A novel risk score for venous thromboembolism in lung cancer patients: a retrospective cohort study

Introduction

Patients diagnosed with cancer are at a higher risk of developing venous thromboembolism (VTE) compared to the general population, with reported incidence rates ranging from 20% to 30%.¹ Among cancer types, lung cancer is particularly associated with thrombosis, with incidence rates ranging from 14% to 30.²

The risk of thrombosis is significantly increased in cancer patients, by 4 to 12 times higher compared to individuals without cancer, and further elevated to 6.5 to 23 times with the addition of chemotherapy or targeted therapy.^3–5

Various factors contribute to this increased risk, including patient-related factors such as advanced age, previous history of VTE, and obesity; tumor-related factors such as cancer type, stage, and aggressivenessr; and treatment-related factors like surgery, radiation, or systemic anticancer.^1,6

The reported incidence of VTE in the literature varies, possibly due to differences in study design, selection of study participants, varying definitions of VTE, and the exclusion of patients with previous thrombosis from most clinical trials.¹

The correlation between cancer and venous thromboembolism (VTE) has been extensively established and is known to have deleterious effects on both morbidity and mortality among cancer patients.⁵

VTE, including pulmonary embolism (PE) and deep vein thrombosis (DVT), as well as arterial thromboembolism, rank as the second most common cause of death in cancer patients.¹ Moreover, cancer patients with superficial vein thrombosis (SVT) face a high risk of death comparable to those with DVT.⁶

Preventing VTE is an essential component of managing lung cancer patients to improve their prognosis. Therefore, several medical scoring systems have been developed to predict the risk of VTE, such as the Khorana, Caprini, Vienna CATS, PROTECHT, and COMPASS-CAT scores.⁷ Nevertheless, these risk prediction scores have been developed with inherent limitations and have undergone limited external validation in lung cancer patients. Additionally, their applicability to the Tunisian population is restricted. Consequently, it is a priority to develop and validate a score specifically tailored to the Tunisian population. This will facilitate the guidance of prophylactic anticoagulation in patients at risk of VTE.

Methods

Results

Population characteristics and incidence of VTE

A total of 381 patients were included in our study after the exclusion of 20 patients (Figure 1). Thirty-nine patients had lung resection. Three hundred and nine patients (81.8% of cases) received first-line chemotherapy, which was combined with curative radiotherapy in 32.4% of cases.

Figure 1. Flow chart of patient inclusion and exclusion.

The incidence of VTE was 26.8% (102/381). It was similar in the development and validation groups (25.8% vs 27.6%; p=0.690). There was no significant difference in the incidence of pulmonary embolism (PE) (14.3% vs 14.6%; p=0.954), DVT (12.6% vs 12.5%; p=0.966), or SVT (1.1% vs 1.5%; p=0.731) between the 2 groups (Figure 2).

Figure 2. Incidence of VTE in patients followed for lung cancer.

The main characteristics of the two databases are summarized in Table 1.

Table 1. Characteristics of patients in the development group and the validation group.

Variables	Development group n=182	Validation group n=199	χ2	p
Sex
Female	30 (16.5)	21 (10.6)	2.884	0.089
Male	152 (83.5)	178 (89.4)
Age (≥65 years)
Yes	92 (50.5)	76 (38.2)	5.890	0.015
No	90 (49.5)	123 (61.8)
BMI (≥25 kg/m2)
Yes	32 (17.6)	51 (25.6)	2.561	0.110
No	106 (58.2)	111 (55.8)
WHO status (≥2)
Yes	29 (15.9)	28 (14.1)	0.283	0.595
No	152 (83.5)	171 (85.9)
TNM stage
I-II	17 (9.3)	15 (7.5)	0.402	0.526
III-IV	165 (90.7)	184 (92.5)
Lymph node status (≥N2)
Yes	131 (72)	127 (69.8)	0.092	0.761
No	51 (28)	50 (37.9)
Metastasis
Yes	113 (62.1)	130 (65.3)	0.432	0.511
No	69 (37.9)	69 (34.7)
Histological type
Small-cell carcinoma	24 (13.2)	27 (13.6)	0.012	0.913
Poorly or undifferentiated	21 (11.5)	19 (9.5)	0.401	0.527
non-Small-cell carcinoma
Squamous cell Carcinoma	48 (26.4)	46 (23.1)	0.543	0.461
Adenocarcinoma	84 (46.2)	95 (47.7)	0.096	0.757
Large cell carcinoma	3 (1.6)	7 (3.5)	1.300	0.254
Active smoking
Yes	101 (55.5)	115 (57.8)	0.204	0.652
No	81 (45.5)	84 (42.2)
Hypertension
Yes	35 (19.2)	24 (12.1)	3.735	0.053
No	147 (80.8)	175 (87.9)
Diabetes mellitus
Yes	26 (14.3)	33 (16.6)	0.383	0.536
No	156 (85.7)	166 (83.4)
Dyslipidemia
Yes	9 (4.9)	9 (4.5)	0.038	0.846
No	173 (95.1)	190 (95.5)
Coronaropathy
Yes	17 (9.3)	14 (7)	0.676	0.411
No	165 (90.7)	185 (93)
Cardiac failure
Yes	6 (3.3)	5 (2.5)	0.208	0.648
No	176 (96.7)	195 (97.5)
Renal failure
Yes	3 (1.6)	1 (0.5)	1.201	0.273
No	179 (98.4)	198 (99.5)
History of surgery
Yes	43 (23.6)	52 (26.1)	0.352	0.553
No	139 (76.4)	146 (73.4)
WBC (≥10 × 109/l)
Yes	70 (38.5)	102 (51.3)	6.781	0.009
No	112 (61.5)	97 (48.7)
Hb (<10 g/l)
Yes	22 (12.1)	20 (10.1)	0.402	0.526
No	160 (87.9)	179 (89.9)
Platelets (≥300 × 109/l)			0.034	0.854
Yes	95 (47.8)	100 (50.3)
No	87 (52.2)	99 (49.7)
Creatinine (>84 mmol/l)
Yes	40 (22)	52 (26.1)	0.789	0.374
No	128 (70.3)	134 (67.3)
C-Reactive protein (>5 mmol/l)
Yes	134 (73.6)	153 (76.9)	0.021	0.886
No	22 (12.1)	24 (12.1)
Tumor resection of lung cancer
Yes	18 (10.4)	21 (10.6)	0.045	0.831
No	164 (90.1)	178 (89.4)
Chemotherapy
Yes	49 (81.9)	160 (80.4)	0.230	0.632
No	32 (18.5)	39 (19.6)
Platinum based chemotherapy
Yes	142 (78)	151 (75.9)	0.246	0.620
No	40 (22)	48 (24.1)
Radiation
Yes	55 (27.6)	54 (29.7)	0.198	0.656
No	142 (71.4)	126 (69.2)

Development of the predictive risk score system for VTE by multivariate analysis

Determination of risk score system items

Variables with a significance level of p<0.250 were entered into the binary logistic regression.

A hemoglobin level <10 g/l, the presence of metastases, the histological type of poorly or undifferentiated NSCLC, and active smoking are the significant variables (p<0.05; therefore, they have been selected as the items of the score.

Based on the weight of the different regression coefficients, we established a risk score system as follows (Table 3):

Table 3. Predictive factors for VTE determined from the development group by Multivariate analysis.

Variables	Score	B	Wald	p	OR, IC 95%
Hemoglobin level <10 g/l	+1	1.508	4.258	0.039	4.520 [1.079-18.937]
Presence of metastasis	+1	1.197	5.670	0.017	3.311 [1.236-8.871]
Histological type poorly or undifferentiated NSCLC	+2	1.875	6.390	0.011	6.522 [1.524-27.911]
Active smoking	+1	1.431	9.309	0.002	4.183 [1.668-10.488]

The risk of VTE was significantly correlated with the risk score in the development group (Pearson contingency coefficient=26.757, p<10^-3).

Determination of risk groups

The cut-off of our VTE predictive score was identified via the ROC curve. Indeed, a total score <2 allows the classification of patients in the “low risk of VTE” group. The other risk classes were developed on the basis of the predicted incidence of VTE.

As a result, patients were classified into 3 risk groups (Table 4): “low risk” (score 0-1 [predicted incidence <29%, n=92]), “moderate risk” (score 2-3 [predicted incidence 29-43%, n=85]), and “high risk” (score 4-5 [predicted incidence >43%, n=5]).

Table 4. Classification of patients according to the predicted risk of the risk score and the actual incidence of VTE.

Score	Development group					Validation group
	Predicted incidence	Number of patients	Actual incidence (%)	Risk group	PPV/NPV (%)	Predicted incidence	Number of patients	Actual incidence (%)	Risk group	PPV/NPV (%)
0	4.7%	25	2 (8%)	low risk (<29%, n=92)	NA/87 Se=NA Sp=100%	10.1%	30	4 (13.33%)	low risk (<29%, n=101)	NA/82 Se=NA Sp=100%
1	18.1%	67	10 (14.9%)			21.4%	71	14 (19.71%)
2	38.8%	64	22 (34.4%)	Moderate risk (29-43%, n=85)	48/82.9 Se=80% Sp=52.7%	37.33%	76	27 (35.5%)	Moderate risk (29-43%, n=86)	80/63.3 Se=12.1% Sp=98%
3	53.1%	21	8 (38.1%)			40%	10	6 (60%)
4	83.9%	5	5 (100%)	High risk (>43%, n=5)	100/NA Se=100% Sp=NA	35.47%	11	4 (36.3%)	High risk (>43%, n=12)	NA/66.7 Se=NA Sp=100
5	0 %	0	0 (0%)			48.7%	1	0 (0%)

The incidence of VTE according to the 3 risk classes in the development group and in the validation group is detailed in Figure 3.

Figure 3. Distribution of VTE according to the 3 risk classes between the development group and the validation group.

The percentages of VTE in “low-risk” and “moderate-risk” patients are similar between the development and validation groups (low-risk: 13% vs 17.8%, moderate-risk: 35.3% vs 38.4%; respectively). In contrast, the percentage of VTE in “high-risk” patients is much higher in the development group compared to the patients in the validation group (100% vs 36.3%, p=0.012)

Validation of VTE predictive score

Internal validation

In the development group, the risk score system has good discrimination. It can distinguish “high-risk” from “low-risk” VTE patients (c statistic=0.703 [0.618-0.789], (Figure 4 A)). This prediction model also showed good calibration according to the Hosmer-Lemeshow test (χ²=2.381, p=0.882).

Figure 4. (A) ROC curve of VTE prediction model using the development group. (B) ROC curve of VTE prediction model using the validation group and after applying the bootstrap method (1000 samples).

External validation

The risk score system shows low discrimination in the validation group (c-statistics=0.641 [0.557-0.726], (Figure 4)).

However, despite being poorly discriminating, it was well calibrated according to the Hosmer-Lemeshow test (χ²= 6.250; p=0.396).

Considering the aforementioned classification, patients in the validation group were classified into 3 risk groups (Table 4): “low risk” (score 0-1 [predicted incidence <29%, n=101), “moderate risk” (score 2-3 [predicted incidence 29-43%, n=86]), and “high risk” (score 4-5 [predicted incidence >43%, n=12).

Discussion

We created a novel prediction model to assess the risk of venous thromboembolism (VTE) in patients diagnosed with lung cancer in the development cohort. Subsequently, we conducted an external validation of the model using a separate validation cohort. Our study involved a total of 381 patients, and our findings indicated that the prevalence of VTE in lung cancer patients was 26.8%, surpassing the rates reported in previous studies.^9–11

The variability in the incidence of VTE is due to several risk factors. In our study, lymph node status (≥N2), presence of metastasis, active smoking, surgical history, hypertension, and absence of coronaropathy were correlated with the occurrence of VTE. However, in other Tunisian studies, TNM stage IV and non-small squamous carcinoma were associated with high VTE incidence.^9,10

Thus, to reduce the occurrence of VTE, it is imperative to identify and assess all possible risk factors while determining the appropriate prophylactic measures for these patients. Various predictive models have been suggested to anticipate VTE occurrence in individuals with lung cancer. These models have incorporated several factors based on existing literature, and it is important to consider biomarkers associated with thrombosis.¹²

The Khorana risk score, a widely recognized predictive scoring system, categorizes cancer patients into distinct risk groups and identifies a high-risk group for thromboprophylaxis. It is considered the prevailing and valuable tool for predicting VTE in the cancer population.^12,13

The majority of factors included in various risk scoring systems have been incorporated into our risk score system. However, we did not include the D-dimer test, despite its significance, due to the limited number of observed values in our dataset.¹³ and, most importantly, due to the unavailability of this test in our current practice.

Furthermore, these aforementioned predictive risk scores were developed using data from patients diagnosed with various types of cancer, whereas our scoring system is specifically tailored to lung cancer patients.¹³ Thus, lung cancer data were included in our score but not adopted in most models. Some of these variables were present in our definitive risk score (i.e., the presence of metastasis and histological type of poorly or undifferentiated NSCLC).

In previous retrospective studies reported in the literature, adenocarcinoma has been identified as a strong predictor of VTE onset.¹³ Blom et al. conducted a study involving 537 NSCLC patients to investigate thrombotic risk and observed a 20-fold higher risk of VTE compared to the general population. Among the patients, those with adenocarcinoma had a three-fold higher risk (incidence=66.7%) than those with squamous cell carcinoma of the lung (incidence=21.2%).¹⁴ Similarly, in another cohort of 493 NSCLC patients, Tagalakis et al. reported a high incidence of DVT (13.6%).¹⁵ However, in our study, we found that poorly or undifferentiated NSCLC was associated with a higher prevalence of VTE, while adenocarcinoma was not predictive of the occurrence of VTE.¹³ Blom et al.^14,15

Body mass index (BMI), which is incorporated in both the Khorana score and Caprini VTE risk assessment, was integrated into our risk system. However, we used a lower cut-off (≥25 kg/m²) possibly due to the prevalence of poor nutritional status among lung cancer patients.^5,16

The hemogram parameter (hemoglobin, platelets, and leukocytes) included in the other risk score models was used in our study. Patients with a hemoglobin level <10 g/l had a four-fold higher risk of VTE. Other biomarkers, namely CRP and creatinine, were also included in our system; these biomarkers were not used by most VTE risk models. A cohort study investigating VTE risk among 3159 patients with newly diagnosed solid tumors concluded that elevated CRP and creatinine levels were predictive of VTE.¹⁷

Our scoring system incorporates cancer therapy and surgery as variables. Previous studies have demonstrated that cancer therapy, including chemotherapy, antiangiogenic therapy, and hormonal therapy, increases the risk of VTE.^18–21 Christensen et al., after reviewing 19 studies involving 10,660 patients with primary lung cancer undergoing curative-intent operations, found that the risk of VTE appears to be highest during the early postoperative period, with a subsequent decrease in risk.²²

As mentioned earlier, since risk factors vary from one population to another, there is a need to develop a risk score system specific to our population, in order to assist healthcare practitioners in developing appropriate prophylactic strategies for patients at risk of developing VTE.

After conducting logistic regression analysis in our study, we identified four items that were included in our risk score system. To the best of our knowledge, this risk score represents the first attempt to predict the potential incidence of VTE specifically for Tunisian patients with lung cancer. The developed risk score suggests that the incidence of VTE is expected to increase exponentially.

Scores below 2 were associated with a low risk of VTE, while scores of 4 or higher were associated with a high risk. The discriminant validity of this VTE score system was confirmed in the validation group, although there was a moderate decrease in predictive power. However, the classification of patients based on the score remained clinically meaningful. The notable variation in prognosis among the three risk groups should aid physicians in determining the appropriate therapeutic approach. Therefore, we strongly recommend thromboprophylaxis for Tunisian patients with moderate and high VTE risks.

Despite its poor predictive discrimination, this score presented several strengths. Firstly, to our knowledge, it is the first risk prediction model that included the occurrence of SVT as a predictable event. In fact, our decision to add SVT among outcomes was not arbitrary but based on several studies. A recent study conducted in 2022 highlighted the significance of SVT as a condition and revealed that patients with cancer and SVT are at an increased risk of thromboembolic complications.²³ In addition, Galanaud et al. suggested that cancer patients with SVT exhibit a poor prognosis, comparable to those with cancer-related DVT, with a heightened risk of recurrence of DVT-PE.

Secondly, we believe that this score can be applied to other populations whose characteristics are quite similar to those of our Tunisian population and with poor means on board, especially in underdeveloped countries. Nevertheless, it is essential to perform external validation of this score using data from diverse populations in order to ensure its generalizability and reliability.

The present study has several limitations that should be acknowledged. First, this study was based on 381 patients from three tertiary centers in northern Tunisia. Given the limited sample size, it is important to note that our patients may not adequately represent the diversity of our population. Secondly, we did not conduct an assessment of the reproducibility of our risk score in the prospective validation cohort. Thirdly, it is important to consider that personal and family history of VTE, as well as the use of anticoagulant or antiplatelet treatment at the time of lung cancer diagnosis, could potentially impact our findings. Finally, it should be noted that the cut-off values for the potential VTE variables included in our risk model were determined based on clinical experience or existing literature, rather than individualized threshold values determined by ROC curve analysis.

We have tried to respond to a need specific to the characteristics of our country where the economic crisis makes it very difficult to provide care according to international standards. Our score contains simple items available to any Tunisian practitioner.

The collection of the four items is straightforward: smoking history can be obtained through an interview, a complete blood count (CBC) is a readily available test in Tunisia, even in primary care settings, determining the histological type, and conducting staging assessments are commonly practiced.

Conclusion

In conclusion, the frequency of VTE in this study was high, at 26.8%. A predictive score for VTE was developed and validated by including epidemiological, clinical, and biological data. This score, despite its low discrimination, has a good positive and negative predictive value for a moderate risk of VTE. The stratification of risk in this newly developed risk system may guide the clinician in prescribing preventive treatment for VTE. However, our study has limitations, particularly the retrospective nature of the analysis and the small sample size. Hence, the need to conduct a prospective study on a nationwide scale for the validation of this score.