Aneuploidy and prognosis of non-small-cell lung cancer: a meta-analysis of published data

In lung cancer, DNA content abnormalities have been described as a heterogeneous spectrum of impaired tumour cell DNA histogram patterns. They are merged into the common term of aneuploidy and probably reflect a high genotypic instability. In non-small-cell lung cancer, the negative effect of aneuploidy has been a subject of controversy inasmuch as studies aimed at determining the survival–DNA content relationship have reported conflicting results. We made a meta-analysis of published studies aimed at determining the prognostic effect of aneuploidy in surgically resected non-small-cell lung cancer. 35 trials have been identified in the literature. A comprehensive collection of data has been constructed taking into account the following parameters: quality of specimen, DNA content assessment method, aneuploidy definition, histology and stage grouping, quality of surgical resection and demographic characteristics of the analysed population. Among the 4033 assessable patients, 2626 suffered from non-small-cell lung cancer with aneuploid DNA content (overall frequency of aneuploidy: 0.65; 95% CI: (0.64–0.67)). The DerSimonian and Laird method was used to estimate the size effects and the Peto and Yusuf method was used in order to generate the odds ratios (OR) of reduction in risk of death for patients affected by a nearly diploid (non-aneuploid) non-small-cell lung cancer. Survivals following surgical resection, from 1 to 5 years, were chosen as the end-points of our meta-analysis. Patients suffering from a nearly diploid tumour benefited from a significant reduction in risk of death at 1, 2, 3 and 4 years with respective OR: 0.51, 0.51, 0.45 and 0.67 (P< 10−4for each end-point). 5 years after resection, the reduction of death was of lesser magnitude: OR: 0.87 (P = 0.08). The test for overall statistical heterogeneity was conventionally significant (P< 0.01) for all 5 end-points, however. None of the recorded characteristics of the studies could explain this phenomenon precluding a subset analysis. Therefore, the DerSimonian and Laird method was applied inasmuch as this method allows a correction for heterogeneity. This method demonstrated an increase in survival at 1, 2, 3, 4 and 5 years for patients with diploid tumours with respective size effects of 0.11, 0.15, 0.20, 0.20 and 0.21 (value taking into account the correction for heterogeneity;P< 10−4for each end-point). Patients who benefit from a surgical resection for non-small-cell lung cancer with aneuploid DNA content prove to have a higher risk of death. This negative prognostic factor decreases the probability of survival by 11% at one year, a negative effect deteriorating up to 21% at 5 years following surgery. © 2001 Cancer Research Campaign http://www.bjcancer.com

Ploidy status predicts disease-free intervals and short-term survival in numerous human malignancies (Barlogie et al 1983;Friedlander et al 1984). In lung cancer, the prognostic value of ploidy is controversial. The effect of ploidy status on patient outcome has been investigated particularly in non-small-cell lung cancers (NSCLC), a group of different histologies including squamous cell carcinoma (SQC), adenocarcinoma (ADC) and large cell carcinoma (LCC). Although authors from different laboratories have suggested that patients presenting an aneuploid tumour have a shorter survival than patients presenting a nearly diploid tumour, others did not find such a difference. Many factors could explain the differences in estimating the ploidy-survival relationship. Possible interpretation could lie in the heterogeneity of the abnormal DNA patterns gathered under the common term of aneuploidy, i.e. hyperdiploidy, hypodiploidy and multiploidy . Differences in methods used to analyse DNA content could also have been responsible for the above-mentioned controversy inasmuch as some studies were founded on flow cytometry data whereas others used static cytometry. Finally, the difference might be related to the histological and clinical characteristics of the studied patient population.
We therefore made a meta-analysis of published studies aimed at determining the prognostic effect of aneuploidy in surgically resected NSCLC.

Eligibility criteria
The most conventional definition of ploidy status is as follows: ploidy was determined for each specimen using DNA index which represents the ratio of the cell DNA content of tumour G 0/1 cells to the diploid G 0/1 peak (2n). A coefficient of variation (CV) for this DNA index (DI) has been specifically defined for each study. Thus, a DNA index = 1 defined a near diploid specimen i.e. only one peak of G 0/1 cells in the near diploid region (2n, DNA index = 1) with few G 2 M tumour cells in the tetraploid region (4n). Conversely, tumours presenting a DI of less than [1 -CV] (hypodiploidy) or over [1 + CV] (hyperdiploid) were classified as aneuploid NSCLCs, as were tumours sharing multiple aneuploid peaks.

Aneuploidy and prognosis of non-small-cell lung cancer: a meta-analysis of published data
To be included in this meta-analysis, studies had to fulfil the following criteria: estimation of overall survival-DNA content relationship as aim of the study; inclusion of patients suffering from histologically proven NSCLC and operated upon in an attempt at surgical resection; no anticancer treatment (neither chemotherapy nor radiotherapy) prior to surgical resection; specification of the nature of both tested specimens and methods of DNA content assessment; standard ploidy and aneuploidy definitions; clear definition of overall survival as the time from surgery to the date of death (or date of survival update for censored patients), comprehensive survival analysis including the Kaplan-Meier method (Kaplan and Meier, 1958) estimation for incomplete observations separately analysing patient groups according to the DNA content. Alternative survival analysis using actuarial survival curves were also considered as eligible for this analysis. Finally, a 2-year minimal median follow-up was required. Studies mainly devoted to the analysis of NSCLC patients which also included a small proportion of SCLC were considered as eligible depending on separate survival analyses.
Studies with one or more of the following methodological issues were not included in the meta-analysis: histology not reported; analysis of DNA content aimed at assessing cell kinetic parameters such as identification of the percent of cells in S phase rather than ploidy status; unconventional survival report (lack of estimation of survival according to time, small subgroup analysis); unknown relationship between death and lung cancer; follow-up shorter than 2 years. In addition, studies mainly aimed at analysing survival according to treatment modalities (post-operative treatment, diagnostic only surgery) were not considered as eligible.

Selection of trials
A computerized bibliography was extracted from MEDLINE and CANCERLIT (CancerNet™) databases using medical subject headings for the following terms: lung neoplasm, lung carcinoma, non-small-cell, ploidy or DNA content, prognosis. The search for publications in any language, was carried out from 1966 to the end of 1999 inclusive. Afterwards, the manual selection of relevant studies was based upon summary analysis. The reprint of each study was carefully analysed regarding the different eligibility criteria. In addition to the above-mentioned procedure, bibliographies of selected full papers were screened in order to disclose other relevant articles.

Collection of data
A comprehensive collection of data has been constructed taking into account the following parameters: year of publication, aim of the study, type of sampled specimens (paraffin embedded specimen versus fresh tissue specimens), DNA content assessment method (flow cytometry versus static cytometry), aneuploidy definition, histology sub-groups and stage grouping, quality of surgical resection and demographic characteristics of the analysed population.
Survivals after surgical resection, from 1 to 5 years, were chosen as the end-points of our meta-analysis. These outcomes were assessed as follows: the parameters were directly graphically measured from magnification of publication graphs. When the data were directly reported in the text, comparisons with graphic assessment were in good agreement. An attempt to contact the first author of each selected article was made in order to obtain permission to use the data and to know whether there had been any update of the study following its publication.

Statistics
2 methods were used in order to estimate the effects of ploidy upon survival in surgically resected NSCLC patients. The Yusuf and Peto method produces odds ratio (OR) and a 95% confidence interval together with the value of the heterogeneity test (Yusuf et al, 1985). In addition, the DerSimonian and Laird method was used in order to estimate the size effects upon the different parameters and their 95% confidence intervals were calculated (DerSimonian and Laird, 1986). For this method the value and the 95% confidence interval were corrected taking into account the heterogeneity where this latter parameter was statistically significant.
Some studies reporting ploidy in NSCLC were considered as being outside the scope of this meta-analysis. 4 studies could not be selected owing to the lack of overall survival data although they appropriately reported disease-free (or relapse-free) survival (Volm et al, 1988;Carp et al, 1992;Costa et al, 1996;Roberts et al, 1998). 3 studies aimed at determining the prognostic effect of DNA content characteristics other than the classical dichotomy diploidy-aneuploidy. These characteristics targeted as prognostic factors were: type of aneuploidy patterns (Bunn et al, 1983), percent of cells in the aneuploid G 1 peak (Van Bodegom et al, 1989) and hypodiploid/hypertetraploidy patterns (Stipa et al, 1993). One study (Del Carlo Bernardi et al, 1997) defined diploid tumours as tumours presenting a DI ≤ 1.2 a feature shared by hypodiploid tumours Stipa et al, 1993). 3 publications were preliminary reports of DNA content-survival relationship (Salvati et al, 1989;Ichinose et al, 1991;Ogawa et al, 1991) but otherwise subsequently reported by the respective groups (Ogawa et al, 1992;Ichinose et al, 1993;   Three studies included less than 10% small cell lung cancers. 1994). 7 publications aimed at determining the relationship between ploidy status and other clinical and biological features such as c-erbB-2 and L-myc expressions (Chiba et al, 1993), Ki-67 growth-fraction evaluation (Simony et al, 1990), stage grouping (Ikeda et al, 1995), S-phase (Granone et al, 1993;Visakorpi et al, 1995), p53 gene mutations (Casson et al, 1994), or bromodeoxyuridine labelling (Shimosato et al, 1989). In these studies the survival data were neither exhaustively reported nor considered as the primary end-point of the experiment. Finally one study aimed at describing a new static cytometric method and secondarily reported descriptive relapse data (Blondal and Bengtson, 1981).

Peto and Yusuf
Patients suffering from nearly diploid tumours benefited from a significant reduction in risk of death at 1, 2, 3 and 4 years with respective OR: 0.51, 0.51, 0.45 and 0.67 (P < 10 -4 for each endpoint; Figures 1-4). 5 years after resection, the reduction of death was of lesser magnitude: OR: 0.87 (P = 0.08; Figures 5). The test for overall statistical heterogeneity was conventionally significant (P < 0.01; Table 2) for all 5 end-points, however. None of the recorded characteristics of the studies were able to explain this phenomenon precluding a subset analysis.

DerSimonian and Laird
The DerSimonian and Laird method was applied inasmuch as this method allows a correction for heterogeneity. This method demonstrated an increase in survival at 1, 2, 3, 4 and 5 years for patients with diploid tumours with respective size effects of 0.11, 0.15, 0.20, 0.20 and 0.21 (value taking into account the correction for heterogeneity; P < 10 -4 for each end-point).

DISCUSSION
DNA content analysis has been proposed to assess cell kinetics insofar as the DNA histogram allows the identification of the percent of cells in S phase. However, the assessment of S phase is frequently hampered by the overlap between aneuploid tumourcell population and diploid non-malignant cell population. Thus, in human malignancies, DNA content analysis is mainly used to evaluate the occurrence of aneuploid cell population, an abnormality known to characterize malignant cells . The study of cell kinetics has been proposed as a reliable prognostic parameter in human malignancies Latreille et al, 1980;Friedlander et al, 1984). Aneuploidy can predict short-term survival in many solid tumours (Wolley et al, 1982;Auer et al, 1984;Friedlander et al, 1984;Bondeson et al, 1986;Matsura et al, 1986). In lung cancer the prognostic significance of an abnormal DNA content might be regarded, in a logical way, as consistent with knowledge of tumour biology for this disease. However, published studies have generated conflicting results and currently the negative effect of aneuploidy in NSCLC is still controversial.
In this study we used a meta-analytic approach to this question in an attempt to determine the ploidy-survival relationship and, depending on its existence, the magnitude of the effect. One may hypothesize that, due to the nature of the herein meta-analysis based on published data, a possible bias was introduced as our procedure did not allow the disclosure of unpublished studies. A direct comparison of meta-analysis on medical literature and metaanalysis on individual patient data has been made in the setting of chemotherapy in ovarian cancer (Stewart and Parmar, 1993). This study suggested possible differences in estimated treatment effect due to patient exclusions and shorter length of follow-up in the former technique. However, the meta-analysis reported here takes into account 35 studies which as a whole included over 4000 patients suffering from NSCLC. A long-term survival end-point has been chosen to avoid overestimation of effect by a short follow-up period. No apparent effect of epoch appeared when year of publication was taken into account. The proportion of tumours presenting an aneuploid DNA content is set at 64% and was reported homogeneously throughout the panel of studies analysed. One can therefore speculate that publication bias could be minimal although remaining unknown.
The different study designs are similar. However, statistical heterogeneity has been detected by both meta-analysis methods regarding survival estimation. Using the DerSimonian and Laird  Zimmerman;1987 Tirindelli;1987 Ten velde;1988 Volm;1988 Yamaoka;1989 Cibas;1989 Dazzi;1990 Isobe;1990 Sahin;1990 Shiota;1990 Miyamoto;1991 Mizumoto;1991 Liewald;1992 Ogawa;1992 Filderman;1992 Morkve;1993 Cheon;1993 Pence;1993 Rice;1993 Ichinose;1993 Yu;1993 Lima;1993 Shimizu;1993 Usuda;1994 Salvati;1994 Tanaka;1995 Pujol;1996 Huang;1996 Nagai;1996 Jeanfaivre;1997 Muguerza;1997 Kolodziejski;1997 Graziano;1997 Dalquen;1997 Asamura;1999 Overall iploidy better Aneuploidy better 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 I yr D Figure 1 Odds ratio and 95% confidence interval of mortality at 1 year for patients operated upon for a NSCLC with nearly diploid DNA content (P < 10 -4 ). Results are expressed as individual and overall Ors; vertical bar, and their respective 95% confidence intervals; horizontal bar. ORs lower than 1 indicate a reduction in risk of death for patients affected by a diploid NSCLC. See in Table 2, test for heterogeneity method this heterogeneity has been taken into account. In addition, the magnitude of the survival effect has been set by the latter method as a complementary estimation of prognosis aside from the classical OR determination using the Peto and Yusuf method. There is, so far, no definitive explanation for the heterogeneity. Among the different hypotheses, we investigated whether the method of analysis could be responsible for the difference. Flow cytometry is considered as the standard method to analyse DNA content histograms after propidium iodine staining of a single-cell suspension. This method takes into account thousands of cells. Static cytometry has been proposed as an alternative method. It analyses cytological prints of the tumour after Feulgen staining. This second method only analyses a few hundred cells. Although this number is lower than the one analysed by flow cytometry, the computer-assisted image processor is able to distinguish between tumour cells to be analysed and tumour-infiltrating lymphocytes.
In addition, studies carried out in order to compare the 2 methods demonstrated the reliability of ploidy analysis using static cytometry (Friedlander et al, 1984;Oud et al, 1989). Thus, static computer-assisted cytometry is considered as a reliable means of analysing ploidy in situ. In this meta-analysis, both studies based on this technique produced results in accordance with the final survival-aneuploidy relationship as generated by the metaanalysis. Neither the nature of the samples, nor the histology or stage of the studied population could be a putative reason for the heterogeneity. In particular, published studies seldom reported on difference in aneuploidy based on histological subgroup (i.e. adenocarcinoma, SQC or LCC). Consequently, it is not possible to determine whether or not difference in NSCLC histology from one study to the other could explain the statistical heterogeneity. In addition, the distribution of aneuploid and diploid tumours by stage has been recorded in our database whenever the data was reported by each study. Out of the 18 studies with detail analysis by stages, the respective percentage of aneuploid tumours in stage I, stage II and stage III was 58, 57 and 70%. Therefore, it seems unlikely that the statistical heterogeneity regarding survival estimation of the whole meta-analysis, belongs to the distribution of stage grouping of the different accrued populations, insofar as the frequency of aneuploidy did not differ from one stage group to the other. The type of sample analysed could hardly be considered as a possible inductor of heterogeneity across the panel of studies  Zimmerman;1987 Tirindelli;1987 Ten velde;1988 Volm;1988 Yamaoka;1989 Cibas;1989 Dazzi;1990 Isobe;1990 Sahin;1990 Shiota;1990 Miyamoto;1991 Mizumoto;1991 Liewald;1992 Ogawa;1992 Filderman;1992 Morkve;1993 Cheon;1993 Pence;1993 Rice;1993 Ichinose;1993 Yu;1993 Lima;1993 Shimizu;1993 Usuda;1994 Salvati;1994 Tanaka;1995 Pujol;1996 Huang;1996 Nagai;1996 Jeanfaivre;1997 Muguerza;1997 Kolodziejski;1997 Graziano;1997 Dalquen;1997 Asamura;1999 Overall Diploidy better Aneuploidy better 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 2 yr

Figure 5
Odds ratio and 95% confidence interval of mortality at 5 years for patients operated upon for a NSCLC with nearly diploid DNA content (P < 10 -4 ) (symbols as in Figure 1; P = 0.08). Odds ratios indicate the reduction of risk of death for patients presenting a nearly diploid tumour. owing to the fact that there is a good agreement between the DNA histograms obtained from fresh tissue and paraffin-embedded specimens where flow cytometry has been done in parallel (Hedley et al, 1983). A possible hidden explanation of the statistical heterogeneity lies in the well-established diversity of DNA content histograms and growth fraction estimations in a given specimen (Sara et al, 1986;Oud et al, 1989;Simony et al, 1990;Stipa et al, 1993). Besides this heterogeneity, one can emphasize that aneuploidy is a common terminology designating different histogram patterns. Multiploidy and hypodiploidy might be underestimated in some specimens. In multiple myeloma and breast cancer attention has been paid to the occurrence of these 2 particular ploidy status (Coulson et al, 1984;Smith et al, 1986). A similar observation has been made in lung cancer in which hypodiploidy seems to indicate a particularly poor outcome (Pujol et al, 1996). Hitherto, the prognosis of patients with surgically resected NSCLC is mainly determined using two main prognostic variables, i.e. the stage of the disease and the performance status. During the past decade, the search for genetic markers of NSCLC has emerged in an attempt to predict poor outcome of this disease and to help the clinician in deciding whether there is a case for adjuvant chemotherapy. Abnormal nuclear content has long been known as a conclusive marker of malignancy and was found in increased frequency in solid tumours. It is noteworthy that the 65% overall frequency of aneuploidy, as observed in this metaanalysis is consistent with the 67% frequency reported in 4941 patients suffering from various human malignancies (Barlogie et al, 1983). During the 1990s, after an initial period of enthusiasm for aneuploidy as a putative prognostic marker in NSCLC, the interest seemed to decrease. This might be explained partially by the fact that conflicting results arose from different studies. However, the meta-analysis described above seems to provide additional clues in favour of a prognostic effect of aneuploidy in this disease. DNA content analysis provides a unique insight into the cellular heterogeneity and the genotypic instability of NSCLC. Chromosomal instability leads to abnormal regulation of gene expression and aneuploidy. This latter characteristic might be a critical factor for phenotypic diversification towards a metastatic phenotype (Nicolson, 1987).
Notwithstanding the heterogeneity across the studies, the herein meta-analysis allows the conclusion that patients who benefit from a surgical resection for NSCLC with aneuploid DNA content prove to have a higher risk of death. The survival probability for patients having an aneuploid tumour is decreased by 11% at one year, a negative effect deteriorating up to 21% at 5 years following surgery.