The PARR tool is a huge step forward.
The PARR algorithm as described by Billings et al is a welcome tool
in the bid of the Health Service to cut down on unplanned secondary care
expenditures. Such tools have been developed in various settings before
now but most of these have focused on one condition or have used far fewer
variables to predict risk of rehospitalisation (1).
The authors however do not make explicit the degree of overlap
between the two 10% samples used to develop and test the algorithm. If any
reasonable degree of overlap existed, it could turn to be a case of
‘garbage in, garbage out’. Conclusions on the validity of the algorithm
would more convincing if it was tested on a population not included in its
development. It has to be said however that the use of a 10% sample,
rather than higher, would minimise the likelihood of a significant
In describing the limitations of their approach, the authors call
attention to the inherent quality deficiencies in the Hospital Episode
Statistics (HES) data, stating that there was however a tendency to under-
prediction, rather than over-prediction. One wonders how true this is
considering the tendency of HES data to include Finished Consultant
Episodes (FCEs) rather than actual admissions. Could they have been more
explicit about how this possibility was dealt with, especially as some of
the conditions they focused on could attract more than one FCEs per
admission (e.g. alcohol related diagnoses could attract traumatologist,
hepatologist and psychiatrist attention)?
Overall, the effort was well thought out and the PARR tool promises
to mark a key turning point in the way intensive case management is
thought about, especially when the next phase of the development
incorporating a wider set of variables is completed.
1. Stukenborg G, Wagner DP, Dembling BP, Connors AF. A Method for
Assessing the Risk of Influenza Attributable Rehospitalization. Available
Accessed 15th August 2006.
Competing interests: No competing interests