A Systematic Review of Molecular and Biological Tumor Markers in Neuroblastoma

Home , Homovanillic acid, Vanillylmandelic acid

4 Vol. 10, 4–12, January 1, 2004 Clinical Cancer Research

Review A Systematic Review of Molecular and Biological Tumor Markers in Neuroblastoma

Richard D. Riley,1 David Heney,2 reviews in the future. In particular, collaboration of cancer David R. Jones,1 Alex J. Sutton,1 research groups is needed to enable bigger sample sizes, Paul C. Lambert,1 Keith R. Abrams,1 standardize methods of analysis and reporting, and facilitate the pooling of individual patient data. Bridget Young,3 Alan J. Wailoo,4 and Susan A. Burchill5 Introduction Departments of 1Health Sciences, 2Medical Education, University of Leicester, Leicester; 3Department of Psychology, University of Hull, Neuroblastoma is a neuroblastic tumor of the primordial Hull; 4School of Health and Related Research, University of neural crest and is the most common extracranial solid tumor of Sheffield, Sheffield; and 5Cancer Research United Kingdom Clinical childhood, comprising between 8 and 10% of all childhood Centre, St. James’s University Hospital, Leeds, United Kingdom cancers. It is an enigmatic tumor demonstrating diverse clinical and biological characteristics and behavior (1). Tumors may regress spontaneously, reflecting induction of apoptosis or dif- Abstract ferentiation, or they may exhibit extremely malignant behavior Purpose: The aim of this study was to conduct a sys- with very low cure rates. The spectrum of clinical behavior tematic review, and where possible meta-analyses, of molec- suggests that genetic, biological, and morphological features ular and biological tumor markers described in neuroblas- may be useful markers to stratify children with this disease for toma, and to establish an evidence-based perspective on the most appropriate management. Knowledge of prognostic their clinical value for the screening, diagnosis, prognosis, markers may also help understand the genesis of this disease. and monitoring of patients. Neuroblastoma is predominantly a disease of the first dec- Experimental Design: A well-defined, reproducible ade with ϳ80% of children presenting at Ͻ4 years old; median search strategy was used to identify the relevant literature age is 22 months. The incidence in the United Kingdom and the from 1966 to February 2000. United States is ϳ1 in 7000 live births, and there is slight sex Results: A total of 428 papers studying the use of 195 predominance in most series with a male-to-female ratio of different tumor markers in neuroblastoma were identified. 1.2:1. The disease accounts for 15% of all childhood cancer Small sample sizes, poor statistical reporting, large hetero- deaths, indicating the poor prognosis of many of the tumors geneity across studies (e.g., in cutoff levels), and publication (2–4). Children with stage 1, 2, or 4 s disease, or presenting in bias limited meta-analysis to the area of prognosis only; the first year of life have a good prognosis. In contrast, children MYCN, chromosome 1p, DNA index, vanillylmandelic acid: (Ն1 year of age) with stage 3 and 4 disease have 3-year survival homovanillic acid ratio, CD44, Trk-A, neuron-specific eno- rates of 50% and 15%, respectively. Most children present over lase, lactate dehydrogenase, ferritin, and multidrug resist- the age of 1 year with metastatic (stage 4) disease; this group has ance were all identified as potentially important prognostic an overall survival of 10–20% (3, 4). tools. A number of genetic and biological features have been Conclusions: This systematic review forms a knowledge investigated in recent years in an effort to improve the under- base of the tumor markers studied thus far in neuroblas- standing of the behavior of neuroblastoma and to identify tumor toma, and has identified some of the most important prog- markers that would improve cure rates by facilitating the screen- nostic markers, which should be considered in future re- ing, diagnosis, prognosis, or monitoring of patients. In particu- search and treatment strategies. Importantly, the review has lar, many prognostic studies have identified many tumor mark- also highlighted some general problems across primary tu- ers associated with overall or disease-free survival, including mor marker studies, in particular poor and heterogeneous MYCN copy number, ploidy, and deletion or loss of heterozy- reporting. These need to be addressed to allow better clinical gosity of chromosome 1p and gain of chromosome 17q. How- interpretation and enable more appropriate evidence-based ever, it has proved difficult to identify which prognostic markers are the most useful, reflecting the complex nature of the tumor and the lack of large prospective clinical outcome studies. The aim of this study was to conduct a systematic review, Received 8/23/02; revised 9/15/03; accepted 9/19/03. and where possible meta-analyses, of molecular and biological Grant support: National Health Service Health and Technology As- tumor markers described in neuroblastoma, and to establish an sessment (HTA) Program (Grant 97/15/03). The costs of publication of this article were defrayed in part by the evidence-based perspective on their clinical value for the payment of page charges. This article must therefore be hereby marked screening, diagnosis, prognosis, and monitoring of patients. This advertisement in accordance with 18 U.S.C. Section 1734 solely to should facilitate identification of the most useful tumor markers indicate this fact. for clinical management and the development of future research Requests for reprints: Richard David Riley, Department of Health Sciences, University of Leicester, 22-28 Princess Road West, Leicester, strategies by: (a) establishing the importance of the markers LE1 6TP, United Kingdom. Phone: 44-116-2525427; Fax: 44-116- studied; and (b) identifying the markers that warrant additional 2523272; E-mail: [email protected]. investigation. Downloaded from clincancerres.aacrjournals.org on September 29, 2021. © 2004 American Association for Cancer Research. Clinical Cancer Research 5

Table 1 Set of keywords used in the neuroblastoma literature search of Medline, Embase, and Cancerlit A. Neuroblastoma Clinical Area Neuroblastoma Patient(s) Prognostic Ganglioneuroblastoma Child Diagnostic Ganglioneuroma Children Pediatric Prognosis Paediatric Diagnosis Screening Monitoring Infant(s) Follow-up B. Tumor marker

Tumour marker(s) DOPA Homovanillic acid Tumor marker(s) Neuron-specific enolase HVA Marker(s) NSE Normetanephrine N-MYC Ferritin NM NMYC Lactate dehydrogenase Metanephrine MYCN LDH MN Tyrosine hydroxylase Ganglioside(s) 3-methoxy tyramine TH Monosialganglioside 3-MT Dopa-decarboxylase Disialoganglioside Vanillacetic acid DDC C-neu VPA Phenylethanolamine-N-methyl transferase C-myc Vanillglycol PNMT Neuropeptide(s) VG PGP9.5 Somatostatin receptors Vanillglycol acid Dopamine-␤-hydroxylase Telomerase VGA DBH CD44 Catechol acetic acid Phenylalanine Mitotic index CAA Drug resistance RT-PCR VAA MRP Dopamine Norepinephrine Tyrosine NB84 Vanilalamine 3,4-dihydroxyl phenyl alanine Noradrenaline VA 1p deletion Adrenaline Trk A DNA diploidy Vanillylmandelic acid Trk B 17q VMA Trk C 14q Epinephrine C.

Psychosocial Economic Quality of life Cost Cea Anxiety Cost-effectiveness Cba Psychosocial Econ Cua Adjustment

Materials and Methods A paper was included if a word from {Neuroblastoma}, a word The systematic review followed the guidelines contained in from {Tumor Marker}, and a word from {Clinical Area} were NHS Centre for Reviews and Dissemination (1996), and its included anywhere in the paper. underlying philosophy was to maintain breadth, synthesize the Three investigators independently performed the assess- evidence qualitatively, and then, only where appropriate, use ment of the papers. All three had previous experience of iden- quantitative methods, making procedures explicit and transpar- tifying relevant tumor marker literature for a systematic review ent throughout (5). and subsequent meta-analysis (6, 7). Furthermore, the second Search Strategy. The three on-line bibliographic data- investigator is a pediatric oncology consultant, with a special bases Medline, Embase, and Cancerlit were chosen as a basis for interest in neuroblastoma, and the third investigator is a trans- identifying the relevant literature from 1966 to February 2000. lational scientific research fellow, with a special interest in small An iterative procedure was used to develop an optimal search round cell cancers of childhood. Both of these investigators held strategy, which culminated in the use of three important sets of regular meetings with the first investigator about the review keywords in the strategy (Table 1). The keywords in {Neuro- process and the assessment of the literature. blastoma} related to the family of this disease, whereas those in The first investigator read the available abstract to classify {Tumor Marker} included the named markers thought a priori each paper into one of three categories: “relevant,”“uncertain,” to be potentially important. The set {Clinical Area} included or “not relevant.” The second and third investigators, who had more specific terms for the clinical use of markers in children. more background clinical knowledge in the research area,

checked all of the abstracts where classification was uncertain, provide an important comparative estimate of the risk of death/ and ϳ10% of the papers in each of the relevant and not relevant disease recurrence between two groups of patients. categories. Copies of all of the papers classified as relevant, It was common for a paper to report more than one prog- together with all of the papers in which relevance remained nostic result by relating one or more markers to OS and/or DFS, unclear after assessment of abstracts by the three investigators, and also by providing unadjusted and/or adjusted results (e.g.,

were obtained and then read thoroughly to make a final decision adjusted for age and stage of disease). Estimates of the loge as to their inclusion. (HR) and its variance comparing two groups defined by a single Inclusion. To be included in the systematic review a marker level were sought from all of the OS and DFS reports paper had to provide a quantitative result or give tabulated using the methods described by Parmar et al. (10). An unad- individual patient data (IPD) evaluating the use of a tumor justed estimate was preferred for each report, as adjusted results marker in neuroblastoma. The paper had to be based on a are likely to be highly inconsistent in the factors for which primary research study of humans relevant to the clinical area of adjustment are made (11). An adjusted estimate was sought in screening, diagnosis, prognosis, or monitoring. There was no the absence of an unadjusted result. restriction on age of patients in the study, although ϳ90% of An assessment of the publication bias in the prognosis papers included just 0–18-year-olds. The criteria for classifying literature was made by application of two appropriate statistical the four clinical areas was that the paper had to present data in tests to the MYCN results (12, 13). The Trim and Fill method the form of summary statistics or IPD for: (a) screening: the use was also used to assess the likely impact of publication bias on of tumor markers to screen an apparent healthy population; (b) the pooled MYCN results (14). Economic and Psychosocial Effects of Tumor Markers. diagnosis: tumor marker levels considered of diagnostic value; A set of keywords was used to screen the abstracts of the papers (c) prognosis: tumor marker levels at a measured point in time classified as relevant for any economic or psychosocial results with relation to the outcome of patients at the end of a specific relating to the use of a tumor marker in neuroblastoma for any follow-up period; and (d) monitoring: tumor marker levels taken of the clinical areas (Table 1). repeatedly during a follow-up period with relation to disease status over that period. Exclusion. Papers that reported only laboratory work, Results methodology for identifying new markers, or results from ani- Literature Search Results. We identified 3415 papers mal studies were excluded. Review articles and foreign lan- from the searches; 1536 were first identified in Medline, an guage papers were also excluded. Histological characteristics of additional 473 in Embase, and then an additional 1406 from tumors [such as the presence of differentiated ganglia in neuro- Cancerlit. These were classified by the three investigators (Fig. blastoma (Shimada index)] were not included in the markers 1). The second and third investigator agreed that 85.7% of a sample of the first investigator’s relevant papers were indeed reviewed. relevant or uncertain (42 of 49). They also agreed that 193 Information Extracted. From the included papers, in- (86.9%) of a sample of 222 not-relevant papers checked were formation was extracted on the tumor marker used and to which indeed not relevant, and of the 29 others classified 8 papers as clinical area it related: screening, diagnosis, prognosis, or mon- relevant and 21 papers as uncertain. After obtaining and reading itoring. Among the covariate information extracted from each the entire articles, 15 of these 21 uncertain papers were paper on prognosis was whether survival was overall (OS) or ultimately classified as not relevant. Thus, 208 of the first disease-free (DFS), the marker cutoff level if applicable and, if investigator’s 222 not-relevant papers sampled by the other so, the total number of patients and deaths within each high and investigators were correctly classified (93.7%). Overall, 428 low subgroup. The age range and stages of neuroblastoma papers were considered relevant and included in our review (Fig. 1; disease represented by the patients in each study were also complete list of these references available on the internet).6 recorded, as these were known a priori to be important prog- Tumor Markers Identified Overall and Within Each nostic clinical features (8). Clinical Area. A total of 195 different tumor markers were Meta-Analysis and Assessment of Publication Bias. studied in these 428 papers in relation to the screening, diagno- Meta-analysis was performed, where possible, to combine all of sis, prognosis, or monitoring of neuroblastoma (Table 2). There the relevant results found from the literature search (9). For each were 49 different papers on screening, 288 on diagnosis, 260 on of the areas of screening, diagnosis, and monitoring, only those prognosis, and 51 on monitoring; 201 of the 428 papers covered tumor markers on which 3 or more papers provided data were two or more clinical areas. considered. For the area of prognosis, due to the many prog- Screening. The review identified 49 papers that gave nostic markers and prognostic studies identified, meta-analysis quantitative data relating to the use of tumor markers in screen- Ն was limited to those reported in 10 papers. Both fixed and ing and potentially considered the evaluation of a population- random effects meta-analyses were used, with the latter pre- based screening program for neuroblastoma. These papers ferred if there was evidence of heterogeneity. Meta-analysis for covered programs established in geographical regions of Aus- Ͻ clinically relevant subgroups of patients (e.g., age 1 and stage tria, Canada, France, Germany, Japan, and the United Kingdom 4 disease) was also considered, but was only performed where sufficient data were available. For the meta-analysis of data from the prognosis papers, 6 the extraction of the loge (hazard ratio; HR) and its variance was Internet address: http://www.prw.le.ac.uk/epidemio/personal/rdr3/ the desired target. These statistics were chosen because they paed2.html.

Fig. 1. Flow chart showing the results at each stage of the process used to identify the final set of relevant papers in the neuroblastoma review.

The studies considered a variety of outcomes including: (a) only compared the number of neuroblastoma patients with high/ feasibility/uptake rate; (b) the number of false-positive and positive marker levels to those with low/negative levels respec- false-negative cases; (c) incidence; (d) stage distribution; and (e) tively. Marker levels from patients with neuroblastoma were mortality. In terms of outcomes (c), (d), and (e), some studies rarely compared with those from a sample of healthy controls in have undertaken, or are designed to enable in the future, a the diagnosis papers, e.g., none of the 22 papers reporting levels comparison between screened and control (nonscreened) popu- of serum lactate dehydrogenase at diagnosis compared patients lations. Unfortunately, a quantitative synthesis of results was not with neuroblastoma to healthy controls. Current clinical practice feasible given the heterogeneity in how and which outcomes suggests that urinary catecholamines are important for the dif- were reported. However, a qualitative assessment suggested that ferential diagnosis of neuroblastoma from other small round cell considerable uncertainty still surrounds whether population- tumors; however, the poor quality of the published literature based screening for neuroblastoma is cost-effective overall, and, prevented a quantitative evaluation of this practice. if so, the optimal age at which to screen, and also the optimal Prognosis. The 12 most commonly studied prognostic screening strategy, i.e., one-stage or multistage. Recent studies markers were each selected for an in-depth study to establish have shown that early screening (before 6 months of age) is not their individual value as a prognostic tool; each marker was informative (see “Discussion”). studied in Ն10 prognosis papers (Table 3). The prognostic value Diagnosis. It was not possible to perform a meta-analysis of CD44 expression was also evaluated, because all of its 8 of the data from the diagnosis papers, because the results mostly prognostic studies were contained within those papers of the

Table 2 List of tumor markers in neuroblastoma that were identified by the systematic review together with the number of papers overall and within each clinical area Tumor Marker Overall Screening Diagnosis Prognosis Monitoring MYCN 201 7 148 151 9 VMA 125 44 78 45 18 HVA 105 38 64 35 16 DNA index/ploidy/diploidy/triploidy/aneuploid/hyperdiploidy 56 5 37 44 1 Chromosome 1p or chromosome 1p36 47 4 34 40 1 Ferritin or isoferritin 49 3 36 33 5 NSE 45 2 33 28 9 LDH 32 1 22 26 4 Dopamine 24 2 22 10 4 TrkA (nerve growth factor receptor) 25 0 16 16 0 Adrenaline/epinephrine 15 0 15 5 4 Multidrug resistance/associated protein/p-glycoprotein 16 0 7 16 0 Nonadrenaline/noradrenaline/norepinephrine 13 0 13 5 2 CD44 10 0 7 8 0 Neuropeptide Y 12 0 10 9 0 Tyrosine hydroxylase 12 0 11 3 3 Chromosome 17q 11 0 9 8 0 Ha-ras/P21/H-ras/c-ha-ras 11 0 8 6 1 Telomerase/Telomeric repeats 11 0 6 7 0 Chromosome 14q 8 0 6 7 0 GD2 ganglioside 8 0 7 5 2 S100 7 1 5 5 0 Chromosome 11q 6 0 5 6 1 Low affinity nerve growth receptor (LNGFR) 6 0 3 6 0 Metanephrine 6 0 6 1 1 TrkC 6 0 3 5 0 3-methoxy-4-hydroxyphenyl glycol 5 1 3 1 0 4-hydroxy-3-methoxymandelic acid 5 0 5 1 2 Dihydroxyphenylalanine 5 1 5 1 3 Dopamine ␤ hydroxylase 5 0 3 2 2 Proliferating cell nuclear antigen/proliferation index/Ki67/KiS5 protein 5 0 5 4 0

other 12 markers to be evaluated. Hence, 13 markers overall presented in Table 3 and have been classified into three marker were evaluated as a prognostic tool. These 13 markers were groups: DNA/chromosome abnormalities, biological markers, studied in 211 (81.2%) of all of the prognosis papers. Within and urinary catecholamines. For the vast majority of markers, these there were 575 reports of prognostic power assessment, there is a statistically significant difference (in terms of the HR) where levels of 1 of these 13 tumor markers were related to OS between the groups defined by the markers. For example, there or DFS by summary statistics or IPD. was strong statistically significant evidence that amplification of Weakness of reporting, analysis, and presentation of results the MYCN gene was associated with a worse OS and DFS. The

meant that only 204 (35.5%) estimates of both the loge (HR) and risk of death was 5.48 times greater for patients with MYCN its variance could be extracted. Meta-analyses were additionally amplification compared with those that did not have amplifica- restricted by large variability in both clinical and statistical tion [HR ϭ 5.48; 95% confidence interval (CI), 4.30–6.97], and factors relating to the 204 estimates. For example, for the similarly for risk of disease recurrence (HR ϭ 4.28; 95% CI,

marker MYCN, 94 estimates of the loge (HR) and variance were 3.34–5.49). All of the papers that could be included in the obtained but these involved 9 different cutoff points to dichot- meta-analyses were published from 1985 onwards, with the omize the marker, 9 different stage groups, 4 different age majority after 1989. For example, of the 70 articles used for the groups, 17 adjusted/77 unadjusted estimates, and 2 different MYCN OS and DFS meta-analyses none were published before outcomes (OS and DFS). Type of treatment and method of 1985, only 8 were published from 1985 to 1989, and 62 were marker measurement were not recorded but both would have published from 1990 onwards. Hence, the prognostic studies we added additional heterogeneity to that already noted. A more included have been reported after the improved method for in-depth evaluation of these reporting problems, together with staging and treatment of neuroblastoma that has improved sur- recommendations for improvement, are provided elsewhere vival for children with this disease over the last 20 years. (15). Monitoring. The review identified 51 papers that pro- Whereas acknowledging the problems implied by this large vided quantitative data evaluating the serial use of tumor mark- degree of heterogeneity, it remained important to use the data ers to aid the clinical management of patients with neuroblas- extracted for each marker and determine which were potentially toma. However, there was considerable heterogeneity between the most important markers for future research. Therefore, meta- the studies in terms of: (a) tumor markers considered (e.g., analysis was performed for each of the 13 markers separately for VMA, HVA, MYCN, ferritin, NSE, and lactate dehydrogenase); OS and DFS. The meta-analysis results for each marker are (b) outcome (OS and DFS); (c) statistical analyses undertaken

Table 3 Meta-analysis results for the 13 prognostic markers, grouped by tumor marker class, for overall survival and disease-free survival together with the number of prognosis papers identified overall and the number of estimates of the log (hazard ratio) and variance obtained for each outcome No. No. Pooled 95% Tumor marker prognosis estimates hazard Confidence Marker type (relationship with prognosis) papers Outcome obtained ratio interval P DNA or chromosome MYCN (amplification poor outcome) DFSa 46 4.28 3.34 to 5.49 Ͻ0.0001 abnormalities 151 OS 48 5.48 4.30 to 6.97 Ͻ0.0001 DNA index (diploidy poor outcome) DFS 8 0.33 0.20 to 0.56 Ͻ0.0001 44 OS 11 0.31 0.20 to 0.48 Ͻ0.0001 Chromosome 1p (deletion poor outcome) DFS 9 3.93 2.31 to 6.68 Ͻ0.0001 40 OS 11 3.12 1.95 to 4.98 Ͻ0.0001 Urinary VMA (elevated poor outcome) DFS 1 Not possible Not possible Not possible catecholamines 36 OS 3 0.50 0.19 to 1.29 0.15 HVA (elevated poor outcome) DFS 0 Not possible Not possible Not possible 26 OS 2 1.14 0.65 to 1.98 0.65 VMA:HVA (small ratio (e.g. Ͻ1) poor outcome) DFS 3 0.35 0.17 to 0.72 0.0043 20 OS 2 0.44 0.18 to 1.06 0.068 Dopamine (elevated poor outcome) DFS 1 Not possible Not possible Not possible 10 OS 1 Not possible Not possible Not possible Biological CD44 (high expression good outcome) DFS 3 0.06 0.02 to 0.21 Ͻ0.0001 markers 8 OS 0 Not possible Not possible Not possible TrkA (high expression good outcome) DFS 7 0.26 0.16 to 0.42 Ͻ0.0001 16 OS 4 0.09 0.05 to 0.16 Ͻ0.0001 NSE (high serum levels poor outcome) DFS 5 5.56 2.11 to 14.7 0.0005 28 OS 4 5.22 3.12 to 8.73 Ͻ0.0001 LDH (high serum levels poor outcome) DFS 7 3.20 2.06 to 4.98 Ͻ0.0001 26 OS 5 3.36 1.72 to 6.57 0.0004 Ferritin (high serum levels poor outcome) DFS 4 4.26 2.42 to 7.53 Ͻ0.0001 33 OS 3 2.74 1.92 to 3.91 Ͻ0.0001 MRP (high expression poor outcome) DFS 7 6.37 3.71 to 10.9 Ͻ0.0001 16 OS 9 3.52 1.19 to 10.5 0.023 a DFS, disease-free survival; OS, overall survival.

and reporting of results; (d) length of follow-up, e.g., treatment and their families of using tumor markers clinically in neuro- phase, long-term follow-up; (e) number of patients; (f) age/stage blastoma, even in the monitoring papers. distribution; and (g) number of serial measurements. The combination of these problems with the relatively few papers identified meant that any meta-analysis using these stud- Discussion ies was not worthwhile, statistically or clinically. This is the first systematic review of tumor markers for

Publication Bias. The estimates of the loge (HR) and its neuroblastoma that has been reported, and it forms a knowledge variance obtained for MYCN were used to assess the possibility base for future research. A systematic review is the preferred means of publication bias. Both Begg and Egger tests produced Ps Ͻ of identifying and combining existing evidence (18), and it is 0.001, and therefore publication bias was strongly suspected to particularly important for evidence-based evaluations of tumor be a problem (12, 13). In fact, using the Trim and Fill method, markers in neuroblastoma, and indeed other rare diseases, because 17 studies with smaller HRs were estimated as missing from our (sometimes conflicting) evidence relating to markers is published results, in addition to the 45 studies included in the analysis. across a number of studies, many of which involve small numbers Hence, it appears likely that the effect size from the original of patients. Systematic reviews are also important because they can meta-analyses may be biased upwards, i.e., they may overesti- highlight underlying problems across individual studies and help mate the true underlying loge (HR) for MYCN. It is highly likely identify future research needs (19). Indeed, both of these aspects that this problem is also true for the other markers. are demonstrated throughout our review and form the most impor- Economic and Psychosocial Results. No papers made tant messages of this paper, which we hope will help improve an economic evaluation of the use of tumor markers in neuro- tumor marker studies in the future. blastoma, but 2 papers reported cost data in relation to screening Appraisal of the Systematic Review. During the sys- (16, 17). They are both somewhat dated and contain few details tematic review we classified 3415 papers overall and ultimately about cost calculations, which made it difficult to assess the identified 428 relevant papers, which showed diversity in pri- accuracy of the claims made or the relevance of the findings to mary interest, methodology, analysis of data, and quality of current practice. Furthermore, none of the 428 relevant papers reporting. The search strategy used is likely to have identified reported any data on the psychosocial consequences for children the majority of the available literature, targeting in particular the

databases specializing in scientific and clinical reporting, al- MYCN, chromosome 1p, DNA index, VMA:HVA ratio, CD44, though we acknowledge the possibility that the review may not Trk-A, NSE, lactate dehydrogenase, ferritin, and multidrug re- be fully comprehensive, reflecting publication and reporting sistance were all identified as potentially important prognostic bias. Our initial search strategy included the names of those tools. However, the pooled results must be treated with caution markers known a priori to be potentially important but, in light given the reporting problems and large heterogeneity of clinical/ of the many markers identified during the review, this list was statistical factors across studies. Recent studies have also indi- certainly not exhaustive, although we did include more general cated that chromosome 17q gains also have important prognos- terms such as “marker” that will have limited the number of tic significance (22–28). Unfortunately, many of these studies markers missed. were published after the start of our review, and consequently Initially our strategy had been to check the references chromosome 17q was not among the prognostic markers we quoted in all of the selected papers, to identify any other papers selected for the in-depth evaluation above. However, in light of missing from the three databases we searched. This was not this current knowledge, we have subsequently extracted, wher- feasible given the large literature base, nor was it possible to ever possible, HR results from each of the 8 prognosis papers systematically check for duplicates of patients across papers. our review identified for this marker. Meta-analysis of these We did not include foreign language papers because of the suggests that patients with gain of chromosome 17q have a difficulties in translation, and this may have introduced bias if significantly worse DFS (from 3 studies: HR ϭ 4.16; 95% CI, statistically or clinically significant studies were more likely to 2.56–6.77) and OS (from 3 studies: HR ϭ 4.30; 95% CI, be (re)written for publication in an English language journal 2.70–6.86) compared with those who did not. However, these (20). results are again subject to the problems of poor reporting and Clinical Interpretation. Neuroblastoma is a multi- heterogeneity. faceted disease. The expansion in biological and cytogenetic It was not possible to compare subgroups of patients, markers is an indication that the cancer process is complex with individual prognostic markers, or assess the benefits of using multiple changes taking place with a neuroblastoma cell. Re- any of the markers in combination. Availability of full IPD, search studies have tried to identify which of these markers are including all of the exact values of all markers assessed and initiating factors contributing to the cancer phenotype and which subgroup information (e.g., age and stage of disease), from each may be regarded as secondary. Initial studies for a particular paper would facilitate such assessments in the future (15). marker have inevitably sought to link that one marker with Ideally, large multicenter studies should also be initiated to survival, to establish its importance. However, it is now becom- assess the benefits of using chromosome 17q and the other ing clear that a number of key biological markers are likely to be important prognostic markers, both individually and in combi- interlinked and must be evaluated in combination with one nation, to improve strategies for the stratification of children another and not in isolation. with neuroblastoma for therapy. Results from such studies The poor and heterogeneous reporting restricted any quan- would enable clinicians to clearly see which are the most titative synthesis in the areas of screening, diagnosis, and mon- appropriate individual and combinations of prognostic markers itoring, and therefore any overall clinical evaluation. Recent to use. papers, published since the start of our review, suggest that there Reporting Problems. Importantly, our review has high- is currently insufficient evidence to support a screening program lighted problems in how individual prognostic marker studies for infants up to 6 months of age, and the majority of authors are designed and reported. Many study reports do not allow for conclude that it should be discontinued (21). Clinical and his- adequate extraction of data in order for comparisons to be made, topathological features, which have not been evaluated in our and the large heterogeneity in cutoff levels and other measures review, are the most informative for the diagnosis of neuroblas- limit evaluations of markers across studies and within specific toma, and these include age at diagnosis, tumor histology, and subgroups of patients (e.g., age Ͻ1). There is also the threat of primary tumor site. However, the detection of catecholamine publication bias across the literature. For those researchers metabolites in urine is also used for the differential diagnosis of studying prognostic tumor markers, Altman and Lyman (29) neuroblastoma from other small round cell tumors of childhood. have proposed important guidelines for both conducting and For the use of molecular and biological markers in diagnosis, the evaluating prognostic factor studies that should be considered. few studies comparing a healthy control group to patients with We have also provided guidelines for improved statistical re- neuroblastoma was particularly disappointing, and future stud- porting of primary studies to facilitate the evaluation and com- ies need to address this. Similarly, monitoring studies need to parison of individual markers, identify the additional benefits of report the differences in serial marker measurements between using combinations of markers, and ultimately allow important those who develop a recurrence of disease and those who remain evidence-based reviews to be made (15). In particular, presen- disease-free, preferably for a many patients over a long tation of the HR with some measure of precision (e.g., 95% CI) follow-up period. Where possible, research groups need to col- and availability of IPD, either in the paper or on the internet laborate and pool together resources to enable bigger sample (30), are both highly desirable. The availability of IPD has sizes and achieve consistency across studies, which should be allowed important evidence-based reviews to be made in other targeted to address the important issues. Only then will the cancer settings (31, 32). benefits of using tumor markers for screening, diagnosis, and Psychosocial and Economic Issues. The clinical impli- monitoring be properly ascertainable. cations of the results must also be considered together with This systematic review did produce an evaluation of the psychosocial and economic aspects of tumor markers, for ex- most commonly reported individual markers for prognosis; ample in monitoring and screening, respectively. Our search

found no published evidence on the psychosocial consequences 4. Cotterill, S. J., Pearson, A. D., Pritchard, J., Foot, A. B., Roald, B., for children and their families of using tumor markers clinically Kohler, J. A., and Imeson, J. Clinical prognostic factors in 1277 patients in neuroblastoma, but this probably reflects the few papers with neuroblastoma: results of The European Neuroblastoma Study Group ’Survey’ 1982–1992. Eur. J. Cancer, 36: 901–908, 2000. studying the use of tumor markers for monitoring patients. 5. NHS Centre for Reviews and Dissemination. Undertaking System- Psychosocial evaluations of tumor markers are clearly important atic Reviews of Research on Effectiveness: CRD Guidelines for Those and have been performed for other disease settings (33). Future Carrying Out or Commissioning Reviews. CRD Report 4. University of research on the use of tumor markers in pediatric oncology York: NHS Centre for Reviews and Dissemination, 1996. should seek to include an assessment of the psychosocial out- 6. Riley, R. D., Burchill, S. A., Abrams, K. R., Heney, D., Sutton, A. J., comes of using markers, particularly for prognostic and moni- Jones, D. R., Lambert, P. C., Young, B., Wailoo, A. J., and Lewis, I. J. toring purposes. A systematic review of molecular and biological markers in tumours of We also only identified two studies that included an eco- the Ewing’s sarcoma family. Eur. J. Cancer, 39: 19–30, 2003. nomic evaluation of using tumor markers, and so the cost- 7. Riley, R. D., Burchill, S. A., Abrams, K. R., Heney, D., Lambert, P. C., Jones, D. R., Sutton, A. J., Young, B., Wailoo, A. J., and Lewis, effectiveness of individual markers could not be evaluated. A I. J. A systematic review and evaluation of the use of tumour markers in clear prescription for future trials and studies is therefore to paediatric oncology: Ewing’s sarcoma and neuroblastoma. Health Tech- include an economic evaluation element wherever possible, nol. Assess., 7: 1–162, 2003. either in terms of a cost-effectiveness study or the identification 8. Brodeur, G. M. Molecular basis for heterogeneity in human neuro- of resource use that would permit decision models to be devel- blastomas. Eur. J. Cancer, 31: 505–510, 1995. oped. 9. Sutton, A. J., Abrams, K. R., Jones, D. R., Sheldon, T. A., and Song Summary. This systematic review has emphasized the F. (eds.). Methods for Meta-analysis in Medical Research. London: John uncertainty on the clinical value of the studied tumor markers in Wiley, 2000. neuroblastoma, reflecting the small size of many studies and 10. Parmar, M. K. B., Torri, V., and Stewart, L. Extracting summary statistics to perform meta-analyses of the published literature for sur- poor statistical reporting. We have managed to assess the sig- vival endpoints. Stat. Med., 17: 2815–2834, 1998. nificance of the most commonly reported prognostic markers; 11. Altman, D. G. Systematic reviews of studies of prognostic vari- MYCN, chromosome 1p, DNA index, VMA:HVA ratio, CD44, ables. In: M. Egger, G. Davey Smith, D. G. Altman (eds.), Systematic Trk-A, NSE, lactate dehydrogenase, ferritin, and multidrug re- Reviews in Health Care: Meta-Analysis in Context, Chapter 13, 228– sistance were all identified as potentially important prognostic 247. London: BMJ Publishing Group; 2001. tools. Each of these, and the more recently identified changes in 12. Begg, C. B., and Mazumdar, M. Operating characteristics of a rank chromosome 17q, should be considered in the development of correlation test for publication bias. Biometrics, 50: 1088–101, 1994. future research strategies and for the stratification of children 13. Egger, M., Smith, G. D., Schneider, M., and Minder, C. Bias in meta-analysis detected by a simple, graphical test. Br. Med. J., 315: with neuroblastoma for different treatment strategies. 629–634, 1997. The multiplicity and complexity of tumor markers in neu- 14. Duval, S., and Tweedie, R. A non-parametric “trim and fill” method roblastoma underlines the need for studies to be coordinated, of assessing publication bias in meta-analysis. Biometrics, 56: 455–463, through the cooperation of cancer research groups, using mul- 2000. tiple laboratories and standardizing methods of analysis and 15. Riley, R. D., Abrams, K. R., Sutton, A. J., Lambert, P. C., Jones, reporting. In particular, collaboration is needed to consider D. R., Heney, D., and Burchill, S. A. Reporting of prognostic markers: prospectively planned pooled analyses and facilitate the pooling current problems and development of guidelines for evidence-based practice in the future. Br. J. Cancer, in press. of IPD, a strategy that would adequately ensure a quantitative 16. Sawada, T., Todo, S., Fujita, K., Iino, S., Imashuku, S., and synthesis (rather than merely a qualitative synthesis) to address Kusunok, T. Mass screening of neuroblastoma in infancy. Am. J. Dis. the questions of interest, such as which combinations of markers Child., 136: 710–712, 1982. provide the best prognostic tool, or whether monitoring patients 17. Scriver, C. R., Gregory, D., Bernstein, M., Clow, C. L., Weisdorf, with neuroblastoma using markers is cost-effective. Large, mul- T., Dougherty, G. E., Auray-Blais, C., Giguere, R., Lemieux, B., and ticenter collaborative studies would require agreement as to Laberge, C. Feasibility of chemical screening of urine for neuroblastoma case finding in infancy in Quebec. CMAJ, 136: 952–956, 1987. which markers to measure, and we have provided a base that 18. Egger, M., Davey Smith, G., and Altman, D. G. (eds.). Systematic highlights the ones reported in greatest detail thus far. Reviews in Health Care: Meta-Analysis in Context. London: BMJ Publishing Group, 2001. Acknowledgments 19. Egger, M., Davey Smith, G., O’Rourke, K. Rational, potentials and We thank Suzy Paisley at the School of Health and Related Re- promise of systematic reviews. In: M. Egger, G. Davey Smith, D. G. search (ScHARR) in Sheffield for advice on systematic reviews and Altman (eds.), Systematic Reviews in Health Care: Meta-Analysis in Context, pp. 3–19. London: BMJ Publishing Group, 2001. literature searching. 20. Egger, M., Zellweger-Za¨hner, T., Schneider, M., Junker, C., Len- geler, C., and Antes, G. Language bias in randomised controlled trials References published in English and German. Lancet, 350: 326–329, 1997. 1. Brodeur, G. M., and Castleberry, R. P. Neuroblastoma. In:P.A. 21. Brodeur, G. M., Look, A. T., Shimada, H., Hamilton, V. M., Maris, Pizzo and D. G. Poplack (eds.), Principles and Practice of Paediatric J. M., Hann, H. W., Leclerc, J. M., Bernstein, M., Brisson, L. C., Oncology, pp. 761–797. Philadelphia: J. B. Lippincott Co., 1997. Brossard, J., Lemieux, B., Tuchman, M., and Woods, W. G. Biological 2. Maris, M., and Matthay, K. K. Molecular biology of neuroblastoma. aspects of neuroblastomas identified by mass screening in Quebec. Med. J. Clin. Oncol., 17: 2264–2279, 1999. Pediatr. Oncol., 36: 157–159, 2001. 3. Grovas, A., Fremgen, A., Rauck, A., Ruymann, F. B., Hutchinson, 22. Abel, F., Ejeskar, K., Kogner, P., and Martinsson, T. Gain of C. L., Winchester, D. P., and Menck, H. R. The National Cancer Data chromosome arm 17q is associated with unfavourable prognosis in Base Report on patterns of childhood cancer in the United States. neuroblastoma, but does not involve mutations in the somatostatin Cancer (Phila.), 80: 2321–2332, 1997. receptor 2(SSTR2) gene at 17q24. Br. J. Cancer, 81: 1402–1409, 1999.

23. Bown, N., Cotterill, S., Lastowska, M., O’Neill, S., Pearson, A. D., 29. Altman, D. G., and Lyman, G. H. Methodological challenges in the Plantaz, D., Meddeb, M., Danglot, G., Brinkschmidt, C., Christiansen, evaluation of prognostic factors in breast cancer. Breast Cancer Res. H., Laureys, G., and Speleman, F. Gain of chromosome arm 17q and Treat., 52: 289–303, 1998. adverse outcome in patients with neuroblastoma. N. Engl. J. Med., 340: 30. Hutchon, D. J. R. Publishing raw data and real time statistical 1954–1961, 1999. analysis on e-journals. Br. Med. J., 322: 530, 2001. 24. Cunsolo, C. L, Bicocchi, M. P., Petti, A. R., and Tonini, G. P. 31. Sakamoto, J. Teramukai, S., Koike, A., Saji, S., Ohashi, Y., and Numerical and structural aberrations in advanced neuroblastoma tumours by CGH analysis; survival correlates with chromosome 17 status. Nakazato, H. Prognostic value of preoperative immunosuppressive Br. J. Cancer, 83: 1295–1300, 2000. acidic protein in patients with gastric carcinoma. Findings from three independent clinical trials. Tumor Marker Committee for the Study 25. Lastowska, M. Cotterill, S., Pearson, A. D., Roberts, P., McGuckin, Group of Immunochemotherapy with PSK for Gastric Cancer. Cancer A., Lewis, I., and Bown, N. Gain of chromosome arm 17q predicts unfavourable outcome in neuroblastoma patients. U. K. Children’s Can- (Phila.), 77: 2206–2212, 1996. cer Study Group and the U. K. Cancer Cytogenetics Group. Eur. J. 32. Look, M. P., van Putten, W. L., Duffy, M. J., Harbeck, N., Cancer, 33: 1627–1633, 1997. Christensen, I. J., Thomssen, C., Kates, R., Spyratos, F., Ferno, M., 26. Plantaz, D., Mohapatra, G., Matthay, K. K., Pellarin, M., Seeger, Eppenberger-Castori, S., Sweep, C.G., Ulm, K., Peyrat, J. P., Martin, R. C., and Feuerstein, B. G. Gain of chromosome 17 is the most frequent P. M., Magdelenat, H., Brunner, N., Duggan, C., Lisboa, B. W., abnormality detected in neuroblastoma by comparative genomic hybrid- Bendahl, P. O., Quillien, V., Daver, A., Ricolleau, G., Meijer-van ization. Am. J. Pathol., 150: 81–89, 1997. Gelder, M. E., Manders, P., Fiets, W. E., Blankenstein, M.A., Broet, 27. Vandesompele, J. Speleman, F., Van Roy, N., Laureys, G., Brin- P., Romain, S., Daxenbichler, G., Windbichler, G., Cufer, T., Borstnar, skchmidt, C., Christiansen, H., Lampert, F., Lastowska, M., Bown, N., S., Kueng, W., Beex, L. V., Klijn, J. G., O’Higgins, N., Eppenberger, U., Pearson, A., Nicholson, J. C., Ross, F., Combaret, V., Delattre, O., Janicke, F., Schmitt, M., and Foekens, J. A. Pooled analysis of Feuerstein, B. G., and Plantaz, D. Multicentre analysis of patterns of prognostic impact of urokinase-type plasminogen activator and its DNA gains and losses in 204 neuroblastoma tumors: how many genetic inhibitor PAI-1 in 8377 breast cancer patients. J. Natl. Cancer Inst., subgroups are there? Med. Pediatr. Oncol., 36: 5–10, 2001. 94: 116–128, 2002. 28. Bown, N., Lastowska, M., Cotterill, S., O’Neill, S., Ellershaw, C., 33. Park, K. C. Schwimmer, J., Shepherd, J. E., Phelps, M. E., Czernin, Roberts, P., Lewis, I., and Pearson, A. D. 17q gain in neuroblastoma J. R., Schiepers, C., and Gambhir, S. S. Decision analysis for the predicts adverse clinical outcome. Med. Pediatr. Oncol., 36(1): 14–19, cost-effective management of recurrent colorectal cancer. Ann. Surg., 2001. 233: 310–319, 2001.

Richard D. Riley, David Heney, David R. Jones, et al.

Clin Cancer Res 2004;10:4-12.

Updated version Access the most recent version of this article at: http://clincancerres.aacrjournals.org/content/10/1/4

Cited articles This article cites 26 articles, 4 of which you can access for free at: http://clincancerres.aacrjournals.org/content/10/1/4.full#ref-list-1

Citing articles This article has been cited by 8 HighWire-hosted articles. Access the articles at: http://clincancerres.aacrjournals.org/content/10/1/4.full#related-urls

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://clincancerres.aacrjournals.org/content/10/1/4. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.