Background: The All of Us Research Program represents a rich resource for cancer epidemiology research, with over 400,000 participants with whole genome sequences linked to electronic health records (EHRs). Large cancer datasets often focus exclusively on cases without controls and neglect pre-diagnosis healthcare occurrences. Here, we perform a phenome-wide association study (PheWAS) of pre-diagnosis EHR data between cancer cases and matched controls, revealing co-occurring and mutually exclusive phenotypes that can be subsequently investigated using All of Us genomic data. Methods: Using SNOMED CT codes, we identified 48,000+ cancer cases across 23 cancer types in All of Us version 8. To conduct PheWAS while eliminating temporal ascertainment bias, we implemented a matched truncation strategy: for each cancer case diagnosed at age X, we matched control individuals on birth year, sex, and race and truncated their EHR data at age X, ensuring equal opportunity to accrue diagnoses. We tested associations between cancer diagnosis and approximately 2,100 clinical phenotypes using logistic regression adjusted for age, sex, and EHR metrics including length of EHR and total ICD code count, with Bonferroni correction for multiple testing. For coding aspects of this study, Claude by Anthropic was used for debugging and troubleshooting. Results: Our analysis confirmed established cancer risk factors, validating All of Us as a robust platform for cancer epidemiology research. Notably, we identified unexpected inverse associations with pain-related phenotypes: chronic pain ( P -value=7.7×10⁻⁴⁷, OR=0.67) and general pain ( P -value=2.9×10⁻⁴⁴, OR=0.72). Additional inverse associations included sleep disorders ( P -value=2.9×10⁻²⁷, OR=0.79) and mood disorders ( P -value=5.2×10⁻³², OR=0.77), suggesting potentially protective relationships warranting further investigation. Conclusions: This comprehensive PheWAS of pre-diagnosis EHR data in All of Us reveals a complex landscape of cancer-associated phenotypes extending beyond traditional risk factors. The identification of potentially protective phenotypes, combined with our rigorous approach to temporal bias elimination, establishes a foundation for hypothesis-generating research in precision cancer prevention. These findings underscore the importance of diverse biobanks and rigorous methods for identifying phenotypic relationships. Claude by Anthropic was used to revise text for grammar, clarity, and length optimization. Importantly, all scientific content, hypotheses, analyses, and conclusions remain the original work of the authors.