Predicting Falls.

To the Editor:

Having read the article entitled "Predicting the Probability for Falls in Community-Dwelling Older Adults Using the Timed Up & Go Test" in the September 2000 issue of Physical Therapy, I am concerned about the validity of the message presented. Two issues appear to seriously compromise the message the authors present.

First, the title of the study implies a study of predictive validity of measurements obtained with the Timed Up & Go Test (TUG). However, the study was not a prospective study in which the participants were measured using the TUG and then followed for a period of time during or after which data about the event to be predicted (ie, falls) were recorded. The participants in this study were studied at one point in time, with information collected on their history of falls (eg, in the past 6 months) and current measurements using the TUG. At best, the design of the study could allow for an investigation of the ability of a test score to distinguish between older people who have fallen in the past and older people who have not fallen in the past. The study design does not enable the investigators to make a statement about the predictive ability of the TUG.

The second issue, though less important given the design is not a study of predictive validity, involves the sample selection. The authors describe the ability to use the TUG to distinguish between older people who are at risk for falling and older people who are not at risk for falling by determining the sensitivity and specificity of the measurements for identifying people who are at risk for falling, as indicated by their history of falling. However, the selection criteria the authors used to assign older people to 2 groups--older people with no history of falls and older people with a history of 2 or more falls--result in a sample with spectrum bias.[1-3] The task of identifying people who are at risk for a problem from people who are not at risk for a problem is most difficult in and most useful in conditions in which a measure is applied to clinical decision making for people who are "borderline" with respect to the problem. The people not at the extremes of the spectrum of risk and no risk--those in the middle with "intermediate" characteristics of those with or without the problem--are the people for whom the sensitivity and specificity of the data obtained with a measure are useful in recognizing people who are at risk.

By selection, the authors have clearly excluded people presumably at intermediate risk for falling, those with one fall in the previous 6 months. In fact, given the selection bias, people with a history of falls could be identified nearly as well based on the use of an assistive device alone (sensitivity, 12/15 [80%]); people without a history of falling were more correctly identified by their lack of use of an assistive device (specificity, 15/15 [100%]; overall recognition accuracy, 27/30 [90%]).

The authors describe the sample in the title and text as community-dwelling older adults, but the lack of information about the volunteer sample from which the subjects were chosen, how many were excluded, and characteristics of those excluded further limits an understanding of how representative the sample is of community-dwelling older people. Thus, any message from the study is obscured by spectrum bias and without an initial understanding of the spectrum.

Any further discussion of inaccuracies in the reporting of the research seems useless given the difficulties in the design and sample described. However, the report of identification of 12 older people from the group of 15 people with a history of falls using the TUG with an added manual task ([TUG.sub.manual]) raises further concerns about the accuracy of reporting. The authors indicated that all people using a walker were excluded from the [TUG.sub.manual] trial. Thus, only 10 subjects could have participated in the [TUG.sub.manual] trial, eliminating the possibility of identifying 12 people with a history of falling.

How the reported research contributes to our understanding of identifying older people who are at risk for falling or to the measurement of physical function is unclear. I find the article title misleading and the conclusions inaccurate, as the study provides no information about the ability to predict falls using the TUG. The authors owe readers an accurate report and interpretation of their findings; this report does not appear to meet such a standard.

Jessie VanSwearingen, PT, PhD Associate Professor Department of Physical Therapy University of Pittsburgh 6035 Forbes Tower Pittsburgh, PA 15260

References

[1] Ransohoff DF, Feinstein AR. Problems of spectrum bias in evaluating the efficacy of diagnostic tests. N Engl J Med. 1978; 299:926-930.

[2] Miller TQ, Turner CW, Tindale RS, Posavac EJ. Disease based spectrum bias in referred samples and the relationship between type A behavior and coronary artery disease. J Clin Epidemiol. 1988; 41:1139-1149.

[3] Laths MS, Nachamkin I, Edelstein PH, et al. Spectrum bias in the evaluation of diagnostic tests: lessons from the rapid dipstick test for urinary tract infection. Ann Intern Med. 1992;15:135-140.

Author Response:

We appreciate Dr VanSwearingen taking the time to express her concerns regarding our report. In response to her point regarding the predictive aspect of this research, we continue to feel that the article itself is a clear representation of the research done; however, we realize that a different title might have eliminated the confusion regarding the purpose of our article. Dr VanSwearingen states, "At best, the design of the study could allow for an investigation of the ability of a test score to distinguish between older people who have fallen ... and older people who have not fallen." This was, in fact, the stated purpose of the research: "We investigated the sensitivity and specificity of the Timed Up & Go Test under single and dual task conditions in identifying fall prone older adults living in the community" (page 901). Predicting in the context of this report did not refer to predicting a future outcome, but rather predicting an outcome (being a faller) across a field of people. We are sorry for the confusion related to the title used for this article.

Her second concern related to the sample used in this research. We used a method known as receiver operating characteristic (ROC) curve analysis. We are sorry this was not more clearly stated in our methods. A ROC analysis evaluates the inherent predictive value of a variable to differentiate people known to have a problem (in this case, a history of falls) from those who do not.[1,2] The use of this analysis requires that subjects be previously classified into one of two groups, hence, our use of retrospective classification. Finally, the sample selected was a convenience sample of older adults who had participated in other aging studies. Results might have been different had subjects been drawn at random from the community or, in fact, represented a broader range of fall risk, including intermediate fall risk. This represents a limitation in the research, and, as is true for most research studies, further research is needed to confirm and expand results from this study.

Anne Shumway-Cook, PT, PhD Associate Professor Division of Physical Therapy University of Washington Box 356490 Seattle, WA 98195

Sandy Brauer, PT, PhD Postdoctoral Fellow Department of Exercise and Movement Science University of Oregon Eugene, Ore

Marjorie Woollacott, PhD Professor and Chair Department of Exercise and Movement Science University of Oregon

Robin High Statistical Consultant

References

[1] Swets JA. Indices of discrimination or diagnostic accuracy: their ROCs and implied models. Psychol Bull. 1986;99:100-117.

[2] Swets JA. Measuring the accuracy of diagnostic systems. Science. 1988;240:1285-1293.