Essays /

Cancer Side Effects Prediction Essay

Essay preview

2013 IEEE 15th International Conference on e-Health Networking, Applications and Services (Healthcom 2013)

Clinical-based Prediction of Side Effects in Colon
Cancer Chemotherapy
Mouna Kacimi

Ognjen Savkovi´c

Manfred Mitterer

Faculty of Computer Science
Free University of Bozen-Bolzano
I-39100 Bozen-Bolzano, Italy
[email protected]

Faculty of Computer Science
Free University of Bozen-Bolzano
I-39100 Bozen-Bolzano, Italy
[email protected]

Hospital Franz Tappeiner
I-39012, Meran-Merano, Italy
[email protected]

Abstract—Chemotherapy is used to treat cancer by killing
malignant cells or stopping them from multiplying. However, it can also harm healthy cells, which causes side effects. Some of the side effects of chemotherapy do not pose a serious threat to patients’ health. But, some others can be very serious such as the rapid fall in white blood cells making the patient vulnerable to serious infections. Different approaches have been proposed in the literature to predict the probability of experiencing a certain side effect on a specified day of each cycle of the chemotherapy. In our work, we are interested in predicting the side effects a patient is more likely to experience in each cycle. To this end, we take a different approach where we propose a predictive

model based on patient clinical information, such as concomitant diseases and medications. We have analyzed data from from 75 patients under FOLFOX chemotherapy for colon cancer. The
results show that our model improves the prediction accuracy compared to previously proposed time-based approaches. The
goal of this work is to help healthcare professionals in identifying possible side effects before starting a chemotherapy and taking the necessary actions to improve the quality of the treatment.



A. Motivation
Cancer is a class of diseases characterized by a growth of
abnormal cells that divide uncontrollably and manage to move throughout the body destroying healthy tissue. According to
the World Health Organization [20], cancer is becoming a
leading cause of death worldwide accounting for 7.6 million
deaths (around 13% of all deaths) in 2008. These numbers are projected to continue rising, with an estimated 13.1 million deaths worldwide in 2030. Hence, research targeting cancer
prevention and treatment is of a vital importance. For this
reason, many efforts were invested over the last decades
to improve the quality of medical treatments leading to an
increase of the event free and overall survival rate. More
specifically, adjuvant chemotherapy, employed after tumor removal surgery, was proven to help reducing the recurrence risk of some types of cancer [18], [19]. However, chemotherapy
can damage healthy cells along with cancer cells. Patients
receiving chemotherapy may experience significant side effects, especially if polychemotherapy is used, that could have a negative impact on their quality of life and daily living [17], [1]. Moreover, side effects can influence the overall survival [5] leading in some cases to mortality [10]. To address this problem, a comprehensive study of the causes of side effects is necessary to provide an effective prediction of their occurrence. In this way, health care professionals can take the necessary

978-1-4673-5801-9/13/$26.00 ©2013 IEEE


actions to alleviate the discomfort and anxiety of patients. Moreover, they can tailor the treatment depending on the
patient’s vulnerability to side effects for having a better control on his reactions to the chemotherapy.
B. Related Work
The problem of chemotherapy side effects have called
for the use of technology as a mean of communication
between professionals and patients. The goal is to improve
the symptom control and achieve a better quality of life by
reducing the hospitalizations rate and cost [12], [11], [7]. A practical example of the use of technology in cancer care is the Advanced Symptom Management System [8], [14], [6],
[15]. It consists of a mobile telephone-based remote symptom monitoring system which can be used to register, monitor and predict the side-effects of chemotherapy while the patient is not with a healthcare professional [4]. In this kind of systems, patients are asked to complete a symptom questionnaire on a

mobile phone on daily basis and sent this information directly to their hospital-based healthcare professional. Self-care advice is then given on the basis of the reported symptoms. All
information collected from patients during their chemotherapy is then exploited to develop predictive models for side effects. Examples of this models include the work by Maguire et. al., [13] that aims at estimating the probability of experiencing each symptom on a specified day of treatment, for patients

with breast cancer. This study shows that the probability
of experiencing a specific symptom is time dependent. Its
tendency is to have a peak effect, around the day in which
the treatment is received by patients, and an inverted U-shape effect rising from a low on the day after treatment to a
peak around mid-cycle before falling again. This work has
been improved and generalized to other types of cancer by
Mazzocco et. al., [16].
Very few approaches have looked at patient characteristics
and clinical information to predict side effects. The work by Dranitsaris et. al., [3] proposes a regression model based on patient information to estimate cardiotoxic risk for patients with metastatic breast cancer. Their study has shown that

characteristics such as patient age and weight and the number of cumulative cycles are good predictors for having cardiac
toxicity. Similarly, Kuderer et. al., [10] show that patient characteristics, type of malignancy, comorbidities, and infectious complications are useful in identifying patients at increased risk of serious medical complications and mortality associated

2013 IEEE 15th International Conference on e-Health Ne...

Read more


-1 -39012 -39100 -4673 -5801 -9 /. /13 / 0 0.047 0.136 0.2 0.325 0.4 0.447 0.505 0.580 0.6 0.636 0.670 0.673 0.688 0.705 0.718 0.76 0.760 0.761 0.787 0.796 0.8 0.839 0.848 0.857 0.865 0.877 0.883 0.928 0.930 0.995 0.997 1 10 100 104 106 107 11 12 13 13.1 13th 14 15 15th 16 17 18 19 1991 1995 1998 2 20 2003 2006 2008 2009 2011 2012 2013 2030 22 22582266 250 26.00 2740 28 282 29 298 3 30 306315 332 34 380386 39 4 40 401 433 437444 443450 444453 5 5152 571 571.2 571.8 58 583590 6 617 618 619 62 620 621 65 7 7.6 71 74 742 75 752 77 79 8 86 9 901906 93 97 978 99 abnorm absenc abstract accord account accur accuraci achiev acknowledg action actual adapt add addit address adjuv adult advanc advic affect age aim al alcohol allevi allow along also alway amilorid among analyz anderson anemia angela angio anthracyclin anxieti aorta appli applic approach area around arriv ask aspect assess assign associ asym asyms-serat atherosclerosi author automat avail averag b bacteri baksh base baselin basi basic becom before-hand behind better big binari biol blood bodi bolzano bonadonna bozen bozen-bolzano brambilla breast british build bunin c call cancer cardiac cardiotox care case categori caus cd cell certain chanc chang character characterist chemeotherapi chemotherapi chemotherapy-associ chemotherapy-rel chemotherapyrel children chosen cl class classif classifi clear cleland clinic clinical-bas closer cm code cohen collect colon colorect combin common communic comorbid compar comparison complet complic compon comprehens comput conceiv concept concis conclus concomit conf confer consist contain continu contrast control correct correspond cosler cost could coventri cowi crawford cristina cross cross-valid crossvalid cumul curs curv cycl cyclebasi cyclophosphamid d daili dale damag data databas dataset day day-by-day deal death decad deceas decis depend describ descript design destroy determin develop diabet diagnos differ dimens dimension direct discomfort discret discuss diseas disord disturb divid done dranitsari dtrain due dynam dysgeusia e e-health eabi eager eas easi easili edg efer effect effort employ end eng england equal erform especi estim et european evalu even event exampl exit experi experienc experiment explain exploit explor extract f f1 f2 f3 f4 f5 f6 f7 factor faculti failur fall fals false-posit famili feasibl featur febril ffect field fig figur final find first fix flexibl fluorouracil fn fold folfox follow follow-up forbat form formal foucault franz free frequent function fund g gain gamper gastro gastrointestin gc1 gc2 gci gcl gelmon gender general generat get gianni gibson give given go goal goldwein good gray gretton growth h hand handheld happen hargan harm health healthcar healthcom healthi heart help hemotherapi henc high higher histori hospit hospital-bas howev hussain hydrochlorid hypercalcemia hyperpotassemia hypertens i.e icd9 icd9-cm ide idea identifi ieee ii iii ijmei illingworth impact implement import improv imput includ increas independ indic individu infect infecti influenc inform input instanc interest interior intern interpret introduc invert invest investig iscuss itali iv j joumal journal k kacimi kearney khayati khorrami kidd kill kind knearest knn knowledg known kramer kuder l lab label laboratori lack larg larsen last lazi lead leadbett leaf learn least less life like limb limit list literatur littl live liver longer look loui low lung lyman m maguir main make malign manag manfr [email protected] mani map margin matter maximum may mazzocco mccall mccann mcgee mean meaning measur med medic medicin medulloblastoma meran meran-merano merano metastat methotrex metric mid mid-cycl miller million milvia mine mitter mobil mobile-phone-bas model molassioti moliterni moment monitor morbid moreov mortal motiv mouna [email protected] move muir multipli muscl n name nausea necessari need negat neighbor network neurolog neurosurg neutropenia new node node-posit non non-object norri note ntroduct null number nurs o object observ occurr odel ognjen [email protected] ohlen old onclus oncolog one open organ origin other otherwis outpati outperform output overal p p1 p2 packer page pair palmer paramet paresthesia particular partit path patient pattern peak pearson percentag percept perform perilongo perman person phase phone phonebas pilot pinuccia pj pn pneumonia point polychemotherapi pose posit possibl power practic precis predict predictor prefix prepar prescrib presenc present preval prevent previous principl prior probabl problem proc process produc profession project proport propos proven provid provinc psycholog purpos qualiti queri questionnair r random rank rapid rare rate rates.http rayson reaction reardon reason receiv recent recurr recurs redict reduc regard regist regress relat relev remain remot remov repeat report repres represent requir research respect restrict result retriev review rise risk roc root row rowntre rule rule-bas rulebas run ryan s368 sage sampl sandor savkovi scienc score second see select self self-car sensat sensi6vity sensit sent serat serious serv servic set setup sever shape share show shown si side side-effect signific similar similarity-bas similaritybas simpli simplic simplifi sing singl size slight small smith soc solv sometim space spars sparsiti specif specifi specific split standard start step stop store straightforward stratifi strength stricker structur studi subsampl subset suffix suitabl summari supplement support surgeri surviv sutton swingler symptom system systemat tabl tailor take tappein tarassenko target task taylor techniqu technolog telemonitor telephon telephone-bas ten tendenc term test third though threat three threshold throughout thus thyrotoxicosi time time-bas tired tissu toler tool top toward toxic train transact treat treatment tree trial trigger true true-posit tumor turner two type u u-shap uk ultim uncontrol undergo understand univers use user utilis v valagussa valid valu vari variabl vector velder vincent visual vital vomit vulner way weak weaver weight well whether white word work world worldwid worst would xperiment year yes young zambetti zimmerman