This meta-analysis was not registered and carried out in accordance with the checklist of the Meta-analysis Of Observational Studies in Epidemiology (MOOSE) guidelines and not registered (24) Two researchers separately searched the PubMed, Embase, Web of Science, and China National Knowledge Infrastructure (CNKI) databases from their inception to May 2022 for pertinent studies published in any language. We used the following keywords “oral health,” “dental loss,” “tooth loss,” “periodontitis,” “edentulous,” “missing tooth,” “dental care,” “denture,” “oral disease,” or “poor tooth health” and “dementia,” “Alzheimer's disease,” “AD,” “cognitive impairment,” “mild cognitive impairment,” “MCI,” “cognitive decline,” “neurocognitive disorder,” “cognitive disorder,” “memory disorder,” “vascular dementia,” or “VaD” along with “cohort studies,” “follow-up studies,” “prospective studies,” or “longitudinal studies” (see Supplementary material for details). Additionally, all identified pertinent publications' reference lists were examined.

Two investigators independently assessed each cohort study's eligibility, and any disagreements were settled by consultation with a third investigator. Studies that met the following criteria were included in this meta-analysis: (1) tooth loss served as the exposure, and dementia and its subtypes or cognitive decline served as the outcome; (2) the study design was a cohort study, including nested case-control studies with a prospective design or a retrospective cohort study; and (3) the study provided HR or relative risk (RR) with corresponding 95% CI for the association between tooth loss and dementia and/or cognitive decline.

Studies were excluded if any of the following conditions were met: (1) the study was not published as full reports, such as comments, communication, conference abstracts, and letters to editors; (2) participants were patients with dementia at baseline; (3) the study design was cross-sectional or case-control; and (4) the study was not HR/RR or lacked sufficient data to calculate them.

Two researchers independently collected data from the included studies and discussed the discrepancies; then an entire research group reviewed and confirmed the data. The following information was extracted from the included studies: first author(s) name, year of publication, sex, sample, study name, country, type, baseline age, tooth loss, denture wearing, assessment method, outcome and assessment, sample size, follow-up period, and covariates adjusted for in statistical analysis.

Utilizing three criteria from the Newcastle-Ottawa scale for cohort studies independently, two researchers evaluated the quality assessment of the meta-analysis (25). Scores of 0–3, 4–6, and 7–9 were assigned for low, moderate, and high quality of studies, respectively. By discussing with a third investigator, disagreements on quality assessment were resolved.

RR is regarded as a typical measure of the relationship between tooth loss and the risk of dementia and cognitive decline. We preferentially pooled multivariable adjusted risk estimates when such estimates were reported. We combined the unadjusted estimates in the absence of an adjusted analysis. For studies that reported outcomes separately for different age groups, education levels, dental conditions, or dementia types, we combined the estimates using a fixed-effects model to obtain an overall estimate that was combined in the primary meta-analysis (26). To examine the source of heterogeneity in the primary outcomes, subgroup analyses were conducted, stratified by geographic location, sex, use of dentures, number of teeth, edentulous status, methods of dental assessment, and duration of follow-up. In a sensitive analysis to assess robustness, we performed a leave-one-out analysis to observe the magnitude of the effect on the pooled RR of each study (24).

The I² statistic was used to assess the statistical heterogeneity among studies, with cut-off values of 25%, 50%, and 75%, respectively, designating low, moderate, and high levels of heterogeneity (26). Potential publication bias was evaluated with a funnel plot, the Begg's test (27), and the Egger's test. Stata statistical software (version 12.0; College Station, TX, USA) was used to perform all statistical analyses. All tests were two-sided, and a statistically significant result was defined as P < 0.05.

Patients and public were not involved in this study.

Figure 1 illustrates the study selection process. First, 38,521 articles were reviewed. Of these, 27,552 duplicate articles and 10,911 articles that were ineligible for this review after screening based on title and abstract were removed. Of the remaining 58 articles for full-text review, 31 were excluded based on a cross-sectional design (n = 19) or a case-control design (n = 12). Four studies did not provide sufficient data to calculate, and two studies that had dementia as a risk factor and tooth loss as outcomes were removed. Thus, 21 eligible cohort studies were included in this meta-analysis.

The 21 included studies (9, 22, 23, 25, 28–44) were published between 2001 (28) and 2022 (43, 44), and characteristics of them are shown in Supplementary Table 1. The sample sizes of the cohorts ranged from 133 (30) to 209,806 (22) with a total of 356,297 and the average of 16,967. The length of follow-up duration ranged from 2 (29) to 20 (38) years in our study, with an average of 8.6 years. Among the 14 studies that reported dementia as outcome (9, 22, 23, 29–31, 33–35, 38, 39, 41–43), six studies also reported risk of tooth loss for AD (23, 29, 30, 39, 41, 42), and three studies reported VaD (23, 39, 41), and one reported comorbidity of AD and VaD (23). A various of dementia assessments were applied, including self- or professional-administrated questionnaires or interviews, medical records, and insurance records. A series of tests were also used, and the Mini-mental State Examination (MMSE) was applied in 11 included studies (25, 29, 30, 32, 33, 35–37, 39, 41, 44) as the most frequently used test, while other tests were only used in several studies. Eight included studies reported results with cognitive decline as outcome (25, 28, 32, 35–37, 40, 44), among which one study reported risks of tooth loss for dementia as well (35). Among these studies, ten studies were from Asia (22, 23, 25, 28, 29, 34, 41–44), six studies from North America (30–33, 36, 37), four studies from Europe (33, 38–40), and one study was conducted in 20 countries including Australasia, Asia, Europe and North America (35). Only four studies reported results for male (32, 33, 43), five studies independently reported results for female (30, 33, 39, 43), and the 13 remaining studies (9, 22, 23, 25, 28, 29, 31, 34–38, 40, 41, 44) did not report results by sex.

Supplementary Table 2 shows the results of the quality assessment. The median quality assessment score for the included cohort studies was 8 [range, 5 (31)−9 (22, 25, 28, 29, 41)]. The interobserver agreement (κ) between the two investigators was 0.95.

We excluded a single study and pooled the results of the remaining studies. In the sensitivity analysis between tooth loss and dementia risk, none of the individual studies had a significant effect on the pooled RR, which ranged from 1.15 (95% CI: 1.07–1.23) to 1.20 (95% CI: 1.11–1.29) when omitting individual studies individually. In the sensitivity analysis of tooth loss with cognitive decline, the pooled RR ranged from 1.17 (95% CI: 1.09–1.25) to 1.22 (95% CI: 1.11–1.34) when omitting individual studies individually. In summary, the results of the quantitative synthesis based on the sensitivity analysis were stable.

The possibility of publication bias was assessed using a funnel plot. For the association between tooth loss and risk of dementia or cognitive decline, visual inspection of the funnel plots revealed no significant asymmetry (Figure 3). The Begg's and Egger's tests also showed no significant publication bias in the assessment of tooth loss and risk of dementia (P from Begg's test = 0.83, P from Egger's test = 0.53) or cognitive decline (P from Begg's test = 0.39, P from Egger's test = 0.11).

Our findings showed a positive correlation between tooth loss and dementia incidence (RR = 1.15; 95% CI: 1.10–1.20), which was not as strong as that from the previous meta-analysis (HR or RR of the included cohort studies between 1.22 and 2.07) (8, 16–19). Compared with these earlier studies, this meta-analysis included more cohort studies and a larger sample size, which significantly increased the statistical power to detect potential associations between tooth loss and dementia risk. The previous meta-analysis by Shen et al. (16) was based on 11 published articles with six cohort studies, three cross-control studies, and two cross-control studies. A previous review performed by Fang et al. (8) included seven cohort studies, eight cross-sectional studies, and two cross-control studies. The reviews by Cerutti-Kopplin et al. (17), Chen et al. (18), and Oh et al. (19) only included 8, 8, and 11 cohort studies, respectively. However, our review included 21 cohort studies with a large number of sample size (a total of 356,297) with a long follow-up period (8.6 years on average), which could decrease the sampling error to a great extent. Moreover, apart from dementia, our study also explored the relationship between tooth loss and the risk of cognitive decline and two common subtypes of dementia, AD and VaD. Additionally, the current meta-analysis investigated a more detailed subgroup analysis, including area, sex, dementia type, use of dentures, number of teeth and edentulous status, dental assessment methods, and follow-up duration of the associations between tooth loss and various cognitive disorders, including all types of dementia, cognitive decline, and dementia due to AD (see Table 1 and Supplementary Table 3).

More importantly, some valuable findings were observed. Our meta-analysis showed a significant association between an increased risk of dementia and cognitive decline in Asian populations with tooth loss, whereas the risks were lower or not statistically significant in people from Europe and North America with missing teeth. The World Health Organization's Global Burden of Disease Study 2019 (GBD 2019) anticipated an increasing trend in dementia prevalence due to population growth and aging, and found wide geographic heterogeneity in the projected increase in dementia cases across countries and regions (45). However, the present studies have been conducted in only a few countries and regions. Few studies were conducted in Africa and the Middle East, where the number of predicted dementia cases was greatest. For both dementia and cognitive decline, the risk of tooth loss was higher in people who did not wear dentures than in those with dentures. In subgroup analysis of the number of remaining natural teeth, the association between tooth loss and risk of dementia, cognitive decline, or AD was higher in the edentulous population than in the population with natural teeth, while higher AD and risk of all types of dementia were observed in people with only nine or fewer teeth than in people with 10–19 teeth. The results suggested a dose-response relationship between the number of natural teeth and the risk of dementia risk; however, further dose-response analyses are needed. In a quick summary, Daly et al. (46) noted that current epidemiologic research is generally based on self-report measures of oral health. Considering the low knowledge and awareness of oral health in the public, this would lead to a considerable risk of error and heterogeneity across studies.

Dementia is a chronic disease and has a longer prodromal period than other diseases. However, there is no firm evidence on the duration from exposure to tooth loss to the onset of dementia/cognitive decline, and there are no guidelines or instructions for a standard follow-up duration for research on this topic. The World Health Organization's Global Burden of Disease Study 2019 showed that the prevalence of dementia is increasing “doubling about every 5 years until 85 years of age in both 2019 and 2050” (45), so we divided the follow-up duration into <5 and ≥5 years. Most of the included studies (17/21) had a follow-up of more than 5 years, and their pooled RRs for all outcomes had narrow CI, allowing the results to provide precise estimates of population values. Some studies suggested a follow-up of 10 years (47, 48), so these risk factors in earlier life could be identified, patients in earlier stages of dementia be diagnosed, and reverse causation be minimized, especially in studies, except prospective cohorts. In our meta-analysis, more than half (11/21) of the included studies were followed for <10 years or more (at least 2 years). Future research should investigate the causality with a sufficiently long follow-up period.

Dementia is a disease that results in diminishing language abilities, judgement, memory loss, and behavioral abnormalities due to damage to cerebral neurons caused by progressive, degenerative, and cerebrovascular disease (49, 50). Although recent research suggests a link between poor dental health and accelerated dementia and cognitive decline, there are currently no clear treatments for dementia, and the mechanism by which cognitive impairment is associated with tooth loss remains unknown. Therefore, it is important to identify modifiable risk factors that support dementia prevention.

Several possible mechanisms have been proposed to explain the effects of tooth loss on cognitive function (51–58). Nutrition, inflammation, and neural feedback are the most commonly suggested factors (59). Theories of nutrition and neural feedback are related to masticatory dysfunction caused by tooth loss. Tooth loss means decreased chewing function, which leads to poorer digestive capacity and appetite, and then decreased nutrition, which ultimately damages brain functions such as cognition. Deficiency of several nutrients have been found to be related to the risk of cognitive decline and dementia, like vitamin D deficiency (60, 61) and iron deficiency (62). While higher intake of some nutrients and dietary interventions also seems to be associated with lower risk of cognitive dysfunction (63, 64). The causally link of vitamin D deficiency to dementia was observed, even after excluding the influence of genetic factors by using Mendelian randomization study design (65, 66). The metabolic, endocrine, immune functions and neuronal effects of 1 alpha, 25-dihydroxyvitamin D (1, 25(OH)2D) would be considered an essential contributor to the development of cognitive disorders (67, 68). Tooth loss (10–15) and vitamin D deficiency (61, 69–71) are both found to be related to many systemic health problems, and which further indicates an important role of low 1, 25(OH)2D concentration in the relationship between tooth loss and systemic diseases including cognitive damage. Low-oxygen supply (72) and brain iron dyshomeostasis (73) may explain the iron-related neuropathogenesis of cognitive impairment or dementia. This may also partly explain why evidence-based guidelines recommend dietary interventions to reduce the risk of cognitive decline and dementia (59). However, few interventional studies have proved causality between the deficiency of vitamin D and iron and the risk of dementia. Morever, attention should also be paid to the likely of prolonged supplementation or overload of vitamin D and iron (73, 74).

Besides, fewer interocclusal contacts related to decreased masticatory function indicate impaired proprioception and less somatosensory feedback, and such changes would lead to cognitive decline and dementia (75). Studies in animals and humans have shown the effects of dysfunction of the masticatory apparatus on cognitive damage and the reverse effect of masticatory activities on cognitive processes, indicating preventive and curative effects of chewing on cognitive performance (76–79). The number of teeth remaining rather than lost, e.g., 16, 20, or whatever, was used for the classification because they were thought to be sufficient for mastication. Number 20 without a third molar is usually recommended as the cut-off (23, 28, 34, 39, 41), since it is at least one posterior occlusal unit, the most important factor affecting a person's ability to chew (80). In fact, the number of lost teeth were assessed only in four out of the 18 involved studies (22, 29, 32, 33) in our meta-analysis. A 2018 Australian cross-sectional study found that the number of teeth (<20) was associated with a lower ability to chew hard foods and discomfort with eating, both of which were related to cognitive status, suggesting a chain of tooth loss, chewing ability, and cognitive damage (81). A cross-national study of three large cohorts in Europe and New Zealand confirmed in 2019 that lack of intact teeth and difficulty chewing are predictive of poor cognition and three other aspects of overall health in older people (82). However, there are some controversial results on whether tooth loss explains the association between cognition and mastication. Lexomboon et al. (83) conducted a study based on a nationally representative sample of an elderly Swedish population and reported that the likelihood of cognitive impairment, adjusted for sex, age, and education, was not significantly different between elderly people with and without multiple tooth loss, but remained significantly higher in those with difficulty chewing. The study suggests that difficulty chewing, not tooth loss, is an independent risk factor for cognitive impairment. Further evidence from studies based on a cohort design across different geographic areas and with adequate follow-up time is needed to confirm the relationship between tooth loss, chewing ability, and cognition.

Inflammation is one of the most commonly proposed theory to explain the effects of tooth loss on cognitive function. The theory is that without the protection of the teeth that results from tooth loss, the gingival tissue cannot protect the central nervous system (CNS) from constant exposure to pathogenic oral bacteria that can lead to or exacerbate inflammatory processes, causing endovascular inflammatory mediators and neurodegeneration. Thus, cognitive deficits and dementia may precede inflammation in the CNS (30, 84, 85). Periodontitis may also be related to cardiovascular and cerebrovascular disease, which are associated with the onset of dementia (10, 53, 86). This inflammation theory is also supported by epidemiological studies, which report a higher risk and severity (87, 88), clinical studies that observed bacterial evidence in patients with dementia and periodontitis (89), and animal experiments (90). Excessive nitric oxide levels (NO) are associated with excessive cell death and inflammation. Pang et al. (90) compared the effects of tooth loss with those of chronic cerebral ischaemia on cognitive function in Wistar rats and found that the effects were similar; NO, inducible nitric oxide (iNOS), and endothelial nitric oxide synthase (eNOS) in the hippocampus were involved in both cases. Some of the involved studies in our meta-analysis also paid attention to the problem of periodontitis and dental caries and found an increased risk for cognitive damage (22, 23). However, few studies have examined the relationship between the incidence of dementia and the number of teeth lost after accounting for other dental problems and/or behaviors, such as chewing difficulties and periodontitis (22, 41). It would be prudent for future studies to consider potential confounding or interactions with other dental problems to demonstrate an independent causative role of tooth loss on the risk cognitive impairment.

Further, poor cognitive status may also precede tooth loss. People having dementia are usually observed to have poor oral conditions and tooth loss, even in the early stage of the disease (91). Foley et al. (92) have conducted and demonstrated the effect of tooth loss on cognitive status. It may be caused by impaired cognitive function, memory, and physical ability. The risk of tooth loss resulting from poor cognitive function should also be paid attention to when discussing the relationship between tooth loss and cognitive impairment, i.e., the bi-direction association between them.

The strength of our study is that it includes twice as many studies as previous reviews, with a large sample size of up to 35 thousand and an average follow-up time of 8 years. This makes our findings more reliable and generalisable. Additionally, this study demonstrates associations between tooth loss and various cognitive disorders, including subclinical cognitive decline, AD, VaD, and dementia of all types. Finally, we conducted a series of subgroup analyses focusing on the baseline characteristics of the subjects (sex and region), study design (follow-up period and dental assessment method), and type of cause and outcome, which could provide a wealth of insights for research on the relationship between dental health and cognitive status. The possible confounding effects of age on the relationship between tooth loss and cognitive decline had been also dealt with, as age was used as a covariate in most of the included studies (19/21, see Supplementary Table 1 “Characteristics of included studies in the meta-analysis”).

In interpreting this meta-analysis, the following limitations must be considered: first, the included studies used different methods and criteria to measure tooth loss and outcomes. For example, the number of teeth in some studies was self-reported by participants and not confirmed by a dentist or dental records, which could affect the accuracy of our results. Although we attempted to pool adjusted risk estimates, adjusted confounders were not available or varied widely among included studies. Although the mechanisms and risk factors of dementia and cognitive decline are far from clear, the incidence of dementia increases with age and varies between men and women. The evidence supporting 12 modifiable dementia risk factors was emphasized by the Lancet Commission on Dementia Prevention, Intervention, and Care (93). However, some of the above 12 potentially important confounders, such as sex, education, diet, smoking, hypertension, depression, dental health behaviors, and status, were neither fully adjusted for nor used as stratifying factors in some of the included studies, which may have influenced the assessment of the results. Third, the time of occurrence of tooth loss and the length of the edentulous period cannot be precisely determined, which may introduce bias in the incomparability of results to some extent (19).