July 2011
Feature Story

Anne Paxton

The old aviation expression “flying by the seat of one’s trousers” refers to going aloft without instruments or radio. As a metaphor, it may not quite fit the situation of clinicians faced with assessing chronic kidney disease patients. But some clinicians feel it comes close. With 26 million people in the U.S. estimated to have chronic kidney disease (CKD), there are too many to target for intervention. Identifying the three percent of CKD patients who are most likely to advance to kidney failure, and treating them preventively, would seem to be a matter of necessity. However, there are no widely accepted predictive instruments for CKD progression.

In general, “Physicians must make ad hoc decisions about which patients to treat,” says Navdeep Tangri, MD, research fellow in the Department of Medicine, Division of Nephrology, Tufts Medical Center, Boston. The risk, consequently, is that they will delay treating those who ultimately progress to kidney failure, or unnecessarily treat those who will not progress. “We need a better way to estimate kidney function rather than requiring physicians to make a mental estimate,” says John H. Eckfeldt, MD, PhD, vice chair for clinical affairs, Department of Laboratory Medicine and Pathology, University of Minnesota, Minneapolis.

Two new articles in the Journal of the American Medical Association offer promising approaches for this critical public health need. They report on studies of multiple-marker approaches to risk prediction in CKD. One study outlines a triple-marker approach that adds cystatin C to the customary measures of eGFR (estimated glomerular filtration rate) and urine albumin-to-creatinine ratio. The second study, by Dr. Tangri and colleagues, focuses on the development and validation of several models employing demographic and clinical data and multiple routine laboratory tests.

The triple-marker study by Carmen A. Peralta, MD, et al., “Detection of chronic kidney disease with creatinine, cystatin C, and urine albumin-to-creatinine ratio and association with progression to end-stage renal disease and mortality” (JAMA. 2011;305;1545–1552) looked at a prospective cohort of U.S. adults enrolled in the REGARDS (Reasons for Geographic and Racial Differences in Stroke) study from 2003–2010. Combining creatinine, cystatin C, and urine albumin-to-creatinine ratio, the study found that adding cystatin C improved the predictive accuracy for all-cause mortality by 13.3 percent and for end-stage renal disease (ESRD) by 6.4 percent. “Adding cystatin C to creatinine and albuminuria for risk prediction can more accurately reclassify persons and can distinguish important prognostic differences, namely a 3-fold risk of death and 4-fold risk of ESRD,” the authors concluded.

In the predictive models study by Dr. Tangri, et al., “A predictive model for progression of chronic kidney disease to kidney failure” (JAMA. 2011;305;1553–1559), the authors used demographic, clinical, and laboratory data from two independent Canadian cohorts of patients with CKD stages three to five. They looked at seven separate models for predicting kidney failure, and found that predictions were significantly better in a model that included age, sex, race, and eGFR_creatinine,plus albuminuria, serum calcium, phosphate, bicarbonate, and albumin. The authors’ conclusion: “We have developed and validated highly accurate predictive models for progression of CKD to kidney failure. Our best model uses routinely available laboratory data and can predict the short-term risk of kidney failure with accuracy and could be easily implemented in a laboratory information system or an EHR [electronic health record].”

Both the Peralta study and the Tangri study make useful contributions to risk prediction, says Josef Coresh, MD, PhD, professor of epidemiology, medicine, and biostatistics at Johns Hopkins University. Dr. Coresh chairs the Chronic Kidney Disease Prognosis Consortium, which pools data from 50 cohorts throughout the world to make more precise and generalizable conclusions about prognosis in kidney disease. Although cystatin C has been shown to be a useful risk marker in a number of articles, how to use it optimally in clinical practice in the U.S. has been unclear. “It can be used very early in the course of kidney disease,” Dr. Coresh says. “But among those patients, it’s differentiating those with very low risk from those with low to intermediate risk.”

The Peralta study, he says “nicely put cystatin C in the context of the more established markers of estimated kidney function and albuminuria and found its highest impact in the group that already has an elevation of one of the other markers. In particular, when kidney function estimated by creatinine is moderately reduced—meaning between 30 to 60—then lower cystatin C is useful in indicating lower risk, while higher cystatin indicates higher risk that may cross the threshold for some action.”

The Tangri article on predictive models is helpful as one of the most systematic attempts to quantify absolute risk, Dr. Coresh believes. “It’s known that absolute risk is very important but most of the assessments have been subjective. This study looked at simple models that incorporated the main risk variables including eGFR_creatinine and demographics as well as the lab variables, and together they did quite well, and better than most clinicians could do in their heads qualitatively.”

These new proposed models for risk prediction appear on the scene as the nephrology community prepares to release revised clinical guidelines, and as the prevalence of chronic kidney disease continues to rise. The crude incidence of dialysis was increasing exponentially for a long time, then it changed, says Dr. Coresh. “Recently the sex- and age-adjusted incidence of dialysis has flattened out some. But in terms of early stages of kidney disease we’re actually seeing a different picture. The prevalence seems to be increasing. It could be that we’re seeing the earliest manifestation of the increasing diabetes epidemic. The big question is what will happen in the next 10 or 20 years given this increasing diabetes prevalence, and whether the risk of dialysis will begin climbing again.”

The staging scheme for CKD developed by the Kidney Disease Outcomes Quality Initiative (Am J Kid Dis. 2002;39(suppl 1):S1–S266) uses eGFR to assign staging numbers one through five to patients. It has been commonly used since 2002 to help physicians understand where patients are in their disease progression, so they can funnel the patients into correct treatment categories. “I think prior to eGFR in 2000, we were sort of in the Dark Ages,” Dr. Tangri says. “[Serum] creatinine alone was just an awful way to predict risk, and the use of eGFR was tremendous in bringing CKD in the spotlight and raising awareness. But over the years we’ve realized it’s not enough.”

A recent international consensus conference (Kidney Disease: Improving Global Outcomes, or KDIGO, Controversies Conference) endorsed use of eGFR and albuminuria in combination to predict risk of CKD. “But the actual recommendation to modify staging criteria has still not been released; they’re working toward consensus on that,” says W. Greg Miller, PhD, director of clinical chemistry and director of pathology information systems, Virginia Commonwealth University Medical Center, Richmond. All nephrologists, he notes, will agree that using both eGFR and urinary albumin is now standard practice. “The question is how to present structured guidelines around that—that’s what is still evolving.” This year or next, he predicts, new guidelines will be released.

Both of the new JAMA studies are important in emphasizing markers in clusters, Dr. Miller says. “There is more than one biomarker, and they need to be used in combination for maximum clinical benefit.” Adding cystatin C, as the Peralta study suggests, would improve risk assessment, says Dr. Miller, who is a consultant to the CAP’s Chemistry Resource Committee and chair of the National Kidney Disease Education Program’s Laboratory Working Group. “Creatinine is produced by muscle metabolism, so some differences you see among people are simply related to differences in muscle mass rather than differences in kidney function. Cystatin C is not sensitive to muscle mass. Adding cystatin C would provide three complementary but independent assessments of kidney function because each marker is sensitive to different types of kidney damage.”

However, some experts voice qualms about cystatin C. Marcello Tonelli, MD, and Braden Manns, MD, in a JAMA editorial accompanying the two studies, write: “Cystatin C measures are not readily available and would lead to additional costs and complexity.” Dr. Miller agrees to an extent: “It’s fair to say most hospital labs don’t offer cystatin C. Our lab doesn’t. And it’s a fairly expensive test compared to eGFRcreatinine and albumin. It’s done by immunoassay, and there are relatively few manufacturers with reagent kits available to measure it. And the principal technical limitation with it is that it is not a standardized assay procedure, so numeric value results differ with different manufacturers’ methods.”

The added expense of cystatin C is a potential obstacle, Dr. Eckfeldt agrees. “I’ve seen clinical charges of over $100 because it generally requires a separate immunoassay instrument to perform. With the reagent costs alone often running about $10 to $20 per test combined with instrument costs, technical time, and markup, the clinical charge might be anywhere from $50 to $150. And that’s a real deterrent for some people who think ‘my chances of getting renal disease are less than one percent.’”

In some countries, particularly in Sweden, cystatin C is used widely, says Dr. Coresh, who notes that the biggest gain from cystatin is that it produces better discrimination rather than high sensitivity. “In Sweden, they don’t necessarily have better quantitative risk information; the quantitative risk information in the U.S. is actually quite good. But they have different platforms, ease of use, and lower cost, so the barriers to offering the test are lower.” Here, more companies are making the assay available on higher-throughput general chemistry analyzers that can run it faster and cheaper. “But over the last decade there have been some issues with some drift in the assays that I am hoping will get resolved soon,” he says.

A standardization effort is underway for cystatin C (sponsored by the International Federation of Clinical Chemistry and Laboratory Medicine), and it led to development of reference material that became available last year through the Institute for Reference Materials and Measurements in Belgium, which should help reduce intermethod variability, Dr. Miller says. “Its commutability properties and an estimated GFR equation based on standardized cystatin C are in current active development and expected soon. Once cystatin C measurement procedures become more standardized, they will be ready for general lab use and presumably more available on different immunoassay and general chemistry platforms. That’s already happening and will probably accelerate in the next year or two.”

There is a big difference, Dr. Miller points out, between identifying CKD patients in the general population and identifying patients in the high-risk population. “The reason creatinine remains the primary biomarker for identifying CKD patients is it’s very commonly measured in almost any condition that you go to the doctor for. And if you’re already at increased risk of developing kidney disease, then you will get creatinine and albumin checked.” Cystatin C is a fairly expensive test compared with the other two, so “when cystatin C enters the picture is unpredictable right now, but it will likely be a followup test.”

Another interesting feature of the triple-marker study is that it actually seems to improve all-cause mortality prediction as well as ESRD or kidney failure, says Dr. Eckfeldt. “The main reason for even having estimating equations or predicting GFR better is that if you find people with early renal disease, then through our interventions like tighter control of diabetes or hypertension, we hope either to prevent or at least to slow progressive deterioration of renal function. While the GFR remains above about 50, people typically have virtually no symptoms. Kidney disease is quite a bit like hypertension, which early on is a totally occult disease and nobody knows they have it until major complications occur, like a stroke. Then it’s too late; the game is over. The same with kidney disease: If you don’t start treatment in the asymptomatic stage, you are much more likely to need a transplant or chronic hemodialysis. “

That’s one of the reasons the federal government is interested in risk prediction, he points out. “Kidney failure is the only disease where regardless of age Medicare pays for therapy, and it currently costs nearly $20 billion. Although ESRD accounts for only about one percent of Medicare patients, it consumes nearly seven percent of the Medicare budget, and that’s predicted to go up dramatically as we see an increase in diabetes.”

In regard to the prediction of all-cause mortality risk, Dr. Eckfeldt says, there is a strong tie between cardiovascular disease and CKD and mortality; people with CKD have much higher cardiovascular disease risk and vice versa. “There is one thing that puzzles me a little. It seems that strategies for primarily controlling diabetes and hypertension reduce the risk of stroke and heart disease. But are the focused therapies for preventing kidney disease progression all that different from what you’d give somebody with diabetes or hypertension anyway? ACE inhibitors and the ARB group of anti-hypertensives help reduce albumin secretions, for example, but there is much more limited clinical trial data that they actually slow progression to ESRD.”

Dr. Peralta, a nephrologist who is an assistant professor of medicine at the University of California at San Francisco, undertook the triple-marker study because she wanted to understand the role of new markers in re-stratifying cases that were either missed by creatinine or misclassified by creatinine. “In my practice, I’ve been very interested that we see a lot of people who either get referred too late, or who get referred but never have complications.”

“By using creatinine alone as a basis to classify kidney disease,” Dr. Peralta says, “we actually can label a lot of people as having the disease that don’t have a lot of risk of progressing or mortality. On the other hand, while using eGFR was a great step forward from creatinine alone and captures many people at risk, “we were missing 16 percent of the population that actually had kidney disease. Given that the equations were developed mostly in persons with kidney disease or healthy kidney donors—a mostly white population—a whole set of people have not been included in the equation, such as Hispanics and the elderly.” Improving detection of CKD in those groups, as well as detecting it even earlier, before GFR reaches 60, “is one way we can guide the future of nephrology in terms of prevention.”

So she believes a triple-marker approach has a lot of promise. There has been a movement in nephrology to add albumin in urine to complications of CKD at all stages, and recent work has shown cystatin C to be a stronger and more linear predictor than creatinine, and more correlated to GFR at the higher end (above 60). “It’s very easily measured because it can be done on the same sample as creatinine,” Dr. Peralta notes.

“Essentially by adding two very easy tests to GFR to stratify predicted outcomes of people, we’re picking up this whole new risk group that would otherwise be completely missed.”

“The question is how clinicians handle the patient in front of them,” Dr. Peralta says. “Right now we don’t have anything that would be similar to the Framingham risk score in cardiology. There are a few risk scores that have been developed, and some guidance in the literature, but now there are no guidelines to educate people on who should get creatinine checked unless they’re diabetic or hypertensive.” As research goes forward, she adds, “it will be important to understand more about how cystatin can work in high-risk groups like the elderly and minorities, and how it can help us in disease not generally marked by proteinuria.”

Dr. Tangri, also a nephrologist by training, is a PhD candidate in epidemiology as well. “So risk prediction was my area of concentration in both fields. The question of predicting which patients with CKD will progress to dialysis was, we thought, one of the most pressing questions in medicine, and that’s why we decided to take this on.”

Since 2002, when the eGFR equation was integrated into the lab, its use has grown to the point that 84 percent of labs in North America report eGFR every time they report creatinine, according to CAP Surveys data. “What they did,” Dr. Tangri says, “was tremendously raise the profile and awareness of CKD as a chronic health problem, but the big problem remained. You’ve identified all these people, but most of them are not going to progress to kidney failure. So this became the next natural question.”

Most clinicians treating CKD patients are primary care physicians, not nephrologists. “They’re aware of the risk factors like the amount of protein in the urine and whether there’s a family history of dialysis, but they’ve largely been thinking about the patients in a ‘gestalt’ way, not a systematic way, and trying to guess who’s going to progress.” Clinicians who are guideline-focused may be intervening too often, Dr. Tangri says, and a clinician who wants to wait may not be intervening often enough. “We really don’t know. All we know is there’s no systematic way this is being done.”

His goal in studying predictive models was to help develop a way that, “when the patient is in front of you, you can look at routinely ordered lab results and give them a precise estimate of their risk of kidney failure in five years.” Ideally, Dr. Tangri says, “we’d like to embed our equation into the electronic medical record or laboratory information system, so that whenever these tests are ordered, a report should give the probability of kidney failure in the next five years.” The study includes links to smartphone applications for every platform—iPhone, Blackberry, iPad, etc.—so physicians can employ the equations at the point of care. “But ultimately we want this to be integrated into the lab reports.”

Beyond that, the models could help more effectively direct clinical care. For example, Dr. Tangri and colleagues suggest, in CKD stage three, lower-risk patients could be managed by the primary care physician without additional testing or treatment of CKD complications, while higher-risk patients could receive more intensive testing, intervention, and early nephrology care. In CKD stage four, different risk thresholds could be used to triage patients for decisions regarding dialysis modality education, vascular access creation, and pre-emptive transplantation.

Before dialysis can be started, “Patients typically need a fistula, which is a connection of the artery to a vein, a minor surgery needed at least three months if not six months before they start dialysis. Currently the guidelines say for any patient with CKD stage four, you should consider putting in dialysis access. But since there is no specific risk cutoff, decisions are again made unsystematically. So we are proposing that anybody with a risk of more than 20 percent of kidney failure over the next two years may be a candidate, while for anybody with a risk of less than 10 percent you could probably wait. Those are the kind of things we are looking to evaluate in trials—whether the right people are having surgery.”

Continuing his work on these models, Dr. Tangri is now engaged in validating the findings in 20 to 30 other cohorts. “We are studying groups of anywhere from 50,000 to 300,000 patients with CKD, trying to put a punctuation point on the study—that this equation works across the world and across the spectrum of patients.”

Dr. Tangri predicts that more and more articles will be tackling risk estimation in medicine. “But there are tremendous barriers that prohibit physicians from using risk prediction tools. No. 1, most tools are never validated in an external population, so you never know whether they are truly generalizable. Second, no physician will want to go calculate on a piece of paper or go to a Web site. Ideally you want things to be integrated into the EMR, but EMR adoption is still far below what it should be, compared to other facets of society like banking.”

That was the reason for integrating a formula into a smartphone app. “We really think the eGFR equation was transformative because it was able to be integrated into the lab, and we think what will make our equation transformative is if it can do the same. So you could have a check box for the Kidney Failure Risk Equation that would simply order the chemistry panel and a urine albumin. On the EMR side, the equation could be programmed so any time this panel of tests is ordered it automatically gives you a prediction.”

“The way we envision it, you report the risk based on the variables you have. So if you only have four variables in the lab or EMR, fine. If you have all eight variables, it’s slightly better.” Multiple institutions across Canada and some U.S. sites will soon pilot this system, and over the long term randomized studies will determine their effectiveness.

Such relatively elaborate testing, in fact, was never done on the Framingham equation, which became part of the standard of practice in cardiology without it. “There are no risk prediction tools yet in nephrology, but as the current guidelines for CKD are being updated now, they will include our tool as an alternative method to potentially predict risk,” Dr. Tangri says. In the meantime, he and his colleagues are going to test what impact the equations can have on care.

The predictive models study by Tangri is preliminary information but a good indication of the direction research is taking with respect to identifying and classifying risk, particularly as to rates of progression, Dr. Miller says. Predicting which patients will progress more rapidly than others to end-stage renal disease would be helpful because they can be put on more aggressive treatment. “That is a really challenging question right now and there are not many tools to do it.”

The Tangri article also makes an important step forward in showing quantitative information for one- and three-year risk, Dr. Coresh says. “The stakes for risk of ESRD are actually quite high. There are a number of clinical decisions that clinicians may initiate to prepare you for the start of dialysis.” Because the study was conducted in Canadian systems, its generalizability to other health care systems needs to be tested. “But you want the information to be quantitative and to be as fast at discriminating risk as possible.”

As to the practicality of requiring multiple markers, Dr. Coresh points out that most kidney patients already get these tests. “The best evidence for that is that the Tangri article used an existing database. So patients are getting these tests to be monitored as to whether they’re having complications. And the study shows that given a certain level of kidney function and albuminuria, the more complications you have and higher risk of eventually ending up on dialysis.”

Do the predictive models in the two studies show value in targeted screening? “The Tangri article shows if you have severe CKD, it’s quite reasonable to look at your other comorbidities and other lab tests, and the article gives physicians a quantitative tool to use above and beyond the individual lab tests,” Dr. Coresh says. “However, testing the entire population with cystatin would be premature because creatinine does reasonably well and is already quite widely applicable. The question is finding that middle zone of individuals where creatinine shows intermediate risk; then cystatin will definitely help in discriminating the risk. But we need more trials to give an evidence base for that action.”

Future research needs to proceed in multiple directions, Dr. Coresh believes. “I think it would be useful, first of all, to expand on these nice articles with wider datasets to be better able to assess the risk and generalizability of the findings. We also need research into specific actions tested by clinical trials based on these risk predictions. Sometimes we will have full trial evidence; sometimes physicians will act based on their best judgment. We need both.” Finally, he adds, new interventions and models of care are needed whereby patients with CKD can optimize their care and prevent or delay the onset of dialysis.

For clinical laboratories, Dr. Eckfeldt says, “The main thing is there is going to be increasing pressure from clinicians to add additional biomarkers. In these two papers, they took two approaches and they did it somewhat differently in the way they developed them, but there will probably be other, perhaps new, biomarkers that will help refine risk prediction both for ESRD and for overall mortality, so I think labs are going to need to respond by making these tests available.”

The next step, Dr. Tangri says, will be to encourage laboratory reporting that includes both tests and interpretation relating to kidney failure risk. “We really need support from pathologists and laboratory directors to say we’re interested in integrating risk prediction into laboratory information systems.”