Elham Madreseh, Mahmood Mahmoodi, Seyed Mostafa Hosseini, Hojjat Zeraati, Iraj Najafi,
Volume 11, Issue 4 (3-2014)
Abstract
Background and Aim: In many medical studies along with longitudinal data, which are repeatedly measured during a certain time period, survival data are also recorded. In these situations, using models such as, mixed effects models or GEE method for longitudinal data and Cox model for survival data, are not appropriate because some necessary assumptions are not met. Instead, the joint models have been introduced, to consider: 1- measurement error in time-dependent covariates 2-monotone and non-ignorable missing data which occurs after an event and 3- relation between longitudinal and survival outcomes, simultaneously. At this paper, joint model Puts longitudinal response (i.e. creatinine) as a time dependent variable, along with other covariates in survival sub model, to investigate dialysis patients survival.
Materials and Methods: This research contained information about 417 patients affected to chronic renal failure, under treatment of continuous ambulatory peritoneal dialysis (CAPD) method. Patients were referred to three medical centers in Tehran (Shariati, Modares and Shafa) between 1997 to 2009.In this study longitudinal data and time dependent covariate were used Therefore, different variables for each person at certain time have been measured. In first some information was gathered from patient’s file, and then effective factors on survival of patients have been determined by using joint model. Results were compared with naive analysis (extended Cox model). For data analyzing, R software and significant level of 0.05 have been used.
Results: with using joint model sex, age, diabetes, diastolic blood pressure, haemoglobin, urea, LdL, and creatinine covariates were significant. In extended Cox model, only age and Diastolic blood pressure covariates were considered as effective factors on hazard of death in patients.
Conclusion: Joint model assess the effective factors on both endpoints simultaneously. Also it considers missing data that appeared due to an event, and covariates which were measured with error. Therefore in these cases, using joint models that led to better results and more knowledge about dieses, are necessary.
