Intended for healthcare professionals

Practice Quality Improvement Report

The role of specialists in managing the health of populations with chronic illness: the example of chronic kidney disease

BMJ 2009; 339 doi: (Published 08 July 2009) Cite this as: BMJ 2009;339:b2395
  1. Brian J Lee, nephrologist1,
  2. Ken Forbes, care management analyst2
  1. 1Kaiser Permanente, Hawaii Region, Moanalua Medical Center, 3288 Moanalua Rd, Honolulu, HI 96819, USA
  2. 2Kaiser Permanente, Hawaii Region, Care Management Institute, 2828 Paa Street, Honolulu, HI 96819
  1. Correspondence to: B J Lee brian.j.lee{at}
  • Accepted 29 April 2009


  • Problem Specialty care has been used to manage individual patients at the discretion of generalists but not to drive improvements at the population level.

  • Design Observational longitudinal study.

  • Setting Kaiser Permanente Hawaii, with more than 10 000 members with documented chronic kidney disease.

  • Key measures for improvement Rate of late referrals to nephrology care, defined as occurring within four months of end stage renal disease and the proportions of patients starting haemodialysis with a mature arteriovenous fistula and starting dialysis in the outpatient setting.

  • Strategies for change Risk stratification of the entire population and unsolicited consultations provided by nephrologists to generalists, based on patients’ risk level, enabled by an electronic population management database.

  • Effects of change Between 2004 and 2008, the proportion of referrals occurring within four months of onset of end stage renal disease dropped from 37 of 116 (32%) to 10 of 84 (12%), P=0.001. The proportion of patients starting haemodialysis with a mature arteriovenous fistula increased from 19 of 108 (18%) to 27 of 76 (36%), P=0.006. The proportion of patients who started haemodialysis as outpatients increased from 39 of 113 (35%) to 47 of 84 (56%), P=0.003.

  • Lessons learnt Turning the traditional referral system on its head by providing unsolicited, risk driven nephrology consultations is an effective strategy for increasing the quality of medical management of patients with chronic kidney disease in the primary care setting and improving the cost effective use of nephrology services. This approach may be broadly applicable to other specialty areas.

End stage renal disease is a major source of morbidity and mortality worldwide. In the minority of the 11% of adults with chronic kidney disease who progress to end stage renal disease, management by a nephrologist can slow progression and improve blood pressure control and the use of angiotensin converting enzyme inhibitors or angiotensin receptor blockers.1 2 3 Management by a nephrologist before the onset of end stage disease is associated with reduced short term morbidity and less need for temporary vascular access for dialysis.4

Despite the trend towards earlier intervention by nephrologists, most patients are still managed in primary care. Generally, a referral is appropriate when a nephrologist is substantially more able than a generalist to manage a patient’s chronic kidney disease, by virtue of more experience with and knowledge about preparing patients for renal replacement therapy or having the ability to home in on optimal management in the presence of multiple comorbidities (optimising antiproteinuric therapy to slow progression of disease, for example). A referral can be considered late when clinical opportunities have been lost: when the progression of chronic disease can no longer be slowed, the onset of end stage disease cannot be prevented or delayed, and not enough time remains to prepare patients optimally for dialysis or transplantation. Late referral to a nephrologist (less than 120 days before the onset of end stage disease) occurs in 30-42% of patients.5 6 7 8 9 10 11 12

Late referrals are associated with higher mortality and higher rates of admission to hospital among kidney patients,13 and with lack of permanent access for haemodialysis14 and early failure of arteriovenous fistulas,15 as well as decreased quality of life.16 Attempts to increase timely referrals have been only marginally successful.17 However, referrals may occur close to the onset of end stage renal disease for reasons beyond the control of referring physicians (in patients with acute renal failure, those who have not been identified, and those who refuse nephrology care). Avoidable late referrals are of patients who are at high risk for progression to end stage disease; these were the focus of our effort.


Kaiser Permanente is the largest not for profit integrated healthcare delivery system in the United States, with 8.7 million members in eight regions. It deals with all health care for adults and children, including preventive, routine, specialty, emergency, and inpatient care; ancillary testing; pharmacy and rehabilitative services; and home care.

Kaiser Permanente Hawaii has 214 000 members of Asian, Pacific Islander, or white ethnic origin. Less than 5% of members are black. More than 10 000 members have a documented glomerular filtration rate of <60 ml/min/1.73 m2 or overt proteinuria (estimated or measured daily urine excretion ≥0.3 g protein).

At Kaiser Permanente Hawaii, 110 generalists (58 family practice physicians and 52 internists) refer patients with chronic kidney disease to a division with six nephrologists. The timing of referrals has been highly variable, hinging on factors that include the patient’s renal function, the generalist’s comfort level with managing chronic kidney disease and preferred practice patterns, and interpretation of published guidelines. Generalists are responsible for timely referrals but can overlook patients with chronic kidney disease who are at high risk for end stage renal disease. Before this project, consensus existed in the nephrology division that some patients were referred for consultation too late and others too early.

Our initiative began before Kaiser Permanente’s integrated electronic health record system was fully implemented in 2005.18 KP HealthConnect provides documentation and clinical results reporting in all settings (primary, specialty, urgent, and emergency), immediate availability for all system users, and easy searchability. A KP HealthConnect utility, the Panel Support Tool, came into wide use in December 2005, prompting primary care doctors to order annual laboratory screening tests in patients with documented chronic kidney disease or diabetes.19

Data analysis and interpretation

In 2004, to understand the characteristics of patients with chronic kidney disease, we began extracting data from a regional registry of renal patients.

Patterns of care

In 2004, generalists independently managed more than 90% of patients whose last recorded glomerular filtration rate was <60 ml/min/1.73 m2. At that time, 32% of patients who progressed to end stage renal disease were first seen by a nephrologist within four months of its onset.

One indicator of the timeliness of referral is the use of a mature arteriovenous fistula at the start of haemodialysis. In 2004, of 108 patients starting haemodialysis, 19 (18%) did so using an arteriovenous fistula and 83 (77%) started with an indwelling central venous catheter. Among 113 patients starting any type of dialysis, 39 (35%) did so as outpatients.

Of 10 179 patients with a glomerular filtration rate of <60 ml/min/1.73 m2, 3834 (37.7%) had a last recorded systolic blood pressure at target (<130 mm Hg). Significant proteinuria (>1 g/day) was recorded in 1035 (10.1%) patients, and 611 (59.0%) were taking less than a full dose of an angiotensin converting enzyme inhibitor (40 mg of lisinopril, or equivalent) daily.

Interpretation and key measure

Data indicated that among patients with chronic kidney disease, referrals to nephrology care were not sufficiently timely, which subsequently had an impact on renal replacement therapy at onset of end stage renal disease and hospitalisation rates for starting haemodialysis. We used the percentage of late referrals for nephrology care (<120 days before onset of end stage disease) as a key measure of appropriate risk management of the whole population. Additional measures included the presence of a mature arteriovenous fistula and start of haemodialysis as an outpatient. Improvement in these process measures is likely to improve survival and quality of life, but they do not directly measure clinical outcomes.

Strategy for change

Our goal was to increase referrals for only those patients at the highest near term (two year) risk for end stage renal disease and overall risk for progression of disease, allocating scarce specialist resources to patients who could benefit most (figure). Our strategy for improving the timeliness of referrals consisted of stratifying patients by risk, enabled by an electronic database, and initiating unsolicited consultations.

The original screening criterion was a glomerular filtration rate <30 ml/min/1.73 m2, in accordance with published guidelines and a similar proposal.20 21 However, this standard overlooked patients with higher glomerular filtration rates and heavy proteinuria, who often progress rapidly and sometimes catastrophically. In addition, a number of referred patients with a rate under 30 were remarkably stable, requiring little intervention.

We analysed population data to identify risk factors for progression (see appendix on When considered together, glomerular filtration rate and proteinuria were highly predictive of risk. We identified three specific laboratory criteria for patients at high risk: glomerular filtration rate <20 ml/min/1.73 m2; glomerular filtration rate <40 ml/min/1.73 m2 plus urinary protein:creatinine ratio >2; and urinary protein:creatinine ratio >4. Laboratory criteria for low risk status were glomerular filtration rate ≥30 ml/min/1.73 m2 and <1 g proteinuria. Giving proteinuria and glomerular filtration rate equal weight in determining risk status runs counter to stage based management of chronic kidney disease,22 but this strategy was derived from internal data analysis and is in agreement with emerging research.23 24 These criteria were reviewed and adopted by an internal committee responsible for evidence based guidelines, which also developed and distributed updated guidelines on treating chronic kidney disease to all generalists to support our initiative.25

We designed a computer program to create a concise profile for every patient with chronic kidney disease, risk stratified on the basis of data downloaded monthly from the central database (see box 1). Recognising the impact of other factors, such as age and comorbid conditions, on the potential value of specialty care, we reviewed the complete profile of each high risk patient and often obtained more information from the complete electronic health record after it became available.

Box 1 Key features of the electronic population management database

  • Access to comprehensive, current information on patients—Database contains longitudinal, dated information on age, sex, weight, comorbidities, blood pressure, medications, urine protein quantitation, serum creatinine, estimated glomerular filtration rate, serum potassium, haemoglobin, phosphorus, calcium, albumin, and parathyroid hormone concentrations, and last visit with the generalist

  • Risk stratification with robust documentation of risk status—Numerical risk ranking, estimated two year risk of end stage renal disease (based on Kaiser Permanente Hawaii population data)

  • Ability to annotate patient records to prevent repeating work—Form field for comment log; eg, “PCP referred but patient did not keep appointment”

  • Seamless integration of new data into longitudinal record—KP HealthConnect data updated daily and incorporated into monthly risk stratification database. Messaging to primary care physicians incorporated into KP HealthConnect

  • Electronic messaging between nephrologists and generalists—Form field automatically generates customisable template for message to generalists, based on most common recommendations

  • Electronic alerts for patients with deteriorating values—When the database is updated with new patient data each month, any records containing deteriorating values are flagged with an alert

  • Generation of population statistics—Database records can be sorted by rapid rate of progression to end stage renal disease or development of anaemia, serum haemoglobin or potassium, estimated glomerular filtration rate, 24 hour urine protein, or blood pressure. One click generates population level statistics on these variables

  • Ability to flag patient records by status—One click colour codes the patient’s database record into one of several categories, such as referred back to primary care, already referred, not a referral candidate

Our initial idea was to provide detailed care recommendations to generalists to help them better manage all high risk patients. When we approached our generalist colleagues about this idea in 2004, however, they objected because of overwhelming workloads and the scheduled implementation of KP HealthConnect. Consequently, we focused on improving the timeliness of referrals, and our colleagues responded positively.

Our approach to improving the timeliness of referrals turned the traditional nephrology referral system on its head. Beginning in early 2004, in addition to continuing to receive referrals from generalists, we proactively intervened. For patients whose abnormal glomerular filtration rate, urinary protein:creatinine ratio, or other variables indicated an immediate need for specialty care, we contacted the responsible generalist, typically via secure messaging, and requested a referral for a nephrology visit.

In late 2005, after more than a year of successfully soliciting referrals (and with the electronic health record system well established and a generalist’s encouragement), we also began providing unsolicited consultations. Among the 4000 cases we reviewed, patients would often clearly benefit from improved medical management (for example, they had significant proteinuria and uncontrolled blood pressure, particularly when they were not receiving maximal pharmacotherapy), but most did not immediately need a referral. We contacted the responsible generalist directly, typically via secure messaging, and provided an unsolicited consultation about an individual patient (box 2). The generalist retained authority for the patient’s care—with the added benefit of interactive mentoring from the nephrologist.

Box 2 Example of risk driven consultation with a generalist

While reviewing risk stratified profiles, the nephrologist noted a serum creatinine concentration of 2.5 mg/dl (121 μmol/l), up from 1.2 mg/dl and blood pressure of 188/78 in a 60 year old man who had not been previously identified as high risk. The nephrologist accessed the patient’s electronic health record, finding that he had a diagnosis of benign prostatic hypertrophy but had seen a urologist four months previously. At that time, the urologist noted that the patient was doing well and suggested a return visit in 12 months. The patient’s generalist saw him two weeks before the nephrologist’s review, at which time routine laboratory investigations were obtained.

The nephrologist contacted the patient’s generalist via secure messaging:

  • ----- Message -----

  • From: B____

  • Sent: Apr 18, 2007 4:52 PM

  • To: J___

  • J_____, I see his Cr is up to 2.5 (I don’t follow him but he came up on my database). I wonder if it is due to the prostate getting worse. You may want to recheck it, maybe repeat ultrasound to see if he is with hydronephrosis. If it’s not the prostate, why don’t you refer him to us?

The generalist’s response:

  • ----- Message -----

  • From: J___

  • Sent: Apr 20, 2007 7:23 AM

  • To: B____

  • Thanks. I will order the US and see.

  • J___

The ultrasound scan showed an obstructing kidney stone on the right side. The stone was removed, and the patient’s serum creatinine fell to 1.7 mg/dl, but remained raised from baseline. The generalist shared this information with the nephrologist, who suggested measuring urine protein.

The patient’s 24 hour urine protein was 4818 mg, and he was referred to nephrology. A subsequent biopsy found IgA nephropathy. With aggressive antiproteinuric treatment and weight loss, the patient’s serum creatinine stabilised at 1.7 mg/dl and urinary protein:creatinine ratio dropped to 1.2.

In 2006, the nephrology department also began reviewing all incoming referrals, using the KP HealthConnect patient record. If we judged the patient to be at low risk, we contacted the referring generalist about the possibility of retracting the referral. We never refused a referral that a generalist felt was needed, but low risk patients were most often returned to primary care with specific management advice. We flagged these patients’ records and looked at them first when the risk stratified database was generated each month. This intervention was well accepted by generalists and was crucial to applying scarce nephrology resources to patients at highest risk.

In one half day per week between 2004 and 2008, a single nephrologist reviewed the profiles of about 4000 patients among more than 10 000 patients with chronic kidney disease, using the electronic health record for additional information as needed. We initiated risk driven consultations that led to approximately 280 referrals of patients at high risk. From 2006 to 2007, an estimated 150 low risk referrals and approximately 200 other low risk patients who had been previously referred and were being actively followed by nephrologists were returned to primary care.

Effects of change

We assessed the statistical significance of changes in the proportions of late referrals, haemodialysis starts with a mature arteriovenous fistula in place, and outpatient dialysis starts, using the χ2 test and controlling for patient’s sex and age.

Our key measure, the rate of late referrals, dropped 18-28 percentage points below previously published rates (table 1). Using Stata version 10 software, we assessed whether the significance of our results was affected by clustering of patients within generalists’ practices or medical facilities; it was not a factor.

Table 1

 Impact of risk driven consultations on patterns of care. Values are numbers (percentages) unless otherwise stated

View this table:

In addition, we examined the patterns of all nephrology referrals. The proportions of referrals for low risk patients and high risk patients fell significantly from 2004 to 2007 (table 2).

Table 2

 Impact of risk driven consultations on total number of referrals and referral risk status. Values are numbers (percentages); totals include patients of uncategorised or unknown risk*

View this table:

We carefully monitored low risk patients who returned to primary care, prioritising them for early review each month. We began tracking outcomes for these patients in late 2007, after which 74 referrals were returned to primary care. After an average follow-up of 409 days, none reached end stage renal disease or were referred a second time. Their average change in creatinine was a decrease of 5 μmol/l (0.06 mg/dl). Only 10 patients had an increase in creatinine ≥26 μmol/l; in the worst case, serum creatinine rose from 126 μmol/l to 194 μmol/l after 353 days in the setting of urosepsis.

Lessons learnt and next steps

Within an integrated care delivery system, nephrologist guided management of care effectively supplements primary care management of patients with chronic kidney disease. We believe that putting more decisions about the appropriateness of nephrology care into the hands of nephrologists is as sensible as it is non-traditional. With enough information about individual cases, nephrologists can often have a better sense of when their expertise is needed than do the generalists who refer patients to them.

However, nephrologists are often limited by a lack of information about patients outside their service. When we overcame this barrier with our computer program that stratified patients by risk each month and gave nephrologists access to central electronic records we could assume a more active role in managing the entire population of patients with chronic kidney disease, directing high risk cases into nephrologists’ care and supporting generalists in managing the other patients. Our initiative reduced the rate of late referrals well below previously published rates.5 6 7 8 9 10 11 12

To sustain the initiative, it was important to prevent sharp increases in the nephrologists’ workloads: high risk patients require more intensive management and more frequent follow-up. We monitored the total number of additional high risk referrals and, depending on our nephrology group’s existing workload, sometimes limited referrals to patients with the highest risk profiles. We also balanced our workload and prevented scheduling delays by sending low risk patients back to primary care. The Panel Support Tool ensured their follow-up laboratory tests were completed, allowing us to identify any patients who needed to return for specialty care. In fact, as generalists learned the criteria for referral and gained more expertise at managing early chronic kidney disease, the total number of referrals to specialty care dropped from 426 in 2004 to 322 in 2007 (table 2). This tool uses automatic reporting of estimated glomerular filtration rate, which can lead to a 40-170% rise in referrals, but our system prevented this.26 27

The increase in referrals of patients at high risk and accompanying decrease in referrals of patients at low risk indicate that our initiative increased the efficient use of nephrology resources. Although we did not measure the overall impact on costs or the use of generalists’ time, proactive intervention may offer efficiency benefits.

Identifying patients and providing proactive, risk driven consultations requires the judgment of a nephrologist. Identification begins with criteria based on algorithms but relies on expert judgment. Our database stratified patients and presented them for our attention, but evaluating each case in the context of the patient’s condition, characteristics, and comorbidities transcended rules, requiring a nephrologist’s knowledge and perspective. Other professionals involved in case management, such as nurses, pharmacists, physician assistants, or even dedicated generalists, could follow a simpler, rule based system to provide advice, but they would lack the clinical judgment that comes from years of specialty training, clinical experience, and focused continuing education. Consultations taking place as part of this initiative—whether in a clinic, by secure messaging, or over the telephone—required the same level of deep expertise that generalists rely on nephrologists to provide in any other model of care.

Risk driven consultations provide a novel and highly effective method of educating doctors, using their own patients as examples. The nephrologist is able to give advice to a generalist on an actual case in which the generalist often has no idea that clinical management could be improved. Pointing out specific opportunities for care can be powerful. Most generalists welcomed information from a trusted colleague about how to improve care for a specific patient with chronic kidney disease who did not yet need a referral for specialty care, responding by email with thanks.

Another advantage of education that uses a generalist’s own patients as real life examples is that learning is more likely to persist and influence the management of future cases. For example, on implementation of the new system during 2004 and 2005 we sent one generalist numerous care recommendations. Several of his patients started dialysis during this time. We encountered him in early 2007. Realising we had sent him only one message in the previous year and that fewer of his patients were starting dialysis, we asked how he was managing his patients with chronic kidney disease. He responded, “Thanks so much for your help and advice. I just screen them for proteinuria and GFR at least once a year, get their blood pressure under 130/80, and make sure they’re on a maximal dose of an ACE inhibitor and add an ARB if needed to get the urine protein-to-creatinine ratio down. I get it now.”

Management of patients with chronic kidney disease or other conditions requires comprehensive information technology. We started with an electronic database to risk stratify the population and to present relevant data about individual patients. The advent of KP HealthConnect in 2005 meant that more data were available in electronic form (blood pressure, weight, updated drug lists) to be pulled into our database and that any patient’s entire medical record was available immediately if we needed more information than our database provided. As comparable systems are not commonly available, the generalisability of our approach is limited.

Similar initiatives can be tailored to available resources. Risk stratification allows nephrologists to target patients at greatest need; how far down the stratified population they are able to reach depends on how much time is available. An alternative risk stratification method could be used, provided it accurately identified patients at highest risk for developing end stage renal disease.

Our system could be applied to other chronic conditions in which at least one measurement related to quality is systematically available. Diabetologists could use haemoglobin A1c concentrations to proactively intervene in people with diabetes; pain management specialists might use the number of opioid prescriptions dispensed to identify chronic pain patients with unmet needs; and cardiologists could use the ejection fraction to risk stratify the population of patients with congestive heart failure to ensure management is consistent with guidelines.

Risk stratification and targeted, proactive interventions are essential to achieving improvements in care at the population level. Our experience shows that considerable improvements are possible when nephrologists take a strong role in reviewing and managing the health of all the patients with chronic kidney disease, actively intervening according to individual risk status.


Cite this as: BMJ 2009;339:b2395


  • The outcomes we observed reflect the skill and hard work of the nephrologists and the participation of the primary care physicians at Kaiser Permanente Hawaii. We particularly thank Samir Patel, Geoff Galbraith, and Alan Lau; senior and local leaders within the Kaiser Permanente Hawaii Region; national programme offices; and nephrology and vascular surgery colleagues, and also Harold Kurt, Monty Glover, Christian Balazs, and Rachel Murkofsky, as well as Carol H Cain and Jim Bellows of the Care Management Institute, and Jenni Green for editorial assistance.

  • Contributors: BJL wrote the article with assistance from Jenni Green, a professional writer. The article was reviewed by Ken Forbes, Carol Cain, and Jim Bellows.

  • Funding: This work was jointly supported by the Nephrology Division and Care Management Institute of Kaiser Permanente, Hawaii Region.

  • Competing interests: None declared.

  • Provenance and peer review: Not commissioned; externally peer reviewed.

  • Patient consent not required (patient anonymised, dead, or hypothetical).