Introduction

Multiple sclerosis (MS) has been described as the archetypal autoimmune disease of the central nervous system. In common with many other autoimmune diseases, MS risk appears to be influenced by both genetics and environment [1, 2]. Whilst the gene most strongly associated with MS is the HLA-DRB1*1501 MHC class II haplotype, a number of other genes, many associated with immune function, have been associated with MS in large-scale genome-wide association studies [3]. In addition, MS has been associated with vitamin D deficiency [1], which is in turn associated with a number of other autoimmune diseases [4].

It is well known that the risk of MS is increased in relatives of probands with MS, emphasizing the genetic contribution to disease [5]. The study of the risk of additional autoimmune diseases in both people with MS, and their first-degree relatives has been pursued over many years, with studies employing a variety of designs and yielding conflicting results [6].

The most recent large-scale study to attempt to address this question [7] used the Swedish National MS register together with the Swedish National Patient Register. Roshanisefat et al. [7] found no consistent evidence for an increased risk of autoimmune disease in the parents of people with MS; additionally they found that the risk of a second autoimmune disease appeared to be increased only after the diagnosis of MS [7]. This finding, which suggests that the increased risk seen in MS may be a result of the increased contact that people with MS have with health-care professionals, implies that there may be either surveillance or reporting bias underlying previous reports of an increased risk of additional diagnoses. However, the study by Roshanisefat et al. [7] is not the only one using a national database to attempt to answer the question regarding MS and autoimmune disease. National databases from Denmark [8, 9], the UK [10], California [11] and Taiwan [12] have also been employed to address this question.

There is therefore a large amount of information available examining the frequency of autoimmune disease in both people with MS and their first-degree relatives. We performed a systematic review of the frequency of selected autoimmune diseases, including (autoimmune thyroid disease, type 1 diabetes mellitus, inflammatory bowel disease, psoriasis, systemic lupus erythematosus (SLE) and rheumatoid arthritis (RA), in both people with MS and their first-degree relatives. Heterogeneity between studies was assessed, and, where possible, overall estimates of the frequency of these diseases in both people with MS and their first-degree relatives were calculated.

Methods

Inclusion criteria

Inclusion criteria were prespecified. Papers selected for inclusion were those published after 1980 which gave figures for the prevalence of specified autoimmune diseases in both MS and healthy control populations. The control population had to be matched to the MS population in terms of age and sex, or alternatively a precise local population prevalence of autoimmune disease had to be given (approximations of overall population rates were not felt to be sufficiently precise). The control population for the “relatives of MS” population could be either directly matched, or alternatively the probands matched and their families compared.

Search strategy

PubMed and Web of Science were searched using the terms “multiple sclerosis” AND “thyroid”, “multiple sclerosis” AND “diabetes”, “multiple sclerosis” AND “Crohn’s”, “multiple sclerosis” AND “ulcerative colitis”, “multiple sclerosis” AND “inflammatory bowel disease”, “multiple sclerosis” AND “psoriasis”, “multiple sclerosis” AND “lupus”, “multiple sclerosis” AND “SLE”, “multiple sclerosis” AND “rheumatoid” and “multiple sclerosis” AND “arthritis”. The resulting abstracts were hand-searched for publications meeting the inclusion criteria. The results from each search were cross-referenced, as many of the included papers examined more than one autoimmune disease.

Statistical analysis

A generic inverse variance fixed or random effects model was used for the statistical analysis as appropriate. A random effects model was applied unless I 2 was ≤25 %, in which case a fixed effects model was used [13]. Between-study heterogeneity was assessed for each calculation using Cochran’s Q Chi-squared test and I 2 [14]. Where present, heterogeneity was explored using subgroup analysis. Risks are reported as pooled odds ratios (OR) and 95 % confidence intervals (CI). Bias was assessed using visual inspection of funnel plots, and where more than ten studies were included, quantified using an Egger p value [15]. A p value of <0.05 was considered statistically significant. Analyses were conducted using RevMan 5.1 (Cochrane Information Management Systems).

Results

Included papers

Following the initial searches 254 papers and four conference abstracts were assessed in order to ascertain whether the inclusion criteria were met. All four conference abstracts were rejected, as the same cohorts were used in later published articles. 41 unique papers were initially selected for inclusion (Fig. 1). Two studies were later excluded from the analysis as the number of relatives was not given, only the number of index MS cases [16, 17]. The remaining 39 papers, details of which are given in Table 1, were used in the analysis.

Fig. 1
figure 1

Study selection

Full size image
Table 1 Included studies
Full size table

Autoimmune disease

Overall results are given in Table 2, and discussed in more detail below.

Table 2 Calculated OR for each autoimmune disease in both MS patients and their first-degree relatives
Full size table

Thyroid autoimmunity

Clinical thyroid dysfunction Hypothyroidism in iodine-replete areas is generally autoimmune in nature [18], but the same cannot be said of hyperthyroidism. Some studies have therefore used hypothyroidism as a surrogate diagnosis for autoimmune thyroid dysfunction. There were 19 studies examining thyroid function in MS [9–12, 19–33]. In five of these studies [9, 19–21, 26] there were no cases of thyroid dysfunction in either MS patients or controls, and so these could not be included in the analysis. The remaining 14 studies gave an overall increased risk of thyroid dysfunction in people with MS (OR 1.66, 95 % CI 1.35–2.05, p < 0.00001), without between-study heterogeneity (Cochran’s Q p = 0.16, I 2 = 27 %; Fig. 2a). A funnel plot demonstrated no significant publication bias (Fig. 3a), with an Egger p value of 0.76. When those studies using hypothyroidism as a marker of autoimmune thyroid disease were selected [10, 12, 22, 28, 32], there was an increased risk in people with MS (OR 1.72, 95 % CI 1.00–2.97, p = 0.05) with no heterogeneity. A similar effect was seen when only those studies specifying “autoimmune thyroid disease” were selected [23–25, 27, 31, 33] (OR 1.72, 95 % CI 1.46–2.04, p < 0.00001). When cases of Hashimoto’s thyroiditis were analysed separately [11, 25, 31], there was no increased risk in people with MS (OR 1.42, 95 % CI 0.72–2.79, p = 0.31).

Fig. 2
figure 2

Forest plots demonstrating the OR of clinical thyroid disease in (a) people with MS and (b) relatives of people with MS

Full size image
Fig. 3
figure 3

Funnel plots demonstrating the lack of publication bias when examining the frequency of thyroid disease in (a) people with MS and (b) relatives of people with MS

Full size image

Concerning thyroid function in first-degree relatives of people with MS, seven studies were identified [9, 17, 28, 30, 31, 34, 35]. One study was excluded [17] as only the number of MS index cases was given, rather than the number of relatives, and one included no patients with thyroid dysfunction [9]. There was an overall increased risk of thyroid dysfunction in first-degree relatives of people with MS (OR 2.38, 95 % CI 1.95–2.91, p < 0.00001; Fig. 2b) with no significant heterogeneity or evidence of publication bias (Fig. 3b).

Thyroid autoantibodies

Ten studies examined thyroid autoantibodies in patients with MS [19, 21, 23, 26, 29, 32, 36–39]. One study [19] did not detect any antibodies in either MS patients or controls, and was excluded from the analysis. There was an overall increased rate of thyroid autoantibodies in patients with MS compared to healthy controls (OR 2.36, 95 % CI 1.32–4.20, p = 0.004) but with significant heterogeneity (Cochran’s Q p = 0.0001, I 2 = 74 %). There was no evidence of publication bias (Egger p value = 0.56). Heterogeneity was explored by examining each thyroid autoantibody individually, but each attempt at subgroup analysis resulted in a small number of studies being examined. No studies gave data regarding the rate of thyroid autoantibodies in relatives of MS patients compared to healthy controls.

Type 1 diabetes mellitus

There were 17 studies [7, 10–12, 24, 27, 28, 30–33, 40–45] examining coexisting diabetes in in MS. One study [30] included no cases of diabetes in either people with MS or controls. There was an increased risk of diabetes associated with MS overall (OR 2.02, 95 % CI 1.22–3.40, p = 0.006; Fig. 4a). However, this was associated with significant heterogeneity when all studies were considered together (Cochran’s Q p < 0.00001, I 2 = 91 %). There was no evidence of publication bias (Egger p value = 0.16; supplementary Fig. 1a). The potential reasons for the heterogeneity seen were explored by only including those studies that specified type 1 diabetes or insulin-dependent diabetes (four studies were excluded [33, 43–45]). A study using patients with stroke as the control group was also excluded [40], given the association between stroke and diabetes. This strengthened the relationship between MS and diabetes (OR 2.69, 95 % CI 1.43–5.04), but heterogeneity remained (Cochran’s Q p < 0.00001, I 2 = 94 %). Separating studies using large databases from those using questionnaires did not affect heterogeneity. Funnel plots of the subgroup analyses did not reveal any evidence of publication bias, but supported the high degree of heterogeneity found.

Fig. 4
figure 4

Forest plots demonstrating the OR of diabetes in (a) people with MS and (b) relatives of people with MS

Full size image

The risk of diabetes in first-degree relatives of people with MS was examined in 11 studies [7, 8, 27, 28, 30, 31, 34, 35, 41, 44, 45]. There was an overall increased risk of diabetes in relatives of people with MS (OR 1.49, 95 % CI 1.15–1.94, p = 0.002; Fig. 4b). This was associated with significant heterogeneity (Cochran’s Q p < 0.0001, I 2 = 74 %), and there was evidence of publication bias (Egger p value = 0.003; supplementary Fig. 1b), with the smaller studies showing a greater effect size. When the two studies not specifying type 1 diabetes were excluded [44, 45], a similar result was obtained (OR 1.48, 95 % CI 1.10–2.00, p = 0.01). Only two studies [7, 8] used databases to examine the OR of type 1 diabetes in first-degree relatives of people with MS. When these studies were examined separately no increase in diabetes risk was seen (OR 1.13, 95 % CI 0.82–1.57), with no significant heterogeneity and no evidence of publication bias (supplementary Fig. 1c). Interestingly, studies using a questionnaire design [27, 28, 30, 31, 34, 35, 41] appeared to show an increase in the risk of type 1 diabetes in first-degree relatives of people with MS (OR 1.65, 95 %CI 1.17-2.35, p = 0.005) with no significant heterogeneity between studies. However, a funnel plot revealed evidence of publication bias amongst these studies (supplementary Fig. 1d), severely limiting the applicability of the results.

Inflammatory bowel disease

Crohn’s disease

Four studies [7, 9, 24, 27] examined the number of people with MS and Crohn’s disease. There was a significantly increased risk of Crohn’s disease in people with MS (OR 1.37, 95 % CI 1.12–1.69, p = 0.003). Three studies [7, 9, 27] examined the risk of Crohn’s disease in first-degree relatives of people with MS, but showed no increased risk (OR = 1.13, 95 % CI 0.90–1.41, p = 0.31). There was no significant heterogeneity, and publication bias did not appear to be present in either analysis.

Ulcerative colitis

Six studies examined the number of people with MS and ulcerative colitis [7, 9, 24, 27, 32, 46]. Again, there was an increased risk of ulcerative colitis in people with MS (OR 2.26, 95 % CI 1.23–4.14, p = 0.009), but with significant heterogeneity (Cochran’s Q p = 0.003, I 2 = 72 %). Three studies [7, 9, 27] examined the risk of ulcerative colitis in relatives of people with MS, but showed no increased risk (OR 1.15, 95 % CI 0.95–1.40, p = 0.15). There was no publication bias.

All inflammatory bowel disease

Data from six studies [7, 9, 11, 24, 27, 28] were used to calculate the overall OR associated with MS for inflammatory bowel disease. There was an increased risk of inflammatory bowel disease with MS (OR 1.56, 95 % CI 1.28–1.90, p < 0.0001; Fig. 5a). No increase in risk was seen in relatives of people with MS (OR 1.29, 95 % CI 0.92–1.82, p = 0.14) [7, 9, 27, 28, 34, 35]; Fig. 5b). There was no significant publication bias (supplementary Fig. 2a, b), although heterogeneity was observed between those studies examining relatives (Cochran’s Q p = 0.008, I 2 = 68 %).

Fig. 5
figure 5

Forest plots demonstrating the OR of inflammatory bowel disease in (a) people with MS and (b) relatives of people with MS

Full size image

Psoriasis

Eight studies examined the risk of psoriasis in people with MS [7, 9, 11, 24, 27, 28, 30, 31]. There was a significant increase in the risk of psoriasis in people with MS (OR 1.31, 95 % CI 1.09–1.57, p < 0.0001). There was no significant between-study heterogeneity (Cochran’s Q p = 0.16, I 2 = 34 %; Fig. 6). Six studies examined the risk of psoriasis in first-degree relatives of people with MS [7, 9, 28, 30, 34, 35] but showed no increased risk (OR 1.17, 95 % CI 0.94–1.46, p = 0.16), and no heterogeneity or publication bias was detected (supplementary Fig. 3).

Fig. 6
figure 6

Forest plot demonstrating the OR of psoriasis in people with MS

Full size image

Systemic lupus erythematosus

Studies examining the risk of SLE in MS used either clinical diagnosis or serology (i.e. the presence of autoantibodies). Five studies using clinical diagnosis [9, 11, 12, 27, 28] did not appear to show an increased risk of SLE in MS (OR 2.80, 95 % CI 0.76–10.25, p = 0.12), although heterogeneity was high (Cochran’s Q p < 0.00001, I 2 = 88 %). There appeared to be an increased risk of detectable ANA [21, 32, 47–49] (OR 6.36, 95 % CI 1.36–29.69), but with a high 95 % CI and heterogeneity. There was no increased risk of detectable dsDNA antibodies [49, 50] (OR 1.26, 95 % CI 0.29–5.47) in MS. No significant publication bias was seen in either analysis. All studies examining the risk of SLE in first-degree relatives of people with MS [9, 28, 34, 35] used a clinical diagnosis, but showed no increase in risk (OR 1.53, 95 % CI 0.87–2.69).

Rheumatoid arthritis

Eleven studies examined the risk of RA in MS [7, 9–12, 24, 27, 28, 30–32]. There was no association between MS and RA seen in either MS patients (OR 1.15, 95 % CI 0.77–1.73, p = 0.49) or relatives (OR 0.98, 95 % CI 0.80–1.20, p = 0.87). There was significant heterogeneity between studies examining patients with MS, but no publication bias (Egger p value = 0.28; supplementary Fig. 4). When studies using a questionnaire design were selected [24, 27, 28, 30, 31], the lack of association between MS and RA persisted (OR 1.29, 95 % CI 0.98-1.71, p = 0.07), but without heterogeneity (Cochran’s Q p = 0.43, I 2 = 0 %).

Conclusions and discussion

This study demonstrates a consistent increase in the rate of clinical thyroid disorders amongst both people with MS and their first-degree relatives. This finding remained during most of the subgroup analyses, although not when those studies specifying Hashimoto’s thyroiditis were selected, possibly due to the small number of patients. A similar association was seen between MS and inflammatory bowel disease and MS and psoriasis, although this did not appear to extend to relatives of people with MS. Neither SLE nor RA demonstrated a significantly increased rate in either people with MS or their relatives, but significant heterogeneity was demonstrated, limiting the interpretation of these results.

Aside from clinical thyroid disease, the only autoimmune disease showing a significantly increased risk in relatives of people with MS was type 1 diabetes. However, there was a high degree of heterogeneity between studies examining this relationship. Whilst it was possible to reduce this heterogeneity by separately analysing studies that used a questionnaire design and those that used large databases, these two subgroup analyses yielded opposing results, complicating interpretation. Additionally, publication bias clearly affected the results, with the two studies using large databases showing neither publication bias nor heterogeneity; this is in contrast to the smaller-scale studies in which there was evidence of persisting bias. It is possible that poor differentiation between type 1 and type 2 diabetes may have led to an additional significant error in those studies examining diabetes prevalence, and efforts to overcome this were not made in all studies. Detection and reporting bias may also have contributed to this result.

Thyroid disease is relatively common in the general population. The symptoms of thyroid disease tend to be nonspecific and progress insidiously. The finding of a consistent increase in the rate of thyroid disease in both patients with MS and their relatives should prompt the consideration of baseline testing of thyroid function in people with MS, and alert clinicians to consider thyroid dysfunction in those patients reporting nonspecific symptoms who have not had thyroid function checked recently. The increase in the rate of thyroid autoantibodies, although of interest, should not prompt screening for these in MS patients. Thyroid autoantibodies may be present in healthy people with normal thyroid function, with the prevalence of thyroid peroxidase antibodies reported to be as high as 12 % in some series of healthy individuals [51].

Ultimately, the findings of a meta-analysis such as this are limited by the quality of the studies included. Despite the best efforts of the authors of the studies included here, it is highly likely that diseases were misclassified in the studies included here. Comparing self-reports to diagnoses verified by general practitioners, Broadley et al. [34] found that the positive predictive value of a patient-reported condition varied from 32 % for RA to 85 % for thyroid disease [34]. This is a major limitation of questionnaire-based studies. Similarly, reporting bias may have led to over-estimation of autoimmune disease prevalence amongst people with MS and their relatives. This is particularly apparent in those studies examining the frequency of diabetes in relatives of people with MS, where there was clear evidence of publication bias. However, the majority of the more recent studies used large-scale databases, potentially minimizing these sources of bias. Interestingly, in the case of diabetes in relatives, the effect of MS disappeared when studies using databases were analysed separately, highlighting the benefits of such studies.

This study does not address the potential cause(s) of the increased rate of autoimmune diseases demonstrated. This is likely to be multifactorial, as the diseases studied have differing underlying aetiologies and pathogenesis. Common factors in the development of MS and these diseases include both genetic and environmental factors, including smoking and vitamin D deficiency. However, the conditions studied do not have a single underlying pathogenesis, and as such it is difficult to use this study to shed light on the mechanisms underlying MS development. However, it demonstrates the importance of study design when addressing epidemiological questions such as these, and highlights the need to be vigilant for a second diagnoses in people who have an existing diagnosis of MS.