Renal Function
Conditions
Keywords
Renal function, eGRF, Atrial fibrillation, Coronary artery disease, Type 2 diabetes mellitus, Machine learning, Prognostic modeling
Brief summary
Scientific analyses are frequently performed on e.g. health insurance databases to study the usage and effectiveness of drugs in real life. Kidney function is known to have an influence on a patients disease development and/or drug levels in blood. However, often direct measures for kidney function are not available in databases. This study plans to develop tools to classify the renal function of patients, which helps scientists to identify patient cohorts (groups of patients sharing same characteristics) for scientific analyses.
Detailed description
Renal impairment is a common comorbidity in patients with diverse main underlying diseases and a pathology accompanying increasing age. Renal function might be an important modifier of treatment effects. Population-based administrative claims databases are increasingly used in large-scale comparative outcomes studies of drug treatments. However, claims databases often lack information on laboratory tests results limiting their usefulness in Real-World Evidence(RWE) research of patients with renal impairment. There is a need to develop methods for identification of patients with renal dysfunction from healthcare administrative claims-based proxies. The main objective of this study is the development of algorithms/models to predict eGFR values and/or classes for patients at certain time point based on entries in claims database (demographic characteristics, clinical diagnoses, procedures and drug treatments) for a general population and a variety of use-cases (atrial fibrillation, coronary artery disease, type 2 diabetes mellitus patients sub-populations). To achieve this, modern data-driven machine learning techniques will be applied to discover relationships between renal status, measured by eGFR, and longitudinal patient-level data. Evaluation of models' performance (out of sample validation, benchmark test, performance differences between eGFR value prediction algorithms and classification models tailored for the pre-defined eGFR classes) will be done as well.
Interventions
OTHERNo Intervention
This study is the development of algorithms/models to predict eGFR values and/or classes for patients at certain time point based on entries in claims database (demographic characteristics, clinical diagnoses, procedures and drug treatments) for a general population and a variety of use-cases (AF, CAD, T2DM patients sub-populations).
Sponsors
Bayer
Study design
Observational model
COHORT
Time perspective
RETROSPECTIVE
Eligibility
Sex/Gender
ALL
Age
18 Years to No maximum
Healthy volunteers
No
Inclusion criteria
To be included in the eGFR-population, patients have to have at least one recorded eGFR value in the OPTUM CDM database between January 1, 2007 and December 31, 2016, be adults (\>18 years of age at the time of eGFR test) and have at least 370/180 days (180 days serves as sensitivity analysis) of continuous enrollment in medical and pharmacy insurance plans since eGFR test date.
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Performance of classification to predict eGFR | From eGRF values starting and lasting 180d + 370d | For numeric models cross-validated performance is measured as correlation via r\*2. Class based performances are measured as cross-validated sensitivities given pre-defined false discovery rates with following definition for positives and negatives: Observed eGFR class X: * positive: eGFR measured at begin of time frame is in class X * negative: eGFR measured at begin of time frame is not in class X Class predicted by model: * positive: eGFR predicted is class X * negative: eGFR predicted is not class X |
Countries
United States
Outcome results
None listed