There are two possible and closely related mechanisms that may explain the relationship between emotional stress and injurious fall: impoverished postural control and gaze strategy. Aging is frequently accompanied by deterioration in postural control (and preparation) and it has been observed that in order to maintain balance, age-specific compensatory strategies involving the hip (rather than the ankle as in younger subjects) and local muscles (thigh ones) are used to counterbalance a decrease in anticipation [
30]. Further, age-related deficits in the neuro-musculoskeletal systems may impede ability to effectively execute "change-in-support" (CIS) balance-recovery reactions that involve rapid stepping or reaching movements that play a critical role in preventing falls [
31].
The changing gaze strategy is related to poor vision. It is known that poor visual acuity reduces postural stability and significantly increases the risk of falls and fractures in older people [
33]. It is also possible that older people, in stress situation would adopt a gaze strategy detrimental to their balance control, more specifically, a premature gaze transfer which in turn has been associated with decline in stepping accuracy and precision [
34]. Studies that have included multiple visual measures have found that reduced contrast sensitivity and depth perception are the most important visual risk factors for falls [
34].
Methodological considerations
A peculiar feature of studies of falls is that person time at risk is not continuous in real time. The fact that there is only an outcome opportunity when walking, standing, sitting or changing posture calls for special measures when applying the case-crossover design. Measuring information of the time at risk is not easy and many studies do not have the possibility to control for it when calculating the effect estimates. In this study we made an effort to collect such information in addition to traditional information for making case-crossover analyses. Therefore we had the opportunity to test different analytic strategies and the robustness of the found effects. Each analytic strategy implies, however, different degrees and aspects of methodological problems.
The first analysis (A) we performed is as close to correct as we think is possible. The analysis was based on an exact calculation of exposed time thoroughly controlled for time at risk. The narrow time interval of 24 hours was chosen to minimize the risk for information bias and to be able to impute risk time information for exposed cases. Ignoring this could lead to a small underestimation of the exposed time among the exposed cases which in turn could lead to an overestimation of the relative risk.
Had we not had information regarding this time at risk we would have been limited to basing the exposed time on the reported usual frequency of trigger exposure. Besides any information bias in such a variable, the analysis based on this has the disadvantage that the reported exposure information is not restricted to times at risk. For this reason we tested to restrict the study base time in three different ways. Analysis B is a straight forward case-crossover analysis falsely assuming that the risk of falling is equal over the whole day. This leads to an overestimation of the proportion of unexposed time and an overestimation of the relative risk. Analysis C restricted the study base time to time spent awake. This reduces the overestimation of the unexposed time but still, depending on the exposure, leads to an overestimation of the relative risk. Analyses D and E was based on the reported person time at risk. The methodological objection to these analyses is due to the fact that the reported usual trigger frequency was not restricted to time at risk. Making assumptions regarding the distribution of exposure over time at risk and time not at risk is difficult. In analysis D, we made the conservative assumption that all episodes of emotional stress occurred during time at risk. Under the premise that it is unlikely that emotional stress always takes place during time at risk, this analysis leads to an overestimation of the exposed time leading to underestimated effect estimates. In analysis E, the overestimation of exposed time was reduced as the frequency of exposure was distributed correspondingly to the distribution of time at risk to time not at risk. However, this assumption might for some individuals not be true, for example if a person becoming angry always starts to move around in order to physically get rid of the anger. On the other hand, if a patient becoming angry always tries to calm down by sitting down this would also be an incorrect assumption. To test this further, additional information has to be measured on an individual level, which we prompt future studies to take into account if not using the exact method.
The restriction to patients with a normal cognitive function and in some analyses to an exposure period of 24 hour or 48 hour prior to the injury, together with an interview after as little delay as possible, would diminish recall problems leading to non-differential misclassification. To further improve recall of exposures, the interview charted activities during the two days prior to the event and there were good possibilities to link the fall and also potential triggers to activities, events and things that had happened during this time period. According to information from the research nurses carrying out the interviews, this group of elderly people lives organized lives with little variations in their daily activities which further aid their memory. Only a small number of patients reported that the 48 hour period prior to injury was "an unusual day". Excluding those from the analysis did not alter our conclusions. It is also reassuring that analyses based on short and long recall periods show similar results.
Additionally, to combat differential misclassification of exposure between the case and control period, neither the patients nor the interviewers were aware of the assumptions regarding the induction times and were instructed to pay equal interest to the whole 48-hour period, hence, an attribution of exposures in those periods seems unlikely. As hip or pelvic fracture is an acute event, exposure in the case period might otherwise be less likely to be missed or forgotten.
Instances of emotional stress were assessed by simple interview questions allowing for a dichotomous classification. No intensity scales were used. Because the case information is compared with control information from the same patient, there will be no differential exposure misclassification as long as each individual patient uses the same definition of every exposure in all questions, which seems a reasonable assumption. Exposure to other triggers of falls in the period prior to onset might imply confounding. However, co-exposure between the five different emotions was uncommon in this study. Only three patients were exposed to two or three emotions simultaneously during the three-hour period prior to the injury. Exclusion of these cases in the analyses decreased power but did not alter the direction of the estimated effects.
Non-participation because the research nurses did not manage to contact all cases would foremost be a result of the nurses' work load and working hours and not selective with regard to the case's exposure status in the case period. The patients were not informed about the questions in the interview and unaware of the fact that emotional stress would be covered. Hence, it is unlikely that exposure status would have affected their participation. Survival bias is no problem as mortality in the early period after a hip or pelvic fracture is low. The restriction in the recruitment of all patients with a hip or pelvic fracture caused by the inclusion criteria does not lead to bias.
If the patients have difficulties in differentiating between emotions prior to the injury and emotions arising due to or post the injury, it would imply bias. However, of the ToFa patients reporting emotions prior to the injury (in the case window) only one reported it to have happened directly in relation to the fall. The rest of the patients reported it to happen longer before the event (such as 25 minutes) which most likely cannot be mistaken for the feelings arising due to the fall.
A further limitation of the study is the small number of exposed cases. Although the effects are statistically significant, the confidence intervals are wide and this challenges the robustness of the estimated effects.