We elaborated a protocol that complies with the Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols (PRISMA-p) 2015 checklist [23]. Our systematic review protocol was submitted to the International Prospective Register of Systematic Reviews (PROSPERO), registration number: CRD42018105849.

Ethics approval was not required for this systematic review. The present review complies with the PRISMA checklist [24].

We searched for studies in the following electronic databases: PubMed (which includes MEDLINE), ISI Web of Knowledge, Scopus, and EBSCOhost platform to access PsycINFO, CINAHL Plus with Full Text, Humanities Abstracts (H.W. Wilson), and PsycARTICLES. Reference lists of included studies or relevant reviews were also scanned to ensure a good literature support.

Our search strategy was defined by two of the authors (A.R., L.S.) and implemented by august 2018. Several keywords combinations were tested to ensure that the strategy would retrieve a high proportion of eligible studies. When the platform allowed, we selected human subjects and the previous referred types of study. We present the search strategy in the Protocol referred above.

Retrieved citations from the search were uploaded and End-Note X8^® was used as reference management software. There was a first screening stage and, based on the title and abstract, studies moved to the second screening stage if they met all eligibility criteria or if there was uncertainty about their eligibility. In the second screening stage, the full text was assessed. Reasons for excluding citations in the second screening stage were recorded.

Eligibility assessment for selecting studies was performed independently by 2 reviewers (A.R., L.S.) to reduce the possibility of rejecting relevant reports, at both screening stages. In case of disagreement between the two reviewers, a third member of the team was consulted (L.F.). Data was extracted and confirmed by the 2 reviewers. Statistical analyses were not performed.

We present a systematic review with information presented in the text and in a table to summarize and explain the characteristics and findings of the included studies. The results will be presented first by characteristics of the studies, and then in order of scientific findings, from main to additional outcomes.

Information was extracted from each included study on: 1) characteristics of the study (authors, year, country, study design, setting, aim; research questions and time of follow-up) 2) characteristics of participants (age, gender, grandparents, not grandparents, sample size), and inclusion/exclusion criteria; 3) cognitive impairment (type of measurement, level of impairment) and other outcomes of interest measured; 4) grandparenting (custodial/occasional, reasons for grandchild care, time of caring, family context). Once the age of becoming a grandparent is highly variable, an assumption of searching only for aged persons was not made.

The quality of included articles was evaluated by two reviewers (A.R., L.S.) independently. Disagreements were resolved by discussion. It was used the Downs and Black [25] checklist for randomized and non-randomized studies. Our search did not retrieve any qualitative studies.

Our search identified a total of 224 records. 30 additional records were found at data platforms, when similar studies were suggested by the platforms, and through snowballing process. After exclusion of duplicates, 178 records were considered for further analysis. Of those, seventeen records entered the second phase. The title and/or abstract of excluded records on the first phase showed these didn’t fulfil the inclusion/exclusion criteria to the present study.

After applying eligibility criteria to select the studies, nine were excluded on the second phase. Seven studies [16,26-31] were excluded for not having an assessment of cognition. Two studies [8,32] were excluded once cognition was not an outcome.

A total of eight records were included for this systematic review: five original articles [33-37], one narrative review [22] and two editorials [17,38]. One of the original articles, a working paper, was retrieved during snowballing process [37].

We provide a PRISMA [24] flow diagram that shows selection of papers (Figure 1).

As previously stated, five of the eight selected studies were original articles. All the original studies were published in English and had an observational and retrospective design. Key features are summarized in Table 1.

The overall number of participants was 38,002, from different countries but all from community dwelling. The most frequent inclusion criteria were adults, women, and have grandchildren. As exposure, all the studies evaluated taking care of grandchildren and their primary outcome was cognition. Secondary or additional outcomes were not present in all studies. The characteristics of each original study regarding population, sample, type of grandparental child care, tests used to evaluate cognition and other relevant variables are presented next.

Arpino and Bordone [33] conducted a repeated cross-sectional study, retrieving their sample form the successive waves of 2004/2005 and 2006/2007 from SHARE, a multidisciplinary longitudinal survey, representative of noninstitutionalized population aged 50 and over. Data retrieved was from the following countries: Austria, Belgium, Czech Republic, Denmark, France, Germany, Greece, Ireland, Israel, Italy, Netherlands, Poland, Spain, Sweden and Switzerland. Participants were women and men (5,610 and 4,760, respectively), with one or more child and with 50 to 80 years old. Those who were permanently sick or disabled, with history of stroke/Parkinson’s disease or cancer, grandparents with co-resident grandchildren, and outliers for the outcome variable or missing values were excluded. Grandparental child care evaluated was supplementary care (i.e. complementary to parental care) in the previous 12 months; primary or co-residing child care categories were excluded. High (>2.5 hours/day), moderate (0.8-2.5 hours/day) or low (0.25-0.8 hours/day) care was compared to very low (0-0.25 hours/day) or none care (including not having grandchildren). Cognition was evaluated through the following tests: verbal fluency, numeracy, immediate recall and delayed recall. Several covariables were included: country, age, education, partner, activity status, social activities, health (limitations on activities of daily living, self-reported health, depression), physical inactivity, smoking and alcohol consumption.

Two studies from Burn et al. [34], and Burn and Szoeke [35] with a cross-sectional design, retrieved their sample from the Women’s Healthy Ageing Project (WHAP). This is an Australian project consisting on a follow-up to the Melbourne Women’s Midlife Health Project with postmenopausal women, aged 45 to 55 in 1991, Australian-born, who had menstruated 3 months prior to recruitment, and had not taken oestrogen-containing hormone therapy. Burn et al. [34] selected 186 women who were aged 57 to 68, from the WHAP wave of 2002/2004. The authors included participants who completed all relevant neuropsychological measures and were not missing key demographic data. Regarding grandchild care, the following comparisons were made: having versus not having grandchildren; not having versus not minding grandchildren; and minding versus not minding grandchildren. Time spent minding grandchildren was also evaluated and compared: never, once every few months, once per month, once every couple of weeks, few hours per week; 1, >1, ≥3 or ≥5 days/week. Cognition was evaluated using California Verbal Learning Test (CVLT), with immediate and delayed recall, for verbal episodic memory; Symbol-Digit Modalities Test (SDMT), for working memory and processing speed; and Tower of London (TOL), for executive function. The predictors included were: age, education, employment status and time spent minding grandchildren. The secondary outcome measured was feeling demand in the previous 12 months.

Burn and Szoeke [35] selected 224 women who were aged 64 to 77, from the wave of 2012/2014 of WHAP. The authors excluded participants with neurological impairment, incomplete data on grandparenting or cognitive function, and medical reasons. Regarding grandchild care, the following differences were evaluated: not having grandchildren, having grandchildren and minding grandchildren. Time spent minding grandchildren was investigated as 1 day/week or >1 days/week. Cognition was evaluated using CVLT (with delayed recall and recognition) and Controlled oral word association test (COWAT) for verbal episodic memory; digit span task (DST) and Verbal fluency test (animals and names), for executive function; Rey-Osterreith complex figure test (RCFT), with copy and delayed recall, for visuospatial function. The control variables included were: age, education and time with grandchildren. The secondary outcome measured was feeling demand in the previous 12 months.

Jun [36] conducted a longitudinal study and retrieved their sample form the waves of 2008 (Time 2) and 2010 (Time 3) from the Korean Longitudinal Study of Aging (KLoSA), meaning a follow-up time of 2 years. KLoSA is a nationally representative sample of Koreans with ≥45 years old. In this study, participants were 2,341 women, with 50 to 74 years old. The authors included: women, aged 45 to 74 at Time 2, participation in all 3 waves of KLoSA, living grandchildren at Time 2, grandchildren aged ≤10. The sample was limited to grandmothers who provided more intensive and frequent child care (≥10 hours/week). Cognition was evaluated using the Korean version of the Mini-Mental State Exam (K-MMSE). Several control variables were included: age, marital status, household income, caregiver status, employment status, frequency of participation in social groups at Time 2, general health status and depression (10-item Center for Epidemiologic Studies-Depression scale). Education was considered a moderator in this study.

Reinkowski [37] also designed a longitudinal study, with a follow-up time of 8 years. Their sample originated from the waves of 2004/2005, 2006/2007, and 2011/2012 from SHARE. Data retrieved was from the following countries: Austria, Belgium, Czechia, Denmark, Estonia, France, Germany, Greece, Hungary, Ireland, Israel, Italy, Netherlands, Poland, Portugal, Slovenia, Spain, Sweden, Switzerland. Participants were 29,461 women, with 45 to 90 years old. Besides these two characteristics, were included those with ≥1 grandchild aged ≤16 and who have complete survey records. Grandparental child care evaluated was occasional care (i.e. 200 to 500 hours per year): no care, 0; almost daily, 38.7; almost every week, 11.3; almost monthly, 5.8; less than monthly, 2.4 mean hours/week. Cognition was evaluated through immediate recall and delayed recall tests. Other outcomes measured were: physical health (index) and mental health (euro-D depression index). Control variables used were: age, education, marital status, employment status, wealth, geographic distance and social engagement. Explanatory variables were: number of children and of grandchildren.

We assessed the risk of bias in our five original studies included using Downs and Black [25] checklist and both authors agreed that all those studies had enough quality to be presented. Studies were punctuated with a score considered fair/medium, of 14 to 17 point on a possible total of 27. This result consensual between the two reviewers on three-17 points to Arpino and Bordone [33], 17 to Burn and Szoeke [35], 16 to Burn et al. [34]- and differed on two but maintaining the same level-17 and 14 to Jun [36], 17 and 15 to Reinkowski [37].

Four studies, Arpino and Bordone [33], Burn et al. [34], Burn and Szoeke [35] and Jun [36], found a positive relationship between grandparenting and different cognition assessment tests: verbal fluency (related with a crystallized component of cognition), CVLT (for verbal episodic memory), DST/verbal fluency test (for executive function), K-MMSE (for global cognitive functioning). However, Burn et al. [34] also suggested a limit to the benefits of caregiving, with more intense grandparental care being associated with worst working memory and processing speed (SDMT). Moreover Reinkowski [37] suggested that grandchild care did not have a significant causal effect on grandparents’ cognition. This is a result approximated to the results of longitudinal study developed by Jun [36] except that here it was stated that grandchild care can have different effects on global cognitive functioning (K-MMSE), when analysing by grandmothers’ educational level.

Considering other outcomes, Burn et al. [34] found a positive association between feeling demand and the time spent minding grandchildren and Burn and Szoeke [35] tested for its association with cognition but it was not significant.

Employment status was a negative predictor for verbal episodic memory (CVLT) on Burn et al. [34] and education was a positive predictor for working memory and processing speed (SDMT). Parallel to this, on Jun [36], grandchild care was positively associated with education (p<0.05) and household income (p<0.001), but negatively associated with age (p<0.01), employment status (p<0.001), and depression (p<0.01). Other outcome evaluated by Reinkowski [37] was mental health (using euro-D depression index) and physical health. Although physical health has been positively correlated with grandchild care and cognitive function, after a more precise statistical analysis, the causal effect was not verified for neither of the two variables.

Regarding individual advantages of studies, we consider that those who restricted their sample to a specific type of grandparenting are more accurate once this aspect can have different implications [39]. Arpino and Bordone [33] and Reinkowski [37] evaluated supplementary and occasional care, respectively. The only study that evaluated gender differences was Arpino and Bordone [33]; the other studies evaluated only women and it can be important to consider that cognitive features are modulated in a different way [40]. Also related with sociodemographic characteristics, grandchildren’s age can be important considering different demands on care across different ages. Jun [36] and Reinkowski [37] limited grandchildren’s age to 10 and 16 years old, respectively. Finally, these two studies were the only ones considering social engagement.

Considering health and diseases associated with cognitive impairment, Arpino and Bordone [33] and Burn and Szoeke [35] excluded participants with medical conditions that could affect the outcome. Jun [36], controlled for previous level of grandmothers’ cognition.

Neither of the included studies evaluated what activities grandparents do when they mind grandchildren or whether the care was voluntary or not. All the samples were retrieved from populational longitudinal surveys. Persons who agree to participate on these surveys can be the ones healthier or with more free time than the general population, what can have implications on the outcomes. Also related with the samples sizes, Burn et al. [34] and Burn and Szoeke [35] had significantly smaller samples than the other studies (186 and 224 participants, respectively). In longitudinal studies, Jun [36] had less time of follow-up (two years) than Reinkowski (eight years) [37].

Statistical analysis is also important. Cross-sectional studies can limit the ability to find an accurate relationship between grandparenting and cognition [33-35]. Arpino and Bordone [33] and Reinkowski [37] had an Instrumental Variable Approach as statistic methodology to correct for confounding, what can be an advantage [41].

Lastly, considering our principal question, in all studies there seems to be a trend towards a positive effect on grandparents’ cognition associated with taking care of their grandchildren.

In the studies included, various factors can explain the differences found between groups such as the demand of grandparenting role, employment status of grandparents, educational level of grandmothers, depression and physical health.

Regarding the narrative review and the two editorials, Burn and Szoeke [35] and Woods [38] they were mainly used for snowballing process. The narrative review [22] differs from ours, once our methodology consisted in a more systematic and precise approach, and also more recent. Their findings were based on two of our included articles, and two conference papers that were not included in our review. The latter did not present significantly different conclusions, just adding the importance of differences between genders.

When asking if physicians should have specific advises for older persons on grandchildren care and cognitive impairment, despite the promising results, one should have caution on counselling grandparents regarding the positive effects of grandchild care on cognition since evidence is still weak and it seems that up to a certain level highly frequent grandparenting might as well have some deleterious effects. Also, most of the studies are not restricted to older persons, what can influence the evidence on this subject.

There were substantial differences across studies, not only concerning the design of the study or inclusion/exclusion criteria, but also regarding specificities of grandparenting, measures of the outcomes and statistical analysis. There was insufficient similarity for the development of a quantitative analysis of findings across studies. Instead, we described the studies, indicating which differences were statistically significant, without comparing direct measured results between studies. We were not able to present metanalytic measures due to the great heterogeneity regarding study designs and outcome measures. Results must be considered with caution once there are diverse contextual factors that were not taken into account, specially differences between countries, cultures, families.

In future research, more precise inclusion criteria should be established. Definition of the type of grandparental care is important according to differences between levels of care: custodial (i.e. being the primary caregiver), non-caregiver or caregiver (occasional, intermediate or extensive caregivers, according to hours of grandchild care in an average week and overnight care) [39]. Living in the same household can also be considered as a differentiator. The age of grandparents and previous diseases should be considered once mild cognitive impairment and dementia can develop in older persons due to organic and degenerative processes. Also, grandfathers should systematically be included once it was suggested that there can be differences between genders [33]. Number of grandchildren and their age may also be considered as a predictor variable or as a factor for sample stratification, considering the different demands through a child’s development. As a potential social role, it would be interesting to assess what kind of activities do grandparents practice with their grandchildren: play or practice physical activity can be different from feeding or household shores. It would also be important uniformization of outcome measures in future research on this topic. This would allow a more accurate comparison between studies and find more solid evidence about the effect of grandparenting on grandparent’s cognition.