Can Big Data Shed Light on the
Origins of Pediatric Cancer?
A. Lindsay Frazier, MD, ScM,a Mark Krailo,b Jen Poynterc
Twenty-five years ago, Sir Richard
Doll and Sir Richard Peto estimated in
their landmark monograph, The Causes
of Cancer, that 75% to 80% of adult
cancer could be attributed to lifestyle
choices and environmental exposures.1
In stark contrast, elucidating the
causes of childhood cancer remains
one of the most vexing questions
in medicine. Despite progress in
identifying the molecular causes of
many pediatric tumors, whether there
are extrinsic environmental causes of
pediatric cancer remains an elusive
question. Without this evidence, the
“primary” prevention of childhood
cancer remains out of reach. The
inherent epidemiologic challenge of
studying the etiology of childhood
cancer is simple math. Childhood
cancer is a rare event, and the power to
detect an association is limited by the
small number of incident cases.
risk of hyperbilirubinemia that might
independently also affect cancer risk
and thereby confound the association
between phototherapy and infant
cancer, the authors developed a
“score” for each subject that captured
all the reasons known to increase the
need/propensity for phototherapy.
This propensity score was necessary
because it would not have been
possible to adjust for each of these
individual risk factors given the rarity
of the outcome, infant cancer. The
authors report, after propensityadjustment, a significant association
between receipt of phototherapy and
increased risk of myeloid leukemia
(adjusted odds ratio [aOR], 2.6; 95%
confidence interval [CI]: 1.3–5.0)
and kidney cancer (aOR, 2.5; 95% CI:
1.2–5.1) and “other” cancers (aOR, 1.6;
95% CI: 1.1–2.4) in infants exposed to
phototherapy.
In this issue of Pediatrics, 2 companion
articles by the same set of authors
address this conundrum of rarity by
using “Big Data” to address whether
neonatal phototherapy can be
implicated as a cause of childhood
cancer. Wickremasinghe et al2
report the results of the California
Late Impact of Phototherapy Study
(CLIPS), a study of ∼5 million infants
who were identified by the California
Office of Statewide Health Planning
and Development by probabilistically
linking vital statistics (birth and death
certificates) with patient discharge
data between 1998 and 2007. A total
of 1100 infants (58 of whom had
received phototherapy) developed
cancer between 60 to 365 days of
life. Because there are many factors
previously identified that increase the
In the second paper, Newman et
al3report the results of the Late
Impact of Getting Hyperbilirubinemia
or Phototherapy (LIGHT) study, a
retrospective cohort analysis that
included 500 000 children born at
Kaiser Permanente Northern California
from 1995 to 2011. This study included
all cases of childhood cancer and was
not restricted to infants as in CLIPS.
Overall, 711 children were diagnosed
with cancer; 60 of these children had
received phototherapy. Because this
study was conducted in a single health
care system, the actual individual
covariates could be abstracted from
the patient’s health record, including
the bilirubin levels. Although the crude
incidence rate ratios were elevated for
myeloid leukemia and liver cancer (P ≤
.05) and kidney cancer (.05 < P ≤ .10),
Downloaded from by guest on July 31, 2017
PEDIATRICS Volume 137, number 6, June 2016:e20160983
aDepartment
of Pediatric Oncology, Dana-Farber Cancer
Institute, Harvard Medical School, Boston, Massachusetts;
bDepartment of Preventive Medicine, Keck School of
Medicine, University of Southern California, Los Angeles,
California; and cDivision of Epidemiology and Clinical
Research, Department of Pediatrics, University of
Minnesota, Minneapolis, Minnesota
Opinions expressed in these commentaries are
those of the author and not necessarily those of the
American Academy of Pediatrics or its Committees.
DOI: 10.1542/peds.2016-0983
Accepted for publication Mar 28, 2016
Address correspondence to A. Lindsay Frazier, MD,
ScM, Department of Pediatric Oncology, DanaFarber Cancer Institute, 450 Brookline Ave, Boston,
MA 02215. E-mail: [email protected]
PEDIATRICS (ISSN Numbers: Print, 0031-4005; Online,
1098-4275).
Copyright © 2016 by the American Academy of
Pediatrics
FINANCIAL DISCLOSURE: Dr Frazier has consulted
for Decibel Therapeutics; and Drs Krailo and
Poynter have indicated they have no financial
relationships relevant to this article to disclose.
FUNDING: No external funding.
POTENTIAL CONFLICT OF INTEREST: The authors
have indicated they have no potential conflicts of
interest to disclose.
COMPANION PAPER: Companions to this article can
be found on online at www.pediatrics.org/cgi/doi/
10.1542/peds.2015-1353 and www.pediatrics.org/
cgi/doi/10.1542/peds.2015-1354.
To cite: Frazier AL, Krailo M, Poynter J. Can
Big Data Shed Light on the Origins of Pediatric
Cancer?. Pediatrics. 2016;137(6):e20160983
COMMENTARY
after these rate ratios were adjusted
with the propensity score, none of the
associations remained significant. It
should be noted that, as in CLIPS, the
associations between phototherapy
and cancer in the LIGHT study were
stronger among children with Down
syndrome.
But lest one be swayed by the allure
of “Big Data,” even in these extremely
large datasets, it is important to
hone in on the actual number of
cases observed. In the larger CLIPS
study, the increased odds of myeloid
leukemia were based on 10 cases
in the exposed versus 103 in the
unexposed. In the LIGHT study, the
increased risk of liver cancer is based
on 3 cases in the exposed versus 9
cases in the unexposed. Pediatric
cancer remains a rare disease.
Although “Big Data” solves some
problems, it creates others. The
administrative datasets used in
the Wickremasinghe paper2 were
deidentified. The “probabilistic”
linkage of these datasets means
that the birth and death records
were linked to the most likely
admission and discharge codes,
without certainty that it was truly
the same patient. Because the actual
individual is not identified, the
report of cancer from the admission
or discharge code cannot be crosschecked with another source,
such as the California State Cancer
Registry. This is unfortunate because
the cancer diagnosis reported in
the administrative datasets use
International Classification of
Diseases, Ninth Revision (ICD-9)
codes rather than in the more
precise childhood cancer codes of
the International Childhood Cancer
Classification. ICD-9 codes are usually
abstracted by those responsible
for billing, whereas International
Childhood Cancer Classification codes
used in cancer registries are assigned
by specially trained cancer registrars.
The potential for misclassification
inherent in using the ICD-9 codes
is apparent when one examines the
2
cancer diagnoses contained in the
“other cancer” category in CLIPS:
bone cancer and skin cancer. Neither
bone cancer nor skin cancer occurs
in infants. These diagnoses must be
misclassifications, likely of metastatic
disease from other sites. An
additional limitation in CLIPS is the
lack of individual-level information,
which makes it impossible to exclude
the possibility of confounding by
indication.
The design of the LIGHT study is an
interesting counterbalance to the
CLIPS study. Because this study took
place within a single health care
system, the diagnosis of cancer was
derived from the actual hospitalbased cancer registry, and the
reports of the most prevalent cancers
were further verified by medical
record review. In addition, the
authors could abstract the values of
potential confounders, including total
serum bilirubin, as well as construct
a quasi-measure for the “dose” of
phototherapy. The drawback to
the LIGHT study is that the cohort,
although still huge, is an order of
magnitude smaller than the CLIPS
study (5 × 105 vs 5 × 106). Notably,
after the more precise adjustment
for the potential confounders is
performed in the CLIPS study,
the association of phototherapy
with childhood cancer is no longer
significant.
Study design aside, as practitioners,
we are faced with deciding, based on
these data, whether phototherapy
causes infant and childhood cancers.
Causal inference is, in the end, a
judgment call. Famously, Bradford
Hill4 suggested a set of criteria by
which to judge causality that include:
(1) temporality, (2) strength of the
association, (3) dose-response, (4)
reversibility, (5) consistency, (6)
biologic plausibility, (7) specificity,
and (8) analogy.
If we apply these criteria to the data
at hand, we see that several of these
criteria have been met. The exposure
is a plausible one; phototherapy is
Downloaded from by guest on July 31, 2017
blue light, a type of ionizing radiation
that has been shown to be associated
with DNA damage.5–9 Most of the
previous case-control studies have
suggested an association between
phototherapy and childhood cancer,
principally acute myelogenous
leukemia (AML) (consistency).10–15
Another form of ionizing radiation,
UV light (analogy), is directly linked
to the risk of skin cancer and is now
classified as a type I carcinogen
by the International Agency for
Research on Cancer.16 In the LIGHT
study, the association between AML
and phototherapy did appear to be
dose-related (biological gradient);
however, this conclusion is based
on 2 cases at the highest dose level.
Although these data do support a
potential role of phototherapy in
childhood cancer to some degree,
a number of questions remain. For
example, why would phototherapy
cause myeloid but not lymphocytic
leukemia, and why kidney and liver
cancers but not skin cancer (which
is the cancer associated with UV
radiation). These inconsistencies are
difficult to fathom if phototherapy is
indeed the cause of increased cancer
incidence.
The authors further suggest that
we should be concerned because
of the prevalence of the exposure
in the population and the potential
attributable risk. In fact, based on
data presented in the LIGHT study,
the number of infants receiving
phototherapy within the Kaiser
Permanent Northern California
health care system has increased
from 2.7% of all children in 1995 to
15.9% of children in 2011, in part
because of the introduction of home
phototherapy.3 However, if the rates
of phototherapy use are indeed
increasing and causally related to the
incidence of certain cancers (eg, AML,
Wilms tumor, and hepatoblastoma),
one might expect the incidence
of these tumors to be increasing
in the population in response to
the increasing prevalence of the
FRAZIER et al
exposure. However, several reports
from population-based cancer
registry data suggest that this is not
the case17–19
In the end, acknowledging that the
information is imperfect, general
pediatricians and neonatologists
must make a choice. These data
suggest that phototherapy may
not be harmless, and that the risks
as well as the benefits need to be
weighed before flipping the switch.
ABBREVIATIONS
AML: acute myelogenous
leukemia
aOR: adjusted odds ratio
CI: confidence interval
CLIPS: California Late Impact of
Phototherapy Study
ICD-9: International
Classification of Diseases,
Ninth Revision
LIGHT: Late Impact of Getting
Hyperbilirubinemia or
Phototherapy
REFERENCES
study of phototherapy and childhood
Cancer in Northern California.
Pediatrics. 2016;137(6):e20151354
4. Hill AB. The Environment and Disease:
Association or Causation? Proc R Soc
Med. 1965;58(5):295–300
5. Oláh J, Tóth-Molnár E, Kemény L, Csoma
Z. Long-term hazards of neonatal bluelight phototherapy. Br J Dermatol.
2013;169(2):243–249
6. Speck WT, Rosenkranz HS.
Phototherapy for neonatal
hyperbilirubinemia--a potential
environmental health hazard to
newborn infants: a review. Environ
Mutagen. 1979;1(4):321–336
7. Aycicek A, Kocyigit A, Erel O, Senturk
H. Phototherapy causes DNA damage
in peripheral mononuclear leukocytes
in term infants. J Pediatr (Rio J).
2008;84(2):141–146
8. Tatli MM, Minnet C, Kocyigit A,
Karadag A. Phototherapy increases
DNA damage in lymphocytes of
hyperbilirubinemic neonates. Mutat
Res. 2008;654(1):93–95
9. Karadag A, Yesilyurt A, Unal S,
et al. A chromosomal-effect study
of intensive phototherapy versus
conventional phototherapy in
newborns with jaundice. Mutat Res.
2009;676(1-2):17–20
1. Doll R, Peto R. The causes of cancer:
quantitative estimates of avoidable
risks of cancer in the United
States today. J Natl Cancer Inst.
1981;66(6):1191–1308
10. Kahveci H, Dogan H, Karaman A, Caner
I, Tastekin A, Ikbal M. Phototherapy
causes a transient DNA damage in
jaundiced newborns. Drug Chem
Toxicol. 2013;36(1):88–92
2. Wickremasinghe AC, Kuzniewicz
MD, Grimes BA, et al. Neonatal
phototherapy and infantile cancer.
Pediatrics 2016;137(6):e20151353
11. Olsen JH, Hertz H, Kjaer SK, Bautz A,
Mellemkjaer L, Boice JD Jr. Childhood
leukemia following phototherapy
for neonatal hyperbilirubinemia
(Denmark). Cancer Causes Control.
1996;7(4):411–414
3. Newman TB, Wickremasinghe AC,
Walsh EM, et al. Retrospective cohort
PEDIATRICS Volume 137, number 6, June 2016
Downloaded from by guest on July 31, 2017
12. Cnattingius S, Zack MM, Ekbom A, et al.
Prenatal and neonatal risk factors for
childhood lymphatic leukemia. J Natl
Cancer Inst. 1995;87(12):908–914
13. Roman E, Ansell P, Bull D. Leukaemia
and non-Hodgkin’s lymphoma in
children and young adults: are
prenatal and neonatal factors
important determinants of disease?
Br J Cancer. 1997;76(3):406–415
14. Podvin D, Kuehn CM, Mueller BA,
Williams M. Maternal and birth
characteristics in relation to childhood
leukaemia. Paediatr Perinat Epidemiol.
2006;20(4):312–322
15. Cnattingius S, Zack M, Ekbom A,
Gunnarskog J, Linet M, Adami HO.
Prenatal and neonatal risk factors
for childhood myeloid leukemia.
Cancer Epidemiol Biomarkers Prev.
1995;4(5):441–445
16. International Agency for Research
on Cancer. IARC monographs on the
evaluation of carcinogenic risks
to humans. Available at: http://
monographs.iarc.fr/ENG/Classification/
latest_classif.php. Accessed March 25,
2016
17. Kenney LB, Miller BA, Ries LA,
Nicholson HS, Byrne J, Reaman GH.
Increased incidence of cancer in
infants in the U.S.: 1980-1990. Cancer.
1998;82(7):1396–1400
18. Gurney JG, Ross JA, Wall DA, Bleyer WA,
Severson RK, Robison LL. Infant cancer
in the U.S.: histology-specific incidence
and trends, 1973 to 1992. J Pediatr
Hematol Oncol. 1997;19(5):428–432
19. Gurney JG, Davis S, Severson RK, Fang
JY, Ross JA, Robinson LL. Trends in
cancer incidence among children in
the U.S. Cancer Epidemiol Biomarkers
Prev. 1996;78(3):532–541
3
Can Big Data Shed Light on the Origins of Pediatric Cancer?
A. Lindsay Frazier, Mark Krailo and Jen Poynter
Pediatrics 2016;137;; originally published online May 23, 2016;
DOI: 10.1542/peds.2016-0983
Updated Information &
Services
including high resolution figures, can be found at:
/content/137/6/e20160983.full.html
References
This article cites 16 articles, 2 of which can be accessed free
at:
/content/137/6/e20160983.full.html#ref-list-1
Subspecialty Collections
This article, along with others on similar topics, appears in
the following collection(s):
Hematology/Oncology
/cgi/collection/hematology:oncology_sub
Permissions & Licensing
Information about reproducing this article in parts (figures,
tables) or in its entirety can be found online at:
/site/misc/Permissions.xhtml
Reprints
Information about ordering reprints can be found online:
/site/misc/reprints.xhtml
PEDIATRICS is the official journal of the American Academy of Pediatrics. A monthly
publication, it has been published continuously since 1948. PEDIATRICS is owned, published,
and trademarked by the American Academy of Pediatrics, 141 Northwest Point Boulevard, Elk
Grove Village, Illinois, 60007. Copyright © 2016 by the American Academy of Pediatrics. All
rights reserved. Print ISSN: 0031-4005. Online ISSN: 1098-4275.
Downloaded from by guest on July 31, 2017
Can Big Data Shed Light on the Origins of Pediatric Cancer?
A. Lindsay Frazier, Mark Krailo and Jen Poynter
Pediatrics 2016;137;; originally published online May 23, 2016;
DOI: 10.1542/peds.2016-0983
The online version of this article, along with updated information and services, is
located on the World Wide Web at:
/content/137/6/e20160983.full.html
PEDIATRICS is the official journal of the American Academy of Pediatrics. A monthly
publication, it has been published continuously since 1948. PEDIATRICS is owned,
published, and trademarked by the American Academy of Pediatrics, 141 Northwest Point
Boulevard, Elk Grove Village, Illinois, 60007. Copyright © 2016 by the American Academy
of Pediatrics. All rights reserved. Print ISSN: 0031-4005. Online ISSN: 1098-4275.
Downloaded from by guest on July 31, 2017