The searches retrieved 9631 papers and a further seven papers were identified through citation tracking. Figure 1 summarises the selection process. Forty papers were included in the review [9, 25–63].
Study characteristics
Eleven papers reported RCTs (n = 7) [30, 33, 35, 37, 54, 58, 62] and non-RCTs (n = 4) [27, 29, 34, 57]. As three RCT papers reported different outcomes collected at the same follow-up from the Experience Corps trial [30, 35, 58]; a total of five independent RCTs [33, 37, 54, 62] were identified. Of the non-RCTs [27, 29, 34, 57], one paper was drawn from a small sample (n = 18) of Experience Corps trial participants [29] completing a comprehensive battery of cognitive tests that were not routinely administered in the wider cohort. A second paper [57], reported three year follow-up data for a sample of African-American women recruited to the Experience Corps over a two year period compared with matched controls selected from another cohort study.
Twenty-nine papers reported longitudinal analysis of cohort studies; 13 papers reported unique cohorts. The remaining papers reported data from four cohorts: American Changing Lives study (ACL, n = 8) [40–42, 46–48, 59, 60]; the National Survey of Midlife Development in the US (MIDUS, n = 3) [25, 31, 36]; the Survey of Health, Ageing and Retirement in Europe (SHARE, n = 2) [56, 61]; and the Wisconsin Longitudinal Study (WLS, n = 3) [9, 51, 52].
An in-depth description of each paper (including study design, population, comparator groups, health outcome measured etc.) is summarised in Additional file 1: Table S1 and Additional file 2: Table S2.
Methodological quality
Four out of five RCTs were at moderate or high risk of bias due to the random sequence generation not being described, high attrition rates, and small sample sizes available for analysis (see Additional file 3: Table S3). Quality appraisal of non-RCTs (see Additional file 4: Table S4) found three non-RCTs were at moderate risk of bias, and one non-RCT at low risk. In contrast, most papers reporting cohort studies (see Additional file 4: Table S4) were large and well designed (25/29 low risk, 4/29 moderate risk).
Study participants and interventions
Experimental studies
All studies were based in the USA and recruited people aged 50 years or over except one Israeli study involving people aged between 19–60 years [33] (Additional file 1: Table S1). The study populations were predominantly female. In the final analyses, there were 308 participants from the five RCTs and 307 participants from the four non-RCTs.
Four RCTs [30, 35, 37, 54, 58, 62] and one non-RCT [27] investigated intergenerational volunteering interventions; settings included schools [30, 35, 37, 58], a state hospital [54], a long-term care living facility [62] and a retirement facility [27].
The frequency of volunteering varied from 30 minutes to 15 hours a week. Study duration ranged from five weeks to eight months apart from two studies [34, 57] that followed-up participants for two to three years.
Cohort studies
Most cohort studies (see Additional file 2: Table S2) recruited large samples of community-dwelling adults; only six papers [28, 31, 36, 41, 45, 49] reported sample sizes of less than 1000 participants. Just one study’s sample included participants from community and institutional settings [44]. Most cohorts (13/17) were in North America, with the remainder located in Israel [26], Germany [43], England [63] and a European collaboration (SHARE) of ten [61] or thirteen nations [56]. Although some cohorts (e.g. ACL, MIDUS, WLS) recruited adults of all ages, most papers restricted analysis to participants aged 50 years or over.
The proportion of participants who reported volunteering varied considerably (5.7% [55]-75.6% [45]). Direct comparisons both between- and within-cohorts were problematic due to differences in samples and definition of volunteering status. Some papers describing complex multivariate analyses omitted basic descriptive information on the prevalence of volunteering [9, 28, 41, 42, 51, 52, 63]. Notwithstanding this, volunteering estimates derived from Japan (5.7% [55]), Israel (10.7% [26]), Europe (12.4% [56], 15.41% [61]) and Germany (23% [43]) were generally lower than those from North America; here only 3/17 papers reported volunteering rates below 30% [38, 39, 44]. Participants’ age appeared to influence prevalence rates; studies using baseline samples of predominantly younger adults yielded higher estimates of volunteering rates [40, 45, 53, 60] than those composed mainly of older adults.
Descriptive data (see Additional file 2: Table S2) on the nature (e.g. setting, type of activity, frequency or duration) of the volunteering activities were relatively sparse and no clear patterns emerged. ollow-up length varied considerably between cohorts and papers e.g. 1–10 years (12 cohorts, 22 papers) [9, 25, 26, 31, 32, 36, 38–42, 46–50],[55, 56, 59–61, 63], and 14–30 years (5 cohorts, 7 papers) [28, 43–45, 51–53].
Impact of volunteering on health outcomes and survival
Experimental studies
Additional file 5: Table S5 summarises the impact of volunteering on physical and mental health outcomes. Only outcomes relating to depression, self-rated health, self-esteem and cognitive function were reported by more than one trial.
Vote counting did not find any consistent, significant health benefits arising through volunteering. Three RCTs found no between-group differences in depression [37, 54, 62], one RCT [54] and two non-RCTs [27, 34] found no significant differences in self-esteem, and an RCT [62] and one non-RCT [34] found no difference in self-rated health. Measures of cognitive function varied within the Experience Corps trial [29, 30, 35] and another RCT [37]. Only one RCT [29] found volunteering significantly improved cognitive function. Two trials reporting data on purpose in life found no significant effect [34, 37].
All other health outcomes were only measured by one trial, with volunteering significantly associated with increased physical activity [35, 57, 58], strength [35], walking speed [35], empowerment [33], wellbeing [62], and decreased stress [37]. No significant effects were found for the number of falls in the previous year [35], cane use [35], sense of usefulness [37], and loneliness [54].
Cohort studies
Additional file 6: Table S6 summarises the impact of volunteering on survival, and physical and mental health outcomes reported in cohort studies.
Survival rates were reported in seven studies [9, 26, 38, 44, 47, 49, 50], with most follow-ups ranging from 4–8 years; only one study followed participants for 25 years [44]. Three studies reported no association with volunteering [9, 44, 49]. The remainder found statistically significant associations between at least one measure of volunteering status or intensity (e.g. frequency, hours spent, number of organisations supported) and mortality [26, 38, 47, 50]. Interpretation was difficult, however, as studies reported statistically significant, but contradictory associations.
Sufficient data were available to pool mortality data for five studies [9, 26, 38, 49, 50] with participant follow-ups ranging from four to seven years. After adjusting for important potential socio-demographic and health-related confounders, volunteers had a significantly lower risk of mortality (risk ratio: 0.78; 95% CI: 0.66, 0.90; I2 test: p = 0.65) compared to non-volunteers (Figure 2).
Vote counting was possible for physical functional abilities and self-rated health. Three cohorts [42, 45, 46, 55, 59] reporting functional abilities (activities of daily living) yielded inconclusive evidence, partially due to the way volunteering was measured. The ACL cohort found both volunteering status and the number of hours spent volunteering improved functional dependency [46, 59]. However, more sophisticated path analysis suggested a clustering effect in people aged 60 years and above compared to their younger counterparts [42]. In contrast, no relationship was found between functional ability and volunteering status [55]. The remaining study [45], reported that intermittent volunteering (as opposed to sustained participation) resulted in benefits, while neither the age nor transitions (i.e. starting/stopping volunteering) impacted on functional ability outcomes.
Four cohorts [45, 46, 52, 53, 59, 60] reported self-rated health, three of which assessed outcomes for 20 years or more [45, 52, 53]. Volunteering status was associated with higher levels of self-rated health in two cohorts [46, 52, 59, 60]. A third study [53] found benefits were associated with environmental volunteering rather than civic volunteering or no volunteering, while another [45] reported no benefits.
Single papers reported other physical health outcomes. Environmental volunteering was associated with higher levels of physical activity across a 20 year follow-up compared with either civic volunteering or no volunteering [53]. Neither volunteering status nor hours spent volunteering were associated with frailty [39] (three year follow-up), and no association in the number of chronic conditions reported [59] was found (eight year follow-up).
Vote counting was possible for depression, life satisfaction, wellbeing, and quality of life. Depressive symptoms were assessed in six cohorts [32, 36, 40–42, 46, 48, 51, 53, 63], with follow-ups ranging from 2–20 years. Irrespective of how it was measured, volunteering was associated with reduced levels of depression in four cohorts [32, 40–42, 46, 48, 51, 63], with two cohorts reporting no benefits [36, 53]. Of cohorts reporting benefits, it was difficult to synthesise clear messages as the way volunteering was modelled (status, intensity, consistency etc.) varied considerably. Analyses of the ACL cohort suggested that while volunteering status and the hours spent volunteering were associated with improved outcomes, this benefit may be limited to older volunteers [40, 42, 48]. Data from ACL [48], English Longitudinal Study of Ageing (ELSA) [63] and WLS [51] cohorts suggested that benefits only accrue through sustained rather than intermittent volunteering.
Four [28, 43, 60, 63] of the five cohorts [28, 43, 44, 60, 63] that assessed life satisfaction reported benefit (3–25 year follow-ups). Two studies explored the effect of volunteering intensity, with improvements being associated with greater time spent (hours) [60] and/or a regular, weekly commitment [43]. One study found benefits were associated with sustained rather than intermittent volunteering [63].
Volunteering status was significantly associated with improved wellbeing in three cohorts [25, 28, 31, 51, 52] (10–29 year follow-ups) but findings regarding volunteering intensity were inconsistent. WLS data [51, 52] suggested that greater benefits were associated with sustained or intermittent volunteering, and volunteering for a diverse range of organisations whereas MIDUS data [25, 31] found benefits only accrued with volunteering one to ten hours per month (with no associated benefit with greater time commitments).
Improved quality of life was associated with volunteering status in two cohorts [56, 61, 63] (2–5 year follow-ups) but only if the activity was reciprocal, i.e. the volunteer felt their actions to be appreciated [56, 63]. Some mental health outcomes were only analysed in one cohort. Volunteering was associated with improved self-efficacy for activities of daily living (one year follow-up) [55] but not with ‘happiness’ (25 year follow-up) [44].