Identification of Patients Eligible for Left Ventricular Assist Device
Multicenter, observational, real-world data study using routinely collected hospital data
Research objectives
Primary objective
- To identify patients who could possibly benefit from a left ventricular assist device (LVAD) therapy in participating hospitals from Belgium and the Netherlands.
Secondary objectives
- To characterize the heart failure (HF) patient population in the participating hospitals.
- To determine the prevalence of advanced heart failure (AdHF).
- To provide a detailed profile of patients in the advanced stages of HF.
Research questions addressed
- Describe the characteristics of all advanced-stage HF patients and evaluate the proportion of patients who could benefit from a cardiac assist device. We aim to describe and compare patients who have the device, and those who do not. The following characteristics will be explored:
- Baseline characteristics and comorbidities
- Medication usage
- Medical devices
- Treatment procedures
- Laboratory measurements
- HF outcomes (mortality and hospitalization)
- Describe the characteristics of patients with advanced-stage HF who could benefit from the cardiac assist device as:
• Bridge-to-transplant therapy (patients awaiting a heart transplant)
• Destination therapy (patients who are not candidates for a heart transplant)
Your Project in Numbers
The numbers represent total recorded instances of these medical parameters, with a single patient possibly having multiple entries at different times or across different records.
+3M
records processed
+5M
Drugs
+5M
measurements
Multi-country
+10M
Conditions
+19K
patients processed
+1.7M
procedures
121
datapoints
LynxCare's Disease-Specific Multilingual NLP
Our disease-specific NLP models could extract endpoints that are not available in standard structured data, as well as improve the available data where structured data is incomplete.
- In defining the cohort selection, we applied additional criteria, such as NT-proBNP levels or unstructured mentions of "Heart Failure" and "NYHA class". This allowed LynxCare to refine a cohort comprising all patients potentially with HF - by narrowing the group from potential to probable HF cases.
- Furthermore, by selecting patients not only based on ICD-10 codes but also on left ventricular ejection fraction (LVEF) values, LynxCare could potentially identify undiagnosed HF cases that might have gone unnoticed.
For more details on the cohort selection process, please refer to the section below.
By leveraging LynxCare’s disease-specific NLP workflows, we identified significantly more patients than would have been possible with traditional methods or general NLP solutions.
Methodology used
Cohorts - Methodology
Cohort 1 consists of all patients potentially with HF.
By applying an additional criterion—such as NT-proBNP levels or unstructured mentions of "Heart Failure" and "NYHA class"—we narrowed this group from potential to probable HF cases. All patients in Cohort 1 have a potential HF diagnosis within the time window of January 1, 2018 to December 31, 2023.
Additionally, by selecting patients not only based on ICD-10 codes but also on LVEF values, we identified undiagnosed HF cases that had previously gone unnoticed.
Cohort 1 is further divided into two subgroups: Cohort 2 (without probable HF) and Cohort 3 (with probable HF). From Cohort 3, we identified patients with probable advanced HF, forming Cohort 5, using an in-house scoring system.
Finally, Cohort 5 helped us identify patients eligible for a Left Ventricular Assist Device (LVAD). These patients form Cohort 7, while those not eligible fall under Cohort 6, with ineligibility due to various reasons.
For each cohort, an index date is defined. This index date serves as the time reference, which allowed us to identify the baseline condition by considering all relevant data preceding that date. In addition, it enabled us to characterize the patient population by assessing clinical parameters that were evaluated around the index date, providing a snapshot of the patient's health status at a critical time in their medical history. This helped ensure that subsequent analyses and comparisons are grounded in consistent, time-specific clinical data.
Results
In this study, we leveraged RWD from EHRs to create detailed cohorts of HF patients, ranging from early-stage HF to AdHF, with a focus on identifying those eligible for LVAD. By combining structured data with innovative techniques for processing unstructured data, we aimed to improve patient stratification and uncover clinical patterns to inform future treatment strategies.
The integration of advanced methodologies, particularly NLP and well-defined cohort criteria, allowed us to address limitations inherent in structured data alone. NLP played a crucial role in extracting valuable insights from unstructured clinical notes within EHRs, enhancing the precision of cohort selection.
Approx. 30% of the patients in cohort 3 with probable HF were not yet formally diagnosed, and were identified using NLP and EHR mapping
33% | 37% of patients in cohorts 1 & 3 progressed to advanced stages of heart failure vs. the commonly reported 1-10% [1] and the 13.7% seen in Dunlay et al. [2]
9% | 24% of patients in cohorts 3 & 5 identified as LVAD-eligible vs. 3.3% reported in Dunlay et al. [2]
[1] Crespo-Leiro, M.G., et al., Advanced heart failure: a position statement of the Heart Failure Association of the European Society of Cardiology. European Journal of Heart Failure, 2018. 20(11): p. 1505-1535. [2] Dunlay, S.M., et al., Advanced Heart Failure Epidemiology and Outcomes: A Population-Based Study. JACC: Heart Failure, 2021. 9(10): p. 722-732.
Key components of the study included
- Iterative refinement of data points, conducted in close collaboration with the LynxCare data engineering team and physicians, ultimately led to more accurate classification of patient data and improved cohort accuracy
- Development and refinement of the AdHF scoring system, which was not only instrumental in cohort identification but also holds significant potential for future predictive models
Deliveries
- Clinical dashboard for the hospitals
- Final report with aggregated results for Abbott
Next steps
- Manual chart review process by the physicians will help us fine-tune the AdHF scoring system for future studies, further enhancing its utility and performance. This will further enhance cohort precision, enabling more timely interventions and improving access to advanced therapies like LVAD for patients who might otherwise be overlooked.
- Development of a methodological paper, as well as a research paper.
Key insights & results
- Improved patient profiling: identified key indicators for patients at-high-risk for advanced heart failure
- 1000+ patients identified as potentially eligible for LVAD across 2 hospitals
Value beyond the
results
Scientific value
Clinical Database
Research articles,
insights hospital patient population, …
Patient value
Increase (earlier) access to LVAD
key indicators identification, ...
Medical marketing value
Enforce relation with physicians
Build a database and deliver better care
Commercial value
Scalability towards other hospitals
Datapoint list, trained datapoints, methodology
Your LynxCare Team
Charlotte Evenepoel
Clinical Project Manager
charlotte.evenepoel@lynx.care
Dries Hens, M.D.
Chief Medical Officer, co-founder
dries.hens@lynx.care
Clara L. Oeste
Principal RWE Researcher
clara.oeste@lynx.care
"Refining the cohort definitions, especially for the LVAD cohort, ignited an exciting discussion that promises significant benefits for both physicians and Abbott. By selecting patients based not only on diagnostic codes but also on clinical heart failure characteristics, we can uncover previously undiagnosed heart failure cases that have gone unnoticed. This approach is vital, as timely identification of heart failure can dramatically improve outcomes and pave the way for better patient care!"
Charlotte Evenepoel, Clinical Project Manager @ LynxCare
