On 17-18 August, we had the pleasure of having Bendix Carstensen, Senior Statistician at Steno Diabetes Center Copenhagen come to Aberdeen for a two-day visit.
He worked hard while with us, doing two talks, one generic for an applied health science audience and one more technical for RSS members. These were followed the next day by a full days’ workshop using the R software, Epi, developed by Bendix.
The first talk, 'Estimation and prediction in multistate models', was largely non-technical and clearly mapped out the statistical model concepts for how a person’s status might change over time leading to the so-called multistate models (MS models) - often mistermed Markov-models. Bendix used data from a clinical trial, the Steno-2 study, initiated at Steno Diabetes Center in the 1990s, where in the long-term follow-up the assessment of treatment on incidence rates of cardiovascular disease (CVD) and mortality (hazard ratios) were needed along with the effect on cumulative measures such as survival, lifetime lost, and time spent with cardiovascular disease. He used the MS models to derive hazard ratios for mortality and CVD occurrence as well as baseline hazards, how these can be used to construct the cumulative measures of interest, and what assumptions are needed. He briefly described the Lexis machinery from the Epi package for R facilitating this with the ideas behind the simulation tool simLexis that allows predictions from multistate models.
The second talk, held in the evening to a mixed audience but billed for RSS members, was on 'The resurrection of time as a continuous concept in biostatistics, demography and epidemiology'. This was broadly similar to the earlier talk but with more of the rational and theory behind MS models. He especially indicated the advantages of the MS models over the Kaplan-Meier, Cox and Age-Period-Cohort models, by the use of the very fine subdivision of time used in the non-parametric modelling with assumptions about continuous, smooth effects of time. Besides the ability to show time-effects directly on the rate-scale, this approach also allows for the possibility to accommodate multiple time scales such as age and duration of disease simultaneously and readily extends to estimation of transition rates.
Both talks were well received – the earlier one especially by those from other discipline areas. The workshop on the following day was an intensive session that allowed participants to use the Lexis Epi package in R and try for themselves some of the features.
These events were partly funded by the RSS and as a result of support by the Institute of Applied Health Science within the University of Aberdeen for their internal research themes. The event materials are available on: http://BendixCarstensen.com/AdvCoh/courses/Aberdeen-2017