The association of orthostatic hypotension with falls—an end to the debate?

Orthostatic hypotension (OH) is a common and disabling condition in which a significant blood pressure drop occurs on standing upright from a recumbent position. It is associated with a variety of symptoms, which include dizziness, blurred vision, fatigue, headaches, nausea and tremulousness [1]. It is also thought to be associated with falls, however, this association is widely debated due to conflicting evidence [2]. Previous attempts to pool results evaluating the link between OH and falls have been limited by the reporting methods of included studies [3], making standard meta-analysis redundant. However, meta-analysis of individual participant data (IPD) allows researchers to analyse original study data, instead of analysing aggregate data from published studies. In general, meta-analysis of IPD improves the quality of a meta-analysis, however, it is dependent upon academics agreeing to share their data, which may limit the number of included studies. Nevertheless, the more comprehensive statistical analyses mean that meta-analysis of IPD is now considered the gold standard of systematic review and meta-analysis [4]. Within this method, there are two analysis options—the one-stage and the two-stage approach. As the names suggest, the one-stage approach organizes all IPD into a similar format and then analyses it within a single statistical model, as if it were a huge multi-centre trial. Although this method may seem the most intuitive, it is less appropriate where recruitment or study methods differ between the individual studies. But it is the preferred method when outcomes of interest are rare, study sizes are small or the effect size is large. With a two-stage approach the data from individual studies are first analysed separately then the results are pooled in the second step. The results of the two-stage approach tend to be simpler to interpret and include estimates of heterogeneity. These two approaches typically yield similar results, although not always, as exemplified by Hartog et al.’s study [5]. Hartog et al. perform the first meta-analysis of IPD, exploring the association between OH and falls. The authors identified six prospective observational studies, but were only able to include three in the analysis as original study data were not available in three studies. Nonetheless, by pooling the results of the three studies, the authors had data for 1,022 older people [median age 77 (IQR 73-81)]. Differences between the studies was immediately evident, with the prevalence of OH ranging from 11 to 82%, although the prevalence of fallers was similar between studies (51–62%). In both the one- and the two-stage IPD meta-analysis, there was no association found between OH and falls using logistic analysis. Things were less clear in the time-to-event analysis, perhaps because only two studies provided the data; using the one-stage approach those with OH were 30% more likely to fall during follow-up [n = 592, HR 1.3 (1.08–1.57)]. However, this association was not evident using the two-stage analysis [HR 0.95 (0.74–1.22)]. It is not entirely clear why these findings are conflicting; it may be explained by the small number of studies included but the uncertainty may also be a result of the complexity of both falls and OH. Indeed, falls are typically multifactorial in nature, with complex synergism between risk factors. For this reason, the binary nature of logistic analysis may not be appropriate. Furthermore, the binary nature makes no distinction between an individual who falls on Day 1 of the study and an individual who has their 1st fall 364 days later. An additional consideration which may explain the discrepancy in results are the findings from McDonald's study, recently published in Age and Ageing [6]. In this prospective observational study, the association between OH and falls was evaluated but with a specific focus on the criteria used to diagnose OH. Using the 1996 consensus criteria [7] (systolic/diastolic drop of >20/10 mmHg upon standing) there was no increased risk of falls during observation using both logistic and Cox proportional hazard models. However, when the 2011 consensus criteria [8] were used (which includes updated criteria for initial OH and OH in the presence of hypertension) there was a strong association between OH and falls in both the logistic and the time-to-event analysis. Similarly, a very recently published study arising from TILDA (The Irish Longitudinal Study on Ageing) has identified a significantly increased risk of falls in those with sustained but not initial OH [9]. Unfortunately, these papers were published too recently to be included in Hartog's study. If the association between OH and falls is still in doubt, should we still be including it in multifactorial falls prevention (MFFP)? Well, yes. There is good evidence for the effectiveness of MFFP and most multifactorial interventions have included an assessment and treatment of OH [10] so our evidence-based practice should reflect this. It seems likely that orthostatic hypotension (OH) is a risk factor for falls. Diagnostic criteria may influence the association between OH and falls. As falls are complex and multifactorial a time-to-event analysis is more appropriate than logistic models.