
Update to documentation, PEPI Version 4.04x
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The main changes in the DOS-based programs since version 4.0 are to PAIRS and SCRN (see A and B below).  Slight changes (cosmetic, removal of bugs) have also been made to DIRST, FIT, FREQ, LOGISTIK, LOGX, MATCHED, PAIRS, RANKCORR and RELATED.

For descriptions of changes and additions introduced in the Pepi-for-Windows programs, see the items marked "NEW" in the documentation files supplied with COMPARE2, DESCRIBE, and PAIRSETC (COMPARE2.TXT, DESCRIBE.TXT, and PAIRSETC.TXT).  In WHATIS-C.I., exact Fisher's intervals are now displayed for rates with person-time denominators if there are up to 40 (instead of 20) events.

Errors found in the Version 4.0 manual are listed below (see C).

----------------------------------------------------------------------------------------
A.

In PAIRS v 4.08, t is replaced by 1.96 in the formulae for the repeatability coefficient (Bland and Altman 1999).

Reference
Bland JM, Altman DG (1999) Measuring agreement in method comparison studies. Statistical Methods in Medical Research 8: 136-160.

----------------------------------------------------------------------------------------
B.

Three features have been added to SCRN (which appraises screening and diagnostic tests for the presence of a disease).

(1). If the test provides a range of values, the program now computes the "gray zone" of indecisive results, and the proportion of results falling in this gray zone (Version 4.01).For this purpose, indecisive results are defined as those that cannot yield a post-test probability of 95% or more (indicating "disease present") or 5% or less ("disease absent").  The gray zone is computed separately for various pre-test probabilities (0.5%, 1%, 5%, 10%, 20%, 30%, 40%, 50%, and 60%).  The procedure is based on the method described by Coste and Pouchot (2003).

Method
The range of test results is regarded as an ordinal scale.  If high test results point to the presence of the disease, the lower limit of the gray zone (gLow) is defined as the lowest test result for which the post-test probability of the disease (PPD+) is less than 95%, and the upper limit of the gray zone (gHigh) is the highest test result for which the post-test probability of absence of the disease (PPD-) is less than 95%.
The test's sensitivity (Se) and specificity (Sp) are computed for each possible cutting-point ("at or above" the selected test result) in turn, and PPD+ and PPD- are then computed for each pre-test probability by the formulae
   PPD+ = (Se.P) / (Se.P + (1 - Sp)(1 - P))
   PPD- = (Sp(1 - P) / ((1 - Se)P + p(1 - P))
where P = pretest probability.
The formula for the proportion of results in the gray zone is 
   L'(1 - P) + U'.P
where
   L' = 1 - (specificity at gLow) - U
   U' = 1 - (specificity at glow) - L
   L  = 1 - (sensitivity at gLow)  
   U  = 1 - (specificity at gHigh)
Note: in the paper, the formula for U' is erroneously stated as
   U' = 1 - (sensitivity at glow) - L
The computation is similar, mutatis mutandis, for a test whose low results point to the presence of the disease.

Reference
Coste J, Pouchot J (2003) A grey zone for quantitative diagnostic nd screening tests.  International Journal of Epidemiology 32: 304-313.

(2).  SCRN now appraises a combination of two tests, used in parallel or (in either sequence) in series (Version 4.03).  The computation requires the results ("positive" or "negative") obtained when the tests are evaluated against the same "gold standard" in the same subjects.  Sensitivity, specificity, and the posttest probability after a positive result are displayed, defining a positive result for the combination as either (a) a positive result for either test or (b) positive results for both tests.  If confirmatory "gold standard" tests are not available for subjects with negative test results, SCRN computes only the combination's relative sensitivity and relative false positive rate (in comparison with each of the tests).  

(3).  When comparing two tests on the basis of their results in the same subjects, compared with "gold standard" tests applied to subjects with positive results, SCRN now computes the FP:TP ratio (Chock et al. 1997), that is, the number of extra false positives when the test with a higher sensitivity is used, for each extra true positive detected (version 4.04), as well as comparing the sensitivity and false-positive rates of the two tests.  The FP:TP ratio is not computed if one test is clearly superior (ignoring considerations of cost, convenience, etc.) - i.e., if it has a higher sensitivity and a lower false-positive rate than the other test.  When a higher sensitivity is attained at the expense of a higher false-positive rate, the ratio expresses the trade-off to be considered when deciding which test to use.  Since the FP:TP ratio depends on the prevalence of cases, ratios are also estimated for use of the tests in target populations with prevalences of 5%, 2%, 1% and 0.1%, as well as in a population with the same prevalence as in the samples tested; a formula is provided for use in populations with other prevalences; these other ratio.  The procedure is applicable to tests that yield "positive" or "negative" results (which may be based on a selected cutting-point in a range of values).  The total number of subjects tested, and the findings in comparison with the gold-standard test, must be entered.  

Method
The FP:TP ratio for the data entered is computed by the formula in Chock et al. (1997, p. 1243), and its 95% confidence interval by formula 2.1; these results are appropriate when the tests are to be used in a target population with the same prevalence of cases (subjects with positive gold-standard tests) as in the sample tested.  FP:TP ratios for target populations with a different prevalence are estimated by formula 3.9.  This formula requires estimates of the numbers of cases and noncases with two negative findings in the sampled population (d* and D*, respectively).  The number of cases with negative findings (which is also required for computation of the prevalence of cases in the sampled population) is estimated by the formula
   d* = |b - c| / a
where
   a = cases with both tests positive
   b = cases with only Test A positive
   c = cases with only Test B positive.
This is based on the assumption that the two tests are independent when applied to the cases.  D* is estimated by subtracting the sum total of subjects whose results are entered, plus d*, from the total number of subjects tested.  FP:TP ratios in populations with other prevalences are not computed if the total number of subjects tested is not entered.

Reference
Chock C, Irwig L, Berry G, Glasziou (1997) Comparing dichotomous screening tests when individuals negative on both tests are not verified. Journal of Clinical Epidemiology 50: 1211-1217.

----------------------------------------------------------------------------------------
C.
Errors found in the Version 4.0 manual:

Page 39.
In the heterogeneity chi-square formula, the "squared" is misplaced.  The right-hand term should be:
  SumOf [Zi - (SumOf Z1/k)] squared
  
Page 66.
In the paragraph on the mean square successive difference test, replace "It is done only if" with "It is not done if".  
  
Page 121.
In Example 2, "159" should be "519". 

Page 142.
In example 6, the first five lines of results are incorrect.z

Page 146.
In line 3, "formulae 9-19" should be "formula 9-19".

Page 159.
In Example 4, the odds ratio to be detected is 2, not 3, and power = 53.09%, not 91.35%.

Page 183.
In lines 4-5 of the "Attributable or prevented fraction" section, delete "unless a different value is entered by the user".
In the middle of this section, replace "formula 16-26" with "formula 16-25".

Page 201.
In line 7, "the 2 x k table" should be "the k x k table".

Page 203.
The reference near the foot of the page shold be to Sokal and Rohlf (1981:573).

Page 215.
Third paragraph: If "gold standard" tests are not available for all subjects, the only relative measures computed (Version 4.03) are sensitivity and the false positive rate.
Near the foot of the page, the sentence "Post-test probabilities ... prevalence" is unnecessary.

Page 261.
At top of page, insert "lower" ("the upper and lower confidence limits of P").
The formula in line 2 should be:
  (P.D2) / [(1 - P).D]
  
Page 262
The reference in the 5th-last line should be:
  Smith and Bates (1992) 

---------------------------------------------------------------------------------------


