Simple yet (more?) effective. Venous thromboembolism risk assessment model for germ cell tumour patients receiving first‐line chemotherapy

Abstract Background Germ cell tumours (GCT) are highly curable malignancies. Venous thromboembolism (VTE) is a serious complication, needing better risk assessment models (RAM). Aim Identification of VTE incidence and risk factors in metastatic GCT patients starting first‐line chemotherapy. Developing a RAM and comparing it to Khorana risk score (KRS) and Padua Prediction Score (PPS). Material and methods We retrospectively analysed GCT patients staged IS–IIIC. VTE risk factors were identified with logistic regression. Area under curve of receiver operating characteristic (AUC‐ROC), Akaike and Bayesian Information Criteria (AIC, BIC) were calculated for the developed RAM, KRS and PPS. Results Among 495 eligible patients, VTE occurred in 69 (13.9%), including 40 prior to chemotherapy. Vein compression (OR: 8.96; 95% CI: 2.85–28.13; p < 0.001), clinical stage IIIB‐IIIC (OR: 5.68; 95% CI: 1.82–17.70; p = 0.003) and haemoglobin concentration (OR for 1 g/dL decrease: 1.32; 95% CI: 1.03–1.67; p = 0.026) were significant in our RAM. KRS ≥ 3 (OR: 3.31; 95% CI: 1.77–6.20; p < 0.001), PPS 4–5 (OR: 3.06; 95% CI: 1.49–6.29; p = 0.002) and PPS > 5 (OR 8.05; 95% CI 3.79–17.13; p < 0.001) correlated with VTE risk. Diagnostic criteria (AUC‐ROC, AIC, BIC) for the developed RAM, KRS and PPS were (0.885; 0.567; −1641), (0.588; 0.839; −1576) and (0.700; 0.799; −1585), respectively. In the numerical score, the optimal cut‐off point for high‐risk was ≥9, with sensitivity, specificity, positive and negative predictive value of 0.78, 0.77, 0.35 and 0.96, respectively. Conclusions Our RAM, based on vein compression, clinical stage and haemoglobin concentration proved superior to both KRS and PPS. VTE is frequent in GCT patients.


| BACKGROUND
Germ cell tumours (GCT) account for the majority of malignancies in men aged 15-44. 1 They mostly originate from testicles (testicular germ cell tumours, TGCT) and, in 5% cases, from extragonadal tissues localised in the body's midline, that is, retroperitoneal space, mediastinum or central nervous system (CNS). 2According to Znaor et al, 1 an increase in GCT incidence is prognosed in 20 European countries.Until 2035, a total number of GCT patients is bound to increase by 21%, 13% and 32% in Northern, Western and Eastern Europe, respectively.A decrease by 12% is prognosed for Southern Europe. 1 Curability of GCT is high; it amounts to 100% in clinical stage (CS) I (disease limited to the testis).Multiagent chemotherapy is also curative in disseminated cases so that 80% of patients achieve a durable remission. 2To maintain these outcomes, referring patients to centres of excellence is recommended, particularly advanced or recurrent cases. 1 Due to the good prognosis and patients' young age, there has been a shift towards reducing treatment burden, as well as cancer-unrelated morbidity and mortality.Venous thromboembolism (VTE), comprising deep vein thrombosis (DVT) and pulmonary embolism (PE), has been a marked villain there. 3Nuances in pathophysiology (e.g.involvement of cancer cells, pro-coagulant state, oncological treatment) and clinical issues (e.g.frequent thrombocytopenia, bleeding, central venous access devices) support the term cancer-associated thromboembolism (CAT). 3,4TE may be the first symptom of cancer 5,6 ; 4%-20% cancer patients are predicted to experience VTE at some stage of their disease.VTE events within the first year from diagnosis confer poorer prognosis.4 A broad spectrum of clinical presentations, from asymptomatic cases or incidental computed tomography (CT) findings to sudden cardiac arrest, 7,8 makes VTE diagnosis challenging.Therefore, VTE risk assessment and thromboprophylaxis are crucial, especially in curable malignancies. Khoana risk score (KRS) 9 and Padua Prediction Score (PPS) 10 are among the most recognised risk assessment models (RAM).However, KRS identifies high-risk patients better and is still suboptimal in low-and intermediate-risk groups. Manwhile, it is in the latter that majority of VTE events occur.11,12 The established RAMs will remain the benchmark for future studies in this field.However, we feel that a more precise CAT prediction should rely on newer scores, specific for different primary sites and histopathology.

| AIM
The aim of this study was to determine VTE incidence and clinical presentations among GCT patients commencing first-line chemotherapy, and to identify potential histological, clinical and laboratory risk factors.We also aimed to assess the performance of KRS and PPS in this cohort, and, finally, construct a RAM specifically dedicated to GCT patients. 3| MATERIAL AND METHODS

| Patients' selection
We retrospectively searched our clinical databases for male patients with gonadal and extragonadal (retroperitoneal, mediastinal and intracranial) GCT (International Classification of Diseases, ICD-10 codes: C62, C48, C38 and C71, respectively) who had undergone first-line chemotherapy at our institution between 2011 and 2019.Patients with disseminated disease, that is, CS IS-IIIC according to TNM (eighth edition), who commenced multiagent chemotherapy or at least prephase (if applied) were eligible.Of note, TNM classifications for primary retroperitoneal (C48) and mediastinal (C38) tumours are designed mainly for soft tissue sarcomas.Moreover, tumours of the central nervous system (C71) are not routinely staged with TNM system; their dissemination is defined as radiological lesions along the cerebrospinal axis or positive cerebrospinal fluid cytology.For the sake of this study, we adapted TNM used for TGCT (C62) to describe extragonadal GCT as well (i.e.T0 N0-3 M0-1 S0-3) to maintain consistent staging.
The study was approved by the Ethical Committee at Maria Sklodowska-Curie National Research Institute of Oncology, Warsaw, Poland (consent no.16/2021; 25 March, 2021).

| Material
The following data were extracted: Due to the retrospective nature of this study, VTE was reported: a.On CT, performed routinely for staging or follow-up purposes (in asymptomatic cases).b.On Doppler ultrasound, performed when DVT was suspected clinically.c.On CT angiography, performed when PE was suspected clinically.d.In medical records from the follow-up period.
The medical records were screened for VTE up to 6 months after the end of first-line chemotherapy or until residual lesions surgery, second-line chemotherapy, any radiotherapy or patient's death, whichever occurred first.Variables were coded as continuous (age, BSA, BMI, dimensions of retroperitoneal and mediastinal lesions; all laboratory parameters), ordinal (IGCCCG risk group, PPS: >5 and 4-5 vs. 3) or binary (CS: IIIB-IIIC vs. IS-IIIA, KRS: 3-5 vs. 1-2, and all the remaining).

| Statistics
Stata®15.1 software was used.Potential VTE risk factors were compared between two subgroups, that is, patients with and without VTE.The final set of patients was determined by a dynamic selection of variables, including these with missing data.Significant variables were indicated by the algorithm, which then excluded patients with missing data in these very variables.
VTE risk factors were identified by logistic regression, with significance threshold of 0.05.Variables were first subjected to univariate analyses.
The number of variables subjected to multivariate analyses was limited by the number of VTE events (one variable per 10 events).They were chosen on the basis of univariate analyses and subject-matter-knowledge approach.We analysed six models of seven variables each, with p < 0.05 as significance threshold.All the 'complete models' were subjected to forward stepwise regression, with 0.3 as significance threshold, to form three new 'selected models'.The best 'complete' and 'selected' model was compared to KRS and PPS in terms of area under curve of receiver operating characteristics (AUC-ROC) statistics as well as Akaike and Bayesian information criteria (AIC and BIC).We applied the following interpretation of AUC-ROCs: 0.5 -random, 0.5-0.7 -low, 0.7-0.8acceptable,0.8-0.9-good/very good, >0.9 -excellent predictive value. 13ubsequent analyses were conducted post hoc, after the best-fitting model was found.It comprised Hgb as a continuous variable.We aimed to investigate whether converting Hgb into a binary variable would yield better VTE prediction.Hgb values were arranged according to quartiles.This enabled to assess the coefficient of Hgb as the only explanatory variable in each quartile.Moreover, Hgb concentration was subjected to locally weighted scatterplot smoothing analysis (LOWESS).Estimates of binarily coded Hgb were then introduced into multivariate models and compared to the original model with Hgb as a continuous variable, with odds ratios (OR) calculated for 1 g/dL decrease.
Finally, we assigned weights (according to regression coefficients and odds ratios in multivariate analyses) to the items of the proposed RAM.A cut-off point for low/ high VTE risk was estimated and remains to be confirmed in a validation cohort.
Ordinal variables were compared with chi-squared test; continuous variables were first checked for normality of distribution with Shapiro-Wilk test.Further comparisons were conducted with Student's t test or Mann-Whitney-Wilcoxon test, for variables with normal or non-normal distributions, respectively.
Continuous variables were presented as means and standard deviations; numerical variables-as numbers and percentages.
In 40 of 69 patients (58.8%),VTE was diagnosed prior to chemotherapy and in 29 patients (42.0%) after the start of chemotherapy.In the latter, median time from Day 1 to VTE event was 49 (1-147) days.
Of 322 patients not receiving extended LMWH prophylaxis, 54 (16.8%) experienced VTE.Notably, this subgroup comprised 40 patients with VTE on presentation (hence, without prior prophylaxis) and 14 patients who commenced chemotherapy and had VTE during the treatment.Of 173 patients with extended thromboprophylaxis during chemotherapy, VTE was diagnosed in 15 (8.7%).
The 'complete' model, with AUC-ROC 0.885, AIC 0.567 and BIC (−1641), fared better than the 'selected' model.Both, however, yielded better results in terms of VTE risk prediction than either KRS or PPS (Figure 1).Moreover, overfitting statistics also favoured tested models.

| LMWH prophylaxis
In all six models, LMWH prophylaxis significantly reduced VTE risk.The magnitude of LMWH effect was similar in all the models; in Model 2, OR for VTE risk was 0.04 (95% CI: 0.02-0.12;p = 0.001).

| Risk assessment score
All significant variables from Model 2, excluding LMWH prophylaxis, were used to develop a numerical risk score.As planned, it is on the basis of the risk score that the decision concerning thromboprophylaxis will According to the logistic regression coefficient as well as LOWESS analysis, the correlation between Hgb and VTE risk was nearly linear (Figures 2 and 3).In other words, OR only depended on the difference between any two Hgb concentrations and not on their absolute values.Hence, even within reference laboratory values for Hgb, the associated VTE risk may differ.
We did not find any cut-off value for Hgb that would improve the model's statistics.All estimated models with binary Hgb showed lower AUC-ROC and higher deviance (Table 7).Therefore, Hgb was left as a continuous variable.The risk score was then constructed on the basis of ORs and regression coefficients of all three variables (Table 8).
In 369 patients with complete data regarding vein compression, scores ranged from 0 to 19.In total, 190 patients (51%) scored 0; VTE incidence in this subgroup was 2.6%.
Interestingly, 58% VTE cases in our group were diagnosed prior to chemotherapy.25,30 In the group of Honecker et al, 14 81% of VTE cases occurred before first-or second-line chemotherapy.However, if first-line patients were considered separately, pre-chemotherapy VTE accounted for 54.5%, which is consistent with our paper.Patients' characteristics in the group of Tran et al. matched ours in many aspects, including VTE incidence -13%. 15Thromboprophylaxis was divided into short-and long-term (<7 and ≥7 days, respectively) but patients receiving long-term prophylaxis (7%) were then excluded.T A B L E 8 VTE risk assessment score for metastatic GCT patients.

Vascular compression (CT) 7
Clinical stage IIIB-IIIC 4 Haemoglobin concentration only if below the lower limit of normal: the difference between lower limit of normal and patient's concentration (LLN -[Hgb]), rounded up to integer 1 for every 1 g/dL Abbreviations: CT, computed tomography; GCT, germ cell tumour; Hgb, haemoglobin; LLN, lower limit of normal; VTE, venous thromboembolism.
Follow-up spanned 90 days, 15 half the period that we adopted; still, only 5.8% VTE events occurred in our patients after 90 days.VTE risk does not remain constant in the course of disease.According to Lauritsen et al, 20 hazard ratio (HR) for VTE was 24.7 (95% CI: 14.0-43.6)during the first year following BEP chemotherapy and decreased to 1.4 (95% CI: 0.9-2.2) after 10 years. 20We deliberately analysed prechemotherapy values only, as the goal of our study was to construct a RAM that would guide an upfront decision on thromboprophylaxis in new GCT patients.Still, some on-treatment circumstances may trigger VTE, for example, hospitalisation, sepsis or immobilisation.Periodic re-assessment of VTE risk and indications for thromboprophylaxis is therefore advised.
GCT histological entities have rarely been investigated separately as potential VTE risk factors.In our study, seminoma histology surprisingly decreased VTE risk whereas yolk sac tumour showed a contrary correlation.Other types proved insignificant.However, neither seminoma nor yolk sac tumour retained significance in multivariate analyses.Detailed histology was also investigated by Honecker et al 14 : seminoma correlated with a higher VTE risk but only in univariate analysis.
Most authors restricted themselves to seminomas and non-seminomas.According to Bezan et al, 19 nonseminomatous histology significantly increased VTE risk in univariate (HR: 4.23; 95% CI: 1.88-9.49;p < 0.0001) but not in multivariate analysis.Importantly, the analysis of CS IS-IIIC subgroup did not reveal any correlation from the start. 19Gizzi et al. did not show a correlation of nonseminoma histology with VTE risk (OR: 1.39; 95% CI: 0.43-4.56;p = 0.58). 28On the other hand, seminoma histology increased VTE risk in the groups of Robinson et al 22 (OR: 2.14; 95% CI: 1.11-4.14;p = 0.023) and Piketty et al 27 (OR: 3.4; p = 0.01), but significance was lost in multivariate analyses. 22,27Moreover, only 35% in the former group received chemotherapy. 22s expected, CS significantly correlated with a higher VTE risk, most probably due to the tumour volume, concentration of serum coagulants and vessels compression (analysed here as a separate factor).We assumed both CS IIIB-IIIC (in reference to CS IS-IIIA) and extranodal metastases to be potential VTE risk factors, which proved correct in univariate analyses.CS was then investigated in three of six RAMs and proved significant.Additionally, it retained its significance after the stepwise forward selection procedure, with a stronger correlation than in the available literature.In contrast, extranodal metastases, analysed in one model, lost their effect.
Our findings are in line with other studies.A higher VTE risk was correlated with CS III, 30 CS > I (OR: 16.95; 95% CI: 3.49-82.38;p < 0.001) 32 or CS ≥IIC (OR: 2.259; 95% CI: 1.105-4.618;p = 0.026). 16Moreover, comparable results were obtained for parenchymal extrapulmonary metastases (OR: 3.297; 95% CI: 1.353-5.077;p = 0.025) and regional lymph nodes ≥N2 (OR: 3.478; 95% CI: 1.126-10.745;p = 0.030). 16Bezan et al 19 investigated VTE risk in TGCT patients staged IS-IIC and IIIA-IIIC in reference to CS IA-IB.Both ranges of higher CS correlated with a higher VTE risk in univariate analysis, but this significance was only retained for CS IIIA-IIIC in multivariate analysis (HR: 4.87; 95% CI: 1.97-12.00;p = 0.001). 19onecker et al 14 found that metastases in retroperitoneal and supraclavicular lymph nodes correlated with VTE risk, contrarily to mediastinal, pulmonary and hepatic metastases.Only supraclavicular localisation remained significant in multivariate analysis (p = 0.03). 14][28] Intermediate and poor risk IGCCCG prognostic groups correlated with a higher VTE risk in univariate analyses; three of six models comprised this variable.Together with Hgb, prognostic group retained its significance in the forward step regression analysis.To the best of our knowledge, ours is the second study in which prognostic group yielded such results; most studies investigated TNM CS.These two classifications systems are overlapping to some extent.However, due to dissimilarities, we analysed CS and IGCCCG groups separately.Corresponding results were obtained by Bezan et al 19 ; intermediate or poor risk patients presented with a higher VTE risk (OR: 2.61; 95% CI: 1.16-5.88;p = 0.02), confirmed in a validation group. 19Srikanthan et al. 18 only subjected the prognostic group to univariate analysis.Intermediate or poor risk group significantly increased VTE risk (OR: 3.76; 95% CI: 1.50-9.46;p = 0.005) but not in a validation cohort. 18Similarly, Tran et al 15 observed an increase in VTE risk for intermediate (OR: 1.72; 95% CI: 1.08-2.74;p = 0.02) and poor risk group (OR: 3.26; 95% CI: 2.06-5.17;p < 0.001) but only in univariate analyses. 15ntermediate but not poor risk group increased VTE risk in the study of Honecker et al 14 ; this finding was not confirmed in multivariate analysis.The results may have been affected by including adjuvant chemotherapy and recurring cases (48% of the study population). 14][28] Deceleration of blood flow, usually due to vessels compression, has been a renowned VTE risk factors since Virchow's triad. 7,33In our univariate analyses, vessels compression yielded the highest OR (9.07; 95% CI: 4.37-18.84;p < 0.001) and was further investigated in four of six multivariate models.The significance was confirmed with even higher ORs: 11.28; 8.96; 9.37 and 12.23.However, analyses of this variable have limitations.Retrospectively, we were only able to analyse CT reports, not images.Many patients referred to our department had their CT scans performed elsewhere.High ORs obtained in the multivariate models indicated that vessels compression was independent of GCT-specific risk factors.We hence postulate that it must be stated directly in CT reports whether vessels compression is present or not.
LMWH prophylaxis in our study was administered in 21.7% patients in the VTE (+) group and in 37.1% in the VTE (−) group.In univariate analysis, LMWH reduced VTE risk approximately two-fold.In all multivariate models, this effect remained significant; in the best-fitting model (Model 2) OR was 0.04 (95% CI: 0.02-0.12;p < 0.001).However, administration of LMWH was left to the decision of the attending physician, based on renowned risk factors and RAMs, usually KRS.LMWH prophylaxis was not based on uniform criteria, medications and doses differed and intermittent administration was not infrequent.Therefore, our results on LMWH influence should be treated with caution.
Bezan et al 19 estimated number needed to treat (NNT) for GCT patients on prophylaxis for CS IA-IB, IS-IIB, IIC and IIIA-IIIC: 118, 34, 14 and 9, respectively.Number needed to harm (NNH) was 125. 19Fankhauser et al 23 obtained similar results; the benefit was even more pronounced in patients with central venous access. 23Both trials corroborate safety and efficacy of thromboprophylaxis.
On the other hand, many authors failed to reproduce these results.In GCT patients receiving in-hospital or extended thromboprophylaxis, Solari et al. found no correlation between LMWH administration and VTE incidence. 29n the studies of Gizzi et al 28 and Haugnes et al 21 thromboprophylaxis had no effect on VTE; moreover, bleeding risk was significantly increased in the LMWH group (14.5% vs. 1.1%; p < 0.001). 21][36][37] Anaemia has been a renowned VTE risk factor and one of the KRS items. 9,38In our cohort, mean Hgb was significantly lower (by 1.9 g/dL) in VTE subgroup.It was then analysed as a continuous variable in all six multivariate models and proved significant in five.In comparison with vessel compression and CS, anaemia was the weakest risk factor.
All patients in our study scored one point in KRS due to cancer type; hence, there were no low-risk patients.KRS ≥ 3 correlated with a higher VTE risk but AUC-ROC for KRS was inferior to our model (0.558 vs. 0.885).This might result from strong correlations of vessel compression and CS.Of all VTE (+) patients in our cohort, 26.1% (18 out of 69) had a high-risk KRS (≥3); the majority of VTE cases were diagnosed in intermediate-risk patients.This corresponds with meta-analysis of Mulder et al, 41 where 23.4% VTE (+) patients presented with a high-risk KRS.The authors concluded that KRS sufficiently identified high risk patients, especially within 6 months, but the prediction in the intermediate risk group was suboptimal. 41In our study, VTE incidence in the intermediate risk group was 11.7% (51/436), which is in line with meta-analysis of Bao et al -11%. 42These authors reported on fewer VTE cases among KRS high risk (≥3) patients -14% 42 versus 30.5% (18 out of 59) in our group.
Several studies addressed the utility of KRS in GCT patients.KRS ≥ 3 correlated with an increased VTE risk; OR ranged from 2.62 (95% CI: 1.28-5.35;p = 0.0008) 15 to 11.80 (95% CI: 3.93-35.39;p < 0.001). 18Bezan et al 19 applied the threshold of ≥2; after correction for chemotherapy, VTE risk was on the verge of significance (HR: 2.22; 95% CI: 1.02-4.85;p = 0.05).Furthermore, when corrected for CS, KRS was insignificant. 19In the study of Heidegger et al, only 7.7% of VTE cases occurred in patients scoring ≥3, 17 which was less than in the above-mentioned metaanalysis. 42Other authors 21,43 did not show a correlation between KRS and VTE risk in GCT patients at all.PPS < 4 and ≥4 identify low and high VTE risk patients, respectively.All our patients scored three points for an active malignancy and some scored additional two points for surgery within the last month.In result, even patients without other VTE risk factors scored five points only due to orchiectomy, bearing a low VTE risk.We therefore decided to analyse two cut-off levels, that is, three subgroups: PPS 3 (reference group), PPS 4-5 and >5.Both tested ranges correlated with a higher VTE risk: OR: 2.91 (95% CI: 1.22-6.92;p = 0.016) and 7.64 (95% CI: 3.17-18.38;p < 0.001), respectively.PPS proved inferior to our model (AUC-ROC: 0.700 vs. 0.885) but superior to KRS.
In the study of Germini et al, 44 PPS ≥ 4 significantly correlated with VTE risk (OR: 4.56; 95% CI: 2.68-7.74;p = 0.000) and allowed for a more accurate VTE prediction than individual assessment by attending physicians.However, there were few patients with active malignancies (3.8% and 6.9%). 44Zhou et al 45 found PPS superior (AUC-ROC: 0.716; 95% CI: 0.693-0.740)to Caprini risk score.The authors concluded that routine thromboprophylaxis in patients with PPS ≥4 (OR for VTE: 5.01; 95% CI: 4.03-6.25;p < 0.001) was justified.Moreover, PPS ≥4 correlated with higher mortality during hospitalisation and 6 months after (16.6%; p < 0.001). 45This effect was confirmed by Vardi et al 46 in hospitalised patients with sepsis, of whom 16.7% had a malignancy.PPS ≥4 correlated with in-hospital mortality (OR: 6.08; 95% CI: 3.73-9.92;p < 0.0001) and overall survival (OS) but not with VTE risk; correction for thromboprophylaxis did not affect these results.Moreover, VTE incidence did not correlate with mortality or OS. 46o the best of our knowledge, ours is the third study involving GCT patients where a RAM was developed and compared to established risk scores.Moreover, it seems to be the first in which a numerical scale was constructed.Srikanthan et al 18 proffered retroperitoneal lymph nodes (RPLN) >5 cm as a single factor for VTE risk assessment; it proved superior to KRS (AUC-ROC: 0.71 vs. 0.67 in the training cohort, respectively, and 0.61 vs. 0.57 in the validation cohort). 18In the study of Bezan et al, 19 VTE risk increased with CS: HR 3.45 (95% CI: 1.13-10.53;p = 0.03) for IS-IIB; 8.86 (95% CI: 2.35-33.45;p = 0.001) for IIC and 13.82 (95% CI: 5.88-32.51;p < 0.0001) for IIIA-IIIC.CS was more precise than RPLN in VTE prediction (AUC-ROC: 0.75 vs. 0.63; p = 0.007). 19Meng et al recommended thromboprophylaxis in GCT patients undergoing cisplatin-based chemotherapy in case of KRS ≥ 3, RPLN ≥ 5 cm, CS ≥IIC, intermediate IGCCCG risk group or LDH ≥ 1.5× upper reference limit.On absence of the above factors, prophylaxis was advised if chemotherapy was delivered through a central line.However, this paper was a systematic analysis of eight chosen studies and did not involve an original patients' cohort. 47omprising three rudimentary items, our risk score is straightforward and easy to operate.Whether its superiority over KRS and PPS can be sustained in a validation cohort remains to be seen.Our results echo other studies 18,19 ; VTE risk increased with CS and vessel compression (which is believed to occur more often in RPLN > 5 cm).Hgb was no surprise here; yet, strikingly, its correlation with VTE risk turned out to be approximately linear along any given concentration range.In other words, even within normal Hgb limits (14.0-18.0g/dL) VTE risk can change considerably.Post hoc analyses with Hgb as a binary variable resulted in decreasing model's fitness.In the literature, Hgb has been usually managed categorically, with a given cutoff point, e.g., 10 g/dL, as in KRS.Allegedly, Hgb treated as a continuous variable may be the key to better prediction of CAT.
The cut-off point of ≥9 could suggest possible target patients for prophylaxis; however, we should keep in mind that some patients scored 7 only due to vein compression, which, in line with Virchow's triad, turned out to be the strongest predictive factor.Recommendation of prophylaxis in patients ≥7 would also be sensible.Last but not least, our results must be confirmed in a validation cohort.
The main limitation of our study is its retrospective design.There were missing data, we could only rely on CT reports instead of scans and LMWH prophylaxis was a physician's subjective decision, hence the choice of drug and dosing were not uniform.The number of VTE events limited the maximal number of variables in multivariate analyses.Hence, some variables significant in univariate analyses were not further investigated.
All data were collected from one time point, that is, before chemotherapy.Such approach was in line with the aim to be accomplished, that is, assess VTE risk using baseline parameters.On the other hand, there has been more emerging evidence that dynamic re-assessment over the course of treatment results in better VTE prediction. 24,48,49trengths of our study include the population's size from one centre and the number of variables analysed.
Our VTE risk assessment model may have a substantial contribution to everyday practice in GCT clinic, that is, more targeted thromboprophylaxis.

| CONCLUSIONS
Our RAM based on vein compression, clinical stage and haemoglobin concentration proved superior to both KRS and PPS in terms of discriminatory power and risk of overfitting.A validation cohort is currently being collected and the numerical score will be further tested.Should our outcomes be corroborated, the presented risk score will have the highest predictive value so far in GCT patients.

ETHICS APPROVAL STATEMENT
The study conforms to the Declaration of Helsinki and was approved by the Ethical Committee at Maria Sklodowska-Curie National Research Institute of Oncology, Warsaw, Poland (consent no.16/2021; 25 March, 2021).Only retrospective data were analysed.All prior medical procedures, examinations of biological materials, diagnostic imaging, pathology, radiology and laboratory reports, as well as data from medical charts had constituted standard clinical management of germ cell tumour patients.

F I G U R E 1 F I G U R E 2
Receiver-operating characteristic (ROC) curves for the developed risk assessment model (RAM), Khorana risk score (KRS) and Padua Prediction Score (PPS).AUC, area under curve.Regression coefficient for haemoglobin concentration.g/dL, grams per decilitre.

F I G U R E 3
Locally weighted scatterplot smoothing (LOWESS).g/dL, grams per decilitre; num, number; VTE, venous thromboembolism.T A B L E 7 c-statistics and deviance for Hgb as a continuous or binary variable.-off value; g/dL) PPS ≥ 4, other surgery than orchiectomy prior to chemotherapy, yolk sac tumour histology, retroperitoneal primary, extranodal metastases, CS IIIB-IIIC, intermediate or poor risk according to IGCCCG, maximum dimension in the retroperitoneal space or in the mediastinum, vessels compression, Hgb decrease, Plt, WBC, Neutr and Mono, PLR, NLR, D-dimer, fibrinogen, AFP, aPTT and LDH; b.Negatively correlated: orchiectomy, seminoma histology, testicular primary, extended LMWH prophylaxis, G-CSF, MPV and Lym.
): a. Positively correlated: VTE history, ECOG PS 3-4, surgery or trauma within 1 month prior to chemotherapy, respiratory and cardiac insufficiency, infection or rheumatological disease (the last three being PPS items), KRS ≥ 3, Patients' characteristics (percentages of subgroups with complete data).
T A B L E 1 Univariate analyses -patients characteristics.Univariate analyses -laboratory parameters.