Systematic review and meta-analysis of the diagnostic accuracy of prostate-specific antigen (PSA) for the detection of prostate cancer in symptomatic patients

We included cross-sectional and cohort studies that reported paired data on the diagnostic accuracy of PSA for the detection of prostate cancer in symptomatic men, verified with the use of a reference test (prostate biopsy). We excluded studies if it was not possible to extract data for a complete two-by-two table for the target condition or if the patient cohort was only asymptomatic patients (i.e. a screening cohort). We did not restrict studies by publication date, country, or clinical setting.

The study population of interest was any patient with symptoms of a possible prostate cancer, with no history of the disease. We defined symptoms of prostate cancer as at least one of LUTS (nocturia, hesitancy, poor stream, incomplete voiding, double voiding, terminal dribbling, urgency, incontinence, frequency), haematuria, erectile dysfunction, or lower back pain. Symptoms may have been identified by a standardised tool, such as the International Prostate Symptom Score (IPSS), clinical coding, or through patient self-report. We did not exclude studies based on age of participants or study setting. Where studies included groups of both asymptomatic and symptomatic men, we included men in the symptomatic group.

The index test was prostate-specific antigen (PSA) in a peripheral blood sample, measured in nanograms per millilitre (ng/mL). We did not set an a priori PSA threshold for prostate cancer detection but instead extracted data based on the PSA thresholds used in each study.

The target condition was prostate cancer, regardless of Gleason grade or clinicopathological stage.

The reference test was a biopsy of the prostate with histological examination. We did not set an inclusion criteria on the basis of prostate biopsy approach used in studies, but this was recorded as part of the data extraction.

Medline Ovid, EMBASE, the Cochrane Central Register of Controlled Trials (CENTRAL), and Web of Science databases were utilised to identify relevant studies. Key search terms, informed by the Scottish Intercollegiate Guidelines Network (SIGN) search strategies and pre-existing systematic reviews in the field of prostate cancer, were combined with MeSH terms for each database search. Hand-searching of reference lists from included studies and snowballing techniques were performed to locate any other possibly relevant studies. Please see Additional file 1 for the search strategy used in this review.

The protocol for this systematic review and meta-analysis was registered with PROSPERO (CRD42021257783).

This systematic review was conducted following the PRISMA reporting guidelines for systematic reviews and meta-analyses [13]. A completed PRISMA checklist can be found in Additional file 2.

Published studies assessing the diagnostic accuracy of PSA in symptomatic patients reported high sensitivity and low specificity for the detection of prostate cancer. Eighteen of the included studies were undertaken in hospital urology outpatient populations, with one study focused on a symptomatic cohort within a population screening study. Importantly, there were no studies assessing the performance of PSA in a primary care population. Insufficient data was available to assess the diagnostic accuracy of PSA for clinically significant prostate cancer. Furthermore, all included studies had a high risk of bias in at least one QUADAS domain.

Harvey et al. [10] published a systematic review of the diagnostic accuracy of PSA for prostate cancer in European populations, focused on studies published between 1998 and 2008. Individual study level data from 10 included papers was reported, though without estimating a combined level of accuracy. They considered the accuracy of PSA for all prostate cancer types overall and showed a range of accuracy estimates similar to this study. Over half of the studies included in this review were published since the review by Harvey et al. A review of clinical features of prostate cancer in primary care by Young and colleagues [34] in 2015 identified one study from 1989 of 287 patients referred from primary care with bladder outlet obstruction, of whom 211 had a PSA test. High levels of sensitivity (89.5%) and specificity (90%) were reported, but Young and colleagues considered the true level of accuracy was likely to be lower given few patients with a normal PSA level had the reference test for prostate cancer.

This study benefited from a rigorous, focused, methodological approach in conducting the review. All clinical settings were eligible, ensuring we found as many relevant studies as possible. Most included studies employed PSA in a similar manner, using similar indications and diagnostic thresholds, allowing for cross-study comparisons.

The evidence for the association between lower urinary tract symptoms and prostate cancer, particularly clinically significant prostate cancer, is equivocal. A number of secondary care studies suggest that symptoms do not discriminate well between prostate cancer and benign prostatic hypertrophy [35, 36]. This assumption is largely untested in primary care populations and contrasts with studies showing that the majority of patients diagnosed with prostate cancer present to their GP with LUTS prior to diagnosis [37‐40]. This controversy also means that LUTS and other relevant symptoms may not be reported or be the focus of some potentially relevant studies of PSA for prostate cancer and may have limited the sensitivity of the search strategy employed. However, key papers were picked up by the database searches and the majority of PSA studies will likely be focused on screening in asymptomatic populations.

All included studies employed TRUS biopsy as a reference test, with some also including pathological data obtained from urological procedures on the prostate. TRUS biopsy is recognised as having poor sensitivity as a diagnostic test [41], owing to the inability to visualise lesions within the prostate resulting in a random sampling of the gland, and thus misclassification bias. Reporting of histological classification of prostate cancers was only included in three studies, and each presented this data differently. Insufficient data was available to determine a relationship between PSA and clinically significant prostate cancer, which is a crucial consideration for the optimal use of PSA for prostate cancer detection. Most included studies only performed the reference test on patients with a raised PSA or abnormal prostate examination, introducing partial verification bias. Therefore, the true sensitivity of PSA in symptomatic patients is unknown and likely to be lower than reported.

PSA is a commonly used test to assess for the presence of prostate cancer, mostly in a primary care setting, and is recommended as part of the assessment of patients with LUTS in national guidelines [42‐44]. The lack of primary care evidence for the use of PSA to detect prostate cancer is known and is not the only condition for which secondary care evidence has been applied to primary care guidance [45]. Even so, this is a major gap in knowledge, as spectrum bias means that secondary care data (or screening data) do not translate to primary care. High-quality studies in primary care populations are needed to fill this gap, and future studies should report not just on prostate cancer per se but on clinically significant cancer as well. The introduction of more accurate diagnostic tests for prostate cancer, including multiparametric magnetic resonance imaging [41], increases the need for better understanding of the role of PSA in the early detection of symptomatic prostate cancer. PSA performance could also be enhanced by incorporating additional relevant clinical data in multivariable risk models [46], although only one has been validated in primary care [47].

Primary care clinicians are generally aware of the limitations of PSA testing [48], and clinical guidelines encourage a balanced discussion with patients of the potential benefits and harms of relying on PSA to detect prostate cancer [3, 49]. The findings of this review suggest this is a pragmatic approach in providing care to patients with LUTS. False-positive PSA results can also occur from non-cancer conditions affecting the prostate such as benign prostatic hypertrophy or prostatitis, further limiting the clinical utility of the test for prostate cancer detection. Alternative tests to PSA have been extensively researched [50, 51], and some show promise of improving the level of confidence in detecting prostate cancer, though none has entered primary care practice as yet.