Hanna Kunina(1), Maria C. Kjellsson(1)
(1) Pharmacometrics Research Group, Department of Pharmaceutical Biosciences, Uppsala University, Sweden
Objectives: Type 2 diabetes mellitus (T2D) is a group of metabolic diseases that are associated with long-term damage to and failure of various organs [1]. Recent evidence suggests that the risk of long-term complications for patients suffering from T2D may vary, depending on the presence and severity of comorbidities [2]. T2D is associated with disabling and life-threatening micro- and macrovascular complications, and diseases e.g. chronic kidney disease (CKD) and cardiovascular disease (CVD) are of paramount interest [3]. The aim of this project was to develop a multistate model for competing risks analysis using data from the Swedish National Diabetes Registry (NDR) and characterize the impact of covariates on the competing risks.
Methods: The NDR coverage is approximately 90% of patients with diabetes in Sweden and contains ~360,000 patients. All adult patients with T2D, registered in NDR 2005-2013, without prior events of CKD and CVD, with at least one record of key covariates (e.g. glycated haemoglobin – HbA1c, body mass index – BMI, sex, systolic blood pressure – SBP) were included in our study. In total, 78,951 patients with 603,308 observations were included. To describe disease progression, a multistate, competing risks model with five clinical states was used [4]. All subjects started from the initial state having T2D without comorbidities and moved towards the terminal state (death), either through intermediate states (CVD, CKD, or the dyad state, CVD+CKD) or directly. The Renal Association Guide [5] and the presence of stroke or ischemic heart disease was used to define CKD and CVD, respectively. Cumulative hazards were used to describe the transition intensities between the different states, based on mean transit times (MTTs) through the states [4]. Due to limited and non-regular observations, MTTs were subject to interval censoring. Several hypotheses were tested: 1) risk of CVD is independent of CKD, 2) risk of CKD is independent of CVD and 3) mortality is independent of CVD, CKD and dyad state. Model building was conducted using the likelihood ratio test and visual predictive check. Data management and exploration were performed using R V3.5.1. and the model fitting and evaluation was performed with NONMEM V7.4.3 and PsN.
Results: A disease progression model, taking into account being in any of five different states and competing risks of transitions between these states, was developed. By definition, all individuals started in the initial state with T2D without comorbidities, and the probability of staying in this state declined non-linearly over time. All mortality transitions were implemented using the Gompertz-Makeham formula, adjusted for the Swedish mortality rate and estimating a shift of age for the current state transition. The mortality risk was dependent on state and the initial state was estimated to be equal to a 7-year younger population than the standardized, while the mortality risk of the CVD, CKD and the dyad state were estimated to be equal to a 1.9-year, 2.4-year and 10-year older population. Thus, comorbidities reduce the expected life-span majorly. The transition intensity to CKD was time-constant and dependent on baseline BMI, baseline SBP and age. The transition intensity to CVD was time-varying with higher risk the first 10 years and dependent on baseline BMI, baseline HbA1c and age. The rate of CKD development was estimated to that of 7.9-year higher for patients being at state CVD than at the T2DM state. The occupational probability of CVD state was higher than the occupational probability of CKD, but patients with CKD state had a higher risk of death. The risk of transition to the semi-competing death state was higher for dyad state than for both states separately, and the rate of transition to CVD+CKD state from CKD state was higher than from CVD state. A challenge with these large data was model evaluation, and data was thus split into several parts and simulated separately for creation of VPCs.
Conclusions: A multi-state model for competing risks analysis of T2D long-term complications was successfully developed. This model adequately described the diabetic disease progression in the Swedish patient population. The magnitude of the estimated parameters was reasonable and meaningful. Future work involves model validation and assessment of the treatment impact on the risk of comorbidities and mortality through changes in covariates.
References:
[1] American Diabetes Association. Diabetes Care. 2013 Jan; 36 (Suppl 1): S67–S74. Published online 2012 Dec 10. doi: 10.2337/dc13-S067
[2] Parrinello, C. M., Matsushita, K., Woodward, M., Wagenknecht, L. E., Coresh, J., & Selvin, E. (2016). Risk prediction of major complications in individuals with diabetes: The Atherosclerosis Risk in Communities Study. Diabetes, obesity & metabolism, 18(9), 899-906.
[3] Chawla, A., Chawla, R., & Jaggi, S. (2016). Microvascular and macrovascular complications in diabetes mellitus: Distinct or continuum. Indian journal of endocrinology and metabolism, 20(4), 546-51.
[4] Ibrahim, Moustafa M. A (2019). Pharmacometric evaluation and improvement of models and study designs – applied in diabetes. Uppsala: Acta Universitatis Upsaliensis. Digital Comprehensive Summaries of Uppsala Dissertations from the Faculty of Pharmacy, 264. p. 75. Uppsala. ISBN: 978-91-513-0518-9.
[5] The UK eCKD Guide. The Renal Association 2019. Date of access January 10, 2019, https://renal.org/information-resources/the-uk-eckd-guide/ckd-stages/
Reference: PAGE 28 (2019) Abstr 9083 [www.page-meeting.org/?abstract=9083]
Poster: Drug/Disease Modelling - Endocrine