How should diagnostic assay developers define analytical performance specifications during IVD assay development? Milan Hierarchy

Defining analytical performance specifications is not a one-size-fits-all exercise. It is a deliberate, evidence-based process that directly shapes your assay’s clinical utility and regulatory success. The global consensus framework for this task is the Milan hierarchy. It classifies measurands into three models, letting you set specifications based on the strongest available evidence—ranging from hard clinical outcome data to the realistic limits of current technology.

The Milan hierarchy guides developers to define analytical goals by asking: How much analytical error can we tolerate before it harms patient care? When outcome data exist, you use it; when they don’t, you turn to biological variation. If neither is available, you benchmark against the best-performing assays on the market. Your job is to pick the highest possible model for your measurand and then tie those goals back to every concrete parameter—imprecision, bias, measuring range, and more—using standardized verification protocols.

Understanding the Milan Hierarchy: The Three Models for Specification Setting

The Milan consensus categorizes every measurand into one of three models. The model you select becomes the yardstick for every analytical requirement you will impose on your assay.

Model 1: Direct Clinical Outcome Data

This is the ideal—and the most demanding—source of specifications. You define acceptable analytical error by studying how assay performance directly or indirectly alters clinical decisions and patient outcomes.

For example, cardiac troponin (cTn) assays use outcome-based goals. A coefficient of variation (CV) of 6% at the 99th percentile upper reference limit, compared to a looser 10% CV, dramatically reduces false-positive myocardial infarction misclassifications to just 0.5%. These numbers come from outcome studies, not arbitrary targets. When a measurand plays a central, direct role in a diagnosis or treatment decision, following the outcome evidence is non-negotiable.

Model 2: Harnessing Biological Variation

Biomarkers that maintain a physiological steady state lend themselves to this model. The principle is simple: analytical noise (CVA) should not drown out the natural biological signal. The accepted rule-of-thumb is:

CVA ≤ 0.5 × CVI (where CVI is the within-subject biological variation)

If you meet this, your total analytical error (bias + precision) is small enough to prevent misdiagnosis across sequential patient samples. For glucose monitoring, this translates to an imprecision goal of ≤2.9%, a bias goal of ≤2.2%, and a total error limit of ≤6.9%. These goals are derived from population studies of healthy individuals, making them objective, evidence-based, and widely accepted.

Model 3: Falling Back on State-of-the-Art

When no outcome or biological variation data exist—common for novel biomarkers—you must look outward. Specifications are set at the performance level already achieved by the best commercially available assays or by the top quartile of laboratories in external quality assessment schemes.

While this model guarantees your assay is competitive, it carries a critical risk: the current “state of the art” may not be clinically sufficient. It should always be viewed as a temporary starting point, with the commitment to move to a higher model as clinical evidence accumulates.

A Practical Decision Workflow for Choosing the Right Model

You cannot simply pick a model at random. A logical sequence ensures you always use the strongest possible evidence:

• Assess clinical utility first. If your measurand directly drives a diagnosis or therapy (like HbA1c in diabetes), you search for outcome studies. If robust data exist, Model 1 applies.
• Evaluate biological regulation. For measurands that are tightly controlled physiologically (like electrolytes or hormones), check databases of biological variation components. Valid CV_I and CV_G data let you use Model 2.
• When in doubt, use Model 3 as the floor. For genuinely novel biomarkers without any history, this is your only option. But never confuse it with a clinically validated specification.

Translating the Milan Model into Concrete Analytical Parameters

Choosing a model gives you a goalpost. You then translate that into the six core analytical performance parameters that every IVD assay must document. Each parameter gets its own numeric specification aligned with your chosen model.

• Precision (Repeatability and Reproducibility): Your imprecision (CV) must stay below the limit defined by your model. For a Model 2 measurand, this means the CV must meet the CVA ≤ 0.5 × CVI rule.
• Trueness (Bias): Any systematic shift from a reference method must be small enough to avoid clinically significant misclassification, again defined by your model’s acceptable bias or total error budget.
• Analytical Measurement Range (AMR): The linear reportable range must cover the concentrations critical for clinical decisions, with imprecision and bias meeting specifications at every point.
• Limit of Detection (LoD): The blank limit and LoD must be set low enough to detect the measurand at its lowest clinically relevant level, especially for Model 1 assays like troponin.
• Analytical Specificity (Interference): Specifications for common interferents like hemolysis, icterus, and lipemia must be such that the measured value remains within the model’s total error limits even in a matrix that mimics patient samples.
• Assay Ruggedness and Stability: Reagent lot-to-lot consistency, calibrator commutability, and environmental robustness must be controlled to preserve the established specifications throughout the product lifecycle.

For every one of these parameters, you use standardized CLSI protocols (EP05 for precision, EP06 for linearity, EP07/C56‑A for interference, EP09 for method comparison, EP17 for LoD, etc.). This connects your strategic goals to the daily language of regulatory submissions.

Understanding the Trade-offs and Pitfalls

The Milan hierarchy is powerful, but it is not without blind spots. Developers who misunderstand these limitations create assays that look great on paper but fail in practice.

• Model 1’s data drought. High-quality outcome studies are expensive and rare. Even for troponin, the 6% CV target is debated and may not apply to all patient populations. Misapplying a single outcome study to a new assay without verifying its own clinical context is a dangerous shortcut.
• Model 2’s false assumptions. Biological variation data assume a steady state. For measurands like cortisol or inflammatory markers, CVI changes with disease, circadian rhythm, and stress. Using a fixed CVI from a healthy cohort can set limits that are too wide to catch meaningful clinical changes.
• Model 3’s complacency trap. If the best available assay has a 15% CV, that does not make a 15% CV clinically acceptable. It simply means the market has not yet been forced to improve. Setting a Model 3 specification without a roadmap to migrate upwards risks locking your product into mediocrity.
• Ignoring pre-analytical factors. Statistical models that estimate CVA and CVI from quality control data without screening for outliers, sample heterogeneity, or pre-analytical errors will produce falsely precise specifications. Use rigorous variance decomposition (ANOVA, REML) and validate normality assumptions to prevent your analytical goals from being built on a swamp.

Making the Right Choice for Your Assay Development Goal

Your final specification strategy must be tailored to the specific job your assay is meant to do. Use these decision anchors to guide your next step.

• If your primary focus is a biomarker that directly dictates a life-or-death clinical decision (e.g., troponin, glucose): Prioritize Model 1 outcome data. Define your imprecision goal at the clinical decision limit based on the acceptable misclassification rate, and align all LoD, AMR, and interference specifications to protect that decision.
• If you are developing a general chemistry or endocrine assay with stable physiology and established biological variation databases: Adopt Model 2. Enforce the CVA ≤ 0.5×CVI rule as your non-negotiable core, and calculate total allowable error bounds that confirm your assay can safely track a patient over time.
• If you are working with a novel biomarker where no outcome or variation data exists: Start with Model 3 as your immediate benchmark, but immediately initiate your own clinical evidence generation. Set internal stretch goals that shrink performance limits as you gather more data, and communicate to stakeholders that this is a phase‑I specification only.
• If your primary concern is regulatory submission and supply chain consistency: Embed your chosen model’s targets into every raw material selection and calibrator assignment decision. Use CLSI EP26 to enforce lot-to-lot consistency, and demand that your antibody and enzyme suppliers provide batch records that align with your total error budget.

Your analytical performance specification is not a box to check—it is a hypothesis about how much measurement error your clinical decision can tolerate. Ground that hypothesis in the strongest evidence available, verify it ruthlessly, and let it drive every technical choice from raw material to the final patient report.

Summary Table:

Turn Analytical Specifications into Market-Ready Assays with CamelBio

Defining robust analytical performance specifications under the Milan hierarchy is only the first step—achieving them requires uncompromised raw material quality and rigorous assay design.

At CamelBio, we provide diagnostic manufacturers, laboratories, and research institutes with one-stop access to high-performance IVD raw materials, technical services, and consulting—covering every stage from concept to clinic. Whether you are optimizing total error budgets, establishing lot-to-lot consistency, or navigating regulatory submissions, our team is here to accelerate your development timeline.

Ready to elevate your IVD assay performance? Contact CamelBio Today to discuss your raw material and assay development needs!