How many levels in multi-level modelling? An example from a recent CCHSR paper

A previous blog by Gary Abel here discussed the interpretation of random effects used to account for clustering (i.e. non-independence) of observations within organisations such as general practices or hospitals. We can also use another common term to describe this use of random effect variables, which is ‘multi-level modelling’. For example, patient observations are usually clustered (or ‘nested’) within hospitals, in which case we consider the data to have two levels (patient and hospital). However, at times we would wish to account for additional levels in the data structure. For example, we may wish to take into account clustering that occurs below the level of a given hospital, e.g. between different specialist clinics, hospital wards, or specific doctors (surgeons) within a hospital. An illustration of this application is provided in a recent CCHSR publication here [1]. In that paper, we observed that there was substantial variation in the use of a specific type of surgery between patients with different characteristics such as age and gender – i.e. at the patient level. We also observed large variation in the use of this surgical technique between surgeons, in the 2nd level of our model. Accounting for person-level and surgeon-level variation, there was no evidence of variation at the level of the hospital (cancer) network (within which consultants are nested) – i.e. the 3rd level of our model. In other words the variation in the use of the type of surgery could be attributed to the surgeons themselves rather than to the place where they worked (and this is an expected finding, as only some surgeons are specialising in this type of urological surgery and expected to perform such operations). This example demonstrates the use of a three-level model in health services research. However, often there is no information to support such multi-level modelling beyond the minimum two levels required.

 

 

1. Hounsome LS, Abel GA, Verne J, Neal DE, Lyratzopoulos G. Predictors of the use of orthotopic bladder reconstruction after radical cystectomy for bladder cancer: data from a pilot study of 1756 cases 2004-2011. BJU Int. 2013 Feb 6. doi: 10.1111/j.1464-410X.2012.11644.x. [Epub ahead of print]

This entry was posted in Blog. Group: . Bookmark the permalink. Both comments and trackbacks are currently closed.
  • The Cambridge Centre for Health Services Research (CCHSR) is a thriving collaboration between the University of Cambridge and RAND Europe. We aim to inform health policy and practice by conducting research and evaluation studies of organisation and delivery of healthcare, including safety, effectiveness, efficiency and patient experience.