The Go/No-Go test exhibited the strongest effect size in the direct performance comparisons of fallers and non-fallers and remained as a significant predictor in all regression models making it a potentially promising assessment tool for fall prediction. It is noteworthy, however, that although the AUC metric for Go/No-Go is the second largest of all assessments, its accuracy in classification was only moderate to acceptable [28, 29].

Still, this confirms the special role of inhibition for falls, as observed in other studies. For instance, Schoene et al. (2017) (p.723) [21] looked into the predictive ability of a similar step-based inhibition assessment and found that its effect on falls was “direct and not mediated by processing speed, attention, and balance, further supporting the notion of iCSRT-RT [inhibitory choice reaction time] being an independent fall risk factor.” Similarly, Mirelman et al. (2012) [19] found that among several computerized cognitive assessments, only response inhibition and attention were significantly associated with a future fall risk. This observation is not surprising, given the demands of daily activities and outdoor walking, where distractions frequently require a rapid response or the inhibition thereof.

This need for rapid response in order to avoid fall incidents might also explain the significant differences in RTT results between fallers and non-fallers, coupled with a statistically significant though poor accuracy in classification according to AUC analysis. However, the regression analyses did not yield statistically significant contributions from RTT, suggesting that, compared with inhibitory choice stepping reaction time (iCSRT), simple choice stepping reaction time (CSRT) holds less significance. This is consistent with previous research. For instance, Lord and Fitzpatrick (2001) [30] observed increased CSRT in older individuals prone to falls, yet subsequent studies such as Schoene et al. (2017) [21] who incorporated an inhibitory test component found that this component improved the predictive ability of CSRT.

Conversely, the cognitive flexibility test failed to differentiate fallers from non-fallers. Pieruccini-Faria (2019) [31] identified concept formation (a sub-component of cognitive flexibility) as a predictive factor for falls and also a confounding variable in the association between balance and falls. However, they also found that this confounding effect was more pronounced in individuals with poor balance, and the overall cognitive flexibility score did not emerge as a significant predictor for falls. These findings could explain the outcomes of our study in which participants exhibited a rather good balance and in which cognitive flexibility assessment was based just on a reaction time score.

The instrumented assessments without a cognitive component (Coordinated Stability Test and the Sway Test) demonstrated a poor discriminatory ability in all statistical analyses, which is not in line with previous research showing strong associations between poor balance and an elevated risk of falls [31]. Our findings deviate, either entirely [32,33,34] or at least partly [30, 35], from previous research on fall risk factors.

For instance, similar to our study, Lord et al. (2001) [30] did not discover any significant differences in sway test measures between fallers and non-fallers. However, they did find worse performance in a Coordinated Stability Test in fallers. Noteworthy, the predictive ability of this Coordinated Stability Test turned out to be weaker compared to CSRT, and both the Coordinated Stability Test and the Sway Test exhibited significant associations with CSRT in the study by Lord et al. (2001).

One explanation for the differing results of our study regarding the discriminative ability of the balance assessments could be that, as described above, the balance ability of both fallers and non-fallers was high, for instance higher than that of participants in a prior usability study using the same Dividat Senso assessments [25] which is why a ceiling effect might have occurred. This in turn would explain why, in the current study population, balance is a non-determining factor in terms of fall risk. Findings by Johansson (2017) [33] supported this idea since they found a non-linear relationship between postural sway length and number of falls with a significantly greater fall frequency in the fifth quintile of sway length. Furthermore, the disparities from other studies might also be attributed to the slightly lower mean age in the present study, as age has exhibited strong negative correlations with balance ability and positive associations with the interplay of sensory, motor, and cognitive functions [31, 36]. Finally, as Zhou et al. (2017) [35] pointed out, standing postural sway is complex, as it depends on various inputs (e.g., somatosensory, visual, vestibular). Accordingly, they found that traditional postural sway metrics, such as those applied in our study, did not differ between fallers and non-fallers, whereas measures of CoP entropy—non-linear time-series analytical techniques—were able to predict falls. This was confirmed by a previous retrospective study revealing a stronger discriminative ability of such temporal dynamics as compared to traditional postural measures analyzing spatial dynamics of balance [37].

In summary, our results suggest that especially in a rather high functioning population, balance might play a subordinate role in fall risk and that simple balance metrics are not sufficient. This emphasizes the importance of integrating a cognitive component in the instrumented assessments.

The TUG test results differed significantly between fallers and non-fallers and the test exhibited significant (though weak) accuracy in classification. However, it did not contribute significantly in either of the two main regression models. Thus, the TUG showed some discriminative ability, but, overall, this ability appeared weaker compared to the Go/No-Go Test.

This limited discriminative ability of the TUG aligns with previous literature [38] asserting its usefulness in fall risk assessment primarily in more frail older populations [39]. This is further underscored when examining cut-off values detected in the present (13.5 s) and in previous studies. Although proposed cut-off values vary widely between studies [39], most commonly ≥ 11 s or even ≥ 12.34 were recommended [6, 13, 39,40,41]. In our study, however, the average test completion time (8.6 s) was below most thresholds defined in previous studies. Another explanation is provided by Chiu et al. (2003) [42] who found that the TUG test is highly sensitive in differentiating multiple-fallers from non-fallers, however, less sensitive in differentiating single-fallers from non-fallers. In our study, though, the majority experienced only a single fall.

As mentioned in the introduction, it must be considered that the instrumented TUG might be more reliable in predicting fall status. For instance, Ponti et al. (2017) [13] found that the pure completion time was not significantly different between fallers and non-fallers and reached an AUC value of 0.668, whereas the fusion of features extracted from accelerometer data resulted in a significant group difference and increased the discriminative ability to an AUC value of 0.84.

In summary, as Chiu points out, in a rather healthy population the “low discriminative ability of the TUG might indicate that the task involved could not challenge the mobility and balance functions of older people enough to reveal their risk for falls”  [42 (p.48)].

Due to multicollinearity with the TUG, the TUG-DT had to be excluded from the first main regression analysis. Therefore, this regression analysis was repeated with the TUG-DT as an independent variable and with the TUG excluded instead. Remarkably, TUG and TUG-DT were mutually interchangeable without concomitant alterations in regression model accuracy and amount of explained variance (χ2(10) = 29.21, p = 0.001; Nagelkerke’s R2 = 0.283) and with a similar non-significant (p = 0.914) relative contribution to predicting fall status, suggesting either a diminished discriminative capacity or a shared measurement of the same construct in our population. This observation contrasts a number of prior studies that underscore the predictive superiority of dual-task assessments over single-task conditions [26, 43]. However, according to a recent systematic review comparing the ability of dual-task versus single-task tests to predict falls, only half of the included studies could confirm this superiority of dual-task tests which is why no definitive conclusions can be drawn yet [44].

Overall, there is conflicting evidence regarding whether to consider TUG(-DT) as standard geriatric fall risk assessments in the first place. Nevertheless, owing to their prevalent usage in clinical settings and the absence of universally accepted method, we opted to compare the instrumented assessments to TUG and TUG-DT, guided in part by recommendations such as those by Ambrose et al. [45].

The STS test stood out in the AUC analysis, showing the highest accuracy in classification of all assessments. This accentuates its potential as an efficient clinical fall risk assessment and is in line with previous research [46]. One explanation might be that STS performance reflects various sensorimotor functions, balance, psychological processes, and transfer skills [47], all of which have been associated with falls.

The biggest limitation of this study is that falls were assessed retrospectively. A prospective or longitudinal analysis is imperative for future investigations. Additionally, data on falls was collected based on self-report, leading to potential (although low) risk of recall-bias. Finally, the standardized geriatric assessments examined in this study are not exhaustive. There are other established geriatric tests which are used for fall risk screening such as the aforementioned Berg Balance Scale [9] and the Tinetti Mobility Test [10].