2026-10-15T09:48:30Z https://data.mendeley.com/oai

oai:data.mendeley.com/xkc9x7v2sr.1 2026-09-01T13:36:19Z

Habib, Jakaria Longitudinal eGFR Trajectory and Survival Benchmark Dataset for Chronic Kidney Disease (SMART-C Cohort) Mendeley Data 1. Scientific Overview & Objective This benchmark dataset, developed under the Reproducible Longitudinal Dynamical Patterns (RLDP) framework, provides high-density, multi-year clinical time-series for evaluating continuous-time trajectory forecasting, state-space estimation, and out-of-sample prognostic risk models in Chronic Kidney Disease (CKD). Longitudinal modeling of renal function requires overcoming significant observational challenges, including sparse and irregularly spaced visits, analytical measurement noise, and non-linear trajectory decay. This dataset establishes a standardized, fully reproducible benchmark for machine learning and dynamical systems in nephrology. 2. Cohort Specification & Clinical Design The dataset comprises N = 5,000 adult CKD patients prospectively tracked across a 3-year timeline (Days 1 to 1,096), generating 41,800 longitudinal laboratory evaluations (mean: 8.36 visits per patient; range: 2 to 14 visits). The primary continuous biomarker is Estimated Glomerular Filtration Rate (eGFR, mL/min/1.73m2). Baseline eGFR across the cohort is 55.70 +/- 17.97 mL/min/1.73m2 (range: 6.0 to 115.0 mL/min/1.73m2). Randomization is structured 1:1 into active intervention (n = 2,501) and control placebo (n = 2,499), stratified at screening by baseline disease severity into three clinical strata: - Screening eGFR 30 to <45 mL/min/1.73m2 (n = 1,497, 29.9%) - Screening eGFR 45 to <60 mL/min/1.73m2 (n = 1,466, 29.3%) - Screening eGFR 60 to <90 mL/min/1.73m2 (n = 2,037, 40.7%) Baseline concurrent GLP-1 receptor agonist therapy is documented in 24.9% of participants. 3. File Structure & Data Dictionary The archive contains two harmonized CSV files: - baseline.csv (5,000 records): usubjid (patient ID), trt01pn (arm: 1=active, 0=placebo), randfl (randomized), ittfl (intent-to-treat), trtfl (treated), blgfr (baseline eGFR), blglp1 (baseline GLP-1 status), strata (screening renal category). - followup.csv (41,800 records): usubjid, randfl, ittfl, trtfl, anl01fl, aval (observed visit eGFR), base (baseline eGFR), paramcd (GFRBSCRT), param, ady (analysis day: 1 to 1096), avisitn, avisit (protocol visit name: BASELINE, WEEK 3, WEEK 13, etc.). 4. Data Quality & Harmonization Audit Automated audit confirms: (a) 0.0% single-visit static records (all multi-year trajectories up to 3 years); (b) 0.0% acute ICU stays (<= 14 days); (c) 0.0% pediatric records (age >= 18); (d) hyperfiltration capped at eGFR <= 140; and (e) 0.0% missingness on protocol eGFR. 5. Intended Benchmark Applications Designed for benchmarking: (1) Sparse Functional Data Analysis (PACE / FPCA) via BLUP conditional expectation; (2) Continuous-Time State-Space Kalman Filtering (CT-SSM) with Maximum Likelihood estimation and closed-form uncertainty bounds; (3) Empirical Bayes Linear Mixed-Effects Models (EB-LMM); (4) Neural Velocity ODEs; and (5) Out-of-sample survival models for severe renal failure progression (eGFR < 30 mL/min/1.73m2). Computer Science Health Informatics Medical Informatics Dataset doi:10.17632/xkc9x7v2sr.1 oai:data.mendeley.com/xkc9x7v2sr.1 Creative Commons Attribution 4.0 International http://creativecommons.org/licenses/by/4.0 https://data.mendeley.com/datasets/xkc9x7v2sr 2026-09-01T13:36:19Z