Summary of findings
Using individual-level participant data for 19 trials for three common and important chronic conditions—all with a mean age of less than 65 years—we found that frailty was highly prevalent among trial participants. The frailty index showed the expected relationships with sex, age (apart from in COPD), and disease severity and identified trial participants at higher risk of serious adverse events.
In context of existing literature
Few studies have attempted to measure frailty across multiple clinical trials. To our knowledge, this is the first to include trials not specifically targeting older populations (with most participants aged < 65 years) and the first to do so for T2DM, RA, or COPD. Our findings that frailty can be identified in trials are consistent with two large hypertension trials, HYVET and SPRINT [10, 11], which focused on hypertension management in older people, and one heart failure trial in older people [12], which showed that frailty was relatively common in these trials. We extend these findings by showing that frailty is relatively common in ‘standard’ industry-funded phase 3 trials in younger populations, that it is associated with baseline characteristics, and that frailty at baseline predicts the risk of serious adverse events, even after adjusting for age, sex, and the severity of the index condition.
The frailty index in our analysis showed similar properties to observational studies of frailty using the frailty index approach [22, 23, 26]. As expected, the frailty index had a skewed distribution, was higher in women than men, and for RA and T2DM was associated with age. We have previously shown, using UK Biobank data, that frailty is identifiable in younger as well as older people [1], and the current work shows that this is also true of trials.
While many of the characteristics of the frailty index in the trial data are consistent with studies of frailty using observational cohorts and administrative data [28], the maximum frailty index in the trials (based on the 99th centile of the frailty index distribution) was lower than is typically seen in observational studies [29]. Since this difference was also evident among trial participants aged over 65, it cannot solely be attributed to the younger age of the trial participants. The extent to which this difference is due to trial eligibility criteria [15, 30] (e.g. comorbidities, renal function) or other selection pressures on trial participation (such as the need to be able to undergo multiple trial visits or procedures) is unknown. This suggests that our findings hold for the range of frailty index values we observed in these trials, which is narrower than that observed in unselected populations.
Importantly, while the very frailest patients were rarely included in the clinical trials, we found moderate to severely frail patients—who make up the bulk of those with frailty in the community—were commonly included as participants in clinical trials, despite those trials involving younger people aged under 65 years. Many trials require high disease activity/severity as inclusion criteria, which is one potential explanation for the high prevalence of frailty in some trials, particularly in conditions like RA where there is overlap between functional limitations resulting from active disease and deficits included in the frailty index.
It is notable that the frailty index in the COPD trials was not associated with increasing age, as would be expected. A similar phenomenon was also observed in both the SPRINT and TOPCAT trials (of hypertension and heart failure, respectively), whereby younger trial participants showed relatively higher frailty index values compared to relatively older trial participants [11, 12]. These COPD trials (as well as previous trials showing similar associations) may suggest that to be included in the trial, older people with COPD tended to be relatively less frail than similarly aged people with COPD in the general population. This could arise due to the trial selection process [31], as an example of collider bias, whereby conditioning on a subsequent outcome (trial inclusion) influences the relationship between causally proximal characteristics such as age and frailty [32]. We conducted exploratory analyses of the association between age and the St George Respiratory Questionnaire score and EQ 5D as these are known to increase with age in unselected populations [33, 34]. Like the frailty index, these were not associated with age in the COPD trials. Furthermore, the mean frailty index is lower in the COPD trials, and the range of frailty index values is narrower, compared to the frailty index distribution in previous observational studies of frailty in COPD [35]. This supports our speculation that the unexpected relationship between age and frailty index in these trials reflects differences between the trial population and people in the community with the same condition.
Frailty index was moderately associated with disease severity in COPD and RA. It would have been surprising had there been no association, as functional limitation and frailty, acting across multiple organ systems, are a well-recognised consequence of both diseases [19, 21]. Moreover, FEV1 has long been established as a marker of general physiological reserve as well as of lung disease. The fact that the correlation was not stronger is perhaps of greater interest as it suggests that factors other than the severity of the index disease are important drivers of frailty. Moreover, frailty index predicted adverse events independently of disease severity, indicating that the frailty index contains important clinical information about trial participants beyond that captured by disease-severity measures alone, possibly related to the increasing prevalence of multimorbidity.
Strengths and limitations
A strength of our study is that we used a standard well-validated approach to measure frailty [26], across a large number of trials and a range of conditions, allowing comparison of findings between trials and between conditions. Our analysis also has some important limitations, however. The trials included were not a random sample, but instead were selected from trials that sponsors have made available to third-party researchers for secondary analyses. Not all sponsors share IPD, and those that do share data do not make all trials available. Of the trials we did access, not all trials had sufficient data to identify deficits for inclusion in a frailty index.
The data used to compile the frailty index were not collected for the purpose of identifying frailty, although this is true for most studies using the frailty index. Moreover, medical history data were redacted in most of the included trials, so we were therefore reliant on concomitant medication data to define long-term condition count-based deficits. Consequently, some conditions could only be included as part of a broader group (e.g. cardiovascular disease, obstructive airways disease) rather than as a specific condition, while other conditions (those without specific drug treatments) could not be included [16]. This restricts the number of conditions that could be included in our frailty index, and may result in an under-estimate of the number of conditions present (e.g. in people with multiple cardiovascular conditions which are counted as a single category, or with conditions such as chronic kidney disease which could not be identified using prescribed medications). Furthermore, we used existing instruments, primarily designed to characterise the index condition, to measure functional deficits of frailty (e.g. reduced mobility and difficulty with household tasks were identified using St George Respiratory Questionnaire in the context of COPD, and using the Health Assessment Questionnaire Disability Index in RA). It may be that instruments designed specifically to measure frailty would have improved sensitivity or specificity. Despite these limitations, and especially compared with most administrative data sources, trial data benefits from a wide range of physiological, biochemical, haematological, and functional measures. Moreover, given the regulatory conditions under which trials are conducted, these data were collected, recorded, and processed according to exacting standards.
Implications
Current guidelines caution against the extrapolation of trial evidence to frail people [7, 20], and clinicians lack high-quality evidence about the benefits and harms of common treatments for people living with frailty. Our findings demonstrate that it is feasible to measure frailty, using an established, validated method—the frailty index—in standard industry-funded drug trials, and that on doing so significant numbers of trial participants have mild to moderate frailty. As such, while such trials cannot be claimed to be representative of people with frailty, particularly those with severe frailty who were very rarely found to be present, trials nonetheless contain important under-used information to help address current evidence gaps.
We were able to identify frailty in trials only because we were able to access trial IPD, which is complex and time-consuming. Moreover, several trials redacted data (and, less often, did not collect sufficient data) to allow us to calculate a frailty index. Both to allow clinicians to assess the degree to which frailty is under-represented in particular trials, and to understand whether and how treatment effects differ by frailty (realistically only feasible via meta-analysis of multiple trials), there is a need to expand existing trial conduct and reporting standards [36], to include standard measures of frailty. Our findings suggest that frailty is sufficiently common in trials for this to be a worthwhile exercise.
To that end, standard approaches to the collection and reporting of medical history data (to allow accurate assessment of comorbidities to be included in a frailty index) as well as measures specifically designed to assess frailty (e.g. the frailty phenotype) should be incorporated into international standards for the conduct of trials. Ideally, the adoption of complementary measures such as the frailty index and frailty phenotype measures should be considered. The frailty index can be applied to routinely collected trial data, but is likely to be more influenced by multimorbidity (and in turn, trial inclusion criteria) while the frailty phenotype may identify trial participants with more explicitly defined physiological frailty, some of whom may not have multimorbidity. Given the well-resourced and rigorous measurement and reporting usual in well-conducted trials, the adoption of standard measures of frailty across trials is highly feasible and would allow estimation of the impact of frailty on treatment effects both for individual trials and for meta-analyses of multiple trials. It would also enable identification of participants with increased frailty who are at increased risk of more serious adverse events, who might benefit from closer monitoring.