Advanced risk signature analysis of inflammation markers in predicting prostate cancer using the Swedish Apolipoprotein-related MOrtality RISk (AMORIS) cohort

Abstract

Background

Elevated post-diagnosis levels of C-reactive protein (CRP) and haptoglobin, and low albumin levels, have been associated with poor prostate cancer (PCa) prognosis. Advanced techniques are needed for biomarker-based cancer risk prediction. We evaluated PCa risk using Cox models and risk signature analysis in the Swedish Apolipoprotein-related MOrtality RISk (AMORIS) cohort.

Materials and methods

AMORIS includes biomarker data on >800 000 individuals from 1985 to 1996 in primary care and occupational setting, linked to national health and population registers through 2020. PCa risk was analysed using Cox proportional hazard models for albumin, CRP, haptoglobin and white blood cells at the third time point, adjusting for age, socioeconomic status, education level, Charlson comorbidity index (CCI) and cancer history, and risk signature analysis with training and validation sets (preventing overfitting) including repeated biomarker measurements, covariate interactions and baseline factors. Sensitivity analysis categorised age and CCI.

Results

Cox model showed elevated CRP and CCI ≥2 significantly increased PCa risk, with age consistently predictive. Risk signatures confirmed age as the dominant risk factor (hazard ratio 1.15 per year, 95% confidence interval 1.13-1.18) and highlighted interaction effects: younger men with cancer history had higher PCa risk, while elevated CRP with high CCI amplified risk, demonstrating strong predictive accuracy (receiver operating characteristic area under the curve: 0.82; Harrell’s C: 0.72). Categorising age and CCI further refined risk stratification.

Conclusion

Although elevated CRP was associated with higher PCa risk, no clear associations were detected for other markers. Advanced risk analysis found age to be the sole predictor, indicating conventional methods may overestimate biomarker roles in PCa prediction.