File
Hilfiker R1, Schädler S2, Sattelmayer M1
1HES-SO Valais-Wallis. University of Applied Sciences and Arts Western Switzerland Valais, Physiotherapy, Leukerbad, Switzerland, 2Physiotherapie Stefan Schädler, Sumiswald, Switzerland
Background: Cut-offs are widely used by physiotherapists to dichotomise test results in falls prediction. This common practice using cut-offs has several important problems: i) Loss of information because of dichotomisation (e.g. for a cut-off of ≥ 13.5 seconds in the Timed Up and Go-Test (TUG), an older adult with test-result of 13.6 seconds is classified at the same risk as one with a result of 35 seconds); ii) the true optimal cut-off is unknown (i.e. the identified cut-off in a given study is associated with sampling error, hence repeating the study would give a more or less different cut-off. Sampling error is larger in small studies and most studies are small; iii) in systematic reviews and meta-analyses, the reported cut-offs show large heterogeneity which is often not explainable by clinical factors, such as age or pathology, and settings, such as community dwelling, hospitalised or nursing-homes. Hence, it is quite impossible to define cut-offs and existing examples show that dichotomising is not necessary.
Purpose: Using a systematic review approach, we wanted to illustrate the above-mentioned problems.
Methods: We performed a systematic review and included systematic reviews and meta-analyses of prospective cohort studies. In addition, we included single prospective studies that were not yet included in published systematic reviews. Test accuracy data to predict a fall of the following assessments were extracted: Timed Up and Go-Test (TUG), Berg Balance Scale (BBS), Performance Oriented Mobility Assessment (POMA), Dynamic Gait Index (DGI), Functional reach (FR) and Balance Evaluation System Test (Mini-BESTest). We plotted sensitivity over 1 minus specificity (summary receiver operator plot). The heterogeneity was analysed to evaluate whether the determination and recommendation of a single cut-off is clinically meaningful. Subgroups were analysed to identify sources of heterogeneity (i.e. clinical characteristics or settings).
Results: We extracted data from 25 studies for the TUG, 15 for the BBS, 2 for the POMA, 4 for the DGI, 4 for the FR and 7 for the Mini-BESTest. Large heterogeneity was identified for sensitivities and specificities of all 6 assessment tools. The largest difference between cut-off values was reported for the TUG (8 to 32 seconds). We would have expected to find an ordering of the cut-offs in regard of sensitivity/specificity (i.e. higher cut-offs in TUG should lead to higher specificity); However, the results did not show this picture. A clear pattern of cut-offs was not observed for neither of the assessments, indicating that no cut-off can be determined as the “best” cut-off.
Conclusion(s): Our results confirm that it is not possible to determine meaningful cut-offs for each of the six falls-prediction tests.
Implications: Research and practice should abandon the use of cut-offs when using falls-prediction (or tests in general). Instead, probabilities should be calculated, for example with a logistic regression model and the process of “dichotomisation” should be left to the practitioner and the patient in their shared decision making process, taking into account more than one assessment, other available quantitative or qualitative information, and the concrete situation of the patient . An existing tool using this approach can be found online (http://ffrat.farseeingresearch.eu/runAssessment).
Keywords: Prediction, cut-off, falls risk
Funding acknowledgements: No funding received for this project.
Purpose: Using a systematic review approach, we wanted to illustrate the above-mentioned problems.
Methods: We performed a systematic review and included systematic reviews and meta-analyses of prospective cohort studies. In addition, we included single prospective studies that were not yet included in published systematic reviews. Test accuracy data to predict a fall of the following assessments were extracted: Timed Up and Go-Test (TUG), Berg Balance Scale (BBS), Performance Oriented Mobility Assessment (POMA), Dynamic Gait Index (DGI), Functional reach (FR) and Balance Evaluation System Test (Mini-BESTest). We plotted sensitivity over 1 minus specificity (summary receiver operator plot). The heterogeneity was analysed to evaluate whether the determination and recommendation of a single cut-off is clinically meaningful. Subgroups were analysed to identify sources of heterogeneity (i.e. clinical characteristics or settings).
Results: We extracted data from 25 studies for the TUG, 15 for the BBS, 2 for the POMA, 4 for the DGI, 4 for the FR and 7 for the Mini-BESTest. Large heterogeneity was identified for sensitivities and specificities of all 6 assessment tools. The largest difference between cut-off values was reported for the TUG (8 to 32 seconds). We would have expected to find an ordering of the cut-offs in regard of sensitivity/specificity (i.e. higher cut-offs in TUG should lead to higher specificity); However, the results did not show this picture. A clear pattern of cut-offs was not observed for neither of the assessments, indicating that no cut-off can be determined as the “best” cut-off.
Conclusion(s): Our results confirm that it is not possible to determine meaningful cut-offs for each of the six falls-prediction tests.
Implications: Research and practice should abandon the use of cut-offs when using falls-prediction (or tests in general). Instead, probabilities should be calculated, for example with a logistic regression model and the process of “dichotomisation” should be left to the practitioner and the patient in their shared decision making process, taking into account more than one assessment, other available quantitative or qualitative information, and the concrete situation of the patient . An existing tool using this approach can be found online (http://ffrat.farseeingresearch.eu/runAssessment).
Keywords: Prediction, cut-off, falls risk
Funding acknowledgements: No funding received for this project.
Topic: Outcome measurement
Ethics approval required: No
Institution: n.a.
Ethics committee: n.a.
Reason not required: Systematic review
All authors, affiliations and abstracts have been published as submitted.
