The Prevalence of Diabetes in Autistic Persons: A Systematic Review

2.1. Study Selection

Studies were included provided they satisfied all the following criteria: (a) investigation of the prevalence of DM (either or both of T1DM and T2DM) in individuals with autism (b) published in English language. Exclusion criteria included conference proceedings. No specific limitations were imposed pertaining to the study duration or the study population size (though case reports and case series were excluded as they cannot reasonably be considered investigations of prevalence).

In the conduction of the review, several studies were identified that measured the association of DM with autism via the prevalence of autism in persons with DM, or the development of DM over time in autistic persons (rather than the prevalence). Such studies do not satisfy inclusion criteria, but are still discussed in the findings, within the section entitled ‘Summary of Relevant Excluded Studies.’

2.2. Systematic Search Strategy

The PubMed, MedLine, CINAHL, PsycINFO and EMBASE databases were searched from their respective inceptions from 4^th April 2020. Titles and abstracts were searched for the following terms: (autis* OR PDD* OR Kanner* OR Asperger* OR ASD* OR pervasive development* or ASC* or neurodevelopmental dis*) AND (diabet* OR hyperglyc* OR glucose* OR DM). All titles and abstracts of articles that remained, following the removal of duplicates, were screened against the inclusion criteria by two investigators (ST + RC). If papers potentially satisfied inclusion criteria, full texts were accessed. In instances of uncertainty between ST and RC regarding the eligibility of an article, the final decision was made by TB. The ancestry method was utilised to identify additional studies within the references of eligible papers. Grey literature searches included Google Scholar and manual searches. Experts in the field were consulted to identify any additional published or unpublished data; for all articles identified via expert consultation, the full texts of the articles were assessed. A separate search was conducted by the University Hospitals of Leicester library services on 28^th April 2020 to find studies not identified by the above methods. The systematic review was registered in 2019 on PROSPERO: CRD42019122176, with a subsequent revision submitted and authorised in 2020.

2.3. Data Extraction

For each article, data was extracted pertaining to the year and location of the study, as well as the number of participants, both with an autism diagnosis, and non-autistic controls, where applicable. The methods employed for such assessments were also obtained, as well as information regarding the resultant prevalence estimate for both autism groups and non-autistic controls (where applicable).

2.4. Data Synthesis

The prevalence results obtained were considered both regarding overall findings for DM prevalence across available studies, as well as specific T1DM and T2DM data where reported. Quantitative synthesis (meta-analysis) was considered inappropriate due to the high level of heterogeneity in study design, population, setting and diagnostic criteria used.

2.5. Quality Assessment

All articles that qualified for inclusion were assessed by two investigators (ST + RC) according to the 22-item (34 with subsections) STROBE checklist for cohort, case-control and cross-sectional studies [16], as all studies were of this type. Each of the STROBE items was determined as being complete (1 point), partially complete (0.5), incomplete (0 points) or not applicable [17]. The resultant STROBE score has been expressed as a fraction, as the total applicable items (the denominator) varied between studies, and a corresponding percentage value. No studies qualifying for inclusion were excluded based upon their STROBE score.

3.1. Study Characteristics

Database searches were conducted on 4^th April 2020. The database search yielded a total of n=5368 articles, including n=1535 articles in PubMed, n=1236 articles in MedLine, n=370 articles in CINAHL, n=305 articles in PsycINFO and n=1922 articles in EMBASE. Additional records (n=119) were identified through the ancestry method, grey literature searches and expert consultation techniques. Following duplicate removal, n=2879 articles remained for screening. (Fig. 1) illustrates a PRISMA flow chart summary of the systematic search.

Upon abstract screening, 2728 articles were excluded. Reasons included focusing on unrelated research topics, failing to satisfy inclusion criteria, or being not being published in English. Thus, full texts of 151 articles were assessed, of which 19 qualified for inclusion [6, 13, 19-35](Please refer to the supplementary material for a table of all full-text articles that were assessed). Of the included papers, 10 were identified on the database search [6, 19, 20, 23-28, 34], 6 via the ancestry method [13, 21, 29-31, 35], and 3 via expert consultation [22, 32, 33]. Grey literature searches and the library search failed to identify any additional studies satisfying inclusion criteria.

3.2. General Findings of Included Studies

Table 1 summarises all the eligible studies. Publication dates for the studies ranged from 2006 to 2020. Most studies were conducted in North America (n=12) [6, 13, 20, 21, 23, 25, 27, 30, 31, 34], followed by Sweden (n=2) [26, 28]. The mean participant age across studies that reported such data ranged from 12-48 years. Some studies focussed exclusively on child/adolescent or adult populations, whereas others included both. Most autistic participants in eligible studies were male, ranging from 66-82%. The proportions of autistic participants with co-occurring ID were reported in 9 eligible studies [6, 22, 24, 26-29, 31, 34], with rates ranging from 0-97%. The methodology varied across studies, including clinical/electronic record review [6, 13, 19-23, 25, 27-29, 32, 34, 35], questionnaires [24, 31], surveys [30, 33], and structured interviews [26].

Overall, of 15 studies that compared findings from autistic and non-autistic groups [6, 13, 19-21, 23, 25-28, 30, 32-35], 9 found a heightened prevalence in the autistic group [6, 13, 19, 20, 23, 26-28, 30]. However, the difference was reported as statistically significant in only 4 of these [6, 20, 23, 28]. Furthermore, the studies that found a heightened DM prevalence in autistic persons [6, 13, 19, 20, 23, 26-28, 30] had a far greater mean average study population size (Average number of autistic participants = 12,599; Average number of controls = 541,861) than those finding a greater DM prevalence in the control group [21, 25, 32 - 35]’ (Average number of autistic participants = 6,829; Average number of controls = 325,539). Additionally, the studies showing heightened DM prevalence in autistic persons [6, 13, 19, 20, 23, 26-28, 30] had a higher mean average percentage STROBE score rating than those showing a higher DM prevalence in the control group [21, 25, 32-35] (75% vs. 64%), suggesting that the reporting of these studies was of a generally higher quality.

Three studies did not provide raw data pertaining to DM prevalence [21, 24, 30], reporting only percentage values. This data was requested from the corresponding authors, but either reply were not received, or such data was unavailable. However, these studies were included, as they provided prevalence data in some form.

Of the sixteen studies that included both autistic and non-autistic groups within their respective study populations [6, 13, 19-21, 23-28, 30, 32-35], five showed a statistically significant difference between the autistic and non-autistic groups [6, 20, 23, 28, 35], four of which demonstrated a significantly increased rate of DM in the autistic group [6, 20, 23, 28], and one a significantly reduced risk of DM in this group [35].

3.3. Non-Specific DM Findings

Twelve studies did not report type-specific DM data, rather reporting a prevalence rate that likely encompasses both subtypes, though in many instances, this is not explicitly stated. Eight of these studies were based in the United States, including Croen et al. (2015) [6], who reviewed the electronic records of 1,507 autistic adults and 15,070 age- and sex-matched controls, finding a significantly (p<0.001) increased prevalence of DM in autistic people (n=114; 7.6%) relative to their non-autistic peers (n=653; 4.3%). In contrast, Vohra et al. (2017) [35] reported a significantly (p<0.05) reduced DM prevalence in autistic individuals (3.6%; 63/1772) compared to non-autistic controls (4.7%; 250/5320). They suggested that possible reasons for the stark differences in findings compared to Croen et al. (2015) [6] could be the different demographics of different populations studied (Northern California for Croen et al. (2015) [6] vs. Illinois, New York and Texas for Vohra et al. (2017) [35]), the Croen et al. (2015) [6] data being more recent and thus their heightened autism awareness with improved access to associated services, and Vohra et al. (2017) [35] requiring at least two outpatient (or one inpatient) insurance claims in order to code comorbidity.

Other United States-based studies adopting electronic records approach include a study by Tyler et al. (2011) [34], who found no significant difference in DM prevalence between 108 autistic adults (n=7; 6.5%) and 206 non-autistic controls matched for age, sex and insurance status (n=16; 7.8%). However, as with many such studies, autistic persons were identified via receiving services through their local healthcare system (in this case, the Cleveland Clinic). Thus, autistic adults not utilising such services are not identified by such approach. Davignon and colleagues (2018) [27] similarly used electronic records to investigate co-occurring physical and psychiatric conditions in autistic children and young adults, comparing them with peer groups with Attention Deficit Hyperactivity Disorder (ADHD), DM, and controls with none of three conditions. DM prevalence rates of 0.6% (25/4123) and 0.5% (107/20,615) were reported for the autistic and ADHD groups, respectively (the control group did not have DM by definition, so could not be meaningfully compared).

McDermott and colleagues (2007) [32] used primary care records to compare individuals with disabilities (including autism, ID, cerebral palsy, spinal cord injury, chronic mental illnesses, visual impairment and hearing impairment) with non-disabled persons. Out of the autistic subgroup (n=45), 0 had co-occurring DM, compared to 15.5% (202/1303) of the overall disability group and 14.5% (265/1828) of the non-disabled cohort. This low prevalence finding may be somewhat attributable to the small size of the autistic cohort, much like the United Kingdom-based survey by McManus et al. (2014) [33], where no autistic individuals with co-occurring DM were identified. Still, the autistic cohort size was similarly small (n=19). However, in contrast, Jones et al. (2015) [31] identified a DM prevalence of 9.8% (n=9) from a cohort of 92 autistic adults, who were interviewed as part of a 25-year outcome study of adult autism.

In Sweden, a large electronic database study by Flygare Wallen et al. (2018) [28], demonstrated a significantly heightened risk of DM in both autistic people as well as those with ID, after adjusting for age. However, the quoted prevalence rate of 2.87% (399/13,921) is likely an overestimate relative to the general Swedish autistic population, as the study inclusion criteria stipulated that participants had to have at least one recorded diagnosis of either DM, hypertension or obesity. Focussing exclusively on Swedish children, Alabaf et al. (2018) [26] interviewed the parents of twins born between 31^st June 2002 and 31^st December 2006, including 301 autistic children and 22,028 controls with no autism, ID or ADHD. They found a slightly heightened prevalence of DM among autistic children relative to controls (0.66%; 91/22028 vs. 0.4%; 2/301), though this difference was not statistically significant, and both diabetic autistic individuals also had co-occurring ID. Additionally, the different approach to data collection (database vs. parental interview) introduces a further source of bias when making a comparison with Flygare Wallen et al. (2018) [28]. In the United States, Gurney et al. (2006) [30] similarly interviewed parents of autistic vs. non-autistic children, similarly finding a lack of a statistically significant DM prevalence difference between the two cohorts (autistic group 0.4%; non-autistic group 0.3%).

Akobirshoev et al. (2019) [25] examined the electronic discharge records of North American-based autistic adults who had been admitted to hospital, relative to non-autistic age- and sex-matched controls. Overall, 11.5-13.6% (53-63/462) of the autistic group had DM, compared to 15.0% (145/967) of the control group; not a statistically significant result, though autistic persons were significantly (p<0.05) more likely to have DM without chronic complications. A France-based study by Guinchat et al. (2015) [29] also focussed on an inpatient setting, reporting on the medical records of 58 autistic adolescents who had been admitted to a specialist neurobehavioural unit. Interestingly, out of the entire cohort, only 1 patient had a DM diagnosis, which may be surprising given the high rates of antipsychotic prescribing in such a group (in this cohort, 56 patients had been prescribed an antipsychotic).

3.4. Subgroup Analyses

Five studies reported T1DM-specific prevalence data. In a US-based study, Kohane et al. (2012) [20] reviewed the electronic healthcare records of 14,381 autistic patients and 2,379,397 non-autistic peers. They found that autistic patients were significantly (p<0.001) more likely to have T1DM (Autistic group: n=114, 0.79%; Non-autistic group: n=8058, 0.34%). This observation is further supported by the findings of Chen and colleagues [19], who found that autistic patients were borderline significantly (p=0.056) more likely to have T1DM compared to their non-autistic counterparts (Autistic group = 0.3%; 4/1,598 vs. Non-autistic controls = 0.1%; 4/6,392). More recent US-based findings by Zerbo et al (2015) [13] found an increased T1DM prevalence in autistic patients relative to their non-autistic peers, though this difference was not statistically significant (Autistic group = 0.22%; 12/5,565 vs. Non-autistic controls = 0.19%; 52/27,825). In contrast, another study found conversely found a modestly increased risk of T1DM among non-autistic individuals (Autistic group = 0.42%; Non-autistic controls = 0.59%) [21]. In a postal questionnaire study of 186 individuals with ID, Taggart et al. (2013) [24] reported a considerably higher T1DM prevalence of 13% among the autistic subgroup. However, raw data pertaining to the size of the autistic subgroup, the number with T1DM and the subgroups age distribution was not reported or available on request, making it difficult to draw any clear conclusions from such a prevalence finding in isolation.

Only three studies reported T2DM-specific prevalence data. In a US-based retrospective review of the electronic healthcare records of 48,762 autistic children and 243,810 matched non-autistic patients [23], autistic children were significantly (p≤0.05) more likely to have T2DM relative to their non-autistic peers (Autistic group: n=515, 1.06%; Non-autistic group: n=970, 0.40%). However, both results likely represent an underestimate of T2DM prevalence for both groups, as insulin treatment was listed as an exclusion criterion for the study. Brondino et al. (2019) [22] reviewed the clinical records of 191 autistic patients based in Italy, finding a prevalence rate of 0.5% (n=1). It is difficult to draw any firm conclusions from this finding, however, considering the relatively small sample size and lack of a non-autistic comparison group. Much like for T1DM, Taggart et al. (2013) [24] also reported a considerably higher T2DM prevalence than the aforementioned studies, 5%. However, as was the case for their T1DM prevalence rate, key related data to put this percentage value in context was not reported.

3.5. Summary of Relevant Excluded Studies

The most frequent rationale for rejection of articles during full text screening included not reporting prevalence data for DM in autistic people, reporting prevalence data for autism among people with DM, or focusing on maternal or family history of DM with autism in their offspring.

Several studies estimated the prevalence of autism in people with DM, rather than the prevalence of DM in autistic people. Though not satisfying inclusion criteria for this review, these studies similarly measure the association of autism with DM. Freeman et al. (2005) [47] retrospectively reviewed the charts of 984 children attending a Canadian diabetes clinic, estimating an autism prevalence of 0.9% (9/984) in children with T1DM. The authors reported that this was greater than general population prevalence estimates [48]; however, more recent data suggests this finding aligns with the general population [3]. Further studies found equivocal results; Harjutsalo and Tuomilehto (2006) [49] observed no significant difference in autism prevalence between Finnish children with T1DM relative to the general population. Iafusco et al. (2006) [50] suggested that the autism prevalence in patients with T1DM in Italy appeared inversely associated with the incidence trends of T1DM in the geographical areas being observed.

Lemay et al. (2018) [51] investigated this relationship in a large study population (n=61,749) of German and Austrian children and adolescents with T1DM, estimating an autism prevalence of 0.24% (n=150), concluding that this finding did not suggest an elevated autism risk relative to the general population. Two similar studies based in North America, both also involving child populations, also concluded that the prevalence of T1DM in autistic children was similar to the general population [52, 53].

In contrast, Levitt Katz et al. (2005) [54] investigated the prevalence of neuropsychiatric disorders in North American T2DM patients via a retrospective chart review of 237 children. However, the prevalence of autism within this cohort was not reported, as the authors grouped autistic persons together with those with ID, who had a combined prevalence of 4% (10/237) within this cohort.

Chen et al. (2016) [55] measured the development of T2DM over time rather than prevalence per se. Using the Taiwan National Health Insurance Database, they monitored 6,122 autistic adolescents and adults, and 24,488 non-autistic age- and sex- matched peers from 2002-2009 until the end of 2011. They observed that the autistic individuals were significantly (p<0.001) more likely to develop T2DM over the study period than their non-autistic peers (Hazard ratio 3.25 and 95% CI 2.23-4.75, after adjusting for demographic data, antipsychotic use and medical comorbidities). Additionally, prescription of atypical antipsychotics, far more prevalent in the autistic cohort (n=288; 32.2%) than the non-autistic group (n=117; 0.8%), conferred an additional significantly increased risk (p<0.001) of T2DM development.

4.1. Limitations

Though this review focuses on the prevalence of DM in autistic persons, this is just one means of measuring the association between the two conditions. Measuring this association in the form of prevalence is valuable to a healthcare professional-based readership, as it can be readily translated into policy and costings [67]. However, other means of measuring the association between these conditions, such as the development of DM in autistic persons over time, as well as the prevalence of autism in persons with DM, should also be considered, and such articles have been discussed in the Summary of Relevant Excluded Studies section of this review. Another issue meriting further exploration is the prognosis of T1DM and T2DM among autistic individuals relative to their non-autistic peers. Studies could be underestimating the true association if the relative survival following a diagnosis of DM is lower in autistic persons [28]. This theory is supported by the findings of Chen et al. (2016) [55], who observed that the rate of development of T2DM in autistic persons over time was significantly (o <0.001) greater than for their non-autistic peers.

The designs of eligible studies varied widely, including surveys, case-control and cohort studies. Ideally, the association between autism and DM should be explored in large cohort studies where autism can be analysed alongside other conditions in which the prevalence of DM may differ from the general population, such as ID [68], as well as measuring other factors contributory to DM development. A further limitation of most eligible studies is that they focus on autistic individuals who have interacted with primary and/or secondary care services; thus, the DM risk in these persons may not be entirely representative of all autistic individuals. Additionally, there is a considerable heterogeneity across studies in terms of approaches to data collection, including electronic records, questionnaires, surveys and structured interview, representing a further source of bias. Furthermore, the search was limited to English language, which could have overlooked relevant evidence published in foreign language journals. However, whilst most eligible studies were North American in origin, the review nevertheless included studies from many different countries, where access to health services and diagnostic practices may differ considerably.

CONCLUSION

As with any systematic review, the quality is affected by the quality of its constituent studies. Many studies did not report data differentiating the prevalence of DM subtypes, and as such, their findings are somewhat limited in value. We would recommend that future work involving estimating the prevalence of DM, in autistic individuals or otherwise needs to collect T1 and T2 specific data. Moreover, it is essential that such studies also report other data central to DM aetiology, such as the age, ethnicity and gender distributions of their respective study populations.