Letters

Differences in mortality after fracture of hip

Nevertheless, we conducted fairly extensive interviews with patients in addition to extracting records from case notes and thus were able to enter a reasonably large number of variables into the logistic regression model. Specifically, these were age; sex; the prior activities of daily living score; the type of residence from which patients were admitted; the number of pre-existing illnesses; the type of fracture; the length of, delay to, type of, and timing of operation; the anaesthetic given; the use of prophylactic anticoagulants and antibiotics; and (as dummy variables) individual pre-existing illnesses and hospital. Our analysis used forwards entry of variables in 441 cases for which data were complete and was subsequently repeated on the smaller subset of indicated factors in 547 cases (14 missing). As reported, only the first three variables listed above and (from the last two variables listed) cardiovascular disease and hospital 6 emerged as significant predictors of death. All of these findings, including the significance of male sex--although not the apparent protective effect of one hospital--are consistent with the literature.⁴

Packham questions the adequacy of the model and the proportion of variability explained. In logistic regression, however, there is no measure analogous to the R² of multiple linear regression (used for continuous response variables) to indicate such a proportion. At best a 2 x 2 table can be generated, matching the observed "yes/no" outcome with expected outcome according to whether the predicted probability exceeds an arbitrary cut off value such as 0.5, though this is rarely informative. Our interest is focused on identifying which characteristics of patients or hospitals are associated with the observed differences in mortality.

Finally, we agree that adequate adjustment for case mix is difficult and hence reiterate our previous comments that our results are hypothesis generating rather than definitive. We hope that they give rise to further research.

Director Medical statistician Department of Community Medicine, Institute of Public Health, University of Cambridge, Cambridge CB2 2SR

Consultant in public health medicine Research/audit assistant Senior registrar Directorate of Public Health Medicine, Anglia and Oxford Regional Health Authority, Cambridge

Research registrar Department of Orthopaedics, Peterborough District Hospital, Peterborough PE3 6DA

Consultant Department of Medicine for the Elderly, Norfolk and Norwich Healthcare NHS Trust, Norwich NR1 3SR

Consultant orthopaedic surgeon Orthopaedic Research Unit, University of Cambridge Clinical School, Addenbrooke's Hospital, Cambridge CB2 2QQ

C J Todd, Chris Palmer, C Camilleri-Ferrante, C J Freeman, C E Laxton, M J Parker, B V Payne, N Rushton 

1. Packham CJ. Differences in mortality after fracture of hip. BMJ 1995;311:571-2. (26 August.) [Free Full Text]
2. Todd CJ, Freeman CJ, Camilleri-Ferrante C, Palmer CR, Hyder A, Laxton CE, et al. Differences in mortality after fracture of hip: the East Anglian audit. BMJ 1995;310:904-8. (8 April.) [Abstract/Free Full Text]
3. Todd CJ, Palmer CR, Camilleri-Ferrante C, Freeman CJ, Laxton CE, et al. Differences in mortality after fracture of hip. BMJ 1995;311:572-3. (26 August.) [Free Full Text]
4. Sainsbury R. Hip fracture. Reviews in Clinical Gerontology 1991;1:67-80.