Exploring the impact of age, frailty, and multimorbidity on the effect of ICU interventions: A systematic review of randomized controlled trials

Abstract

Objective:

To date, age, frailty and multimorbidity have been used primarily to inform prognosis in older adults. It remains uncertain, however, whether these patient factors may also predict response to critical care interventions or treatment outcomes.

Data Sources:

We conducted a systematic search of top general medicine and critical care journals for RCTs examining critical care interventions published between January 1, 2011 and December 31, 2021.

Study Selection:

We included RCTs of critical care interventions that examined any one of three subgroups – age, frailty, or multimorbidity. We excluded cluster RCTs, studies which did not report interventions in an intensive care unit, and studies which did not report data examining subgroups of age, frailty, or multimorbidity.

Data Extraction:

We collected study characteristics (single vs. multi-country enrollment, single vs. multicenter enrollment, funding, sample size, intervention, comparator, primary outcome and secondary outcomes, length of follow-up), study population (inclusion and exclusion criteria, average age in intervention and comparator groups), and subgroup data. We used the ICEMAN instrument to evaluate the credibility of subgroup findings.

Data Synthesis:

Of 2037 unique citations, we included 48 RCTs comprising 50,779 total participants. Seven (14.6%) RCTs found evidence of statistically significant effect modification based on age, while none of the multimorbidity or frailty subgroups found evidence of statistically significant subgroup effect. Subgroup credibility ranged from very low to moderate.

Conclusions:

Most critical care RCTs do not examine for subgroup effects by frailty or multimorbidity. While age is more commonly considered, the cut-point is variable and relative effect modification is rare. Although interventional effects are likely similar across age groups, shared decision making based on individual patient preferences must remain a priority. RCTs focused specifically on critically ill older adults or those living with frailty and/or multimorbidity are crucial to further address this research question.

Brief summary (Twitter):

Perrella et al: Systematic review examining critical care interventions in patients of older age, higher frailty status, or increased comorbidities. Despite ongoing recognition of #frailty as an important prognostic predictor in critical illness, data reman lacking in current #EBM in #criticalcare.

Introduction

Intensive care unit (ICU) demographics are shifting. Recent data from the United States demonstrates that patients aged 85 and older account for 20% of all ICU admissions (1). Canada has seen a similar aging of its ICU population compared to 15 years ago (2). These trends coincide with medical advances that have allowed for longer life expectancies and reduced mortality from critical illness.

Older age is presumed to be a commonly associated risk factor for ICU mortality (3–5). However, a post-hoc analysis of the VIP1 and VIP2 studies, the largest prospective studies of elderly ICU patients, found that nonagenarians had a longer length of stay in the ICU (84h versus 54h, p<0.001) compared to octogenarians, with no dose dependent difference in ICU mortality based on age (3,6,7). This suggests that even patients over the age of 90 may have reasonable post-ICU survival rates (8), although these findings could be confounded by patient selection based on decisions surrounding goals of care or triage decisions for admission to ICU.

In addition to age, frailty has gained attention as an important prognostic predictor in critical illness. Frailty can be defined as the “biologic syndrome of decreased reserve and resistance to stressors, resulting from cumulative declines across multiple physiologic systems, and causing vulnerability to adverse outcomes” (9, pp. 146]. The Clinical Frailty Scale (CFS) is a widely used assessment tool, which for critically ill older adults often relies on collateral information from caregivers for a patient’s premorbid functional status (10–12). Increased frailty is associated with adverse outcomes among ICU patients (13). A prospective observation study of 3920 ICU patients demonstrated that each 1-point increase on the CFS was associated with an increased risk of death (hazard ratio 1.10, 95% confidence interval [CI] 1.05 to 1.15) (6). Furthermore, given the current understanding of the implications of frailty, recent studies which demonstrate increasing ICU mortality with older age but do not evaluate frailty as a realistic confounder (14,15) risk overlooking a potentially modifiable risk.

Finally, multimorbidity is an emerging concept in the care of critically ill older adult patient, appreciated as “a set of mostly predictable clusters” rather than a co-occurrence of diseases (16, pp. 3). Multimorbidity features prominently in some definitions of frailty – notably the cumulative deficit model (17) – and lends itself to feature prominently in discussions and outcome assessments for critically ill patients.

To date, age, frailty and multimorbidity have been used primarily to inform prognosis among those older adults with critical illness. However, it remains uncertain whether these factors may also predict response to ICU interventions, especially in the research setting, as very few trials have specifically examined these populations. To address this, we conducted a systematic review of randomized controlled trials (RCTs) published in high impact journals, to investigate the prevalence of subgroup effect modification – statistically significant differences in relative effect – in older patients, those with increased frailty, and/or those with multimorbidity. We hypothesized that although older, frail and multimorbid patients may have a higher baseline risk for poor outcomes, there was unlikely to be differences in the relative effects of ICU interventions amongst these RCTs.

Methods

We report this systematic review in accordance with the Preferred Reporting Items of Systematic Reviews and Meta-Analyses (PRISMA) statement (18). We registered the protocol in PROSPERO (CRD42021283625) on 12/14/2021.

Search strategy

We conducted a systematic search of leading general medicine (Journal of the American Medical Association [JAMA], JAMA Internal Medicine [JAMA-IM], New England Journal of Medicine [NEJM], Lancet, Annals of Internal Medicine, and the Canadian Medical Association Journal [CMAJ]) and critical care journals (Critical Care Medicine, Intensive Care Medicine, American Journal of Respiratory and Critical Care Medicine, CHEST, Lancet Respiratory Medicine) for RCTs published between January 1, 2011 and December 31, 2021. To maximize feasibility, we selected these dates as there were comparatively fewer reports of frailty in the critically ill prior to 2011 (19), with journals felt to be inclusive of the most practicing-changing research.

An experienced health sciences librarian assisted in developing the search strategy. Please see the supplementary for full search strategy (see Supplementary Appendix). We hand-searched the references of each included study for additional eligible RCTs.

Eligibility criteria

We included RCTs conducted in any adult patient population (>18 years of age) admitted to a critical care unit, published in the journals and timeframe listed above, that reported outcomes based on any one of three subgroups of interest: Older adults versus younger adults, frailty status, and multimorbidity (each as defined by the study authors).

We included parallel design RCTs with individual patient level randomization which examined any intervention against any comparator (usual care, active treatment, or placebo) in any critical care population including medical, surgical, or cardiac surgery patients. We excluded non-randomized studies, cluster RCTs, studies which did not report interventions implemented in an intensive care unit, and RCTs which did not report data examining subgroups of age, frailty, or multimorbidity.

Study selection

After removing duplicates, two authors (AP, MA) screened studies independently and in duplicate first by title and abstract, and then in full-text using Covidence software (20). In the first stage, any article identified as potentially relevant by either reviewer was advanced to stage two. In stage two, we resolved any disagreement by discussion or a third reviewer (OG) if consensus could not be reached between the primary reviewers.

Data extraction

From the included studies, two reviewers collected data independently and in duplicate (SS, RK). A third research member (MA) reviewed all abstracted data for consistency, resolving discrepancies when able, with remaining disagreements resolved by group consensus. We collected the following data: study characteristics (single vs. multi-country enrollment, single vs. multicenter enrollment, funding, sample size, intervention, comparator, primary outcome and secondary outcomes, length of follow-up), study population (inclusion and exclusion criteria, average age in intervention and comparator groups), and subgroup data (outcome of interest, statistical data, and p-values for age, frailty, and multimorbidity, as applicable).

Subgroup analysis

We assessed pre-defined subgroups of age, frailty, and multimorbidity, as defined by the respective study authors. We describe results narratively. We analyzed p-values and confidence intervals of subgroup results, and we defined a statistically significant relative subgroup effect as an interaction p-value of <0.05. For any statistically significant subgroup effects identified, we used the ICEMAN (Instrument for assessing the Credibility of Effect Modification Analyses) to evaluate the credibility of subgroup findings (21), as determined by consensus amongst review authors.

Results

Selection process

Of 2037 unique citations, we reviewed 416 full texts and included 48 eligible RCTs, comprising 50,779 total participants. See Figure 1 for the PRISMA flow diagram.

Figure 1:

Hypothetical effect on 30-day mortality if older adults respond differently to ICU interventions.

Characteristics of included studies

Sample sizes ranged from 62 to 5243, with a median of 468 study participants. 85.4% of studies were multicenter (n=41) and 38.5% of total participants were female. The median age of participants across RCTs was 62 years (range: 43 to 75). The 48 RCTs examined 23 different ICU interventions including drug therapies (n=23, 47.9%), temperature management (n=4, 8%), invasive or non-invasive ventilation interventions (n=3, 6%), and nutritional interventions (n=3, 6%). Median length of follow-up was 3 months. See Table 1 for general characteristics of included studies, and Table 2 for summary of papers with predefined age subgroups, intervention, and primary outcome.

Table 1:

Overview of Study Characteristics

Study Characteristic		Number (Percent) of Trials
Year	2011	1 (2%)
	2012	2 (4%)
	2013	4 (8%)
	2014	8 (17%)
	2015	3 (6%)
	2016	5 (10%)
	2017	3 (6%)
	2018	5 (10%)
	2019	7 (15%)
	2020	4 (8%)
	2021	6 (13%)
Journal	AJRCCM	2 (4%)
	Chest	2 (4%)
	Critical Care Medicine	9 (19%)
	Intensive Care Medicine	5 (10%)
	JAMA	13 (27%)
	NEJM	14 (29%)
	The Lancet	3 (6%)
Study Center	Single Center	6 (13%)
	Multicenter	42 (88%)
Funding *	Industry	6 (13%)
	Government	25 (52%)
	Institution (e.g. University)	6 (13%)
	Hospital	0 (0%)
	None	2 (4%)
	Other	8 (17%)
Intervention	Drug	23 (48%)
	Non-drug**	25 (52%)
Comparator	Usual Care	26 (54%)
	Active Comparator	9 (19%)
	Placebo	13 (27%)
Subgroups	Outcomes Analyzed in Age Subgroup	28 days all-cause mortality, Absence of functional impairment at 90 days, Number of Ventilator Free Days, Physical Functional Performance-10 longitudinally at various time points
	Outcomes Analyzed in Frailty Subgroup	Ventilator-Associated Pneumonia
	Outcomes Analyzed in Multi-morbidity Subgroup	28-day All-Cause Mortality, Physical Functional Performance-10

^*Not reported for Igonin 2012.

^**Non drug interventions include: Body temperature targets, blood pressure target, Intensive Physiotherapy, Parenteral Nutrition

AJRCCM: American Journal of Respiratory and Critical Care Medicine. JAMA: Journal of American Medical Association. NEJM: New England Journal of Medicine.

Table 2:

Summary of papers positive subgroup effect

First Author	Sample Size	Average Age (Intervention)	Average Age (Comparator)	Intervention	Primary Outcome	Subgroup	P-value
Igonin	62	52	50	C1-esterase inhibitor	28-day mortality, systemic inflammatory response syndrome criteria (1992 definition)	>65, <65	>65=0.05 <65=0.09
Lamontagne	118	63	66	High (75–80 mmHg) vs low MAP (60–65mmHg) target	Feasibility outcome: between-group difference in MAP during vasopressor therapy	<75, ≥75	0.015
Legriel	268	57	57	Hypothermia (32 to 34°C for 24 hours)	Good functional outcome at 90 days, defined as a Glasgow Outcome Scale (GOS) score of 5	</=65, >65	0.02
Lemkes	538	65.7	64.9	Immediate Angiography	Survival at 90 days	<70, ≥70	0.007
Mazer	5243	72	72	Restrictive (<7.5 g/dL) vs liberal (<9.5 g/dL intraoperatively or in the ICU postoperatively OR <8.5 g/dL if on non-ICU ward) red cell transfusion protocol	Death from any cause, myocardial infarction, stroke, or new onset renal failure with dialysis occurring within 28d of surgery	2017 study: <75, ≥75	0.004
					Death from any cause, myocardial infarction, stroke, or new onset renal failure with dialysis occurring within 6 months of surgery.	2018 study: <45, 45–54, 55–64, 65–74, 75–84, ≥85
Vincent	309	62.9	61.6	Talactoferrin	28-day mortality	</=65, >65	0.04

All included studies reported subgroup data for age which was variably defined, two studies reported subgroup data for multimorbidity (using the Charlson Comorbidity Index [22] or a predetermined list of comorbidities [23]) and one study reported subgroup data for frailty (using the Clinical Frailty Scale) (24). Of the eligible studies, 71% (n=34) dichotomized age into old and non-old subgroups, with the most common dichotomization being age <65 and ≥65 (50%). Four of the studies divided age subgroups into deciles; the remaining studies used other age subgroup classifications.

Seven (14.6%) RCTs found evidence of statistically significant effect modification based on age, while none of the multimorbidity or frailty subgroups found evidence of statistically significant effect modification. Please see the Supplementary Appendix for credibility of subgroup analyses assessment using the ICEMAN tool.

Statistically significant relative subgroup effects

We identified 4 RCTs in which interventions were more effective in older adults compared to younger adults including restrictive blood transfusion in cardiovascular surgery patients (25,26) (moderate credibility), immediate angiography in out-of-hospital cardiac arrest patients without ST elevation (27) (low credibility), lower blood pressure targets for vasopressor therapy in shock (28) (low credibility), and C1-esterase inhibitor use in sepsis (29) (very low credibility). We identified 2 RCTs in which interventions were less effective in older adults compared to younger adults including hypothermia in status epilepticus (30) (low credibility) and talactoferrin use in severe sepsis (31) (very low credibility).

Restrictive vs. liberal blood transfusion in cardiovascular surgery patients (25,26)

The effect of restrictive vs. liberal transfusion targets in patients undergoing cardiac surgery was investigated in a multicenter, multinational noninferiority RCT, published in 2017. In this study, 5243 adults undergoing cardiac surgery were assigned to receive red-cell transfusions at either restrictive (<7.5 g/dL) or liberal (<9.5 g/dL) thresholds. The age-based subgroup analysis (stratified by decades) demonstrated the restrictive transfusion threshold was associated with a lower risk of the composite outcome (death, myocardial infarction, stroke, and acute renal failure) among patients aged 75 years and older (interaction p-value = 0.004). We judged this subgroup finding to be of moderate credibility. A follow-up study examining the same outcomes at 6 months follow-up demonstrated a consistent subgroup effect at this later endpoint (26).

Immediate vs. delayed angiography in out-of-hospital cardiac arrest patients without ST elevation (27)

The use of immediate vs. delayed angiography in patients who presented with an out-of-hospital cardiac arrest without ST elevation was investigated in a multicenter RCT published in 2019. In this study, 552 patients were randomly assigned to angiography at arrival, or delayed angiography. Subgroup analysis demonstrated improved 90-day survival with the delayed angiography approach in those <70 years of age (interaction p=0.007). We judged this subgroup finding to be of low credibility.

Higher vs. lower blood pressure targets for vasopressor therapy in shock (28)

The feasibility of targeting higher vs. lower blood pressure targets in vasopressor-dependent shock was assessed in a multicenter, multinational pilot RCT published in 2016. In this study, 118 patients in vasodilatory shock were randomized to a lower (60–65 mmHg) versus a higher (75–80 mmHg) MAP target. Patients aged 75 years or older had lower hospital mortality (interaction p=0.03) with the lower MAP target as compared to younger patients. We judged this subgroup finding to be of low credibility.

C1-esterase inhibitor use in sepsis (29)

The use of C1-esterase inhibitor therapy in patients with sepsis was explored in a multicenter RCT, published in 2012. In this study, 61 patients were randomized to receive either 12,000 units of C1-esterase inhibitor or conventional treatment only. Age-based subgroup analysis demonstrated improved survival in patients >65 years of age with C1-esterase inhibitor therapy compared to younger patients (interaction p=0.02). We judged this subgroup finding to be of very low credibility.

Induced hypothermia in status epilepticus (30)

The effect of induced hypothermia in patients with status epilepticus was investigated in a multicenter RCT, published in 2016. In this study, 270 patients with convulsive status epilepticus requiring mechanical ventilation were randomized to hypothermia (32 to 34°C for 24 hours) or standard care within 8 hours of ICU admission. Patients ≤65 demonstrated worse neurological outcomes with hypothermia compared to patients ≤65 years of age (interaction p=0.02). We judged this subgroup finding to be of low credibility.

Talactoferrin use in severe sepsis (31)

The use of talactoferrin alfa, a recombinant form of human lactoferrin, was evaluated in a phase II/III multicenter, multinational RCT published in 2015. In this study, patients with a confirmed or suspected infection, presence of ≥3 components of the SIRS syndrome (32), and acute organ dysfunction were randomized to receive either talactoferrin 1.5g (15mL) or placebo three times a day. The study was stopped early after 305 patients had been enrolled due to concerns of safety and futility in the talactoferrin group. The age-based subgroup findings demonstrated patients aged 65 years and older had higher mortality compared to those that were younger (interaction p=0.04). We judged this subgroup finding to be of very low credibility.

Discussion

This systematic review set out to determine whether patients of older age, higher frailty status, or increased comorbidities respond differently to ICU interventions. While the absolute effect may differ due to higher baseline risk, there exists uncertainty whether these populations experience a greater or lesser relative benefit with commonly utilized ICU interventions. This is an important question, as differences in relative effects could meaningfully impact the balance between harms and benefits (see Figure 1).

Of the 48 studies included in this review, only 7 (14.6%) found differences in the relative effect of interventions between age groups. Of these, only one study – the TRICS III trial – demonstrated a subgroup effect that was judged to be of moderate credibility. In this trial, patients >75 years of age had a lower risk of the composite outcome (death, MI, stroke, or acute renal failure requiring renal replacement therapy) at 28 days when provided with blood transfusions at restrictive cutoffs (hemoglobin <7.5) as compared to a more liberal cutoff. The odds ratio comparing the different hemoglobin thresholds was 0.70 (95% CI 0.54 – 0.89) for older adults as compared to 1.17 (95% CI 0.91–1.50) in younger patients (p-value for interaction = 0.004). Other statistically significant subgroup effects were judged to be of low or very low credibility. Reasons for downgrading credibility included a lack of hypothesis, a lack of prior evidence in support of the subgroup finding, and a weakly significant interaction p-value. Some reasons are not surprising – the first two likely reflect an absence of research overall in so far as guiding a priori hypotheses on the effect of ICU interventions on understudied subgroups, while a weakly significant p-value for ‘positive studies’ (defined as an interaction p-value ≤0.05 and >0.01) reflects small subgroup sizes.

Given this, the take-away from this review – and the available data – is that if an intervention or care strategy shows benefit in a more generalized critical care population, the same relative benefit may apply to older patients. Obviously, this conclusion does not consider goals of care, patient values and preferences, and shared decision-making that remain crucial to consider in these settings. An important limitation of this finding is that subgroups are, by definition, underpowered, and therefore absence of subgroup effect does not definitively rule out important effect modification. Furthermore, subgroup definitions based on age are essentially arbitrary, and there remains little effective clinical utility of ‘over-under’ age cut-offs for drawing conclusions. Age subgroups divided into meaningful ranges (e.g. 8 of 48 studies divided age over 65 into 3 or more ranges) or as a continuous variable (which requires sufficient subgroup numbers) may allow for more meaningful interpretations. Finally, the median age in the included trials ranged between 43 to 75 years of age. Therefore, it is possible that within the “oldest” ICU patients – typically defined as those greater than 80 years of age – further signals for effect modification may exist but were not able to be evaluated. Finally, an important consideration is that age cut-offs for subgroup definitions in the included studies were largely arbitrary.

There is also a growing understanding that frailty – rather than age – is the more important factor to consider in predicting response to ICU interventions. Frailty is a state of decreased physiologic, functional, or cognitive reserve resulting in increased vulnerability to acute health stressors (9,33) and holds a more biologically plausible explanation for differences in intervention effects than does age. Although frailty increases with age, it is an independent prognostic factor that can also occur in younger patients. Frailty is present in roughly 30% of older ICU patients (13) and has recently been shown to be a more significant risk factor for poor ICU outcomes than age, illness severity, or comorbidity (34).

Examining the eligibility criteria of the included studies, many intentionally excluded patients that would have likely been of high frailty status – a common practice in intervention-focused RCTs. This review identified only one RCT published in a leading journal in the past 10 years that examined subgroups based on frailty, with no evidence of effect modification. If frailty is as important as we suspect, the older adults included in many of these RCTs would be considered non-frail, and it would therefore be plausible that they respond to ICU interventions similarly to younger patients. On the contrary, the ‘positive study’ results may be confounded by frailty, which is more prevalent in older populations (explaining why the “older” age group performed differently) but not present in all older adults (explaining why age should not be the characteristic of interest). Without having performed subgroup analysis based on frailty, it is uncertain whether the differences in relative effect observed within the older age subgroups were due to age or confounded by frailty. This uncertainty demonstrates why it is imperative that future RCTs in ICU patients include frailty as a subgroup of interest.

Strengths of this review include a comprehensive search facilitated by an experienced health science librarian, a pre-registered protocol, dual and independent screening and data abstraction, and careful assessment of subgroup credibility using the ICEMAN tool. The review also examined important factors beyond age, including frailty and multimorbidity – constructs that are gaining attention and may guide future research direction.

Limitations of this review include ongoing imprecision of individual studies in evaluating subgroups which does not absolutely rule out effect modification, even if present. While all included studies examined age subgroups, there was only a single RCT examining frailty and two evaluating multimorbidity. Furthermore, the search was limited to selected high impact journals, although this was intentional to focus on practice-changing studies and to increase pragmatism.

Conclusion

Most critical care RCTs do not examine for subgroup effects by frailty or multimorbidity. While age is more commonly considered, the cut-point is variable and relative effect modification is rare. As our understanding of frailty and its impact increases, it should become commonplace for RCTs to report subgroups examining frailty as commonly as they currently examine age.

Key Points.

Question:

What do we know about how patients of older age, higher frailty status, or increased comorbidities respond to current evidence-based critical care interventions?

Findings:

The vast majority of critical care randomized controlled trials do not examine for subgroup effects by frailty or multimorbidity. Age subgroups – when provided – have variable cut-points and relative effect modification based on age is rare.

Meaning:

Shared decision-making based on individual patient preferences, and trials focused specifically on patients of older age, higher frailty status, or increased comorbidities, remain a priority.