The bone morphogenic proteins (BMPs) are a member of a large family of growth factors known as the transforming growth factor-β superfamily.¹ Because of their ability to induce new bone formation, BMPs are used clinically as a substitute for iliac crest bone grafting in patients undergoing lumbar fusion surgery. One of these proteins, recombinant human bone morphogenetic protein-2 (rhBMP-2), is licensed in Europe and the United States for anterior lumbar spinal fusion and is delivered via an absorbable collagen sponge carrier. The combination of rhBMP-2 with absorbable collagen sponge is marketed as INFUSE Bone Graft (Medtronic Inc., Memphis, TN). A second product, rhBMP-7, is mixed with bovine collagen and reconstituted with saline and administered as a paste.

In addition to their effect on bone formation, BMPs also have roles in cell lineage commitment, differentiation, proliferation, and apoptosis, and receptors are present in multiple cell types, including cancer cells. A large number of laboratory-based in vitro and in vivo studies have examined the role of BMP in promoting tumorigenesis and metastasis and have yielded conflicting results.²

In the initial published clinical trials of the long-term safety of rhBMP, there seemed to be no association of rhBMP with subsequent cancer risk.^3,4 However, because a postmarketing analysis indicated a nonsignificantly increased risk of pancreatic cancer in patients who received rhBMP, we previously performed a retrospective cohort study in the Medicare population.⁵ Although this analysis found no association of rhBMP with subsequent pancreatic cancer incidence, the study was restricted to 1 tumor type and had a relatively short duration of postsurgical follow-up. In addition, 2 recently published analyses of clinical trial data reported a higher rate of cancers in the rhBMP-treated patients than in those who had undergone bone grafts,^6,7 but only 1 found the differences to be statistically significant.⁷ Given the discordant findings, our goal was to compare the incidence of all cancers after lumbar spinal fusion among a population-based sample of patients treated with rhBMP with the incidence among those who did not receive rhBMP.

MATERIALS AND METHODS

Patients

The cohort was obtained from all Medicare beneficiaries who underwent lumbar fusion surgery between October 2003 (first month in which Medicare provided reimbursement for rhBMP) and December 2008. The relevant files included the Medicare Provider Analysis and Review files, which included claims from inpatient hospitals; the Carrier file, which included claims from physicians and freestanding ambulatory surgical centers; and the Outpatient file, which included claims from institutional outpatient providers. Patients were identified if they had a procedure code for a lumbar fusion operation by one of the following International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) or Current Procedural Terminology, Fourth Edition (CPT-4) codes: ICD-9-CM 81.06, 81.07, 81.08, 81.36, 81.37, 81.38; CPT-4 22558, 22630, 22612.

To obtain complete claims history, patients were excluded if they were not continuously enrolled in fee-for-service Medicare for at least 2 years prior to the index surgery date. Patients who did not continuously participate in Medicare part B, which provides coverage for physician charges and outpatient services, were also excluded because their claims histories may have been incomplete. In addition, patients younger than 67 years were excluded—those younger than 65 years who were enrolled in Medicare because of end-stage renal disease or chronic disability not being representative of the general Medicare population and those aged 65 or 66 years with less than 2 years of enrollment data prior to the spinal surgery.

In addition, to exclude prevalent cases of cancer, as well as the inability to differentiate a newly treated cancer from the treatment of cancer recurrence, any patient with a claim indicating a previous malignant tumor diagnosis during the 2-year period prior to surgery was excluded. A 2-year cutoff was used to maximize the sensitivity of capturing and excluding patients who are long-term cancer survivors. Previous malignant neoplasm diagnoses were identified from one or more ICD-9-CM diagnosis codes in any file. Patients were also excluded if they had one or more ICD-9 diagnosis codes indicating a “personal history of a malignant neoplasm” (V10.00–10.9) or had one or more codes for radiation or chemotherapy.

Measures

Consistent with our previous analysis,⁵ a claim for rhBMP (ICD-9-CM 84.52) on the same day as fusion surgery was used as a surrogate for exposure, which cannot be ascertained directly using Medicare data. This code also includes the administration of rhBMP-7, but the overwhelming majority of procedures use rhBMP-2. Because Medicare did not provide additional reimbursement for these products until October 2003, to reduce exposure misclassification, we limited our study to patients who underwent fusion surgery from this date onward.

A diagnosis of a malignant neoplasm after surgery was the major outcome of interest and was identified by one or more of the ICD-9-CM codes listed in any of the Medicare files in follow-up (see Supplemental Digital Content Appendix 1, available at https://links.lww.com/BRS/A801). We included codes consistent with any of the 26 cancer types that are included in the National Cancer Institute's Surveillance Epidemiology and End Results (SEER) classification,⁸ and followed patients through the end of calendar year 2010. This time interval was consistent with that of 2 recently published systematic reviews and meta-analyses^6,7 and would be more than adequate to detect a potential effect of rhBMP in promoting the growth of subclinical tumors.

Because a single code may not be valid and may reflect “rule out” or other diagnoses that were ultimately found to represent benign diseases, as in our pancreatic cancer analysis, we used 2 secondary definitions of cancer. These included (1) an ICD-9-CM diagnosis code for the same type of cancer on more than 1 date of service and (2) 2 or more ICD-9 diagnosis codes for the same type of cancer on different dates of service and at least 1 procedure code consistent with cancer therapy. The latter codes included site-specific procedure codes (see Supplemental Digital Content Appendix 1, available at https://links.lww.com/BRS/A801) as well as procedure codes for radiation therapy and chemotherapy. Per Centers for Medicare & Medicaid Services policy, because of patient confidentiality issues, any cell sizes with a frequency less than 11 were suppressed.

In addition to data about exposure and outcomes, we included potential confounders such as age (at the time of index surgery), sex, race (Caucasian, African American, other), and length of follow-up. The presence of comorbid conditions was measured using a previously validated index, which includes diagnoses present in Medicare Provider Analysis and Review, Outpatient, and Carrier files.⁹ In order to differentiate complications from comorbidities, only diagnoses that were present from 2 years through 30 days prior to date of surgery were included. Using this algorithm, a weighted score was assigned for each individual.

Analysis

Patients were followed from the date of index lumbar fusion surgery until the diagnosis of cancer, death, disenrollment, or end of the study period (December 31, 2010). Individuals who underwent an initial operation without rhBMP and a subsequent procedure with rhBMP were followed in the nonexposed group to the date of the second surgery and thereafter in the exposed group.

The association of demographic variables, comorbid conditions, and rhBMP administration with each cancer site and with overall cancer risk was examined using the primary cancer definition (one or more diagnoses). Chi-square analysis was used to determine statistical significance. In addition, to account for variable length of follow-up, a series of univariate Cox proportional hazards models were constructed to examine the association of rhBMP administration and risk of individual cancer types.

The independent association of rhBMP administration and cancer risk was then determined in multivariable analyses using Cox proportional hazards regression. In all models, covariates included demographic factors (age, race, and sex, if appropriate for that cancer), comorbidity score, and rhBMP administration.

Finally, to determine whether the observed incidence of cancer was different than expected from the general population, we used the standardized incidence ratio (SIR). The SIR determines the number of observed cases divided by the number of expected cases in both rhBMP-exposed and unexposed patient groups. The expected numbers were obtained by applying age- and sex-specific incidence rates for all cancers from the SEER Program⁸ to the corresponding person-time.

RESULTS

We initially identified 295,493 patients who underwent lumbar spinal fusion during the time period of interest. From that sample, patients were excluded for the following nonmutually exclusive reasons: not continuously enrolled in Medicare parts A and B (n = 69,398), enrolled in Medicare health maintenance organizations (n = 53,107), age less than 67 years (n = 56,699), and previous cancer diagnosis (n = 29,765). The remaining 146,278 patients were the subject of this analysis.

Characteristics of the cohort are shown in Table 1. The mean age was 74.5 ± 5.1 years, 66.5% were female, and 93.7% were Caucasian. Most patients had comorbidity scores of 0 or 1. A code for rhBMP administration was documented in 15.1% of surgical procedures. Compared with others, patients who received rhBMP were younger, somewhat more likely to be female or Caucasian, and had higher comorbidity scores. The proportion of patients who received rhBMP generally increased during the study period. The average length of follow-up was 4.8 ± 1.5 years (range: 1.23–7.25 yr) in the rhBMP-treated patients and 4.4 ± 1.3 years (range: 1.16–7.25 yr) in others. Death rates during the follow-up period were 3.27% in the rhBMP group and 3.45% in others.

One or more diagnosis codes for cancer were documented in follow-up in 24,481 patients including 21,079 in the non-rhBMP group (17.0%) and 3402 in the rhBMP-treated patients (15.4%). Consistent with the known incidence of cancers in the older US population,⁸ the most commonly recorded cancer diagnoses were prostate, breast, lung, and colon and rectum.

Using a proportional hazards model, we determined the risk of cancer as a whole and within individual tumor types (Table 2). Overall, there was no association of rhBMP administration with cancer incidence (hazard ratio: 0.98, 95% confidence interval [CI]: 0.95–1.02). Similarly, when individual cancer sites were considered, there were no significant differences between the 2 groups. In an adjusted analysis, the risk of cancer was similar between rhBMP-treated patients and others (hazard ratio: 0.99, 95% CI: 0.96–1.03). As with the unadjusted analysis, there were no significant differences among specific cancer sites.

In a secondary analysis, we considered 2 other definitions of incident cancer. Using a criterion of a diagnosis on 2 or more different dates, we identified 18,942 cancer cases, with similar frequencies in rhBMP-treated (12.0%) and other patients (13.1%) (see Supplemental Digital Content Appendix 2, available at https://links.lww.com/BRS/A801). The overall cancer risk was similar in unadjusted (hazard ratio: 0.97, 95% CI: 0.93–1.01) and adjusted (hazard ratio: 0.98, 95% CI: 0.94–1.02) proportional hazards models. The risk was also similar among individual tumor types. Using the most stringent definition of 2 or more diagnoses and cancer treatment codes, we identified 14,362 cases with an incidence of 8.7% and 10.0% in rhBMP-treated and untreated patients, respectively (see Supplemental Digital Content Appendix 3, available at https://links.lww.com/BRS/A801). With this definition, there was a somewhat lower overall cancer risk with rhBMP use in both unadjusted (hazard ratio: 0.94, 95% CI: 0.89–0.98) and adjusted (hazard ratio 0.95, 95% CI: 0.90–0.99) proportional hazards analysis. When individual sites were examined, there was a lower risk of brain tumors in rhBMP-treated patients in unadjusted (hazard ratio: 0.65, 95% CI: 0.46–0.90) and adjusted (hazard ratio: 0.64, 95% CI: 0.46–0.90) models. No other significant differences were observed.

In the SIR calculations, using the primary case definition of 1 diagnosis code, the incidence of all cancers combined was higher than expected in both the rhBMP-treated and untreated groups. For rhBMP-treated patients, the SIR was 177.79 (95% CI: 172.00–185.00) and for untreated patients, the SIR was 177.00 (95% CI: 175.00–179.00). Consistent with more stringent diagnostic criteria, the SIRs were lower for both rhBMP-treated and untreated patients, using the other 2 case definitions. For a diagnosis code on 2 or more service dates, the SIR in the treated patients was 135.26 (95% CI: 130.00–141.00) and in untreated patients was 136.30 (95% CI: 134.00–138.00). For the criteria of 2 or more diagnosis codes as well as treatment, the SIRs closely approximated that of SEER. The SIR in rhBMP-treated patients was 96.99 (95% CI: 92.00–101.00) and in untreated patients was 101.66 (95% CI: 100.00–103.00).

DISCUSSION

rhBMP is commonly used as an adjunct to orthopedic surgical procedures to promote bone growth and is used as an alternative option to bone grafting. Although most safety reports have focused on local events such as bony overgrowth, wound healing, and neurological events,^3,10–12 there is at least a theoretical concern about the increased risk of malignant tumors among patients treated with rhBMP. Cancer case ascertainment through analysis of clinical trial data is limited by issues of power and sample size as well as relatively short duration of follow-up. In this study, which included a population-based sample of a large number of surgical patients with a median follow-up of more than 4 years, we did not demonstrate an increased risk of malignancy across multiple tumor sites.

In theory, given the presence of BMP receptors of a variety of tumors, including lung, pancreatic, renal, brain, osteosarcoma, ovarian, breast, and prostate cancers,² there is a potential risk of BMP in promoting tumorigenesis as well as metastases. In different models, BMP promotes angiogenesis,^13,14 cancer cell growth,¹⁵ bone metastases,¹⁶ and cancer cell motility and invasiveness.¹⁶ However, BMP can also act as a growth and proliferation inhibitor and thus have antineoplastic effects.^16,17 The only individual tumor type for which we found a significant difference in incidence was brain tumors, where there was a somewhat higher incidence in patients who were not treated with rhBMP-2. This finding is consistent with previous studies that showed an antitumor effect of rhBMP-2.^18,19 Although clinical trials to date have failed to show a conclusively increased risk of malignancy, given the theoretical risk of cancer progression, rhBMP is not indicated in the vicinity of a resected or extant tumor in patients with active malignancy or undergoing treatment of malignancy.²⁰ Our study results included the entire spectrum of SEER cancers and a relatively long follow-up period. However, if a potential risk of BMP is in mutagenesis, a longer observation period may be required to provide additional clinical evidence against the malignant potential of rhBMP.

As with other analyses, our methodology has a number of strengths and limitations. The strengths of the study include the large sample size, the representation of diverse practice sites, and the ability to follow patients for an average of 4.7 years after spinal surgery. Limitations include that the sample was restricted to patients aged 67 years and older, and generalizability of the findings to younger patients is uncertain. However, most cancers increase in incidence with age, and the older population accounts for a significant proportion of patients undergoing spinal fusion surgery. The study was also limited to fee-for-service beneficiaries who were enrolled in Medicare part B. The diagnoses of previous and subsequent cancer were ascertained through ICD-9-CM codes, which are used for billing purposes and not research. However, the results were consistent across 3 different definitions of incident cancer, increasing the face validity of the findings. The study lacked data on other risk factors such as smoking, alcoholism, obesity, and family history of cancer, which are either significantly under-reported or absent in Medicare data. As the study was observational and not randomized, there may have been systematic differences between groups that could have biased the results. However, in our previous analysis of pancreatic cancer risk,⁵ we found on medical record review that there was no association of rhBMP use with established cancer risk factors such as smoking and obesity. Despite the large sample size, the study lacked precision to measure the incidence of rare cancers in the elderly such as testicular cancer and Kaposi sarcoma. Finally, because we used a procedure code as a proxy for exposure to rhBMP-2, misclassification could be a concern. However, a previously conducted chart review study demonstrated a specificity of the code for rhBMP-2 (as opposed to other forms of rhBMP) of 95% and a positive predictive value of 100%.⁵

In summary, in this large cohort of older patients, the study provides evidence that treatment with rhBMP at the time of lumbar spinal fusion surgery does not increase the risk of subsequent malignancy. The results should be reassuring to providers and patients.

Key Points

• Use of rhBMP has been postulated to increase the risk of subsequent cancer.
• In a population-based sample of Medicare beneficiaries undergoing lumbar spinal fusion, we found no association of rhBMP administration with subsequent risk of cancer.
• The lack of risk of cancer was consistent across all tumor types and different definitions of cancer incidence.

Supplemental digital content is available for this article. Direct URL citations appearing in the printed text are provided in the HTML and PDF version of this article on the journal's web site (www.spinejournal.com).

Supplemental Digital Content

• BRS_2013_07_18_COOPER_205752_SDC1.doc; [Word] (355 KB)