Gel Berg 1997
Gel Berg 1997
Gel Berg 1997
2
© International Epidemiological Association 1997 Printed in Great Britain
Gelberg K H (New York State Department of Health, Bureau of Environmental and Occupational Epidemiology, Albany,
NY, USA), Fitzgerald E F, Hwang S and Dubrow R. Growth and development and other risk factors for osteosarcoma in
children and young adults. International Journal of Epidemiology 1997; 26: 272–278.
Background. Risk factors for osteosarcoma in young people were investigated in a population-based case-control study
among residents of New York State, excluding New York City.
Methods. Cases (n = 130) were diagnosed between 1978 and 1988 at <24 years of age. Controls were randomly selected
from birth certificates and were pair matched to cases on year of birth and sex. Exposure information was obtained by
telephone interview with a subject and/or parent, and from birth certificates and school and medical records.
Results and Conclusions. A significant positive association was observed with height one year before diagnosis ( P-value
for trend = 0.02). No significant associations were observed between osteosarcoma and weight or body mass index one
year before diagnosis, birth length, birthweight, gestational age, having reached puberty, having begun growth spurt, age
at puberty, age growth spurt began, medical x-rays, antenatal exposures, family history of cancer, birth defects, or
parental occupation.
Keywords: osteosarcoma, bone neoplasms, child development, growth, case-control study, height
Bone cancer is the fourth most common cancer in breeds of dogs which suggests that a relationship may
people under 25 years of age,1 and osteosarcoma is the exist between human bone cancer and a large body size
most common type of bone cancer in this age group. at the time of diagnosis.13,14
Very little is known about the aetiology of osteo- A number of antenatal environmental exposures such
sarcoma in humans; the only known aetiologic agent is as infectious agents, drugs and radiation are capable of
radiation.2,3 Other factors suggested to be related to altering the normal development of an embryo which
osteosarcoma include tall stature, previous bone trauma could contribute to the development of osteosarcoma in
and viruses.3–9 One case-control study observed an young people. Parental occupation is also of interest
association between osteosarcoma and short length at because parents can bring home chemicals or dusts
birth.9 Genetic factors have been identified in a small from the workplace on their clothes, thus exposing their
percentage of cases.6,10,11 children.
Osteosarcomas are distributed throughout the entire The purpose of this population-based case-control
skeleton, with the rapidly growing long bones of the study was to identify risk factors for osteosarcoma
extremities affected most often among young people.12 among people <24 years of age. The specific aims were
An increase in osteosarcoma incidence around puberty to investigate the relationship between osteosarcoma in
has led many investigators to postulate that this cancer young people and growth and development, radiation
may be associated with the onset of puberty and may be exposure, antenatal exposures, family history of cancer,
a function of growth.4–6 Animal studies have demon- birth defects, and parental occupation. In a previous
strated an excess risk of bone sarcomas among larger publication using data from this study, the relationship
between osteosarcoma in young people and fluoride
exposure was investigated.15
* New York State Department of Health, Bureau of Environmental and
Occupational Epidemiology, Albany, NY, USA.
** Department of Epidemiology and Public Health, Yale University METHODS
School of Medicine, New Haven, CT, USA.
†
Current address: New York State Department of Health, Bureau
Cases and Controls
of Occupational Health, 2 University Place Room 155, Albany, Details of the selection of cases and controls for
NY 12203, USA. this study have been described elsewhere.15 Briefly,
272
RISK FACTORS FOR OSTEOSARCOMA IN YOUTHS 273
osteosarcoma cases who were ,25 years of age and could be located for 118 (83%) of these birth certific-
residents of New York State, excluding New York City, ates. Nine controls resided in New York City or out-
were retrospectively ascertained from the New York side of New York State at the time of diagnosis of the
State Cancer Registry for the years 1978 through 1988. matched case.
Seven cases with previous cancers were excluded,
resulting in a case population of 171. Potential controls Collection of Data from Records
were randomly selected from live birth records in New Birth certificates were acquired for 118 cases (91%)
York State, excluding New York City, and were one-to- and all controls. Of the cases for whom no birth certi-
one pair-matched to each case by year of birth and sex. ficate was obtained, six were born outside of the US,
Controls were assigned a reference age that was the age and six were not found.
of diagnosis of the matched case, and they had to have Height and weight information was obtained from
survived until that reference age. medical and school records, in addition to the inter-
views. Information on the subject’s primary physicians
Interviews and the schools the subject attended was collected
Telephone interviews were requested from all study during the interview. Consent forms were mailed to all
subjects (cases and controls) who were >18 years of age interviewed subjects or their parents, and were returned
and living. Interviews were also requested from parents by 85% of the cases and 72% of the controls. The doc-
of study subjects (including those ,18 years of age). tors and schools were then traced and asked to complete
While subjects were able to provide some information data forms on the subject’s height and weight at various
about childhood exposures, their mothers or fathers ages. Of the forms mailed, forms were returned with at
were often able to provide more precise and detailed least some information by 52% of the doctors and 82%
information. of the schools. The per cent returned was similar for
The interview focused on the subjects’ social, med- cases and controls.
ical and exposure histories prior to diagnosis/reference.
Issues regarding growth and development were ad- Analysis
dressed with questions about the subjects’ heights and Odds ratios (OR), 95% confidence intervals (CI), and
weights at various ages up to age 18, the age at which P-values were computed by creating conditional logistic
puberty occurred (age at menarche/voice change), and the models in EGRET.16 P-values for trend were calculated
age the growth spurt began. Questions were asked about for continuous variables by including the variables in a
exposure to x-rays, family history of cancer, and birth model in the continuous form. The P-value for the like-
defects. Information on parental occupational expos- lihood ratio statistic reflecting the difference between
ures was obtained by asking whether either parent had the model with and without that variable was interpreted
worked in any of a list of occupations since the subject as the P-value for trend.
was born. In addition, parents were asked about ma- To compare cases and controls with respect to height
ternal exposures during pregnancy. and weight one year before diagnosis/reference, a data-
Interviews were obtained from a case and/or a parent set was created that combined height and weight data
for 130 (76%) of the 171 osteosarcoma cases ascer- from all available sources. This was necessary due to a
tained. Of the 130 cases for whom interviews were large amount of missing data from records and the
obtained, 64 (49%) were from the cases themselves and inability of subjects and parents to recall past height
126 (97%) were from parents. The low number of inter- and weight. Since records were considered to supply
views from the cases themselves was due to 72 (42%) accurate height and weight information, the primary
of the 171 ascertained cases being deceased. Thirty-five source of information used was the school records.
of the 130 cases participating in the study were born When this information was missing, the next source
in New York City or outside of New York State (27%), used was the medical record forms; then the subjects’
making them not comparable to the controls with interviews and finally the parents’ interviews were used.
respect to place of birth. Height and weight data were converted to height-
Interviews were obtained from a subject and/or a for-age and weight-for-age percentiles based on the
parent for 130 matched controls, 118 (91%) from the age- and sex-specific growth standards of the National
controls themselves and 126 (97%) from the parents. Center for Health Statistics using ANTHRO.17 Body
An average of 2.1 birth certificates were selected before mass index (weight/height2 ) was also calculated.
an adequate control was interviewed. Of the 143 birth ANTHRO only calculates the height- and weight-
certificates that were selected but did not result in a for-age percentiles for ages <17. For subjects aged .17,
control interview, neither the control nor the parents no meaningful increase in height after age 17 or 18 was
274 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
TABLE 1 Number of osteosarcoma cases and controls, aged 3–24 years, odds ratios (OR), 95% confidence intervals (CI), and P-values for
trend for height percentiles, weight percentiles and body mass index (BMI) one year before diagnosis, matched and unmatched,a New York
State 1978–1988
a
Unmatched analyses controlled for year of birth and sex.
b
Percentiles based on the age- and sex-specific growth standards of the National Center for Health Statistics17
c
Both height and weight data were available for 45 matched pairs and 137 individuals (64 cases and 73 controls).
assumed, so the age 17 growth standard was applied. allowing more subjects to be included. These analyses
Because of the high likelihood that weight would con- were performed adjusting for the matching variables,
tinue to change with age, no extrapolations were made year of birth and sex.
for the weight data. If height or weight was missing for
the time one year before the diagnosis/reference age,
but was available for the surrounding years, the mean of RESULTS
these two points was used as height or weight one year Due to the matching, the cases and controls had the same
before the diagnosis/reference age. distributions for sex (68% female), age at diagnosis/
Birthweight was abstracted first from the birth certi- reference (mean ± s.d. = 15.1 ± 4.4) and year of birth
ficates, then from the parent interviews, and finally (1966.9 ± 5.4). There were 18 non-white cases (13%)
from the subject interviews. Birthweight was recorded and four non-white controls (3%).
in pounds/ounces and was converted to metric units. The height and weight percentiles and body mass
Birth length was obtained from the medical records, indices one year before diagnosis/reference were each
then from the parent interviews, and last from the categorized into quartiles, based on the subjects included
subject interviews. Birth length was recorded in inches in the respective matched analyses (Table 1). A sig-
and was converted to metric units. Gestational age was nificant trend (P = 0.02) was observed of increasing risk
obtained from the birth certificates. of osteosarcoma with increasing height one year before
Despite the use of multiple sources of information, diagnosis. Neither weight nor body mass index one year
there was still only a limited number of matched sets before diagnosis were associated with osteosarcoma.
in which both the case and the control had information The results of the analyses for height, weight and
on height the year before diagnosis/reference (n = 67), body mass index with the matching broken are also
weight the year before diagnosis/reference (n = 47), shown in Table 1. Again, a significant trend of increas-
and birth length (n = 68). Consequently, in addition to ing risk with increasing height was observed (P = 0.01).
the conditional logistic regression analyses using the In this analysis, 91 cases were compared with 96 con-
matched pairs, unconditional logistic regression ana- trols, as opposed to 67 pairs in the matched analysis.
lyses were performed for the growth and development The likelihood that the association observed between
variables in which the matching was not maintained, height and osteosarcoma could be explained by biases
RISK FACTORS FOR OSTEOSARCOMA IN YOUTHS 275
TABLE 2 Number of osteosarcoma cases and controls, aged 3–24, odds ratios (OR), 95% confidence intervals (CI), and P-values for trend
for body size at birth and gestational age, matched and unmatcheda analyses, New York State 1978–1988
a
Unmatched analyses controlled for year of birth and sex.
b
Low birthweight is defined as ,2500 g (ICD-9).
c
This corresponds to birth lengths of 20 and 21 inches.
in the data was explored. Comparisons were made with fairly similar. In the matched analysis, height informa-
respect to sex, age at diagnosis/reference, year of birth, tion came from school or medical records, as opposed
race, and vital status between the subjects included in to subject or parent interviews, for 81% of cases and
the matched and unmatched height analyses and the sub- 88% of controls. There were 48 pairs (72%) in which
jects excluded from these analyses. Compared to subjects the height data came from the records for both pairs.
not included in the analysis, those included were signi- The trend of increasing risk of osteosarcoma with
ficantly older at diagnosis/reference (16.1 ± 4.2 versus increasing height was borderline significant when the
14.0 ± 4.3 matched data; 15.6 ± 4.2 versus 13.9 ± 4.6 un- matched analysis was restricted to these 48 matched
matched data), had earlier years of birth (1966.4 ± 5.4 pairs (P = 0.08). In the unmatched analysis, height
versus 1967.5 ± 5.5 matched data); (1966.6 ± 5.2 versus information came from records for 84% of cases and
1967.9 ± 5.9 unmatched data), and were borderline sig- 91% of controls. The trend of increasing risk of osteo-
nificantly more often white (95% versus 88% matched sarcoma with increasing height remained significant
data; 94% versus 86% unmatched data). In the matched when the unmatched analysis was restricted to these
analysis, there was no differential selection of cases subjects (P = 0.045; 76 cases, 87 controls).
versus controls according to age or year of birth due to Finally, it was considered whether the cases born
the matching. Furthermore, for those cases and controls in New York City or outside of New York State or the
included in the unmatched analysis, the ages at controls who resided in New York City or outside of
diagnosis/reference were similar (15.5 ± 4.2 for cases, New York State at the reference age introduced a bias.
15.8 ± 4.2 for controls), as were the years of birth The trend of increasing osteosarcoma risk with increas-
(1966.7 ± 5.3 for cases, 1966.4 ± 5.2 for controls). The ing height remained significant after excluding these
trend of increasing risk of osteosarcoma with increas- cases and controls from the matched analysis (P = 0.015;
ing height remained significant when non-whites were 50 pairs) and the unmatched analysis (P = 0.02; 71 cases,
excluded from the analysis (P = 0.01, 60 matched pairs; 94 controls), respectively.
P = 0.03, 82 cases, 93 controls unmatched). No significant associations were observed between
Cases and controls were compared with respect to the birthweight, birth length, or gestational age and osteo-
source of height information, and were found to be sarcoma in the matched analyses (Table 2). In an
276 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
TABLE 3 Number of osteosarcoma cases and controls, aged 3–24, odds ratios (OR), 95% confidence intervals (CI), and P-values for trend
for puberty and growth spurt variables, information from subjects’ interviews onlya, matched and unmatchedb analyses, New York State
1978–1988
a
64 case subjects and 118 control subjects were interviewed. There were 64 matched pairs for which subject interviews were obtained for both
members of the pair.
b
Unmatched analyses controlled for year of birth and sex.
c
Includes those subjects who had reached puberty one year prior to the diagnosis/reference age or earlier. The matched analysis includes pairs in which
both the case and control had reached puberty one year prior to the diagnosis/reference age or earlier.
d
Includes those subjects who began their growth spurt one year prior to the diagnosis/reference age or earlier. The matched analysis includes pairs in
which both the case and control began their growth spurt one year prior to the diagnosis/reference age or earlier.
e
Reached puberty one year prior to the diagnosis/reference age or earlier. In the matched analysis, one or both members of eight pairs did not know
when they reached puberty (age at menarche/voice change).
f
Began growth spurt one year prior to the diagnosis/reference age or earlier. In the matched analysis, one or both members of 19 pairs did not know
when they began their growth spurt.
unmatched analysis, however, there was a borderline To further examine whether puberty or growth spurt
significant trend of increasing osteosarcoma risk with were associated with osteosarcoma, analyses were con-
increasing birth length (P = 0.07) (93 cases, 100 con- ducted on whether the subject had reached puberty or
trols versus 68 pairs in the matched analysis). The had begun their growth spurt at least one year before
primary source of information for birth length was the the diagnosis/reference age (Table 3). No significant
parents’ interviews (80% of cases, 84% of controls). associations were observed.
The primary source of birthweight information was the The remaining analyses reported here were only per-
birth certificates (91% of cases, 100% of controls). formed with the matching maintained, as there were few
Subjects were better than their parents at providing missing values. No significant associations were ob-
information on puberty and growth spurt, as indicated served between osteosarcoma and exposure to medical
by a lower percentage of ‘don’t know’ responses; there- x-rays, history of cancer in first-degree relatives, or
fore, only information from the subjects’ interviews birth defects (data not shown).
was used for these analyses. In the matched analyses, In order to examine the association between antenatal
age of puberty and age growth spurt began were exam- exposures and osteosarcoma, only information collected
ined for pairs in which both the case and control had from the parents was used. This included maternal use
reached puberty or began their growth spurt one year of alcohol and cigarettes, exposure to dental and med-
prior to the diagnosis/reference age or earlier (Table 3). ical x-rays, intake of vitamin or mineral supplements,
No significant associations were observed between and intake of hormones (oestrogens, growth hormone,
osteosarcoma and age of puberty or age growth spurt thyroid hormone, steroids, diethylstilboestrol [DES])
began. and drugs (tetracycline in particular) during pregnancy.
RISK FACTORS FOR OSTEOSARCOMA IN YOUTHS 277
No significant associations were observed between In the unmatched analysis, a borderline significant trend
maternal exposures during pregnancy and osteosarcoma of increasing osteosarcoma risk with increasing birth
(data not shown). length was observed. This association was in the op-
No significant associations were observed between posite direction of that previously found. 9 Given these
parental occupations from birth of the subject to the diag- contradictory findings from two relatively small studies,
nosis/reference age and osteosarcoma (data not shown). the authors conclude there is presently no convincing
Parents were asked about work in various industries evidence for an association between birth length and
and whether they worked with various chemicals or osteosarcoma.
materials. The number of affirmative responses for each No association was found between osteosarcoma and
occupation/exposure was quite small, with a maximum exposure to medical x-rays. Osteosarcomas have been
of eight affirmative responses to handling lead, mercury, reported to arise in a linear dose-response pattern after
chromium, or nickel. radiation therapy, but the amount of exposure received
It has been reported previously that race and maternal was orders of magnitude higher than the medical x-ray
age are potential confounders in these data.15 Adjust- exposures experienced by the subjects in this study.2,3
ment for race and for maternal age did not materially One major limitation in this study was that cases
change any of the results of the analyses reported here. were identified retrospectively for the period 1978–
1988. Problems with recall became exacerbated due to
the long period of time that may have passed since
DISCUSSION the exposures. The number of pairs analysed for a given
Because of the increase in osteosarcoma incidence variable was often diminished due to responses of ‘don’t
around puberty, many investigators have postulated know’ or to missing data from birth, medical or school
that osteosarcoma may be associated with the onset of records. Prior studies on risk factors for osteosarcoma
puberty and may be a function of growth.4–6 The shape in young people were also based on small sample sizes
of the age-incidence curve for osteosarcoma in young ranging from 64 to 85 cases.5,9 Lack of power due to the
people bears a remarkable resemblance to the velocity small sample size could render it difficult to detect
curve for height,18 with a shift to the right of several associations; biases in the selection of subjects included
years. Females develop osteosarcoma at a younger age in a given analysis could result in spurious associations.
than males, consistent with the earlier skeletal develop- A second limitation of this study was the relatively
ment and maturation of females.19 The increase in bone large number of potential controls who were selected
cancer incidence/mortality during the first 20 years of for the study but never located. These potential controls
life is followed by a decline in incidence/mortality after may have changed their address more frequently than
the cessation of growth.3 It is well known that rapidly controls who were located and participated in the study.
growing tissue is particularly susceptible to carcino- If, in fact, the controls in the study were not represent-
genesis.20 ative of potential controls with respect to mobility, and
In both the matched and unmatched analyses, a sig- if mobility was appreciably related to height or to other
nificant association was observed between height one variables of interest (perhaps through a joint relation-
year before diagnosis/reference and osteosarcoma. This ship with socioeconomic status), then results of this
association was found to be robust when various biases study would be influenced by a selection bias.
were considered. However, due to missing data, partic- In summary, the association between height and osteo-
ularly in the matched analysis, the possibility that sel- sarcoma in young people observed in three of four studies,
ection bias influenced the results cannot be completely including the present study, is intriguing. However, as
ruled out. in previous studies, examination of the growth and devel-
Nevertheless, this finding was consistent with earlier opment issues in this study relied on relatively small
reports that osteosarcoma cases are taller at the time of sample sizes. In addition, due to a substantial amount of
diagnosis than controls,5,7 although a more recent study missing data, the possibility that selection bias in-
did not observe this association.9 Weight, body mass fluenced the results cannot be ruled out. Strategies for
index, age of puberty, age growth spurt began, having investigating these issues with larger sample sizes and
reached puberty, and having begun the adolescent higher rates of information retrieval need to be developed.
growth spurt were not associated with osteosarcoma.
A previous study found an association between
osteosarcoma in young people and short birth length.9 ACKNOWLEDGEMENTS
In the present study, no association was found between The authors thank John Kushner, Daniel Bonomo,
osteosarcoma and birth length in the matched analysis. and Raina Josberger for their assistance with the data
278 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY