Anti-Drug Antibodies (ADA): Tiered Testing, Cut Point and Worked Example

Anti-drug antibodies (ADA) are antibodies that a person’s immune system makes against a therapeutic protein product such as a monoclonal antibody, an enzyme or a hormone. Detecting them is an immunoassay and biostatistics problem: a sample is called ADA-positive only if its signal exceeds a statistically derived cut point, and the result is then characterised in further assay tiers. This guide explains the FDA’s tiered testing sequence, the cut-point formula, a fully worked calculation, and how to report and compute it.

It describes assay methodology for researchers and assay scientists. It is not a guide to interpreting an individual patient’s result; that belongs to the treating clinician.

Definition: what are anti-drug antibodies?

Anti-drug antibodies are antibodies directed against a therapeutic protein product. FDA’s immunogenicity guidance separates them into two practical groups. Neutralizing antibodies can block the efficacy of a product by targeting domains critical for its activity. Non-neutralizing antibodies do not block activity directly, but both groups may change the pharmacokinetics of the product by speeding its clearance and shortening serum half-life, or, conversely, by prolonging half-life (FDA, 2014). A non-neutralizing antibody at high titer may also mistarget the product into cells that carry Fc receptors, and non-neutralizing responses may promote neutralizing ones through epitope spreading.

Safety is part of the reason regulators ask for ADA testing. FDA describes immunologically based events such as anaphylaxis and cytokine release syndrome, and cross-reactive neutralization of an endogenous counterpart protein. Its example is neutralizing antibodies to therapeutic erythropoietin causing pure red cell aplasia by also neutralizing the patient’s own erythropoietin (FDA, 2014). For background on the antibody classes involved, see the comparison of polyclonal and monoclonal antibodies; for the regulatory context of follow-on products, see FDA biosimilars.

The tiered testing sequence: screening, confirmatory, titration, neutralization

FDA recommends a multi-tiered approach (FDA, 2019). Each tier answers a different question, and a sample only moves to the next tier if it passes the previous one.

1. Screening assay. A sensitive assay that flags potentially positive samples. The cut point that separates screen-positive from screen-negative is set statistically (next section).
2. Confirmatory assay. Screen-positive samples are tested to demonstrate that the antibodies are specific for the therapeutic protein product, for example with a competition assay.
3. Titration assay. Confirmed positives are further characterised to describe the magnitude of the response.
4. Neutralization assay. Confirmed positives are also tested for neutralizing activity. FDA notes that neutralization assays may not reach the sensitivity expected of screening and confirmatory assays.

FDA recommends that screening and confirmatory IgG and IgM ADA assays achieve a sensitivity of at least 100 ng/mL. Historically it recommended at least 250 to 500 ng/mL. For IgE ADA assays the expectation is in the high pg/mL to low ng/mL range (FDA, 2019). Sensitivity is expressed as the mass of antibody detectable per mL of undiluted matrix. In FDA’s own illustration, an assay sensitivity of 50 ng/mL with a minimum required dilution (MRD) of 20 is reported as 1,000 ng/mL (50 × 20).

Formula: how the ADA cut point is calculated

The screening cut point is the signal above which a sample is called potentially positive. FDA’s approach targets a 5% false-positive rate among treatment-naive samples, so the cut point is an estimate of the 95th percentile of the negative (naive) population. Under a normal distribution, that percentile is estimated as:

Cut point = x̄ + 1.645 × s

• x̄ is the mean signal of the treatment-naive samples (each sample’s reportable value is the average of its replicate measurements).
• s is the sample standard deviation of those values. See standard deviation: formula and worked example.
• 1.645 is the one-sided standard normal critical value (z) that leaves 5% of the distribution above it. The same logic underlies a z-score.

FDA gives the design around it: generally around 50 treatment-naive samples, each tested by at least two analysts on at least three days, so each sample yields at least six measurements, and the reportable value is the average of those six. To give high assurance of the 5% false-positive rate, one approach is a 90% one-sided lower confidence interval for the 95th percentile of the negative population (Shen et al., 2015, as cited by FDA). When the data are not normal, FDA also describes using the median and the median absolute deviation instead of the mean and standard deviation (FDA, 2019).

Worked example with the arithmetic shown

Suppose a screening assay is run on n = 50 treatment-naive serum samples, each measured by two analysts on three days, and each sample’s six measurements are averaged. The 50 averaged signals, in optical density (OD) units, have a mean of 0.100 and a standard deviation of 0.012. These are illustrative numbers chosen for the arithmetic, not data from a real assay.

1. Multiply: 1.645 × 0.012 = 0.01974.
2. Add the mean: 0.100 + 0.01974 = 0.11974.
3. Round to the precision of the readout: cut point = 0.120 OD (0.1197 before rounding).

Now classify two study samples, each reported as an average signal:

• Sample A = 0.131 OD. Its distance above the naive mean is (0.131 − 0.100) / 0.012 = 2.58 standard deviations. Because 0.131 is above 0.1197, it is screen-positive and goes to the confirmatory assay.
• Sample B = 0.115 OD. Its z is (0.115 − 0.100) / 0.012 = 1.25. Because 0.115 is below 0.1197, it is screen-negative.

Interpretation: reading the result at each tier

Result	What it means	Next step (FDA, 2019)
Signal below the screening cut point	Screen-negative; treated as ADA-negative at the assay’s sensitivity	None within the tiered scheme
Signal above the screening cut point	Potentially positive; about 5% of truly naive samples are expected to land here by design	Confirmatory assay
Confirmed positive	Antibodies are specific for the therapeutic protein product	Titration and neutralization assays
Positive in the neutralization assay	Antibodies show neutralizing activity in that assay	Assess against pharmacokinetic, pharmacodynamic and clinical data (FDA, 2014)

A positive ADA result alone does not establish clinical impact. FDA’s guidance points to pharmacodynamic biomarkers for assessing interference and says correlation with clinical response is usually needed (FDA, 2014).

Assumptions and common mistakes

• Using the wrong population. The cut point comes from treatment-naive samples. Mixing in treated or pre-exposed samples shifts both the mean and the spread.
• Assuming normality without checking. The mean + 1.645 SD formula estimates the 95th percentile only when the negative-sample signals are approximately normal. See how to check the normal distribution assumption; FDA describes the median and median absolute deviation as an alternative.
• Too few samples or runs. The FDA design of around 50 samples, two analysts and three days exists so that analyst and day variability are in the estimate.
• Reporting diluted sensitivity. Sensitivity is stated per mL of undiluted matrix, so multiply by the MRD (50 ng/mL × 20 = 1,000 ng/mL in FDA’s example).
• Treating a screen-positive as confirmed. Specificity for the product is shown only by the confirmatory tier.
• Confusing the cut point with a detection limit. The cut point is a statistical threshold on naive samples, not a limit of detection; see limit of detection versus limit of quantitation, and for plate-assay calibration four-parameter logistic ELISA curve fitting.

How to report an ADA cut point (APA example)

In a methods section, state the population, the sample size, the statistic and the target error rate:

The screening cut point was set at 0.120 OD, calculated as the mean plus 1.645 standard deviations of 50 treatment-naive samples (M = 0.100, SD = 0.012), corresponding to a targeted 5% false-positive rate (U.S. Food and Drug Administration [FDA], 2019).

APA 7th edition reference entry:

U.S. Food and Drug Administration. (2019). Immunogenicity testing of therapeutic protein products: Developing and validating assays for anti-drug antibody detection [Guidance for industry]. https://www.fda.gov/media/119788/download

How to calculate the cut point in Excel, R or Python

Excel

With the 50 averaged signals in cells A2:A51:

=AVERAGE(A2:A51) + 1.645 * STDEV.S(A2:A51)

R

x <- c(...)   # 50 averaged naive-sample signals
cut_point <- mean(x) + qnorm(0.95) * sd(x)   # qnorm(0.95) = 1.6449

Python

import numpy as np
x = np.array([...])   # 50 averaged naive-sample signals
cut_point = x.mean() + 1.645 * x.std(ddof=1)   # ddof=1 gives the sample SD

With a mean of 0.100 and a standard deviation of 0.012, each returns 0.1197 to four decimals (qnorm(0.95) = 1.6449 gives the same value at that precision).

Frequently asked questions

What is the difference between ADA and neutralizing antibodies?

ADA is the broad group of antibodies that bind a therapeutic protein product. Neutralizing antibodies are the subset that block its activity; non-neutralizing ADA can still alter clearance and half-life (FDA, 2014).

Why is the cut point set at the 95th percentile?

It accepts a 5% false-positive rate in the screening tier, as one approach to a sensitive screen, with the confirmatory tier then testing screen-positive samples for specificity (FDA, 2019).

How sensitive does an ADA assay need to be?

FDA recommends at least 100 ng/mL for screening and confirmatory IgG and IgM assays (FDA, 2019).

What does a positive ADA test mean for a patient?

That depends on the product, the assay and clinical context, and is a matter for the treating clinician and the product’s prescribing information.

Sources

• U.S. Food and Drug Administration. (2019, January). Immunogenicity testing of therapeutic protein products: Developing and validating assays for anti-drug antibody detection. Guidance for industry. fda.gov/media/119788/download. Date checked: 2026-10-29. Guidance is non-binding.
• U.S. Food and Drug Administration. (2014, August). Immunogenicity assessment for therapeutic protein products. Guidance for industry. fda.gov/media/85017/download. Date checked: 2026-10-29.