**5. Implementation of Biomarkers in Observational Studies**

Owing to the importance of improving risk stratification beyond traditional kidney measures and to help clinical decision-making in CKD patients, the evaluation of novel biomarkers acquired great emphasis in clinical research, as witnessed by the growing number of publications on this topic [114]. However, before a biomarker can find full application in clinical practice, several steps need to be satisfied and reported. The first questions that should be addressed are the following: What will be the clinical intended use of the biomarker? Is the assay analytic performance acceptable for the intended use? To answer these important questions, the development process should start with assessing the analytic and clinical validity of the biomarker. Analytic validity refers to evaluating whether the characteristics of the measured biomarker are acceptable in term of precision, accuracy, and reproducibility [115]. It is indeed important to be aware that biomarker levels may vary in clinical practice due to factors not linked to the disease of interest being classified as pre-analytical and analytical factors [116]. Pre-analytical variation depends on several factors that include lifestyle (exercise, smoking habit, obesity), age, race, influence of gender, specimen collections (fasting, time, and temperature of storage) [117]. For instance, albuminuria, measured with the available methods, such as 24-h urine collection or albumin-to-creatinine ratio, is extremely influenced by physical exercise and other conditions that determine a day-by-day variation, defined as random variation [37]. Urinary NGAL concentration is stable in urine for up to seven days, but it is increased by the presence of white blood cells that are an important confounder [118]. Analytic variation is mainly defined by two parameters, which are bias and precision [119]. Bias is the amount by which an average of many repeated measurements made using the assay systematically over- or underestimates the true value. Precision represents the repeatability of measurements under unchanged assay conditions in a laboratory. While analytic validation is often discussed, it is seldom handled in a proper fashion. It was suggested to deepen analytic validation, while developing a biomarker, and to report metrics, such as precision, reproducibility, accuracy, analytic sensitivity, limits of detection and quantification, linearity, and analytic specificity [120]. A descriptive summary of these measures is reported in Table 2.


**Table 2.** Principal tools used to assess analytic validation.

Clinical validity is the next important step and consists of demonstrating that biomarker measurement is associated with a clinical characteristic of interest [115]. The first steps of clinical validation are the proof of concept and prospective validation [114]. Proof of concept means to assess whether biomarker levels differ between subjects who develop the event of interest vs. non-events. This phase is essential since it allows understanding if the biomarker can play a role in the context of disease, and continuing its development is convenient. To this aim, a cross-sectional design could be sufficient [121]. Next, it is necessary, in prognostic validation, to evaluate if the biomarker is significantly associated with the event of interest, with a prospective analysis. Moreover, it is important that the magnitude of this association is not attenuated when the analysis is adjusted for traditional risk factors, such as age, gender, and proteinuria and eGFR levels in CKD patients. This step provides other useful information such as the distribution of the biomarker and, therefore, how to incorporate the biomarker levels in multivariable analyses. It is suggested to start by adding to the model the biomarker variable as a continuous variable, before applying a categorization (e.g., tertiles or quartiles) [122]. For instance, proteinuria has a skewed distribution and is often added as a log-transformed variable or restricted cubic spline in CKD prognostic models [27]. The prospective validation step is also important for selecting variables to be included in the model. This can be done by using a knowledge-driven (or a priori) method, based on the already known biological association of the variables with the outcome, or data-driven methods, which are automated tools that select a small set of variables from a larger one, in order to maximize the model fit [123]. In the case of a large number of predictors, as often happens during the development of proteomic/metabolomic classifiers, regularization or dimension reduction methods can be used [124]. The metabolomic classifier for the prediction of response to ARB treatment, which we described in Section 4, was developed by means of least absolute shrinkage and selection operator (LASSO), a regularization technique that shrinks the variables regression coefficients through a tuning parameter and retains the best predictors in the model. LASSO was also shown to work very well with small sample sizes [112]. The third phase of clinical validation is focused on the incremental value of the biomarker on the previous assessed risk models. In nephrology, what is essentially required in biomarker research is to demonstrate that a biomarker adds information, in the prediction of a defined endpoint, on top of already assessed risk factors. This process needs a hierarchical assessment, since a likelihood ratio test (LR test) should be firstly reported to determine if the biomarker remains associated with the endpoint after controlling for previously established risk factors. Next, three measures of performance should be reported: discrimination, calibration, and reclassification [125]. These three domains are important to warrant the applicability of the biomarker predictive performance to the individual patient. Discrimination refers to the ability of the model to attribute a high risk to patients who develop the outcome of interest and, accordingly, a low risk to those who do not [126]. A measure that depicts sensitivity and specificity for all possible thresholds of a biomarker is the receiver operating characteristic (ROC) curve. To evaluate discrimination, it is, thus, suggested to present the ROC derived from the model together with the area under the curve (AUC), also labeled the *c*-statistic [127]. The difference in *c*-statistic between models with and without the biomarker should also be presented. Calibration is the degree of agreement between observed and predicted outcomes. It is suggested to depict calibration graphically by plotting the mean predicted versus mean observed outcome probability for intervals (usually deciles) of risk in a predictiveness curve or by representing observed event rates versus mean predicted risk, thus creating a calibration plot, with points that should lie along a 45◦ line if the model is well calibrated [126]. Reclassification metrics provide useful information on the proportion (%) of patients that are reclassified in the true risk category (lower or higher risk), whether or not the new biomarker is added to a traditional risk prediction model. The most used reclassification metrics are the net reclassification improvement (NRI), the integrated discrimination index (IDI), and reclassification tables which directly depict the movement of patients between risk categories based on the risk predicted by models with and without the biomarker [127]. After showing measures of performance, it would be necessary to internally or externally validate the model. External validation

implies that the risk prediction model, including the biomarker, is re-run within an external cohort of patients with similar characteristics (e.g., CKD patients) to confirm predictive accuracy in all sequences. Alternatively, several methods of internal validation, such as bootstrapping or cross-validation, can be computed [128]. The appropriateness of methodology used to develop a biomarker is a key element to obtain useful clinical results. This is particularly true if we consider that only a few prediction models in nephrology reported these measures appropriately [129]. However, this is not the only limitation. Most of the proposed biomarkers are yet to complete the sequence from discovery to clinical application, because they were developed in studies with a small sample size without validation, thus providing heterogeneous results. The ISN prompted that biomarker research would take advantage from the setting up of large, observational cohort studies and possibly a long-term follow-up in which biomarker development and validation could be strengthened and provide robust evidence for clinicians [18]. This also requires the standardization of data collection, storage, and database structure across countries, as well as a collaboration among academia, industry, and regulatory authorities in order to warrant a correct dissemination of results.

#### **6. Biomarkers in Intervention Studies**

Projecting clinical trials, which test the effect of novel pharmacological treatments on prognosis of CKD patients, is always an important challenge. In the past few decades, all nephrology communities expressed the need for clinicians to have more therapeutic tools, with each one specific for a particular etiology of CKD, in order to improve the care of patients with CKD and to deal promptly with the complexity of kidney disease, abandoning the "reductionist" approach [10,130]. The milestone of intervention studies in nephrology dates back to the years 1990–2000 when the Collaborative Study Group, the RENAAL, and the IDNT trials showed the efficacy of RAAS-i (ACEi and ARBs) in reducing CV and renal risk in patients with diabetes and CKD [76,131,132]. Since then, a number of clinical trials were carried out with an attempt to reduce the high residual risk in CKD patients, but they missed the target [133]. The reasons for this breakdown are several and include the enrolment, in clinical trials, of a large number of CKD patients with heterogeneous etiologies and the add-on strategy. The add-on strategy consists of adding a pharmacological agent to patients who are already being treated with a drug belonging to the same class. This was adopted, for example, in the Veterans Affairs Nephropathy in Diabetes (VA-NEPHRON-D) clinical trial, which tested the effect of dual RAAS blockade ACEi + ARB, or in the ALTITUDE trial, with the addition of Aliskiren, a renin inhibitor, to RAAS-i [134,135]. In these studies, the intensification of RAAS blockade did not result in further CV or renal risk protection and even increased the risk of these endpoints. Hence, a series of initiatives were started to improve clinical trial designs. The focus is indeed to move from large trials to smaller studies that enroll similar patients so that the treatment effect can be adequately measured [26]. Biomarkers play a central role in this context (Figure 2), being useful to enrich clinical trial CKD populations through at least three important ways called biomarker-based approaches: (1) by identifying patients at increased risk for developing an event (risk-based enrichment); (2) by selecting a population based on the response to a drug of interest (predictive response enrichment or adaptive enrichment); (3) by detecting subgroup of similar patients within a master trial protocol [136].

Risk-based enrichment was used in the proteomic prediction and renin angiotensin aldosterone system inhibition prevention of early diabetic nephropathy in type 2 diabetic patients with normoalbuminuria (PRIORITY) study. The PRIORITY study enrolled patients with diabetes mellitus and normal albuminuria at increased risk for developing albuminuria [137]. High or low risk was established based on urine CKD273 levels, and only high-risk patients were then randomized to receive spironolactone or placebo. Although the trials did not show a significant effect of spironolactone on preventing the development of albuminuria, high-risk patients identified with CKD273 were at increased risk of CKD progression vs. low risk patients (*p* < 0.001). PRIORITY was an innovative design, since it anticipated the treatment of albuminuria in patients who were only likely to develop albuminuria, but not yet with albuminuria. The adaptive enrichment design consists of exposing all patients to a short-term period (usually called run-in) of treatment with the drug of interest before randomization. In this case, biomarkers could inform on the response/non-response to treatment. Such a design was adopted in previous trials like the Study of Heart and Renal Protection (SHARP) study and more recently in the study of diabetic nephropathy with the endothelin receptor antagonist atrasentan (SONAR) trial [94,138]. Patients enrolled in SONAR underwent a six-month treatment period with atrasentan, and only patients who manifested a 30% reduction in albuminuria levels (measured as albumin-to-creatinine ratio) were then randomized. Hence, albuminuria worked as a biomarker for the prediction of response to treatment. Moreover, the SONAR trial included the assessment of "secondary" risk markers, such as the B-type natriuretic peptide levels. Patients who showed a significant increase in this marker during run-in were excluded from the study. This strategy allows assessing the individual response to treatment, including the effect of a drug on primary and secondary markers and, thus, to capture in a reliable manner the effect of treatment after randomization. An extension of the adaptive enrichment trial is given by the master trial protocol [139]. Master protocols can be planned to test the efficacy of multiple interventions, each targeting a subgroup of patients defined by a biomarker. Master protocols encompass umbrella, basket, and platform trials. Platform trials aroused the interest of the nephrology community [136]. The platform is an experimental cohort of patients followed periodically to assess laboratory and clinical measurements. Within the platform, multiple treatments can be started or withdrawn and, if a defined treatment shows benefits in a defined subgroup of the platform, it can be introduced in clinical practice [140]. This approach allows the acceleration of the experimental phase of drug development, to improve the application of biomarkers and to save time and financial sources. Future perspectives around the implementation of available biomarkers are depicted in Table 3.

**Figure 2.** Biomarker-based approaches for patient selection in clinical trials.



**Table3.**ValidationscoreandfutureperspectivesinthedevelopmentofbiomarkersinCKDpatients.

+, fully present; +/−, partially present; −, absent. CKD, chronic kidney disease; eGFR, estimated glomerular filtration rate; RRI, renal resistive

 index; CV,

cardiovascular.
