Childhood Cancer Survivor Study

Allison King king_a@wustl.edu Kevin Krull kevin.krull@stjude.org Wendy Leisenring wleisenr@fhcrc.org Marilyn Stovall mstovall@mdanderson.org Greg Armstrong greg.armstrong@stjude.org Nicole Ullrich nicole.ullrich@childrens.harvard.edu Les Robison les.robison@stjude.org Lonnie Zeltzer lzeltzer@mednet.ucla.org Dan Green daniel.green@stjude.org Charles Sklar sklarc@MSKCC.ORG Roger Packer rpacker@cnmc.org Elizabeth M. Wells ewells@cnmc.org

We propose to complete an analysis of the updated 2007 questionnaires of medulloblastoma/PNET survivors and their siblings to address chronic medical conditions, cognitive function, employment status, perceived health status, and anxiety. We will also measure the level of concern regarding future health and fertility. These results should provide data to address the challenges and concerns of long-term survivors of childhood medulloblastoma. This will likely be the last cohort of survivors who were treated systematically with 3600 cGy of craniospinal radiation. Comprehensively characterizing long term outcomes in this cohort will provide critical background information against which to compare survivors treated with more contemporary regimens that reduced craniospinal dose to 2340 cGy and utilize chemotherapy for standard risk patients.

4.
Specific Aims: That would be chest wall dosing like the Mertens paper in 2002.

Planned Analyses
Descriptive characteristics for all outcomes and covariates will be summarized separately for cancer survivors and sibling controls. Where necessary, frequency distributions will be examined to aid in forming reasonable groupings for categorical variables that will be used in statistical modeling.
The type of statistical model utilized for comparison of survivors to controls for each outcome will depend upon the structure of the outcome measure. For outcomes for which CCSS collects data on age of onset and which were ascertained on at least the baseline questionnaire (neurologic problems, cardiac disease, respiratory disease), Cox regression models of time-to-firstonset will be used. Event times will be censored at time of loss-to-follow-up or death. For sibling control subjects, their entry into the at-risk group will be at age 5, the youngest age a survivor can enter the cohort. Hazard ratios for the outcome event in cancer survivors versus sibling controls will be estimated, along with the accompanying 95% confidence intervals. Robust estimation procedures will be utilized to account for any dependence among observations on subjects from the same family. Cumulative incidence curves may also be presented as graphical summaries for these outcome measures.
One notable exception to the analysis of medical outcomes is the cognitive impairment/memory measure. This item was only asked on the 2007 questionnaire. Consequently we do not have appropriate follow-up on the whole cohort of 5 year medulloblastoma/PNET survivors that would be necessary to do a Cox regression analysis. This outcome will instead be evaluated as a dichotomous Y/N outcome in a cross-sectional analysis that is limited to subjects who responded to the cognitive impairment questions on the FU2007 questionnaire.
The remaining outcome measures are interval, ordinal or nominal scale observations. Most of these measures have longitudinal data collected at multiple time points. In order to utilize all available observations, these outcomes will be analyzed using Generalized Estimating Equation modeling. Outcome measures will first be dichotomized [e.g. living arrangement (7 categories) condensed into living independently Y/N], and the GEE model will estimate the relative risk parameter and 95% confidence interval for cancer survivors versus siblings. Either a log-binomial or a Poisson implementation of the relative risk estimation can be used for the relative risk estimation.
All of the above regression analyses for the primary aim will be carried out with the goal of evaluating the comparison of survivors to siblings, with appropriate adjustment for the demographic covariates, age, gender and race.
The analyses relating to the secondary study aim will utilize data from the cancer survivor cohort only. The same types of modeling approaches described above will be used to evaluate the impact of categories of cancer treatment exposures on the neurological, cognitive/memory and fertility outcomes. Univariate analyses will be conducted first, and additional covariates for which p<=0.20 in univariate analyses will then be evaluated in multivariable models. Sibling controls All subjects who completed at least the baseline questionnaire will be included in the full analysis data set, and will be utilized for most outcome analyses. For a subset of the outcomes that were only ascertained on specific questionnaires, only those subjects who responded to the relevant questionnaire will be utilized, and analyses will be carried out in a cross-sectional fashion.