Once the data are collected, one has the proverbial chicken-or-egg decision; in this case, accuracy or precision. Since accuracy
and precision go hand-in-hand, the decision of which to assess first involves personal preference. Here, for the sake of simplicity,
accuracy will be assessed first, then precision. Accuracy is calculated by combining all data across analysts and days for
each level of analyte. Depending on the method being validated, acceptance criteria should be established. For the following
example, one can use a percent recovery of 95–105%. A typical accuracy analysis is shown in Table 6.
Table 7. A typical precision analysis
Without a precision analysis, one cannot confirm accuracy claims. The intermediate precision includes analyst and day, while
the repeatability includes the variability within analyst per day. Each source of variability is assessed and then combined
to yield the intermediate precision and repeatability. A typical precision analysis is contained in Table 7.
Table 8. A typical linearity analysis
Once one has shown acceptable precision and accuracy, one can assess if the bias is constant by performing a linearity analysis.
Using the same data from the accuracy and precision analysis, an ordinary least squares (OLS) estimate can be calculated.
Two coefficients are estimated using OLS: the slope and intercept. A lack-of-fit test confirms that the linear model is appropriate
for the data set. Combining all the data similar to the accuracy analysis yields the linearity analysis contained in Table
8. A graphical representation of the data is shown in Figure 2. The model illustrates a statistically significant slope with
a lack-of-fit test showing that the linear model is sufficient (for lack-of-fit test, a p-value greater than 0.05 is indicative that the model is sufficient). The intercept is not statistically significant (p > 0.05), indicating that the assay would run through the origin.
Since the accuracy, precision, and linearity all meet the requirements, one can state that the range of the assay is 50–150%.
Using a well-designed experiment and statistically relevant methods, method validation can be accomplished in accordance
with the ICH guidelines. Precision analysis is the most critical component because it allows the claims of accuracy and linearity
to be made.
Steven Walfish is president of Statistical Outsourcing Services and BioPharm Editorial Board Member, 403 King Farm Boulevard, Suite 201, Rockville, MD 20850, 301.325.3129, fax: 301.330.2143,
1. Krause SO. Analytical method validation for biopharmaceuticals: a practical guide. Guide to Validation. Supp to BioPharm
Int. 2005 Mar; 26–34.
2. Kanarek AD. Method validation guidelines. Guide to Bioanalytical Advances. Supp to BioPharm Int. 2005 Sep; 28–33.
3. US Food and Drug Administration. Guidance for industry: Q2B validation of analytical procedures: methodology. Rockville,
MD: Nov 1996.
4. US FDA. Guideline for industry: text on validation of analytical procedures: ICH Q2A. Rockville, MD: Mar 1995.
5. Rodbard D. Statistical estimation of the minimal detectable concentration (sensitivity) for radioligand assays. Analytic
6. Hauck WW, Capen RC, Callahan JD, De Muth JE, Hsu H, Lansky D, et al. Assessing parallelism prior to determining relative
potency. PDA J Pharma Sci Technol. 2005;59(2):127–37.
7. Daniel C, Wood FS. Fitting equations to data. 2nd ed. New York: Wiley & Sons; 1980.
8. Dixon WJ, Massey FJ. Introduction to statistical analysis. New York: McGraw-Hill; 1969.
9. Box GEP, Hunter WG, Hunter JS. Statistics for experimenters. New York: Wiley & Sons; 1978.