Prostate cancer is the most commonly diagnosed non-skin cancer in men [1]. It is also the cancer for which the phrase "you have cancer" carries the widest possible range of meanings.
At one end are tumours that will kill. At the other are tumours that a man will die with rather than of — cancers so slow that, left entirely alone, they would never have produced a symptom in his lifetime. Both are histologically malignant. Both look like cancer to a pathologist. And the second kind is common enough that finding them is a genuine harm, because a man told he has cancer will usually be treated, and the treatment has permanent consequences [2] [3].
This is the disease where detection and benefit come apart most sharply, and where the central clinical skill is not finding cancer but deciding which cancers to leave alone. The centerpiece quantifies that.
Start here: what prostate cancer is
The prostate is a walnut-sized gland below the bladder that contributes fluid to semen. Its growth and maintenance depend on androgens — testosterone and its more potent derivative dihydrotestosterone — and so, crucially, does the growth of most prostate cancers. That dependence, established in 1941 by Huggins and Hodges in work showing that castration and estrogen affected the disease [4], is the foundation on which almost all systemic therapy still rests.
Most prostate cancers are adenocarcinomas arising in the peripheral zone. They typically cause no symptoms while confined to the gland; urinary symptoms in older men are far more often due to benign enlargement. When prostate cancer does cause symptoms, it is usually because it has spread — characteristically to bone.
Risk rises steeply with age, and is higher in men of African ancestry and in those with affected first-degree relatives. The genetic architecture is polygenic, with association analyses of more than 140,000 men identifying 63 new susceptibility loci [5] alongside a smaller set of rarer high-penetrance variants in DNA-repair genes that matter therapeutically [6].
Pillar 1: measurement and diagnosis
PSA, and what it is actually telling you
Prostate-specific antigen is a protein made by prostate tissue — normal, enlarged, inflamed and malignant alike. It is organ-specific, not cancer-specific, which is the root of every difficulty that follows. A raised PSA can mean cancer; it can equally mean benign enlargement, prostatitis, or recent instrumentation. A normal PSA does not exclude cancer.
The screening debate, stated honestly
Two large randomised trials dominate the evidence and reached different headline conclusions.
The European trial (ERSPC) randomised 162,243 men aged 55 to 69 and found, at a median of 9 years, a prostate-cancer death rate ratio of 0.80 (95% CI 0.65 to 0.98) — a 20 percent relative reduction — with an absolute risk difference of 0.71 deaths per 1,000 men. Its own summary is the fairest one-line statement available: PSA screening "reduced the rate of death from prostate cancer by 20% but was associated with a high risk of overdiagnosis" [7]. Longer follow-up strengthened the mortality finding [8].
The American trial (PLCO) found no significant mortality benefit — a result heavily complicated by the fact that a large proportion of men in its control arm were screened anyway, so it substantially compared screening with screening [9]. A large UK trial of a single low-intensity PSA invitation likewise found no significant mortality reduction [10].
The evidence therefore supports a modest mortality benefit at a substantial cost in overdiagnosis, which is why guidance has converged on shared decision-making rather than blanket recommendation or blanket rejection [11] [12].
Grading, which matters more than detection
Once cancer is found, the single most important number is its grade. The Gleason system scores the two most prevalent architectural patterns and sums them, and its long-standing defect was that the lowest score used in practice was 6 out of 10 — which, as the 2014 ISUP consensus conference noted plainly, "implies that their prognosis is intermediate and contributes to their fear of having a more aggressive cancer" [13].
The replacement is Grade Groups 1 to 5, validated in over 20,000 radical prostatectomy specimens, over 16,000 needle biopsies and over 5,000 biopsies followed by radiation, and adopted by 90 percent consensus. The stated rationale includes, explicitly, that "the lowest grade is 1 not 6 as in Gleason, with the potential to reduce overtreatment of indolent cancer" [13] [14], and the system has since been validated in population-based cohorts [15] and updated [16] [17].
Grade Group 1 is the lowest of five prognostically distinct groups, and the explicit purpose of renaming it from "Gleason 6 out of 10" was to make leaving it alone a conversation that could actually be had [13]. That renaming, more than any imaging or biomarker, is what made active surveillance clinically sayable.
MRI before biopsy
The traditional pathway sent every man with a raised PSA to a systematic 10-to-12-core transrectal ultrasound-guided biopsy [18], which sampled the gland blindly: it missed significant cancers in some men and found insignificant ones in others.
Multiparametric MRI changed this, and the trial evidence is unusually clean. In PRECISION, 500 men with suspected prostate cancer were randomised to MRI with targeted biopsy or to standard biopsy. In the MRI arm, 71 of 252 men (28 percent) had a non-suspicious MRI and avoided biopsy altogether. Clinically significant cancer was detected in 38 percent versus 26 percent — and, critically, clinically insignificant cancer was diagnosed 13 percentage points less often [18]. PROMIS had established the diagnostic accuracy underpinning that strategy [19], with further work on how to combine targeted and systematic sampling [20] [21].
That is the rare intervention that finds more of what matters while finding less of what does not.
Centerpiece: a simple simulatable model of what screening does
The screening debate is usually conducted with adjectives. It can be conducted with arithmetic, because the European trial reported enough numbers to constrain each other.
It reported an absolute risk difference of 0.71 prostate-cancer deaths per 1,000 men, cumulative prostate-cancer incidence of 8.2 percent in the screened group against 4.8 percent in the control group, and the consequence that 1,410 men would need to be screened and 48 additional cases treated to prevent one death [7].
Those are not four independent facts. The number needed to screen is one over the absolute risk difference; the number of additional cases is that figure multiplied by the excess cumulative incidence. Both can be recomputed, and both reproduce: 1,408 against a published 1,410, and 47.9 against a published 48.
Scaled to 1,000 men invited, the balance sheet is stark: 34 extra cancers detected, against 0.71 deaths prevented — 48 additional diagnoses for every death averted.
The obvious question is how many of those 34 are real cancers found earlier and how many would never have surfaced at all. A simulation model built on the Rotterdam arm of the same trial estimated the overdetection rate — the share of screen-detected cancers that would never have been diagnosed without screening — and found it depends heavily on how screening is done. For a single screen at age 55 it was 27 percent, with a mean lead time of 12.3 years. For four-yearly screening from 55 to 67 it was 48 percent; for annual screening over the same ages, 50 percent. For a single screen at age 75 it was 56 percent, with the lead time halved to 6.0 years. Extending screening to age 75 "would result in at least two cases of overdetection for every clinically relevant cancer detected" [22].
That last relationship is the teaching point, and it is counterintuitive. More screening does not merely find more cancer; it finds proportionally more harmless cancer. The reason is length bias: slow-growing tumours spend far longer in a detectable-but-asymptomatic state, so any screening test preferentially catches them, and the more often you look the more the yield is dominated by the tumours that were never going anywhere. Screening older men makes it worse still, because the lead time needed to matter exceeds their remaining life expectancy.
The resolution is not to abandon detection but to decouple it from treatment. If roughly half of what screening finds is harmless, and the grading system now names the lowest group in a way that permits inaction [13], then the harm of overdiagnosis is almost entirely the harm of overtreatment — and that is avoidable by not treating. This is precisely what active surveillance does, and why it is the single most important development in the management of this disease.
Three honest limits. The excess incidence is measured at a median of nine years, so part of it is lead time rather than overdiagnosis — cancers advanced in date, which the control arm would eventually reach — making 34 an upper bound at that horizon; the overdetection estimates are lifetime model projections from a related but not identical population, so combining them illustrates magnitude rather than measuring one quantity. Corrections for lead-time and length bias are themselves methodologically contested [23] [24] [25]. And these figures come from a screening programme run in the 1990s and 2000s, before MRI-first pathways, which demonstrably reduce insignificant-cancer detection [18] — the modern balance sheet should be better than the one drawn here.
Pillar 2: treatment
Active surveillance
For low-risk, low-grade, localised disease the standard of care is now to monitor rather than treat: serial PSA, repeat imaging and biopsy, with definitive treatment triggered by evidence of progression. Long-term series of conservatively managed localised disease showed the natural history was frequently indolent [26], and the grade-group system exists partly to make this conversation possible [13].
The advantage is that a man with Grade Group 1 disease keeps his continence and erectile function, which definitive treatment frequently costs [2] [3]. The cost is the burden of monitoring and of living with an untreated cancer, which is not nothing.
Definitive treatment for localised disease
Radical prostatectomy removes the gland. Its benefit over conservative management was established in SPCG-4: at 15 years, cumulative prostate-cancer mortality was 14.6 percent with surgery against 20.7 percent with watchful waiting, a relative risk of 0.62, with a number needed to treat of 15 overall and 7 for men under 65 — and the benefit was confined to men younger than 65 [27] [28] [29].
That age restriction is the whole risk-stratification argument in one finding: the same operation that saves lives in younger men does not measurably help older ones, because they die of something else first.
Radiation therapy, delivered as external beam or brachytherapy, is the alternative with broadly comparable cancer outcomes and a different side-effect profile — more bowel toxicity, less immediate incontinence.
Neither is free. Urinary and sexual dysfunction after prostatectomy are common and durable [2], quality-of-life and satisfaction data across modalities have been collected systematically [3], and surgical approach affects the profile without eliminating it [30]. These harms are the reason overdiagnosis matters: they are what an overdiagnosed man is exposed to for no possible benefit.
Advanced disease: taking the androgens away, then away again
Androgen deprivation therapy is the backbone, and it works because of Huggins' finding [4]. It is achieved medically rather than surgically now, and it reliably controls the disease — for a while.
Castration-resistant disease is what emerges when the tumour adapts, and the key insight of the last fifteen years is that it is usually still androgen-driven, via intratumoural androgen synthesis and receptor amplification. Two drug classes exploit that. Abiraterone inhibits CYP17 and shuts down androgen synthesis wherever it occurs, including within the tumour — a phase I trial confirmed that castration-resistant disease remains hormone-driven [31] and it extended survival after chemotherapy [32]. Enzalutamide is a potent androgen-receptor antagonist, with survival benefit after chemotherapy [33] and before it [34]. Moving these agents earlier, into hormone-sensitive metastatic disease, improved survival again [35] [36].
Chemotherapy with docetaxel was the first agent to extend survival in castration-resistant disease [37], with cabazitaxel available after it [38]. Immunotherapy with sipuleucel-T showed a survival benefit without the expected PSA response [39]. Bone-targeted agents address the skeletal complications that dominate advanced disease [40] [41] and the bone loss that androgen deprivation itself causes [42].
PARP inhibitors exploit a genomic vulnerability: a substantial minority of men with metastatic prostate cancer carry defects in DNA-repair genes including BRCA2, and olaparib produced responses concentrated in exactly those men [43] [44] [45], with rucaparib showing the same in BRCA-altered disease [46]. This is the disease's first genuinely genomically-selected therapy, and it makes germline and tumour sequencing part of advanced-disease care.
PSMA-targeted radioligand therapy is the newest mechanism. Prostate-specific membrane antigen is abundant on prostate cancer cells, and attaching a beta-emitting isotope to a PSMA-binding ligand delivers radiation to metastases wherever they are. In a randomised trial, [177Lu]Lu-PSMA-617 was compared with cabazitaxel in metastatic castration-resistant disease [47]. The same target also supports imaging, so the diagnostic and therapeutic agents are matched — the property that makes it a theranostic.
Pillar 3: what is unresolved
How to screen without overdiagnosing. The most promising answer is not a better PSA threshold but a better pathway: risk-stratified testing, MRI before biopsy to avoid sampling men who do not need it [18] [19], and biomarkers to triage further. Whether this shifts the balance sheet enough to change population policy is the open question [11] [12].
Who can safely stay on surveillance, and for how long. The grade-group system stratifies well [13] [15], but predicting which Grade Group 1 cancers harbour unsampled higher-grade disease remains imperfect.
Sequencing the advanced-disease agents. With hormonal agents, taxanes, PARP inhibitors and radioligands all active, the order is largely unresolved [48] [49], and cross-resistance between the hormonal agents is real.
Genomic selection. DNA-repair defects identify PARP-inhibitor responders [43] [46], and the androgen receptor itself may induce a BRCA-like state exploitable by combination [50] — but most men have no actionable alteration.
Dig deeper in lmmol
Prostate cancer belongs with the other screened cancers in this collection, and the comparison is the point. Breast cancer has the same structure of argument — a real mortality benefit, a substantial overdiagnosis cost, and a long literature on separating the two [23]. Colorectal cancer is the instructive contrast: screening there removes precursor lesions and so prevents cancer rather than merely finding it earlier, which is why its benefit-to-harm ratio is far more favourable and why the debates are quieter. Androgen deprivation causes bone loss and fractures, connecting directly to osteoporosis [42]. The lifelong-monitoring logic of active surveillance has a close parallel in inflammatory bowel disease, where objective targets replaced symptom-led management, and the burden of living with an untreated diagnosis connects to depression. The androgen receptor that drives this disease and its treatments is at androgen receptor. The full collection is at health.