The proposed systematic review will be conducted and reported in accordance with the reporting guidance provided in the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) statement [
22] and the Meta-analysis Of Observational Studies in Epidemiology (MOOSE) reporting guideline [
23]. This systematic review protocol is registered with the International Prospective Register of Systematic Reviews (PROSPERO) database (registration number: CRD42017059955) and has been reported in accordance with the PRISMA for Protocols (PRISMA-P) statement [
24,
25] (see Additional file
1 for the checklist).
Eligibility criteria
Studies will be selected according to the following criteria: study design and participants (exposure), comparator(s) or control group, outcome(s) of interest, and setting.
Study design and participants of interest
Eligible studies will be observational studies reporting study-specific data for all-cause mortality or cause-specific mortality in children, adolescents, or adults with ASD or ADHD and/or their first-degree relatives. Case-control studies, prospective cohort studies with a reference comparator, and retrospective cohort studies with a reference comparator (also known as historical cohort studies) will be included. Randomized controlled trials will be unavailable for our research question. For participants (index subjects will be patients or first-degree relatives), we will use investigator-reported definitions of ASD or ADHD according to the accepted diagnostic criteria (e.g., ninth or tenth revisions of the ICD coding system or the third, fourth, or fifth edition of the Diagnostic and Statistical Manual of Mental Disorders [DSM] criteria): ASD (ICD-9: 299.0, 299.8; ICD-10: F84) and ADHD (ICD-9: 314.00, 314.01; ICD-10: F90). We will exclude studies in which ASD or ADHD is not the exposure of interest and all-cause or cause-specific mortality is not the outcome of interest. Observational studies not presenting study-specific data (e.g., relative risks, 95% confidence intervals, numbers of cases/population, observed, and expected cases) or sufficient data for an outcome measure to be calculated (see the “
Outcomes” section below) will be also excluded.
Comparator(s) or control group
The comparator group will be based on subjects with no history of ASD or ADHD (e.g., the general population, the community, unexposed outpatient or hospital-based controls, parents or siblings of people without ASD or ADHD).
Outcome(s)
The primary outcome will be all-cause mortality based on valid information sources (e.g., national vital registration system, hospital records, death certificates, verbal autopsy standards). Effect measures will include the maximally adjusted effects reported as the standardized mortality ratio (SMR), the relative risk (RR), the odds ratio (OR), and the hazard ratio (HR) for the risk of mortality between ASD and ADHD compared with controls. The secondary outcome measure will be cause-specific mortality (according to the main causes of death grouped into the ICD categories of diagnoses).
Setting
There will be no restriction by study setting.
From each eligible report, two reviewers will extract information independently on first author, year of publication, neurodevelopmental disorder (ADHD or ASD), index subjects (patients, parents of patients, or siblings of patients), the general characteristics of participants (age, gender, ethnicity, social status, and comorbidities such as intellectual disability, substance abuse, epilepsy, and other medical or psychiatry conditions), epidemiological design (cohort or case-control, prospective, or retrospective), the country of study, the follow-up period, the setting (mixed, inpatient, outpatient, or community), the sample size, the outcomes assessed (including definitions and confounding factors that were taken into consideration), the number of cases and controls (in case-control studies) or the number of cases and population participants (in cohort studies), and/or the maximally adjusted effect measure (e.g., relative risk, hazard ratio, or standardized mortality ratio) with 95% confidence intervals. We will use pre-designed forms that will be piloted initially on a small number of included studies. We will also attempt to contact authors of primary publications and/or collaborators for missing outcome data or unclear information in order to maximize available data.
Data synthesis
We will calculate summary effect estimates comparing the mortality in people with ADHD or ASD with the mortality in the reference control group. Statistical combination of data from two or more observational studies in a meta-analysis will be conducted and reported separately for ADHD and ASD. We will estimate the summary effect size and its 95% confidence interval using the inverse variance method based on the DerSimonian and Laird random-effects model [
27]. The random-effects model is selected a priori to synthesize the epidemiological data, as it considers both within-study and between-study variation by incorporating the heterogeneity of effects into the overall analyses. If primary studies report results separately for men and women (or other subgroups), we will combine the subgroup-specific estimates using a fixed-effects model to generate an estimate for both subgroups combined so that each study was represented only once in the analyses. Patient studies with mortality data will be included in the main meta-analyses and first-degree relative studies separately in different (secondary) meta-analyses. We will evaluate statistical heterogeneity by estimating the variance between studies using the Cochran’s
Q test [
28] and
I
2 statistic [
29]. The Cochran’s
Q test is obtained by the weighted sum of the squared differences of the observed effect in each study minus the fixed summary effect. The
I
2 statistic is the ratio of variance between studies over the sum of the variances within and between studies and ranges between 0 and 100% (with values of 0–25% taken to indicate low heterogeneity and 75–100% to indicate considerable heterogeneity). We will also estimate the 95% prediction interval [
30] where appropriate, which further accounts for between-study heterogeneity and evaluates the uncertainty for the effect that would be expected in a new observational study addressing that same association. Small study effects will be investigated by visual inspection of funnel plots (where appropriate) and formally tested using Egger’s [
31] and Begg’s test [
32], with the results considered to indicate potential small study effects when
P < 0.10.
If sufficient studies are identified, we will present subgroup analyses to attempt to explain any potential observed between-study heterogeneity. The potential moderators (covariates) considered will be gender (male or female), diagnostic criteria used (e.g., DSM or ICD and their versions), age group at first diagnosis (children and adolescents [< 18 years old] versus adults [≥ 18 years old]), ethnicity (e.g., white or non-white), number of comorbidities (e.g., 0, 1, 2, or ≥3), medical or neuropsychiatric comorbidity (e.g., presence or absence of a comorbid disorder with ASD or ADHD), index subjects (patients, parents of patients, or siblings of patients), country or geographic region, cohort/sample size (< 500, 500–1000, or > 1000 participants), setting (mixed, inpatient, outpatient, or community/general population), population-based (yes or no), follow-up period (0–1, > 1–5, or > 5 years), year of publication as a proxy of changes in clinical practice over time (before 2000 or in 2000 and after), study quality (low or high-moderate quality), and effect measure adjusted for potential confounders (age, sex, or other). We will also explore the contributions from natural and unnatural causes of death. Unnatural deaths will be defined on the basis of ICD codes (ICD-10: V01-Y98 or ICD-9: E800-999); the remaining deaths will be classified as natural.
To further assess the consistency of evidence over time, we will conduct cumulative meta-analyses in the order of publication year [
33,
34].
Finally, we will apply a set of criteria to conclude whether the evidence may be considered convincing, probable, limited-suggestive, limited-not conclusive, or unlikely. For this, we will follow the Global Burden of Disease Study approach for risk factors [
35‐
37] which is also being applied in ongoing evidence syntheses of central nervous system disorders and cancer [
38,
39]. “Convincing evidence” consists of biologically plausible associations between exposure and outcome established from multiple epidemiological studies in different populations. Evidentiary studies must be substantial, include prospective observational studies and, where relevant, epidemiological studies of sufficient size, duration, and quality, and show consistent effects. A convincing relationship should be robust enough to be highly unlikely to be modified in the foreseeable future as new evidence accumulates. “Probable evidence” is similarly based on epidemiological studies with consistent associations between exposure and outcome but shortcomings in the evidence exist, such as insufficient trials or prospective observational studies available. “Limited-suggestive evidence” represents too limited evidence to permit a probable or convincing causal judgment, but where there is evidence suggestive of a direction of effect. “Limited-not conclusive evidence” consists of information that is so limited that no firm conclusion can be made for a number of reasons (e.g., the evidence might be limited by the amount of evidence in terms of the number of studies available, by inconsistency of direction of effect, by poor quality of studies, or by any combination of these factors). “Substantial effect on risk unlikely” consists of evidence strong enough to support a judgment that a particular exposure is unlikely to have a substantial causal relation to a cancer outcome. The evidence should be robust enough to be unlikely to be modified in the foreseeable future as new evidence accumulates [
38,
39].
We will also use the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) methodology for evaluating the quality of evidence [
30‐
42] as follows: high-quality evidence (further research is very unlikely to change our confidence in the estimate of effect), moderate-quality evidence (further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate), low-quality evidence (further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate), or very low-quality evidence (very uncertain about the estimate of effect).