Exploring the impact of mental health conditions on vaccine uptake in high-income countries: a systematic review

The studies of interest in this systematic review were those that included any type of population who has access to a routinely delivered vaccination programme in a high-income country, as defined by the World Bank [12], of any age, or population subgroup. This also includes groups such as refugees and homeless people in cases with a vaccination programme for those groups in place.

We included studies covering a broad range of mental health issues, such as diagnoses covered in the 10^th edition of the International Classification of Diseases (ICD-10) (e.g., major depressive disorder, autism spectrum disorder, etc.) or symptom-based reports (e.g., items from the SF-12 [13] or PROMIS Global-10 [14], etc.) and severe psychological distress.

Although we included studies with alcohol dependence or abuse defined as such by the study authors, we did not include studies reporting smoking and drinking behaviours per se because definitions of abnormal substance use are highly influenced by context and social norms. Furthermore, we did not include studies covering potential risk factors for mental health issues such as family stress and including life-circumstances such as prison detention which as usually closely linked to mental health issues.

For our analysis, we grouped the mental health issues into depressive disorders, anxiety (including mixed anxiety and depression), substance use disorder and alcohol abuse, autism spectrum disorders and severe mental illness including bipolar disorder, schizophrenia and other forms of psychoses.

We included a control without mental health issue or the general population. In case–control studies, a vaccinated versus a non-vaccinated group should be compared with mental health as a covariate in the statistical model.

We included all studies covering the uptake and timing of a recommended vaccine for an individual or a relative with a mental health issue in comparison to a control group.

Every type of observational study published in a peer-reviewed journal was eligible for inclusion, e.g., cohort studies, cross-sectional studies, case–control studies. Conference abstracts and PhD theses were also eligible if they contained enough information which was the only deviation from the PROSPERO protocol. We made this decision during the search and screening process as several of these studies added value to the broad range of study populations and mental health conditions and seemed of comparable study quality to other studies published in journals.

Any type of relative measure of effect or enough information to calculate a relative measure of effect was sufficient for inclusion. There were no constraints to language of the studies.

We searched Embase, MEDLINE and PsycINFO via Ovid, CINAHL via EBESCO, the Cochrane Library, Scopus and Open grey for studies published from 1994 up to the 26/03/2021. The study inclusion was limited to 1994 and later, as the publication of the Diagnostic and Statistical Manual of Mental Disorders IV (DSM-IV) in 1994 led to a significant change in how mental disorders were defined and diagnosed [15].

The search terms were based on terms covering all mental health issues in the 10^th edition of the International Classification of Diseases (ICD-10) and other systematic reviews of mental health issues [10, 16]. The search terms for vaccine uptake were based on search terms used in Jain et al. [17] expanded to cover all vaccines given routinely in high income countries. Terms for preventive health and childcare visits were also included to detect vaccination as part of a preventive care appointment.

The search terms for every database can be found in the supplementary material. Backwards citation tracking was conducted for all included papers and encountered systematic reviews on the topic.

Duplicates were removed using a R-script written by AS (https://​github.​com/​Eyedeet/​Systematic_​review/​tree/​main) which uses the revtools R-package [18]. The script removes duplicates in a first step using string matching of the digital object identifiers. In a second step, the titles of the papers are compared using a string distance algorithm and a fuzzy string-matching algorithm. AS tested her deduplication algorithm including different sensitivity parameters of the string distance algorithm and the fuzzy string algorithm against a conventional deduplication algorithm as used in the free software Mendeley.

Every study’s title and abstract were screened by two authors independently (OOA & AS). In case of a disagreement, the study was included into a full-text screening which was conducted by the same two reviewers blinded to each other’s decision. Any difference was resolved through discussion by OOA and AS. The data extraction was conducted by AS using a standardized table. A list of extracted items can be found in the supplementary material. If a study had more than one estimate, we extracted the estimate adjusted for the largest number of confounders and the crude estimate. If the estimates were for different types of mental health issues or different types of vaccination, all estimates were extracted.

The studies were grouped by age (adults over 65, general adult population, mothers and their children, and children), by different mental health issue and by type of vaccine.

A tool for assessing the risk of bias was developed based on the ROBINS-I [19] tool and was adapted according to the topic of the review. Detailed decision criteria for each type of bias can be found in the supplement (see Table S.1). The risk of bias assessment was conducted by AS and HM in parallel and blinded to each other’s rating for five studies. As there were no disagreement for the piloted studies, AS continued with the rating of the remaining studies.

Additionally, a funnel plot was used to graphically explore potential including reporting bias [20, 21].

Crude effect sizes in form of odds ratios and confidence intervals were calculated for all studies which did not provide any adjusted measure of effect based on their proportions of vaccinated individuals in the exposed and control groups. For this purpose, the sample sizes of Howard et al. [22] and Lawrence et al. [23] had to be approximated as it was unclear whether the grouping into mental health issues was mutually exclusive. Studies could present effect estimates for different vaccines and different mental health issues. In order to make different studies better comparable, we averaged the uptake over 3 years for each type of vaccine and mental health condition for the study by Browne et al. [24]. For the Howard et al. [22] paper, we derived the confidence intervals from the given p-values.

All results were analysed and presented by the underlying study population grouped by age group (e.g., older adults, adults, children, and mother and children), the underlying mental health issue and type of vaccine. We also explored potentially different outcomes for different definitions of being vaccinated, e.g., vaccinated in the last year or ever vaccinated. If a study had more than one effect estimate for different vaccines or mental health conditions, all of them were presented separately. In studies adjusting for different confounders, we presented the effect estimate adjusting for the maximum number of confounders.

Heterogeneity, clinical and statistical, was deemed too high to reasonably apply meta-analytic methods. A grade approach was used to present the certainty of the evidence.

A total of 4,069 titles were identified of which 64 were included in the full-text screening resulting in 23 eligible studies published between 2000 and 2021 (see Fig. 1). One of the 23 included studies [25] was identified through backwards citation tracking.

About half of the eligible studies (n = 12) were cross-sectional, and 11 of the studies were cohort studies. Studies covered a range of population groups with 12 studies looking at adults over 50 and older, one at the general adult population, four studies at children with mental health issues and six studies exploring the impact of mental health issues in mothers on vaccine uptake in their children. There were 14 studies conducted in the United States, four in the United Kingdom, one single multi-national study in 10 different European countries, and one study each in Australia, Canada, and Denmark. Many studies included more than one mental health issue as exposure and allowed their subjects to fall under more than one category. Depression was the most common mental health issue explored (n = 12), followed by autism spectrum disorder (n = 4), anxiety (n = 4) and substance use disorder including alcoholism (n = 4), two studies using general mental health scores, two studies on psychological distress, two studies on dementia, and one study each on eating disorders, psychosis, learning disability, post-traumatic stress disorder. The studies focused on vaccines either given at older age or during childhood: 12 studies looked at influenza vaccine uptake, 4 at pneumococcal vaccine, 10 at different childhood immunizations and one study at tetanus vaccine. The definition and time period of being vaccinated varied between studies as national vaccination schedules differed.

An overview of all included studies and more details about the study population can be found in the in Table 1.

According to Table 2, most studies showed a high risk of selection bias. This was due to selection of subgroups of populations who were already actively engaging with the health care system or covered by a certain health care insurance in the United States.

Potential confounding bias may have resulted from a lack of adjustment in some studies which were not designed for linking mental health with vaccine uptake but were part of surveillance.

Some of the studies from the United States did not provide sufficient information on whether a vaccine was covered by the respective health insurance or looked at Medicare use only which could still require co-payment for some vaccines [25].

A few studies were adjusting for factors closely linked to health care seeking behaviour such as health insurance and health care utilisation which might have resulted in collider bias in those studies. This means that it was adjusted for something which is a consequence of both outcome and exposure, which might result in a distorted association [36]. Whilst most studies on electronic health records provided relatively reliable information on vaccine uptake, those were more likely to suffer from moderate biases towards the diagnosis of health care issues in primary care. The majority of cross-sectional studies assessed all participants for mental health issues using questionnaires but collected data about vaccine receipt by participant self-recall and were consequently potentially biased by differential misclassification of vaccination status.

A graphical bias exploration using a funnel plot (see supplementary Figures S.5, S.6, S.7) indicated some asymmetry. We interpreted this finding as being attributable to the high heterogeneity between the studies caused by very different study populations, different ways of assessing mental health and vaccine receipt.

Overall, a wide variety of different subgroups of the population was covered in the studies. These ranged from elderly adults, general adult population, children and mothers with their children. All effect estimates can be found in the supplement material ordered by the different subgroups (see supplementary Figures S.1, S.2, S.3, S.4).

Expecting different mechanisms of how a mental health issue could impact and influence vaccine uptake in different subgroups, we also explored results grouped by different mental health issues.

Despite the heterogeneity across different studies, several studies found substantial health inequalities for individuals with mental health issues. Osam et al. [31] found a substantially lower vaccine uptake in children of mothers with substance use disorder, similarly adults over 65 with substance use disorder showed a significantly lower influenza uptake in comparison to healthy controls [24, 37]. Some studies found a drastically lower vaccine uptake in people with depression or their children [26, 31].

During childhood, there was no consistent trend in differences of uptake noticeable for different types of vaccines. This also applied to different vaccines for adults of all ages.

Only four studies explored the effect of mental health issues on vaccine delay or being partially vaccinated [29, 33, 34]. Turner et al. [32] found higher odds of delaying vaccines for children of mothers with mental health issues but not for eventual coverage later. In contrast, Gilbert et al. [29] found that women with lower self-rated mental health status were less likely to have children with up-to-date vaccinations but no difference in delaying vaccination. One study indicated higher odds of being only partially vaccinated for children with learning disability [33] and another study showed both, a lower uptake and a higher proportion of delayed vaccines in younger siblings of children with autism [34].