal, MD

Articles

Primary Care Referrals to Nephrology in Patients With Advanced Kidney Disease

Advancing the Learning Health System by Incorporating Social Determinants

Elderly Veterans With Dual Eligibility for VA and Medicare Services: Where Do They Obtain a Colonoscopy?

Associations With Reduced Length of Stay and Costs on an Academic Hospitalist Service

Does Length of Hospital Stay During Labor and Delivery Influence Patient Satisfaction? Results from a Regional Study