Note: Descriptions are shown in the official language in which they were submitted.
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
1
METHOD FOR PREDICTION OF A PLACEBO RESPONSE IN AN INDIVIDUAL
TECHNICAL FIELD
The invention pertains to the technical field of methods for providing
improved
therapeutic treatments and improved clinical trials for therapeutic
treatments.
More particularly this relates to methods for predicting placebo response or
effect
and to systems providing such predictions and using the generated data of the
predictions.
BACKGROUND
The clinical development of new drugs or treatments in major therapeutic
indications such as chronic pain (including neuropathic pain, migraines...),
mental
disorders, depression, epilepsy, Parkinson, asthma is complex and is not
efficient.
This is mainly due by the fact that many Phase 2 and 3 clinical trials are
abandoned or fails because of safety or the inability to demonstrate clear
superiority of the tested drug versus a placebo despite promising results
observed
in vitro and/or in pre-clinical studies. The reason for this is that, in
therapeutic
fields such as e.g. pain or depression, the placebo response by itself has a
pronounced effect on the primary outcomes of the clinical studies. More
specifically, one recognizes today that the investigator behavior vis-à-vis
its
patient as well as the patients expectations (in terms of drug efficacy and
overall
well-being) have a profound impact on the patient assessment regarding the
efficacy of the medication.
Hence the steep rise in attrition rate of drug development is a major concern
for
both clinicians and pharmaceutical companies that face major difficulty of
obtaining market authorization of new drugs in nowadays prominent therapeutic
fields such as e.g. pain and depression. On the standpoint of the health care
attendant, managing properly the placebo effect/response may positively
contribute to the better well-being of its patients. On the standpoint of the
pharmaceutical companies, controlling the placebo effect is essential to
design
appropriately clinical trial to allow a clear differentiation between, on the
one
hand, the physiological effect of the studied drug and, on the other hand, the
other effects collectively referred to as the placebo effect.
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
2
Altogether, (i) the high impact of the placebo response on the drug efficacy
evaluation and (ii) the absence of common traits among patients that allow to
measure, at the level of a population, to which extent the placebo response
interferes with the physiological assessment of a new drug candidate make it
very
difficult to demonstrate its superiority. As a result, both clinical research
scientists
and pharmaceutical companies need improved clinical studies designs and
improved patient's characterization able to differentiate the placebo response
from
the physiological effect of the tested drug.
It was found that the placebo effect is multifactorial in nature. On the one
hand
the effect is a learning phenomenon, which is influenced by the manipulation
of
different variables including patient expectation, (bio)physical, prior
experiences,
observational and social learning as well as personal traits. Hence, the
placebo
effect is mainly patient-dependent. Each individual may demonstrate a
different
response based on his/her therapeutic history and personality related aspects.
It was furthermore found that the placebo-effect is disease dependent, whereby
an individual will show an effect which differs from disease to disease.
It was furthermore found that the placebo-effect is time dependent, whereby an
individual will show a placebo response which evolves with time or time of
treatment. Hence patients may respond to a placebo effect differently at the
start
of a treatment compared to the level of response during or at the end of a
treatment. Individuals who respond to placebo or who demonstrate a propensity
to said 'response shift' or response drift may be more amenable to lower
dosages,
improved therapeutic outcomes, higher self-reported perceived improvements,
quality of life or the like.
Similarly, an individual may show a nocebo effect which evolves with time or
time
of treatment. Hence patients may respond to a nocebo effect differently at the
start of a treatment compared to the level of response during or at the end of
a
treatment. Individuals who respond to nocebo or who demonstrate a propensity
to
said 'response shift' or 'response drift' may be more amenable to higher
dosages,
decreased therapeutic outcomes, lower self-reported perceived improvements,
quality of life or the like.
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
3
Several questionnaires, biophysical tests or virtual reality tools have
already been
developed and used to assess some aspects of the placebo effect in an
individual.
However, because of their stand-alone and very narrow nature, these
questionnaires and biophysical tests do not allow giving an accurate
estimation of
a placebo effect present in the individual.
WO 2005027719 describes a method for predicting the predisposition to a
placebo
effect, based on biological markers. The method is very one-sided, and does
not
take into account the multifactorial nature of the placebo effect.
WO 2013039574 describes a method for selecting participants for a clinical
trial
whereby the participants are screened based on their responsiveness to placebo
treatment. The method in WO 2013039574 thereto makes use of an assessment
of the bodily self-image or self-identity, e.g. an individual's perception of
their own
self in relation to, or in relationship with their body. The method described
in WO
2013039574 is one of the methods available in the prior art to classify
subjects
among placebo responders and non-responders but relies only on the assessment
of the adaptability of a subject's perception of its bodily self-image. The
assessment according to WO 2013039574 fails to provide a method relying on the
proper understanding of the inter-relationships between various factors either
psychological or physiological in nature that contribute to a placebo effect.
Accordingly WO 2013039574 fails to describe a subject's global and unbiased
placebo response signature or pattern.
U520140006042 describes a methodology for conducting studies, thereby
generating a placebo responder index. The index is obtained by comparison of
data obtained from a patient with previously obtained data. Having to use a
comparative approach for determining a putative placebo response is not
desired
as such comparison has to rely on previously obtained data. If such previous
data
is flawed or there is even the slightest difference in the test circumstances,
than
the comparison may lack in trustworthiness. Moreover, a deviation in result
can
occur if the compared data does not originate from the same individual. This
can
give a distortion in the obtained result.
Currently, either for decreasing the level of attrition rates in clinical
trials or for
improving the accuracy of the contribution of the physiological effect of a
(drug)
treatment to the overall response of a patient when treating diseases where
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
4
placebo effect intervenes or, more generally, for improving a treatment of
diseases where placebo effect intervenes, the prior art inappropriately solves
the
problem of accurately defining the propensity of a subject to raise a placebo
response or to reveal a placebo effect. Secondly, the existing methods,
especially
the questionnaires, are time-consuming and put a heavy burden on the patient
having to undergo the testing.
The present invention aims to resolve at least some of the problems mentioned
above.
SUMMARY OF THE INVENTION
The current invention aims to provide a method and tool, for predicting the
propensity of a placebo effect in an individual, said prediction is built on a
multifactorial approach of traits which are related to the placebo effect. The
methodology and tool start from a predefined amount of data, obtained from the
individual, which is used in a mathematical model to define a correlation
between
the input data, whereby the correlation enables to provide a measure of the
placebo response. The invention offers a means for generating an accurate
placebo score using a limited number of input variables. It has been
surprisingly
observed that the relationship between input variables (correlations or other
forms
of mathematical relationships between two or more random variables or data
points) can be used to have a "straightforward" prediction of the placebo
response
(without "undue" questioning the patients).
Because of the multi-facet approach of the current invention, said prediction
is
more reliable than the other methods currently known in the art. As it is
based on
a correlation of inherent characteristics and data obtained from one
individual,
often on a specific time point, thereby omitting the necessity of comparing
the
latter with previously obtained data (e.g. from other individuals), the latter
is
more trustworthy. Hence, the results of the current method can be deployed in
various stages of patient treatment and/or clinical trials, including for
balancing
the placebo responders in various groups (arms) of a clinical study, all of
which
are known to be affected by a placebo effect. The current invention relates
thereto
to a method for predicting a placebo response in an individual, according to
claim
1. In further aspects, the current invention also relate to a computer
implemented
method and product and a companion diagnostic tool. The current invention also
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
relate to methodologies that can be used in clinical trials or for improving
the
quality of the results of the latter.
DESCRIPTION OF FIGURES
5
Figure 1 shows a schematic overview of an embodiment of the methodology
according to the current invention.
Figure 2 shows a screenshot of a computer interface according to an embodiment
of the current invention, whereby the intensity of a placebo response is
predicted
based on input traits.
Figure 3 shows a decision tree following example 2.4.
DETAILED DESCRIPTION OF THE INVENTION
The present invention concerns methodologies for determining a placebo effect
in
an individual, or to determine the propensity that an individual has to
respond to a
placebo effect. The importance of the placebo effect in clinical trials and in
patient
therapy has only begun to be acknowledged in the last decade. Some of the
neuroanatomical and neurophysiological substrates of the placebo effect have
been elucidated in the past years, but development of prediction tools for
placebo
effect have until now been largely underexposed. It is the aim of the current
invention to develop a methodology and system for predicting a placebo
response
in an individual and for implementing the latter in drug design and clinical
trials.
Unless otherwise defined, all terms used in disclosing the invention,
including
technical and scientific terms, have the meaning as commonly understood by one
of ordinary skill in the art to which this invention belongs. By means of
further
guidance, term definitions are included to better appreciate the teaching of
the
present invention.
As used herein, the following terms have the following meanings:
"A", "an", and "the" as used herein refers to both singular and plural
referents
unless the context clearly dictates otherwise. By way of example, "a
compartment" refers to one or more than one compartment.
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
6
"About" as used herein referring to a measurable value such as a parameter, an
amount, a temporal duration, and the like, is meant to encompass variations of
+/-20% or less, preferably +/-10% or less, more preferably +/-5% or less, even
more preferably +/-1% or less, and still more preferably +/-0.1 /0 or less of
and
from the specified value, in so far such variations are appropriate to perform
in the
disclosed invention. However, it is to be understood that the value to which
the
modifier "about" refers is itself also specifically disclosed.
"Comprise," "comprising," and "comprises" and "comprised of" as used herein
are
synonymous with "include", "including", "includes" or "contain", "containing",
"contains" and are inclusive or open-ended terms that specifies the presence
of
what follows e.g. component and do not exclude or preclude the presence of
additional, non-recited components, features, element, members, steps, known
in
the art or disclosed therein.
The recitation of numerical ranges by endpoints includes all numbers and
fractions
subsumed within that range, as well as the recited endpoints.
The expression " /0 by weight" (weight percent), here and throughout the
description unless otherwise defined, refers to the relative weight of the
respective
component based on the overall weight of the formulation.
The current invention thereto provides for a method for predicting a placebo
response in an individual. Said method may comprise collecting data via the
following steps:
- querying said individual on personality and health traits; and/or
- performing one or more social learning and/or (bio)physical tests on said
individual.
In a preferred embodiment, a Scoring Factor will be attributed to said
individual,
whereby said Scoring Factor is a measure of propensity to raise a placebo
response and a measure of the intensity of the response. To that purpose, the
data obtained is used in a mathematical model, the output of said model being
the
Scoring Factor.
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
7
This is different from what is currently known in the art. To date, no
mathematical
model or tool for qualifying, quantifying and/or predicting the placebo effect
of an
individual exists which takes into account a subset of aspects that contribute
to
the placebo effect such as the individual's personality traits, health traits,
(bio)physical measures etc. Questionnaires taken alone or (bio)physical tests
used
alone currently used never give a value for a placebo effect, as they are
stand-
alone approaches. Not only do they fail to take into account the
multifactorial
nature of the placebo effect but if the skilled person of the art decides to
use them
all (together or sequentially), he will fail in providing a measure of the
placebo
effect since conducting the corresponding surveys and tests is not feasible.
In the context of the current invention, the terms 'predicting' and any
derivatives
thereof (predictive, prediction...) is to be understood as providing a
probabilistic
picture of an analysed feature, said picture is preferably computed by a
model.
Alternatively or in addition, predicting is to be understood as anticipating
the
evolution of said feature in time or during a predefined time period.
For the purpose of current invention, the term 'pain disorder' is to be
understood
as an acute or chronic pain experienced by a patient. Said pain disorders may
be
subdivided in three groups:
- Pain associated with psychological factors
- Pain associated with psychological and a general medical condition
- Pain disorder associated with a general medical condition
Hence, said pain:
- may be caused by damages or diseases that affect the somatosensory system
(neuropathic pain);
- from activation of nociceptors (nociceptive pain);
- caused or increased by mental, emotional or behavioural factors
(psychogenic
pain);
- breakthrough pain, e.g. caused by cancer; or
- arising from a sudden activity (incident pain).
For the purpose of the current invention, said 'correlation' or 'correlating'
is to be
understood as a mathematical relationship between two or more random variables
or data points. Preferably, said correlation is predictive or allows
identifying a
predictive relationship between the analysed variables.;
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
8
In the context of the current invention, the term 'Placebo' can be any of
typically
inert or active substances, formulations, drug-based therapies or non-drug-
based
therapies administered to, given to or used by a patient, e.g., tablets,
suspensions
or injections of inert ingredients, e.g., sugar pills or starch pills, or
other mock
therapies, e.g., fake surgeries, fake psychiatric care, or others that have
been
used, typically as controls, for a putative "real" treatment (in order to
obtain a
purported, supposed, or believed therapeutic effect on a symptom, disorder,
condition, or disease, or prescribed, recommended, endorsed or promoted,
knowingly or unknowingly, to another, notwithstanding that the therapy is
actually
ineffective for, has no known physiologic effect on, or is not specifically
effective
for the symptom, disorder, condition, or disease being treated).
In the context of the current invention, the term 'Placebo effect' means any
specific or non-specific psychobiological phenomenon attributable to the
placebo
and/or to the treatment context irrespective to the fact that the placebo is
actually
administered or not. The placebo effect as meant in the context of the current
invention highlights the central role of expectations and suggestions in
placebo-
related phenomena and diseases.
In the context of the current invention, the term 'Placebo response' means the
outcome of the placebo effect as expressed, perceived or measured by one or
more individuals for qualifying or quantifying either the improvement or the
deterioration (nocebo response) in a symptom or a physiological condition in
the
context of the administration of a placebo and/or a treatment.
Said Placebo response not only includes the presence or the absence of the
response itself but equally relates to the intensity of the response that is
given or
expressed by the individual.
Said placebo response may be disease-dependent and/or time-dependent.
In the context of the current invention, the term 'response shift' or
'response drift'
means a change in the placebo response along a treatment, a clinical trial or
any
health-related intervention.
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
9
In the context of the current invention, 'trait or traits' is to be understood
as all
kinds of variables, whether or not directly linked to an individual, which can
be
inputted in the model according to the current invention, and which are used
to
come to the Scoring Factor. More in detail, said traits are identified by a
skilled
person based on current understanding of different aspects potentially related
to a
placebo aspect, and commonly collected with existing questionnaires and/or
tests.
In the context of the current invention, 'personality traits' is to be
understood as
the characteristics of an individual which relate to the psyche of the
individual, the
physical characteristics of said individual and/or the personal background
information of that individual. Said characteristics of the psyche may
include, but
are not limiting to emotional characteristics, behavioural characteristics,
general
beliefs of the individual and/or emotional traits.
Said health traits may include all health related information of the
individual, as
well as of family of the individual. Said health traits may for instance
include, but
are not limiting to past and current diseases, received treatments, current
and
past medicinal use, potential health risks, genetic predisposition for disease
development, etc.
Within the context of the current invention, said social learning might be
understood as a process in which individuals observe the behaviour of others
and
its consequences, or specific situations and models to modify their own
behaviour
accordingly. Said social learning test includes providing an individual with
behavioural, environmental and/or exemplary information or stimuli, thereby
eliciting (or not) a response in said individual, based on the information
received.
In the context of the current invention, said (bio)physical test is to be
understood
as any test, relating to the measurement or detection of a biophysical
parameter.
For instance, said (bio)physical test may include but is not limited to
measuring or
analysing a biological compound of said individual; measuring or detecting a
biological reaction of said individual; performing a neurological test on said
individual; measuring or detecting a sensory reaction; performing a tactile
test on
said individual.
For instance, the Somedic Thermotest apparatus (Somedic AS, Stockholm,
Sweden) may be used to deliver quantified and reproducible heat impulses via a
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
2.5 x 5 cm (12 cm2) - Peltier thermode applied to the thenar eminence of the
non-dominant hand.
By preference, said (bio)physical test involves a neurological, somatosensory,
5 tactile or analytical test, or virtual reality tools or any combination
thereof.
Examples of such objective tests may include heart rate monitoring, blood
pressure monitoring, monitoring respiration, measuring one or more components
or metabolites of blood (e.g. blood chemistry) or other bodily fluid,
measuring skin
10 parameters such as blood flow, temperature, or conductance; or other
physiological measures including measuring any brain or neurological activity,
skin
conductance resonance (SCR), electroencephalography (EEG), quantitative EEG
(QEEG), magnetic resonance imaging (MRI), functional MRI (fMRI), computed
tomography (CT), positron emission tomography (PET), electronystagmography
(ENG), single photon emission computed tomography (SPECT),
magnetoencephalography (MEG), superconducting quantum interference devices
(SQUIDS), electromyography, eye movement tracking, and/ or pupillary diameter
change, pain tests such as for instance heat pain procedure.
In the context of the current invention, said Scoring Factor is to be
understood as
a measure for a certain analysed feature (in the current case the propensity
to
exhibit a placebo effect or response). Said Scoring Factor may be a numerical
factor or parameter, being an indication of the analysed feature based on a
specific scale, whereby the higher (or lower) the numerical factor resides on
the
scale, the more likely it is that the analysed feature is present. For
example, in the
context of the current invention, said Scoring Factor may provide a scale with
regard to the propensity of an individual to be eligible for a placebo effect.
In
another embodiment, said Scoring Factor may be a classification of an analysed
individual. For example, in the context of the current invention, said Scoring
Factor may determine whether an individual is a responder or non-responder to
a
placebo effect ('yes' or 'no'). In yet another embodiment, said Scoring Factor
is a
profile or outline of the Placebo response. In general, said Scoring Factor is
a
(predictive) value (e.g. a colour code, a definition, a term, a numerical
factor...) of
the placebo response or placebo effect of an individual.
In an embodiment, said Scoring Factor will be compared to one or more cut-off
values or thresholds, in order to determine whether a placebo response is
present
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
11
in an individual. If said Scoring Factor is higher than a predefined cut-off
value,
this indicates the presence of a placebo response, or a high propensity of
developing the latter.
If the scoring Factor is situated in below the cut-off value, but above a
second cut-
off value, then a placebo response might be present. Below the second cut-off
value, a placebo response is not present.
In another embodiment, said Scoring Factor will be mapped on or compared to a
predefined scale, whereby the height of the Scoring Factor is directly
proportional
to the propensity of developing a placebo response or the presence of a
placebo
response in the individual.
The current method has as advantage that it offers a model for a placebo
effect or
response, thereby adopting a multifactorial and multi-integrated approach.
Models
for the placebo response have focussed until now on a very limited amount of
information, and studies have failed to provide a coherent link with the data
gathered and the placebo response as such. The current methodology and tools
derived thereof strive to take into account multiple facets of the placebo
effect,
thereby offering a reliable tool, for predicting a placebo response in a vast
amount
of medical indications. To that purpose, the current invention describes a
methodology and tools which make use of objectified data (e.g. obtained by
testing and/or questioning an individual), and which is to be considered as
the
'input' for the final prediction.
In a preferred embodiment, said method will include data from:
- one or more personality queries;
- one or more health queries;
- one or more social learning tests; and
- one or more (bio)physical tests
relating to or performed on an individual.
In another embodiment, said method comprises any combination of 2 or 3 of
above queries and/or tests.
Figure 1 shows a schematic overview of a possible methodology according to the
current invention.
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
12
In an embodiment, said personality query comprises one or more questions
selected from clusters of questions for characterizing an individual's
personality
traits or characteristics which are stable over time and attributable to a
person
itself and not to the effect of its environment. Said cluster of questions
related to
personality comprises one or more questions for measuring the Big Five
components (readily known in the art) of personality namely individual's
openness
to experience, conscientiousness, extroversion, agreeableness, and neuroticism
(or emotionality), all of which are well-known to the skilled person in the
art.
In another embodiment, said query comprises one or more questions selected
from clusters of questions for measuring or evaluating the impact of an
individual's
surrounding on its perception of health-related issues.
Said cluster of questions related to the impact of the surrounding comprises:
- one or more questions for measuring the impact of the caregiver's
behaviour (agreeable, open, severe..) or intervention (oral, acts...),
- one or more questions relating to the sensation of contagion,
suggestibility
or any other factor likely to influence the balance between deliberate and
automatic processing of information on a health symptom onset, evaluation,
relief, evolution...
- one or more questions for evaluating the level of anxiety, fear,
discouragement, hopelessness, depression related to the environment of a
clinical
setting or a caregiver.
In another embodiment, said query comprises one or more questions selected
from clusters of questions for evaluating the impact of an individual's
environment
on his belief of a just world, psychological well-being, psychological quality
of life,
life satisfaction, resistance to stress and depression...
In another embodiment, said query comprises one or more questions selected
from clusters of questions for measuring the individual's expectations with
respect
to an external stimulus, positive and negative outcomes of an intervention or
a
treatment, and for evaluating his propensity to have a positive or a negative
attitude with respect to external factors or health symptoms, specific
treatments
to relief health symptoms...
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
13
In another embodiment, said query comprises one or more questions which are
asked after exposing said individual to either expectation-influential or
neutral
information. For the purpose of the current invention, said information
includes all
information, directly or indirectly related to the performed test and/or the
placebo
given and mode of action of said placebo.
In another embodiment, said query comprises one or more questions selected
from clusters of questions for evaluating the attitudinal and emotional
response of
an individual to external stimuli. Said cluster of questions comprises
questions for
measuring the level of control that the individual believes to have on his
life, the
level control of external factors or health symptoms on his life such as luck,
fate,
life events or powerful others (such as e.g. relatives, health professionals,
colleagues at work etc.) and for measuring the level of control of powerful
others
such as relatives or social learning.., on his attitude to resist, fight or
overcome
aggressive external factors or health symptoms.
In another embodiment, said query comprises one or more questions selected
from clusters of questions for evaluating the level (severity) of health
symptoms.
Said such cluster of questions may comprise one or more questions for
measuring
to which extent the individual estimates that health symptoms influence his
general physical and psychological condition comprising his body function,
activity,
mobility, working ability, relations with other people, sleep, life
satisfaction, mood,
..., and to which extent the influence of the health symptoms on his general
condition evolve with time.
In yet another embodiment, said such cluster of questions may comprise one or
more questions for evaluating to which extent the caregiver estimates that
health
symptoms influence a patient's general physical and psychological condition
comprising his body function, activity, mobility, working ability, relations
with
other people, sleep, life satisfaction, mood..., and to which extent the
influence of
the health symptoms on his general condition evolve with time.
In another embodiment, said query comprises one or more questions selected
from clusters of questions for evaluating the level (severity) of pain. Said
cluster
of questions comprises one or more questions for measuring:
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
14
- to which extent the individual estimates that said pain influences his
general physical and psychological condition comprising his body function,
activity,
mobility, working ability, relations with other people, sleep, life
satisfaction,
mood..., and to which extent the influence of the pain on his general
condition
evolve with time;
- to which extent the caregiver estimates that said pain influences a
patient's
general physical and psychological condition comprising his body function,
activity,
mobility, working ability, relations with other people, sleep, life
satisfaction,
mood..., and to which extent the influence of said pain on his general
condition
evolve with time.
In another embodiment, said query comprises one or more questions selected
from clusters of questions for characterizing the typology and localisation of
pain.
Said cluster of questions comprises one or more questions for defining:
- the painful areas,
- how the individual translates pain in terms and qualifications such as
painful cold, burning, electric shocks, mechanical shocks, tingling, pins,
needles,
numbness, itching etc.
- the physical status of the painful area such as hypoesthesia to touch,
hypoesthesia to prick, pain caused or increased by mechanical actions on the
body
such as brushing, pinching etc.
In a further embodiment, said query comprises one or more questions chosen
from any of the clusters of questions as outlined above. The clusters as
described
above may come in the form of questionnaires known in the art (e.g. big Five,
Belief in Just World, etc.) or may comprise questionnaires that are
specifically
designed by the inventors of the current invention.
The Scoring Factor describing the propensity of a placebo response will
preferably
be computed by a mathematical function of the input data. Said model will be
built
such that based on the input data, the propensity of the placebo effect may be
calculated for each tested individual.
The current method thereto offers one or more algorithms which allow
correlation
of the input data with the propensity of having a placebo effect. By
preference,
said mathematical model is computer implemented.
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
Let P be a population defined by a n-row and p-columns matrix X of input data
and
a n-sized Y vector of observed placebo responses. Each of the n rows of X
5 corresponds to a patient. Each of the p columns of X corresponds to a
trait i.e. a
personality trait. A signature S is defined as a subset of the p input traits.
S is of
size p' smaller or equal to p. S is used to define a new n-rows and p'-rows
matrix
called X' which together with Y defines P'.
10 A model estimation occurs on P'. The resulting model is called M. M is a
function
which maps a vector x of size p' to an output y. This output y is the
predicted
placebo response, in the current invention being the Scoring Factor.
15 TRAITS
The p traits constituting the columns of matrix X described herein were
identified
by a skilled person based on current understanding of different aspects
potentially
related to placebo effect, and commonly collected with existing questionnaires
and/or tests. A person of the art will understand that the traits captured by
such
queries and/or tests might be captured as well by other but similar queries or
tests. Thus queries and/or tests capturing the same traits but formulated
differently than herein described may be employed in X as well instead of
restricting the definition of X to the questionnaires and/or tests described
above.
TYPE OF PREDICTION
In one embodiment, entries of the Y vector are binary variables corresponding
to
placebo responders and non-responders respectively.
In another embodiment, entries of the Y vector are ordinal variables with a
finite
number of modes corresponding to different placebo response levels (for
example
non-responders, low responders, mild responders, strong responders).
In another embodiment, entries of Y are continuous variables corresponding
either
to placebo response likelihood or placebo response intensity.
In another embodiment entries of the y vector are categorical variables with a
finite number of modes corresponding to different forms of placebo responses.
MODEL
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
16
In one embodiment, the model M has the form of a linear model for regression
or
classification.
In another embodiment, the model M has the form of a k-Nearest Neighbour.
In yet another embodiment, the model M has the form of a decision tree.
In another embodiment, the model M is a set of models of the forms defined
above built on various sub-samplings of the columns and or rows of P'.
Alternatively, classification or regression can be achieved using other
mathematical methods that are well known in the art.
In all cases, the sensitivity and specificity trade-off of the models can be
tuned via
a meta parameter according to the applicative context. The current invention
covers all possible trade-offs.
As described herein, methods to predict a placebo response or to identify
individuals more likely to respond to placebo, is not meant to imply a 100%
predictive ability, but is meant to indicate whether individuals with certain
traits
are more likely to experience a placebo response than individuals who lack
such
characteristics. However, as will be apparent to one skilled in the art, some
individuals identified as more likely to experience a response may nonetheless
fail
to demonstrate measurable placebo response. Similarly, some individuals
predicted as non-responders may nonetheless exhibit a placebo response.
By preference, attribution of the Scoring Factor is computer implemented. The
latter allows quick and accurate analysis of input data. In one embodiment,
said
attribution can be performed on a place remote from of the site of data
collection.
Said data can be obtained on one specific site and transferred to a second
site
(e.g. via electronic ways, systems stored in the cloud, etc.), where data
analysis
and Scoring Factor attribution occurs.
Hence, the current invention also relates to a computer implemented method for
predicting a placebo response in an individual. By preference, said computer
implemented method comprises:
(a) inputting data obtained from personality and health-related queries,
social
learning and/or (bio)physical test performed by an individual;
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
17
input data(b) computing a measure of propensity to respond to a placebo
effect.
In an embodiment, one or more correlations may be calculated between the input
data. Said 'correlation or correlations' is to be understood as the
relationship
between each of the individually collected data points or the whole data
collection
with the feature to be investigated. Said correlation may equally be
understood as
the mutual relationship of the collected data with said feature. In the
current
invention, the feature to be investigated is the propensity to respond to a
placebo
effect, which will be defined by virtue of an attributed Scoring Factor.
A screenshot of a possible embodiment of a computer implemented interface
according to the current invention is shown in figure 2. Based on certain
input
traits, the intensity (Scoring Factor) of a placebo response is predicted. In
the
embodiment as shown in figure 2, the Scoring Factor is given by means of a
percentage.
In a further aspect, the current invention also relates to a computer program
product for predicting of a placebo response in an individual. By preference,
said
computer program product comprises at least one computer-readable storage
medium having computer-readable program code portions stored therein, the
computer-readable program code portions comprising instructions for comparing
data obtained from personality and health-related queries, social learning
and/or
(bio)physical tests performed by an individual and/or with a data collection
obtained from previously tested individuals, thereby computing a Scoring
Factor
for said individual, whereby said Scoring Factor is a measure of propensity to
respond to a placebo effect.
In a further embodiment, the input data from said individual, as well as the
Scoring Factors thereof may be stored in a database; said database may be
stored
on an external server. Such database may serve for further analysis and for
further fine-tuning of the algorithms and queries used for determining said
Scoring
Factor. In another embodiment, query or queries used are equally stored on an
external server. The latter allows third parties to make use of the
methodology
and system, e.g. by remotely logging in to the system. In another more
preferred
embodiment, said database and queries are applicable for cloud computing and
being stored and/or computed in the cloud.
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
18
In a preferred embodiment, the obtained Scoring Factor and optionally the
imputed test and/or query results will be summarized in a report, said report
may
be a digital report sent to the person making use of the methodology.
The method of the current invention is specifically useful for predicting a
placebo
effect in an individual or for predicting the propensity of an individual to
raise a
placebo response, said individual suffering from or prone to a therapeutic
indication where a placebo is used as comparator in clinical development
trials or
where a placebo effect is found relevant for said therapeutic indication. More
in
particular, it is related to indications where a high rate of placebo response
has
been detected. These indications may include but are not limited to developing
asthma, depression, Peripheral Neuropathic Pain, chronic pain, terminal
cancer, a
neurodegenerative condition, a spinocerebellar ataxia, encephalopathy, or
other
condition causing cerebellar degeneration, congestive heart failure, muscular
dystrophy, cirrhosis of the liver, Parkinson's disease, schizophrenia,
Huntington's
disease, multiple sclerosis (MS), amyotrophic lateral sclerosis (ALS),
osteoarthritis,
rheumatoid arthritis or other form of arthritis, diabetes mellitus, emphysema,
macular degeneration, or glomerulonephritis.
The latter indications are known to have a link with a placebo effect. Hence,
by
implementing the current invention in view of these therapeutic indications,
treatment of a patient may be optimised, unnecessary treatments may be avoided
and side-effects may be minimised. Therefore the current invention also
relates to
a method of identifying individuals for a therapeutic treatment based on their
propensity to respond to a placebo effect, thereby predicting a Scoring Factor
according to the method as described above.
In another, preferred embodiment, said method is particularly useful for
predicting
a placebo effect or response in an individual suffering from or prone to
developing
a pain disorder. It was found that especially in the field of pain treatment;
the
placebo effect may be for over 50% responsible for the 'activity' of an
administered pain-management drug.
The method of the current invention is specifically useful for predicting a
placebo
response in an individual suffering from or prone to a pain disorder where a
placebo is used as comparator in clinical development trials or where a
placebo
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
19
effect is found relevant for said pain disorder. More in particular, it is
related to
pain disorders where a high rate of placebo response has been detected.
The methodology according to the current invention can be applied in a fast
way,
if necessary even multiple times a day. This is a big amelioration with regard
to
the methodologies currently been used, which are tedious and require a
significant
amount of time. The methodologies used to date to evaluate a possible placebo
response do not allow multiple testing on one day.
In that respect, the methodology according to the current invention can be
performed within a time frame of about or less than 3 hours, preferably less
than
2 hours, more preferably less than 1 hour. More preferably, said methodology
can
be performed at least two times a day, e.g. 2 or 3 times a day. Said
methodology
according to the current invention can be performed multiple times a week, at
least 7 times a week, more preferably 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
18, 19,
20, 21, etc. or more times a week.
The methodology according to the current invention comprises less than 250
questions and/or tests which have to be completed by the individual, more
preferably 160 questions and/or tests or less than 160 questions and/or tests,
more preferably less than 100, more preferably between 1 and 99 more
preferably
between 1 and 90, more preferably between 1 and 80, between 1 and 70,
between 1 and 60, between 1 and 50, less than 50, less than 40, less than 30,
between 1 and 20, less than 20, between 1 and 15, less than 15, between 1 and
10.
As a result thereof, the methodology can be performed in a very fast manner,
without causing any undue burden to the individual or patient.
In a further aspect, the method of the current invention may be equally used
for
selecting participants in a clinical trial. As used herein a "clinical trial"
or "clinical
study" is to be understood as encompassing all types of health-related studies
in
which obtaining data regarding safety and efficacy is a pre-requisite. As
such, said
clinical trial or study may refer to any research study, such as a biomedical
or
health-related research study, designed to obtain data regarding the safety or
efficacy of a therapeutic treatment such as a drug, device, or treatment. Said
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
clinical trial or study may equally relate to epidemiological or observational
studies, market studies and surveys.
Such studies can be conducted to study fully new drugs or devices, new uses of
5 known drugs or devices, or even to study old or ancient treatments that
have not
been used in Western-style medicine or proven effective in such studies.
Clinical
studies frequently include use of placebo treatments for one group of
individuals.
Clinical studies are in some embodiments conducted as double blind studies
wherein the individuals do not know whether they received a putative active
10 ingredient or treatment for the condition being tested, or a placebo
with no known
physiologic effect on the condition. In addition, in such double-blind
studies, the
researchers collecting the data also do not know which individuals received
placebo or active treatment. Double blind studies help prevent bias for or
against
the test treatment. Moreover, while the use of placebos can help prove the
15 efficacy of new drugs, if a research study turns out to include many
people who
respond to the placebo, it is much more difficult to establish the efficacy of
what
may well be a worthwhile therapeutic compound. Another pitfall is that on
small
cohorts (typical phase I and II), the distribution of the placebo-responders
is very
likely unbalanced. This might turn out to favour or disfavour the treatment
under
20 study, but in any case, it represents a lack of control over the placebo
response.
Hence, clinical trials often suffer from the fact that obtained data and
conclusions
made thereof are stained by the influence of the placebo-effect which was not
(or
not adequately) taken into account. As a consequence, the obtained results
might
lack reliability. Often the problems are traced back to an inadequate
selection of
participants or non-optimised stratification of the participants in the trial.
By
starting with incorrectly stratified or non-optimal groups of participants,
the whole
set-up of the trial may be compromised. Hence, there is thus a need in the art
for
an improved method for selecting participants for a clinical trial or for
allocating a
trial's patient into various arms of the trial.
Said method for selecting or managing participants of a clinical trial
comprises
preferably the following steps:
(a) establishing at least one inclusion and / or exclusion criterion for the
clinical
trial that encompasses a measure of a participant's propensity to respond to a
placebo;
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
21
(b) eliminating, a priori, from the clinical trial any participant who does
not meet
the required criteria for inclusion or exclusion.
In an embodiment of the current invention, said clinical trial relates to a
pain
disorder.
For the purpose of the current invention, said managing includes allocation of
participants in a balanced way into various arms of the trial.
In a preferred embodiment, a measure of propensity to respond to a placebo
effect is predicted according to the method as described above. By preference,
said only those candidates will be selected which show a Scoring Factor
conform to
or within a specific predefined range or profile.
Because of the potential for added time or expense to qualify a candidate for
a
clinical study, it is useful to first establish that the candidate is
otherwise qualified
to be a participant in the clinical trial based on the inclusion and exclusion
criteria
for the clinical trial. It is also useful in some applications of the methods
that
likelihood of being prone to a placebo effect be used as an additional
criterion for
inclusion in, or exclusion from, the study or for allocating a participant
into a
specific arm of the trial.
In an embodiment, said clinical trial relates to a pain disorder.
As a consequence, the current invention also relates to a drug approved for
the
therapeutic treatment by a regulatory agency, said drug has been tested in one
or
more clinical trials whereby said participants were selected according to
abovementioned method.
In a further preferred embodiment, said drug is approved for the therapeutic
treatment of a pain disorder. Such drug may include, but is not limiting to
paracetamol, non-steroidal anti-inflammatory drugs, COX-2 inhibitors, opioids,
flupirtine, tricylic antidepressants, selective serotonin and norepinephrine
reuptake
inhibitor, NMDA antagonists, anticonvulsants, cannabinoids, adjuvant
analgesics,
such as nefopam, orphenadrine, pregabalin, gabapentin, ketamine,
cyclobenzaprine, duloxetine, scopolamine or any combination of the latter.
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
22
In another aspect, the current invention also relates to a method of improving
data analysis of data from a clinical trial for a therapeutic treatment.
Said method of improving data analysis of data from a clinical trial for a
therapeutic treatment comprises the steps of:
(a) obtaining a set of raw clinical data;
(b) evaluating the raw clinical data by standard methods to generate
preliminary
results;
(c) obtaining the identity for each participant in the trial (i.e. unblinding
the data);
(d) assessing the likelihood of a placebo response in each participant
according to
the methodology and/or the computer program described above;
(e) creating a modified clinical data set by modifying the raw clinical data
by
retraction of said placebo effect for each participant.
In a preferred embodiment, said treatment is a therapeutic treatment of a pain
disorder.
The skilled person will appreciate that step (a) is a prerequisite to the
method, in
that the method cannot be applied until clinical trial data are available,
e.g. a
clinical trial is either complete, or underway to at least the point of an
initial data
collection. It is to be understood that step (b), i.e. evaluating the data by
standard
methods is not essential to the method and may be eliminated however, it is
believed it will be generally employed by the researchers or analysts and
generally
expected by regulators.
In step (d), the predisposition of the participants to be responsive to a
placebo
effect and, accordingly, to raise a placebo response is determined by the
method
described above. The participants are attributed a Scoring Factor as defined
above. The results pertained to those participants who have a Scoring Factor
conform to or within a specific predefined range, or conform to one or more
inclusion and / or exclusion criterions, are identified, eliminated, or
statistically
adjusted to account for the fact that these were likely to be prone to a
placebo
effect or to raise a placebo response during the clinical trial. The skilled
person will
understand that the data modified (identified, eliminated, or statistically
adjusted)
will be those related to the clinical trial for those participants. Data that
would not
be modified would include data not related to a likely placebo effect. Also
not
modified would be the collected data and basic factual information relating to
likely
be prone to a placebo effect (e.g. raw data would remain intact).
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
23
Data that may be modified would include response data to the therapeutic
treatment or placebo. The least preferable modification is to merely identify
suspect data that comes from a likely placebo effect, for example with a
series of
footnotes or other explanatory notes. If the data for a likely placebo effect
can be
eliminated from the data set without compromising the integrity of subsequent
statistical analyses, that may be most preferred. Alternatively, data for
individuals
likely to be prone to a placebo effect may be statistically adjusted.
Statistical
models are available and skilled persons will be readily able to apply
appropriate
or suitable statistical adjustments to the collected data to allow the
modified data
set to be created.
In an alternative step (e), or additional step (f) modified data are created
by
suppressing or re-interpreting the results of the individuals which were
wrongly
attributed to a specific arm of the trial or which caused unbalanced arms. By
creating fair comparative arms (e.g. arms with balanced placebo effect), the
data
can be normalised.
The method described above is equally suitable for improving the data quality
arising from clinical trials by reassessing this data on a regular basis on an
individual's placebo effect and its propensity to raise a placebo response,
including
its response drift/shift during the treatment. This can happen at the end of
the
clinical trial, but preferably reassessment is done on a regular basis
throughout
the course of the clinical trial on the basis of the response of the
individual.
A much clearer picture of therapeutic efficacy of a treatment may emerge from
the
study or analysis of the modified clinical data as compared to the
understanding
that comes from the raw data. By eliminating or adjusting for the likely
placebo
effect or the response shift/drift in the placebo response, confounding
effects may
be removed.
In some embodiments, the methods comprise a further step of comparing the
preliminary results and the modified results to generate a comparison, and
optionally using the comparison in connection with seeking approval from a
regulatory agency.
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
24
In another aspect, the current invention relates to a method of identifying
individuals for a therapeutic treatment based on their propensity to respond
to a
placebo effect, the method comprising the prediction of a Scoring Factor
according
to the methodology and/or computer system as described above.
For this aspect of the invention, the therapeutic treatment comprises for
example
a modified or reduced dosing regimen, a modified or reduced time of
therapeutic
treatment, a therapeutic treatment with fewer side effects than a standard of
care
therapy, an alternative to a standard of care therapy, or a placebo.
Because the method is selecting for likely placebo responders, it is expected
that
for certain therapeutic treatments with active ingredients, lower dosages,
shorter
time courses, and/ or lower circulating blood levels of active ingredient, or
the like
may work as well or provide the same clinical benefits in the likely placebo
responders as higher doses, longer time courses, and/ or higher circulating
blood
levels of active ingredient work in non-placebo responders. Because
populations of
likely placebo responders could not previously be determined a priori, it was
not
possible to consider the benefits that could accrue to this population such as
reduced side effects, reduced exposure time, reduced clearance periods, as
well as
the potential benefits for medical providers of reduced costs for such
populations.
Surprisingly, as a result of the inventor's discovery, clinical trials
designed to test
such hypotheses are now possible.
Such methods may have particular benefits where an individual is suffering
from a
health-related condition comprising anxiety, or depression or an anxiety-
related or
depression-related disorder, a neuropathy, or chronic pain and where the
therapeutic treatment is for treating the condition. Since likely placebo
responders
are more likely to notice and / or report improvements in their personal state
of
anxiety, depression, or pain (in theory by being more readily in the
"experiencing
self") - it is expected that these and related types of conditions would be
well
suited to therapeutic treatment according to the method.
These methods are meaningful for scientifically clarifying the therapeutic
role of a
proposed therapy by eliminating or minimizing confounding results, and
accordingly are valuable to the pharmaceutical industry and for the regulatory
agencies tasked with ensuring that new drugs and other therapeutic treatments
are safe and effective. The methods generally comprise the steps of assessing
a
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
Scoring Factor of a candidate thereby determining the likelihood that the
candidate will respond to a placebo based on the estimation.
The current invention equally relates to a companion diagnostic tool. Said
5 companion diagnostic tool is to be understood as a tool to predict
whether a
patient will respond to a certain therapy. In an embodiment, said companion
diagnostic tool according to the current invention is a companion diagnostic
tool
for predicting a placebo effect in an individual. Said tool preferably
comprises
instructions for computing a Scoring Factor for said individual, whereby said
10 Scoring Factor is a measure of propensity to respond to a placebo
effect, based on
data obtained from personality traits and/or health traits and/or social
learning
tests and/or on or more (bio)physical tests performed by said individual.
The latter will help improve patient outcomes and decrease healthcare costs.
For
15 patients with a certain disease, those that are identified as "not
likely to respond"
can quickly move on to other¨perhaps more effective¨therapies if they exist.
Furthermore, the companion diagnostic tool according to the current invention
helps the healthcare system save costs by identifying the patient population
that
20 will most likely benefit from the therapy, and ruling out therapies that
are not
likely to be effective. This is especially important as some higher-priced
therapeutics (e.g. for cancer) enter the market. An additional benefit can be
realized by decreased costs related to managing side effects or
hospitalizations
due to unnecessary treatments.
In another aspect, said current invention relates to the use of the companion
diagnostic tool as described above for patient specific treatment or for
stratification of individuals in view of a clinical trial for a specific
treatment,
preferably for a pain disorder.
As outlined above, the tool may be used for deciding on the optimal treatment
of a
patient. Secondly, said tool may also serve to classify/stratify individuals
enrolled
in a clinical trial or specific treatment. Prior to being enrolled in a
clinical trial, the
propensity of a placebo effect being present may first be evaluated in an
individual, after which it may be decided in which group the individual may be
categorized.
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
26
In another embodiment, said companion diagnostic tool will be useful as a tool
for
predicting whether or not, during a treatment or a trial, the outcome of the
trial is
void of a placebo response (including shift/drift). The tool according to the
current
invention is fast and reliable, can be used multiple times throughout the
course of
the trial and is suitable for qualifying and/or quantifying a placebo response
d rift/sh ift.
Finally, the current invention equally relates to a set of questions or
queries, or a
combination of the latter, for use in either a method as described above, or
for a
companion diagnostic tool as explained above.
The invention will further be described by examples which are not limiting for
the
invention.
Example 1:
Description of a clinical study aimed to collect the "input variables/data"
and to
estimate real values of a placebo response in an experimental situation where
the
level of placebo response can be evaluated a posteriori
The first example was aimed to collect among a sample of patients with
neuropathic pain i.e.,
- the input variables deemed a priori to be essential for predicting a
placebo
response and
- a real estimation of a placebo response measured in specific situations
where the level of placebo response can be evaluated ,
This Example 1 was aimed to show that the input variables/data, in the absence
of
the method and tool of the invention, are not able to predict the placebo
response
of such patients.
Thus a clinical study has been conducted [hereafter Clinical study A].
Clinical study
A had as objective to predict an individual placebo response (the Scoring
Factor)
after investigating the relationship between the patient's profile (as defined
by
his/her medical history, personality traits, expectation or general
characteristics
like age, Body Mass Index (BMI), ...) and his/her placebo response. The study
was performed in the field of peripheral neuropathic pain, and is deemed to
serve
as a model for other fields of applications.
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
27
The patients were subjected to 245 questions or queries known in the art [212
queries (expressing several trait variables and pain symptoms) have been asked
before placebo treatment and 33 queries were repeated during the study]; the
answers to these questions were defined as the "input data/variables". These
variables were found to be unable to predict a placebo Score (Scoring Factor)
as
such (that is, without any mathematical modelling), only a description of the
individual is provided. In a further attempt, the variables were used in
mathematical modelling approaches in order to arrive to a predictive score.
The inventors of the current invention then surprisingly found that the number
of
input variables can be limited, resulting in a less tedious test for the
patients,
thereby still allowing an accurate and precise prediction.
Randomization
The study was performed on 41 patients.
Patients were stratified based up on 4 different traits of personality in 2
cohorts.
Patients in Cohort 1 followed the studied placebo-reinforcing procedure
consisting
of positive expectation directed information about T4P1001 drug (in fact
placebo
pills), positive social observational learning and modulation of pain
conditioning.
Accordingly, the enrolling Investigator communicated expectation of treatment
improvement to patients. Each patient watched a video presenting T4P1001
(placebo) drug properties and describing heat pain procedure, pre-treatment
stimuli and modified post-treatment pain stimuli. The patient was then
undergoing
pre-treatment heat pain stimuli. After the pain stimuli, patients received
their first
placebo capsule and underwent a new heat pain conditioning approximately one
hour after dosing. The post-treatment heat pain conditioning protocol was
intentionally modified from the pre-treatment one as the mean intensity was
reduced to induce a patient's belief in analgesic efficacy.
Patients randomized to Cohort 2 followed the sham procedure consisting of no
expectation improvement, neutral social observational learning and no
modulation
of pain stimuli. Accordingly, the enrolling Investigator communicated neutral
information about the treatment. Patients watched a video presenting neutral
properties of T4P1001 drug (in fact placebo pills) and describing only the pre-
treatment heat pain procedure without post-treatment stimuli. They underwent a
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
28
pre-treatment heat pain stimulus protocol and they received their first
placebo
capsule thereafter. Approximately one hour after dosing, they underwent a post-
treatment heat pain stimulus intentionally set at the same intensity as before
dosing.
A posteriori evaluation of the placebo response of the included patients
This study design permitted to establish an estimation of the "really
experienced"
placebo response a posteriori for each of the patients included in Cohort 1
and
Cohort 2. Said a posteriori estimation will be used in Examples 2 and 3 for
testing
the ability of the Scoring Factors on the invention to correctly predict a
placebo
response (by comparison of the a posteriori response with the obtained Scoring
Factor).
Thus, the a posteriori placebo response has been measured by monitoring the
patient's change from baseline of pain severity after treatment, as measured
by
the Weekly mean of the daily Average Pain Scores (APS) in the last 24 hours.
In
practice, the intensity of pain using the Average Pain Score (APS) has been
measured as follows: Patients of both cohorts assessed every day their pain
intensity in a diary by answering the question "Could you please indicate us
how
was your average pain during the last 24 hours? For this, circle the most
descriptive number on this scale" [i.e;., a 11 NRS scale ranging from 0 (no
pain)
to 10 (pain as bad as you can imagine)].
The average Weekly means of the APS [WAPS] were calculated for each of the 41
patients both before [baseline] and after the treatment [placebo pill+ placebo-
reinforcing procedure (for the patients in cohort 1); placebo pill+neutral
procedure
(for the patients in cohort 2)].
It is well-known to the skilled person of the art that, for patients receiving
a
placebo drug/pill, when the change of the WAPS from baseline (AWAPS) is >0 on
the 11 NRS scale, this means that pain had increased at the end of the study
compared to the baseline. When the change of the WAPS from baseline (AWAPS)
is <0, this means that pain decreased at the end of the study. When the
decrease
of WAPS is > 1 (thus when AWAPS is lower than -1, e.g., AWAPS-1.5, -2.0, -
5.3 to give some numerical examples) this indicates not only a significant
pain
decrease at the end of the study but also a significant contribution of the
placebo
effect to the response of the patient to the pain treatment. Thus in the
Examples 2
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
29
and 3, when AWAPS values are < -1, this indicates that there is a placebo
response.
In the clinical study of Example 1 [41 randomized patients], 24 patients had a
AWAPS >0 [pain have increased after the treatment] and 17 patients had a
AWAPS <0 [pain have decreased after the treatment]. Among the latter 17
patients, 11 patients had a decrease of WAPS > 1 indicating that they were in
fact
placebo responders.
Description of the questions and tests performed to collect the input
data/variables
To each of the 41 patient, 212 queries (expressing several trait variables and
pain
symptoms) have been asked before placebo treatment and 33 queries were
repeated during the study. These queries have been selected among the clusters
of validated questionnaires known in the art. The answers of the patients to
each
of the 245 questions have been scored on a scale from 0 to 5 (scaled
responses)
or 0 to 10 (4 questions).
The following table lists the main category of queries that have been asked to
the
patients.
Table 1.1: Types of questionnaires and questions selected from clusters of
questions used to collect the "input variables/data"
Type of Clusters of questions to which Number of
Questionnaire the questionnaires are related such
questions
used in the
study
Perception of health Cluster of questions for evaluating 18
related issues the attitudinal and emotional
response of the patient to e.g.:
- external factors such as the
level of control of powerful
others on his attitude to resist,
fight and overcome pain;
- internal factors such as the level
of control that the individual
believes to have on his life
Big Five Components Cluster of questions for 85
characterizing the patient's
personality traits or characteristics
which are stable over time and
attributable to the person itself and
not to the effect of its environment
e.g.,
- Extraversion,
- Agreeableness,
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
- Conscientiousness,
- Openness,
- neuroticism
Suggestibility Cluster of questions related to the 12
impact of the
surrounding/environment on
health-related and/or psychological
issues e.g.:
- Sensation of contagion,
- Factors likely to influence the
balance between deliberate and
automatic processing of
information on pain
Influence of the Clinical Cluster of questions related to the 20
settings impact of the surrounding e.g.:
- Anxiety,
- Fear
- Discouragement, hopelessness
Belief in a Just World Cluster of questions for evaluating 6
the impact of the patient's
environment on his belief of a just
world, psychological well-being,
psychological quality of life, life
satisfaction, resistance to stress
etc.
Expectation Cluster of questions for measuring 4
the patient's expectations and
desire with respect to an external
stimulus, positive and negative
outcomes of an
intervention/treatment.
Pain Compliance Perceived doctor¨patient 26
Questionnaire (PCQ) relationship,
Positive beliefs on analgesics,
Partner agreement
Severity of health Intensity, interference, medication 37 +33
symptoms
Caregiver Assessment Evaluation of the patient's 2
of symptoms condition
Patient Assessment of Evaluation of the patient's 2
health condition
Total number of questions used during the whole study 245
period
Example of queries asked relating to an individual's expectations, evaluate an
individual's attitudinal and emotional response:
5 - How much do you expect this treatment will change your current pain?
How strong is your desire for pain relief?
Examples of queries that have been asked relating to an individual's
personality
traits and the impact of its surroundings
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
31
- I think I'm tending to find fault with others.
- I think I have a forgiving nature.
Examples of queries that have been asked relating to an individual's
personality
traits ("extraversion")
- I think I have an assertive personality
- I think I'm outgoing, sociable
Examples of queries that have been asked relating to "the evaluation of the
attitudinal and emotional response of an individual to external stimuli"
- I control my health.
- Doctors control my health.
Examples of queries that have been asked relating to "the impact of an
individual's
environment on health-related and/or psychological issues"
- I feel that people treat me fairly in life.
- I feel that my effort are noticed and rewarded.
Examples of queries that have been asked relating to the level of health
symptoms
("self-assessed health")
- "If you take into consideration all the various ways the pain influence
you
and your life how do you then evaluate your condition over the last week?".
In total, 245 response scores (on a scale of 0 to 5 or 0 to 10, e.g. Likert
Score)
where collected for each patients examined (21 receiving a conditioning
placebo
and 22 receiving neutral information). For each patient, the time needed to
collect
the answers to all the queries has been estimated to be approximately 3 hours.
As such, the biophysical scores and the answers to the queries are not able to
provide a single scoring value of the placebo response. The collected data are
able
to provide a caregiver with a general description of a patient, but not more.
There are no indications as such that can predict the propensity of these
patients
to have a placebo response, thus to predict a placebo response, in particular
to
predict the estimated AWAPS as measured a posteriori in the Clinical study A.
CA 02946808 2016-10-24
WO 2015/169810
PCT/EP2015/059875
32
Example 2.Comparison between the prediction of the placebo response of the
patients in Example 1 (the Scoring Factor) and their real placebo
response measured a posteriori
Example 2.1 - Use of a linear regression algorithm (LRA) for generating a
Scoring Factor by using the input variables collected in Example 1
Example 2.1 shows the ability of a linear regression algorithm such as LRA-1
(see
below) to use the data [demographic data, answers to the 212 queries at
baseline
and the data from the biophysical test of Example 1] collected among 30
patients
(out of the 41 patients included in the Clinical Study A of Example 1) in
order to
predict a placebo response [the Scoring Factor] for each of the 30 patients.
LRA-1 used:
f(x)= -(-6.309 + 0.030*x1 + 0.268*x2 + 1.308*x3 - 0.058*x4 + 0.031*x5 -
0.220*x6 + 0.297*x7
where:
f(x) 9,
0 9 being the Scoring Factor
o y is the "real" placebo response based on the variation of the WAPS
score [AWAPS]
o f(x) is the model, a function of x, and
x are the input variables, x = {x1=[Age] , x2=[Expectation] ,
x3=[Agreeableness], x4=[Extraversion], x5=[Internal factor of perception of
health-related issues], x6=[Beliefs in a Just World], x7=[Self-assessed
health]}
The LRA-1 has been used for processing the input data of 30 patients of the
clinical study A and has predicted the placebo response in the form of
continuous
output. The corresponding Scoring Factors [named "9" in the LRA-1 of the
example] have been compared to the a posteriori "real" placebo response ["y"]
based on the variation of the WAPS score [AWAPS]. The comparison between the
Scoring Factor and the a posteriori placebo response is given in Table 2.1
The Scoring Factor in Example 2.1 is a continuous value.
Table 2.1: Comparison between the predicted placebo response [the Scoring
Factor] and the real placebo response measured a posteriori
Patient [Scoring] Comment Y [AWAPS Comment
1019 -3.30 Placebo Responder -4.43
Placebo Responder
CA 02946808 2016-10-24
WO 2015/169810
PCT/EP2015/059875
33
1032 -3.19 Placebo Responder -5.43
Placebo Responder
I011 -3.12 Placebo Responder -1.86
Placebo Responder
1029 -2.84 Placebo Responder -3.00
Placebo Responder
1049 -2.67 Placebo Responder -0.86
1044 -2.21 Placebo Responder -2.57
Placebo Responder
1047 -1.66 Placebo Responder -1.29
Placebo Responder
1038 -1.35 Placebo Responder -2.71
Placebo Responder
1050 -1.33 Placebo Responder -2.43
Placebo Responder
1036 -1.23 Placebo Responder -1.86
Placebo Responder
1005 -1.08 Placebo Responder -0.57
1051 -1.07 Placebo Responder 0.43
1025 -1.00 -0.71
1016 -0.97 +0.29
1006 -0.93 -1.43
Placebo Responder
1042 -0.70 0.43
1014 -0.54 +0.29
I001 -0.53 -0.43
1023 -0.52 -0.29
1028 -0.48 0.57
1033 -0.44 0.00
1052 -0.23 -2.29
Placebo Responder
1031 -0.15 0.00
1034 -0.06 0.14
1027 -0.05 0.00
1013 0.01 0.86
1012 0.03 0.14
1026 0.29 0.00
1039 0.61 -0.14
1022 2.05 1.14
Based on the statistical analysis of the results in Table 2.1, the accuracy of
the
predictive value of the Scoring Factor is 0.775, measured by Pearson
correlation
between the Scoring Factor and the a posteriori placebo response.
In Table 2.1, when the Scoring Factor is lower than -1 which is a predefined
cut-
off value, this indicates the presence of a placebo response, or a high
propensity
of developing the latter.
Example 2.2 - Use of a linear classification algorithm (LCA) for generating a
binary Scoring Factor by using the input variables collected in
Example 1
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
34
Example 2.2 shows the ability of a linear classification algorithm such as LCA-
1
(see below) to use the data [demographic data, answers to the 212 queries and
the data from the biophysical test of Example 1] collected among the subset of
30
patients from the Clinical Study A of Example 1
LCA-1 used:
f(x)= sign(-2.026 + 0.011*x1 + 0.004*x2 + 0.501*x3 - 0.128*x4 + 0.022*x5 -
0.067*x6 + 0.214*x7) ¨> 9
where:
V f(x) ¨> 9,
O 9 being the binary Scoring Factor
o y is the "real" binarized placebo response, measuring whether the
decrease of WAPS score is greater than 1. When the decrease of
WAPS is > 1 (thus when AWAPS is lower than -1) this indicates not
only a significant pain decrease at the end of the study but also a
significant contribution of the placebo effect to the response of the
patient to the pain treatment.
o x are the input variables, x = {x1, x2, ... , xn}, with x = {x1=[Age]
, x2= [Duration of symptoms] ,
x3= [Agreeableness],
x4=[Extraversion], x5=[Internal factor of perception of health-
related issues], x6= [Beliefs in a
Just World], x7=[
Discouragement]}, and
o f(x) is the model, a function of x
The LCA-1 has been used for processing the input data of 30 patients of the
clinical study A. The corresponding binary Scoring Factors [named "9" in the
LCA-1
of the example] have been compared to the a posteriori "real" binary placebo
response ["y"] based on the variation of the WAPS score [AWAPS]. The
comparison between the binary Scoring Factor and the a posteriori binary
placebo
response is given in Table 2.2
The Scoring Factor in Example 2.2 is a categorical value.
Table 2.2: Comparison between the predicted placebo response [the binary
Scoring Factor] and the real placebo response measured a posteriori.
In column 4, when the AWAPS was <-1 then "real" binarized score [Y] was set as
TRUE (Placebo responder), When AWAPS was >-1 then the binarized score was
set as FALSE (Placebo non-responder).
Column 4
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
35
Patient [Scoring Factor] Comment Y Comment
[Real
binarized
score]
1042 sign(-0.689) = -1 FALSE
1022 sign(-0.681) = -1 FALSE
1023 sign(-0.633) = -1 FALSE
1033 sign(-0.550) = -1 FALSE
1026 sign(-0.502) = -1 FALSE
1039 sign(-0.466) = -1 FALSE
1031 sign(-0.464) = -1 FALSE
1025 sign(-0.446) = -1 FALSE
1027 sign(-0.339) = -1 FALSE
1001 sign(-0.336) = -1 FALSE
1013 sign(-0.275) = -1 FALSE
1012 sign(-0.159) = -1 FALSE
1016 sign(-0.087) = -1 FALSE
1034 sign(-0.080) = -1 FALSE
1036 sign(-0.073) = -1 TRUE Placebo responder
1005 sign(-0.067) = -1 FALSE
1014 sign(-0.023) = -1 FALSE
1051 sign(-0.022) = -1 FALSE
1006 sign(0.082) = 1 Placebo TRUE Placebo responder
responder
1028 sign(0.084) = 1 Placebo FALSE
responder
1052 sign(0.102) = 1 Placebo TRUE Placebo responder
responder
1011 sign(0.237) = 1 Placebo TRUE Placebo responder
responder
1049 sign(0.285) = 1 Placebo FALSE
responder
1050 sign(0.313) = 1 Placebo TRUE Placebo responder
responder
1029 sign(0.341) = 1 Placebo TRUE Placebo responder
responder
1044 sign(0.377) = 1 Placebo TRUE Placebo responder
responder
1019 sign(0.380) = 1 Placebo TRUE Placebo responder
responder
1032 sign(0.396) = 1 Placebo TRUE Placebo responder
responder
1038 sign(0.427) = 1 Placebo TRUE Placebo responder
responder
1047 sign(0.547) = 1 Placebo TRUE Placebo responder
responder
Based on the statistical analysis of the results in Table 2.2, the accuracy of
the
predictive value of the binary Scoring Factor is 0.90.
Example 2.3 - Use of an instance-based non-linear classification algorithm
for generating a binary Scoring Factor by using the input variables
collected in Example 1
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
36
Example 2.3 shows the ability of a non-linear classification algorithm such as
the
1-nearest-neighbor model presented in NCA-1 (see below) to use the data
[demographic data, answers to the 212 queries and the data from the
biophysical
test of Example 1] collected among the 30 patients included in the Clinical
Study A
of Example 1. Other non-linear models including but not limited to decision
trees
or artificial neural networks have shown similar results.
NCA-1 used:
f(x) is computed as follows:
o The distance between a new patient x and each of the 30 reference
patients is computed
o The closest reference patient is chosen
o His/her class (responder/non responder, observed a posteriori) is
returned as prediction for the class of x
where:
V f(x) ¨> 9,
o 9 being the binary Scoring Factor
o y is the "real" binarized placebo response, measuring whether the
decrease of WAPS score is greater than 1 [AWAPS<-1]
o x are the input variables, x = {x1=[Age] , x2=[Duration of
symptoms] , x3=[Agreeableness], x4=[Extraversion], x5=[Internal
factor of perception of health-related issues], x6=[Beliefs in a Just
World], x7=[Discouragement]},
0 f(x) is the model, a function of x, and
o distances between patients are measured by the Euclidean distance
between their standardized input variables
The NCA-1 has been used for processing the input data of 30 patients of the
clinical study A. The corresponding binary Scoring Factors [named "9" in the
NCA-
1 of the example] have been compared to the a posteriori "real" binary placebo
response ["y"] based on the variation of the WAPS score [AWAPS]. The
comparison between the binary Scoring Factor and the a posteriori binary
placebo
response is given in Table 2.3
The Scoring Factor in Example 2.3 is a binary value.
Table 2.3: Comparison between the predicted placebo response [the binary
Scoring Factor] and the real placebo response measured a posteriori.
In column 4, when the AWAPS was <-1 then "real" binarized score [Y]
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
37
was set as TRUE (Placebo responder). When the AWAPS was >-1 then
the binarized score was set as FALSE (Placebo non-responder). [NN
stands for nearest neighbor].
Column
4
Patient Y [Scoring Factor] Comment Y [Real Comment
binarized
score]
I001 FALSE (NN is FALSE
1027)
1005 FALSE (NN is FALSE
1013)
1006 TRUE (NN is 1050) TRUE Placebo
responder
I011 FALSE (NN is TRUE Placebo
1049) responder
1012 FALSE (NN is FALSE
1014)
1013 FALSE (NN is FALSE
1005)
1014 FALSE (NN is FALSE
1012)
1016 FALSE (NN is FALSE
1013)
1019 TRUE (NN is 1044) Placebo responder TRUE Placebo
responder
1022 FALSE (NN is FALSE
1034)
1023 FALSE (NN is FALSE
1042)
1025 FALSE (NN is FALSE
1031)
1026 FALSE (NN is FALSE
I001)
1027 FALSE (NN is FALSE
1039)
1028 TRUE (NN is 1036) Placebo responder FALSE
1029 TRUE (NN is 1038) Placebo responder TRUE Placebo
responder
1031 FALSE (NN is FALSE
1025)
1032 FALSE (NN is TRUE Placebo
1028) responder
1033 FALSE (NN is FALSE
1031)
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
38
1034 TRUE (NN is 1052) Placebo responder FALSE
1036 FALSE (NN is TRUE Placebo
1028) responder
1038 TRUE (NN is 1052) Placebo responder TRUE Placebo
responder
1039 FALSE (NN is FALSE
1027)
1042 FALSE (NN is FALSE
1023)
1044 TRUE (NN is 1047) Placebo responder TRUE Placebo
responder
1047 TRUE (NN is 1044) Placebo responder TRUE Placebo
responder
1049 FALSE (NN is FALSE
1028)
1050 TRUE (NN is 1006) Placebo responder TRUE Placebo
responder
1051 FALSE (NN is FALSE
1012)
1052 TRUE (NN is 1038) Placebo responder TRUE Placebo
responder
Based on the statistical analysis of the results in Table 2.3, the accuracy of
the
predictive value of the binary Scoring Factor is 0.83.
Example 2.4 - Use of a rule-based non-linear classification algorithm for
generating a binary Scoring Factor by using the input variables
collected in Example 1.
Example 2.4 shows the ability of a non-linear classification algorithm such as
the
1-nearest-neighbor model presented in NCA-2 (see below) to use the data
[demographic data, answers to the 212 queries and the data from the
biophysical
test of Example 1] collected among the 30 patients included in the Clinical
Study A
of Example 1. Other non-linear models including but not limited to decision
trees
or artificial neural networks have shown similar results.
NCA-2 used:
f(x) is computed as follows (as Presented in figure 2.1):
o The feature at the root (top) of the tree is tested.
o The test indicates in which branch the patient falls.
o The next node indicates which test is to be performed next.
0 The reasoning is pursued up to a point where the patient reaches a
leaf node (bottom).
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
39
O Each leaf node corresponds to a particular category (placebo
responder or not)
where:
V f(x) ¨> V,
0 9 being the binary Scoring Factor
o y is the "real" binarized placebo response, measuring whether the
decrease of WAPS score is greater than 1 [AWAPS<-1]
o x are the input variables, x = {x1=[Beliefs in a Just World] , x2=[
discouragement] , x3= [Age] , x4=[Extraversion]l
0 f(x) is the model, a function of x
To make a prediction for a particular patient x, the feature at the root (top)
of the
tree is tested. The test indicates in which branch the patient falls. The next
node
indicates which test is to be performed next. The reasoning is pursued up to a
point where the patient reaches a leaf node (bottom). Each leaf node
corresponds
to a particular category (placebo responder or not).
In a first case, the NCA-2 has been used for processing the input data of 30
patients of the clinical study A. The corresponding binary Scoring Factors
[named
"V" in the NCA-2 of the example] have been compared to the a posteriori "real"
binary placebo response ["y"] based on the variation of the WAPS score
[AWAPS].
The comparison between the binary Scoring Factor and the a posteriori binary
placebo response is given in Table 2.4
The Scoring Factor in Example 2.4 is a binary value.
Table 2.4: Comparison between the predicted placebo response [the binary
Scoring Factor] and the real placebo response measured a posteriori.
In column 4, when the AWAPS was <-1 then the "real" binarized
score [Y] was set as TRUE (Placebo responder). When the AWAPS was
>-1 then the binarized score was set as FALSE (Placebo non-
responder). [NN stands for nearest neighbor].
Column 4
Patient it [Scoring Comment Y [Real Comment
Factor] binarized
score]
1001 FALSE FALSE
1005 FALSE FALSE
1006 FALSE TRUE Placebo responder
1011 TRUE Placebo responder TRUE Placebo responder
1012 FALSE FALSE
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
1013 FALSE FALSE
1014 TRUE Placebo responder FALSE
1016 FALSE FALSE
1019 TRUE Placebo responder TRUE Placebo responder
1022 FALSE FALSE
1023 FALSE FALSE
1025 FALSE FALSE
1026 FALSE FALSE
1027 FALSE FALSE
1028 FALSE FALSE
1029 TRUE Placebo responder TRUE Placebo responder
1031 FALSE FALSE
1032 FALSE TRUE Placebo responder
1033 FALSE FALSE
1034 FALSE FALSE
1036 TRUE Placebo responder TRUE Placebo responder
1038 TRUE Placebo responder TRUE Placebo responder
1039 FALSE FALSE
1042 FALSE FALSE
1044 TRUE Placebo responder TRUE Placebo responder
1047 TRUE Placebo responder TRUE Placebo responder
1049 FALSE FALSE
1050 TRUE Placebo responder TRUE Placebo responder
1051 FALSE FALSE
1052 TRUE Placebo responder TRUE Placebo responder
Based on the statistical analysis of the results in Table 2.4, the accuracy of
the
predictive value of the binary Scoring Factor is 0.9.
Above examples show that it is possible to determine the placebo response on
the
5 basis of input variables relating to an individual by mathematical
modelling.
10 Example 3. Reduction of the number of questions needed to obtain the
same
placebo scores as in Example 2.
Surprisingly, the inventors of the current invention learned that the number
of
questions asked to a patient or individual can be reduced whilst still
maintaining a
very accurate prediction of the placebo response. This enables a fast
execution of
15 the test, even multiple times a day/week thereby reducing any negative
side-
effects for the patient or individual whilst taking the test.
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
41
In a first case, all 41 patients completed the 212 queries performed at the
baseline. Using feature selection techniques', the total number of queries
related
to trait personality was decreased from 167 to 117, without decreasing the
number of personality traits measured. The impact of queries reduction on the
measure of each personality trait was minimal (average R-squared >0.5 and p-
value of signature stability < 0.10).
In a second case, it was possible to reduce the number of personality traits
associated to the prediction of the placebo response. As a result, a reduced
subset
of only 99 questions related to personality traits and less than 60 related to
health
were found sufficient to predict the placebo response in future patients with
the
same level of confidence as those obtained in Example 2.
Example 3.1 - Use of a linear regression algorithm for generating a Scoring
Factor by using the reduced set of input variables
This example shows the ability of a linear regression algorithm such as LRA-1
(see
Example 2.1) to generate accurate Scoring Factors based on the reduced set of
input variables [demographic data, answers to the 99 questions, answers to
less
than 60 questions related to health and the data from the biophysical test of
Example 1] collected among the 30 patients included in the Clinical Study A of
Example 1.
The predictive model LRA-1 has been used for generating Scoring Factors based
on the input data of 30 patients of the clinical study A, with the reduced set
of
input variables introduced above. The corresponding Scoring Factors [named
"Sr]
have been compared to the a posteriori "real" placebo response ["y"] based on
the
variation of the WAPS score [AWAPS]. The comparison is given in Table 3.1
The Scoring Factor in this example is a continuous value.
Table 3.1: Comparison between the predicted placebo response [the Scoring
Factor] obtained on a shortlist of variables (column 2), the Scoring
Factor as obtained in Example 2.1 (column 4) and the real placebo
response measured a posteriori (column 5)
Column 2 Column 4 Column 5
Patient [Scoring Comment '1' Y Comment
Factor] [Scoring [AWAPS
"reduced" Factor] ]
"Ex. 2.1"
1 See Guyon, I., & Elisseeff, A. (2003). An introduction to variable
and feature
selection. The Journal of Machine Learning Research, 3, 1157-1182.
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
42
1032 -3.59 Placebo responder -3.19 -5.43
Placebo responder
1019 -3.42 Placebo responder -3.30 -4.43
Placebo responder
1011 -3.14 Placebo responder -3.12 -1.86
Placebo responder
1049 -2.80 Placebo responder -2.67 -0.86
1029 -2.73 Placebo responder -2.84 -3.00
Placebo responder
1044 -1.80 Placebo responder -2.21 -2.57
Placebo responder
1047 -1.66 Placebo responder -1.66 -1.29
Placebo responder
1006 -1.44 Placebo responder -0.93 -1.43
Placebo responder
1050 -1.33 Placebo responder -1.33 -2.43
Placebo responder
1025 -1.25 Placebo responder -1.00 -0.71
1036 -1.24 Placebo responder -1.23 -1.86
Placebo responder
1038 -1.23 Placebo responder -1.35 -2.71
Placebo responder
1005 -0.93 -1.08 -0.57
1016 -0.84 -0.97 0.29
1042 -0.83 -0.70 0.43
1028 -0.72 -0.48 0.57
1051 -0.42 -1.07 0.43
1033 -0.41 -0.44 0.00
I001 -0.38 -0.53 -0.43
1052 -0.37 -0.23 -2.29
Placebo responder
1023 -0.28 -0.52 -0.29
1013 -0.10 0.01 0.86
1014 -0.04 -0.54 -0.29
1034 0.08 -0.06 0.14
1031 0.12 -0.15 0.00
1012 0.15 0.03 0.14
1027 0.19 -0.05 0.00
1026 0.29 0.29 0.00
1039 0.49 0.61 -0.14
1022 2.58 2.58 1.14
Based on the statistical analysis of the results in Table 2.1, the accuracy of
the
predictive value of the Scoring Factor is 0.787, measured by Pearson
correlation
between the Scoring Factor and the a posteriori placebo response.
Example 3.2 - Use of a linear classification algorithm for generating a binary
Scoring Factor by using the reduced set of input variables
Example 3.2 shows the ability of a linear classification algorithm such as LCA-
1
(see Example 2.2) to generate accurate binary Scoring Factors based on the
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
43
reduced set of input variables [demographic data, answers to the 99 questions
and less than 60 health-related questions and the data from the biophysical
test of
Example 1] collected among the 30 patients included in the Clinical Study A of
Example 1.
The predictive model LCA-1 has been used for generating binary Scoring Factors
based on the input data of 30 patients of the clinical study A, with the
reduced set
of input variables introduced above. The corresponding binary Scoring Factors
[named "9"] have been compared to the a posteriori "real" placebo response
["y"]
based on the variation of the WAPS score [AWAPS]. The comparison is given in
Table 3.2
The Scoring Factor in this example is a binary value.
Table 3.2: Comparison between the predicted placebo response [the binary
Scoring Factor] and the real placebo response measured a posteriori.
In column 4, when the AWAPS was <-1 then the "real" binarized score
[Y] was set as TRUE (Placebo responder). When the AWAPS was >-1
then the binarized score was set as FALSE (Placebo non-responder).
Column 4
Patient [Scoring Factor]Comment Y Comment
"reduced" [Real
binarized
score]
1022 sign(-0.910) = -1 FALSE
1023 sign(-0.694) = -1 FALSE
1042 sign(-0.642) = -1 FALSE
1033 sign(-0.624) = -1 FALSE
1031 sign(-0.590) = -1 FALSE
1026 sign(-0.512) = -1 FALSE
1039 sign(-0.438) = -1 FALSE
1001 sign(-0.425) = -1 FALSE
1027 sign(-0.382) = -1 FALSE
1025 sign(-0.375) = -1 FALSE
1051 sign(-0.282) = -1 FALSE
1013 sign(-0.270) = -1 FALSE
1012 sign(-0.187) = -1 FALSE
1014 sign(-0.169) = -1 FALSE
1005 sign(-0.159) = -1 FALSE
1034 sign(-0.156) = -1 FALSE
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
44
1016 sign(-0.134) = -1 FALSE
1036 sign(-0.051) = -1 TRUE Placebo
responder
1028 sign(0.123) = 1 Placebo FALSE
responder
1052 sign(0.168) = 1 Placebo TRUE Placebo
responder responder
1044 sign(0.194) = 1 Placebo TRUE Placebo
responder responder
1006 sign(0.260) = 1 Placebo TRUE Placebo
responder responder
1011 sign(0.282) = 1 Placebo TRUE Placebo
responder responder
1050 sign(0.313) = 1 Placebo TRUE Placebo
responder responder
1049 sign(0.326) = 1 Placebo FALSE
responder
1029 sign(0.336) = 1 Placebo TRUE Placebo
responder responder
1038 sign(0.393) = 1 Placebo TRUE Placebo
responder responder
1019 sign(0.414) = 1 Placebo TRUE Placebo
responder responder
1047 sign(0.538) = 1 Placebo TRUE Placebo
responder responder
1032 sign(0.552) = 1 Placebo TRUE Placebo
responder responder
Based on the statistical analysis of the results in Table 2.2, the accuracy of
the
predictive value of the binary Scoring Factor is 0.90.
Example 3.3 - Use of an instance-based non-linear classification algorithm
for generating a binary Scoring Factor by using the reduced set of
input variables
Example 3.3 shows the ability of a non-linear classification algorithm such as
NCA-
1 (see Example 2.3) to generate accurate binary Scoring Factors based on the
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
reduced set of input variables [demographic data, answers to the 99 questions
related to personality, answers to less than 60 questions related to health
and the
data from the biophysical test of Example 1] collected among the 30 patients
included in the Clinical Study A of Example 1.
5 The predictive model NCA-1 has been used for generating binary Scoring
Factors
based on the input data of 30 patients of the clinical study A, with the
reduced set
of input variables introduced above. The corresponding binary Scoring Factors
[named "9"] have been compared to the a posteriori "real" placebo response
["y"]
based on the variation of the WAPS score [AWAPS]. The comparison is given in
10 Table 3.3.
The Scoring Factor in this example is a binary value.
Table 3.3:Comparison between the predicted placebo response [the binary
Scoring
15 Factor] and the real placebo response measured a posteriori. In
column
4, when the AWAPS was <-1 then the "real" binarized score [Y] was
set as TRUE (Placebo responder). When the AWAPS was >-1 then the
binarized score was set as FALSE (Placebo non-responder).
[NN stands for nearest neighbor].
Column
4
Patient "( [Scoring Factor] Comment Y [Real Comment
binarized
score]
1001 FALSE (NN is FALSE
1039)
1005 FALSE (NN is FALSE
1013)
1006 TRUE (NN is 1050) TRUE Placebo
responder
1011 FALSE (NN is TRUE Placebo
1025) responder
1012 FALSE (NN is FALSE
1014)
1013 FALSE (NN is FALSE
1005)
1014 FALSE (NN is FALSE
1012)
1016 FALSE (NN is FALSE
1013)
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
46
1019 TRUE (NN is 1044) Placebo responder TRUE Placebo
responder
1022 FALSE (NN is FALSE
1034)
1023 FALSE (NN is FALSE
1042)
1025 FALSE (NN is FALSE
1031)
1026 FALSE (NN is FALSE
1001)
1027 FALSE (NN is FALSE
1039)
1028 TRUE (NN is 1036) Placebo responder FALSE
1029 TRUE (NN is 1036) Placebo responder TRUE Placebo
responder
1031 FALSE (NN is FALSE
1025)
1032 FALSE (NN is TRUE Placebo
1028) responder
1033 FALSE (NN is FALSE
1025)
1034 TRUE (NN is 1038) Placebo responder FALSE
1036 FALSE (NN is TRUE Placebo
1028) responder
1038 TRUE (NN is 1052) Placebo responder TRUE Placebo
responder
1039 FALSE (NN is FALSE
1027)
1042 FALSE (NN is FALSE
1023)
1044 FALSE (NN is Placebo responder TRUE Placebo
1039) responder
1047 TRUE (NN is 1019) Placebo responder TRUE Placebo
responder
1049 FALSE (NN is FALSE
1028)
1050 TRUE (NN is 1006) Placebo responder TRUE Placebo
responder
1051 FALSE (NN is FALSE
1012)
1052 TRUE (NN is 1038) Placebo responder TRUE Placebo
responder
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
47
Based on the statistical analysis of the results in Table 3.3, the accuracy of
the
predictive value of the binary Scoring Factor is 0.80.
Although the illustrative embodiments of the present invention have been
described in greater detail, it will be understood that the invention is not
limited to
those embodiments. Various changes or modifications may be effected by one
skilled in the art without departing from the scope or the spirit of the
invention as
defined in the claims.
Example 3.4 - Use of a rule-based non-linear classification algorithm for
generating a binary Scoring Factor by using the reduced set of
input variables
Example 3.4 shows the ability of a non-linear classification algorithm such as
NCA-
2 (see Example 2.4) to generate accurate binary Scoring Factors based on the
reduced set of input variables [demographic data, answers to the 99 questions
related to personality, answers to less than 60 questions related to health
and the
data from the biophysical test of Example 1] collected among the 30 patients
included in the Clinical Study A of Example 1.
The predictive model NCA-2 has been used for generating binary Scoring Factors
based on the input data of 30 patients of the clinical study A, with the
reduced set
of input variables introduced above. The corresponding binary Scoring Factors
[named "9"] have been compared to the a posteriori "real" placebo response
["y"]
based on the variation of the WAPS score [AWAPS]. The comparison is given in
Table 3.4.
The Scoring Factor in this example is a binary value.
Table 3.4:Comparison between the predicted placebo response [the binary
Scoring
Factor] and the real placebo response measured a posteriori. In column
4, when the AWAPS was <-1 then the "real" binarized score [Y] was
set as TRUE (Placebo responder). When AWAPS was >-1 then the
binarized score was set as FALSE (Placebo non-responder).
Column 4
Patient it [Scoring Comment Y [Real Comment
Factor] binarized
score]
1001 FALSE FALSE
CA 02946808 2016-10-24
WO 2015/169810 PCT/EP2015/059875
48
1005 FALSE FALSE
1006 FALSE TRUE
1011 TRUE Placebo responder TRUE Placebo responder
1012 FALSE FALSE
1013 FALSE FALSE
1014 TRUE Placebo responder FALSE
1016 FALSE FALSE
1019 TRUE Placebo responder TRUE Placebo responder
1022 FALSE FALSE
1023 FALSE FALSE
1025 FALSE FALSE
1026 FALSE FALSE
1027 FALSE FALSE
1028 FALSE FALSE
1029 TRUE Placebo responder TRUE Placebo responder
1031 FALSE FALSE
1032 FALSE TRUE Placebo responder
1033 FALSE FALSE
1034 FALSE FALSE
1036 TRUE Placebo responder TRUE Placebo responder
1038 TRUE Placebo responder TRUE Placebo responder
1039 FALSE FALSE
1042 FALSE FALSE
1044 TRUE Placebo responder TRUE Placebo responder
1047 TRUE Placebo responder TRUE Placebo responder
1049 FALSE FALSE
1050 TRUE Placebo responder TRUE Placebo responder
1051 FALSE FALSE
1052 TRUE Placebo responder TRUE Placebo responder
Based on the statistical analysis of the results in Table 3.4, the accuracy of
the
predictive value of the binary Scoring Factor is 0.90.
Although the illustrative embodiments of the present invention have been
described in greater detail, it will be understood that the invention is not
limited to
those embodiments. Various changes or modifications may be effected by one
skilled in the art without departing from the scope or the spirit of the
invention as
defined in the claims.
