Prostate MRI Was Negative—What's Next?

Men with clinical suspicion of prostate cancer based on PSA or digital rectal exam can often safely avoid a biopsy if no suspicious lesion is detected on prostate multiparametric MRI (mpMRI). Their risk of harboring clinically significant PCa (csPCa) is even lower if PSA density (PSAD; i.e., PSA divided by prostate volume) is low (1). What is less clear is how these patients should be followed clinically at this point. Are they put back into the same category as the general, unscreened population? Are they to be followed more closely because we know neither MRI nor PSAD has perfect sensitivity? Answers to these questions have important implications for the individual patients and for leaders interested in implementing organized programs for screening or early detection of csPCa.

Reported in this issue, Pylväläinen and colleagues performed a large-scale retrospective analysis of men in Finland (2). By comparing a cohort of men with nonsuspicious MRI (nMRI) to population registry data, they were able to evaluate diagnosis rates for men in the nMRI group relative to the general population. Limitations of retrospective data, of course, apply. The authors accurately note that it is likely the nMRI cohort was followed clinically more closely than the general population and therefore diagnosis rates were higher. The nMRI cohort also had some clinical indication for prostate MRI in the first place. Unsurprisingly, they had a higher risk of csPCa than the population, with age-standardized incidence rate ratio (IRR) for csPCa (over 2–5 years of follow-up) of 3.4 (95% CI, 2.8–4.1). The IRR was considerably lower (but still probably elevated) if PSAD was <0.15 ng/mL/cm³: 1.4 [0.9–2.0]. More interestingly, men with nMRI were at much higher risk of clinically insignificant PCa (ciPCa), with IRR 8.2 (5.8–11.3) overall and 7.1 (4.2–11.3) among those with low PSAD. In other words, they have a higher overall risk of prostate cancer compared with the general population, but their relative risk of overdiagnosis is much higher.

We currently lack prospective trials with long follow-up to guide us in how to manage patients with elevated PSA but nMRI, especially with respect to meaningful oncologic outcomes like metastasis-free survival. Thus, we need to be cognizant of two risks: (i) missing an aggressive cancer that was invisible on MRI and (ii) increasing overdiagnosis by overreacting to elevated PSA alone. The solution likely lies in comprehensive risk stratification by incorporating multiple pieces of information. For example, Pylväläinen and colleagues confirmed the well accepted utility of PSAD to further risk stratify patients. Prostate volume is readily measured on prostate MRI, making PSAD nearly universally available in this population. Additional risk factors and biomarkers could be combined to increase the precision and accuracy of risk estimates, including age, genetic features, ancestry, any prior biopsy results (nearly 37% of the nMRI cohort had a prior negative biopsy), and biomarkers derived from urine, serum, or additional imaging (3, 4). The authors of the work in question have incorporated MRI, PSAD, and the 4K Score into an ongoing randomized controlled trial, ProScreen, powered for prostate cancer mortality.

Age is particularly interesting in this clinical context. Incidence of both csPCa metastatic prostate cancer increase exponentially with age (5). However, early onset may have adverse long-term prognostic implications, and, importantly, MRI interpretation may be more difficult at younger ages. Pylväläinen and colleagues found that men in the nMRI cohort aged 50 to 59 years had an IRR for csPCa of 10.4 (7.3–14.5), versus 3.4 (2.8–4.1) for all ages. Thus, deintensification of clinical follow-up after nMRI may be more readily implemented among older patients (despite their baseline higher risk of csPCa).

If we can decide what we are looking for, we stand a better chance of finding it. Ultimately, we want to detect prostate cancer that is localized to the prostate (i.e., curable) but destined to become metastatic if left undetected. Unfortunately, it is not possible to know whether a given cancer is destined to become metastatic, and we are left to grapple with probabilities. The Gleason system carries powerful risk stratification, but both pathology and imaging are subjectively interpreted and heavily dependent on expertise, creating reproducibility problems. Moreover, migration of clinical practices over time lead to evolving significance of these presumed surrogates. Artificial intelligence (AI), if carefully applied, may usher in a new era of objective and reproducible results for both pathology and imaging. Genomic testing and pathology AI tools, combined with clinical factors, allow better prediction of metastatic risk (3). Using more accurate predictors of metastatic disease will allow us to better evaluate the accuracy of prostate MRI to detect truly aggressive cancers while minimizing overdiagnosis.

Finally, the biggest flaw in prostate MRI is not a lack of sensitivity but a lack of reproducibility in its interpretation. As the authors note, pooled negative predictive value of prostate MRI is quite high (91%; ref. 6) yet still varies widely across studies. MRI quality and reader expertise are significant concerns. Positive predictive value varies enormously, even among centers of excellence (7). Features like tumor location (e.g., anterior tumors may be more easily missed), age of the patient, and prior medical history (e.g., implanted devices, procedures, or medications for benign prostatic hyperplasia) may all impact sensitivity and reliability of MRI with unknown impact on future cancer risk after nMRI. Efforts to ensure high-quality imaging are expected to improve overall imaging performance (8). Biparametric (noncontrast) MRI might improve cost-effectiveness and adoption, though reader expertise has been even more important for noncontrast prostate MRI. Advanced MRI techniques improve radiologists’ accuracy and negative predictive value (9). These techniques may also be implemented as quantitative imaging that becomes more like a lab test: the ordering physician receives not only the radiologist's critical interpretation but also an objective, numerical probability of csPCa (10).

While we await full implementation of the exciting new developments discussed above, we still must answer the question for our current patients: what should be done for patients with nMRI? Pylväläinen and colleagues and others have produced data that suggest we must not interpret MRI in isolation. Consideration of age, PSAD, overall health, and other risk factors is key. Older patients with nMRI and PSAD <0.15 ng/mL/cm³ often do not need intense follow-up. Younger patients may warrant closer attention. And all patients with nMRI, especially those with low PSAD, must be carefully cautioned that they carry a significant risk of being diagnosed in the future with a low-grade prostate cancer that would be managed not with aggressive treatment but with active surveillance.

T.M. Seibert reports grants and personal fees from GE Healthcare; personal fees from Varian Medical Systems, Janssen; and personal fees and other support from Cortechs Labs outside the submitted work.