The likelihood chi-square test statistic is 24.87 which is statistically significant at the 1% level (p=≤0.00) (see Appendix 8a). Therefore, the overall model is statistically significant.
Predictor Estimated Odds Ratio p-value 95% CI
Age 0.93 0.19 0.83, 1.04
Sex (male) 0.59 0.45 0.15, 2.36
Cognition (MoCA) 1.07 0.46 0.88, 1.31
Gait speed (m/s) 0.004 0.02 0.00, 0.35
mDTC (back-count) 0.99 0.69 0.95, 1.04
Balance (BBS) 1.03 0.63 0.91, 1.16
Falls risk (PPA) 1.63 0.05 0.99, 2.67
Legend: CI=confidence intervals, MoCA=montreal cognitive assessment, m/s=meters per second, mDTC=mean dual-task cost, BBS=berg balance scale, PPA=physiological profile assessment
Table 2.8: Estimated odds ratios for variables influencing falls experienced
60 | P a g e Gait speed (p=0.02) and falls risk scores (p=0.05) were statistically significant (Appendix 8b) (Table 2.8). The likelihood of having a fall was reduced as gait speed increases. The odds ratio for gait speed (m/s) of 0.004 was transformed using natural log to determine an OR 0.58 per 0.1m/s (Appendix 2.8b).
Therefore an increase in gait speed of 0.1 m/s decreases the odds of falling by 42% (OR 0.95). The likelihood of having a fall is increased as the falls risk score (PPA) increases. A unit increase in falls risk score increases the odds of falling by 63% (OR 1.63). For both of these coefficients, we can be 95% certain that these results are not due to chance. However, the other coefficients (age, sex, MOCA, mDTC, and balance) demonstrate no significant difference in falls categorisation. These results indicate that falls experienced in this sample are not significantly influenced by age, gender, general cognition, or the average cost accrued in gait and cognition while completing a secondary task.
Figure 2.4 demonstrates that all the predicted values span between 0 (no fall) and 1 (fall) but that most are featured closer to the 0 with a mean of 0.33 (Appendix 8c).
Figure 2.4: Dot plot of history of falls frequency in previous six months
61 | P a g e Predicted probabilities can also be described according to the independent variables (Appendix 8c). At the mean gait speed of the whole sample
(mean=0.88m/s), there is a 24% chance that a fall will be experienced (95% CI 11, 37). Going from the fastest gait speed to the slowest gait speed will
increase the chance of falling by 96%. As gait speed reduces by one standard deviation (centered on the mean, ±0.32), there will be a 76% increased chance of falling (Appendix 8d). The range of gait speed in this sample is limited, without a linear relationship between gait speed and falls experienced.
Therefore the chance of falling will not continue along a linear trajectory.
At the mean falls risk score (mean PPA=2.5), there is a 25% chance that a fall will be experienced (95% CI 12, 38). Going from the highest falls risk score to the lowest will increase the chance of falling by 57%. As the falls risk score reduces by one standard deviation (centred on the mean, ±1.67), there will be a 75% chance of falling. For every unit reduction in falls risk score, there is a 7%
increased chance of falling (Appendix 8e). As with gait speed, there is not a linear relationship between falls risk and falls experienced.
62 | P a g e Model Explanatory variables included in model LR chi2
(within fmodel)
p-value
Fmodel Sex, age, cognition, gait speed, mDTC, falls
risk and balance. - -
n1model Sex, age, cognition, gait speed, falls risk
and balance. 0.17 0.68
n2model Sex, age, gait speed, falls risk and balance. 0.81 0.67
n3model Sex, age, cognition, and mDTC. 22.39 ≤0.01
intercept_only n/a 24.87 ≤0.01
Legend: LR=likeihood ratio, mDTC=mean dual-task cost.
Table 2.9: Likelihood ratio models for analysis of robustness of models for falls experienced
Two of the LR tests were not statistically significantly different from the full model (n1model nested in fmodel, LR chi2 = 1.17, p=0.68; n2model nested in fmodel, LR chi2 = 0.81, p=0.67) (Table 2.9). The null hypotheses that neither the mDTC or the mDTC and cognition had an independent effect on falls risk was not rejected. The LR test between the n3model and full model is statistically significantly different (n3model nested in fmodel, LR chi2 = 22.39, p=≤0.01).
Therefore, the hypothesis that gait speed, falls risk, and balance have no effect on falls risk can be rejected. The full model provides the best fit for the data and indicates that gait speed, falls risk, and balance variables influence the likelihood of an individual experiencing falls (Appendix 8f).
A further LR test was used to test the null hypothesis that the effects of all the independent variables being simultaneously equal to zero. A model with no independent variables (intercept_only) was statistically significantly different from fmodel (intercept_only nested in fmodel, LRX2 = 24.87, p=≤0.01) and therefore the null hypothesis (that none of the variables influence the likelihood of experiencing a fall) can be rejected (Appendix 8g).
63 | P a g e In consideration of the LR modelling, the hypothesis that cognition and mDTC will influence falls, cannot be accepted. There is statistical significance
attributed to the inclusion of the physical outcome measures. However, because of the statistical difference of the intercept-only model, we can also infer that the cognitive components do have some influence, although not to the same degree as the physical components. Therefore, the full model is the better model. This model can be further tested for robustness before conclusions are drawn. The predicted values from our full model correctly classified 75% of the predicted scores against the actual scores (Appendix 8h). The overall fit of the model using McFadden’s Pseudo-R2 demonstrates a good fit, as the Pseudo-R2 of 0.29 is within the recommended values of 0.2 to 0.4 (Appendix 8i).
64 | P a g e
2.5. Discussion
The aims of this study were to describe physical characteristics of gait, mDTC, balance, and falls risk in older adults with mild CI; to explore associations between these characteristics and their relation to cognition; and compare to these characteristics between those that are or are not already reporting falls.
Results from a cross-sectional survey are presented, and the key findings summarised and compared with the literature.