Predictive Cancer Model
Conditions
Brief summary
The goal of this prospective observational cohort study is to validate previously developed Hepatocellular Carcinoma (HCC) risk prediction algorithms, the Liver Risk Computation (LIRIC) models, which are based on electronic health records. The main questions it aims to answer are: * Will our retrospectively developed general population LIRIC models, developed on routine EHR data, perform similarly when prospectively validated, and reliably and accurately predict HCC in real-time? * What is the average time from model deployment and risk prediction, to the date of HCC development and what is the stage of HCC at diagnosis? The risk model will be deployed on data from individuals eligible for the study. Each individual will be assigned a risk score and tracked over time to assess the model's discriminatory performance and calibration.
Detailed description
The investigators will conduct a prospective observational cohort study, separately deploying three separate LIRIC models (the general population, cirrhosis, and no\_cirrhosis models) on retrospective de-identified EHR data of 44 HCOs in the USA, using the TriNetX federated network platform. LIRIC will generate a risk score for each individual. All risk-stratified individuals will be prospectively, electronically followed for up to 3-years to assess the primary end-point of HCC development. At the end of this period, model discrimination will be assessed, using the following metrics: AUROC, sensitivity, specificity, PPV/NPV. Risk scores generated by the model will be divided into quantiles. For each quantile, the investigators will evaluate the following: number of individuals in each quantile, number of HCC cases, PPV, NNS, SIR. Model calibration will be used for assessing the accuracy of estimates, based on the estimated to observed number of events. The model will dynamically re-evaluate all individual data every 6 months, re-classifying individuals (as needed).
Interventions
OTHERLiver Risk Computation Model (LIRIC)
A neural network model (LIRIC-NN) and a logistic regression model (LIRIC-LR) that use routinely collected EHR data to stratify individuals into HCC risk groups for the general population
OTHERLiver Risk Computation Model (LIRIC)_cirrhosis
A neural network model (LIRIC-NN) and a logistic regression model (LIRIC-LR) that use routinely collected EHR data to stratify individuals into HCC risk groups for the population with liver cirrhosis
OTHERLiver Risk Computation Model (LIRIC)_no_cirrhosis
neural network model (LIRIC-NN) and a logistic regression model (LIRIC-LR) that use routinely collected EHR data to stratify individuals into HCC risk groups for the population without liver cirrhosis
Sponsors
Beth Israel Deaconess Medical Center
Massachusetts Institute of Technology
TriNetX, LLC
Study design
Observational model
COHORT
Time perspective
PROSPECTIVE
Eligibility
Sex/Gender
ALL
Age
40 Years to 100 Years
Healthy volunteers
Yes
Inclusion criteria
The investigators will utilize the following criteria for all 3 models: Inclusion criteria: * Male and females age ≥40 years from all US HCOs available on the platform * at least at least 2 clinical encounters to the HCO, within the last year, before the study start date
Exclusion criteria
* Personal history of HCC or current HCC (ICD-9: 155.0; ICD-10: C22.0) * Age below 40. The same dataset will be utilized for the non-cirrhosis validation, with exclusion of all cases with cirrhosis (ICD-9: 571.2, 571.5; ICD-10: K70, K70.3, K70.30, K70.31, K74, K74.0, K74.6, K74.60, K74.69). For the cirrhosis validation, the investigators will include only patients with the above cirrhosis codes.
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Area under the receiver operating characteristic curve (AUROC) of LIRIC for all groups stratified | 6 months from index date, at 1 year, 2 years and 3 years | To assess the discriminatory performance of LIRIC for prospective identification of high-risk individuals for HCC development. ROCs and AUROC numbers will be calculated for the whole population and groups stratified by age, sex, race, and geographical location. |
| Calibration of LIRIC for all groups stratified | 6 months from index date, at 1 year, 2 years and 3 years | To assess how well the risk prediction by LIRIC aligns with observed risk without recalibration. Calibration plots will be created for the whole population and groups stratified by age, sex, race, and geographical location. |
| Performance metrics for LIRIC model risk quantiles | 6 months from index date, at 1 year, 2 years and 3 years | To evaluate the sensitivity, specificity, number of individuals/number of HCC cases, PPV, NNS in each predicted risk quantile for multiple risk thresholds |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Timing of incident HCC occurrence | 6 months from index date, at 1 year, 2 years and 3 years | To evaluate how long in advance before HCC occurrence should be expected for LIRIC models to make high-risk predictions based on different thresholds for high-risk. Distribution plots of the date of HCC incidence for multiple risk thresholds will be created. |
| Tumor stage at HCC diagnosis | 6 months from index date, at 1 year, 2 years and 3 years | TNM staging at HCC diagnosis |
Countries
United States
Outcome results
None listed