There are few established osteosarcoma risk factors apart from early exposure to high-dose radiation, and Paget's disease, hereditary retinoblastoma, and Li–Fraumeni Syndrome (Miller et al, 1996). The bimodal age–incidence curve reflecting peak rates occurring both in adolescence and in older age suggests two separate aetiologies. Enhanced carcinogenic susceptibility during the adolescent growth period is suggested by higher radiogenic bone cancer risk among children than adults, and the characteristic development of childhood tumours in the long bone epiphyses of the lower limbs (Fraumeni, 1967). An excess osteosarcoma risk in larger compared with smaller dog breeds may be consistent with this hypothesis (Withrow et al, 1991). Higher male than female incidence rates in puberty, and the early age at which osteosarcoma incidence first peaks – 10–14 years of age in girls, and 15–19 years in boys – may indicate the importance of accelerated growth and hormonal differences, and raise the possibility that very early-life exposures play a role as well (SEER, 2005).

In this study, we focused on the associations of osteosarcoma with factors related to growth and development from the in utero period through puberty and adolescence.

Materials and methods

Study population

Participants were drawn from orthopaedic surgery departments in 10 US medical centres between 1994 and 2000 (Massachusetts General Hospital, Boston, MA, USA; Creighton University/St Joseph's Hospital and University of Nebraska, Omaha, NE, USA; Children's National Medical Center and Washington Hospital Center, Washington, DC, USA; University of Chicago and Rush Presbyterian St Luke's, Chicago, IL, USA; University of Florida, Gainesville, FL, USA; University of California, Los Angeles, CA, USA; Cleveland Clinic, Cleveland OH, USA). Cases were patients with newly diagnosed primary osteosarcoma admitted for evaluation of eligibility for limb salvage surgery. Controls were orthopaedic patients from the same departments with benign tumours (26%), or non-neoplastic conditions such as inflammatory diseases, cysts, and trauma (ICD-9-CM codes 289.1, 277.8, 354.0, V711, 682.6, 714.3–756.59, 810.0–885.0, and 959.7). Nurse coordinators were responsible for control selection by identifying the next patient matching the case on age (+/−5 years), sex, hospital, and distance from the respective medical centre based on the participant's residential zip code. Pathology reports from surgery and/or biopsy were obtained for all cases and for controls for whom this documentation was applicable to confirm diagnosis, whereas hospital medical records confirmed diagnoses for controls without a pathology report (with conditions such as injury, reconstruction, revision, or pain).

Institutional review boards at each of the medical centres approved this study, and informed consent, which included a check list for a questionnaire, blood draw, and toenail collection, was obtained from all participants. Of eligible cases, 13.6% were not enrolled and a further 3% declined to participate in any of the study components. Of those enrolled in the study, 93.5% of cases and 98% of controls completed interviews. Reasons for non-participation in the questionnaire component included refusal, death, extreme illness following surgery and failure to return for follow-up care. The present analysis was restricted to 169 cases and 144 controls less than 40 years of age to focus on the aetiology of the adolescent-young adult peak in osteosarcoma incidence.

Exposure and covariate information

Interviews ascertaining information on growth and development, physical activity, and medical history were conducted in the hospital, clinic, or ward with cases after surgery, and after surgery, or other therapy in controls. A supplemental interview on pregnancy exposures was conducted with parents of participants less than 20 years of age, whereas participants over 20 were asked directly.

Participants 9–21 years of age were asked to give consent for growth records acquisition. Of 109 cases and 91 controls, respectively, all of the 78 (71.5%) and 57 (62.6%) participants who reported a regular health-care provider gave permission for contact, and a total of 402 growth records were obtained for 49 cases and 46 controls (average of 4.2 records per participant). Growth records validated participants' self-reports of height (shorter, about the same, and taller) compared with their peers at various ages. Among the combined cases and controls, mean height from records was 9% lower and 1.5% higher among participants who reported being shorter and taller, respectively, compared with being about the same height as their peers at age 9 or 10. Similarly, recorded height was 7% lower and 4.5% higher at age 12 or 13, and 3.5% lower and 6% higher at age 15 or 16. Results were similar when evaluated separately for cases and controls.

US born participants were asked for their consent to acquire birth records from state vital records offices with 97% (n=154) of cases and 99% (n=139) of controls providing permission. For states where at least six participants were born records were obtained for a total of 289 participants. Equal proportions of cases (55%; n=87) and controls (55%; n=78) were missing information in the birth records for pregnancy length. Pregnancy complications were missing for 54% (n=85) of cases and 59% (n=83) of controls, and of those with these data, complications were rare (two cases/five controls). Birth weight as reported by the participant or mother (mean=3392; s.d.=589) and from records (mean=3389; s.d.=556) were highly correlated in those with both sources of information (r=0.95; n=176). Therefore, reported birth weight was used when birth records were unavailable.

Statistical analysis

Current height for all participants was converted to percentiles based on sex- and age-specific growth standards provided by the National Center for Health Statistics (CDC, 2000). Unconverted height was analysed in participants 21 years of age or older who were assumed to have attained their final height. Tertiles were based on sex-specific distributions in the control group. Birth weight percentiles were based on gestational age and sex (Oken et al, 2003) and quartiles were based on the control distribution. Unconditional logistic regression models including age and sex were used to estimate odds ratios (OR) and 95% confidence intervals (CI). Analyses were based on male and female subjects combined to increase the sample size but are presented separately when findings differed by sex. Further inclusion of study centre, geographic ring, education, and family income did not affect the estimates. Linear trends were assessed using orthogonal polynomial contrasts (Winer, 1962) or by including the continuous variable in the model. The data analysis was generated using SAS/STAT software (1999).


Similar proportions of cases (64%) and controls (62%) were under 20 years of age, and they were comparable in sex and race/ethnicity (Table 1). Cases had a lower combined family income. There were no trends in age- and sex-adjusted osteosarcoma risk with increasing participant's education, or with mother's or father's education (data not shown).

Table 1 Distributions of characteristics for osteosarcoma cases and controls

Neither height percentile nor absolute height was associated with osteosarcoma risk (Table 2). In participants 21–39 years of age, attaining final height at a younger age was associated with a reduced risk in female subjects (OR=0.53; CI=0.15–1.8 for <17 vs 17+ years), but with an elevated risk in male subjects (OR=7.1 CI=1.6–50 for <19 vs 19+) (P-value for interaction=0.01). The association of participant's height percentile with risk showed no consistent pattern in any subgrouping according to mother's height (data not shown).

Table 2 Age- and sex-adjusted ORs and 95% CIs for participant's and parent's height and osteosarcoma risk

Cases appeared less likely to be shorter than their peers at ages 9–10 and 12–13, and more likely to be taller at ages 15–16, compared with being about the same height (Table 2). The reduced ORs for being shorter at younger ages were consistent in male and female subjects, but the elevated OR for being taller at age 15–16 resulted from an OR of 2.4 in male subjects and 1.0 in female subjects. Among male subjects, controls were more likely than cases to have started shaving, and developing pubic hair early in male controls followed a similar pattern (Table 3). Age at menarche was not associated with risk. Results for the puberty variables were unchanged with adjustment for height percentile (data not shown).

Table 3 Age- and sex-adjusted ORs and 95% CIs for development measures and osteosarcoma risk

There was no consistent association between sports participation at various ages during childhood and osteosarcoma risk (data not shown). Comparing frequent with less frequent activity (4+/week vs <4/week), the OR were generally elevated in female subjects, in particular, at ages 15 or 16 (OR=2.9; CI=1.2–7.4), whereas there was no association in male subjects (data not shown). Excluding controls whose condition on study entry was fracture attenuated the elevated estimate for more frequent activity in female subjects (OR=1.8; CI=0.8–4.6).

Excluding controls whose condition on study entry was a fracture or fibroma, the OR for ever fracturing a bone was 0.68 (CI=0.41–1.1); with further exclusion of fractures occurring within 2 years of questionnaire completion the OR was 0.65. The site of prior fracture was examined to identify whether for any case, the tumour occurred in the same bone; they matched in only two cases (within 1 year and 15 years of diagnosis).

Diagnostic or therapeutic radiation before the present illness was not associated with osteosarcoma risk (OR=1.2; CI=0.75–1.9). Radiation exposure was mainly in the form of routine medical X-rays (88% of cases and 87% of controls).

High birth weight was associated with an increased osteosarcoma risk (Table 4) that was similar when stratified by age <21 vs 21+ years (data not shown), but stronger among female subjects (OR=7.2; CI=1.7–50) than male subjects (OR=2.9; CI=1.0–9.1). There was no trend in risk with gestational age (P=0.63), and adjustment for gestational age did not change the OR for birth weight. Birth weight percentiles confirmed the increased risk among heavy babies (highest vs lowest quartile OR=4.6; 95% CI=1.4–16.4). Longer birth length appeared protective though the association was not linear or statistically significant. The estimates for both birth weight and length were slightly stronger with mutual adjustment (OR=7.6 for birth weight >4000 g vs 3000–3500 g; OR for birth length=0.40, 0.32, and 0.34 for 20, 21, and 22+ inches, respectively, vs <20 inches). There were no associations between risk and mother's primary job, smoking, and alcohol intake during pregnancy and father's smoking during pregnancy (data not shown).

Table 4 Age- and sex-adjusted ORs and 95% CIs for pregnancy and birth characteristics and osteosarcoma risk


Osteosarcoma appears to be positively associated with bone growth, based primarily on the rapid rise and fall of incidence rates from adolescence into young adulthood, and buttressed by the typical occurrence of this tumour (approximately 70% (Miller et al, 1996)) in the long bone epiphyses and the strong positive association in canines with breed size (Withrow et al, 1991), and in particular, height (Ru et al, 1998).

Findings for growth and development in human populations, however, are equivocal. There is some evidence (Fraumeni 1967; Gelberg et al, 1997; Cotterill et al, 2004) of cases being taller than controls, although the differences often derive from inconsistent subgroup findings defined by age, gender, or anatomic site. Other data suggest no differences in height or average growth rate (Operskalski et al, 1987; Pui et al, 1987; Glasser et al, 1991; Buckley et al, 1998). Our data show no association with height, but suggest that cases are less likely to be shorter than their peers before and during the early years of adolescent growth. The positive association we observed with birth weight may also be consistent with earlier growth being adverse. The few previous studies of birth size, like those of height, have had conflicting results (Operskalski et al, 1987; Gelberg et al, 1997; Buckley et al, 1998). The increased risk we observed in the lowest birth length category when adjusted for birth weight could be a chance finding given the lack of overall trend, the increased risk with elevated birth weight, and the inconsistent findings from prior studies (Operskalski et al, 1987; Gelberg et al, 1997). However, our estimate is similar to that of an earlier study (RR for 21.5 vs 19.5–20.5 inches=0.59) (Operskalski et al, 1987).

The development of secondary sexual characteristics has also been a focus of studies of growth and development in relation to osteosarcoma risk (Gelberg et al, 1997; Buckley et al, 1998). We found no associations among female subjects, but earlier development of facial and pubic hair appeared protective in male subjects. Like the growth data, previous findings are inconsistent with respect to timing of sexual development (Gelberg et al, 1997; Buckley et al, 1998).

Together with those of earlier studies, our results may suggest that there is an underlying relationship of growth and maturation patterns with osteosarcoma risk that is likely too weak to explain a meaningful portion of the remarkable age–incidence curve. The shape of this curve, instead of resulting from growth characteristics, may simply be a weakly correlated marker of the true aetiologic event. The abrupt rise and decline in incidence would be consistent with exposure occurring at a specific common time before the peak. Because of the young age of cases, and resultant requirement for a short common exposure interval, the period during in utero development seems likely. In fact, the age–incidence curve is similar to that of vaginal clear-cell adenocarcinoma, caused by in utero diethylstilbestrol exposure. In that malignancy, the teratogenic/carcinogenic error occurs in the foetus, but is not manifest until the normal hormonal development and maturation of the reproductive system following puberty. Similarly, carcinogenic events related to in utero bone development may only become manifest during their major development and maturation in adolescence. We saw no evidence of adverse effects of in utero exposure to alcohol or tobacco. Given the above discussion of growth patterns, perhaps the most likely intrauterine exposures are those with independent effects on subsequent growth and development. Prime candidates for consideration might be gene variants responsible for foetal bone development, and/or environmental exposures such as nutritional factors or infections.

Another commonly suggested risk factor for osteosarcoma is prior bone trauma, although epidemiologic studies have not found evidence of an association with the exception of one study (Operskalski et al, 1987). We attempted to assess this based on histories of prior fractures, and by frequency of sports participation although neither was associated with excess risk, and prior fracture was associated with decreased risk. Excluding conditions for which prior fractures or other trauma might be a risk factor did not alter our results, but residual bias is possible. However, because only one case reported a fracture of the same bone more than a year before the diagnosis of osteosarcoma, if such trauma were a risk factor, it would seem to account for only a very small portion of disease.

This investigation has several strengths, including a high cooperation rate with an in-person interview conducted by trained interviewers, and a control group of orthopaedic patients, minimising the opportunity for selection bias, particularly for early-life exposures. There are several limitations as well. The case series included only those patients commonly treated by orthopaedic surgeons (those with limb tumours), with other sites underrepresented, for example, skull, jaw, rib, and vertebral column. Because the hospitals are academic centres with a high rate of referral patients, the opportunity for referral bias exists. For this reason we stratified the controls to the distribution of cases according to hospital, and residential distance from the hospital. Risk estimates were virtually unchanged with adjustment for these variables, as well as for family income. The use of orthopaedic controls has potential disadvantages as well. If the factors we investigated are associated with risk for orthopaedic conditions as well as osteosarcoma, we might fail to identify differences. To mitigate this possibility we included a wide variety of diagnoses in the control group. Finally, the relatively small number of cases limited statistical power, however, owing to the rarity of this tumour, our study includes more cases than all but one of previously published studies (Cotterill et al, 2004). Larger studies, however, are needed, particularly to investigate subgroup risks.

In conclusion, our results together with those of previous studies do not suggest a clear growth and development pattern that could explain the dramatic young-age peak of this malignancy. However, there are enough signals to speculate that growth and development characteristics may be markers of true aetiologic events to which they are weakly correlated. Given the sharpness of the young incidence peak, the most likely timing of these events is in-utero development. A more concerted effort to evaluate this aspect of aetiology is warranted.