Risk factors for bone metastasis in patients with primary lung cancer

Supplemental Digital Content is available in the text


Introduction
Bone metastases (BM) are prevalent among lung cancer (LC) patients. [1] Around 30% to 40% of LC patients develop BM in the disease course. [2] BM would cause severe complications, like pathological fractures, spinal cord compression, hypocalcemia and other skeletal-related events (SREs). [3] Each of them would bring about a rising cost of healthcare and the impaired quality of life. [4,5] Skeletal metastases account for approximately 350,000 deaths in the United States every year, [6] and nearly 3 times this number if patients in the European countries and Japan are also included. Early treatments are effective to lower the incidence of complications and medical expenses. [5] Applications of bisphosphonates and denosumab might relieve suffering and save money for every LC patients. [7][8][9] It would be a turning point for every patient's wellbeing if we can find out risk factors for BM/SREs. We have many high-tech types of equipment that can find out bone lesions of BM/SREs, but no one could identify the latent hazard. Therefore, it is very imperative for us to identify risk factors of BM/SREs before things get worse than before. Thankfully, many researchers have done plenty of work on this topic.
There have been some studies of risk factors for BM in lung cancer. In 1999, Kobayashi et al [10] identified that the aminoterminal propeptide of Type I collagen (PINP) and carboxyterminal telopeptide of Type I collagen (ICTP) correlated with BM and survival time. They appeared to be of great value for the prediction of BM. In 2005, Brown et al [11] found that bone biomarker levels were an indicator of SREs, disease progression and death in patients with BM secondary to nonsmall cell lung cancer (NSCLC). In the next year, Coleman et al [12] published an article that the bone resorption marker NTX provided predictive information in BM patients. They found that high NTX levels (≥100 nmol/mmol creatinine) were related to high risk of SREs and disease progression compared with low NTX levels (<50 nmol/mmol creatinine).
However, these studies all focused on a few factors. Previous studies have shown that expression of some biochemical compounds (e.g., bone sialoprotein, osteopontin, and Ntelopeptide of type I collagen (NTX), serum cross-linked carboxyterminal telopeptide of type I collagen (ICTP) and the aminoterminal propeptide of type I collagen (PINP)) strongly associated with development and progression of BM in lung cancer patients. [10][11][12][13][14][15][16] It is not enough to involve these biomarkers in predicting the incidence of BM/SREs. We need more evidence to recognize factors to apply them in identifying the high-risk population. This systematic review intended to help clinicians generate a basic conceptual structure to better understand the relationship between potential risk factors and BM/SREs.

Electronic search
We applied PubMed, MEDLINE, Web of Science, EMBASE, the Cochrane Library (Cochrane Database of Systematic Reviews) and the Cochrane Central Register of Controlled Trials (CENTRAL) (from January 1990 to November 2017) to search the relevant literature without any language restrictions. We used predefined keywords to run searches: "primary pulmonary neoplasm," "risk factors," and "bone metastases." We described search strategy for PubMed in detail in Supplementary File 1, http://links.lww.com/ MD/C768. Primary and secondary outcomes should be BM and SREs. We summarized the effect estimates of risk factors and used random-effect models to pool the estimates if the outcomes and characteristics in studies were comparable. The quality of the study was assessed using the Newcastle-Ottawa Scale and the Cochrane Collaboration tool. [17]

Selection criteria and data collection
We included case-control, cohort studies, randomized controlled trials (RCTs) and systematic reviews in adults and elderly patients with primary lung cancer. Descriptions of risk factors are adequate. The primary and secondary outcomes are BM and SREs separately. BM is defined as one or more radiographically confirmed bone metastases. Diagnostic methods include plain radiography, myelography, MRI, CT, radionuclide bone scanning (scintigraphy with technetium-99m-labeled diphosphonates), single-photon emission CT and positron emission tomography. [18] SREs include the first SRE, time-to-the first SRE, all SREs, SRE-free survival, skeletal progression and related death (our protocol described SREs in detail [17] ).
We retrieved information for eligible studies (the PRISMA guidelines, www.prisma-statement.org) [19] using a predefined procedure and collection form. [17] The heterogeneity of study design and outcomes did not fit for a meta-analysis, so we undertook a systematic narrative review to synthesize potential risk factors of BM/SREs. Experimental procedures were approved by the Institutional Review Board of the Fourth Military Medical University.

Study characteristics
Through database searching, we identified 13,148 references. We used Endnote (Microsoft, Redmond, WA) to remove 11,192 duplicates. Then 2 review authors (W-WS, Y-TW) separately examined 1956 publications. After exclusion of inconsistent titles/abstracts, getting full-text of publications, uniting different articles of the same study together, and analyzing full-text based on eligibility criteria, we listed 12 final selected publications. After assessment of eligibility, 6 records were duplicate reports from the same study population; 18 records were nonlung cancer case/control groups; 8 references had no control group; 9 studies did not conform to the specified study design. The flowchart (Fig. 1) presented the specific selection processes. [12][13][14][20][21][22][23][24][25][26][27][28] [12] shared the same database of a randomized controlled trial [29] with Hirsh et al, [22] their purposes and populations were varied. These studies had different limits of NTX levels among patients with placebo on zoledronic acid. [12,22,29]

Risk of bias assessment
Two investigators (SW, XB) separately assessed the risk of bias (ROBs) using the Newcastle-Ottawa scale [30] and the Cochrane Collaboration tool [31] to value observational studies and RCTs, separately. We contacted authors of publications with openended questionnaires for additional information if some data were needed.
The publication of Lee et al [28] got the least score using the Newcastle-Ottawa Scale. Although both articles of Coleman et al [12] and Hirsh et al [22] came from one original study, [29] we treated them as 2 studies because they had diverse purposes and populations. However, bias from selective reporting of outcomes was likely to occur in the study of Hirsh et al. Table 2 presents ROB ratings and scores for included observational studies, which indicated the need for high-quality articles.

P-stage III.
From the study of 374 NSCLC patients, Wang et al [24] evaluated that P-stage III was a high-risk factor influencing bone metastasis. Univariate analysis suggested that P-stage III (P = .007) was an independent factor for BM comparing P-stage I + II. A multivariate analysis found that patients with P-stage III had a higher risk for bone metastasis (HR: 2.410; 95% CI: 1.265-4.593; P = .008) than P-stage I + II. There were no significant differences between patients with Pstage I disease and patients with P-stage II disease (HR: 1.089; 95% CI: 0.482-2.461; P = .838). All above suggested P-stage III was related to high risk of BM in NSCLC patients.

No history of EGFR-TKIs therapy.
A cohort study [21] reported a higher risk of all SREs in patients with no history of treatment with epidermal growth factor receptor (EGFR) tyrosine kinase inhibitors (TKIs)   Table 1 Characteristics of studies related to bone metastases/skeletal-related events in lung cancer patients.   [20] found the number of BM at the time of the initial diagnosis influenced the occurrence of SREs. They reported that single BM (vs none BM) (OR: 3.08, 95% CI: 1.60-5.94; P < .001) and multiple BM (vs none BM) (OR: 4.27, 95% CI: 2.66-6.86; P < .001) associated with an increased risk of SREs. For time-tothe first SRE, it showed similar effects of single BM (HR: 3.00, 95% CI: 1.68-5.35; P < .001) and multiple BM (HR: 4.43, 95% CI: 2.91-6.76; P < .001). But for SRE-free survival, only the result of multiple BM was statistically significant (HR: 1.80, 95% CI: 1.40-2.31; P < .001) ( Table 4). Then, in 2016, Ulas et al identified the number of BM as a significant factor in predicting SREs (OR: 3.05), such as the need for radiotherapy and malignant hypercalcemia.
3.2.9. Elevated BAP levels. Coleman et al [12] reported the effect of bone-specific alkaline phosphatase (BAP) on the risk of SREs among patients treated with zoledronic acid. Generally, elevated BAP levels (≥146 U/L) correlated with a higher risk for SREs irrespective of outcomes. However, the correlation was weak for All SREs (unadjusted RR: 1.29, 95% CI: 0.89-1.88; P = .180). Table 4 Association between exposure to potential risk factors and risk of skeletal-related events.

Exposures
Outcomes  [26] followed up 835 NSCLC patients and found the presence of bone metastasis at diagnosis was a predictive factor of SREs (OR: 12.6). The most common SREs were the need for radiotherapy (43.2%) and malignant hypercalcemia (17.6%). The median time to first SRE was 3.5 months at the median follow-up of 17 months.
3.2.11. Three or more metastatic vertebrae. An article [27] revealed lung cancer patients with 3 or more metastatic vertebrae had a great risk of developing metastatic spinal cord compression (MSCC) than those who have up to 2 involved vertebrae (OR:6.1, 95% CI:2.5-15.1, P < .001).

Discussion
Bone metastasis and SREs are frequent and burdensome among lung cancer patients. Although we can apply multiple approaches to diagnose BM/SREs, clinicians need comprehensive and systematic information to predict the risk factors of BM/SREs and to decide the suitable strategies for preventing and treating disease. Therefore, it is imperative to identify the potential risk factors of BM/SREs from previous studies. Recently, new predictors provided directions to prevent BM/SREs; we need an accurate prediction model to estimate risk.
In an exploratory cohort analysis published in 2005, Coleman et al [12] suggested that elevated NTX or BAP levels positively correlated with SREs. In 2006, another retrospective casecontrol study reported that BSP protein expression was positively associated with higher risk of BM progression and may be a useful predictor in identifying a high-risk population in the primary resected NSCLC. [14] In a retrospective cohort study published in 2009, Sekine et al [20] found that multiple BM was strongly correlated with increased risk of SREs in advanced NSCLC. They also suggested that male and poor PS were additional predictors for SRE-free survival. In 2010, Zhang et al [13] confirmed that positive BSP expression in the primary resected Chinese NSCLC positively associated with a higher risk of BM. It was coincident with Papotti et al, [14] who concluded that there was a positive correlation between BSP expression and BM. Besides, this cohort study also revealed that the T4 stage and N3 stage were independent risk factors for BM. In a retrospective cohort study published in 2011, Sun et al [21] found that patients ever-smoking had a significantly higher risk of SREs than neversmokers. Another cohort study from China indicated that decreased blood calcium levels at initial care associated with an increased risk of BM versus normal levels in 2012. [23] Besides, in 2014, Lee et al [28] showed that increased serum CEA levels could be a predictor of increased bone metastatic potential in stage IV lung cancers.
Thus, despite the individual researches reporting developments of predicting BM/SREs, our review set out to provide a pooled analysis of the expected improvements. We ran this systematic review and evaluated articles published from 1990 to 2014. There seemed to be progressive developments in risk factors of BM/SREs in lung cancers. We found that T/N staging and positive BSP expression positively associated with the occurrence of BM. Moreover, results showed that ever-smoking and multiple BM correlated with the proportion of SREs.
The literature supports these associations. Our results are consistent with the large-scale correlative literature of bone turnovers in patients with solid tumors (including NSCLC) published in 2005. [11] Brown et al [11] found that recent bone turnover assessments (e.g., NTX, BAP) were better indicators for SREs than baseline bone marker levels. Then in 2013, Sutcliffe et al [18] reported that there was an increased likelihood of SREs with ever-smoking, lack of history of therapy with EGFR TKIs, poor ECOG status, and nonadenocarcinoma. Their review found that the greater the number of BM, the higher was the risk of SREs. In an exploratory analysis published in 2013, Lipton et al [32] showed that biomarkers of bone metabolism could provide insight into ongoing rates of bone destruction among patients with malignant skeletal diseases. [33] However, from their previous studies, inconsistent results gave us confused interpretations about baseline NTX levels. [34,35] In this retrospective analysis, among patients with BM from prostate cancer, breast cancer, NSCLC or other solid tumors who received zoledronic acid treatment, Lipton et al [32] suggested that NTX elevations not precede SREs. By contrast, our study focuses on lung cancers, despite mixed solid tumors in previous studies.
We admit that our study has some limitations. Originally, in our protocol, we intended to conduct a systematic review. [17] Only when several studies have the same risk factor, we would perform a meta-analysis. [17] Because included studies have different risk factors, we cannot synthesize their results. Second, the majority of included studies were retrospective studies. These data are probably lack of accuracy because patients cannot always remember when and how frequently they were exposed to risk factors. Third, it is possible that we did not include studies which could affect the result and conclusion in the current analysis. Another limitation of the observational studies was that it was not possible to control all potential confounding covariates. Because most of the included studies were observational studies, this bias was inevitable but did not have a major effect on the results of the analysis.
Despite these limitations, the evidence from this review may help establish risk prediction models (RPMs) for BM/SREs in lung cancers and apply these predictors to identify the high-risk population. Our finding provides comprehensive and systematic information to help oncologists identify patients who might obtain a benefit from systematic therapy and to help clinicians to prevent BM and SREs in future works.
In conclusion, this review has made several conclusions about clinical problems. Lung cancer patients with T4 stage, N3 stage, and positive BSP expression may experience a higher risk of BM. Furthermore, our data showed that ever-smoking and multiple BMs significantly associated with an increased risk of SREs in lung cancer patients with BM.