Community-based complex interventions to sustain independence in older people, stratified by frailty: a systematic review and network meta-analysis

Databases and trial registers

We searched the following databases from inception between 9 and 11 August 2021:

• Cochrane Central Register of Controlled Trials (CENTRAL) Wiley (1992 to 11 August 2021)

• MEDLINE Ovid (1946 to 6 August 2021)

• Embase and Embase Classic Ovid (1947 to 6 August 2021)

• CINAHL EBSCOhost (1972 to 9 August 2021)

• APA PsycINFO Ovid (1806 to August Week 1, 2021).

We also searched trial registers from inception on 10 August 2021:

• US National Institutes of Health Ongoing Trials Register ClinicalTrials.gov (www.clinicaltrials.gov; to 10 August 2021);

• World Health Organization International Clinical Trials Registry Platform (https://trialsearch.who.int; to 10 August 2021).

Other sources

We searched for additional reports of all included studies in the reference lists of their reports, by searching Google Scholar (scholar.google.co.uk) with the intervention name, project name and trial register number separately, and, if available, on the study’s website. We scanned the reference lists of included studies to identify potentially eligible trials not already identified (backward citation searching). We exhaustively continued backward citation searching and identification of additional reports, including for all reports and studies included in the review through these same processes.

Types of studies

Only RCTs and cRCTs were eligible. Where only one unit of randomisation (an individual or cluster) was allocated to an arm of a trial, we excluded the trial as the treatment effect is completely confounded with the unit. We accepted minimisation as a method of sequence generation, in keeping with Cochrane risk-of-bias (RoB) guidance. 23 We did not exclude variants of randomised trials such as stepped-wedge trials. Crossover and waiting-list designs were also eligible, but we only used outcome data from the pre-crossover period because it is likely that the older people’s independence evolves over time and that, were interventions to be effective at modifying outcomes such as activities of daily living (ADL), this would be a long-term modification whose effects may carry over into the subsequent period of the trial.

Types of participants

We included studies involving older people living at home (mean age of study participants 65 years or older). We excluded trials in residential/nursing homes. If not all participants were living at home, we only included the study if data could be extracted specifically for these participants.

Types of intervention

Aligned with our focus on community-based complex interventions, trials were considered eligible if:

• the intervention was both initiated and mainly provided in the community

• the intervention included two or more interacting components (intervention practices, structural elements and contextual factors)

• the intervention was targeted at the individual person, with provision of appropriate specialist care

• a focus of the intervention was sustaining (maintaining or improving) the person’s independence.

A broad range of interventions was eligible, differing in terms of how the service was organised and what was done to or for the older person. Our criterion of including two or more interacting components could be met in multiple ways. Eligible interventions could include multiple discrete practices, such as exercise sessions and nutritional advice. Alternatively, interventions could include one practice that interacted with other structural elements such as being reliant on general practice or other services; or interaction with contextual factors by being substantially tailored to the person’s physical and social environment such as ‘comprehensive geriatric assessment’24 or rehabilitation interventions.

Interventions that were not eligible for inclusion were those where:

• the intervention was either not initiated, or not mainly provided, in the community, or neither for example interventions delivered in outpatient, day hospital, inpatient and intermediate (post-acute) care settings

• the intervention included only one component (intervention practices, structural elements and contextual factors), for example, if any of the following were delivered as single component interventions: a drug; treadmill training; yoga; provision of information; cataract surgery for visual impairment; hearing aid for hearing impairment; medication-review; nutritional supplements

• the intervention was not targeted at the individual person, with provision of appropriate specialist care for example general staff education (not training in a patient-level intervention), practice-level reorganisation, operational, managerial or information technology (IT) interventions, public health messages

• the intervention was not focused on sustaining (maintaining or improving) independence; for example, we excluded interventions that primarily addressed cognitive deficits, mood disorders or both, unless they also aimed to improve overall independence

• disease-focused case management of older people with specific long-term conditions; for example, diabetes, chronic obstructive pulmonary disease or depression

• interventions in which the primary focus was fall prevention were excluded as the evidence base for fall prevention is well established including via NMA. 25 Nonetheless, falls were an additional outcome of this review.

Initially we intended to only include interventions for which sustaining independence was the main aim, but this was broadened as complex interventions were rarely described as having one main aim.

Comparators

Usual care, ‘placebo’ or attention control or a different complex intervention which met our criteria were eligible comparators.

Outcomes

Studies were only included where outcome data were measured at a minimum 24-week (approximately 6 months) time point. We included studies that met the above criteria whether or not they measured or reported our outcomes of interest (see below).

Main outcomes

• Living at home.

• Activities of daily living (personal/instrumental).

• Hospitalisation.

• Care-home placement.

• Homecare services (non-healthcare professional) usage.

• Costs.

• Cost-effectiveness.

Additional outcomes

• Health status/health-related quality of life.

• Depression.

• Loneliness.

• Falls.

• Mortality.

We worked with our PPI FOG to identify key outcomes, and prioritise them as ‘main’ or ‘additional’, from the perspective of older people. Originally (prior to analyses beginning) these outcomes were proposed as one primary outcome (living at home) with the others as secondary outcomes.

We sought binary data for living at home, homecare services usage, hospitalisation, care-home placement, falls and mortality; and continuous data for activities of daily living, homecare services usage, hospitalisation, costs, cost-effectiveness, health status, depression, loneliness and falls. We excluded bespoke metrics without evidence of evaluation of measurement properties, or metrics where significant problems with their use are recognised.

Specifically, for binary outcomes, we sought to extract the total number of participants in each intervention and control group, and the number of outcome events, alongside any reported effect measures and their 95% confidence intervals (CIs), such as risk ratios (RRs) or odds ratios (ORs). For continuous outcomes, we sought to extract the total number of participants in each intervention and control group, the mean and standard deviation (SD) of the outcome in each group at baseline and at follow-up, and the mean and SD of the change score for each group. These estimates were extracted alongside any reported effect measures, such as the mean difference (MD) in final score or change score, or the ANCOVA (adjusted for baseline) result and a measure of variance such as the standard error. Where the above estimates were unavailable within publications, we attempted to recover these using alternate available information so as to avoid selection bias (e.g. using reported elements of the five-number summary, CIs, p-values, 2 × 2 contingency tables etc., to recover mean (SD) or ORs/RRs).

Where living at home was not a reported study outcome but care-home placement and mortality were, we calculated living at home as the remainder (not dead or living in a care home) where it was possible to do so without double-counting participants. Where care-home placement results related to all participants including those who had subsequently died, we could not disaggregate those who had died while living at home and so could not use these results.

Care-home placement as a standalone outcome was only included where it related to care-home residence at that time point, that is the figures excluded mortality during the period of reporting.

Time points

For all outcomes of interest, data were extracted and categorised for three time frames shown in Table 1.

Where more than one time point was reported for an outcome within a range specified above, we used the time point nearest to the target time point.

Because of the large number of outcomes and multiple time frames, we limited analyses of additional outcomes to the medium-term time frame only. We anticipated that the medium term would be of particular interest to commissioners and older people, would be most likely to allow sufficient time for effects to be realised but not washed out, and when most data would be available. We planned to conduct sensitivity analyses and meta regression for the medium-term time frame initially and to extend these to short- and long-term networks in the presence of significant findings.

Throughout the rest of the report we refer to results of interest, which we define here as reported data about a comparison between two arms of a trial for an outcome of interest in one of our analysis time frames.

Age

The pooled mean age of participants was 77.3 (SD 7.5) years, reported in 97 studies. Four studies had study populations aged 85 years or over on average (mean or median);74,95,102,113 only one study’s population was < 70 years on average (mean 68.4 years). 66 Most of the studies explicitly targeted older adults, except three where all adults were eligible. 66,71,174 A minimum age was either reported or an eligibility criterion in all but 8 studies: under 65 years in 20 studies (16–18 years: n = 3; 50 years: n = 3; 60 years: n = 14), 65–79 years in 90 studies (65–69 years: n = 46; 70–74 years: n = 20; 75–79 years: n = 24), 80 years or over in 5 studies,74,85,96,102,113 and 6 studies had criteria where minimum age differed in conjunction with ethnicity,116,145–147 medical condition143 or both. 130

Frailty

One-hundred and eight study populations were classified for frailty (21 were insufficiently described to be classified64,71,72,77,80,82,83,86,100,106,110,111,115,124,126,128,140,142,143,156,174): 31 included all frailty levels (robust, pre-frail and frail), 3 populations were robust, 5 were robust and pre-frail, 8 were pre-frail, 25 were pre-frail and frail and 36 were frail. The studies that had populations classified as ‘all’ typically had broad inclusion criteria, although some purposefully sampled multiple risk or frailty groups, and others were selective on health-related criteria but in a way that would include all frailty categories. Ninety-one populations were classified on the basis of characteristics and criteria, while 17 were classified on the basis of a validated measure, of which 2 were classified as all,102,176 4 as pre-frail,89,96,157,186 7 as pre-frail and frail69,121,122,135,144,153,180 and 4 as frail. 75,136,170,179

Components

We identified 19 separate components that were intended to be delivered to all participants among the 266 interventions. With our PPI group we developed public-facing names and plain language descriptions for these components, and organised them into domains (see Appendix 2). Fourteen components were ‘action’ components (e.g. therapy, support, education) and the remaining five components involved some kind of ascertainment or assessment process with the potential to lead to multiple actions.

The 14 action components, their short (in bold) and abbreviated labels were:

• formal homecare (hmcr, identified in 41 interventions)

• physical exercise (exrc, 30 interventions)

• health education (educ, 26 interventions)

• ADL training (ADL, 13 interventions)

• providing aids and adaptations (aids, 12 interventions)

• nutritional support (ntr, 10 interventions)

• psychological (mood) therapy (psychology, psyc, 4 interventions)

• technology for communication and engagement (telecoms, comm, 4 interventions)

• cognitive training (cgn, 3 interventions)

• engagement in meaningful-activities (eng, 3 interventions)

• care voucher provision (vchr, 2 interventions)

• alternative medicine such as homeopathy and naturopathy (hmnt, 1 intervention)

• social skills training (sst, 1 intervention) and

• welfare rights advice (wlfr, 1 intervention).

Homecare involved frequent visits at home by health or care professionals to provide services including support with household tasks, self-care and nursing care. Formal homecare was a component of usual care intended to be received by all participants in 40 of the interventions and any other components typically interacted, or were integrated, with it. In the remaining intervention, formal homecare was part of the programme of ‘early’ interventive social services for older people moderately at risk of becoming unable to remain in their own homes. 158

The five components involving a process of ascertainment or assessment were:

• multifactorial-action from care planning (mfa, 117 interventions)

• routine follow-up review (following multifactorial-action from care planning; mfar, 82 interventions)

• medication-review (med, 4 interventions)

• monitoring (mntr-mfa-, 2 interventions) and

• routine risk-screening (rsk-mfa-; 7 interventions).

Multifactorial-action from care planning is the term we gave to a process of individualised multidomain assessment and management intended to lead to subsequent action, where the components of action are tailored to the individual. This did not apply to interventions where the subsequent action was intended to relate to one component only.

Routine review is a process of scheduled, regular follow-ups that follow on from multifactorial-action. This does not include additional contacts that were ad hoc or determined by need. We labelled the remaining 35 interventions with mfa- to denote the absence of routine review (although not necessarily further contact).

Because multifactorial-action encompassed a broad range of approaches and differing patterns of follow-on action, we sought meaningful and identifiable criteria by which to further delineate it. Based on discussion with experts, we identified medication-review (w/med; 54 interventions), and specific psychological strategies to support behaviour change and self-management (w/slfm; 29 interventions) in the assessment and care planning of multifactorial-action, including their combination and absence. The specific self-management strategies included reframing, motivational interviewing, problem-solving, goal setting and action planning,592 but not practical support with adherence (e.g. encouragement phone calls, prompts and reminders), or general comments about encouragement of behaviour change or self-management. This helped to characterise approaches to multifactorial-action that incorporated medical and psychosocial orientations.

We treated medication-review as a component in its own right when present in an intervention that did not include multifactorial-action.

Monitoring and routine risk-screening are two components where participants only receive multifactorial-action from care planning when they meet a particular trigger or threshold. We did not treat risk-screening as an intervention component where it was used solely to identify those eligible for the trial.

Available care was the label we applied to interventions (or conditions) where there was no particular component intended to be delivered to all participants. This acknowledged the availability of a wide range of primary, secondary, tertiary and wider community services that would be accessed by some, but not all, participants without specifying their nature. We also included here actions that were not intended or anticipated to affect an individual’s independence such as attention control, placebo and other minor actions such as giving a leaflet. Additionally, we did not identify trial-specific procedures related with data collection, or safeguarding such as suicide monitoring, as separate components.

Intervention groups

Our intervention groups were formed from the 63 combinations of these components among the 266 interventions, as detailed in Table 4. The largest group is ac (comprised of 98 interventions). Other groups including more than 10 interventions are multifactorial-action and review with medication-review (24 interventions), multifactorial-action and review (15 interventions) and homecare (12 interventions). Eight groups contained two interventions while 37 groups were formed of one intervention only. We have provided TIDieR summaries of the 26 intervention groups with more than 1 intervention in Report Supplementary Material 4.

Usual care, the standard care provided in that context in the absence of a trial, was often ac or homecare but in some contexts it was multifactorial-action; multifactorial-action and review; aids; or another intervention. Not all trials included a usual care comparator.

Comparisons

The 144 within-trial comparisons between eligible arms formed 80 types of direct comparison (edges or links) between the intervention groups (nodes or points; illustrated in Figure 2). There were 20 comparisons of multifactorial-action and review with medication-review versus ac, 8 comparisons of multifactorial-action and review versus ac and 6 comparisons of risk-screening versus ac. Sixty-one edges were formed of a single within-trial comparison.

FIGURE 2.

Network of 63 intervention groups (nodes) and the 80 types of direct comparison (edges) included in the review (all included studies regardless of availability of results data). ADL, activities of daily living training; aids, provision of aids and adaptions; cgn, cognitive training; comm, technology for communication and engagement; educ, education; eng, engagement in meaningful-activities; exrc, exercise; hmcr, homecare; hmnt, alternative medicine such as homeopathy and naturopathy; med, medication-review; mfa, multifactorial-action; mfar, multifactorial-action routine review follow on; mntr-mfa, monitoring, which may trigger multifactorial-action; ntr, nutritional support; psyc, psychological therapy; rsk-mfa, risk-screening, which may trigger multifactorial-action; sst, social skills training; vchr, care voucher provision; wlfr, welfare rights advice; w/med, with medication-review; w/slfm, with self-management.

Most of the edges had ac as one of their nodes (n = 45). Nine of the edges had formal homecare as one of their nodes. Forty-three nodes were only connected by one edge to the rest of the network, 34 of which were only connected by a single trial’s comparison.

The nodes and edges formed one connected network in principle. However, single studies linked different parts of the network, thus lack of data meant the network was divided in many analyses into two networks: one network of comparisons with ac and one network of comparisons with homecare (hmcr). Additionally, among the nodes with homecare as a component, there were multiple comparisons on the path between the homecare node and the homecare, multifactorial-action and review node, which was connected to ac. This meant there were some analyses where some nodes that included homecare as a component were connected to the ac network but still disconnected from the homecare network.

Nine comparisons formed self-loops, where we had categorised two or more arms of a trial into the same intervention group. These comparisons, therefore, do not form part of the network for meta-analysis. Three of these were in three-arm trials, with a pair of comparisons that still contributed to the analysis. 69,106,171 Morey 2006139 was a three-arm trial that did not report any results of interest where we categorised all three interventions in the same group; all three arms featured a minimal but active intervention, one was a high-intensity alternative and one added an attention control element. The remaining three self-comparisons all included multifactorial-action as a component. Kono 2012120 compared two forms of multifactorial-action and review. Stewart 2005161 compared multifactorial-action from care planning performed by either a social worker or an occupational therapist. Ryvicker 2011156 compared two forms of homecare, multifactorial-action and review.

Individually randomised studies

Cluster-randomised studies

Risk of bias in results of interest

Risks of bias arising in the other domains varied according to the outcome assessed. For missing outcome data, differences particularly related to whether the outcome was continuous or dichotomous and the proportion of people experiencing the event, with rare, dichotomous outcomes more likely to be at higher risk. For bias in measurement of the outcome, differences largely related to whether the outcome was self-reported or sourced from records and whether the self-reporting was about the individual’s perception (such as depressive symptoms) or memorable, observable events (such as hospitalisation). For bias in selection of the reported result, we very rarely had access to a sufficiently detailed analysis plan, so differences largely related to whether we had access to numbers of events and cases where there were no plausible alternative definitions for the measure such that the same data could be recut into different groups such as mortality as a defined outcome.

Overall risk of bias

Overall, no results of interest were judged to have low RoB; there were some concerns about 28%, with the remaining 72% at high RoB. We further judged those results at high RoB to present either serious concerns (53% of results) or very serious concerns (19% of results). Because there were differences in RoB by outcome, some outcomes were reported by more studies, and different studies report different outcomes, we have not presented a more detailed breakdown here. The RoBs are summarised for each analysis in the text and details are provided in Report Supplementary Material 10. Results were included in analyses regardless of RoB, but we conducted sensitivity analyses excluding results for which we had very serious concerns. We judged whether to downgrade our certainty in the evidence for an effect estimate based on the contributions of results judged as serious concerns or very serious concerns about RoB.

Short-term time frame

The available-care network for living at home in the short term included eight studies (n = 4013) with eight intervention groups (see Table 5 for a summary of findings). 78,113,121,126,136,165,166,190 Each comparison included one study. Two study populations included all frailty categories, two were pre-frail and frail, three were frail and one was unclassifiable. We had some concerns regarding RoB in six study results, with two at high RoB (one serious concerns, one very serious concerns).

TABLE 5 - Living at home in the short term: comparisons with ac summary of findings table

Calculated from OR and an assumed comparator risk of 0.972, the median ac risk among these studies.

Very serious concerns about imprecision as CI includes substantial benefit and substantial harm. Downgrade twice.

Very serious concerns about RoB due to significant contamination in both groups, particularly in the intervention arm where serious deviations from the intended interventions happened. Already downgraded twice for imprecision, therefore downgrade once.

Serious concerns about RoB due to randomisation process. Downgrade once.

Extremely serious concerns about imprecision as CI is extremely wide. Downgrade three levels.

Population: Older people Interventions: Community-based complex interventions Comparator: Available care (ac) Outcome: Living at home Time frame Short term; range of follow-up 24 weeks to 6 months Setting: Community Total studies: 8 Total participants: 4013 Comparator rank: Mean 4.7, 95% CI 2 to 7
Intervention group	Relative effect (95% CI)		Anticipated absolute effect (95% CI)				Certainty of the evidence (GRADE)	Ranking (95% CI)	Interpretation
	Network summary estimate	Calculated RRa	High-risk population (952 per 1000 with ac)		Low-risk population (980 per 1000 with ac)
			With intervention	Difference	With intervention	Difference
Multifactorial-action and review (mfar)	OR 1.34 (0.75 to 2.39) Mixed estimate	RR 1.01 (0.99 to 1.02)	964 per 1000 (937 to 979)	12 more per 1000 (15 fewer to 27 more)	985 per 1000 (974 to 992)	5 more per 1000 (6 fewer to 12 more)	⊕⊕⊝⊝Lowb	3.1 (1 to 6)	May result in a very slight increase in chance of living at home
ADL, nutrition and exercise (ADL & ntr & exrc)	OR 1.01 (0.25 to 4.13) Mixed estimate	RR 1.00 (0.92 to 1.02)	953 per 1000 (831 to 988)	1 more per 1000 (121 fewer to 36 more)	980 per 1000 (924 to 995)	0 per 1000 (56 fewer to 15 more)	⊕⊕⊝⊝Lowb	4.5 (1 to 8)	May result in little to no difference in chance of living at home
Multifactorial-action and review with medication-review [mfar(w/med)]	OR 0.95 (0.58 to 1.55) Mixed estimate	RR 1.00 (0.98 to 1.01)	949 per 1000 (920 to 969)	3 fewer per 1000 (32 fewer to 17 more)	979 per 1000 (966 to 987)	1 fewer per 1000 (14 fewer to 7 more)	⊕⊕⊝⊝Lowb	5.0 (2 to 7)	May result in little to no difference in chance of living at home
Education, multifactorial-action and review with medication-review and self-management strategies [educ & mfar(w/med + slfm)]	OR 0.52 (0.13 to 2.06) Mixed estimate	RR 0.98 (0.85 to 1.01)	912 per 1000 (725 to 976)	40 fewer per 1000 (227 fewer to 24 more)	962 per 1000 (867 to 990)	18 fewer per 1000 (113 fewer to 10 more)	⊕⊕⊝⊝Lowb	6.2 (2 to 8)	May result in a reduction in chance of living at home
Aids, multifactorial-action and review (aids & mfar)	OR 3.06 (0.31 to 30.42) Mixed estimate	RR 1.02 (0.94 to 1.03)	984 per 1000 (859 to 998)	32 more per 1000 (93 fewer to 46 more)	993 per 1000 (938 to 999)	13 more per 1000 (42 fewer to 19 more)	⊕⊝⊝⊝Very lowb,c	2.2 (1 to 7)	The evidence is very uncertain about the effect on chance of living at home
Multifactorial-action and review with self-management strategies [mfar(w/slfm)]	OR 1.34 (0.56 to 3.25) Mixed estimate	RR 1.01 (0.98 to 1.02)	964 per 1000 (917 to 985)	12 more per 1000 (35 fewer to 33 more)	985 per 1000 (965 to 994)	5 more per 1000 (15 fewer to 14 more)	⊕⊝⊝⊝Very lowb,d	3.4 (1 to 7)	The evidence is very uncertain about the effect on chance of living at home
ADL, aids, education, exercise, multifactorial-action and review with medication-review and self-management strategies [ADL & aids & ed & ex & mf(w/med + slfm)]	OR 0.18 (0.01 to 3.69) Mixed estimate	RR 0.89 (0.24 to 1.02)	779 per 1000 (145 to 987)	173 fewer per 1000 (807 fewer to 35 more)	897 per 1000 (295 to 994)	83 fewer per 1000 (685 fewer to 14 more)	⊕⊝⊝⊝Very lowe	7.0 (1 to 8)	The evidence is very uncertain about the effect on chance of living at home

The consistency assumption was found to hold for the network according to both the global Wald test (p = 0.238) and the node-splitting method (all contrasts p > 0.05). As the network contained only a single study measuring each comparison, there was no potential source of heterogeneity and therefore a common-effect model was fitted.

There was low-certainty evidence that multifactorial-action and review may result in a very slight increase in the odds of living at home compared with ac in the short term (OR 1.34, 95% CI 0.75 to 2.39). There was low-certainty evidence that ADL, nutrition and exercise (OR 1.01, 95% CI 0.25 to 4.13); and multifactorial-action and review with medication-review (OR 0.95, 95% CI 0.58 to 1.55) may result in little to no difference in odds of living at home. Education, multifactorial-action and review with medication-review and self-management may result in a reduction in odds of living at home [OR 0.52 (moderate), 95% CI 0.13 to 2.06; low certainty]. Three other comparisons with ac were of very low certainty.

Medium-term time frame

The available-care network for living at home in the medium term consisted of 21 studies (n = 16,937) with 14 intervention groups (see Table 6 for a summary of findings). 70,84,88,103,104,106–108,116,118,119,121,136,138,142,143,150,153,158,165,178 Multifactorial-action and review with medication-review versus ac included data from six studies; two comparisons had data from two studies and the remainder had one. Four study populations included all frailty categories: one was robust, one pre-frail, six pre-frail and frail and six frail. There were some concerns with RoB in 12 of the results; the other 12 were high RoB (6 serious concerns, 3 very serious concerns).

TABLE 6 - Living at home in the medium term: comparisons with ac summary of findings table

Calculated from OR and an assumed comparator risk of 0.935, the median ac risk among these studies.

Serious concerns about imprecision as CI crosses the no effect line and includes substantial benefit. The CI for the absolute effect with a high-risk population also includes our pre-specified definition of very small harm but given this was marginal and in light of the small lower CI for the RR (0.9955), we did not judge this as very serious. Downgrade once.

Very serious concerns about imprecision as CI includes substantial benefit and substantial harm. Downgrade twice.

Very serious concerns about RoB due to excluding participants in per-protocol analysis and missing outcome data in the indirect evidence via the homecare, multifactorial-action and review vs. ac comparison. Downgrade twice.

Very serious concerns about imprecision as CI is very wide, there is no closed loop and the direct comparison is based on indirect evidence from 122 persons in homecare, multifactorial-action and review and 81 persons in homecare, multifactorial-action and review with self-management, which does not meet the optimal information size. Downgrade twice.

Very serious concerns about RoB due to excluding participants in per-protocol analysis and missing outcome data. Downgrade twice.

Very serious concerns about imprecision as CI is very wide, and there is no closed loop and the direct comparison is based on indirect evidence from 122 persons in homecare, multifactorial-action and review, which does not meet the optimal information size. Already downgraded twice, downgrade once.

Serious concerns about RoB due to missing outcome data. Downgrade once.

Serious concerns about Rob due to the recruitment process of participants and missing outcome data in one study. Downgrade once.

Very serious concerns about RoB due to randomisation process and missing outcome data. Already downgraded twice, downgrade once.

Population: Older people Interventions: Community-based complex interventions Comparator: Available care (ac) Outcome: Living at home Time: Medium term; range of follow-up 12 to 18 months Setting: Community Total studies: 21 Total participants: 16,937 Comparator rank: Mean 9.9, 95% CI 7 to 12
Intervention group	Relative effect (95% CI)		Anticipated absolute effect (95% CI)				Certainty of the evidence (GRADE)	Ranking (95% CI)	Interpretation
	Network estimate	Calculated RRa	High-risk population [833 per 1000 with ac]		Low-risk population [981 per 1000 with ac]
			With intervention	Difference	With intervention	Difference
Multifactorial-action and review with medication-review [mfar(w/med)]	OR 1.22 (0.93 to 1.59) Mixed estimate	RR 1.01 (1.00 to 1.02)	859 per 1000 (823 to 888)	26 more per 1000 (10 fewer to 55 more)	984 per 1000 (980 to 988)	3 more per 1000 (1 fewer to 7 more)	⊕⊕⊕⊝Moderateb	7.5 (5 to 11)	Probably results in a slight increase in chance of living at home
Multifactorial-action with medication-review [mfa-(w/med)]	OR 2.55 (0.61 to 10.60) Mixed estimate	RR 1.04 (0.96 to 1.06)	927 per 1000 (754 to 981)	94 more per 1000 (79 fewer to 148 more)	992 per 1000 (969 to 998)	11 more per 1000 (12 fewer to 17 more)	⊕⊕⊝⊝Lowc	4.6 (1 to 13)	May result in an increase in chance of living at home
Cognitive training, medication-review, nutrition and exercise (cgn & med & ntr & exrc)	OR 1.93 (0.79 to 4.77) Mixed estimate	RR 1.03 (0.98 to 1.05)	906 per 1000 (797 to 960)	73 more per 1000 (36 fewer to 127 more)	990 per 1000 (976 to 996)	9 more per 1000 (5 fewer to 15 more)	⊕⊕⊝⊝Lowc	5.3 (2 to 12)	May result in an increase in chance of living at home
ADL, nutrition and exercise (ADL & ntr & exrc)	OR 1.79 (0.67 to 4.76) Mixed estimate	RR 1.03 (0.97 to 1.05)	899 per 1000 (771 to 960)	66 more per 1000 (62 fewer to 127 more)	989 per 1000 (972 to 996)	8 more per 1000 (9 fewer to 15 more)	⊕⊕⊝⊝Lowc	5.9 (2 to 13)	May result in an increase in chance of living at home
Risk-screening (rsk-mfa-)	OR 0.90 (0.66 to 1.23) Mixed estimate	RR 0.99 (0.97 to 1.01)	818 per 1000 (768 to 860)	15 fewer per 1000 (65 fewer to 27 more)	979 per 1000 (972 to 984)	2 fewer per 1000 (9 fewer to 3 more)	⊕⊕⊝⊝Lowc	11.1 (7 to 13)	May result in a very slight reduction in chance of living at home
Education, multifactorial-action and review with medication-review [educ & mfar(w/med)]	OR 0.88 (0.60 to 1.29) Mixed estimate	RR 0.99 (0.96 to 1.01)	814 per 1000 (749 to 865)	19 fewer per 1000 (84 fewer to 32 more)	978 per 1000 (969 to 985)	3 fewer per 1000 (12 fewer to 4 more)	⊕⊕⊝⊝Lowc	11.2 (6 to 14)	May result in a very slight reduction in chance of living at home
Education, multifactorial-action and review with medication-review and self-management strategies [educ & mfar(w/med + slfm)]	OR 0.41 (0.14 to 1.17) Mixed estimate	RR 0.91 (0.72 to 1.01)	670 per 1000 (413 to 854)	163 fewer per 1000 (420 fewer to 21 more)	955 per 1000 (879 to 984)	26 fewer per 1000 (102 fewer to 3 more)	⊕⊕⊝⊝Lowc	13.5 (8 to 14)	May result in a reduction in chance of living at home
Homecare, multifactorial-action and review with self-management strategies [hmcr & mfar(w/slfm)]	OR 8.89 (1.90 to 41.62) Indirect estimate	RR 1.06 (1.03 to 1.07)	978 per 1000 (904 to 995)	145 more per 1000 (71 more to 162 more)	998 per 1000 (990 to 1000)	17 more per 1000 (9 more to 19 more)	⊕⊝⊝⊝Very lowd,e	1.5 (1 to 4)	The evidence is very uncertain about the effect on chance of living at home
Homecare, multifactorial-action and review (hmcr & mfar)	OR 5.71 (1.65 to 19.83) Mixed estimate	RR 1.06 (1.03 to 1.07)	966 per 1000 (891 to 990)	133 more per 1000 (58 more to 157 more)	997 per 1000 (988 to 999)	16 more per 1000 (7 more to 18 more)	⊕⊝⊝⊝Very lowf,g	2.4 (1 to 5)	The evidence is very uncertain about the effect on chance of living at home
Multifactorial-action (mfa-)	OR 2.23 (0.63 to 7.91) Mixed estimate	RR 1.04 (0.96 to 1.06)	917 per 1000 (758 to 975)	84 more per 1000 (75 fewer to 142 more)	991 per 1000 (970 to 998)	10 more per 1000 (11 fewer to 17 more)	⊕⊝⊝⊝Very lowc,h	4.9 (1 to 13)	The evidence is very uncertain about the effect on chance of living at home
Multifactorial-action and review (mfar)	OR 1.15 (0.60 to 2.18) Mixed estimate	RR 1.01 (0.96 to 1.04)	851 per 1000 (751 to 916)	18 more per 1000 (82 fewer to 83 more)	983 per 1000 (969 to 991)	2 more per 1000 (12 fewer to 10 more)	⊕⊝⊝⊝Very lowc,i	8.4 (4 to 13)	The evidence is very uncertain about the effect on chance of living at home
Education and risk-screening (educ & rsk-mfa-)	OR 1.09 (0.60 to 2.01) Mixed estimate	RR 1.01 (0.96 to 1.03)	845 per 1000 (748 to 909)	12 more per 1000 (85 fewer to 76 more)	983 per 1000 (968 to 990)	2 more per 1000 (13 fewer to 9 more)	⊕⊝⊝⊝Very lowc,h	8.9 (4 to 13)	The evidence is very uncertain about the effect on chance of living at home
Multifactorial-action with medication-review and self-management strategies [mfa-(w/med + slfm)]	OR 1.00 (0.63 to 1.58) Mixed estimate	RR 1.00 (0.96 to 1.02)	833 per 1000 (759 to 888)	0 per 1000 (74 fewer to 55 more)	981 per 1000 (970 to 988)	0 per 1000 (11 fewer to 7 more)	⊕⊝⊝⊝Very lowc,j	9.8 (5 to 13)	The evidence is very uncertain about the effect on chance of living at home

The consistency assumption was found to hold for the network according to both the global Wald test (p = 0.796) and the node-splitting method (all contrasts p > 0.05). Between-study variance was estimated to be non-zero, indicating some degree of heterogeneity between studies included in the network (τ = 0.0856).

There was moderate-certainty evidence that multifactorial-action and review with medication-review probably results in a slight increase in the odds of living at home compared with ac in the medium term (OR 1.22, 95% CI 0.93 to 1.59; moderate certainty). There was low-certainty evidence that multifactorial-action with medication-review [OR 2.55 (large), 95% CI 0.61 to 10.60]; cognitive training, medication-review, nutrition and exercise [OR 1.93 (large), 95% CI 0.79 to 4.77]; and ADL, nutrition and exercise [OR 1.79 (large), 95% CI 0.67 to 4.76] may result in an increase in the odds of living at home compared with ac. There was low-certainty evidence that risk-screening (OR 0.90, 95% CI 0.66 to 1.23); and education, multifactorial-action and review with medication-review (OR 0.88, 95% CI 0.60 to 1.29) may result in a very slight reduction in odds of living at home. There was low-certainty evidence that education, multifactorial-action and review with medication-review and self-management may result in a reduction in odds of living at home [OR 0.41 (very large), 95% CI 0.14 to 1.17]. Other comparisons with ac were of very low certainty.

Long-term time frame

The available-care network for living at home in the long term included 13 studies (n = 14,843) with 10 intervention groups (see Table 7 for a summary of findings). 76,92,93,106,116,118,121,136,162,164,165,172,173 Multifactorial-action and review with medication-review versus ac included data from three studies; two comparisons had data from two studies and the remainder had one. Four study populations included all frailty categories, one included the robust and pre-frail categories, three the pre-frail and frail categories, one was pre-frail, three frail and one was unclassifiable. There were some concerns regarding RoB in nine study results and four were high RoB (three with serious concerns, one with very serious concerns).

TABLE 7 - Living at home in the long term: comparisons with ac summary of findings table

Calculated from OR and an assumed comparator risk of 0.816, the median ac risk among these studies.

Very serious concerns about imprecision as CI includes substantial benefit and substantial harm. Downgrade twice.

Very serious concerns about RoB due to randomisation process and missing outcome data. Downgrade twice.

Serious concerns about imprecision as no closed loop and direct comparison is based on 110 persons which does not meet optimal information size. Downgrade once (already downgraded twice for RoB).

Serious concerns about RoB due to missing outcome data. Downgrade once.

Population: Older people Interventions: Community-based complex interventions Comparator: Available care (ac) Outcome: Living at home Time frame: Long term; range of follow-up 24–43 months Setting: Community Total studies: 13 Total participants: 14,843 Comparator rank: Mean 7.5, 95% CI 5 to 9
Intervention group	Relative effect (95% CI)		Anticipated absolute effect (95% CI)				Certainty of the evidence (GRADE)	Ranking (95% CI)	Interpretation
	Network estimate	Calculated RRa	High-risk population [607 per 1000 with ac]		Low-risk population [932 per 1000 with ac]
			With intervention	Difference	With intervention	Difference
Multifactorial-action and review (mfar)	OR 1.29 (0.63 to 2.63) Indirect estimate	RR 1.04 (0.90 to 1.13)	665 per 1000 (492 to 803)	58 more per 1000 (115 fewer to 196 more)	946 per 1000 (896 to 973)	14 more per 1000 (36 fewer to 41 more)	⊕⊕⊝⊝Lowb	5.0 (1 to 10)	May result in an increase in chance of living at home
Education, multifactorial-action and review with medication-review [educ & mfar(w/med)]	OR 1.23 (0.72 to 2.10) Mixed estimate	RR 1.04 (0.93 to 1.11)	654 per 1000 (525 to 764)	47 more per 1000 (82 fewer to 157 more)	944 per 1000 (908 to 966)	12 more per 1000 (24 fewer to 34 more)	⊕⊕⊝⊝Lowb	5.2 (2 to 10)	May result in an increase in chance of living at home
Multifactorial-action and review with medication-review [mfar(w/med)]	OR 1.17 (0.94 to 1.47) Mixed estimate	RR 1.03 (0.99 to 1.06)	645 per 1000 (592 to 694)	38 more per 1000 (15 fewer to 87 more)	942 per 1000 (928 to 953)	10 more per 1000 (4 fewer to 21 more)	⊕⊕⊝⊝Lowb	5.2 (3 to 9)	May result in a slight increase in chance of living at home
ADL, nutrition and exercise (ADL & ntr & exrc)	OR 1.15 (0.64 to 2.05) Mixed estimate	RR 1.02 (0.91 to 1.10)	639 per 1000 (499 to 760)	32 more per 1000 (108 fewer to 153 more)	940 per 1000 (898 to 966)	8 more per 1000 (34 fewer to 34 more)	⊕⊕⊝⊝Lowb	5.8 (2 to 10)	May result in a slight increase in chance of living at home
Engagement in meaningful-activities and multifactorial-action with self-management strategies [eng & mfa-(w/slfm)]	OR 1.03 (0.85 to 1.25) Mixed estimate	RR 1.01 (0.97 to 1.04)	614 per 1000 (567 to 658)	7 more per 1000 (40 fewer to 51 more)	934 per 1000 (921 to 945)	2 more per 1000 (11 fewer to 13 more)	⊕⊕⊝⊝Lowb	6.9 (3 to 10)	May result in a very slight increase in chance of living at home
Risk-screening (rsk-mfa-)	OR 0.91 (0.77 to 1.07) Mixed estimate	RR 0.98 (0.95 to 1.01)	584 per 1000 (543 to 624)	23 fewer per 1000 (64 fewer to 17 more)	926 per 1000 (913 to 936)	6 fewer per 1000 (19 fewer to 4 more)	⊕⊕⊝⊝Lowb	8.8 (6 to 10)	May result in a slight reduction in chance of living at home
Aids (aids)	OR 2.64 (1.02 to 6.88) Mixed estimate	RR 1.13 (1.00 to 1.19)	803 per 1000 (611 to 914)	196 more per 1000 (4 more to 307 more)	973 per 1000 (933 to 990)	41 more per 1000 (1 more to 58 more)	⊕⊝⊝⊝Very lowc,d	1.8 (1 to 7)	The evidence is very uncertain about the effect on chance of living at home
Multifactorial-action (mfa-)	OR 2.13 (0.85 to 5.33) Mixed estimate	RR 1.11 (0.97 to 1.18)	767 per 1000 (568 to 892)	160 more per 1000 (39 fewer to 285 more)	967 per 1000 (921 to 986)	35 more per 1000 (11 fewer to 54 more)	⊕⊝⊝⊝Very lowb,e	2.5 (1 to 9)	The evidence is very uncertain about the effect on chance of living at home
Education, multifactorial-action and review with medication-review and self-management strategies [educ & mfar(w/med + slfm)]	OR 1.08 (0.78 to 1.49) Mixed estimate	RR 1.01 (0.95 to 1.06)	625 per 1000 (547 to 697)	18 more per 1000 (60 fewer to 90 more)	937 per 1000 (915 to 953)	5 more per 1000 (17 fewer to 21 more)	⊕⊝⊝⊝Very lowb,e	6.3 (3 to 10)	The evidence is very uncertain about the effect on chance of living at home

The network contained no loops and therefore the consistency assumption could not be tested and a ‘consistency’ model was fitted, setting the inconsistency parameter in the model to zero for all comparisons. Between-study variance was estimated to be very small but non-zero, indicating some degree of heterogeneity between studies included in the network (τ = 9.92 × 10⁻⁷).

There was low-certainty evidence that multifactorial-action and review [OR 1.29 (moderate), 95% CI 0.63 to 2.63]; and education, multifactorial-action and review with medication-review [OR 1.23 (moderate), 95% CI 0.72 to 2.10] may both result in an increase in the odds of living at home compared with ac in the long term. Multifactorial-action and review with medication-review (OR 1.17, 95% CI 0.94 to 1.47); and ADL, nutrition and exercise (OR 1.15, 95% CI 0.64 to 2.05) may both result in a slight increase in the odds of living at home. Meaningful-activities and multifactorial-action with self-management may result in a very slight increase in the odds of living at home (OR 1.03, 95% CI 0.85 to 1.25). Risk-screening may result in a slight reduction in odds of living at home (OR 0.91, 95% CI 0.77 to 1.07). Three other comparisons with ac were of very low certainty.

Across time frames

Multifactorial-action and review with medication-review [mfar(w/med)] was estimated to make little to no difference in the short term (low certainty) but result in small (moderate or low certainty) increases in living at home in the medium- and long-term time frames, respectively. Similarly, ADL, nutrition and exercise (ADL & ntr & exrc) may make little to no difference in the short term but result in large or small increases in living at home in the medium- and long-term time frames, respectively (low certainty). Conversely, there was low-certainty evidence that education, multifactorial-action and review with medication-review and self-management [educ & mfar(w/med + slfm)] may result in moderate or very large reductions in living at home in the short- and medium-term time frames, respectively. Risk-screening (rsk-mfa-) may result in a very slight or slight reduction in odds of living at home in the medium and long term, respectively (low certainty). There was contrasting evidence of low certainty that education, multifactorial-action and review with medication-review [educ & mfar(w/med)] may result in a slight reduction or moderate increase in living at home in the medium- and long-term time frames, respectively.

Short-term time frame

The homecare network for living at home in the short term included four studies (n = 704) and four intervention groups (see Table 8 for a summary of findings). 91,117,146,154 The comparison of homecare, ADL, multifactorial-action and review with self-management strategies versus homecare included the data of two studies, with the other two comparisons comprised of one study each. One study population was classified as pre-frail and frail, and three were frail. We had some concerns with RoB in three study results and one was at high RoB (serious concerns).

TABLE 8 - Living at home in the short term: comparisons with homecare summary of findings table

Calculated from OR and an assumed comparator risk of 0.924, the median ac risk among these studies.

Very serious concerns about imprecision as CI includes substantial benefit and substantial harm. Downgrade twice.

Very serious concerns about imprecision as the optimal information size is not met. The CI is very wide: OR CI ratio 8.0; 588 to 919 per 1000 in the high-risk population. There is no closed loop and the direct comparison is based on evidence from 163 persons. Downgrade twice.

Serious concerns about RoB due to missing outcome data. Downgrade once.

Very serious concerns about imprecision as the optimal information size is not met. The CI is very wide: OR CI ratio 8.8; 512 to 902 per 1000 in the high-risk population. There is no direct evidence, the indirect evidence coming from the comparison of hmcr & mfa - vs. hmcr & ADL & mfar(w/slfm) with 104 participants. Downgrade twice.

Population: Older people Interventions: Community-based complex interventions Comparator: Homecare (hmcr) Outcome: Living at home Time frame: Short term; range of follow-up 6–7 months Setting: Community Total studies: 4 Total participants: 704 Comparator rank: Mean 1.1, 95% CI
Intervention group	Relative effect (95% CI)		Anticipated absolute effect (95% CI)				Certainty of the evidence (GRADE)	Ranking (95% CI)	Interpretation
	Network estimate	Calculated RRa	High-risk population (923 per 1000 with hmcr)		Low-risk population (953 per 1000 with hmcr)
			With intervention	Difference	With intervention	Difference
Homecare, ADL, multifactorial-action and review with self-management strategies [hmcr & ADL & mfar(w/slfm)]	OR 0.63 (0.31 to 1.26) Mixed estimate	RR 0.96 (0.86 to 1.02)	882 per 1000 (789 to 938)	41 fewer per 1000 (134 fewer to 15 more)	927 per 1000 (863 to 962)	26 fewer per 1000 (90 fewer to 9 more)	⊕⊕⊝⊝Lowb	2.1 (1 to 3)	May result in a reduction in chance of living at home
Homecare and nutrition (hmcr & ntr)	OR 0.34 (0.12 to 0.95) Mixed estimate	RR 0.87 (0.64 to 1.00)	801 per 1000 (588 to 919)	122 fewer per 1000 (335 fewer to 4 fewer)	872 per 1000 (707 to 951)	81 fewer per 1000 (246 fewer to 2 fewer)	⊕⊕⊝⊝Lowc	3.2 (2 to 4)	May result in a reduction in chance of living at home
Homecare and multifactorial-action (hmcr & mfa-)	OR 0.26 (0.09 to 0.77) Indirect estimate	RR 0.82 (0.56 to 0.98)	756 per 1000 (512 to 902)	167 fewer per 1000 (411 fewer to 21 fewer)	840 per 1000 (639 to 940)	113 fewer per 1000 (314 fewer to 13 fewer)	⊕⊝⊝⊝Very lowd,e	3.6 (3 to 4)	The evidence is very uncertain about the effect on chance of living at home

The network contained no loops and therefore the consistency assumption could not be tested and a ‘consistency’ model was fitted. Between-study variance was estimated to be very small but non-zero, indicating some degree of heterogeneity between studies included in the network (τ = 2.85 × 10⁻⁷).

There was low-certainty evidence that homecare, ADL, multifactorial-action and review with self-management [OR 0.63 (moderate), 95% CI 0.31 to 1.26]; and homecare and nutrition [OR 0.34 (very large), 95% CI 0.12 to 0.95] may result in reductions in the odds of living at home compared with homecare in the short term. There was very low-certainty evidence regarding homecare and multifactorial-action versus homecare.

Medium-term time frame

Overall characteristics

The homecare network for living at home in the medium term included five studies (n = 1978) and six intervention groups (see Table 9 for a summary of findings). 91,125,146,154,189 Each comparison consisted of one study. All study populations were frail. There were some concerns about RoB in one study and four were at high RoB (three with serious concerns, one with very serious concerns).

TABLE 9 - Living at home in the medium term: comparisons with homecare summary of findings table

Calculated from OR and an assumed comparator risk of 0.738, the median ac risk among these studies.

Very serious concerns about imprecision as CI includes substantial benefit and substantial harm. Downgrade twice.

Very serious concerns about RoB due to randomisation process and missing outcome data. Already downgraded twice for imprecision, downgrade once.

Serious concerns about RoB due to missing outcome data. Downgrade once.

Population: Older people Interventions: Community-based complex interventions Comparator: Homecare (hmcr) Outcome: Living at home Time frame: Medium term; range of follow-up 12–13 months Setting: Community Total studies: 5 Total participants: 1978 Comparator rank: Mean 3.0, 95% CI 1 to 5
Intervention group	Relative effect (95% CI)		Anticipated absolute effect (95% CI)				Certainty of the evidence (GRADE)	Ranking (95% CI)	Interpretation
	Network estimate	Calculated RRa	High-risk population (649 per 1000 with hmcr)		Low-risk population (843 per 1000 with hmcr)
			With intervention	Difference	With intervention	Difference
Homecare, ADL, multifactorial-action and review with self-management strategies [hmcr & ADL & mfar(w/slfm)]	OR 0.76 (0.40 to 1.45) Mixed estimate	RR 0.92 (0.72 to 1.09)	585 per 1000 (426 to 728)	64 fewer per 1000 (223 fewer to 79 more)	804 per 1000 (683 to 886)	39 fewer per 1000 (160 fewer to 43 more)	⊕⊕⊝⊝Lowb	3.8 (1 to 6)	May result in a reduction in chance of living at home
Homecare, education, multifactorial-action and review (hmcr & educ & mfar)	OR 1.17 (0.85 to 1.59) Mixed estimate	RR 1.04 (0.96 to 1.11)	683 per 1000 (612 to 747)	34 more per 1000 (37 fewer to 98 more)	862 per 1000 (821 to 895)	19 more per 1000 (22 fewer to 52 more)	⊕⊝⊝⊝Very lowb,c	1.8 (1 to 4)	The evidence is very uncertain about the effect on chance of living at home
Homecare, multifactorial-action and review with medication-review [hmcr & mfar(w/med)]	OR 1.11 (0.82 to 1.51) Mixed estimate	RR 1.03 (0.94 to 1.10)	672 per 1000 (601 to 736)	23 more per 1000 (48 fewer to 87 more)	856 per 1000 (814 to 890)	13 more per 1000 (29 fewer to 47 more)	⊕⊝⊝⊝Very lowb,d	2.1 (1 to 4)	The evidence is very uncertain about the effect on chance of living at home
Homecare and multifactorial-action (hmcr & mfa-)	OR 0.51 (0.17 to 1.49) Indirect estimate	RR 0.80 (0.45 to 1.09)	485 per 1000 (243 to 734)	164 fewer per 1000 (406 fewer to 85 more)	732 per 1000 (483 to 889)	111 fewer per 1000 (360 fewer to 46 more)	⊕⊝⊝⊝Very lowb,d	5.1 (1 to 6)	The evidence is very uncertain about the effect on chance of living at home
Homecare and nutrition (hmcr & ntr)	OR 0.50 (0.23 to 1.07) Mixed estimate	RR 0.79 (0.54 to 1.02)	480 per 1000 (301 to 664)	169 fewer per 1000 (348 fewer to 15 more)	729 per 1000 (556 to 852)	114 fewer per 1000 (287 fewer to 9 more)	⊕⊝⊝⊝Very lowb,d	5.2 (2 to 6)	The evidence is very uncertain about the effect on chance of living at home

Long-term time frame

There were only two homecare comparisons in the long term (homecare and multifactorial-action vs. homecare, multifactorial-action and review; and homecare, multifactorial-action and review vs. homecare, multifactorial-action and review with self-management), so there was insufficient evidence to conduct an NMA. 103,147

Across time frames

In the short- and medium-term time frames, there was low-certainty evidence that homecare, ADL, multifactorial-action and review with self-management may result in a moderate or large reduction in the odds of living at home compared with homecare alone.

Short-term time frame

A total of six studies and seven intervention groups were analysed for the short-term IADL available-care network (see Table 10 for a summary of findings). 79,99,136,141,152,166 Intervention groups ranged from 595 participants (ac) to 24 (ADL, aids, education, exercise, multifactorial-action and review with medication-review and self-management). The populations were pre-frail, frail or both in all but one study. 79 All results were at high RoB (five: serious concerns, one: very serious concerns).

TABLE 10 - Instrumental activities of daily living in the short term: comparisons with ac summary of findings table

Calculated from the estimated SMD using a SD of 2.62, the pooled SD across intervention groups reporting the Lawton IADL.

Serious concerns about RoB due to missing outcome data. Downgrade once.

Serious concerns about imprecision as no closed loop and direct comparison is based on 316 persons which does not meet optimal information size. Downgrade once.

Very serious concerns about imprecision as CI includes substantial benefit and harm (SMD ± 0.05). Downgrade twice.

Very serious concerns about RoB due to randomisation process, missing outcome data and reported results were not analysed according to allocation. Downgrade twice.

Very serious concerns about imprecision as confidence interval includes substantial benefit and harm (SMD ± 0.05). Already downgraded twice for risk of bias, downgrade once.

Serious concerns about RoB due to ceiling effect in the outcome measurement for a substantial proportion of participants. Downgrade once.

Population: Older people Interventions: Community-based complex interventions Comparator: Available care (ac) Outcome: Independence in IADL Time frame: Short term; range of follow-up 24 weeks to 9 months Setting: Community Total studies: 6 Total participants: 1155 Comparator rank: Mean 4.4, 95% CI 2 to 6
Intervention group	Anticipated absolute effect (95% CI)		Certainty of the evidence (GRADE)	Ranking (95% CI)	Interpretation
	SMD	MD (Lawton IADL 0–8)a
Education, multifactorial-action and review with medication-review and self-management [educ & mfar(w/med + slfm)]	SMD 0.22 lower (0.45 lower to 0.00) Mixed estimate	MD 0.59 lower (1.17 lower to 0.01 lower)	⊕⊕⊝⊝ Lowb,c	6.5 (5 to 7)	May result in a slight reduction in IADL independence
ADL, aids, education, exercise, multifactorial-action and review with medication-review and self-management [ADL&aids&ed&ex&mf(w/med + slfm)]	SMD 0.38 higher (0.26 lower to 1.01 higher) Mixed estimate	MD 0.98 higher (0.69 lower to 2.66 higher)	⊕⊝⊝⊝ Very lowb,d	1.9 (1 to 7)	The evidence is very uncertain about the effect on IADL independence
ADL training, aids-adaptations and physical exercise (ADL & aids & exrc)	SMD 0.14 higher (0.09 lower to 0.36 higher) Mixed estimate	MD 0.36 higher (0.24 lower to 0.95 higher)	⊕⊝⊝⊝ Very lowb,d	2.8 (1 to 6)	The evidence is very uncertain about the effect on IADL independence
Engagement in meaningful-activities and education (eng & educ)	SMD 0.06 higher (0.18 lower to 0.30 higher) Mixed estimate	MD 0.16 higher (0.46 lower to 0.79 higher)	⊕⊝⊝⊝ Very lowe,f	3.5 (1 to 6)	The evidence is very uncertain about the effect on IADL independence
Exercise (exrc)	SMD 0.00 (0.60 lower to 0.60 higher) Mixed estimate	MD 0.00 (1.58 lower to 1.58 higher)	⊕⊝⊝⊝ Very lowd,g	4.2 (1 to 7)	The evidence is very uncertain about the effect on IADL independence
Multifactorial-action with medication-review [mfa-(w/med)]	SMD 0.05 lower (0.37 lower to 0.27 higher) Mixed estimate	MD 0.13 lower (0.98 lower to 0.71 higher)	⊕⊝⊝⊝ Very lowb,d	4.8 (2 to 7)	The evidence is very uncertain about the effect on IADL independence

The network contained no loops and, therefore, the consistency assumption could not be tested, and a ‘consistency’ model was fitted. As the network contained only a single study for each comparison, there was no potential source of heterogeneity and therefore a common-effect model was fitted for the short-term IADL available-care network. There was low-certainty evidence that education, multifactorial-action and review with medication-review and self-management may result in a slight reduction in IADL independence (SMD −0.22, 95% CI −0.45 to 0.00). All other short-term estimates were of very low certainty due to imprecision and RoB.

The 95% CI for the true rank was typically wide, encompassing six of the seven ranks for all intervention groups except education, multifactorial-action and review with medication-review and self-management (95% CI 5 to 7).

Medium-term time frame

Overall characteristics

For the medium-term IADL available-care network there was a total of 16 studies (n = 5309) with 14 intervention groups in the network, including the largest node ac (n = 3136; see Table 11 for a summary of findings). 70,73,74,79,86,88,96,99,108,136,138,152,155,167,172,177 Within the network a single study provided a direct comparison for each experimental intervention group, except multifactorial-action and review with medication-review which has four studies. In the network, the number of participants ranged from 34 (aids) to 702 (multifactorial-action and review with medication-review) excluding the comparator ac. The populations of three studies spanned all frailty categories, one study was of a robust population, five a frail population, six studies covered the other categories and one was unclassifiable. There were some concerns regarding RoB in 1 study result, with the other 15 results being at high RoB (10 serious concerns, 5 very serious concerns).

TABLE 11 - Instrumental activities of daily living in the medium term: comparisons with ac summary of findings table

Calculated from the estimated SMD using a SD of 2.62, the pooled SD across intervention groups reporting the Lawton IADL.

Serious concerns about RoB due to missing outcome data. Downgrade once.

Serious concerns about imprecision as CI crosses the no effect line and includes substantial harm. Downgrade once.

Serious concerns about imprecision as no closed loop and direct comparison is based on evidence from 260 persons which does not meet optimal information size. Downgrade once.

Serious concerns about imprecision as no closed loop and direct comparison is based on evidence from 124 persons which does not meet optimal information size. Downgrade once.

Very serious concerns about RoB due to recruitment of participants and missing outcome data. Downgrade twice.

Very serious concerns about imprecision as CI includes substantial benefit and harm (SMD ± 0.05). Downgrade twice.

Very serious concerns about RoB due to the randomisation process, missing outcome data and selection of the reported result. Downgrade twice.

Very serious concerns about imprecision as confidence interval includes substantial benefit and harm (SMD ± 0.05). Already downgraded twice for risk of bias, downgrade once.

Very serious concerns about RoB due to excluding participants in per-protocol analysis and missing outcome data. Downgrade twice.

Very serious concerns about RoB due to the randomisation process and missing outcome data. Downgrade twice.

Population: Older people Interventions: Community-based complex interventions Comparator: Available care (ac) Outcome: Independence in IADL Time frame: Medium term; range of follow-up 10–18 months Setting: Community Total studies: 16 Total participants: 5309 Comparator rank: Mean 7.2, 95% CI 5 to 9
Intervention group	Anticipated absolute effect (95% CI)		Certainty of the evidence (GRADE)	Ranking (95% CI)	Interpretation
	SMD	MD (Lawton IADL 0–8)a
Multifactorial-action and review with medication-review [mfar(w/med)]	SMD 0.11 higher (0.00 to 0.21 higher) Mixed estimate	MD 0.28 higher (0.01 higher to 0.55 higher)	⊕⊕⊕⊝ Moderateb	4.4 (2 to 7)	Probably results in a very slight increase in IADL independence
ADL, aids and exercise (ADL & aids & exrc)	SMD 0.19 lower (0.42 lower to 0.04 higher) Mixed estimate	MD 0.50 lower (1.11 lower to 0.11 higher)	⊕⊕⊝⊝ Lowb,c	11.2 (5 to 13)	May result in a slight reduction in IADL independence
ADL, aids, education, exercise, multifactorial-action and review with medication-review and self-management [ADL&aids&ed&ex&mf(w/med + slfm)]	SMD 0.56 lower (0.81 lower to 0.31 lower) Mixed estimate	MD 1.47 lower (2.12 lower to 0.82 lower)	⊕⊕⊝⊝ Lowb,d	13.9 (13 to 14)	May result in a reduction in IADL independence
Multifactorial-action and review (mfar)	SMD 0.50 higher (0.15 higher to 0.86 higher) Mixed estimate	MD 1.32 higher (0.38 higher to 2.26 higher)	⊕⊝⊝⊝ Very lowe,f	1.2 (1 to 4)	The evidence is very uncertain about the effect on IADL independence
Medication-review, nutrition and exercise (med & ntr & exrc)	SMD 0.21 higher (0.08 lower to 0.51 higher) Mixed estimate	MD 0.56 higher (0.22 lower to 1.34 higher)	⊕⊝⊝⊝ Very lowb,g	3.2 (1 to 10)	The evidence is very uncertain about the effect on IADL independence
ADL (ADL)	SMD 0.10 higher (0.12 lower to 0.33 higher) Mixed estimate	MD 0.27 higher (0.31 lower to 0.86 higher)	⊕⊝⊝⊝ Very lowb,g	4.9 (2 to 10)	The evidence is very uncertain about the effect on IADL independence
Multifactorial-action with medication-review [mfa-(w/med)]	SMD 0.02 higher (0.30 lower to 0.35 higher) Mixed estimate	MD 0.06 higher (0.79 lower to 0.92 higher)	⊕⊝⊝⊝ Very lowb,g	6.7 (2 to 13)	The evidence is very uncertain about the effect on IADL independence
Education and risk-screening (educ & rsk-mfa-)	SMD 0.00 (0.19 lower to 0.19 higher) Mixed estimate	MD 0.00 (0.50 lower to 0.50 higher)	⊕⊝⊝⊝ Very lowb,g	7.1 (3 to 12)	The evidence is very uncertain about the effect on IADL independence
Engagement in meaningful-activities and education (eng & educ)	SMD 0.01 lower (0.26 lower to 0.23 higher) Mixed estimate	MD 0.03 lower (0.68 lower to 0.61 higher)	⊕⊝⊝⊝ Very lowh,i	7.5 (2 to 13)	The evidence is very uncertain about the effect on IADL independence
Exercise and psychology (exrc & psyc)	SMD 0.12 lower (0.60 lower to 0.37 higher) Mixed estimate	MD 0.30 lower (1.58 lower to 0.98 higher)	⊕⊝⊝⊝ Very lowi,j	8.9 (2 to 14)	The evidence is very uncertain about the effect on IADL independence
Multifactorial-action with medication-review and self-management [mfa-(w/med + slfm)]	SMD 0.07 lower (0.20 lower to 0.06 higher) Mixed estimate	MD 0.19 lower (0.53 lower to 0.15 higher)	⊕⊝⊝⊝ Very lowi,k	9.2 (5 to 12)	The evidence is very uncertain about the effect on IADL independence
Aids (aids)	SMD 0.15 lower (0.60 lower to 0.30 higher) Mixed estimate	MD 0.39 lower (1.56 lower to 0.78 higher)	⊕⊝⊝⊝ Very lowi,k	9.6 (2 to 14)	The evidence is very uncertain about the effect on IADL independence
Education, multifactorial-action and review with medication-review and self-management [educ & mfar(w/med + slfm)]	SMD 0.13 lower (0.35 lower to 0.09 higher) Mixed estimate	MD 0.34 lower (0.92 lower to 0.24 higher)	⊕⊝⊝⊝ Very lowb,g	10.1 (5 to 13)	The evidence is very uncertain about the effect on IADL independence

Long-term time frame

Six studies and five intervention groups were analysed in the long-term IADL available-care network, incorporating 1727 participants (see Table 12 for a summary of findings). 73,115,136,155,162,172 The available-care node was the largest (n = 858) and aids the smallest (n = 34).

TABLE 12 - Instrumental activities of daily living in the long term: comparisons with ac summary of findings table

Calculated from the estimated SMD using a SD of 2.62, the pooled SD across intervention groups reporting the Lawton IADL.

Serious concerns about imprecision as CI crosses the no effect line and includes substantial harm. Downgrade once.

Very serious concerns about imprecision as CI includes substantial benefit and harm (SMD ± 0.05). Downgrade twice.

Serious concerns about RoB due to missing outcome data. Downgrade once.

Very serious concerns about RoB due to randomisation process and missing outcome data. Downgrade twice.

Very serious concerns about imprecision as confidence interval includes substantial benefit and harm (SMD ± 0.05). Already downgraded twice for risk of bias, downgrade once.

Population: Older people Interventions: Community-based complex interventions Comparator: Available care (ac) Outcome: IADL (higher is better) Time frame: Long term; range of follow-up 24–36 months Setting: Community Total studies: 6 Total participants: 1727 Comparator rank: Mean 3.4, 95% CI 2 to 5
Intervention group	Anticipated absolute effect (95% CI)		Certainty of the evidence (GRADE)	Ranking (95% CI)	Interpretation
	SMD	MD (Lawton IADL 0–8)a
Multifactorial-action and review with medication-review [mfar(w/med)]	SMD 0.08 lower (0.21 lower to 0.05 higher) Mixed estimate	MD 0.21 lower (0.56 lower to 0.13 higher)	⊕⊕⊕⊝ Moderateb	4.5 (3 to 6)	Probably results in a very slight reduction in IADL
Risk-screening (rsk-mfa-)	SMD 0.23 higher (0.13 lower to 0.60 higher) Mixed estimate	MD 0.61 higher (0.35 lower to 1.57 higher)	⊕⊕⊝⊝ Lowc	1.7 (1 to 5)	May result in a slight increase in IADL
Education, multifactorial-action and review with medication-review [educ & mfar(w/med)]	SMD 0.14 higher (0.08 lower to 0.36 higher) Mixed estimate	MD 0.37 higher (0.21 lower to 0.95 higher)	⊕⊕⊝⊝ Lowc	2.1 (1 to 5)	May result in a very slight increase in IADL
Education, multifactorial-action and review with medication-review and self-management [educ & mfar(w/med + slfm)]	SMD 0.21 lower (0.44 lower to 0.01 higher) Mixed estimate	MD 0.56 lower (1.14 lower to 0.02 higher)	⊕⊕⊝⊝ Lowb,d	5.5 (4 to 6)	May result in a slight reduction in IADL
Aids (aids)	SMD 0.03 lower (0.48 lower to 0.42 higher) Mixed estimate	MD 0.07 lower (1.25 lower to 1.10 higher)	⊕⊝⊝⊝ Very lowe,f	3.8 (1 to 6)	The evidence is very uncertain about the effect on IADL

The network contained no loops and, therefore, the consistency assumption could not be tested and a ‘consistency’ model was fitted. Between-study variance was estimated to be very small but non-zero, indicating some degree of heterogeneity between studies included in the network (τ = 5.31 × 10⁻⁶). There was moderate-certainty evidence that multifactorial-action and review with medication-review was associated with a very slight reduction in IADL independence (SMD −0.08, 95% CI −0.21 to 0.05). There was low-certainty evidence that risk-screening (SMD 0.23, 95% CI −0.13 to 0.60); and education, multifactorial-action and review with medication-review (SMD 0.14, 95% CI −0.08 to 0.36) each increased IADL independence (slightly and very slightly, respectively). There was low-certainty evidence that education, multifactorial-action and review with medication-review and self-management (SMD −0.21, 95% CI −0.44 to 0.01) was associated with a slight reduction in IADL independence.

For the long-term IADL available-care network, the 95% CI for the true ranking placed ac 2nd–5th, multifactorial-action and review with medication-review in the bottom four, and education, multifactorial-action and review with medication-review and self-management in the bottom three. For the other three interventions, the 95% CI covered at least five of the six places.

Across time frame

There was moderate-certainty evidence that multifactorial-action and review with medication-review was associated with a very slight increase in IADL independence in the medium term but also a very slight reduction in IADL in the long term. There was low-certainty evidence that education, multifactorial-action and review with medication-review and self-management was associated with a slight reduction in IADL independence in the short and long term.

Short-term time frame

In the short-term IADL homecare network, there were a total of four studies and five intervention groups, with the homecare node having the largest number of participants (n = 301) and homecare, multifactorial-action and review the smallest (n = 121; see Table 13 for a summary of findings). 65,87,117,146,147 Three study populations were frail and one was pre-frail and frail. Two results were at high RoB (serious concerns) and two results were rated as some concerns.

TABLE 13 - Instrumental activities of daily living in the short term: comparisons with homecare summary of findings table

Calculated from the estimated SMD using a SD of 2.62, the pooled SD across intervention groups reporting the Lawton IADL.

Very serious concerns about imprecision as CI includes substantial benefit and harm (SMD ± 0.05). Downgrade twice.

Serious concerns about RoB due to missing outcome data in indirect evidence via the homecare, ADL, multifactorial-action and review with self-management vs. homecare and multifactorial-action comparison. Downgrade once.

Serious concerns about RoB due to missing outcome data. Downgrade once.

Serious concerns about RoB because multiple analyses were conducted but the results from only one analysis were reported. Downgrade once.

Population: Older people Interventions: Community-based complex interventions Comparator: Formal homecare (hmcr) Outcome: Independence in IADL (higher is better) Time frame: Short term; range of follow-up 6–7 months Setting: Community Total studies: 4 Total participants: 970 Comparator rank: Mean 2.5, 95% CI 1 to 4
Intervention group	Anticipated absolute effect (95% CI)		Certainty of the evidence (GRADE)	Ranking (95% CI)	Interpretation
	SMD	MD (Lawton IADL 0–8)a
Homecare, ADL, multifactorial-action and review with self-management [hmcr & ADL & mfar(w/slfm)]	SMD 0.02 lower (0.33 lower to 0.29 higher) Mixed estimate	MD 0.06 lower (0.88 lower to 0.76 higher)	⊕⊕⊝⊝ Lowb	2.8 (1 to 5)	May result in little to no difference in IADL independence
Homecare, multifactorial-action and review (hmcr & mfar)	SMD 0.04 higher (0.51 lower to 0.58 higher) Indirect estimate	MD 0.09 higher (1.33 lower to 1.52 higher)	⊕⊝⊝⊝ Very lowb,c	2.4 (1 to 5)	The evidence is very uncertain about the effect on IADL independence
Homecare and multifactorial-action (hmcr & mfa-)	SMD 0.07 lower (0.55 lower to 0.42 higher) Indirect estimate	MD 0.18 lower (1.44 lower to 1.09 higher)	⊕⊝⊝⊝ Very lowb,d	3.4 (1 to 5)	The evidence is very uncertain about the effect on IADL independence
Homecare and medication-review (hmcr & med)	SMD 0.13 lower (0.31 lower to 0.06 higher) Mixed estimate	MD 0.33 lower (0.82 lower to 0.15 higher)	⊕⊝⊝⊝ Very lowb,e	3.9 (1 to 5)	The evidence is very uncertain about the effect on IADL independence

The network contained no loops and therefore the consistency assumption could not be tested and a ‘consistency’ model was fitted. As the network contained only a single study for each comparison, there was no potential source of heterogeneity and therefore a common-effect model was fitted.

One comparison with homecare had low certainty. Homecare, ADL, multifactorial-action and review with self-management may result in little to no difference in IADL independence (SMD −0.02, 95% CI −0.33 to 0.29; low certainty). There were no serious concerns about RoB but the CIs for all estimated effects included benefit and harm.

There was little difference in the mean rank of intervention groups.

Medium-term time frame

Long-term time frame

It was not possible to conduct analysis of IADL in the long term for the homecare network as it only had two intervention groups. 146,147

Across Time frames

The only evidence that was not very uncertain was from the short-term time frame.

Short-term time frame

A total of eight trials (n = 4075 participants) compared a short-term PADL outcome across nine intervention groups including ac, which was the largest node (see Table 14 for a summary of findings). 69,79,99,136,152,166,168,186 There was a maximum of one trial for a single direct comparison (compared to ac), with the number of participants within any one intervention group in the network ranging from 24 to 1983. All studies were at high RoB (five serious concerns, three very serious concerns).