Limit of Detection
The ability of a medical device to detect low concentrations of an analyte can be critical in diagnosing medical conditions or exposure to toxins or other harmful agents. While there is a general agreement among regulators and practitioners on the concept of the limit of detection, there is less agreement on how it should be estimated. There is also some confusion stemming from the use of other terms (e.g., sensitivity) to describe the limit of detection. The limit of detection of a medical device is influenced by several factors such as the type of measurement (potentiometric vs. spectrometric), the specificity of the detection scheme and the sample matrix. Because the sample matrix is highly influential, it is necessary to evaluate the limit of detection using samples that are representative of those encountered during intended use. It is also important to capture as much measurement variation as possible by using multiple samples and analyzing them over multiple days with multiple devices.
The Food and Drug Administration (FDA) regulates medical devices in the U. S. A. For estimating the limit of detection, the FDA recommends the protocols published by the Clinical and Laboratory Standards Institute (CLSI.) The approach suggested by CLSI relies on characterizing the results obtained when blank and low-level samples are analyzed. Blank samples are samples that do not contain the analyte of interest but they are otherwise representative of samples encountered during intended use (e.g., blood or sputum.) To characterize the results of blank samples, a relatively large number of blank samples (usually 60 or more) is used. The results of the blank samples represent a distribution of results obtained when the analyte is not present. A certain percentile of this distribution (typically the 95th) is defined as the Limit of the Blank (LoB). By doing so, we are implying that there is a 5% chance that the analyte will be reported as present (i. e., detected) when it is not present in the sample. This 5% represent the probability of false positive and is referred to as α . Setting the LoB at the 95th percentile ( α = 0.05) is common, but other values (99th percentile, α = 0.01) may be used.
Once the LoB is established, the results of samples containing a low-level of the analyte are used estimate the Limit of Detection (LoD). This step also requires a relatively large number of samples (usually 60 or more) with a concentration between LoB and 4 times the LoB. (Since we do not know the LoD, we may need to try several concentrations.) The LoD is defined as the concentration whose distribution of results shows no more than 5% of the results below the LoB. The 5% represents the probability of not detecting the analyte when it is present. This is known as β, the probability of false negative. Setting β = 0.05 is common, but other values may be used.
The estimation of LoB and LoD described above makes no assumption about the distribution of the results of the blank samples or the results of the low-level samples. There are many equations in the literature that are meant to estimate the LoB and LoD, but they make certain assumption about the nature of the distributions of results (usually normal distributions are assumed) and the uniformity of the variance of the results. You can use these parametric methods, but it is recommended that you verify that your data satisfy the assumptions involved.