【分享】环境监测中的分析灵敏度与功能灵敏度

Not only at the analytical sensitivity, but even at concentrations significantly above it, imprecision may be so great that results do not reproduce well enough to be of real clinical utility. Consequently, analytical sensitivity does not typically represent the lowest measurable concentration that is clinically useful.

This limitation of analytical sensitivity has always been with us, for RIA and IMMULITE, and applies to all methods by all manufacturers. Since patient samples are not typically run in replicates, the lack of reproducibility may not be readily apparent in routine testing. However, the overall quality and usefulness of the results are affected. This is why the lower limit of the reporting range in the IMMULITE and IMMULITE 2000 software is often set at a concentration above the analytical sensitivity. DPC sets the reporting limits for every assay to what a comprehensive assessment suggests is the range of effective and reliable performance for the assay, relative to its intended clinical use.

The limitations of analytical sensitivity, for describing the lower limit of clinically effective assay performance, led to the development of another concept.

Functional sensitivity
About a decade ago, in reaction to the limited utility of analytical sensitivity as a measure of assay performance, a group of researchers evaluating TSH assays developed a concept they termed functional sensitivity.1 They defined it as "the lowest concentration at which an assay can report clinically useful results." Clinically useful results for TSH were deemed to be good accuracy with a day-to-day CV of not more than 20%. While this choice of CV limit was somewhat arbitrary, the researchers felt that, for TSH, a CV of 20% was the most imprecision that could be tolerated for clinical purposes.1

Since CV is the standard deviation expressed as a percentage of the mean, a CV of 20% implies the SD would be 20% of the mean. For a sample with a TSH concentration of 0.1 µIU/mL, for example, the range encompassing 95% of the expected results from repeat analysis would be ±40% (±2 SD), or 0.06 µIU/mL to 0.14 µIU/mL.

Although originally developed only for TSH assays, the concept of functional sensitivity and the use of a 20% CV as the limit of clinical usefulness have been widely applied to other immunoassays. The concept has gained acceptance because it provides the laboratory with an objective and clinically meaningful indication of the practical lower limit of an assay.

When developing a new assay, DPC uses essentially the same approach, evaluating both precision and accuracy to establish the concentrations at which the limits of clinical usefulness are likely to be reached. The software reporting range is based on this evaluation. For competitive assays especially, there is usually a significant difference between the analytical sensitivity and the lower reporting limit. The reporting range, as set in the IMMULITE and IMMULITE 2000 software, represents DPC's recommendation for the CLIA'88* "reportable range"-which is the concentration range over which assay performance is documented as valid.

Verifying assay performance
Currently, for laboratories using automated immunoassay systems in the US, the only sensitivity-related performance characteristic that CLIA'88 requires to be verified by the laboratory is the lower limit of the reportable range. Some laboratories may also choose to estimate the functional sensitivity of a new assay; and, historically, some have wanted to verify analytical sensitivity. Each of these assessments is a different experiment with distinct protocols and requirements. So the first step is to decide what is to be verified and then use the appropriate protocol and evaluate the data accordingly.

If a laboratory chooses to evaluate analytical sensitivity, the goal is typically to verify the value given for that performance measure in the package insert. It is essential that the sample used for an analytical sensitivity study be a true zero concentration sample with an appropriate sample matrix. Any other type of sample may bias the results. The usual protocol involves assaying 20 replicates of the zero sample, followed by calculating the mean and SD of the CPS. The analytical sensitivity is estimated as the concentration equal to the mean counts of the zero sample plus 2 SD for immunometric ("sandwich") assays like TSH, or minus 2 SD for competitive assays like T4. Technical Services can assist in calculating this concentration. This protocol yields an initial estimate, which is usually adequate for comparison with the analytical sensitivity listed in the package insert. However, multiple experiments encompassing several kit lots are necessary to obtain a robust and accurate assessment.

In assessing functional sensitivity, the aim is to determine the lowest concentration corresponding to a laboratory-specified goal for day-to-day (interassay) imprecision representing the limit of clinical usefulness for a given assay. Commonly, a CV of 20% has been used as the goal, based on the original application of the concept to TSH. However, this CV may not always be the most appropriate limit. For some assays, a CV greater than 20% may be consistent with clinically reliable and informative results, while for others, a CV less than 20% may represent the limit of clinical usefulness. The performance goal needs to be set for each assay, based on its intended clinical application.

Having determined the day-to-day CV representing the clinically useful limit of reproducibility, the next step is to estimate the concentration at which the CV might reach this limit. On the basis of prior studies, package insert data, and estimates made from the assay's precision profile, Technical Services can usually help to identify a "target range" of concentrations bracketing the predetermined CV limit.

Ideally, this study should be performed using several undiluted patient samples, or pools of patient samples, with concentrations that span the target range. However, these samples may be difficult to obtain. Reasonable alternatives include patient samples diluted down to concentrations spanning the target range, or control materials in or near this range. If there is a need to dilute any type of sample for the study, the diluent used is critical. The routine sample diluents are intended only for diluting very high concentration samples; for some assays, they may have a measurable, though very low, apparent concentration. Use of these diluents can bias the results of the study.

The samples should be analyzed repeatedly over a number of different runs, ideally over a period of days or weeks, to assess the day-to-day precision. (A single run of 20 replicates does not provide a valid assessment of functional sensitivity.) Having collected the data, calculate the CV for each sample tested. The functional sensitivity is the concentration at which the CV reaches the predetermined limit. This concentration can be estimated from the study results by interpolation, if it doesn't happen to coincide with one of the levels tested.

Verifying the lower limit of the reportable range is one part of the process of verifying the entire reportable range. This is typically accomplished by performing replicate analysis on a series of three to five samples with known concentrations spanning the reportable range. These samples can be obtained using a single sample, with a concentration near the upper limit of the range, which is then diluted to give additional samples spanning the entire reportable range. The results obtained are evaluated for both reproducibility and recovery of expected values to determine that the assay's performance meets clinical usefulness needs across the reportable range.

Conclusion
So, why is the lower limit of the software reporting range 1.0 µg/dL (13 nmol/L) when the package insert says the sensitivity is 0.3 µg/dL (3.9 nmol/L)? In this example, the assay is a competitive assay and the imprecision of the assay exceeds clinically useful limits at a concentration well above the analytical sensitivity.

If the imprecision is such that you cannot say with certainty that results of, say, 0.4 µg/dL (5 nmol/L) and 0.7 µg/dL (9 nmol/L) are in fact different, might it not be better to report both as "< 1.0 µg/dL" ("< 13 nmol/L") rather than to risk having a physician interpret explicit results as showing a clinically meaningful difference?

Ultimately, it is usually not the assay's detection limit (analytical sensitivity) but rather the reproducibility of results which determines the lower limit of clinically reliable assay performance in routine practice.

第三篇


分析灵敏度（检测限）

一、 分析灵敏度（检测限）
1． 检测的最低分析物浓度为检测系统的分析灵敏度或称检测限。这个浓度限值对毒检验在法庭上特别重要，希望通过检测知道样品中究竟有无药物，这是很关键的。另外，肿瘤标志物及许多特定蛋白应该有一个可检测的最低浓度或某个量；如：前列腺特异蛋白（PSA），这是病人治疗后监视复发的重要信息；长期以来，临床对报告的前列腺特异蛋白有意义的最小量要求予以明确。核酸检测报告的阴、阳性也要求说明，能检出的最小拷贝的核酸量相当于多少病毒。因此，确定检测系统的可报告低限是重要的分析性能。
2． 当前，检测限术语混乱。厂商使用各种词语，如：灵敏度（sensitivity）,分析灵敏度(analytical sensitivity),最小检测限(minimum detection limit)，功能灵敏度(functional sensitivity)，检测限度(limit of detction)定量限度（limit of quantitation）等。迄今尚无标准定义，所以有必要了解每个词语的实际含义和确定这个含义的实验方式，怎样处理数据，怎样由数据作出估计，以及这个估计对该检验的医学应用是否有用。以下介绍的分析灵敏度分为具有定性含义的检测低限，和具有定量含义的生物检测限及功能灵敏度。
二、 检测低限（Lower Limit of Detection，LLD）
每次检测，总是做一个空白样品。检测方法常以空白响应量校准至零点，再检测各个检测样品的反应响应量。这些样品的反应响应量在扣除了空白样品响应量后，是分析物的对应响应量。但是，空白响应量也有波动。若重复多次作空白检测，以空白（响应量）均值和标准差表示这些空白均值的离散程指标。在确定方法性能或绘制标准曲线时，常常以空白均值表示空白响应量大于或小于空白均值，各有50%的可能性。当空白响应量小于空白均值，对同一个样品检测响应量（未扣除空白响应量），似乎反映分析物要多一点，检测方法好象灵敏些。当空白响应量大于空白均值，似乎原先可以检测出来的分析物现在测不出了。因此，检测方法必须说清楚：究竟怎样才算是可检测出来的分析物量？标准曲线从零开始，是不是报告的分析物量 可以从零开始？这就是检测低限要回答的问题。统计说明，如果空白响应量的波动服从正态分布规律：各个单次检测的空白响应量x空白有95%的可能性为：
—2.s空白≤x空白≤ —空白 2.s空白
即:∣x空白- 空白∣≤2.s空白
其中较空白均值小的一半会使分析物更易检测出来,这不是检测不出,不必考虑.若有一个检测响应量较空白响应量均值大2s空白，仍然认为是空白响应量的可能性只有5%;有95%的可能性属于样品内有分析物形成的检测响应检测响应量；它较空白均值差2s空白
以上。同理，响应量较空白均值相差3s空白以上的，还认为是空白响应量的可能性仅0.3;而有99.7%的可能性是样品内有分析物形成的响应量.所以若检测样品的反应响应量较空白均值大的,但和空白均值相差2s空白或3s空白以下的,只能说这些响应量是空白样品单次检测的响应量,样品没有分析物,或者表示:分析物量为零.超过2s空白或3s空白的响应量才认为样品中真的含有分析物.
检测低限定义为样品单次检测可以达到的检测响应量对应的分析物量。检测系统或方法对小于或等于检测低限的分析物量只能报告“无分析物检出”。通常估计95%或99.7%的两种可能性：
95% 可能性：LLD= 空白 2.s空白
99.7% 可能性LD= 空白 3.s空白
要注意的是,直接读出浓度单位的检测系统对低于零的检测将报告零,其分布不是正态的,因此计算的均值和标准差不能如实表达检测低限的真实情况.若检测响应可以用初始值表示,如:吸光度、荧光、等，此时s空白是有效的。所以应使用初始值来计算均值和标准差，然后再转换成浓度单位。

三、 生物检测限度（Biologic Limit of Detection，BLD）
大于检测低限的响应信号说明这个样品内有分析物，但是方法还不能正确报告定量结果。因为在这样低的浓度或其他量值范围内，单次检测样品的反应响应量重复性较差。那么在多少检测响应量时才能较好地报告定量结果呢？现介绍两种方式：生物检测限度和功能灵敏度。原则上，对多个近于检测限浓度的样品（肯定不是空白样品）作重复检测，对扣除了空白响应量后的样品检测响应量以均值和标准差归纳。按正态分布规律，单次检测样品具有的响应量有95%或99.7%的可能性与不检测响应量均值相差2或3倍的响应量标准差。较均值还大的单次响应量肯定没问题；但是较均值小的那些单次响应量，若和检测低限（空白响应量上限）交叉，说明检测方法还不能单凭一次检测区分出这是空白还是有分析物。因此，这些样品检测响应量的95%或99.7%单次检测量最低值也较检测低限(LLD)大,这样就可保证样品在任何情况下,单次检测响应量一定不是空白的响应量;这个样品具有的分析物浓度可以定量地报告出来.在多个近于检测限的样品中,符合这样条件的最低分析物浓度(或其他量值)就是检测系统或方法的生物检测低限.
生物检测低限定义为:以检测低限加2或3倍检测限样品标准差的方式,确定检测系统或方法可定量报告分析物提取低浓度或其他量值的限值。
生物检测低限（BLD）的具体度量方式为：
95% 的可能性：BLD=LLD 2s检测限样品
99.7%的可能性: BLD=LLD 3.s检测限样品
本词语较完善地表示实际样品的检测限度,如什么浓度才是零值或没有分析物有差异.在证实厂商的BLD说明时，应使检测限样品和浓度的厂商的说明相同。
四、 功能灵敏度（Functional Sensitivity ，FS）
功能灵敏度定义为：以天间重复CV为20%时对应检测限样品具有的平均浓度，确定检测系统或方法可定量报告分析物的最低浓度或其他量值的限值。为了估计FS，须用多个检测限浓度来确定在低浓度处的精密度表现，从中选择具20%CV的浓度。在证实厂商的FS的明时，使用的检测限样品浓度应和厂商的说明相同。
五、 实验须考虑的因素
一般制备两种不同类型的样品。一个是“空白”样品，即不含有分析物，分析物浓度为零。另一个是“检测限”样品，即含有低浓度的分析物。在有些场合下，需要制备几份“检测限”样品。空白和检测限样品由检验方法作重复检测。计算各自的均值和标准差。不同的检测限由空白和检测限样品数据计算出来。
1． 空白液：一份空白液用作空白，其他用于制备检测限样品。理想的空白液应具有和检验的病人样品相同的基体。但是，常使用检测系统的系列标准品中的“零标准”作为空白。对某些项目，可使用术后无某疾病的病人样品（如：前列腺肿瘤术后病人的无PSA血清）为空白样品。
2． 检测限样品：在证实某方法的灵敏度性能时，对空白液加入分析物配制成检测样品，。加入的分析物量应是厂商说明的检测浓度。在建立检测限度时，有必要制备几份检测限样品，它们的浓度应界于预期检测限度高低一些的范围内。重复检测数：没有具体规定，但常推荐做20次。符合临床检验对重复检测实验的要求。厂商常推荐10次，为减少开支实验室也常采纳作10次
3． 实验需要时间：如果主要从空白液的重复性了解检测低限，常常就做批内或短期实验。如果主要从“检测限”样品的重复性了解定量的检测限，推荐作较长时间的实验，代表天间测定性能。实际就做10次检测（10d）。
计算示例：
某酶标法的甲状腺球蛋白试剂盒的分析灵敏度实验结果如下：
1． 用甲状腺切除后恢复健康的病人血清为无甲状腺球蛋白的空白血清。将纯化的甲状腺球蛋白加入该血清，并用空白血清作系列稀释，组成0，0.25,0.50,1.0,2.0,4.0,8.0ug/L的甲状腺球蛋白样品.
2． 空白血清作12次批内重复测定;对其它系列样品作天间重复测定,做了12d.记录所有结果,

1. 检测低限
空白”0”组的均值为0.098,标准差为0.0137。这里采用99.7%的可能性来计算检测低限。如上所述：
LLD= 0 3s0，现 0=0.098,SO=0.0137,
∴ LLD= 0 3s0=0.098 3×0.0137=0.139
实际上,若以空白吸光度均值为检测的起点0,那么有99.7%的可能性,每次只做一个空白时,出现最高的空白吸光度较该空白吸光度均值高0.0137的3倍,即0.041A。这些吸光度相当于样品具有的甲状腺球蛋白即为本法的检测低限。
其实，上述的实验结果可以画出该检测系统在低浓度范围的标准曲线。将上面所有的吸光度减去空白吸光度均值，然后求每组的均值和标准差。见表2。
现在0.25ug/L组减去空白后的吸光度均值为0.07625A.假定这段范围内甲状腺球蛋白量和吸光度间呈线性.因此0.041A相当于
0.25× 0.041 =0.13ug/L
0. 07625
所以本法的检测低限为0.13ug.L

甲状腺球蛋白由检测系统分析灵敏度计算

1、生物检测限
由前述,空白血清如以现有的吸光度均值为准,有99.7%的可能性出现的空白吸光度可高到0.139A.0.25ug/L甲状腺蛋白组减去空白后的吸光度均值为0.07625A,标准差为0.0253A;可推测,有99.7%的可能性出现最小的吸光度为0.07625 -3×0.0253A,即0.0019A;而空白吸光度减去空白吸光度均值可能有的最高值为0.0411A.这0.0019A在空白吸光度涉及的范围内.因此0.25ug/L不是本法的生物检测限。0.5ug/L的99.7%的最低吸光度0.025A(已减去空白),也在空白涉及的范围内.1.0ug/L的99.7%的最低吸光度为0.0518A(已减去空白),在0.0411A之外,说明1.0ug/L的样品吸光度有99.7%的可能性,一定大于空白的吸光度,能定量地报告结果。1.0ug/L为本法的生物检测限.
3、功能灵敏底
从上面的表中可知,1.0ug/L组的重复检测变异系数CV为19.3%,最接近20%.所以,本法的功能灵敏度为1.0ug/L.它也正好等于生物检测限估计的限值.