PROGNOSTIC MODELS FOR MORTALITY AND HOSPITALIZATION IN ADULTS WITH ADVANCED CHRONIC KIDNEY DISEASE ON RENAL REPLACEMENT THERAPY IN COLOMBIA.

In Colombia, death and hospitalization rates of patients with advanced chronic kidney disease undergoing renal replacement therapy are progressively increasing. Consequently, three-year mortality and one-year hospitalization prognostic models were developed and validated

Literature search for prognostic factors, variable selection through a modified Delphi experts’ panel, assembly of retrospective cohorts of patients with advances chronic kidney disease from 2014 for development of mortality and hospitalization prognostic extended cox proportional hazard models, sensitivity analysis of missing data with multiple imputation procedures. Internal validation using bootstrapping re-sample, external validation with retrospective cohorts from 2015-2017. Cohorts created using Colombian data from a kidney care clinics’ network. Evaluation included discrimination (Harrell's C-index), calibration (calibration plots and joint statistical test) and clinical utility (decision curve analysis). Based on the PROGRESS framework.

54 prognostic variables found in literature search, 23 variables for mortality and 20 variables for hospitalization selected by experts’ panel.

Three-year mortality model development cohort included 2.343 patients, 983 (41,95%) women, median age 61,6 years (IQR= 21,6), the prognostic model was built with statistically significant variables from bivariate analysis: dialysis initiation age, coronary artery disease, diabetes mellitus, congestive heart failure, residual urine volume, Charlson index, weight (time-dependent transformation), corporal mass index (1/sqrt*X transformation), serum calcium, serum phosphorus, hemoglobin, serum albumin and random blood glucose. Discrimination C-index: 0,9404; calibration joint test: p=0,649; net benefit with probability threshold: 12%. Internal validation: C-index: 0,9447; calibration: p= 0,542; net benefit with probability threshold: 11%. External validation cohort included 3.688 patients, 560 (42,30%) women, median age 61,5 years (IQR= 20,9); C-index: 0,9463; calibration: p=0,184; net benefit with probability threshold: 13,5%.

One-year hospitalization development model cohort with 2.819 patients, median age 62,1 years (IQR: 21,6), 42,36% women. Selected variables: dialysis initiation age (time-dependent), dialysis therapy type, diabetes mellitus (time-dependent), peripheral arterial disease, residual urine volume, Kt/V, hemoglobin, serum albumin and random blood glucose; discrimination C-index: 0,8962; calibration: p=0,948; net benefit with probability threshold: 25%. Internal validation: C-index: 0,8959; Calibration: p=0,776; net benefit with probability threshold: 24%. External validation cohort: 4.054 patients, mean age 61,0 years (IQR: 20,6), 43,62% women; C-index: 0,9089; Calibration: p=0,135; net benefit with probability threshold: 25%

A mortality prognostic model with good discrimination level, low probability threshold and a hospitalization prognostic model with good discrimination level, intermediate probability threshold was obtained, both with good calibration and stable coefficients in sensitivity analysis; both models were developed and validated in the Colombian samples.