Review

Correspondence *Division of Nephrology, 950 West Walnut Street, R2-202 Indianapolis, IN 46202, USA

Email bmolitor@iupui.edu

Introduction

Acute kidney injury (AKI) is a devastating clinical disease associated with unacceptably high mortality rates.¹ With its incidence increasing at a rapid rate,^{2, 3, 4} AKI remains a diagnostic and therapeutic dilemma.^{1, 2, 5} Recent attention has focused on understanding the epidemiology of AKI and on translating basic pathophysiologic understanding into clinical advances. Central to this translational effort is the need to develop 'biomarkers' that will improve the precision of the early diagnosis of AKI and facilitate effective therapeutic intervention. A number of epidemiologic studies have identified clinical and laboratory parameters that have advanced our understanding of the at-risk population and of the overall importance of AKI severity as it relates to clinical outcomes. The pioneering study by Thakar et al. identified clinical risk factors for patients undergoing cardiovascular surgery who developed AKI requiring renal replacement therapies.⁶ In this high-risk population, prevention therapies could be used to reduce the incidence of AKI.

The Acute Dialysis Quality Initiative developed a set of criteria for defining AKI, based upon serum creatinine and urine output, called the RIFLE (Risk, Injury, Failure, Loss, and End-stage) criteria.⁷ Pioneering epidemiologic studies that used the RIFLE criteria have shown conclusively that the morbidity and mortality associated with AKI are correlated with the severity of kidney injury as defined by elevations in serum creatinine level.^{1, 8} These data are important as they indicate that prevention of AKI or attenuation of the extent of injury could result in marked decreases in morbidity, mortality and the associated costs of AKI. The studies re-emphasize the need for biomarkers that are sensitive, specific and quantitative in facilitating the early diagnosis of AKI. The numerous potential biomarkers that have been identified in preclinical studies are only now beginning to be tested in clinical validation studies. This Review will, therefore, propose a framework for the validation of biomarkers that affect therapeutic decisions. Specifically, we describe the utility of biomarkers for the identification of patients at risk of developing AKI, the surveillance of early events leading to or preceding AKI, and the diagnosis and prognostication of AKI. The importance of this endeavor cannot be overemphasized, as both primary prevention and secondary therapeutic decisions depend on this information.

What is a Biomarker?

The development of new biomedical technologies is rapidly advancing, creating potentially novel methods to diagnose and monitor disease states. A more prominent role for biomarkers is proposed in assessing overall disease risk, in early non-invasive screening and detection, in disease stratification and prognostication, and in assessing response to therapeutic interventions ('theragnostics'). Novel biomarkers also have considerable potential utility for research and clinical purposes.

The term biomarker (shortened from 'biological marker') was introduced in 1989 as a Medical Subject Heading (MeSH) term; biomarkers are defined as "measurable and quantifiable biological parameters (e.g. specific enzyme concentration, specific hormone concentration, specific gene phenotype distribution in a population, presence of biological substances) which serve as indices for health- and physiology-related assessments, including disease risk, psychiatric disorders, environmental exposure and its effects, disease diagnosis, metabolic processes, substance abuse, pregnancy, cell line development, epidemiologic studies, etc."⁹ Generally speaking, a biomarker is a measurable indication of a specific biologic state that is relevant to a specific disease process. In 2001, an NIH working group standardized the definition of a biomarker as "a characteristic that is objectively measured and evaluated as an indicator of normal biologic processes, pathogenic processes, or pharmacologic responses to a therapeutic intervention."¹⁰ The FDA uses the term 'biomarker' to describe any measurable diagnostic indicator that is used to assess the risk or presence of disease.¹¹

The concept of biomarkers is related to, but distinct from, that of disease risk factors or intermediate and surrogate end points. Preferred definitions and a conceptual framework for biomarkers and surrogate end points have been defined by a consensus working group, and a basic glossary of related terminology is provided in Box 1.¹⁰ Biomarkers can sometimes serve as surrogate end points; for example, to evaluate response to therapy in clinical trials in lieu of clinically meaningful 'hard' outcomes such as morbidity and mortality. The validity of surrogate end points in clinical trials is, however, often questioned, due to risks of residual confounding.¹² As such, the threshold for surrogate end point validity is often set higher than that for biomarker validity. It is recommended that a surrogate end point should explain more than 50% of the effect of the exposure (intervention) on the outcome of interest.^{9, 13}

Utility of Biomarkers

Desirable performance characteristics for biomarkers have been well described. A perfect biomarker is easily measurable, accurate, reproducible, cost-effective and easy to interpret by the clinician. Measures of the biomarker should be widely separable in patients with the disease state compared with individuals who are in good health. The biomarker should be acceptable to the patient, and applicable to men and women over a range of different ages. The biomarker should be detectable early in the disease process to allow diagnosis and effective therapy to take place before permanent sequelae occur.^{9, 14} The desirable properties of a biomarker will vary according to the intended use.

There is growing awareness of the importance of rigorous analysis and systematic testing of prototypical biomarkers. Proper design of validation studies for candidate biomarkers is crucial, but has often been lacking in biomarker research.^{15, 16} As an example, it is no longer sufficient for a novel biomarker simply to be associated with risk in a study population, such as patients with AKI. To be useful, the biomarker should independently provide information about risk or prognosis that is additive to that provided by conventional clinical factors, and/or the biomarker should account for a large proportion of the risk associated with the disease. Biomarker assays should be reproducible. If the biomarker is to be used as a diagnostic test, it should be sensitive and specific; that is, it should have high predictive value. The generally accepted measure of performance is the receiver operating characteristic (ROC) curve, which compares test sensitivity to 1 – test specificity for all possible cutoff values. The area under the ROC curve (also known as the C statistic), is the proportion of pairs for which the model assigns greater probability to patients who will experience the event than to individuals who will not.

When viewed from the perspective of ROC curves (as opposed to risk ratios), current biomarkers for many clinical conditions unfortunately often add only marginal predictive power to that provided by conventional clinical factors in assessing risk. In this regard, it is worth examining data on biomarkers of cardiovascular risk, one of the better-studied areas of biomarker development. Wang et al. assessed the performance of 10 biomarkers for predicting first major cardiovascular events and death in the Framingham population and found a marginal increase in the C statistic (from 0.795 to 0.816), even when adding the five most powerful biomarkers to a model based on conventional risk factors.¹⁷ Honda and co-workers evaluated the C statistics for serum albumin, C-reactive protein, interleukin (IL)-6 and fetuin-A in an end-stage renal disease population and found that these biomarkers had marginal utility for predicting cardiovascular events or mortality.¹⁸ At present, therefore, available biomarkers of cardiovascular risk—although associated with hazard for adverse outcomes in large populations—have limited utility in the management of individual patients.

An ideal biomarker for AKI would be inexpensive, easy and rapid to quantify, repeatable, capable of determining the severity of injury, unique to the kidney, would increase early in the course of injury, and would have a high degree of sensitivity and specificity. Sensitivity, specificity and ROC curves are commonly used to measure the discrimination of a biomarker. 'Discrimination' refers to the ability of a biomarker to distinguish 'case' from 'non-case'. An additional consideration in evaluating biomarker utility is that sensitivity, specificity and the C statistic are not directly dependent on the relative frequency or pre-test probability of the disease. As such, appropriate interpretation of a biomarker requires a Bayesian approach in which pre-test probability of disease is integrated with biomarker test results to estimate the post-test probability of disease.^{19, 20} This combination yields a predictive value for a positive or negative test. Nomograms that use Bayesian analysis to generate post-test probability of disease or outcomes are readily available.²¹ The positive predictive value of a biomarker falls as the prevalence (or pre-test probability) of the disease or outcome falls, so that a biomarker validated in a population with a high prevalence of a disease might not perform as well in populations in which the disease is less prevalent. The lower the prevalence of disease or outcome in a population, the higher the rate of false-positive results; the higher the disease prevalence, the higher the rate of false-negative results.

In addition to having good discrimination, an ideal biomarker will also exhibit a high level of calibration. 'Calibration' refers to the capacity of a biomarker to predict risk in sub-groups of the population relative to the actual observed risk.^{22, 23} Calibration is frequently assessed using the Hosmer–Lemeshow goodness-of-fit statistic.²⁴ In well-calibrated models, there are only small differences between predicted and actual risks, resulting in Hosmer–Lemeshow statistics that are small in absolute value with nonsignificant probability values. Risk prediction algorithms that uniformly overestimate or underestimate risk can be recalibrated; for example, recalibration of the Framingham Coronary Heart Disease risk score for a Chinese cohort substantially improved risk prediction.²⁵

Use of Multiple Biomarkers

AKI is a complex syndrome with multiple diverse etiologies, and often occurs in the setting of systemic diseases such as sepsis. Sepsis is a syndrome that presents enormous challenges for biomarker development (Figure 1).²⁶ The ability to predict risk in such complex clinical circumstances might be enhanced by strategies that employ multiple simultaneous biomarkers of risk,²⁷ or by integration of biomarkers with other risk estimates.²⁸ For example, cardiovascular risk biomarkers, and sensitive markers of inflammation, myocyte necrosis, vascular damage, platelet activation, and hemodynamic stress, have all been suggested to be of benefit in assessing risk for acute coronary syndrome. Some, but not all, studies have indicated that combining different pathogenic components of risk pathways adds utility to risk assessment.²⁷ Similarly, it has been suggested that multimarker approaches enhance risk prediction in oncology,²⁹ including in diffuse large B cell lymphomas³⁰ and glioblastoma multiforme.³¹ Most RNA-based studies employ multigene fingerprints on molecular patterns. Pattern-based analyses have been associated with differences in breast cancer survival rates.^{32, 33, 34, 35} One strategy—combining multimarkers, nonparametric data analysis, and integration of biomarker data with clinical variables (i.e. age, diastolic blood pressure, serum albumin, and method of vascular access)—augmented the capacity of plasma cytokine levels to predict cardiovascular outcomes in dialysis patients.³⁶ This type of approach might well be effective to define risk and prognosis in AKI.

Figure 1 Chronology of biomarker development.

After identification, a biomarker is evaluated for a particular clinical indication. Both analytical and clinical validation are necessary before FDA evaluation in the US (not necessary for research-only purposes). Following FDA approval, the CMS determines if the biomarker is "reasonable and necessary" for improvement of patient care and, therefore, reimbursable. Abbreviations: AKI, acute kidney injury; CMS, Center for Medicaid and Medicare Services. Permission obtained from Nature Publishing Group © Ludwig JA and Weinstein JN (2005) Nat Rev Cancer 5: 845–856

Full figure and legend (28K)Figures & Tables indexDownload Power Point slide (140K)

The Long and Uncertain Path to Biomarker Utility

Despite intensified interest and investment in biomarker development in recent years, surprisingly few biomarkers are used in clinical practice.³⁷ Moreover, the rate of introduction of valid biomarkers into clinical medicine is falling over time rather than increasing. Since 1998, the rate of introduction of new protein analytes that have received FDA approval has fallen to an average of one biomarker per year.³⁸ This drop has led many observers to conclude that the potential clinical utility of biomarkers has yet to be realized.^{39, 40} Even those biomarkers that are FDA-approved are rarely used in standard clinical practice. For example, of the more than 20 FDA-approved cancer biomarkers, only two have been incorporated into clinical practice guidelines for the Tumor, Node, Metastasis (TNM) Staging System. There is frequently a gap in the progress of biomarker development between discovery (candidate biomarkers are often discovered during investigator-initiated, government-funded research) and the more distal stages of clinical validation (often facilitated by manufacturers of commercial in vitro diagnostics). Weaknesses have been identified in the middle section of the biomarker pipeline, a stage at which development is often not funded specifically; as a consequence, knowledge gaps often occur at this point.³⁷

Despite much recent attention in biomarker development being directed towards high throughput '-omic' technologies (e.g. genomics, proteomics and metabolomics), and towards in silico analyses of databases, advances in basic biologic understanding of multiple disease processes have not led to a considerable increase in the number of biomarkers approved by the FDA. An additional challenge for these high throughput methodologies is the lack of technology standardization. Further, most attempts at biomarker validation have been made using relatively small retrospectively-collected datasets. Few prospective studies have validated initial findings in diverse patients in multiple institutions, or done so using well-established standardization of all steps in the biomarker development process.⁴¹ Finally, the requisite multivariate statistical approaches are prone to overfitting, particularly when more flexible, non-linear algorithms are used. The problem of overfitting data is particularly relevant when multimarker patterns are being evaluated.⁴² Biomarker identification and testing with high throughput technologies also faces enormous statistical issues related to multiple comparisons, particularly when the search for a biomarker is 'unbiased' (e.g. when genome-wide comparisons are made between cases and controls).⁴³ Concerns about false-positive findings are particularly relevant when biomarker testing is performed in a population with a low expected rate of disease prevalence, a phenomena which has been referred to as the 'incidentalome'.⁴⁴

Studies employing information-rich patterns in which individual analytes are not identified are particularly problematic due to issues of technology standardization. An example is the use of surface enhanced laser desorption/ionization (SELDI)-proteomic profiles of disease, where few of the 'unidentified flying particles' that are components of published patterns have been identified, and data reproducibility has been problematic.^{38, 45, 46, 47} Ultimately, the extreme nature of human biologic variability, coupled with stochastic uncertainties in many disease processes, contributes to tremendous observed heterogeneity in clinical outcomes. These formidable challenges to fulfillment of the promise of biomarkers are frequently underestimated during the early discovery phase. To be clinically useful, a biomarker must be relevant to the individual patient, not just to a population of patients with the same disease process.

Phases of AKI for which specific biomarkers are needed

For the purpose of this Review, we discuss biomarkers of AKI within the framework of the following four distinct phases: risk identification; surveillance; diagnosis; and prognosis. Although this is an arbitrary classification, and some biomarkers might be useful in several disease phases, it provides a pathophysiologic and clinical framework that is a useful basis for therapeutic decisions. We have also grouped potential biomarkers into four broad categories, again for clinical reasons; these categories are clinical, functional, structural and inflammatory biomarkers. These groupings according to AKI phase and biomarker type are shown in Table 1, which includes some examples of known and potential biomarkers. Note that different categories can be complementary, such that utilization of all information will yield improved approaches. The ability to combine different biomarkers from different categories, and even within the same category, will be important for the development of specific diagnostic panels, which is currently underway. For example, the combination of a functional biomarker such as glomerular filtration rate (GFR) and a structural biomarker such as sodium hydrogen exchanger isoform 3 (NHE3) could have synergistic effects. Both biomarkers are altered in prerenal azotemia and acute tubular necrosis, but the change in GFR would be rapidly reversible, and NHE3 urinary levels lower, in prerenal azotemia. As such, prerenal azotemia and acute tubular necrosis could be rapidly differentiated.

Table 1 Clinical phases of acute kidney injury and types of biomarkers that are useful in each phase.

Full tableFigures & Tables indexDownload Power Point slide (180K)

Surveillance phase

Monitoring patients to detect the early events that warn of impending AKI has been a very challenging area of clinical investigation. As kidney injury, for the most part, lacks clinical signs or symptoms, the physician is left with biochemical tests that are nonspecific, insensitive and often not ordered in a timely fashion. Decreased urine output and blood pressure (Table 1) have been associated with development of both prerenal azotemia and acute injury. The use of insensitive markers of AKI (i.e. urine output, blood urea nitrogen and serum creatinine) has been blamed for the failure of therapeutic trials in AKI despite promising preclinical studies.^{52, 53}

Serum creatinine level has traditionally been used as a surveillance marker of AKI. Recent prospective data indicate that increases in serum creatinine level of as little as 25.5mol/l (0.3 mg/dl) portend a worse prognosis, and that patient outcomes worsen as the concentration of serum creatinine rises above this point.⁵⁴ Unfortunately, many factors that are independent of changes in renal function affect plasma creatinine, making it a relatively poor biomarker of GFR in AKI. Creatinine concentration is an insensitive indicator of baseline GFR and of alterations in GFR under physiologic conditions. As AKI is a non-steady-state condition, the increasing serum creatinine level during AKI cannot be used to estimate GFR using standard formulas. The rate of generation of creatinine in serum varies with age, race, state of hydration, muscle mass, metabolic state and muscle injury. Creatinine excretion is principally a result of glomerular filtration; however, tubular secretion is responsible for creatinine excretion to an increasing relative degree as GFR decreases. Further, some drugs and organic compounds block creatinine excretion. Finally, the increase in serum creatinine level after the onset of AKI is slow, and can be retarded further by enhancing the hydration status of the patient. This latter phenomenon often masks reductions in GFR in postsurgical patients who are given large volumes of fluid for blood pressure control. For example, mortality rates increased in one study of post-cardiovascular surgery patients as creatinine levels decreased by more than 25.5mol/l (0.3 mg/dl) from baseline.⁵⁴

Also important is the fact that by the time a change in serum creatinine concentration has been observed, a critical therapeutic window might have been missed. Cystatin C concentration has been shown to be superior to serum creatinine level for prediction of the occurrence of AKI in patients with sepsis.⁵⁵ Finally, it is standard practice to measure serum creatinine levels on a daily basis, meaning that AKI occurring today might not be detected until tomorrow's blood work results are reviewed. The delay in diagnosis that is attributed to inadequate sampling can be easily remedied.

These shortcomings in the utility of serum creatinine level as a biomarker of AKI have hampered clinical trials, and led to the genesis of the Acute Dialysis Quality Initiative and the Acute Kidney Injury Network. These organizations represent the efforts of workgroups that seek to develop consensus and evidence-based statements in the field of AKI. The Acute Dialysis Quality Initiative developed a set of criteria, based upon serum creatinine and urine output, that has recently been modified by the Acute Kidney Injury Network.⁵⁶ These criteria are standards for diagnosing and classifying AKI, and have become the subject of early validation studies in selected groups of patients.^{57, 58, 59} Figure 2 indicates how the phases of AKI for which biomarkers are required can be related to a working clinical model of the disorder. Data accumulated using the RIFLE classification system have repeatedly shown that the severity of kidney injury, as determined by serum creatinine level, correlates with patient mortality.^{1, 8} Unfortunately, this classification system does not allow stratification of early-stage AKI patients by severity of injury in a timeframe that is useful to the clinician for therapeutic purposes. Also, the data on the association between change in serum creatinine concentration and mortality do not prove cause and effect. Until an intervention is shown to reduce both the magnitude of changes in serum creatinine level and mortality, the validity of creatinine as an end point will continue to be questioned.

Figure 2 Association between clinical stages of acute kidney injury and phases of biomarker utility.

Increasing severity of the clinical stages of acute kidney injury is indicated by increasing color intensity. Phases of potential biomarker utility are aligned with their most relevant clinical stages. Abbreviation: GFR, glomerular filtration rate. Permission obtained from BioMed Central Ltd © Mehta RL et al. (2007) Crit Care 11: R31

Full figure and legend (13K)Figures & Tables indexDownload Power Point slide (127K)

Biomarkers developed for diagnostic purposes are now in early clinical studies (see next section). Of these, three might be useful for patient surveillance, as their concentrations increase very early in the course of AKI; unfortunately, the specificity of these markers in studies of adults is not high. Early elevation of the urine levels of neutrophil gelatinase-associated lipocalin (NGAL) and IL-18, and of NGAL in serum, have been observed in clinical studies.^{60, 61, 62, 63} These markers are hallmarks of inflammation, which might be one of the first signs of systemic or local kidney injury. Another early-stage indicator of altered renal function or kidney injury is urinary NHE3, which has been found at increased levels in both prerenal azotemia and AKI, and, therefore, has potential as a surveillance marker.⁶⁴ The lack of overlap between urinary levels of NHE3 in patients with prerenal azotemia and in those with AKI could provide a means of differentiating between these two common (and commonly overlapping) disease states. As NHE3 is expressed in the proximal tubule brush border membrane, increased levels in urine provide information about the site of injury.

Conclusions

Extensive work needs to be done if biomarkers are to be successfully incorporated into clinical practice for AKI. There is an urgent need to continue discovery, characterization and validation of novel biomarkers. The current repository constitutes a fraction of the serum and urinary biomarkers that could be discovered using advanced molecular and analytical methods to study genes (transcriptomics),^{76, 77} proteins (proteomics)^{78, 79} and metabolites (metabolomics).⁸⁰ Biomarkers of AKI need to provide early evidence of risk and injury, and differentiate between types of injury (e.g. that caused by nephrotoxins, ischemia, sepsis and interstitial nephritis). Biomarkers must have the capacity to discriminate between different forms of AKI in both elderly and pediatric populations, as well as in patients with a variety of comorbidities. It is particularly important for biomarkers to have the capacity to differentiate between prerenal azotemia and cellular injury, as the therapies for these two conditions differ considerably. Sophisticated analytical algorithms will be needed in order to predict specific etiologies of AKI on the basis of patterns of urinary biomarker abundance. An artificial neural network was recently employed for this purpose in glomerular diseases.⁸¹

Biomarkers for AKI must be validated in well-designed, cost-effective and adequately powered clinical trials, and compared with a standardized definition of AKI. There is growing consensus that RIFLE and the AKI score provide the long-sought consistent definition of AKI.^{56, 57, 58} A collaborative network of centers for the study of AKI is needed so that clinical information can accumulate and repositories of biomarkers can be maintained; independent biomarker studies could be prohibitively expensive. Biomarkers will need to be identified using assays that are simple, rapid and reproducible, such that timely and cell-specific clinical intervention is possible.

Acknowledgments

The authors acknowledge that portions of this manuscript were derived from a meeting held by the Acute Kidney Injury Network in Vancouver, BC, Canada, 11–14 September 2006. This work was supported by NIH grants DK56223, DK62324 and DK58413 to MDO, and by DK53465 and DK61594, and VA Merit Review, to BAM. Charles P Vega, University of California, Irvine, CA, is the author of and is solely responsible for the content of the learning objectives, questions and answers of the Medscape-accredited continuing medical education activity associated with this article.

References

1. Hoste EA et al. (2006) RIFLE criteria for acute kidney injury are associated with hospital mortality in critically ill patients: a cohort analysis. Crit Care 10: R73 | Article | PubMed |
2. Xue JL et al. (2006) Incidence and mortality of acute renal failure in Medicare beneficiaries, 1992 to 2001. J Am Soc Nephrol 17: 1135–1142 | Article | PubMed | ISI |
3. Xue JL et al. (2007) Longitudinal study of racial and ethnic differences in developing end-stage renal disease among aged Medicare beneficiaries. J Am Soc Nephrol 18: 1299–1306 | Article | PubMed |
4. Hsu CY et al. (2007) Community-based incidence of acute renal failure. Kidney Int 72: 208–212 | Article | PubMed |
5. Chertow GM et al. (2006) Mortality after acute renal failure: models for prognostic stratification and risk adjustment. Kidney Int 70: 1120–1126 | Article | PubMed | ISI | ChemPort |
6. Thakar CV et al. (2005) A clinical score to predict acute renal failure after cardiac surgery. J Am Soc Nephrol 16: 162–168 | Article | PubMed | ISI |
7. Bellomo R et al. (2004) Acute renal failure—definition, outcome measures, animal models, fluid therapy and information technology needs: the Second International Consensus Conference of the Acute Dialysis Quality Initiative (ADQI) Group. Crit Care 8: R204–R212 | Article | PubMed | ISI |
8. Uchino S et al. (2006) Pulmonary artery catheter versus pulse contour analysis: a prospective epidemiological study. Crit Care 10: R174 | Article | PubMed |
9. Vasan RS (2006) Biomarkers of cardiovascular disease: molecular basis and practical considerations. Circulation 113: 2335–2362 | Article | PubMed |
10. Biomarkers Definitions Working Group (2001) Biomarkers and surrogate endpoints: preferred definitions and conceptual framework. Clin Pharmacol Ther 69: 89–95 | Article | PubMed | ISI |
11. Gutman S and Kessler LG (2006) The US Food and Drug Administration perspective on cancer biomarker development. Nat Rev Cancer 6: 565–571 | Article | PubMed | ISI | ChemPort |
12. De Gruttola VG et al. (2001) Considerations in the evaluation of surrogate endpoints in clinical trials: summary of a National Institutes of Health workshop. Control Clin Trials 22: 485–502 | Article | PubMed | ISI | ChemPort |
13. Freedman LS et al. (1992) Statistical validation of intermediate endpoints for chronic diseases. Stat Med 11: 167–178 | Article | PubMed | ISI | ChemPort |
14. Roberts MA et al. (2006) Cardiovascular biomarkers in CKD: pathophysiology and implications for clinical management of cardiac disease. Am J Kidney Dis 48: 341–360 | Article | PubMed | ChemPort |
15. Sargent D and Allegra C (2002) Issues in clinical trial design for tumor marker studies. Semin Oncol 29: 222–230 | Article | PubMed |
16. Gion M et al. (1999) A guide for reviewing submitted manuscripts (and indications for the design of translational research studies on biomarkers). Int J Biol Markers 14: 123–133 | PubMed | ChemPort |
17. Wang TJ et al. (2006) Multiple biomarkers for the prediction of first major cardiovascular events and death. N Engl J Med 355: 2631–2639 | Article | PubMed | ISI | ChemPort |
18. Honda H et al. (2006) Serum albumin, C-reactive protein, interleukin 6, and fetuin A as predictors of malnutrition, cardiovascular disease, and mortality in patients with ESRD. Am J Kidney Dis 47: 139–148 | Article | PubMed | ISI | ChemPort |
19. Diamond GA et al. (1995) Prior restraint: a Bayesian perspective on the optimization of technology utilization for diagnosis of coronary artery disease. Am J Cardiol 76: 82–86 | Article | PubMed | ChemPort |
20. Diamond GA and Kaul S (2004) Prior convictions: Bayesian approaches to the analysis and interpretation of clinical megatrials. J Am Coll Cardiol 43: 1929–1939 | Article | PubMed |
21. Fagan TJ (1975) Letter: nomogram for Bayes theorem. N Engl J Med 293: 257 | PubMed | ISI | ChemPort |
22. Justice AC et al. (1999) Assessing the generalizability of prognostic information. Ann Intern Med 130: 515–524 | PubMed | ISI | ChemPort |
23. van Houwelingen HC (2000) Validation, calibration, revision and combination of prognostic survival models. Stat Med 19: 3401–3415 | Article | PubMed | ChemPort |
24. Lemeshow S and Hosmer DW Jr (1982) A review of goodness of fit statistics for use in the development of logistic regression models. Am J Epidemiol 115: 92–106 | PubMed | ISI | ChemPort |
25. Liu J et al. (2004) Predictive value for the Chinese population of the Framingham CHD risk assessment tool compared with the Chinese Multi-Provincial Cohort Study. JAMA 291: 2591–2599 | Article | PubMed | ISI | ChemPort |
26. Ludwig JA and Weinstein JN (2005) Biomarkers in cancer staging, prognosis and treatment selection. Nat Rev Cancer 5: 845–856 | Article | PubMed | ISI | ChemPort |
27. Morrow DA and Braunwald E (2003) Future of biomarkers in acute coronary syndromes: moving toward a multimarker strategy. Circulation 108: 250–252 | Article | PubMed | ISI |
28. Koenig W (2007) Cardiovascular biomarkers: added value with an integrated approach? Circulation 116: 3–5 | Article | PubMed |
29. Hanash S (2003) Disease proteomics. Nature 422: 226–232 | Article | PubMed | ISI | ChemPort |
30. Alizadeh AA et al. (2000) Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling. Nature 403: 503–511 | Article | PubMed | ISI | ChemPort |
31. Liang Y et al. (2005) Gene expression profiling reveals molecularly and clinically distinct subtypes of glioblastoma multiforme. Proc Natl Acad Sci USA 102: 5814–5819 | Article | PubMed | ChemPort |
32. Perou CM et al. (2000) Molecular portraits of human breast tumours. Nature 406: 747–752 | Article | PubMed | ISI | ChemPort |
33. Sorlie T et al. (2001) Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications. Proc Natl Acad Sci USA 98: 10869–10874 | Article | PubMed | ISI | ChemPort |
34. Sotiriou C et al. (2003) Breast cancer classification and prognosis based on gene expression profiles from a population-based study. Proc Natl Acad Sci USA 100: 10393–10398 | Article | PubMed | ChemPort |
35. van't Veer LJ et al. (2002) Gene expression profiling predicts clinical outcome of breast cancer. Nature 415: 530–536 | Article | PubMed | ISI | ChemPort |
36. Knickerbocker T et al. (2007) An integrated approach to prognosis using protein microarrays and nonparametric methods. Mol Syst Biol 3: 123 | Article | PubMed |
37. Rifai N et al. (2006) Protein biomarker discovery and validation: the long and uncertain path to clinical utility. Nat Biotechnol 24: 971–983 | Article | PubMed | ChemPort |
38. Anderson NL (2005) The roles of multiple proteomic platforms in a pipeline for new diagnostics. Mol Cell Proteomics 4: 1441–1444 | Article | PubMed | ChemPort |
39. Dalton WS and Friend SH (2006) Cancer biomarkers—an invitation to the table. Science 312: 1165–1168 | Article | PubMed | ISI | ChemPort |
40. Wilson JF (2006) The rocky road to useful cancer biomarkers. Ann Intern Med 144: 945–948 | PubMed |
41. Hammond ME and Taube SE (2002) Issues and barriers to development of clinically useful tumor markers: a development pathway proposal. Semin Oncol 29: 213–221 | Article | PubMed | ISI |
42. Ransohoff DF (2004) Rules of evidence for cancer molecular-marker discovery and validation. Nat Rev Cancer 4: 309–314 | Article | PubMed | ISI | ChemPort |
43. HunterM DJ and Kraft P (2007) Drinking from the fire hose—statistical issues in genomewide association studies. N Engl J Med 357: 436–439 | Article | PubMed | ChemPort |
44. Kohane IS et al. (2006) The incidentalome: a threat to genomic medicine. JAMA 296: 212–215 | Article | PubMed | ISI | ChemPort |
45. Petricoin EF et al. (2002) Use of proteomic patterns in serum to identify ovarian cancer. Lancet 359: 572–577 | Article | PubMed | ISI | ChemPort |
46. Baggerly KA et al. (2005) Signal in noise: evaluating reported reproducibility of serum proteomic tests for ovarian cancer. J Natl Cancer Inst 97: 307–309 | PubMed | ChemPort |
47. Ransohoff DF (2005) Lessons from controversy: ovarian cancer screening and serum proteomics. J Natl Cancer Inst 97: 315–319 | PubMed | ChemPort |
48. Palomba H et al. (2007) Acute kidney injury prediction following elective cardiac surgery: AKICS score. Kidney Int 72: 624–631 | Article | PubMed | ChemPort |
49. Thakar CV et al. (2005) Influence of renal dysfunction on mortality after cardiac surgery: modifying effect of preoperative renal function. Kidney Int 67: 1112–1119 | Article | PubMed |
50. Jaber BL et al. (2004) Cytokine gene promoter polymorphisms and mortality in acute renal failure. Cytokine 25: 212–219 | Article | PubMed | ChemPort |
51. Jaber BL et al. (2005) Polymorphism of host response genes: implications in the pathogenesis and treatment of acute renal failure. Kidney Int 67: 14–33 | Article | PubMed | ChemPort |
52. Hirschberg R et al. (1999) Multicenter clinical trial of recombinant human insulin-like growth factor I in patients with acute renal failure. Kidney Int 55: 2423–2432 | Article | PubMed | ISI | ChemPort |
53. Allgren RL et al. (1997) Anaritide in acute tubular necrosis. Auriculin Anaritide Acute Renal Failure Study Group. N Engl J Med 336: 828–834 | Article | PubMed | ISI | ChemPort |
54. Lassnigg A et al. (2004) Minimal changes of serum creatinine predict prognosis in patients after cardiothoracic surgery: a prospective cohort study. J Am Soc Nephrol 15: 1597–1605 | Article | PubMed |