TL;DR
This paper introduces an "Informatics Consult" service that aims to rapidly generate tailored clinical evidence for complex patient cases, addressing the significant lag between evidence generation and clinical practice. By integrating data from multiple sources (RCTs, observational studies, EHRs, genetics), they demonstrate a proof-of-concept for informing treatment decisions in situations with conflicting indications and contraindications, like anticoagulation for patients with both atrial fibrillation and liver cirrhosis.
What Problem Does This Paper Address?
Generating robust clinical evidence, especially for patients with multiple conditions (multimorbidity), is a slow and often incomplete process. Traditional evidence generation, like Randomized Controlled Trials (RCTs), frequently excludes patients with comorbidities, leaving a gap in evidence tailored to real-world patient populations. This means clinicians often make decisions with limited or no high-quality evidence, leading to uncertainty and potentially suboptimal care. The paper addresses the lack of readily available, on-demand clinical evidence that is specifically relevant to individual patients facing complex treatment dilemmas.
Methods
The authors developed and tested an "Informatics Consult" framework, which involves:
- Prototype EHR Forms: Designing electronic health record (EHR) forms for clinicians to request consultations and receive summarized evidence.
- Multi-Source Evidence Generation:
- RCT Search: Identifying existing and recruiting RCTs.
- Meta-Analysis of Observational Studies: Synthesizing data from published studies.
- Target Trial Emulation: Using large-scale EHR data to simulate an RCT design and analyze treatment effects.
- Mendelian Randomization (MR): Employing genetic data to validate drug targets and infer causal relationships.
- Automation Potential Assessment: Identifying steps within each evidence stream that could be automated to accelerate evidence generation and delivery.
- Clinician Acceptability Survey: Gauging the usefulness and feasibility of the Informatics Consult among practicing clinicians.
Datasets
- Electronic Health Records (EHRs): A large dataset of 3,854,710 adults aged ≥30 years from the Clinical Practice Research Datalink (CPRD) linked to secondary care Hospital Episode Statistics and the Office for National Statistic death registration was used for target trial emulation.
- PubMed: Used for searching and identifying relevant observational studies for meta-analysis.
- ClinicalTrials.gov: Used to identify reported and currently recruiting RCTs.
- Genome-Wide Association Study (GWAS) Data: Used for Mendelian randomization analyses, specifically from the MEGAS-TROKE study for stroke outcomes.
Key Insights and Ideas
- Bridging the Evidence Gap: The Informatics Consult offers a novel way to generate evidence rapidly, directly addressing the limitations of traditional methods for complex patient scenarios.
- Concordance Across Diverse Evidence Sources: The study demonstrated a degree of agreement between different evidence streams (observational studies, target trial emulation, genetic data) in suggesting potential benefits of warfarin for stroke reduction.
- RCTs Exclude the "Difficult" Patients: A key finding was that existing and recruiting RCTs for anticoagulation in atrial fibrillation explicitly exclude patients with cirrhosis, highlighting the critical need for alternative evidence generation methods.
- Clinician Demand is High: A significant majority of surveyed clinicians found the concept of an Informatics Consult highly useful and believed it should be integrated into healthcare systems.
- Automation is Crucial for Scalability: The paper emphasizes that automating various stages of evidence synthesis is essential for making such a service practical and scalable across different clinical questions and datasets.
Key Results
- No Relevant RCTs Found: Zero previously reported RCTs were identified for anticoagulants in patients with cirrhosis and atrial fibrillation.
- Observational Studies Suggest Benefit: A meta-analysis of observational studies showed warfarin use was associated with a lower stroke risk (HR = 0.71, CI 0.39–1.29), though with high heterogeneity.
- Target Trial Emulation Shows Significant Benefits: In a large EHR-based emulation, warfarin was associated with significantly lower all-cause mortality (HR=0.61, CI 0.49–0.76) and ischaemic stroke (HR=0.27, CI 0.08-0.91).
- Genetic Evidence Supports Drug Target: Mendelian randomization suggested that lower vitamin K1 levels (which warfarin inhibits) are associated with lower stroke risk, corroborating warfarin's potential mechanism.
- High Clinician Acceptance: 85% of clinicians found prognosis information useful, and 79% believed they should have access to the Informatics Consult service. 74% would discuss the report with patients.
Limitations and Future Work
- EHR Forms Not Yet Implemented: The prototype EHR forms have not been integrated into live clinical systems.
- Limited Scope of Evidence: The study did not assess bleeding outcomes or newer non-vitamin K oral anticoagulants due to data access limitations.
- Unmeasured Confounding: The authors acknowledge the persistent difficulty in fully assessing unmeasured confounding in observational data, and the possibility of missing prescription data.
- Limited GWAS Data: Access to summarized GWAS data for stroke outcomes in patients with cirrhosis was limited, impacting the MR analysis.
- Automation Not Yet Implemented: While potential automation pathways were identified, they have not yet been implemented.
- Future Work: The authors suggest future analyses on more recent and larger datasets, exploring bleeding outcomes, and implementing the identified automation opportunities to scale the Informatics Consult service. They also propose building a library of consults to facilitate knowledge sharing and identify needs for new RCTs.
Summary updated