Wayne S. Kendal

A population-based analysis of second primary cancers after irradiation for rectal cancer

Objective:To investigate the possible association between pelvic irradiation for rectal cancer and subsequent second primary cancers. Patients and Methods:A population-based analysis of 20,910 individuals with rectal cancer from the Surveillance, Epidemiology, and End Results registry, for whom follow-up times were at least 5 years, was performed. Kaplan-Meier estimates for the development of second cancers within irradiated and nonirradiated cohorts provided a comparison that accounted for censored data. Cox proportional hazards analyses were further conducted to compensate for patient and tumor-related factors. Results:A total of 656 (12%) and 2368 (16%) second primary cancers were enumerated from the irradiated and nonirradiated cohorts, respectively, with the proportion of second primary cancers within the irradiated cohort being significantly decreased (P < 0.001) on crude analysis. However, Kaplan-Meier and Cox analyses revealed no significant difference between the 2 cohorts when all second primary cancer sites were considered together (hazard ratio = 1.02; 95% confidence interval [CI], 0.92–1.12). Proportional hazards analysis for specific second primary sites revealed a decreased risk after pelvic irradiation for cancer of the prostate (hazard ratio = 0.63; 95% CI, 0.48–0.84), and an increased risk for cancers of the uterine corpus & cervix (hazard ratio = 2.5; 95% CI, 1.6–4.0). Conclusion:Second primary cancers after irradiation for rectal cancers appear relatively infrequent compared with the background incidence of spontaneous cancers, and should not factor into treatment decisions for this older population. We hypothesize that the incidence of second primary tumors within adjacent organs could represent a balance between the radiation-induction of tumors and the radiation-inhibition of spontaneously occurring tumors.

Chance mechanisms affecting the burden of metastases

BackgroundThe burden of cancer metastases within an individual is commonly used to clinically characterize a tumors biological behavior. Assessments like these implicitly assume that spurious effects can be discounted. Here the influence of chance on the burden of metastasis is studied to determine whether or not this assumption is valid.MethodsMonte Carlo simulations were performed to estimate tumor burdens sustained by individuals with cancer, based upon empirically derived and validated models for the number and size distributions of metastases. Factors related to the intrinsic metastatic potential of tumors and their host microenvironments were kept constant, to more clearly demonstrate the contribution from chance.ResultsUnder otherwise identical conditions, both the simulated numbers and the sizes of metastases were highly variable. Comparable individuals could sustain anywhere from no metastases to scores of metastases, and the sizes of the metastases ranged from microscopic to macroscopic. Despite the marked variability in the number and sizes of the metastases, their respective growth times were rather more narrowly distributed. In such situations multiple occult metastases could develop into fully overt lesions within a comparatively short time period.ConclusionChance can have a major effect on the burden of metastases. Random variability can be so great as to make individual assessments of tumor biology unreliable, yet constrained enough to lead to the apparently simultaneous appearance of multiple overt metastases.

Characterization of the frequency distribution for human hematogenous metastases: Evidence for clustering and a power variance function

When groups of mice are injected with cells from a metastasizing tumor, a minority of individuals within a given group tends to sustain disproportionately larger numbers of metastases relative to the remaining group members. This clustering of metastases obeys a power function relationship, σ2 = aμb, between the variance σ2 and the mean number of lung metastases per animal μ (a and b constant). To see whether such clustering occurs with human lung, brain, and liver metastases, a meta-analysis of clinical and pathological series was performed. Thirty-three published series were identified that provided data regarding the numbers of organ metastases sustained by 5582 people. The data were grouped according to the primary tumor, site of metastasis and method of detection of metastases. Clustering of metastases within individuals of each subgroup (similar to the murine systems) was demonstrated by variance to mean ratios greater than 1, and by a strong correlation to the variance to mean power function (a∼0.49, b∼2.24, r2 = 96%, p < 10−6). The cause of this clustering remains unclear, but it may in part relate to heterogeneities in regional blood flow. As a consequence of this clustering, limited metastases would be expected to occur more frequently than predicted from random chance-providing for some optimism in the management of limited metastasis. As well, the frequency distribution for metastases revealed certain scaling symmetries, likely reflective of the underlying mechanisms of metastasis, that could be of interest to both clinicians and experimentalists working with metastasis.

Statistical kinematics of axillary nodal metastases in breast carcinoma

The number of involved lymph nodes in individuals with breast cancer is highly variable, and of both prognostic and therapeutic importance. A statistical description for the frequency distribution of the numbers of involved nodes in an affected population could potentially reveal mechanisms of axillary metastasis, and eventually facilitate predictive models for tumor control and axillary sampling. A meta-analysis of 15 studies involving 24,757 axillary dissections was performed, including conventional dissections, sentinel node dissections and studies of occult metastases. Frequency histograms for the numbers of involved axillary lymph nodes from the populations were tested for clustering and they were fitted, as a first approximation, to a negative binomial distribution. Although the number of involved nodes per individual was quite variable, some individuals sustained more involved nodes than could be expected from a random (Poisson) distribution. The negative binomial distribution, however, provided acceptable descriptions for the distributions of involved nodes in all populations studied. Two mechanisms could explain these observations: (1) an apparent contagion model, where involved nodes seeded further nodal metastases, and (2) a spurious contagion model where the number of involved nodes per individual was randomly (Poisson) distributed and population heterogeneity accounted for the more severe cases. Both models were consistent with the hypothesis that the nodal metastasis is a chance event, with the probability of involvement greatest for nodes contiguous to the primary tumor and proportioned by lymphatic flow.

Large-scale genomic correlations in Arabidopsis thaliana relate to chromosomal structure.

BackgroundThe chromosomes of the plant Arabidopsis thaliana contain various genomic elements, distributed with appreciable spatial heterogeneity. Clustering of and/or correlations between these elements presumably should reflect underlying functional or structural factors. We studied the positional density fluctuations and correlations between genes, indels, single nucleotide polymorphisms (SNPs), retrotransposons, 180 bp tandem repeats, and conserved centromeric sequences (CCSs) in Arabidopsis in order to elucidate any patterns and possible responsible factors for their genomic distributions.ResultsThe spatial distributions of all these elements obeyed a common pattern: the density profiles of each element within chromosomes exhibited low-frequency fluctuations indicative of regional clustering, and the individual density profiles tended to correlate with each other at large measurement scales. This pattern could be attributed to the influence of major chromosomal structures, such as centromeres. At smaller scales the correlations tended to weaken – evidence that localized cis-interactions between the different elements had a comparatively minor, if any, influence on their placement.ConclusionThe conventional notion that retrotransposon insertion sites are strongly influenced by cis-interactions was not supported by these observations. Moreover, we would propose that large-scale chromosomal structure has a dominant influence on the intrachromosomal distributions of genomic elements, and provides for an additional shared hierarchy of genomic organization within Arabidopsis.

Empirically-based estimates for the burden of subclinical metastases

Purpose: To describe the frequency distribution for the number of residual subclinical metastatic tumor cells after removal of the primary cancer. Materials and methods: Previously obtained autopsy, surgical pathological and laboratory data were used to characterize the size and number distributions for hematogenous and lymphatic metastases. Monte Carlo simulations were used to estimate the numbers of residual tumor cells based upon the assumption of a lognormal distribution for the sizes of metastases and Poisson, Poisson negative binomial, or negative binomial distributed numbers of metastases (corresponding to lymphatic metastases within individuals, hematogenous metastases within individuals, and lymphatic metastases within populations, respectively). Results: In each of the scenarios the resultant distribution for the numbers of subclinical tumor cells was unimodal and positively skewed, with a tail extending to the higher numbers of metastases. When plotted with equal sized counting bins and according the logarithm of the number of tumor cells, the distributions showed deviations from the normal form no greater than several percentage points – a result considered acceptable given the variabilities inherent to metastasis data. Conclusions: The distribution for the number of residual subclinical metastases may be extrapolated from data and models derived from the size and number distributions for metastases. In the absence of a closed form description for this distribution, the lognormal distribution could provide a crude, but practical, approximation for cases limited to occult microscopic residual disease. These analyses will facilitate the definition of the dose-response for the adjuvant therapy of subclinical metastases.

Age Bias in Time From Diagnosis Comparisons of Prostate Cancer Treatment.

Objectives: Observational studies of prostate cancer treatment have demonstrated a major survival benefit with prostatectomy; randomized trials have been less certain in this regard. This discrepancy is hypothesized to be due to the use survival calculations based on time from diagnosis (TFD), which can bias toward better survival for younger cohorts. Attained age is an alternative timescale that can mitigate this effect. A Surveillance, Epidemiology and End Results comparison of prostatectomy, radiotherapy (XRT), and conservative management for localized prostatic cancer was conducted to compare these 2 timescales. Methods: Kaplan-Meier analysis was used to contrast overall survival based on TFD and attained age from 279,064 prostate cancer cases. Proportional hazards models were constructed and baseline hazard functions estimated. Results: The prostatectomy cohort averaged 9 to 12 years younger than the radiotherapy or conservative management cohorts, and the baseline hazard depended more strongly upon age than TFD. Survival calculations based on TFD demonstrated a major benefit with prostatectomy compared with XRT and conservative management, consistent with prior observational studies. Calculations based on attained age, however, demonstrated lesser differences between treatment cohorts and were more consistent with published randomized trials. Conclusions: The survival benefit apparent to prostatectomy in conventional observational cohort studies could reflect an age-related bias attributable to their use of TFD analysis. Care is warranted in the choice of timescale in observational analysis if large age differences exist between treatment cohorts. Randomized controlled trials remain the most reliable means to compare prostate cancer treatments.

Pancreatectomy Versus Conservative Management for Pancreatic Cancer: A Question of Lead-time Bias.

Objectives:Pancreatectomy is regarded as the only curative treatment for cancer of the pancreas. A population-based study was conducted to examine its efficacy within the general community. Methods:Overall and cancer-specific survivals were compared between individuals treated with pancreatectomy and those managed nonsurgically. Kaplan-Meier analysis was used, based on both time from diagnosis and attained age (age at diagnosis plus time from diagnosis). Results:A total of 7830 Surveillance Epidemiology and End Results cases of localized cancer of the pancreatic head were retrieved, diagnosed from 2000 to 2008. Median follow-up was 12 months; the pancreatectomy cohort was 5 years younger and had 7-fold less stage III disease. Overall and cancer-specific survivals were 17% and 21% at 5 years from time of diagnosis in the pancreatectomy cohort versus 2% and 4% in the nonsurgical cohort, respectively (P<0.001). However, the overall and cancer-specific survival curves were nearly superimposed on each other when based on attained age. Moreover, the proportion of deaths attributable to pancreatic cancer exceeded 85% in both cohorts. Conclusions:A lead-time bias is hypothesized to explain the survival discrepancies seen between time from diagnosis and attained age analyses; the pancreatectomy cohort was diagnosed earlier, with less disease. If most of these individuals had occult metastases at diagnosis, which manifested later and caused death at similar ages as the nonsurgical cohort, their survival from time of diagnosis would appear speciously improved. A randomized controlled trial would be necessary to confirm whether or not the survival advantage ascribed to pancreatectomy should be attributed to lead-time bias.