We observed an inverse U-shaped relationship between Rankin score and the risk of post-stroke falls. The highest risk was at an mRS of 2. This may be explained by an interplay between exposure to circumstances where falls may occur and one’s physiological susceptibility to falls. Persons with low functional status are less likely to be mobile, are more likely to be physically inactive and hence are exposed to fewer circumstances where they may sustain a fall. On the other hand, persons with higher functional states are more likely to have intact motor and sensory functions for maintaining balance, and hence less likely to fall. Persons with moderate functionality experience a mix of these. They have an impaired ability to maintain balance yet retain some reasonable mobility, and thus attempt to be mobile. This may explain why persons in the middle of the spectrum of physical function are most prone to falls.
This finding corroborates with what has been found in three other studies which showed a similar relationship between function and post-stroke falls(17–19). Two of these studies are well-powered and utilised cohort study designs(17,18), one of which used a large registry-based cohort with data on long-term outcomes(17). In contrast, most other studies demonstrate that worse disability and function are associated with increased falls risk(5,20,21). These studies often dichotomize physical function(21,22) or analyse functional scores as continuous variables on a linear model(20) which may obscure this observation. Future studies may assess if patterns of falls (e.g. mechanism of fall, location) sustained by patients differ between functional levels. Targeted interventions could be potentially developed based on these and evaluated.
The potential positive association between antithrombotic therapy (anticoagulation and antiplatelets) and falls likely arises from a bias in the detection of falls. Patients on antithrombotic therapy who fall are managed with greater caution due to greater risks of haemorrhage and are more likely to undergo brain scans and receive inpatient care(23,24). Also, these patients may be more likely to present for medical attention for falls due to precautions advised in case of trauma while on antithrombotic therapy. We are further reassured that neither agents were associated with worse post-fall survival in the sensitivity analysis. Of note, only 4 (0.18%) of the patients with modified Rankin score sustained a head injury with AIS score of 3 or more during the study period, and of all the ischemic stroke survivors in the registry, only 243 (1.1%) sustained a head injury with an AIS score of 3 or more. Concerns that anti-coagulation may lead to major intracranial haemorrhage after a fall are allayed by the low incidence of serious head injuries and the no-worse survival profile.
Limitations and Strengths
The injurious fall rate in our study is 2.4%, much lower than the proportion of all fallers among community stroke survivors observed in other studies ranging between 23.0% and 55.0% (5) in the literature. This is because the NTR captures only falls leading to emergency department presentations, capturing significant incidents but is not representative of all falls. In contrast, most studies use self-reported fall outcomes either through interviews or diaries(5,25), which would detect minor falls not picked up by the NTR. In this respect, this study complements the literature on prospective falls, by focusing on falls serious enough to present to hospital. However, it also carries the limitation of omitting near-miss falls or seemingly minor falls, events that are known to have implications for frailty and future falls.
In the multivariable regression model, 15 individual factors were included for being established risk factors for falls. While this may risk over-fitting with the additional variables, the explanatory model maintains the inverted U-shaped trend (in relation to mRS) observed in the univariate analysis (Table 1) and confirms that the trend is not due to confounding.
A low proportion of all stroke patients within the registry had mRS recorded (10.3%). There was a significant baseline difference in the history of TIA or stroke (14.4% vs 20.3%), a small but statistically significant difference in age (66.6 years vs 67.7 years), but no significant differences in haemoglobin levels at admission and cardiovascular co-morbidities (diabetes mellitus, hypertension, hyperlipidaemia and atrial fibrillation).
Of note, the patients with recorded mRS included in our analysis were different from those without mRS in several ways. Patients in our study, who had a complete mRS, were more likely to be on a stroke pathway (93.6% vs 73.2%) than those that were excluded, and more likely to be discharged home. They also had a lower mortality (4.1% vs 20.7%), and lower risks of falls (2.4% vs 10.2%), suggesting that there were uncaptured baseline differences in other co-morbidities and functional status that contributed to differences in outcome.
These differences could explain why, at discharge, they were more likely to receive antiplatelet agents (85.1% vs 82.9%) or oral anticoagulation (11.8% vs 8.3%) than the patients missing mRS. Furthermore, patients missing mRS might include patients assessed by clinicians to have higher fall risk or lower mortality benefit, patients with poor compliance to anticoagulation follow-up and patients who might also not be compliant to rehabilitation follow-up where the mRS would be scored. Another explanation is that patients missing mRS likely included patients with subclinical stroke diagnosed incidentally when presenting for another medical problem, or those presenting with multiple medical issues, hence accounting for the much lower proportion treated in stroke units. Hence, our study findings may be generalizable primarily to patients presenting with acute stroke as the main presenting complaint, and less generalizable to patients where stroke is not their primary presenting complaint, as they were more likely to be missing mRS in our registry and excluded from our study.
Another potential bias comes from the timing of the score. As scoring of functional status is commonly done in rehabilitation(26), the sample may select for persons who are deemed more appropriate for rehabilitation and may reflect better health and prognosis. Finally, as with all registry-based studies, we were only able to include risk factors in our model that were mandated in the registries used.
The strengths are that we used nation-wide registry data involving multiple centres with a large number of patients, with data from a 5-year period.