Statistical Power and Sample Size Size in Personalized Medicine

72 International Journal of Systems Biology and Biomedical Technologies, 2(2), 72-88, April-June 2013
Statistical Power and Sample

Size in Personalized Medicine
Alexander Rompas, Biomedical Engineering Laboratory, National Technical University of
Athens, Athens, Greece
Charalampos Tsirmpas, Biomedical Engineering Laboratory, National Technical University
of Athens, Athens, Greece
Athanasios Anastasiou, Biomedical Engineering Laboratory, National Technical University of
Dimitra Iliopoulou, Biomedical Engineering Laboratory, National Technical University of
Dimitris Koutsouris, Biomedical Engineering Laboratory, National Technical University of
ABSTRACT
Personalized medicine (PM) is a rapidly growing field of healthcare and medicine. The advantage of a
personalized medicine is the availability of each person’s unique genetic and genomic print. The healthcare
that incorporates personalized medicine provides coordinated, continuous patient-specific data. The goal of
personalized medicine is to promote health wellness, satisfaction, and to increase the likelihood of a successful
disease prevention, detection and treatment. This form of medicine, apart from patient’s personal data and
medicine-biological measurements, uses genomic information data to understand the molecular structure of
the disease and to optimize health care strategies and drug therapies. Clinical trials that investigate person-
alized approaches are subject to special rules, for example, pertain the selection of participating patients.
In personalized medicine, a certain genetic profile must be identified so that the treatment can work. This is
why potential participants are first screened and selected accordingly for clinical trials. The special design
of such clinical trials has an impact on the evaluation of data collected during the given study.
Keywords: Cancer, Clinical Trials, Personalized Medicine (PM), Pharmacogenetics, Phase III, Targeted
Therapy
INTRODUCTION molecular predictors of drug safety and efficacy

up front and to stratify patients and diseases into
Traditionally drug discovery and development subsets based on their unique characteristics.
attempts to find medicines that show benefit This approach allows for an optimal response
across an entire disease population. Personal- to therapy by enabling physicians to prescribe
ized Medicine however, attempts to define therapies more likely to benefit patients and
minimizes patient exposure to drugs that are
DOI: 10.4018/ijsbbt.2013040105
Copyright © 2013, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
International Journal of Systems Biology and Biomedical Technologies, 2(2), 72-88, April-June 2013 73
not likely to provide a benefit. The aim of breast cancer drug Herceptin, for instance, is
personalized medicine is to treat select patient effective only to women whose tumors have an
groups with medicines that match the genetic abundance of a protein called Her2 (Lazakidou
profile of their cells. For example, special agents & Daskalaki, 2012).
have been developed for lung cancer patients
whose tumor cells exhibit specific genetic
profiles. They disable certain gene mutations CANCER MANAGEMENT
or gene rearrangements, thereby inhibiting the
Oncology is a primary section of medicine
proliferation of cancer cells. Molecular biologi-
consisting of various cancer phases, measured
cal profiling is part of the diagnostic process.
in centimeters in most cases, and types with
Performed before start of a treatment regimen,
regard to their anatomy and pathology. Cancer
this type of screening determines if a medicine
genetics is a subgroup falling under the category
will achieve the desired effect in a given patient.
of oncology that is focusing on genes and is
There are various synonyms for personalized
associated with inherited cancer risk (Mansour
medicine such as tailored, stratified, precision
& Schwarz, 2008). There is a limited number
and individualized medicine (Veer & Bernards,
of cancerous disorders in which homogeneity
2008; Willard, 2012). They often mean the same
separates in an autosomal dominant fashion,
thing – the targeted application of drugs tailored
leading to considerably higher risk for certain
to the genetic traits of certain diseases. In this
cancers types. It is considered that inherited
context, personalized medicine does not mean
cancer genetics factors explain only about
neither the doctor-patient relationship, nor the
5-10% of all cancers cases. Nevertheless, other
personal attention and care that are so important,
genetic modifications with more indirect effects
too. Personalized medicine has the potential for
associated to cancer risk may trigger detailed
great benefit because it enables targeted treat-
cancer risk valuation to patients who are not
ments with fewer side effects. These therapies
associated with a family history (Yan, 2008).
are tailored to the needs of certain patient groups
Examples of personalized cancer manage-
or populations and therefore not all patients can
ment:
benefit, so it does give rise to several ethical
questions (Haile, 2008; Shastry, 2006). These
concern access to medical services, the handling 1. Mutations in the BRCA1 and BRCA2 genes,
of personal patient data, and the compatibility associated with inherited breast–ovarian
of personalized research and treatments with cancer disorders. Findings of a disease
the principle of solidarity in healthcare. The leading to genes modifications to a con-
design of clinical studies in personalized trolled group of individuals could provide
medicine is also changing for ethical reasons. information as to whether they potentially
In many cases, patients receiving standard or hold a higher probability for cancer and may
placebo treatments within the framework of the prompt personalized prophylactic therapy
study will soon cross over to the personalized such as mastectomy as well as removal
and therefore far more effective agent (Sadée of the ovaries. Such testing incorporates
& Dai, 2005). complex personal decisions and is com-
Personalized medicine often involves de- menced in the context of in depth genetic
veloping genetic or other tests that can be used treatment.
to determine which patients are most likely to 2. Minimal residual disease (MRD) tests are
benefit from a drug or which are most likely to used to compute residual cancer, permitting
suffer serious side effects (Wong, 2008). detection of tumor markers prior to any
Relatively few drugs are now accompa- occurrence of physical sings or symptoms.
nied by such so-called companion diagnostic This permits doctors in taking clinical
tests. They are most common in oncology. The decisions earlier than before.
3. Targeted therapies is the use of remedies Genetic maps can assist doctors to dis-
designed to tackle eccentric molecular tinguish subgroups of patients with different
pathways among a subgroup of patients forms of cancer and other difficult to treat
within a given cancer type. To mention a diseases, assisting doctors to formulae a pre-
few, trastuzumab (marketed as Herceptin) dictive medicine choice. Furthermore, through
is used as a treatment for women suffering personalize medicine (PM) physicians are able
from breast cancer whose HER2 protein to optimally select between pools of treatment
is overexpressed. Xalkori, is used in case protocols and/or avoiding applying unneces-
of lung cancer patients and shows great sary medical treatments to the patient. Also,
promise in this area. Imatinib (a tyrosine ki- personalized medicine, when used properly,
nase inhibitor) targets patient with chronic points to potential early detection of patient’s
myeloid leukemia (CML), in which the specific metabolize variations in pharmaceuti-
BCR-ABL fusion gene is existent in >95% cals. Personalized medicine is aiming toward
of cases and produces hyperactivated abl- identifying the optimal dose for a patient;
driven protein signaling. avoid risks based on hereditary characteristics,
environmental effects, and genetic deviations
(Clayton et al., 2006; Galas & Hood, 2009).
USABILITY OF
UNDERSTANDING THE
GENETIC MAP CLINICAL TRIAL PROCESS
Developing a patient’s genetic map and ac- Clinical trials involves three steps:
counting for potential genetic variations could
potentially lead to identification of remedies and 1. Phase I: (PK {pharmacokinetics}, PD
treatment courses. Hence, analysis of genetic {pharmacodynamics}).
maps could eliminate side effects and generate 2. Phase II: (proof of concept (usually small
efficient outcome driven medical treatment sample for identifying safety/efficacy)
strategies. Helping physicians examine all 3. Phase III: (pivotal {validation regarding
the possible roots and pathways is essential. to efficacy/safety} trial
In addition, genetic mapping reveals patient’s
tendency of developing specific diseases prior A more detailed description is presented
to the existence of any recognizable symptoms in Table 1.
giving to both the physician and patient to come In our research we will examine the Phase
up with a plan for observation and prevention III of a clinical study. Phase III clinical trials
(Willard, 2012). are usually randomized (Hanna et al., 2004), in
The ability to identify genes sequencing the sense that patients receive either the inves-
and expressing mechanism form the basis tigational treatment or the standard treatment
for defining diseases classification and treat- in arbitrary way (Saglio et al., 2004). Usually
ments discovery, letting GPs to go beyond phase III clinical trials recruit patients of differ-
“a unified” model that could be ineffective ent ages, ethnicities, and femininities so that the
or lead to unwanted side effects. Through results may provide the desired diversification
further categorization, usage of personalized (Oldenburg, Watzka, Rost, & Muller, 2007;
medicine, GP’s and principal medical inves- Saglio et al., 2004). Lastly, the number of
tigators are developing subgroup populations people enrolled in a phase III clinical trial can
for complex diseases and physical conditions range depending on minimum statistical power
such as diabetes, Alzheimer’s, cancer and heart required according to the clinical protocol (or
disease (Acharya et al., 2008; Cichon, Nothen, SPA if in place).
Rietschel, & Propping, 2000).
Table 1. Clinical trial process
Phase I Phase II Phase III

Description A Phase I trial is the first Once a treatment is found A Phase III trial tests
test of a new treatment to be safe (often in a a treatment that has
to see if it is safe to Phase I trial), it can be been shown to help
use in people. The new tested to see if it helps some patients (often
treatment is tested because patients. in a Phase II trial). It
it showed promise in lab usually compares a newer
tests. treatment to the standard
or best known treatment.
Goals Learn: Learn: Learn:
Whether the treatment Whether the treatment Whether the new
is safe works treatment is better than, as
The best way to give the Whether there are any good as or worse than the
treatment (for example, as less common side effects, standard treatment
a pill or a shot) which may appear when Phase III trials may be
The right dose — the more patients get the more complex and look at
amount that causes the treatment more aspects of treatment
fewest side effects than Phase I or II trials.
Number of Patients Often 20-30 Often 100 or more Several hundred to several
thousand
What to Expect Patients often have many Patients often have many Phase III trials are usually
physical exams and tests physical exams and tests randomized. This means
(blood tests, for example) (blood tests, for example). half the patients in the
so doctors can find out This may take a lot of study are chosen at
how the treatment affects time. random to get the newer
them. This may take a lot treatment. The other half
of time. get the standard treatment.
Patients and doctors
do not decide which
treatment the patient
gets. As in Phase I and II
trials, patients may have
many physical exams and
tests that may take a lot
of time.
CATEGORIES OF 3. Bcl-2 inhibitors

TARGETED THERAPY 4. PARP inhibitors
5. PI3K inhibitors.
The main categories of targeted therapy are 6. VEGF inhibitor
listed below and fall under two main categories; 7. Braf inhibitors for metastatic melanoma
small molecules (Figure 1) and monoclonal harbors BRAF V600E mutation
antibodies (Mansour & Schwarz, 2008). 8. MEK inhibitors are used often in com-
bination with BRAF inhibitors to treat
Small Molecules melanoma
9. CDK inhibitors, e.g. PD-0332991, LEE011
To mention a few:
1. Janus kinase inhibitors (JAK)

2. ALK inhibitors
Figure 1.Mechanism of action
A CASE OF PERSONALIZED plan- Meier (Xu et al., 2012) uses assumptions

MEDICINE (XALKORI) given below:
Xalkori, is used in case of lung cancer patients 1. At any time both censored patients and
and shows great promise in this area. Specifi- those who continue to be followed have
cally, XALKORI is indicated for the treatment the same survival prospects.
of adults with previously treated anaplastic 2. Both early and late recruited subjects have
lymphoma kinase (ALK)-positive advanced the same survival probabilities
non-small cell lung cancer (NSCLC). There- 3. The event happens at the time specified.
fore, Xalkori falls under personalized medicine This creates problem in some cases when
domain. As referred, in a previous section the event would be detected at a regular
molecules such as Xalkori are used to treat examination, but in fact the event happened
specific subgroups of cancer patients in most between two examinations.
cases. It can then be expected that the efficacy
levels of the investigational drug applied to a The survival probability at any particular
carefully selected sample will be exceptionally time is calculated by the following formula:
high (Pfizer, 2013).
In the rest of this article we will examine Number of subjects
the above assumption by applying statistical living at the start
−Number of subjectsdied
St
methods and efficacy estimators used through Number of sublectsliving at thestart
academia and adjusting for different scenarios.
For each time interval, survival probability

THE KAPLAN- MEIER is calculated as the number of subjects surviving
SURVIVAL CURVE divided by the number of patients at risk. The
subjects, who died or moved out or dropped out,
The Kaplan-Meier is known as product limit are not included as “at risk”. Total probability
estimator and is defined as the probability of survival till that time interval is calculated
of surviving in a given length of time while by multiplying all the probabilities of survival
considering time in many small intervals. Ka- at all-time intervals preceding that time.
Table 2. Cycles vs. dropouts
For example, a placebo-controlled random- we divide one calendar year, i.e. the duration
ized trial proposes (Vong, Bergstrand, Nyberg, of the trial to 58 cycles {6.3 days}, the number
& Karlsson, 2012) to assess the effectiveness of of withdraws occurring within the cycle goes
Drug A in curing infants suffering from sepsis. to 0.This partially solves the third assumption
A previous study showed that proportion of referred previously.
subjects cured by Drug A is 50% and a clinically
important difference of 16% as compared to
placebo is acceptable (Xu et al., 2012). OPTIMAL SAMPLE
We observe in Table 2 that if D {Duration} SIZE SELECTION
is constant, set at 12 months, by increasing
Level of significance = 5%, Power = 80%, Type
gradually the number of cycles, the pace of
of test = two-sided
patients withdraws in each cycle starting at 21
days equals to one cycle until the number of 2
patients withdraws during the first cycle is less  
Z − Z  × (p1 × 1 − p1 + p2 × (1 − p2)
than one (see Figures 2, 3, and 4). Therefore  a b ( )
 2 
it can be assumed that the number of cycles n (p1) =
(p1 − p2)
2
needed as indicated in Table 2 is 58. So as, if

(1)
Figure 2. Mechanism of action (cycles=67)
Figure 3. Mechanism of action (cycles=112)
where Zα/2: Depends on level of significance, for 5%

this is 1.96
n: sample size required in each group, Zβ: This depends on power, for 80% this is 0.84
p1: proportion of subject cured by Drug A = 0.50,
p2: proportion of subject cured by Placebo = Therefore, by minimizing the first derivate
0.34, (Sakpal, 2010) of [1] we conclude that the
p1-p2: clinically significant difference = 0.16 number of patients must be recruited in the
Figure 4.Mechanism of action (cycles=18)
experimental arm for achieving (5% in this As it can be seen in Figure 5 the skew
case) statistical significance is: and steepness of Sample size function can be
approached by f (x)= 1/x2 & f (x)= 1/xπ in a
max (p1 ) = 2p22 , “precise” manner. Such an observation gives
an analytical estimating power for calculating
the first and second order derivatives for any
In other words p1 can be expressed as a given point within our sample. Our scope is to
find the optional point in the above graphical
function of p2 .Given the above formula for
interpretation, which requires the minimum
any prespecified levels of power Zβ and statis- number of patients and retains the maximum ef-
tical significance Zα we can calculate the ficacy between the two arms in a clinical setting.
minimum number of patients required in the We will call this point efficient frontier
clinical trial. So as to validate the above, we and will be denoted by ӗ. Since, Sample size
simulate for different levels of clinical efficacy. function is constructed empirically ӗ will be
For example in our first scenario we cal- approached by maximizing the first derivatives,
culate the minimum number of patients n (p1) for every point of our sample size range, averag-
as follows. ing the two maximum points accordingly for
1/x2 & 1/xπ and sequencing the result towards
p1 =0.3{constant} (proportion of subject a “closed neighbor” in our subjects.
cured by Placebo) Without loss of analysis since both our
p2 = 0.4 + Step1 (proportion of subject cured functions are monotonically decreasing we
by Drug) can “draw” the efficient frontier empirically
Step1= 0.001 from our dataset as the point ӗt in which ӗt+1
requires the same or greater number of patients.
In Figure 6 we plot the change in Sample
where Step is set to 0.001 ranging from (0.3, 0.4)
size required for every p2 = p2 + 15 Step2, as
& Step2=15
defined previously, versus efficacy, with p2 at
where Step2 is the difference between the mini-
Step0 equals to 0.4 and p1 equals 0.3 constant.
mum number n (p2) of patients required for every
Since our explicit estimator functions are
15 different levels of efficacy {i.e. 15Step1}.
Figure 5. Efficacy vs. sample size (p1= 0.3,p2=0.4)
monotonically decreasing, Sample size will Accordingly in Figure 9 we plot the change
behaves in the same fashion. Approaching, in Sample size required for every p2 = p2 + 15
efficient frontier empirically from the Table 3 Step2, as defined previously, versus efficacy,
(and Figure 6) our ӗ is as presented in Box 1. with p2 at Step0 equals to 0.2 and p1 equals 0.1
So forth the minimum number of patients constant. Since our explicit estimator functions
required is 269, with efficacy p2 + Stept ≈ p2 + are monotonically decreasing, Sample size will
Stept+1. at point ӗ. The ΔΕ needed for producing behaves in the same manner. Approaching, ef-
ӗ was 0.215 or else 16 Steps. ficient frontier empirically from Table 5 (and
Similarly in Figure 7 we plot the change Figures 9 and 10) our ӗ is as presented in Box 3.
in Sample size required for every p2 = p2 + 15 So forth the minimum number of patients
Step2, as defined previously, versus efficacy, required is 220, with efficacy p2 + Stept ≈ p2 +
with p2 at Step0 equals to 0.3 and p1 equals 0.2 Stept+1. at point ӗ. The ΔΕ needed for producing
constant. Since our explicit estimator functions ӗ was 0.225 or else 16 Steps.
are monotonically decreasing, Sample size will
behaves in the same manner. Approaching, ef-
ficient frontier empirically from Table 4 (and CONCLUSION
Figures 7 and 8) our ӗ is as presented in Box 2.
Personalized Medicine attempts to define mo-
So forth the minimum number of patients
lecular predictors of drug safety and efficacy
required is 220, with efficacy p2 + Stept ≈ p2 +
up front and to stratify patients and diseases
Stept+1. at point ӗ. The ΔΕ needed for producing
into subsets based on their unique characteris-
ӗ was 0.24 or else 16 Steps.
tics. This approach allows for an optimal re-
Box 1.
269 0,415 3 0,625
Figure 6. Δn vs. Δe (p1= 0.3,p2=0.4)
Table 3. Sample size required number (p1= 0.3,p2=0.4)
N Efficacy ΔN ΔΕ
353 0,4 84 0,4
347 0,401 57 0,415
340 0,402 41 0,43
333 0,403 30 0,445
327 0,404 23 0,46
321 0,405 18 0,475
315 0,406 14 0,49
310 0,407 11 0,505
304 0,408 9 0,52
299 0,409 8 0,535
293 0,41 7 0,55
288 0,411 5 0,565
283 0,412 5 0,58
278 0,413 4 0,595
274 0,414 4 0,61
269 0,415 3 0,625
264 0,416 2 0,64
260 0,417 3 0,655
256 0,418 2 0,67
252 0,419 2 0,685
247 0,42 2 0,7
243 0,421 1 0,715
240 0,422 2 0,73
236 0,423 1 0,745
232 0,424 1 0,76
228 0,425 1 0,775
225 0,426 0,69
222 0,427 0,705
218 0,428 0,72
215 0,429 0,735
212 0,43 0,75
208 0,431 0,765
205 0,432 0,78
202 0,433 0,695
199 0,434 0,71
197 0,435 0,725
194 0,436 0,74
Table 3. Continued
N Efficacy ΔN ΔΕ
191 0,437 0,755
188 0,438 0,77
186 0,439 0,785
183 0,44
Figure 7. Δn vs. Δe (ρ1= 0.3,ρ2=0.2)
Figure 8. Efficacy vs, sample size (ρ1= 0.3,ρ2=0.2)
N Efficacy ΔN ΔΕ
291 0,3 68 0,3
285 0,301 46 0,315
280 0,302 33 0,33
275 0,303 24 0,345
270 0,304 19 0,36
265 0,305 14 0,375
260 0,306 12 0,39
256 0,307 9 0,405
251 0,308 8 0,42
247 0,309 6 0,435
243 0,31 6 0,45
239 0,311 5 0,465
235 0,312 4 0,48
231 0,313 3 0,495
227 0,314 3 0,51
223 0,315 3 0,525
220 0,316 2 0,54
216 0,317 2 0,555
213 0,318 2 0,57
209 0,319 2 0,585
206 0,32 1 0,6
203 0,321 2 0,615
200 0,322 1 0,63
197 0,323 1 0,645
194 0,324 1 0,66
191 0,325 1 0,675
188 0,326 1 0,69
185 0,327 1 0,705
183 0,328 1 0,72
180 0,329 0 0,735
177 0,33 1 0,75
175 0,331 1 0,765
172 0,332 0 0,78
Box 2.
220 0,316 2 0,54
Figure 9. Δn vs. Δe (ρ1= 0.2,ρ2=0.1)
Figure 10. Efficacy vs. sample size (ρ1= 0.2,ρ2=0.1)
Box 3.
220 0,316 2 0,54
N Efficacy ΔN ΔΕ
196 0,2 42 0,2
193 0,201 30 0,215
190 0,202 21 0,23
187 0,203 16 0,245
183 0,204 12 0,26
180 0,205 10 0,275
177 0,206 8 0,29
175 0,207 7 0,305
172 0,208 5 0,32
169 0,209 5 0,335
166 0,21 4 0,35
164 0,211 3 0,365
161 0,212 3 0,38
159 0,213 2 0,395
156 0,214 3 0,41
154 0,215 2 0,425
152 0,216 1 0,44
149 0,217 2 0,455
147 0,218 2 0,47
145 0,219 1 0,485
143 0,22 1 0,5
141 0,221 1 0,515
139 0,222 1 0,53
137 0,223 1 0,545
135 0,224 1 0,56
133 0,225 1 0,575
131 0,226 0 0,59
130 0,227 1 0,605
128 0,228 1 0,62
126 0,229 0 0,635
124 0,23 1 0,65
123 0,231 0 0,665
121 0,232 1 0,68
120 0,233 0 0,695
118 0,234 1 0,71
117 0,235 0 0,725
115 0,236 0 0,74
114 0,237 1 0,755
112 0,238 0 0,77
sponse to therapies, which are more likely to Lazakidou, A., & Daskalaki, A. (2012). Quality as-
provide a benefit. In this manner in order to surance in healthcare service delivery, nursing and
personalized medicine: Technologies and processes.
validate the expected superior efficacy of
Hershey, PA: IGI Global.
molecules referring to PM it is sufficient to
apply to the statistical methods presented to Mansour, J. C., & Schwarz, R. E. (2008). Molecular
this study. In other words, assuming that mechanisms for individualized cancer care. [Review].
Journal of the American College of Surgeons, 207(2),
max (p1 ) = 2p22 and calculating the efficient 250–258. doi:10.1016/j.jamcollsurg.2008.03.003
frontier ӗ explicitly via 1/x2 & 1/xπ or empiri- PMID:18656055.
cally through the sample size dataset. Oldenburg, J., Watzka, M., Rost, S., & Muller, C. R.
(2007). VKORC1: Molecular target of coumarins.
[Research Support, Non-U S Gov’tReview]. Journal
REFERENCES of Thrombosis and Haemostasis, 1, 1–6. doi:10.1111/
j.1538-7836.2007.02549.x PMID:17635701.
Acharya, C. R., Hsu, D. S., Anders, C. K., Anguiano, Pfizer. (2013). XALKORI.
A., Salter, K. H., & Walters, K. S. et al. (2008). Gene
expression signatures, clinicopathological features, Sadée, W., & Dai, Z. (2005). Pharmacogenetics/ge-
and individualized therapy in breast cancer. [Research nomics and personalized medicine. Human Molecu-
Support, N I H, Extramural Research Support, Non- lar Genetics, 14(suppl 2), R207–R214. doi:10.1093/
U S Gov’t Retracted Publication]. Journal of the hmg/ddi261 PMID:16244319.
American Medical Association, 299(13), 1574–1587.
doi:10.1001/jama.299.13.1574 PMID:18387932. Saglio, G., Morotti, A., Mattioli, G., Messa, E.,
Giugliano, E., & Volpe, G. et al. (2004). Rational
Cichon, S., Nothen, M. M., Rietschel, M., & approaches to the design of therapeutics targeting
Propping, P. (2000). Pharmacogenetics of schizo- molecular markers: The case of chronic myelog-
phrenia. [Review]. American Journal of Medical enous leukemia. [Research Support, Non-U S
Genetics, 97(1), 98–106. doi:10.1002/(SICI)1096- Gov’t Review]. Annals of the New York Academy
8628(200021)97:1<98::AID-AJMG12>3.0.CO;2-W of Sciences, 423–431. doi:10.1196/annals.1322.050
PMID:10813809. PMID:15650267.
Clayton, T. A., Lindon, J. C., Cloarec, O., Antti, H., Sakpal, T. V. (2010). Sample size estimation in
Charuel, C., & Hanton, G. et al. (2006). Pharmaco- clinical trial. Perspect Clin Res, 1(2), 67–69.
metabonomic phenotyping and personalized drug PMID:21829786.
treatment. [Research Support, Non-U S Gov’t].
Nature, 440(7087), 1073–1077. doi:10.1038/na- Shastry, B. S. (2006). Pharmacogenetics and the
ture04648 PMID:16625200. concept of individualized medicine. [Review].
The Pharmacogenomics Journal, 6(1), 16–21.
Galas, D. J., & Hood, L. (2009). Systems biology doi:10.1038/sj.tpj.6500338 PMID:16302022.
and emerging technologies will catalyze the transition
from reactive medicine to predictive, personalized, Veer, L. J., & Bernards, R. (2008). Enabling per-
preventive and participatory (P4) medicine. IBC, 1, 6. sonalized cancer medicine through analysis of
gene-expression patterns. [10.1038/nature06915].
Haile, L. A. (2008). Making personalized medicine Nature, 452 (7187), 564-570.
a reality. Engineering & Biotechnology News, 28.
Vong, C., Bergstrand, M., Nyberg, J., & Karlsson,
Hanna, N., Shepherd, F. A., Fossella, F. V., Pereira, M. O. (2012). Rapid sample size calculations for a
J. R., De Marinis, F., & von Pawel, J. et al. (2004). defined likelihood ratio test-based power in mixed-
Randomized phase III trial of pemetrexed versus effects models. The AAPS Journal, 14(2), 176–186.
docetaxel in patients with non-small-cell lung cancer doi:10.1208/s12248-012-9327-8 PMID:22350626.
previously treated with chemotherapy. [Clinical Trial
Clinical Trial, Phase III Randomized Controlled Willard, G. (2012). Genomic and personalized
Trial. Research Support, Non-U S Gov’t]. Journal of medicine (2nd ed.). Academic Press.
Clinical Oncology, 22(9), 1589–1597. doi:10.1200/ Wong, S. Y. (2008). Pharmacogenomics and person-
JCO.2004.08.163 PMID:15117980. alized medicine. In A. Dasgupta (Ed.), Handbook of
drug monitoring methods (pp. 211–223). Humana
Press. doi:10.1007/978-1-59745-031-7_11.
Xu, S., Shetterly, S., Powers, D., Raebel, M. A., Tsai, Yan, Q. (2008). Pharmacogenomics in drug discovery
T. T., Ho, P. M., & Magid, D. (2012). Extension and development. Humana Press. doi:10.1007/978-
of Kaplan-Meier methods in observational stud- 1-59745-205-2.
ies with time-varying treatment. Value in Health,
15(1), 167–174. doi:10.1016/j.jval.2011.07.010
PMID:22264985.
Alexander Rompas received his bachelor degree in Mathematics from University of Patras, in
2006. In 2007 he obtained his M.Sc. in Finance from Imperial College London, pursuing his
thesis in options & derivatives securities with honors. He joined Moody’s KMV as a credit ana-
lyst & product support specialist with focus on Banking industry, until 2011. He is a member
of American Financial Association and a Charter Financial Engineer. Since 2013, he is a PhD
candidate and a member of Biomedical Engineering Laboratory at N.T.U.A. His current research
interest is in the area of clinical trials with adaptive statistical designs with a focus on orphan
and difficult to treat indications.
Athanasios Anastasiou received his Bachelor degree in Electrical & Electronic Engineering
from the University of Newcastle Upon Tyne, in 2005. In 2006, he received his M.Sc. degree
in Digital Signal & Image Processing from the University of Central Lancashire. Likewise, in
2009 he obtained his second M.Sc. degree in Biomedical Engineering held by the University of
Patras and the National Technical University of Athens (N.T.U.A.). Since 2011, he is a member
of the Biomedical Engineering Laboratory of N.T.U.A., and a Ph.D. candidate in Biomedical
Engineering (University of Patras and N.T.U.A.). His current research interests comprise Bioin-
formatics, Biosignal Processing, Digital Image Processing and technologies for assisted living.
He is a Member of the Institute of Electrical and Electronics Engineers (IEEE) - EMBS Society
and Computer Society, Member of the Institution of Engineering and Technology (IET) and a
member of the Technical Chamber of Greece (TEE).
Charalampos Tsirmpas was born in Athens in 1984. In 2009 he graduated from the School of
Electrical and Computer Engineering of the National Technical University of Athens (NTUA),
specializing in telecommunications. Since 2009 he is a PhD candidate at the Biomedical Engineer-
ing Laboratory of NTUA. His research interests include areas such as Bioinformatics, Telemetry,
Telemedicine, and secure transmission of medical data. Also, his PhD research focuses on the
“Internet of Things” in healthcare applications. He is a member of the Institute of Electrical
and Electronics Engineers (IEEE) and the Technical Chamber of Greece (TEE).
Dimitra Iliopoulou, graduated from the School of Electrical and Electronic Engineering at
NTUA and is a post doctorate associate of the Biomedical Engineering Laboratory. Her research
involves disease modelling (mainly diabetes mellitus and cervical cancer) and bioinformatics
focusing on self-adaptive treatment planning, large data manipulation and systems interoper-
ability. She has published 10 scientific papers. She has worked in IT and Telecoms in several
companies (Multimedia A.E. – DOL, Bull ATS, Datamed, Omnis M.). She has been active in
several Research projects in regards to the application of IT in health.
Dimitris Koutsouris was born in Serres, Greece in 1955. He received Diploma in Electrical En-
gineering in 1978 (Greece), DEA in Biomechanics in 1979 (France), Doctorat in Genie Biologie
Medicale (France), Doctorat d’ Etat in Biomedical Engineering 1984 (France). Since 1986 he
was research associate on the USC (Los Angeles), Renè Dèscartes (Paris) and Assoc. Professor
at the Dept. of Electrical & Computers Engineering of NTU of Athens. He is currently Professor
& head of the Biomedical Engineering Laboratory. He has published over 100 research articles
& book chapters & more than 150 conference communications. He has been the former elected
president of the Hellenic Society of Biomedical Technology, principal investigator in many
European & National Research programs, especially in the field of Telematics in Healthcare.

Statistical Power and Sample Size Size in Personalized Medicine

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Statistical Power and Sample Size Size in Personalized Medicine

Uploaded by

Copyright:

Available Formats

72 International Journal of Systems Biology and Biomedical Technologies, 2(2), 72-88, April-June 2013

Statistical Power and Sample

INTRODUCTION molecular predictors of drug safety and efficacy

Table 1. Clinical trial process

Phase I Phase II Phase III

CATEGORIES OF 3. Bcl-2 inhibitors

1. Janus kinase inhibitors (JAK)

Figure 1.Mechanism of action

A CASE OF PERSONALIZED plan- Meier (Xu et al., 2012) uses assumptions

For each time interval, survival probability

Table 2. Cycles vs. dropouts

needed as indicated in Table 2 is 58. So as, if

Figure 2. Mechanism of action (cycles=67)

Figure 3. Mechanism of action (cycles=112)

where Zα/2: Depends on level of significance, for 5%

Figure 4.Mechanism of action (cycles=18)

Figure 5. Efficacy vs. sample size (p1= 0.3,p2=0.4)

269 0,415 3 0,625

Figure 6. Δn vs. Δe (p1= 0.3,p2=0.4)

Table 3. Sample size required number (p1= 0.3,p2=0.4)

Figure 7. Δn vs. Δe (ρ1= 0.3,ρ2=0.2)

Figure 8. Efficacy vs, sample size (ρ1= 0.3,ρ2=0.2)

Table 4. Sample size required number (p1= 0.2,p2=0.3)

220 0,316 2 0,54

Figure 9. Δn vs. Δe (ρ1= 0.2,ρ2=0.1)

Figure 10. Efficacy vs. sample size (ρ1= 0.2,ρ2=0.1)

220 0,316 2 0,54

Table 5. Sample size required number (p1= 0.1,p2=0.2)

193 0,201 30 0,215

190 0,202 21 0,23

187 0,203 16 0,245

183 0,204 12 0,26

180 0,205 10 0,275

177 0,206 8 0,29

175 0,207 7 0,305

172 0,208 5 0,32

169 0,209 5 0,335

166 0,21 4 0,35

164 0,211 3 0,365

161 0,212 3 0,38

159 0,213 2 0,395

156 0,214 3 0,41

154 0,215 2 0,425

152 0,216 1 0,44

149 0,217 2 0,455

147 0,218 2 0,47

145 0,219 1 0,485

143 0,22 1 0,5

141 0,221 1 0,515

139 0,222 1 0,53

137 0,223 1 0,545

135 0,224 1 0,56

133 0,225 1 0,575

131 0,226 0 0,59

130 0,227 1 0,605

128 0,228 1 0,62

126 0,229 0 0,635

124 0,23 1 0,65

123 0,231 0 0,665

121 0,232 1 0,68

120 0,233 0 0,695

118 0,234 1 0,71

117 0,235 0 0,725

115 0,236 0 0,74