Prognostic factors of the overall survival in laryngeal cancer

  1. Introduction

So far many risk factors for laryngeal cancer have been described. The American Cancer Society stressed the role of smoking, alcohol intake, the interaction of smoking and alcohol consumption, age, sex, ethnic origin, poor diet, papillomavirus infection, and genetic predispositions [1]. The purpose of the analysis was to assess whether some of these risk factors affect also the overall survival of patients with laryngeal cancer.

  1. Material and methods

The statistical analysis was performed on a database of 101 patients with laryngeal cancer. Participants were evaluated for:

  • the occurrence of polymorphism of p53, CCND, and p16 alleles related to laryngeal cancer;
  • factors included to TNM tumour progression scale: tumour size, a number of lymph nodes involved;
  • alcohol consumption and smoking;
  • histopathological features of a tumour: grading, pleomorphism, figures, invasion, keratosis, lymphocytic infiltration.

The analysis was performed using the Survival Analysis module in Statistica 13.0.

2. 1. Life expectancy tables

The life expectancy tables in the study group were constructed for 12 compartments and step 1. The compliance with theoretical distribution (exponential, linear, Gompertz and Weibull) was assessed by chi square test.

2.2 Analysis of the influence of genetic factors

In the first step, Kaplan-Meier curves were created for each of the three genes separately. To investigate whether there is Mendelian inheritance or dependency of survival time from allelic dose, and to assess whether the dominant or recessive allele is related to the increased risk, the statistical significance of the difference in survival time between the groups has been assessed: A) for the CCND gene: AA + AG vs. GG (assessment of the allele A impact) AA vs. AG + GG (assessment of the allele G impact) B) for the p53 gene arg/arg + arg/pro vs. pro/pro (assessment of the impact of allele arg dose) arg/arg vs. arg/pro + pro/pro (assessment of the impact of allele pro dose) C) for the p16 gene CC vs. CG (assessment of substitution impact).

2.3 Analysis of the influence of histopathological factors

Kaplan-Meier curves were constructed for all histopathological features. The significance of survival differences was determined by the F Cox assay, which in the analysis of genetic factors impact was the most sensitive of Wilcoxon-Gehan test, Cox-Wilcoxon test, Cox-Mantel test, Peto-Peto-Wilcoxon and log-rank test. For p<0.05 the result was also checked with the other four tests and an interpretation consistent with the majority of the tests was taken. Subgroups that did not differ in survival were integrated into the same risk classes that were then used in the analysis as alternative variables for raw data. After constructing a Cox proportional hazard model, correlations of histopathological variables were analysed using the Factor Analysis module.

2.4. Construction of the Cox proportional hazard model

We introduced into the model quantitative variables: age, tumour size (T) and number of occluded lymph nodes (N), as well as qualitative variables: genotype (or class of genetic risk), smoking, alcohol consumption, the class of histopathological risk (alternatively: cell malignancy in a histopathological examination: pleomorphism, figures, invasion, keratinization, lymphocytic infiltration). Based on the literature data we tested also: smoking and alcohol intake, an interaction of these factors with genotype, interactions between tumour size and a number of occluded lymph nodes, and between sex and age.

Variables were introduced into the model gradually. For each set of variables, the model was constructed using the stepwise backwards method that gives the best chance of including all relevant variables. Qualitative variables were tested against the reference class, which were: female; GG genotype for CCND gene, arg/arg for p53 gene, CG for p16 gene; class of genetic risk 0, the lowest grades in histopathological and TN assessment, the class of histopathological risk 0. The estimation was performed by Breslov’s reliability method. The maximum number of iterations was 500. For each model, the goodness of fit was assessed and Akaike was used as a benchmark. For the best model, the proportional hazard assumption was also controlled by evaluating the association between Schoenfeld’s residues and time.

  1. Results

3.1. Life expectancy tables In the study group survival was consistent with Gompertz distribution (p1 = 0.63, p2 = 0.84, p3 = 0.93) and Weibull distribution (p1 = 0.73, p2 = 0.67, p3 = 066). The life tables are shown in Table 1.

3.2. Analysis of the influence of genetic factors

There were individuals with AG genotype (N = 68, 67%), GG (N = 13, 13%) and AA (N = 20, 19%) in CCND gene. The rarest genotype was AA, which suggests that this allele is recessive. For the p53 gene, the arg/arg (N = 57, 56%),

There were individuals with AG genotype (N = 68, 67%), GG (N = 13, 13%) and AA (N = 20, 19%) in CCND gene. The rarest genotype was AA, which suggests that this allele is recessive. For the p53 gene, the arg/arg (N = 57, 56%), arg/pro (N = 41, 41%) and pro/pro allele combinations were recorded. For the p16 gene, we detected genotypes CC (N = 76, 75%) and CG (N = 25, 25%). Kaplan-Meier curves for individual genetic variants are shown in Figures 1A-D and 2A. Differences in survival between patients with A (+) vs. alleles A (-) in the CCND gene has been statistically insignificant in each of the two survival curves (Wilcoxon-Gehan, F Cox, Cox-Mantel, Peto-Peto-Wilcoxon, and log-rank). Differences in survival between G (+) vs. G (-) were also insignificant in all above tests and the difference between arg (+) and arg (-) too. The difference in the course of the curves was visually large, probably due to the small number of homozygous subgroups. In contrast, the survival of pro (+) patients was different in Gehan-Wilcoxon test (p = 0.0234), F Cox (0.0426), and Peto-Peto-Wilcoxon (p = 0.03678). None of these tests showed any differences between the CG vs. CC alleles in p16 gene. However, the assessment of the impact of each gene alone cannot be the last stage of the analysis. Since proteins encoded by individual proto-oncogenes and anti-oncogenes are a part of different mitotic cycle checkpoints, interactions between individual genes are expected. Patients were therefore classified into 12 groups based on their genotype in all three studied loci. Five combinations of alleles occurring in a total of 7 patients were excluded due to the subgroup size of less than or equal to 2. Subsequently, Kaplan-Meier curves were prepared for each combination of alleles (Figure 2B). There are three risk groups in this chart: • AG arg/pro CG (the red group) • GG arg/pro CC (the black group) • AA arg/pro CC (the violet group). In the red and purple groups, mortality in the first month of observation is high. The black group is initially characterised by a small mortality rate that rapidly increases after 2 months of observation and reaches the highest level at the end of the follow-up period (less than 20% of patients live after 9 months). Based on the visual assessment of this graph, the risk groups were compiled into the following risk groups:

• High-risk – less than 50% of patients live in 9 months: Genotypes: AG arg/pro CG, GG arg/pro CC, AA arg/pro CC

• Average-risk – 50-60% of patients lived after 9 months: Genotypes: AA arg/arg CC, AG arg/arg CC

• Low-risk – 70% of patients lived after 9 months Genotypes: AG arg/pro CC, GG arg/arg CC, AG arg/arg CG, AA arg/arg CG

In the high-risk group survival was worsened compared to the other two groups in Wilcoxon-Gehan test (p = 0.031), Cox (p = 0.0022), Cox-Mantel (p = 0.0009), Peto-Peto-Wilcoxon (p = 0.0029) and log-rank (p = 0.0026).  In the low-risk group survival was improved compared to the other two groups in Cox-Mantel (p = 0.044) and log-rank (p = 0.043) test. The average-risk group did not differ in any of the above low-risk groups, but differed significantly from the high-risk group in each of the above tests (p = 0,0157; p = 0,01068; p = 0,01066; p = 0,01282; p = 0,01773). Thus, patients with the genotype AA arg/arg CC and AG arg/arg CC were included in the low-risk group. High-risk groups differ from the low-risk group for the presence of allele pro, which – as reported – significantly worsens the prognosis of patients. It is worth noting, however, that allele pro is also present in the blue group, with a survival rate of 70% after 10 months. This may be due to the aforementioned allele interactions or the impaired recruitment to the sample (subgroup numbers are small). It should also be borne in mind that genetic factors can interact not only with each other but also with other, non-genetic risk factors. For this reason, the next step in the analysis was the regression analysis using the Cox proportional hazard model.

3.3 Assessment of the impact of histopathological features

Results of differences in survival and histopathological risk classes thus created are presented in Table 2.

3.4. Cox proportional hazard model

The comparison of different Cox hazard models was shown in Table 3. The best fit was obtained for models 19 and 20. The evaluation of parameters for all built models is shown in Table 4. Values of Wald statistic and significance levels are included in Appendix I.

  1. Discussion

4.1. Genetic risk factors

The CCND gene encodes cyclin D; the database did not specify which numerical variant of this cyclin was investigated. There are three types of cyclin D in mammals, but cyclin D1 is involved in the literature on the risk of larynx cancer. Cyclins are part of the cell cycle checkpoints system. Their concentration in each phase of the cell cycle is variable. They regulate the activity of cyclin-dependent kinases, affecting the duration of individual phases of the cell cycle, and determine whether a cell can move from G1 to S-phase [2]. Mielcarek-Kuchta et al. have shown a correlation between cyclin D1 expression and larynx malignancy [3], but Vielbe et al. did not confirm the cyclin D1 prognostic value for this cancer [4]. Rydzanicz et al., however, have shown that important for the pathogenesis of laryngeal cancer is the G870A mutation, which involves the substitution of guanine by adenine at position 870 of exon 4. This mutation changes the frequency of alternative splicing, extending the half-life of the protein, which results in overexpression. The GG genotype was mainly found in healthy individuals while GA or AA genes were more common in people with larynx cancer. In addition, they found that the AA genotype increases the risk of lymph node metastasis.

The p53 gene encodes a p53 protein that participates in transcriptional regulation, micro-RNA processing, cell cycle control, apoptosis induction, and angiogenesis; it is also associated with oxidative stress and pluripotent stem cell differentiation [6]. P53 gene mutations have been identified in numerous cancers and are associated with poor prognosis. The database sent for analysis did not specify in which position appeared the resulting in the conversion of arginine to proline. It is known from the literature that this substitution in codon 72 is associated with increased oral and cervical cancer in people infected with papilloma virus (which is also a risk factor for laryngeal cancer) [7], colorectal cancer [8] and breast cancer [9]. The effect of p53 mutations on prognosis in laryngeal carcinoma was studied by Pastuszewski et al., who in the group of 65 patients with laryngeal carcinoma reported elevated p53 levels associated with decreased survival [10]. Chomehai et al., who described a mutation in the p53 gene in 33% of patients in the laryngeal cancer trial, had confirmed the association of this mutation with total survival, but the direction of this association was opposite — patients with a mutation had a better prognosis. As predicted by the authors of this paper, the predictive power of the mutation in the p53 gene depends on the loss of protein function and the inactivation of conserved regions thereof [11].

The p16 gene codes the p16 protein, which is one of the control points between the G1 and S phases and the inhibitor of the cyclin D/Cdk4,6 complex [12]; its structure is often altered in tumours of various origins [13]. In a study published by Pietruszewska et al., 57% of 67 patients with laryngeal carcinoma showed this protein inhibition [13]. This study also found a relationship between the inhibited expression of this protein and lymph node metastases and the shorter survival time of tumour recurrence. It is suspected that changes in p16 protein play a role not only in the formation of such larynx but also in its progression [13]. However, this study did not confirm these assumptions. The absence of the GG variant in the test sample may be caused by the lack of representativeness of the sample caused by the rarity of this variant (insufficient sample size) or lethality of the combination.

Genes tested are not the only genetic risk factors. Guo-Kang et al. described the role of p27 protein — a negative regulator of the cell cycle — as a prognostic factor in laryngeal cancer [14]. In this study, lack of expression of p27 was one of the strongest prognostic factors in the Cox proportional hazard model, as well as gender, tumour size, lymph node involvement, disease stage, p53 expression, and G-CSF-R expression. The role of p21 protein was showed in the Pietruszewska’s study [13].

Peschos et al. examined the expression of p53/p21, cyclin D1/cyclin E and p21/p27 genes in 57 laryngeal cancer patients. Analysis of the p53/p21 gene revealed an abnormal phenotype in 67% of patients and normal in 33% of patients. Cycle lesions were overexpressed in 35% of patients and normal expression in 65% of patients. Evaluation of p53/p21, cyclin D/E and p21/p27 was not useful as a prognostic factor in laryngeal cancer [14]. Unfortunately, this study did not analyse the subgroups identified due to environmental risk factors.

In light of the above, differences in survival between individual genotypes are understandable. Low-risk patients (relatively low, as it is a group of patients who have had a laryngeal carcinoma, which demonstrates the increased risk of cancer in patients with this allele combination) have retained the p53 protein production. Affiliation of the AG arg/pro CC variant to the lowest risk group is difficult to explain on the basis of the three analysed genes. This is a combination of unfavourable alleles for the CCND gene with reduced expression of the p53 gene (heterozygote), and differences in survival based on the variant p16 gene were not statistically significant. Perhaps, in this case, an important role play other genes, not included in this experiment. The alternative explanation is that the sample size was inadequate to show the statistical significance of the differences visible in the Kaplan-Meier curves.

The group with the worst prognosis combines reduced expression of p53 (all three variants). Two of them also have unfavourable alleles for the CCND gene.

4.2. Cox proportional hazard model

Cigarette smoking is one of the main risk factors for head and neck cancers, including larynx cancer [15]. Consumption of over 1 drink (14 g ethanol) per day also increases the risk of this group of cancers, but to a lesser extent than smoking, and the scale of this increase is proportional to the dose [15]. However, both behaviours, often co-occurring, increase the synergistic risk [15].

The obvious association between prognosis and TNM scale in laryngeal cancer has been indicated by Vlachtsis et al. [17]. Our study confirmed the fundamental importance of this classification for prognosis in this cancer. The size of a tumour and the number of lymph nodes involved proved to be the best model among models that have been built on one group of related factors. This model was better at Akaike 21 points than the model based on a histopathological raw data and better at 23 Akaike points than the model based on histopathological data classified in risk classes. The AIC for the model based on tumour size model and lymph node involvement was only one point worse than the model based on all variables assessed in the study.

The model based on raw genotypic data in the three studied loci proved to be completely useless as opposed to the risk-based model built on differences in survival. Nevertheless, this model has been found to be significantly worse than both the histopathological model and the TN model.

The introduction the histopathological and genetic data into a model improved the AIC compared to those models in a single perspective; the addition of two variables from TNM scale to the same model has dramatically improved its quality. When instead of TN we tested variables regarding demography and addictions, the quality of the model has not improved. It also did not increase after the addition of alcohol and cigarette interaction to the model, despite the literature reports on the role of such interactions in the larynx cancer pathogenesis. Apparently, this factor is only conducive to starting oncogenesis and does not affect the progression of existing cancer.

The hypothesis about the potential interaction of the genotype for oncogenesis and addiction has also not been confirmed. Introducing variables which describe the habit into the genetic model has drastically reduced its quality. Also, sex and gender interaction with genotype and additions deteriorated the quality of the model. Adding gender as well as gender and addiction interaction has worsened a model based on genetic, histopathological and TN characteristics. The quality of model that contained only the p53 gene (significant, as described above), TN, and the most important histopathological variable (grading) was also poor.

Best quality had two models created by the random method, most likely due to the mixing of variables of different types that were not correlated. Histopathological variables were characterised by a high correlation coefficient (Table 2), which was determined only after the models were built. One of these models was the first to show the significant interaction of high-risk genetic and cigarette smoking. In view of the insignificance of such variables in all created models, this may be an artefact. The possibility of the appearance of artefacts in models suggests the existence of the variable ‘infiltration’ among the significant variables of model No. 1, although the differences in survival between the Kaplan-Meier curves for this variable proved to be insignificant. Therefore, extreme caution should be exercised in concluding that cigarette smoking has sped up the death of only those patients with laryngeal cancer who have a genetic predisposition.

In conclusion, the study confirmed the multi-factor nature of models predicting a prognosis in laryngeal cancer. Risk factors for 9 months of follow-up (established according to inclusion criteria) remain for the criteria of tumour progression and histopathological criteria (related to tumour type). Genetic predispositions (in particular arginine substitution by proline in p53 protein and guanine substitution by adenosine in CCND gene), and alcohol consumption appear to play some role, but these relationships are too weak to draw any concrete conclusions from this study.

Figure 1. Kaplan-Meier curves in genetic subgroups



Table 1. Life span chart in the study group (on request).

Table 2. Classification of histopathologic risk classes (on request)


  1. [Access on 17 April 2017]
  2. Stenzel A, Żuryń A, Grzanka AA, i wsp. Cykliny jako markery chorób nowotworowych. Nowotwory, 2012;Vol 62 (2);115–122.
  3. Mielcarek-Kuchta D, Olofsson J, Golusiński W. p53, Ki67 and cyclin D1 as prognosticators of lymph node metastases in laryngeal carcinoma. Eur Arch Otorhinolaryngol 2003;260:549–554.
  4. Vielba R, Bilbao J, Ispizua A et al. p53 and cyclin D1 as prognostic factors in squamous cell carcinoma of the larynx. Laryngoscope 2003;113:167–172.
  5. Rydzanicz M, Golusiński P, Mielcarek-Kuchta D et al. Cyclin D1 gene (CCND1) polymorphism and the risk of squamous cell carcinoma of the larynx. Eur Arch Otorhinolaryngol 2006;263:43–38.
  6. [Access on 18 April 2017]
  1. Katiyar S, Thelma BK, Murthy NS, Hedau S, Jain N, Gopalkrishna V, Husain SA, Das BC. Polymorphism of the p53 codon 72 Arg/Pro and the risk of HPV type 16/18-associated cervical and oral cancer in India. Mol Cell Biochem. 2003 Oct;252(1-2):117-24.
  2. Lima JM, Serafim PV, Silva ID, Forones NM. [Role of the genetic polymorphism of p53 (codon 72) gene in colorectal cancer]. Arq Gastroenterol. 2006 Jan-Mar;43(1):8-13. Epub 2006 May 8.
  3. Bonafé M, Ceccarelli C, Farabegoli F, et al. Retention of the p53 codon 72 arginine allele is associated with a reduction of disease-free and overall survival in arginine/proline heterozygous breast cancer patients. Clin Cancer Res. 2003 Oct 15;9(13):4860-4.
  4. Pastuszewski W, Dziegiel P, Krecicki T, et al. Prognostic Significance of Metallothionein, p53 Protein and Ki-67 Antigen Expression in Laryngeal Cancer. Anticancer Research, Jan-Feb 2007; vol. 27, no. 1A, 335-342
  5. Chomehai JS, Wei D, Sarkar FH, et al. Prognostic significance of p53 gene mutations in laryngeal cancer. The Laryngoscope. 1999 March;109,3:455–459.
  6. [Access on 18 April 2017]
  7. Pietruszewska W, Durko M, Kobos J. Alterations of cell cycle regulating proteins: Rb, p21 and p16 in laryngeal cancer. Otolaryngologia Polska 2007, LXI, 6, 951-957.
  8. Peschos D, Stefanou D, Vougiouklakis T, Assimakopoulos DA, Agnantis NJ. Cell cycle proteins in laryngeal cancer: role in proliferation and prognosis. J Exp Experimental & Clin Cancer Res 2005;24(3):431–437.
  9. [Access on 18 April 2017]
  10. Muscat JE, Wynder EL. Tobacco, alcohol, asbestos, and occupational risk factors for laryngeal cancer. Cancer, 1992; 69(9]:2244–2251.
  11. Vlachtsis K, Nikolaou A, Markou K, et al. Clinical and molecular prognostic factors in operable laryngeal cancer. European Archives of Oto-Rhino-Laryngology and Head & Neck, 2005 Nov; 262;11:890–898.