Abstract
Cardiorespiratory fitness has been linked to both lower and higher risks of cancer, but the evidence comes from observational analysis which may be influenced by unobserved confounders and bias processes. We aimed to examine the associations between adolescent cardiorespiratory fitness and risk of cancer in late adulthood while addressing the unknown influence of unobserved familial confounders and diagnostic bias processes.
We conducted a sibling-controlled cohort study with registry linkage based on all Swedish men who participated in mandatory military conscription examinations from 1972 to 1995 and who completed standardized cardiorespiratory fitness testing. The outcomes were overall cancer diagnosis and cancer mortality, and 14 site-specific cancers (diagnosis or death), ascertained using the National Patient Register and Cause of Death Register until 31 December 2023. A total of 1,124,049 men, including 477,453 full siblings, with a mean age of 18.3 years at baseline, were followed until a median (maximum) age of 55.9 (73.5) years, during which 98,410 were diagnosed with cancer and 16,789 died from cancer (41,293 and 6,908 among full siblings respectively). In cohort analysis, individuals in the highest quartile of fitness had a lower risk of overall cancer mortality (adjusted hazard ratio [HR]: 0.71, 95% confidence interval [CI] 0.67, 0.76; P < 0.001) compared to the lowest quartile, corresponding to a standardized cumulative incidence (1-Survival) difference of -0.85 (95% CI [-1.00, -0.71]) percentage points at 65 years of age. Individuals in the highest quartile of fitness also had lower risks (HRs ranging from 0.81 to 0.49, incidence differences ranging from -0.13 to -0.32 percentage points; P < 0.001 for all) of rectum, head and neck, bladder, stomach, pancreas, colon, kidney, liver, bile ducts, and gallbladder, esophagus, and lung cancer. Yet, individuals in the highest quartile of fitness had higher risks of prostate (HR: 1.10, 95% CI [1.05, 1.16]; P < 0.001, incidence difference: 0.48 percentage points, 95% CI [0.23, 0.73]) and skin cancer (e.g., non-melanoma HR: 1.44, 95% CI [1.38, 1.50]; P < 0.001, incidence difference: 1.84 percentage points, 95% CI [1.62, 2.05]). Individuals in the highest quartile of fitness had a higher risk of overall cancer diagnosis (HR: 1.08, 95% CI [1.06, 1.11]; P < 0.001, incidence difference: 1.32 percentage points, 95% CI [0.94, 1.70]), results driven by prostate and skin cancer being the most common types of cancer. When comparing full siblings, and thereby controlling for unobserved shared confounders, the lower risk of overall cancer mortality remained (HR: 0.78, 95% CI [0.68, 0.89]; P < 0.001, incidence difference: -0.61 percentage points, 95% CI [-0.93, -0.28]), while the excess risk of prostate (HR: 1.01, 95% CI [0.90, 1.13]; P = 0867, incidence difference: 0.05 percentage points, 95% CI [-0.50, 0.60]), skin (e.g., non-melanoma HR: 1.09, 95% CI [0.99, 1.20]; P = 0.097, incidence difference: 0.40 percentage points, 95% CI [-0.07, 0.87]), and overall cancer diagnosis (HR: 1.00, 95% CI [0.95, 1.06]; P = 0.921, incidence difference: 0.04 percentage points, 95% CI [-0.80, 0.88]) attenuated to the null. For other site-specific cancers, sibling comparisons results varied, with more attenuation for melanoma, kidney, stomach, bladder, pancreas, and liver, bile ducts, and gallbladder cancer, while associations with lung, colon, head and neck, and esophagus cancer seemed to attenuate less. The findings were confirmed through an extensive set of sensitivity analyses. The main limitations of this study include the lack of inclusion of female participants, lack of data on other risk factors such as smoking, alcohol consumption, and physical activity, and only adjustment for the unobserved confounders which are shared between full siblings.
Higher levels of adolescent cardiorespiratory fitness were associated with lower overall cancer mortality in late adulthood, a finding that persisted in sibling comparisons. However, the influence of unobserved familial confounding appeared to vary by cancer type and be more pronounced for cancer diagnoses than for mortality. This may suggest a need for robust causal methods to triangulate results, rather than relying on correlations alone, to better inform public health efforts.
