Paper 119-2024

Limit of Detection Calculation Methods Comparison for PCR-based

Quantitative Studies using the SAS® System

Jesse A. Canchola, Roche Diagnostics Solutions, Pleasanton, California, USA

ABSTRACT
In assay performance evaluation for quantitative assay studies based on polymerase chain reaction (PCR), the Limit
of Detection (LoD) is defined as the lowest concentration or amount of analyte that is consistently detectable (in our
case, in at least 95% of the samples tested; CLSI EP17-A2). In practice, the estimation of the LoD uses a parametric
curve fit to a set of panel member (PM1, PM2, PM3, etc.) data where the responses are binary (i.e., percent detected
at a certain level). Typically, the parametric curve fit to the percent detection levels takes on the form of a logistic or
probit distribution. Additional methods for estimation of the LoD in PCR-based studies include using the “based on hit
rate” and maximum likelihood estimation (MLE) methods (Singh & Nocerino, 2001). Amongst all methods, the MLE
method is preferred since the MLE is taken to be sufficient (Rice, 1988) given the selected parent probability distribution
assumed to model the data. Moreover, it can be shown that the MLE for a Poisson-distributed random variable has
the minimum variance unbiased estimator (MVUE) property (Cassella and Berger, 2002), thus allowing it to be used
as a reference in comparison with other methods. We performed a comparison of the LoD calculation methods using
various realizations of a real seven-member panel for HIV-1 PCR assay results.

INTRODUCTION AND BACKGROUND

Performance evaluation for PCR-based quantitative assay studies includes reproducibility, linearity, accuracy and
analytical sensitivity of the assay (CLSI EP17-A2, CLSI EP05-A2, CLSI EP06-A). Reproducibility speaks to the
variance of the measured results across the linear range of the assay whilst linearity and accuracy endeavor to measure
the linear relationship (typically on the log10 scale) and bias (or off-set) as compared to a target (or expected) value.
Finally, for analytical sensitivity, the main topic of interest here, we would like to know the lowest amount of analyte we
can reliably detect (typically taken as 95% of samples tested). This is arguably most important in the diagnosis and
monitoring of target viruses (e.g., HIV, HBV, HCV)* or bacteria (e.g., Listeria, MRSA, MAP)**. For example, the
analytical sensitivity of the assay test for the HIV-1 virus in the bloodstream can be used to declare disease in a patient
(diagnosis) who is then placed on a treatment drug regimen. Thereafter, the viral load of the patient is measured
regularly (monitoring) to be sure the treatment is efficacious.
For quantitative assays, the analytical sensitivity is measured by the limit of detection (LoD) and sometimes called the
lower limit of detection (LLoD). From this point forward, we use the more common LoD acronym and “sensitivity” to
mean “analytical sensitivity” throughout this document. For this, an experiment is performed that collects information
about the lower end of the quantitative assay. Typically, several levels are targeted at the lower assay range in order
to fit a parametric model curve to obtain an estimate of the LoD where the curve crosses 95% detection (Figure 1).

___________________
* HBV=Hepatitis B Virus, HCV=Hepatitis C Virus, HIV=Human Immunodeficiency Virus.
** MRSA=Methicillin-resistant Staphylococcus aureus, MAP=mycobacterium avium subspecies paratuberculosis.

In the experiment, the levels are chosen to include at least one panel member at 100% detection, another to anchor
the parametric curve at the bottom end (not including zero) with the remaining three or more levels targeting the

1
Limit of Detection Calculation Methods Comparison for PCR-based Quantitative Studies using the SAS® System, continued

region where one believes the LoD might be. Table 1 shows an example of six panel levels (not including zero) with
the top end anchored by an HCV RNA assay level at 50 IU/mL and one anchor at the bottom at 2.5 IU/mL. There are
four levels in between both the top and bottom ends: 25, 15, 10 and 5 IU/mL.

Table 1. LoD in EDTA Plasma from “Empower change in HCV” for COBAS® AmpliPrep/COBAS® TaqMan®
HCV Qualitative Test, v2.0

An Aside: A related but natural question then becomes, “How many samples shall one test at each level?” The
answer, of course, usually depends on the purpose. If one is performing first-time development work for determining
positioning of the LoD for future targeting, one needs only a minimum amount of testing for each level (say, ~20 samples
per level with possibly more at the lower end if we expect “dropouts” or samples that do not all quantitate at each level).
On the other hand, if the regulatory agency requires testing with material that is in short supply (e.g., World Health
Organization (WHO) material), or is expensive, the researcher may, by default, be forced to use the same minimum
number of samples per level as in the development case. Furthermore, considerations for level sample sizes are also
made in order to more fully characterize the true LoD. That is, the more sample size one uses at each level, the tighter
the confidence interval about the estimated LoD. Table 1 shows that the experiment used about 252 samples for each
level. However, typical experiments will have 40 samples or more per level. Note that the mention of “sample” in this
discussion can be taken to be, for example, a contrived sample that is diluted to its specific level from a common pool
of inactivated analyte.
Figure 1 shows the results of a parametric curve fit to the six-member panel “hit rate” data using the probit distribution
with panel concentrations transformed using the logarithm base 10 (i.e., log10) scale. The estimated LoD is 11.7 IU/mL
which rounds to 12 IU/mL as indicated at the bottom of the published Table 1. The goodness-of-fit (GOF) statistics
show statistical non-significance thereby indicating a good fitting model (Pearson chi-square=2.9, p-value=0.58 and LR
chi-square=3.1, p=0.55). Note that the panel member with zero IU/mL concentration is not used in the calculations as
the log10 of zero is undefined. Moreover, in PCR-type studies, we can momentarily define the target response as binary
(i.e., detected or not detected) so that there is no background “noise” that needs to be measured in estimating the LoD.
Therefore, the Limit of Blank (LoB; CLSI EP17-A2) has no intrinsic meaning in PCR testing so we exclude any further
discussion about the LoB.
Of course, the parametric probit curve fit method is one of several that will be discussed in this paper. We include the
parametric logistic curve fit, the method of percent detection and the method of maximum likelihood using a Poisson
probability distribution to model the observations in the vicinity of the LoD. The reader should note that these methods
are not exhaustive.

METHODS & RESULTS

PARAMETRIC CURVE FIT METHOD
As indicated prior, the medical diagnostic “standard” parametric curve fit to the percent detection levels takes on the
form of a logistic or probit distribution. We use our SAS macro LoD5 (Canchola and Hemyari, 2016) which incorporates
the SAS PROBIT procedure for fitting either distribution and assess fit using the Pearson chi-square and Likelihood
Ratio (LR) chi-square statistics. Figure 1 shows the result of fitting the probit distribution to the hit rate data of Table 1.
Note that the rates shown in Table 1 are in percentage format while the realization in Figure 1 on the y-axis is in “rate”
or decimal format. We next show how our LoD5 SAS macro is used to fit these data in Table 1.
First, we read in the relevant data from Table 1 into the SAS/System:

2
Limit of Detection Calculation Methods Comparison for PCR-based Quantitative Studies using the SAS® System, continued

data LoD ;
input Target $3. Titer Hits TotalTests @ ;
Log10_Titer = log10(Titer) ; Obs_HitRate = Hits / TotalTests ;
Obs_HitPcnt = Obs_HitRate * 100 ;
datalines ;
HCV 50 252 252
HCV 25 251 252
HCV 15 246 251
HCV 10 233 252
HCV 5 183 252
HCV 2.5 124 251
; run ;

Next, we load our LoD5 SAS macro and specify the macro call (for details, please see our SAS paper introducing the
LoD5 SAS macro):
%include "<<your path>>/LoD5.sas" ;
%LOD5 (
datain = LoD
,study = CTM_1
,target = &target
,subset = target = "HCV"
,units = IU/mL
,distn = Probit
,alpha = 0.05
,sided = 2
,TableNo = 1
,FigureNo = 1
,ODStype = RTF
,RTFout = <<your path>>/LoD_HCV_CTM_Probit.RTF
,GraphMin = -0.5
,GraphMax = 3.0
,GraphByInc = 0.5
,GraphTitle1 = Figure 1 Limit of Detection Estimation for HCV Target
,GraphTitle2 = Probit Curve Fit
,YaxisTitle = Hit Rate
,XaxisTitle = Titer (log10 &Units) ) ;

This SAS code produces the graph in Figure 1 (without the additional annotation) where the LoD is estimated to be
11.7 IU/mL, with 95% CI from 10.3 to 13.6 IU/mL, using the probit curve fit. In addition, the SAS code generates a
percent detection table (aka “hit rate” table) as shown in Table 2 with associated 95% (Clopper-Pearson) exact
confidence intervals for the panel level hit rates. From Table 2, one can see that the estimated LoD should be between
10 and 15 IU/mL (second column) since the hit rate at those levels is between 92.4% and 98.0%, respectively (column
6).
Table 2. Percent Detection Output Results using the LoD5 SAS Macro for Data in Table 1 with Associated
95% Exact (Clopper-Pearson) Confidence Intervals for Panel Level Hit Rates.
Titer Titer No. Total Percent
Target (IU/mL) (log10 IU/mL) Detected Tests Detection 95% Exact LL 95% Exact UL
50 1.6990 252 252 100 98.5 100
25 1.3979 251 252 99.6 97.8 100
15 1.1761 246 251 98.0 95.4 99.3
HCV
10 1.0000 233 252 92.4 88.4 95.3
5.0 0.6990 183 252 72.6 66.6 78.0
2.5 0.3979 124 251 49.4 43.0 55.7
HCV=Hepatitis C Virus, IU=International Units, LL=Lower Limit, UL=Upper Limit

3
Limit of Detection Calculation Methods Comparison for PCR-based Quantitative Studies using the SAS® System, continued

To obtain a logistic curve fit to the hit rate data, one needs only to change the distribution (viz., distn) portion of the
LoD5 SAS macro as shown above to “distn = Logistic”. The results of which are shown in Figure 2.

Again, we see that the LoD estimate is 11.7 IU/mL – the same as when
using the probit fit. However, the 95% CI for the LoD is slightly different:
10.2 to 13.9 IU/mL for the logistic fit as compared to 10.3 to 13.6 IU/mL
using the probit fit (see Figure 1), although the GOF statistics are better
for the probit than for the logistic curve fit (viz., Pearson chi-square=7.8,
p-value=0.1 and LR chi-square=8.8, p-value=0.07 (see Figure 2).

The SAS PROBIT procedure has additional non-standard distributions

available for fitting (viz., Extreme Value and Gompertz). Because SAS
PROBIT is the main engine for the LoD5 SAS macro, these additional
distributions are also available for use. For example, Figure 3 shows
the Extreme Value distribution curve fit to the Table 1 data. From Figure
3, we see that the LoD is 11.8 IU/mL with 95% CI from 10.6 to 13.5
IU/mL with GOF statistics Pearson chi-square=0.6 (p-value=0.97) and
LR chi-square (p-value=0.97) being better fit than both the probit and
logistic distributions. Choosing the Gompertz distribution for the hit-rate
fit will give the same results as the Extreme Value since the two
distributions are related under certain constraints.

HIT RATE METHOD

The hit-rate method uses the actual level or panel closest to, but above,
the 95% hit rate. Since this method is “visual”, there are no confidence
interval estimates that are produced. From Table 2, one can easily see
that the 15 IU/mL level/panel is closest (i.e., 98% point estimate, column
6), but above, the 95% hit rate. One can see that this method can be
somewhat conservative in the LoD estimate.

MAXIMUM LIKELIHOOD (ML) METHOD

As mentioned prior, amongst all methods noted above, the ML method
is a robust statistical tool for determining the Limit of Detection (LoD).
For the Poisson probability distribution, the Maximum Likelihood
Estimator (MLE) is particularly powerful because it is a sufficient
statistic and also is the Minimum Variance Unbiased Estimator
(MVUE). These properties, along with the completeness of the
sufficient statistic, ensure that the MLE for the Poisson distribution is optimal for many statistical purposes, including
LoD determination.
ML estimation works on the principle of estimation of model parameters (e.g., from Poisson, Exponential, Normal
models) given the collected data. That is, “what is the best estimate of the parameter estimate (i.e., the maximum
likelihood) given the data at hand?” (Pawitan, 2013; Purcell 2007).
Our LoD_MLE SAS macro (Canchola and Hemyari, 2017) is used to provide the maximum likelihood estimate (MLE).
The macro call code for running the LoD_MLE SAS macro is shown below (for details, please see our SAS paper
introducing the LoD_MLE SAS macro10):
%include "<<your path>>/LoD_MLE.sas" ;
%LOD_MLE (
datain = LoD
,target = HCV
,subset = target = "HCV"
,units = IU/mL
,alpha = 0.05
,sided = 2
,TableNo = 1
,ODStype = RTF
,RTFout = <<your path>>/LoD_HCV_CTM_Probit.RTF ) ;

This SAS code produces the estimated LoD as 11.5 IU/mL with 95% CI from 10.5 to 12.5 IU/mL.

4
Limit of Detection Calculation Methods Comparison for PCR-based Quantitative Studies using the SAS® System, continued

PERFORMANCE EVALUATION FOR PROBIT METHOD USING MLE AS TRUTH

We perform a comparison of two of the LoD calculation methods presented above: (1) the Probit method that is in
wide use and (2) the ML method that is preferred We use various realizations of a real seven-member panel for HIV-
1 PCR assay results. Table 3 shows the reference set of starting values for the evaluation. The evaluation uses all
possible combinations of 3, 4, 5, 6, and 7 of the possible 7 levels to perform the LoD estimations via Probit and ML
methods. Furthermore, we perform the evaluation on total sample sizes (per level) of 40, 50, 70, 100 and 200.
Table 3. A Seven Member Panel Reference Set of Starting Values for Panel Level Combinations of Size 3 to
7 used in LoD Estimation using Probit and ML Methods
Number Detected or Number of Hits
Detec (out of the total per level)
tion %
Panel Titer or Hit Detection (n=40) (n=50) (n=70) (n=100) (n=200)
Target Level (cp/mL) Log10(Titer) Rate or % Hits
7 50.0 1.6990 1.000 100.0 40 50 70 100 200
6 25.0 1.3979 0.990 99.0 40 50 69 99 198
5 15.0 1.1761 0.993 99.3 40 50 70 99 199
HIV-1 4 10.0 1.0000 0.953 95.3 38 48 67 95 191
3 5.0 0.6990 0.780 78.0 31 39 55 78 156
2 2.5 0.3979 0.557 55.7 22 28 39 56 111
1 1.0 0.0000 0.100 10.0 4 5 7 10 20

Figure 4 shows the graphical results for panel levels with n=40 at each level. For example, the graph shows 35 blue
open-circles representing the LoD estimates for all three level combinations out of the seven total levels.
Mathematically, there should be (73) (i.e., 7 choose 3) or 35 possible combinations [e.g., (1,2,3), (1,2,4),
(1,2,5),…,(5,6,7)] where the LoD is estimated using Probit for each of the 35 combinations of size 3. Similarly, the red
crosses represent the LoD estimates for (74) or all 35 possible combinations of size 4. Next, the green Xs represent the
(75) or 21 possible panel level combinations of 5 from 7 levels. It is easy to see that the brown triangles and purple
square represent the remaining panel level combinations of (76)=7 and (77)=1, respectively. Finally, the ML estimate is
calculated using the full 7-member panel and is shown on Figure 4 as a violet line at 10.5 IU/mL with associated 95%
CI at 8.2 cp/mL (green line at the lower limit) and 12.7 cp/mL (red line at the upper limit).

5
Limit of Detection Calculation Methods Comparison for PCR-based Quantitative Studies using the SAS® System, continued

DISCUSSION

TABLE 1 DATA
Table 3 shows the summary of the various LoD estimation methods shown above when using the data in Table 1. If
we take the MLE of 11.5 IU/mL to be the reference or “true” value, then the other methods can be compared in
performance using the difference of the estimate for that method with that of the MLE (i.e., the bias). The last column
in Table 3 indicates that the least bias comes from the Probit and Logistic methods (bias: 0.2 IU/mL) followed by the
Extreme Value and Gompertz methods (bias: 0.3 IU/mL). The Hit Rate method produces the largest bias (bias: 3.5
IU/mL).
The second to the last column of Table 3 shows the 95% confidence interval length. The narrowest interval comes
from the ML method (95% CI Length: 2 IU/mL) whilst the largest comes from the Logistic method (3.7 IU/mL).
Table 3. Limit of Detection and associated 95% CI using Six Methods.
LoD 95% LL 95% UL Interval Bias
Method From MLE
(IU/mL) (IU/mL) (IU/mL) Length
Probit 11.7 10.3 13.6 3.3 0.2
Logistic 11.7 10.2 13.9 3.7 0.2
Extreme
11.8 10.6 13.5 2.9
Value 0.3
Gompertz 11.8 10.6 13.5 2.9 0.3
Hit Rate 15.0 --- --- --- 3.5
ML 11.5 10.5 12.5 2 ---

PERFORMANCE EVALUATION OF PROBIT METHOD USING MLE AS TRUTH

Figure 4 shows that when choosing 3 levels for LoD estimation (blue open circles) using the Probit method, the LoD
estimates can be quite variable. Using 4 levels (red plus signs), the variability is reduced. When choosing 5 levels
(green Xs), the variability diminishes further. Using 6 levels, the LoD estimates via Probit are much closer and better
defined. These results are similar for n=50, 70, 100 and 200 (see Appendix A). Overall, the average bias when
choosing 3, 4, 5, 6 and 7 levels to estimate the LoD as compared to the MLE is shown in Table 4.
Table 4. Bias of LoD Estimates (cp/mL) when Choosing 3, 4, 5, 6 and 7 Panel Levels from the
Maximum Likelihood Estimate.

n=40 n=50 n=70 n=100 n=200

Levels ̅̅̅̅̅ ̅̅̅̅̅ ̅̅̅̅̅ ̅̅̅̅̅ ̅̅̅̅̅

𝐿𝑜𝐷 ̅̅̅̅̅̅ 𝐿𝑜𝐷 ̅̅̅̅̅̅ 𝐿𝑜𝐷 ̅̅̅̅̅̅ 𝐿𝑜𝐷 ̅̅̅̅̅̅ 𝐿𝑜𝐷 ̅̅̅̅̅̅
(# LoD MLE 𝐵𝑖𝑎𝑠 MLE 𝐵𝑖𝑎𝑠 MLE 𝐵𝑖𝑎𝑠 MLE 𝐵𝑖𝑎𝑠 MLE 𝐵𝑖𝑎𝑠
(SD) (SD) (SD) (SD) (SD)
combos)
9.7 9.7 9.7 9.7 9.7
3 (35) 10.5 -0.8 10.2 -0.5 10.7 -1.0 10.9 -1.2 10.8 -1.1
(1.7) (1.7) (1.7) (1.7) (1.7)
9.8 9.8 9.8 9.8 9.8
4 (35) 10.5 -0.7 10.2 -0.4 10.7 -0.9 10.9 -1.1 10.8 -1.0
(0.9) (0.9) (0.9) (0.9) (0.9)
9.7 9.7 9.7 9.7 9.7
5 (21) 10.5 -0.8 10.2 -0.5 10.7 -1.0 10.9 -1.2 10.8 -1.1
(0.6) (0.6) (0.6) (0.6) (0.6)
9.7 9.7 9.7 9.7 9.7
6 (7) 10.5 -0.8 10.2 -0.5 10.7 -1.0 10.9 -1.2 10.8 -1.1
(0.4) (0.4) (0.4) (0.4) (0.4)
9.6 9.6 9.6 9.6 9.6
7 (1) 10.5 -0.9 10.2 -0.6 10.7 -1.1 10.9 -1.3 10.8 -1.2
(---) (---) (---) (---) (---)
̅̅̅̅̅=Average Limit of Detection estimate for within-level combinations (cp/mL), MLE=Maximum Likelihood Estimate
𝐿𝑜𝐷
̅̅̅̅̅̅ = average of (LoDi – MLE).
calculated from the full 7-member panel for each per-level sample size, 𝐵𝑖𝑎𝑠

The Table 4 LoD standard deviations (SD) buttress the equivalent Figure 4 graph in terms of the dispersion of
individual LoD estimated values for each combination of levels (viz., 3, 4, 5, 6 and 7). Interestingly enough, the
average bias levels seem to be optimal (for this example) when n=50 per level. Of course, this determination

6
Limit of Detection Calculation Methods Comparison for PCR-based Quantitative Studies using the SAS® System, continued

assumes that the MLE is the “true estimate” of the LoD. Moreover, within each of the levels (for any sample size), the
least bias occurs when choosing four levels to estimate the LoD.

CONCLUSION
The results using different methodologies for estimating the Limit of Detection for one example of hit rate data in
PCR-type studies showed that, using the Poisson distribution to model observations in the vicinity of the LoD where
the distribution is not likely to be Normal, and given that the MLE is MVUE, one can use the MLE as the reference to
compare with parametric methods with respect to the bias and confidence interval lengths. We showed that the LoD
methods presented were biased with the least biased LoD estimation methods of those presented were those using
the Probit and Logistic model fit to the hit-rate data. When consideration is for how many levels to use in the LoD
experiment at hand, a small performance evaluation study using the Probit model with average bias from the MLE
showed the average LoD/bias combined optimal number to be 6 levels (SD=0.4, bias= -0.5 at n=50 per level).

7
Limit of Detection Calculation Methods Comparison for PCR-based Quantitative Studies using the SAS® System, continued

REFERENCES
1. Canchola JA, Hemyari P. 2016. “Limit of Detection (LoD) Estimation Using Parametric Curve
Fitting to (Hit) Rate Data: The LoD_Est SAS Macro.” Proceedings of the SAS Global 2016
Conference held in Las Vegas, NV USA. Cary, NC: The SAS Institute.
2. Canchola JA, Vaks JE, Tang S (2019). “Limit of Detection (LoD) Estimation Using Maximum
Likelihood from (Hit) Rate Data: The LoD_MLE SAS Macro.” Proceedings of the Western Users
of SAS Software Annual Conference, September 2019, Seattle, WA.
3. Casella G, Berger RL (2002). Statistical Inference. 2Ed. pp. 334-339. Duxbury Thomson
Learning. Pacific Grove, CA: Wadsworth Group.
4. Clinical and Laboratory Standards Institute (CLSI) document EP05-A2. “Evaluation of Precision
Performance of Quantitative Measurement Methods; Approved Guideline--Second edition.”
Wayne, PA. 2004.
5. Clinical and Laboratory Standards Institute (CLSI) document EP06-A. “Evaluation of the Linearity
of Quantitative Measurement Procedures: A Statistical Approach; Approved Guideline.” Wayne,
PA. 2003.
6. Clinical and Laboratory Standards Institute (CLSI) document EP17-A2. “Evaluation of Detection
Capability for Clinical Laboratory Measurement Procedures; Approved Guideline--Second
edition.” Wayne, PA. 2012.
7. Hogg RV, McKean JW & Craig, AT (2013). Introduction to mathematical statistics. Pearson
Education India.
8. Johnson NL, Kemp AW & Kotz S (2005). Univariate discrete distributions (Vol. 444). John Wiley &
Sons.
9. Lehmann EL & Casella G (2006). Theory of point estimation. Springer Science & Business
Media.
10. Mood AM, Graybill FA and Boes DC (1974). Introduction to the Theory of Statistics. McGraw-Hill.
11. Pawitan Y (2013). In All Likelihood: Statistical Modelling and Inference Using Likelihood.
Oxford, United Kingdom: Oxford University Press. 528 pp.
12. Purcell S (2007). “Maximum Likelihood Estimation”. Accessed 05Jun2015. Available at:
http://statgen.iop.kcl.ac.uk/bgim/mle/sslike_3.html
13. Rice JA (2007). Mathematical statistics and data analysis (Vol. 371). Belmont, CA:
Thomson/Brooks/Cole.
14. Roche Molecular Systems, Inc. 2011. “Empower change in HCV” for COBAS®
AmpliPrep/COBAS® TaqMan® HCV Qualitative Test, v2.06. Accessed on 05Jun2015. Available
at: http://www.roche-diagnostics.ch/content/dam/corporate/roche-
dia_ch/documents/broschueren/molecular_diagnostics/virology/06611656001_EN_EA_COBAS-
AmpliPrep_COBAS-TaqMan-HCV-Qualitative-Test-v2.0.pdf
15. Singh A & Nocerino J. 2001. “Robust Estimation of Mean and Variance Using Environmental
Data Sets with Below Detection Limit Observations”, U. S. Environmental Protection Agency
internal article. Clearance Number 01-062. Washington, DC: U. S. EPA.

CONTACT INFORMATION
Your comments and questions are valued and encouraged. Contact the first author at:
Name: Jesse A. Canchola
Company: Roche Molecular Systems, Inc
Address: 4300 Hacienda Drive

8
Limit of Detection Calculation Methods Comparison for PCR-based Quantitative Studies using the SAS® System, continued

City, State ZIP: Pleasanton, California, 94588

E-mail: [email protected]

SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS
Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are
trademarks of their respective companies.

APPENDIX A
N=40 at each level.

9
Limit of Detection Calculation Methods Comparison for PCR-based Quantitative Studies using the SAS® System, continued

N=50 at each level.

10
Limit of Detection Calculation Methods Comparison for PCR-based Quantitative Studies using the SAS® System, continued

N=70 at each level.

11
Limit of Detection Calculation Methods Comparison for PCR-based Quantitative Studies using the SAS® System, continued

N=100 at each level.

12
Limit of Detection Calculation Methods Comparison for PCR-based Quantitative Studies using the SAS® System, continued

N=200 at each level.

13