DEVELOPMENT AND VALIDATION OF THE COMPLICATION MODEL FOR OBESITY AND OVERWEIGHT (COMET): A PATIENT-LEVEL STOCHASTIC MICROSIMULATION MODEL FOR OBESITY-RELATED COMPLICATIONS AND MORTALITY
Author(s)
Tiange Tang, MPH1, Lizheng Shi, PhD2, Hui Shao, MHA, PhD, MD3, Charles Stoecker, PhD2, Vivian Fonseca, MD4.
1Student, Tulane University, metairie, LA, USA, 2Tulane University School of Public Health and Tropical Medicine, New Orleans, LA, USA, 3Emory University, Atlanta, GA, USA, 4Tulane University, New Orleans, LA, USA.
1Student, Tulane University, metairie, LA, USA, 2Tulane University School of Public Health and Tropical Medicine, New Orleans, LA, USA, 3Emory University, Atlanta, GA, USA, 4Tulane University, New Orleans, LA, USA.
OBJECTIVES: This study aimed to develop and validate the COMET model for predicting the risks of obesity-related complications and mortality.
METHODS: The COMET microsimulation model was developed using longitudinal data from 3,081 participants in the Diabetes Prevention Program (DPP) and Diabetes Prevention Program Outcomes Study (DPPOS), a racially and ethnically diverse randomized trial cohort of adults with overweight/obesity and impaired glucose tolerance. The risk engine included 12 event-specific risk equations for onset of diabetes, chronic kidney disease, retinopathy, neuropathy, atherosclerosis, major adverse cardiovascular events, cardiovascular death, all-cause mortality, cancer, cognitive impairment, frailty, and physical function decline. Candidate predictors included demographic characteristics, cardiometabolic risk factors, laboratory measures, and histories of prior complications. Model performance was evaluated using apparent C-statistics, Brier scores, and calibration plots comparing observed and predicted cumulative incidence of DPP/DPPOS for internal validation. The risk equations were then implemented into the individual-level stochastic microsimulation model. External validation was conducted using linear regression to compare COMET-predicted event rates with 48 observed event rates from 13 obesity outcome trials: semaglutide STEP trial series and OASIS trial, liraglutide SCALE trial, and tirzepatide SURMOUNT trial series.
RESULTS: The COMET risk equations demonstrated varying discrimination across obesity-related outcomes, with C-statistics ranging from 0.67 for retinopathy to 0.90 for cognitive impairment. Discrimination was highest for cognitive impairment, physical function decline, all-cause mortality, and diabetes. Apparent Brier scores ranged from 0.0003 to 0.0133 across outcomes, indicating low prediction error. Internal calibration plots showed close agreement between observed and predicted cumulative incidence for DPP/DPPOS outcomes. The external validation found the COMET predicted event rates were closely aligned with observed event rates, with a regression slope of 1.108, intercept of -0.004, and R-squared of 0.771.
CONCLUSIONS: The COMET model demonstrated good internal and external validity in modeling obesity-related complications and mortality to improve obesity risk evaluation and management.
METHODS: The COMET microsimulation model was developed using longitudinal data from 3,081 participants in the Diabetes Prevention Program (DPP) and Diabetes Prevention Program Outcomes Study (DPPOS), a racially and ethnically diverse randomized trial cohort of adults with overweight/obesity and impaired glucose tolerance. The risk engine included 12 event-specific risk equations for onset of diabetes, chronic kidney disease, retinopathy, neuropathy, atherosclerosis, major adverse cardiovascular events, cardiovascular death, all-cause mortality, cancer, cognitive impairment, frailty, and physical function decline. Candidate predictors included demographic characteristics, cardiometabolic risk factors, laboratory measures, and histories of prior complications. Model performance was evaluated using apparent C-statistics, Brier scores, and calibration plots comparing observed and predicted cumulative incidence of DPP/DPPOS for internal validation. The risk equations were then implemented into the individual-level stochastic microsimulation model. External validation was conducted using linear regression to compare COMET-predicted event rates with 48 observed event rates from 13 obesity outcome trials: semaglutide STEP trial series and OASIS trial, liraglutide SCALE trial, and tirzepatide SURMOUNT trial series.
RESULTS: The COMET risk equations demonstrated varying discrimination across obesity-related outcomes, with C-statistics ranging from 0.67 for retinopathy to 0.90 for cognitive impairment. Discrimination was highest for cognitive impairment, physical function decline, all-cause mortality, and diabetes. Apparent Brier scores ranged from 0.0003 to 0.0133 across outcomes, indicating low prediction error. Internal calibration plots showed close agreement between observed and predicted cumulative incidence for DPP/DPPOS outcomes. The external validation found the COMET predicted event rates were closely aligned with observed event rates, with a regression slope of 1.108, intercept of -0.004, and R-squared of 0.771.
CONCLUSIONS: The COMET model demonstrated good internal and external validity in modeling obesity-related complications and mortality to improve obesity risk evaluation and management.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
MSR240
Topic
Epidemiology & Public Health, Methodological & Statistical Research, Real World Data & Information Systems
Disease
Diabetes/Endocrine/Metabolic Disorders (including obesity), No Additional Disease & Conditions/Specialized Treatment Areas