In article ,
Henry Kuska wrote:
billo, your answer satisfies my suspicion that your quest is meaningless
since your first reply clearly shows that it excludes meaningful real world
groups.



Here are two examples of how it should be done.



The first study was a large randomized study of the Atkin's Diet,
which found that the Atkins Diet resulted in good weight loss
for those who stayed on it, but had a very high level of attrition.
In other words, if one stayed with the original group it didn't
work -- because few people stayed on the diet. It is important
to actually test whether people "on the diet" actually are
doing what they are supposed to do:

Foster, et al. A Randomized Trial of a Low-Carbohydrate Diet for
Obesity NEJM 348:2082-2090, 2003.

(begin excerpt)

A total of 49 subjects completed 3 months of the study (28 on the
low-carbohydrate diet and 21 on the conventional diet), 42 subjects
completed 6 months (24 on the low-carbohydrate diet and 18 on the
conventional diet), and 37 subjects completed 12 months (20 on the
low-carbohydrate diet and 17 on the conventional diet). The percentage
of subjects who had dropped out of the study at 3, 6, and 12 months was
higher in the group following the conventional diet (30, 40, and 43
percent, respectively) than in the group following the low-carbohydrate
diet (15, 27, and 39 percent, respectively), but these differences were
not statistically significant. Overall, 59 percent of subjects
completed the study, and 88 percent of those who completed the
six-month assessment completed the full study. When the analysis
included data on subjects who completed the study and data obtained at
the time of the last follow-up visit for those who did not complete the
study, the pattern of weight loss was similar to that obtained when the
base-line values were carried forward in the case of missing data.
Subjects on the low-carbohydrate diet lost significantly more weight
than the subjects on the conventional diet at 3 months (P=0.002) and 6
months (P=0.03), but the difference in weight loss was not
statistically significant at 12 months (P=0.27)

(end excerpt)

See, Henry, you don't just assume that because someone has
received instructions that they follow them. You ask the question,
and compare those who do and do not follow directions.


In this study, compliance
was measured for studying the efficacy of a birth control pill:

F. D. Anderson, H Hait. A multicenter, randomized study of an extended
cycle oral contraceptive. Contraception Volume 68, Issue 2 , August
2003, Pages 89-96

(begin excerpt)

3.2. Compliance

There were two measurements of compliance, which were evaluated by
assessing patient diary data as to whether or not a patient took her OC
pill every day. Pill compliance within each extended or conventional
cycle was determined by observing if the patient missed 2 consecutive
days of pill-taking and, if so, the patient was considered to be
noncompliant for that cycle. Overall, study compliance was determined
by counting the percentage of total days in the 1-year study when the
patient took the designated pill for a given day. Overall compliance of
80% would exclude a patient altogether from the Pearl Index
calculation. Otherwise, noncompliance within a particular cycle would
exclude that cycle only from the Pearl Index. For the life-table
calculation, only the overall compliance criterion was used to exclude
"noncompliant" patients from the cumulative pregnancy rate calculation,
since exclusion of individual cycles from the patient's total would
lead to a noncontinuous, intermittently truncated time frame.

The overall treatment compliance rate in each of the study groups was
very high with 95.4% of extended cycle regimen patients and 93.4% of
conventional regimen patients assessed as compliant. A total of 22
(4.8%) extended cycle regimen patients and nine (4.0%) conventional
regimen patients were discontinued from the study due to noncompliance.
The number of clinically significant protocol deviations was minimal
and no protocol deviations were used to exclude any patients from the
analysis of efficacy or safety. Most protocol deviations were related
to inclusion/exclusion criteria at study enrollment and were not
observed during the active study interval.


(end excerpt)



Now, I know, Henry, that you do consider comparing compliant
vs noncompliant groups "meaningless," but when studying the
effects of therapy -- or the toxicity in adverse effects --
recognizing that one can both measure and separate compliant
from noncompliant groups is important. In the case of looking
at pesticide/herbicide toxicity when used as directed, that
means actually looking at whether or not it is used as
directed.


billo