Authors defend their study to develop tool to predict falls in elderly peopleBMJ 1998; 316 doi: https://doi.org/10.1136/bmj.316.7140.1318a (Published 25 April 1998) Cite this as: BMJ 1998;316:1318
- P T Seed, Lecturer, department of statistics (division of public health sciences),
- D Oliver, Senior registrar,
- F C Martin, Consultant physician,
- A H Hopper, Consultant physician
EDITOR—Altman1 has raised several valid points regarding our study to develop and assess a tool for predicting falls in elderly people (STRATIFY).2 As he says, we sought a practical measure that would be simple to use and yet powerful enough to guide nursing practice in the daily care of elderly people. STRATIFY is based on five factors that predict falls in hospital: falls as the presenting complaint, visual impairment, agitation, need for frequent toileting, and a combined Barthel transfer and mobility score of 3 or 4. We defined “high risk of falling” as any two of the five symptoms. We chose to use falls rather than patients as the unit of analysis on medical grounds. We were interested in a pragmatic tool to predict the risk of any fall, repeated or single.
Altman suggested fitting a multiple logistic regression to the data from phase 1 (a case-control study at St Thomas's Hospital). This gave the following weightings: fall as presenting complaint, 1.20, visual impairment, 1.02, agitation, 2.95, need for frequent toileting, 0.34, and combined Barthel transfer and mobility score of 3 or 4, −0.81. A total above 1.6 would indicate a high risk. Such a scale would be awkward in practice and would be dominated by agitation.
As we were convinced that each of the elements made a genuine contribution to the patient's risk of falling, the negative sign for the mobility score makes no sense. It was not borne out by the later studies. The range of weightings for the other four items is large. A simple unweighted score was preferred in the validation studies (cohort studies at St Thomas's Hospital and in Canterbury).
We applied the original STRATIFY and the weighted scale to the case-control study, the local (St Thomas's) cohort, and the remote (Canterbury) cohort (table). STRATIFY had lower sensitivity than the weighted scale but identical specificity in the original case-control study. However, it had better sensitivity and indistinguishable specificity in each of the cohort studies. It seems that there was overfitting in the logistic model. The high weighting given to agitation was not borne out in the later studies. This may have been a chance effect or because a relatively large number of falls in the original case-control study were by a small number of agitated patients. Altman ends by approving our external validation of STRATIFY and calling for further evaluation elsewhere. We hope to continue the process in further cohorts.