153
© Quarterly Homœopathic Digest, Vol. XXX1, 3 & 4, 2014. Private Circulation only.
Homeopathy, each subject is considered unique, with
peculiar responses and dynamics. The selection of other
people as controls substantially biases the analyses,
once the variables used for matching are insufficient to
express the supposed similarity among individuals.
Gender, age, smoking and physical inactivity, for
example, do not correspond to the measures used for the
prescription of homeopathic medicines. The actual
variables, characteristic and individual, make the cases
to be so distinct from potential controls that the search
for an appropriate control would be formidable, if not
impossible. This way, we have the confounding by
indication, in which the condition that determined that
therapeutics is so peculiar that it is, by itself, an
important predictor of the effect under study. This
makes practically unfeasible the use of other individuals
as controls. At the brink of a new paradigm, the designs
described here intend to coexist with other alternate
methodologies and to be a suitable model for clinical
homeopathic research.
The use of cases as their own controls removes
confounding from many unchangeable characteristics,
that means, uncontrollable and intrinsic properties of
each person, such as intelligence, genetics,
susceptibility. Moreover, the self matching designs are
more powerful from a statistical perspective and offer
savings in the sample size. The case-crossover design
was introduced in 1991 for examining the transient
effects of a brief exposure on the onset of an acute
outcome.
Such analyses can be completed with minimal
ethical worries, at low cost and are usually quick.
Common and rare outcomes can be tested, and the
clinical and statistical significance determined. They are
appropriate to rapidly fluctuating processes, such as
activity, emotion or pain (Redelmeier & Tibshirani,
1997). Each case corresponds to a stratum, and
contributes one case window and one or more control
windows. The case window is defined as the period just
preceding or during the event under study. The control
windows are periods of the same length as, and not
overlapping with, the case window, and provide an
estimate of the expected frequency of exposure for each
case. The case window and the control window derive
from the same person at different times (MacLure,
1991; Hernandéz-Dias et al.
In 1995, the case-time-control design was
proposed, very similar to the case-crossover, used,
however, to the study of chronic exposures. The time
trends in exposure are limitations to the results of self-
matching studies. Thus, the casetime-control design
includes an adjustment for time trend from the controls.
The self-controlled designs have some limitations.
The use of subjects as their own controls already adjusts
for fixed factors, but it does not account for the
variations over time. The adjustment for time trends in
the case-time-control studies is limited to the measured
variables, infrequentist analyses.
This adjustment may be imprecise for time trends
that vary over immeasurable variables. There is a
tendency to overstate the exposures in the current period
and to underestimate in the reference periods. The
model is based on the assumptions that the exposures
are independent and that there is no carryover effect
from one period to the other. There is, also, the selection
and the information bias, as in any case-control study.
The use of the subjects as their own controls has been
limited in clinical trials, for many reasons. One of them
is the ethical worry of allowing patients with placebo,
only, during the reference period (control window). This
is not possible in lots of situations under study (Louis et
al., 1984; Suissa, 1998; Greenland, 1996; Greenland,
1999; MacLure & Mittleman, 2000). This way, the
bidirectional studies may be an option to this limitation.
Statistical approaches
In order to make the self-controlled clinical trials
widespread, there is the urgent need for the
development of effect estimates that are free from bias
caused by time trends. MacLure (1991) proposes the
Mantel-Haenszel estimator of the rate ratio, with
confidence intervals for sparse data. Since that, other
choices have been used, like maximum likelihood
(Marshall & Jackson, 1993) and logistic regression
(Suissa, 1995). Recently, Navidi & Weinhandl (2002)
have compared four sampling schemes for case-
crossover designs, to evaluate which one would better
control for confounders that vary predictably with time.
The best scheme, according to this study, was the semi-
symmetric bi-directional. Other new methods for
achieving control of unmeasured confounding have
been studied, from more elaborate extensions of Poisson
regression to hierarchical modeling. It is also
fundamental the development of methods that can
handle exposures which occur concurrently and non-
independently.
It is also possible to use less orthodox statistical
approaches. The increasing power of computers is
bringing Bayesian methods to the fore. It allows
handling unknown heterogeneity sources or, at least, not
attributable to any specific variable (Struchiner, 2004),
dealing with the uncertainty. The proposal of the self-
controlled studies also allows the review of the
variables, going beyond the simple dichotomies,
searching for a deep understanding of the patient as
human being, once he will be compared only to himself.
The fuzzy logic handles this imprecision at the analysis
and the variables construction.
It works the concept of partial truth, where the
transition between sets is gradual, and not abrupt (Shaw
& Simões, 1999). The fuzzy sets theory is an important