Simple nucleic acid detection methods could facilitate the progress of disease diagnostics for clinical uses. An attractive strategy is label-free surface-enhanced Raman scattering (SERS) due to its capability of providing structural fingerprinting of analytes that are close to or on nanomaterial surfaces. However, current label-free SERS approaches for DNA/RNA biomarker detection are limited to short and synthetic nucleic acid targets and have not been fully realized in clinical samples due to two possible reasons: (i) low target copies in limited patient samples and (ii) poor capability in identifying specific biomarkers from complex samples. To resolve these limitations and enable label-free SERS for clinical applications, we herein present a novel strategy based on multiplex reverse transcription-recombinase polymerase amplification (RT-RPA) to enrich multiple RNA biomarkers, followed by label-free SERS with multivariate statistical analysis to directly detect, identify and distinguish between these long amplicons (∼200 bp). As a proof-of-concept clinical demonstration, we employed this strategy for non-invasive subtyping of prostate cancer (PCa). In a training cohort of 43 patient urinary samples, we achieved 93.0% specificity, 95.3% sensitivity, and 94.2% accuracy. We believe that our proposed assay could pave the way for simple and direct label-free SERS detection of multiple long nucleic acid sequences in patient samples, and thus facilitate rapid cancer molecular subtyping for personalized therapies.