Stage I testicular germ cell tumour: relapse risk on surveillance

• Danish nationwide models (Wagner, J Clin Oncol 2024, seminoma n=924; Eur J Cancer 2024, non-seminoma n=453). All Danish patients, centrally reviewed pathology, surveillance only, diagnosed 2013 to 2018, follow-up to 2022. Internally validated (C-index 0.70 seminoma, 0.75 non-seminoma); no external validation. Seminoma model has no tumour-size term (size was not an independent predictor there). Hazard ratios are published; baseline risk is calibrated to the published 5-year endpoints.
• EAU seminoma groups (Boormans, Eur Urol Oncol 2024, n=1,016, nine institutions): tumour size ≤2, >2 to 5, >5 cm, rete testis invasion, LVI. 5-year relapse 8%, 20%, 44% (95% CI 6 to 11, 16 to 24, 27 to 66). Not population-based (nine institutions, diagnosed 1994 to 2019). Harrell's C-index 0.65 in the discovery cohort. In an independent validation cohort (n=285) 5-year relapse was 5% (low), 25% (intermediate) and 27% (high, only 8 patients) with C-index 0.72, so only low and intermediate were validated.
• SWENOTECA non-seminoma (Tandstad, J Clin Oncol 2009): 745 clinical stage I patients, 1998 to 2005; vascular invasion was the only factor used. On surveillance, 13.2% of 338 vascular-invasion-negative patients relapsed; only 12 invasion-positive patients were untreated. Shown by vascular invasion only.
• SWENOTECA seminoma (Tandstad, J Clin Oncol 2011): prospective population-based cohort from Sweden and Norway, 512 clinical stage I patients on surveillance, 2000 to 2006, median follow-up 5.0 years. Shown as the published 5-year relapse rate by vascular invasion (19.5% present, 13.0% absent; overall 14.3%), with its own, later timing. Rete testis invasion and LDH were not recorded, and no factor was significant, so it is a population anchor and not a risk model.
• LVI-only reference for non-seminoma: 5-year relapse 59% with LVI and 17% without, same Danish cohort.
• Primary RPLND panel is separate from the Danish model. It uses published series: pooled occult metastasis 16.9% without LVI (Blok, BJU Int 2020); pN0 outcomes (Baky, Clin Genitourin Cancer 2024, n=346) and relapse after pathologic stage II without adjuvant chemotherapy (Tachibana, J Clin Oncol 2022, n=97); pathologic stage II in 37% with one risk factor and 54% with both (verified, p=0.009; n=267 clinical stage I-IIA, 73% stage I), with 54% low-volume and 16% teratoma among pN+ (Stephenson, J Urol 2005); the 16.9% pooled occult-metastasis rate without LVI is verified from the Blok meta-analysis (24 studies; LVI+ 47.5%, OR 4.33; EC present 33.2% vs 16.2%; EC >50% 40.0% vs 20.0%), whose outcome combines relapse on surveillance and pN+ at RPLND; Stephenson J Clin Oncol 2005 (n=453) supplies the relapse rates by nodal stage and 54% with both predominant embryonal carcinoma and LVI, plus nodal volume, teratoma and relapse after RPLND (Stephenson, J Urol 2005, n=267); the German RPLND vs BEP trial (Albers, J Clin Oncol 2008, n=382); EAU guidelines. Hilar invasion and tumour size are not in these estimates.

Risk in the window from month a to b, given relapse-free at a, is (C(b) − C(a)) / (1 − C(a)), where C is cumulative relapse incidence. C(60) is the model's 5-year risk; the shape over time comes from a published timing distribution, not from patient-level data.

• Seminoma timing: median 12.7 months to relapse (Danish cohort 1.05 years; DaTeCa 13 months, IQR 7 to 25; Kollmannsberger 14 months). The model declares 47%, 76% and 87% of relapses by months 12, 24 and 36, against 44%, 74% and 86% in DaTeCa and 42%, 75% and 92% in Kollmannsberger. About half of 5-year relapses are declared by month 12 (Boormans: 4% of 8%, 10% of 20% and 30% of 44% at 1 year). A late component (about 6% of relapses after year 5) is fitted to the Princess Margaret landmark tables.
• SWENOTECA non-seminoma (Tandstad 2009; 338 vascular-invasion-negative patients on surveillance, 1998 to 2005): 13.2% relapsed at 5 years, against 17% for LVI-negative patients in the Danish cohort. Relapses came slightly later than in the Kollmannsberger data (median 0.7 years; 64% by 12 months and 90% by 24, against the model's 73% and 89%), so this column has its own timing (median 8.4 months). Only 12 vascular-invasion-positive patients stayed on surveillance (5 relapses, 41.7%), so that estimate is very imprecise and uses the pooled LVI-positive timing (n=5 cannot define one). Final staging was done 6 to 8 weeks after orchiectomy, which may move some patients to stage II before surveillance starts.
• SWENOTECA (Tandstad 2011; 512 surveillance patients 2000 to 2006) is the one population-based cohort that disagrees. Its crude 5-year relapse risk of 14.3% is above the model's average for a cohort with that LVI and hCG mix, and relapses came somewhat later (median 1.4 years; 69% by 24 months against 76%). It did not record rete testis invasion or LDH and found no significant predictor (tumour size over 4 cm 16.0% vs 14.1%; vascular invasion 19.5% vs 13.0%, P=.10; elevated hCG 17.0% vs 12.9%, P=.10), so it cannot test the Danish risk-factor effects. Relapses may be somewhat later and more frequent in this older Scandinavian era than the Danish model implies; treat the seminoma risks as possibly low by a few points.
• The seminoma Combined column pools the three sources on the log-hazard scale: its cumulative hazard at each month is 0.5 × Danish + 0.3 × EAU + 0.2 × SWENOTECA, each with its own timing. Weights are a judgement (Danish sees the most risk factors; EAU is pooled institutional data and its high group was not validated; SWENOTECA gives only crude rates by vascular invasion), not fitted to outcome data, and the sources are not fully independent. Treat it as a summary of the spread, not a validated prediction. The range chip shows the lowest and highest source. Non-seminoma has no combined estimate.
• The seminoma SWENOTECA column uses its own timing (single lognormal, median 16.8 months: 69% of relapses by 24 months, 3.8% after 60), because that cohort's relapses came later than in the Danish data. For both histologies its risk is the published 5-year relapse rate by vascular invasion status only; neither paper reports a combined risk model.
• Non-seminoma timing depends on LVI. LVI-positive: median 4 months, 90%, 95% and 98% of relapses by months 12, 24 and 36. LVI-negative: median 8 months, 72%, 89% and 93% (Kollmannsberger, 2,483 patients from five centres). Pooled, these reproduce the Danish median of 4.9 months and the DaTeCa timing (80%, 94%, 96%; Mortensen). Each shape keeps a late component because the Princess Margaret cohort declines more slowly than the Danish data.
• Beyond month 60 the curve is extrapolated and is least certain.

Princess Margaret figures are the 5-year conditional risks printed in Nayan et al. (Fig. 2 and Fig. 4), for a cohort treated in 1980 to 2014 with different imaging schedules. The model matches them within about 2 points except where the published group was very small. The DaTeCa and Kollmannsberger rows assume a case mix (30% LVI-positive with 55% 5-year risk, 70% LVI-negative with 20%, for the Danish row), because those papers did not report risk by group. Late risk for high-risk non-seminoma is probably overstated by the timing curve, so treat it as conservative.

Baseline is the 5-year risk at the first landmark (4 months for seminoma, 2 months for non-seminoma). The 95% intervals are wide for the small groups; see the paper.

• Outcome is relapse on surveillance, not death. Relapses are almost always cured; 5-year cancer-specific survival was 99.7% in the Canadian cohort of 2,483 patients.
• Time to relapse depends on the imaging and marker schedule used in each cohort. Different follow-up schedules shift detection times.
• Combinations of factors seen rarely (for example hilar soft tissue invasion with all other factors) rest on few patients.
• Pagetoid rete involvement is not counted as rete testis invasion. Pathology should follow ICCR reporting.
• Unknown inputs produce a range between the lowest and highest possible value.
• Individual 95% intervals cannot be computed from the published tables; published intervals are shown above where they exist.

• Wagner et al. Seminoma, J Clin Oncol 2024 and hazard-ratio table
• Wagner et al. Non-seminoma, Eur J Cancer 2024 (open access)
• Boormans et al. EAU risk groups, Eur Urol Oncol 2024
• Nayan et al. Conditional risk of relapse, Eur Urol 2017
• Mortensen et al. Late relapses, Eur Urol 2016
• Kollmannsberger et al. Surveillance outcomes, J Clin Oncol 2015
• Baky et al. Long-term outcomes of pN0 at primary RPLND, Clin Genitourin Cancer 2024 (doi 10.1016/j.clgc.2024.102217)
• Tachibana et al. Primary RPLND for pathologic stage II NSGCT, is adjuvant chemotherapy necessary? J Clin Oncol 2022;40:3762 (doi 10.1200/JCO.22.00118)
• Tandstad et al. Management of seminomatous testicular cancer: SWENOTECA, J Clin Oncol 2011;29:719 (doi 10.1200/JCO.2010.30.1044)
• Tandstad et al. Risk-adapted treatment in clinical stage I NSGCT: the SWENOTECA management program, J Clin Oncol 2009;27:2122 (doi 10.1200/JCO.2008.18.8953)
• Stephenson et al. (J Urol 2005, n=267; verified from the full paper) RPLND with embryonal predominance and/or LVI, J Urol 2005
• Stephenson et al. RPLND for NSGCT: impact of patient selection factors on outcome, J Clin Oncol 2005;23:2781 (n=453)
• Blok et al. LVI and embryonal carcinoma meta-analysis, BJU Int 2020
• Albers et al. RPLND vs one cycle BEP, J Clin Oncol 2008
• Tandstad et al. One course of adjuvant BEP, Ann Oncol 2014
• EAU Guidelines on Testicular Cancer, disease management