Venous thromboembolism (VTE)

Pulmonary embolism (PE)

Lower extremity deep vein thrombosis (LE-DVT)

VTE is a serious and preventable complication of systemic therapy in cancer patients

Utilizing retrospective data on patients diagnosed with cancer from 2011 to 2020

HHS

RAM

VA

A RAM was derived using least absolute shrinkage and selection operator regression of HHS (n = 9769)

External validation using the Department of Veterans Affairs (VA) health care system (n =79517)

Bootstrapped c-statistics and calibration curves are used to evaluate the discrimination and fit of external models.

Create dichotomous risk stratification using integer scores and compare with Khorana score (KS)

Baseline covariates were compared between cohorts using standardized mean differences (SMD), where SMD greater than 0.1 was considered significantly different.

Finally, the model performance was tested across demographic subgroups to ensure generalizability

New RAM has similar discriminatory power across age, sex, and racial/ethnic subgroups in HHS and VA health care system cohorts (Data Supplement)

At 6 months, 590 (6.2%) and 437 (4.6%) patients in HHS developed VTE and PE/LE-DVT, respectively

VTE and PE/LE-DVT occurred in 4,027 (5.1%) and 3,331 (4.2%) patients, respectively, in the VA healthcare system

Based on the final list of 11 predictors in the HHS derivation cohort, we created a new VTE risk score by adding a weighted integer score (median, 3 [IQR, 2-5])

To mimic the KS classification implemented in previous clinical trials, risk groups were dichotomized based on a predetermined clinical threshold of 7%-8% overall VTE at 6 months, which corresponded to a score of 31 in the high-risk group

New RAM grouping improves discrimination and classification compared to KS with cutoff of 2

Assessment at initiation of systemic therapy

New RAM includes components of KS

The c-statistics for the HHS and VA healthcare systems were 0.71 and 0.68, respectively (and KS were 0.65 and 0.60, respectively)

The new RAM appropriately reclassified 28% of patients and increased the proportion of VTE in the high-risk group in the validation dataset from 37% to 68%

Simple additive fractions can be easily implemented and calculated in real time without the need for external websites or nomograms

The novel RAM stratifies cancer patients into high-risk groups, with a cumulative incidence of VTE of 8%-10% and PE/LE-DVT of 7% at 6 months (3% and 2%, respectively, in the low-risk group)

This model improves performance over original KS and doubles the number of VTE events in high-risk tiers

Prospective studies are encouraged for additional external validation

Baseline clinical characteristics of patients with acute heart failure

Derivation of a new VTE risk assessment model 6 months after initiation of systemic therapy in a Harris Health System cohort

Simplifying the performance of VTE risk assessment models in derivation and validation datasets

Comparison between the new risk assessment model and the original KS

Patient selection and exclusion in derivation and validation cohorts

Incidence of VTE stratified in derivation and validation cohorts by new risk assessment model

Risk stratification based on XGBoost

(A) Features that have the greatest impact on predictions (ordered from most important to least important)

(B) Distribution of the impact of each feature on model output

(C and D) Individualized predictions for two patients are shown.

Data extraction and reconciliation between two healthcare systems

Validation of computable phenotypes for venous thromboembolism (VTE) outcomes

Derivation and validation of new risk assessment models

Baseline variables included in model derivation

Systemic treatment classification

International Classification of Diseases (ICD) algorithm used to determine study results

Natural language processing (NLP) algorithm for determining research results

Performance of combined ICD and NLP algorithms at each site

Logistic regression model of cancer subtype reclassification on thrombosis risk

Ranking of importance of LASSO-selected variables for overall VTE and PE/LE-DVT in the HHS-derived cohort at λ 1se

Variables in the multivariable logistic regression model of overall VTE in the HHS-derived cohort

Original KS score vs. new model performance in derivation and validation cohorts

Sociodemographic subgroup analysis of a new risk assessment model in derivation and validation cohorts

Comparison with previous landmark studies

Covariate selection by LASSO model for overall VTE and PE/LE-DVT outcomes

Calibration plot of the new risk assessment model in the validation cohort

Cumulative incidence of VTE in derivation and validation cohorts by KS score and new model

Derivation and validation of overall survival for cohorts