Error bars indicate 95% CI.
eMethods. Methods for Genealogy Reconstruction, Threshold Liability, Model, and Definitions of Socioeconomic Data
eTable. Code List for Sjögren Syndrome and Other Autoimmune Diseases
Kuo C, Grainge MJ, Valdes AM, See L, Luo S, Yu K, Zhang W, Doherty M. Familial Aggregation of Systemic Lupus Erythematosus and Coaggregation of Autoimmune Diseases in Affected Families. JAMA Intern Med. 2015;175(9):1518–1526. doi:10.1001/jamainternmed.2015.3528
Relatives of patients with systemic lupus erythematosus (SLE) appear to be at higher risk of SLE and other autoimmune diseases, but estimates of individual familial risks are largely unavailable or unreliable. Furthermore, relative contributions of genetic, shared, and unshared environmental factors to SLE susceptibility remain unclear.
To examine familial aggregation and heritability of SLE and the relative risks (RRs) of other autoimmune diseases in relatives of patients with SLE.
Design, Setting, and Participants
A population-based family study using the Taiwan National Health Insurance Research Database was conducted. Participants included all individuals (N = 23 658 577) registered with that database in 2010; of these, 18 283 had SLE. We identified 21 009 551 parent-child relationships, 17 168 340 full sibling pairs, and 342 066 twin pairs. Diagnoses of SLE were ascertained from March 1, 1995, to December 31, 2010, and analysis was conducted between March 1 and August 15, 2014.
Main Outcomes and Measures
The prevalence and RRs of SLE and other autoimmune diseases in relatives and spouses of patients with SLE as well as the relative contributions of heritability, shared, and nonshared environmental factors to SLE susceptibility.
Among the more than 23 million participants, the RRs (95% CIs) for SLE were 315.94 (210.66-473.82) for twins of the patients, 23.68 (20.13-27.84) for siblings, 11.44 (9.74-13.43) for parents, 14.42 (12.45-16.70) for offspring, and 4.44 (2.38-8.30) for spouses without genetic similarity. The accountability for phenotypic variance of SLE was 43.9% for heritability, 25.8% for shared environmental factors, and 30.3% for nonshared environmental factors. The RRs (95% CIs) in individuals with a first-degree relative with SLE were 5.87 (4.89-7.05) for primary Sjögren syndrome, 5.40 (3.37-8.65) for systemic sclerosis, 2.95 (2.04-4.26) for myasthenia gravis, 2.77 (1.45-5.32) for idiopathic inflammatory myositis, 2.66 (2.28-3.11) for rheumatoid arthritis, 2.58 (1.16-5.72) for multiple sclerosis, 1.68 (1.22-2.32) for type 1 diabetes mellitus, 1.39 (0.66-2.91) for inflammatory bowel diseases, and 0.86 (0.43-1.71) for vasculitis.
Conclusions and Relevance
The individual risks of SLE and other autoimmune diseases were increased in families that included patients with SLE. The heritability of SLE was estimated to be 43.9%. These data should be considered when counseling families with affected members.
Systemic lupus erythematosus (SLE) is the prototype autoimmune disease with features of autoantibody production, immune complex deposition, and multiple target organ damage. The disease can affect any part of the body,1 and the course of the disease is diverse and unpredictable.2 The prevalence of SLE ranges from 0.02% to 0.15%.3 Our group estimated that the prevalence of SLE was 0.10% in the United Kingdom4 and 0.07% in Taiwan.5
Early family studies6- 13 documented familial aggregation of SLE, and a classic twin study14 found a 10-fold increased concordance in monozygotic compared with dizygotic twins. Furthermore, SLE has been reported10,12 to coaggregate with other autoimmune diseases. The tendency of SLE and other autoimmune diseases to cluster within families suggests a significant role for genetic or shared environmental factors in the pathogenesis of autoimmune diseases. Heritability, defined as the proportion of the phenotypic variance explained by genetic factors, is estimated to be 66% in SLE,15,16 suggesting a strong genetic component to its pathogenesis. Consequently, efforts to define the pathogenesis of SLE focus on genetic factors, and genome-wide association studies17 have successfully identified more than 30 susceptibility loci for SLE. However, these findings account for less than 10% of the phenotypic variation observed.18 A large, unexplained proportion of heritability leads to the issue of the relative contribution of genetic factors to SLE susceptibility.
Two studies15,16 have reported on the heritability of SLE; however, both failed to differentiate genetic and shared environmental factors. Therefore, the studies more accurately estimated familial transmission, which is a measure of the combined contribution of genetic and shared environmental factors to disease susceptibility. Such estimates overestimate the heritability of SLE. In contrast to heritability, quantitative estimates of an individual’s risk of SLE and other autoimmune diseases are more useful for genetic counseling if there is a family history of SLE. However, reliable measures, such as relative risks (RRs), are largely unavailable or of limited reliability.
Therefore, we conducted a study composed of nearly the entire population of Taiwan in 2010. Using genealogy and linked health information derived from a comprehensive database, we determined familial clustering of SLE by estimating the risks of SLE according to specific affected kinship and assessed the relative contribution of genetic, shared, and unshared environmental factors to SLE susceptibility. In addition, we estimated the RRs of other autoimmune diseases associated with a family history of SLE.
This study was approved by the institutional review board of the Chang Gung Memorial Hospital, Taoyuan, Taiwan, with waiver of informed consent. A cohort of all individuals registered in the Taiwan National Health Insurance (NHI) Research Database in 2010 was established using data from the registry for NHI beneficiaries, registry for patients with catastrophic illness, and data sets of ambulatory care expenditures and details of ambulatory case orders, all of which are parts of the NHI Research Database. The database is anonymous. Individuals without valid insurance status were excluded from analysis. The NHI coverage rate was more than 99.5% in 2010.19
Since 1995, the NHI Research Database has recorded sex, date of birth, place of residence, details of insurance (employment categories, sum of insurance amount, enrollment, discharge date), family relationships, vital status, and details of clinical information, including dates of inpatient and outpatient visits, medical diagnoses, medical expenditures, prescription details, vaccination status, examinations, operations, and procedures. The National Health Research Institute acquires all data from the Department of Health and Welfare and implements the data into an electronic database. All of the information is linked using a unique personal identification number assigned to each resident in Taiwan. To ensure confidentiality, this number is encrypted before releasing the data to researchers, but the identification remains unique for each beneficiary in the database to facilitate internal linkage of records.
Methods to identify first-degree relatives in the NHI Research Database have been reported.20 In brief, the registry of beneficiaries contains the identifiers of the relationships between the insured person (who paid the insurance fee) and his or her dependents. Only blood relatives and spouses are eligible to be claimed as dependents of an insured person. A birth certificate issued by the medical facility that delivered the child or a DNA parentage testing for those who were not born in medical facilities is required for a child to be registered as a dependent of the parents. This allows us to establish family relationships (parents, offspring, full siblings, twins, spouse) using the identifiers and unique personal identification of parent, grandparent, child, grandchild, and spouse. In general, parent-offspring relationships and spouses can be identified directly. An algorithm allowing indirect identification of the parent-offspring relationship is also used to maximize possible family links (eMethods in the Supplement).20 Full siblings of an individual were identified if they had the same parents. Twins were full siblings with the same date of birth (±1 day), but twin zygosity cannot be derived from the database. To consider the correlation among people from the same family, individuals were grouped into families according to their relationships (eMethods in the Supplement).
Among 28 402 865 beneficiaries in the NHI (both alive and dead between March 1, 1995, and December 31, 2010), 8 186 069 individuals were registered alone without any identifiable relative. The remaining 20 216 796 individuals were classified into 4 229 301 families. Overall, 21 009 551 parent-child relationships, 17 168 340 full sibling pairs, and 342 066 twin pairs were identified. Each individual may appear multiple times in different categories of family relationships depending on family structure. Data analysis was conducted between March 1 and August 15, 2014.
In Taiwan, patients with suspected autoimmune diseases are referred to specialists for diagnosis and treatment. Patients with a diagnosis of SLE or other autoimmune diseases included in this study (ie, rheumatoid arthritis, systemic sclerosis, primary Sjögren syndrome, idiopathic inflammatory myositis, type 1 diabetes mellitus, multiple sclerosis, myasthenia gravis, inflammatory bowel diseases, vasculitis) are entitled to waive medical co-payment. Diagnostic information is sent to the insurance administration for review by commissioned expert panels to confirm the diagnosis before approval of waivers. In general, the panel reviews the diagnosis in compliance with the updated classification criteria. For instance, the American College of Rheumatology–revised criteria21,22 for classification of SLE were used to assist with the review of certificate applications for SLE. The Registry for Catastrophic Illness Patients23 contains information on these patients with unique personal identification codes, diagnosis, demographics, application date, diagnosing physician, hospital, and other administration data. We used this registry to identify patients with SLE, rheumatoid arthritis, systemic sclerosis, primary Sjögren syndrome, idiopathic inflammatory myositis, type 1 diabetes mellitus, multiple sclerosis, myasthenia gravis, inflammatory bowel diseases, and vasculitis (eTable in the Supplement).
We considered age, sex, occupation categories, income level quintiles, level of urbanization of residence,24 and family size that might confound or modify the familial associations. Details on socioeconomic factors are summarized in the eMethods in the Supplement.
The prevalence of SLE was calculated for the general population and for individuals with affected first-degree family members. We calculated RRs of SLE as the adjusted prevalence ratios between first-degree relatives of an individual with SLE and the general population. The RR estimated in this study is essentially the relative recurrence risk according to the original Risch definition,25 which was the prevalence ratio between individuals with a specific type of affected relative and the general population. Several established methods are available for the estimation of prevalence ratios, including the Breslow-Cox proportional hazards models,26 log-binomial model,27 and robust Poisson methods.28 Cox proportional hazards models are a well-recognized statistical technique to handle censored survival data and estimate instantaneous hazard ratios based on varying follow-up time. Breslow26 adapted the Cox proportional hazards models to estimate prevalence rate ratios in a cross-sectional study by applying an equal follow-up time for all subjects. This method has been proven to produce consistent estimates for prevalence ratios close to true limits.27,29 The Cox proportional hazards model assumes independence between participants. However, family members naturally cluster with each other. Both the marginal model and the shared frailty model are designed to handle bias caused by within-family clustering. The shared frailty model estimates cluster-specific hazard functions before producing joint hazard function, the marginal model focuses on the mean population hazard function.30 Previous studies have documented the comparability of RRs and 95% CIs between the marginal model (given a robust sandwich method to adjust CIs)31 and the frailty model.30 In addition, one study32 suggested that the marginal model produces a more precise measure if the proportion of families containing discordant pairs of disease is low as is the case in our study. The RR was adjusted for age, sex, socioeconomic factors, and family size. This approach has been applied before and validated in other diseases.33
We calculated RRs and tetrachoric correlations for individuals with an affected first-degree relative of any kinship and also for individual kinship (parent, offspring, sibling, twin). Because the kinship and sex of the affected relative may influence familial risk, we fitted models separately according to the kinship and sex of affected relatives (mother, father, daughter, son, sister, brother, twin sister, twin brother). We excluded twins from the sibling analyses. In addition to first-degree relatives, we estimated RRs for spouses. The RR was estimated for the number of affected first-degree kinships (father, mother, son, daughter, brother, sister). In this model, we compared the risk of SLE in individuals with 1 or 2 affected first-degree relatives with the risk in the general population. To measure the degree of similarity in different types of relatives, we estimated tetrachoric correlations for each category of first-degree relationships stratified by the sex of patients with SLE and their relatives, assuming that there is a continuous normally distributed liability underlying the diagnosis of SLE.
Heritability was defined as the proportion of phenotypic variance that is attributable to genetic factors; the theoretical definition of familial transmission is the proportion of genetic and shared environmental contributions. Familial transmission and heritability can be calculated using the polygenic liability model to calculate both measures.34- 36 This model assumes a normally distributed liability of disease resulting from small and additive influences from a large number of unspecified genes and environmental factors. The liability of the affected individuals is greater than a critical threshold, the value of which can be determined with the information of the disease prevalence in the affected and the general populations.
In the present study, familial transmission was defined as the function of the difference of normal deviation of the threshold from the mean liability between individuals with affected relatives and the healthy population (model is described in the eMethods in the Supplement).16,20 The original model assumes no common environmental variance; therefore, familial transmission equals heritability. To account for contributions of shared environmental factors to phenotypic variance, we used the spouse as a control, assuming that spouses share the family environment but have no close genetic similarity with blood-related family members. We restricted family history to first-degree relatives and assumed a mean of 2 siblings in a family.
An alternative way to estimate heritability was based on comparing tetrachoric correlations that were used as an index of phenotypic similarity between siblings and spouses, assuming that they have a similar shared environment but have 50% and 0% genetic similarity.37 Heritability was calculated as 2 × (tetrachoric correlation for full siblings − tetrachoric correlation for spouse).
We further estimated the extent of familial coaggregation of other autoimmune diseases in SLE-affected families by a marginal Cox proportional hazards model, with an equal follow-up time for all participants adjusting for age, sex, place of residence, income levels, occupation, and family size. Relative risks of rheumatoid arthritis, systemic sclerosis, primary Sjögren syndrome, idiopathic inflammatory myositis, type 1 diabetes mellitus, multiple sclerosis, myasthenia gravis, inflammatory bowel diseases, and vasculitis were estimated as the adjusted prevalence ratio of specified autoimmune diseases between individuals with a first-degree relative with SLE and those without a family history of SLE.
All tests of statistical hypothesis were performed on the 2-sided 5% level of significance. All analyses were performed using SAS, version 9.3 (SAS Institute Inc).
The study population comprised 23 658 577 individuals enrolled in the NHI in Taiwan in 2010. Among this group, 18 283 patients had a diagnosis of SLE, giving a crude prevalence of 0.08%. Women had a significantly higher prevalence (0.14%) than men (0.02%), with a female to male ratio of 8.6:1 (Table 1). Overall, 19 085 610 individuals (80.7%) had at least 1 known first-degree relative. The proportions of individuals in the study population with known parent, children, siblings, and twins were 51.3%, 38.2%, 42.2%, and 1.1%, respectively. In the general population of Taiwan in 2010, a total of 45 718 individuals (0.2%) had at least 1 first-degree relative with SLE: 20 343 with affected parents, 12 435 with affected offspring, 13 115 with affected siblings, and 101 with affected twins. Among the individuals with affected family members, 607 had SLE, giving a prevalence of 1.3%. For individuals with affected first-degree relatives with SLE, the age-specific prevalence of SLE was significantly higher than the age-specific prevalence in the general population (Figure).
In the overall analysis, individuals with affected relatives of female, male, and both sexes had respective RRs (95% CIs) for SLE of 16.31 (14.60-18.23), 20.35 (16.01-25.87), and 65.24 (27.36-155.55). Stratified analysis of the prevalence (RR) of SLE in individuals with affected first-degree relatives of specific types is presented in Table 2. Overall, having an affected first-degree relative with SLE was associated with an adjusted RR (95% CI) of 16.92 (15.23-18.80). Table 2 also presents adjusted RRs (95% CIs) for SLE and for different affected first-degree relatives stratified by sex. Although it seems that the sex of affected relatives did not influence the RR, point estimates (RRs) in Table 2 suggest that there may be trends related to sex; in particular, men with a male affected relative tended to have a higher RR.
The RRs (95% CIs) for SLE were associated with the degree of genetic distance between family relatives, with values of 315.94 (210.66-473.82) for twins (with the highest genetic similarity) of patients with SLE, 23.68 (20.13-27.84) for siblings, 11.44 (9.74-13.43) for parents, 14.42 (12.45-16.70) for offspring, and 4.44 (2.38-8.30) for spouses without genetic similarity. In addition, the RRs increased with the number of types of affected first-degree relatives. Compared with the general population, individuals with 1 type of affected first-degree relative had an RR of 17.04 (95% CI, 15.31-18.96) and those with 2 or more of these first-degree relatives had an RR of 35.09 (95% CI, 14.89-82.70) for SLE.
Overall, the tetrachoric correlation (95% CI) for first-degree relatives was 0.33 (0.32-0.34), and the correlations were substantially higher for first-degree relatives compared with those for spouses (Table 2). Tetrachoric correlations (95% CIs) were estimated to be 0.59 (0.54-0.64) for twins, 0.35 (0.33-0.36) for full siblings, 0.27 (0.25-0.29) for parents, 0.25 (0.23-0.26) for offspring, and 0.07 (0.02-0.11) for spouses. Using a threshold liability model, we estimated an accountability for phenotypic variance of SLE of 43.9% for genetic factors (heritability), 25.8% for shared environmental factors, and 30.3% for nonshared environmental factors. By comparing tetrachoric correlations between siblings and spouses, heritability was estimated to be 56.0%.
Table 3 presents adjusted RR (95% CI) values for other autoimmune diseases in individuals with affected first-degree relatives compared with the general population. The RRs (95% CIs) in individuals with a first-degree relative with SLE were 5.87 (4.89-7.05) for primary Sjögren syndrome, 5.40 (3.37-8.65) for systemic sclerosis, 2.95 (2.04-4.26) for myasthenia gravis, 2.77 (1.45-5.32) for idiopathic inflammatory myositis, 2.66 (2.28-3.11) for rheumatoid arthritis, 2.58 (1.16-5.72) for multiple sclerosis, 1.68 (1.22-2.32) for type 1 diabetes mellitus, 1.39 (0.66-2.91) for inflammatory bowel diseases, and 0.86 (0.43-1.71) for vasculitis.
We performed a sensitivity analysis using rheumatologist-based diagnosis as an alternative diagnosis. Using this information, we identified 36 431 patients with SLE and found a higher prevalence of SLE (0.15%), probably owing to the inclusion of patients with a less severe disease or incomplete lupus. Overall, a family history of SLE was associated with RR (95% CI) values of 16.74 (15.77-17.77) for SLE, 22.35 (20.31-24.60) for siblings, and 6.22 (4.77-8.12) for spouses. The heritability for SLE was 41.7% based on these values and the threshold liability model.
The pathogenesis of SLE is multifactorial, including genetic and environmental factors as well as abnormalities of both innate and adaptive immunity.38 Genetic predisposition plays a crucial role in susceptibility, and environmental exposure can cause epigenetic change39 or trigger activation of innate and adaptive immune response to induce or accelerate the development of SLE in susceptible individuals.40
Strong familial aggregation in SLE has been reported6- 16 but, to the best of our knowledge, this is the first population-based study investigating the familial aggregation of SLE and coaggregation of other autoimmune diseases in first-degree relatives of people with SLE. We found that the first-degree relatives have a 17-fold increased risk of SLE compared with the general population and that genetic relatedness is associated with the magnitude of risk of SLE. Sex differences in familial risks are not apparent despite men with a male affected relative tending to have a higher RR.
Heritability in this study was estimated to be 43.9%, which is significantly lower than previous estimates of 66%.15,16 However, both previous studies did not identify shared environmental contributions to the risk of SLE. The extensive family data in our study indicate that shared environmental factors contribute to SLE. To estimate the heritability of SLE, we compared the liability threshold among individuals with an affected sibling and individuals with an affected spouse with that of the general population. Because spouses only share the family environment, the differences in liability threshold between siblings and spouse are contributed to by heritability. We further estimated heritability using methods based on the tetrachoric correlation of disease status; this correlation gave a slightly higher estimate of heritability (56.0%). The difference between estimates of heritability using different methods is probably the result of the tetrachoric correlation coefficient not adjusting for other potential confounders. Therefore, our findings support the contention that familial factors are predominant contributors to SLE susceptibility and that genetic factors explain approximately half of the phenotypic variance of SLE.
Previous studies have documented familial aggregation of SLE. One study41 surveying 570 patients with SLE in the United States found that 27% of the patients had a family history of autoimmune diseases. Another US study42 reported that 10% of patients with SLE had affected first-degree relatives compared with only 1% of controls. A tendency for familial aggregation of autoimmune diseases other than SLE has also been suggested. In a multicenter study of 1177 patients with SLE in 9 Latin American countries,12 97 individuals had at least 1 relative with SLE, and the sibling recurrence risk ratio was 29. Furthermore, a study14 comprising 107 twin pairs found a concordance rate of 24% in 45 monozygotic twin pairs compared with only 2% in 62 dizygotic twin pairs. The reported differences between the present study and earlier research may be attributed to study design, including sampling, case ascertainment, and analytical approach. Previous reports are often based on less robust sampling strategies and case ascertainment, such as hospital records, self-reported diagnosis, and disease registries, therefore limiting generalizability. In contrast, our study used nearly the entire population of Taiwan, and the case definitions of SLE and other autoimmune diseases were based on physician diagnoses, which were scrutinized by expert panels.
Although recent efforts using genome-wide association studies have identified more than 30 susceptibility loci for SLE,17 these loci account for less than 10% of phenotypic variations observed.18 Previous studies generally attribute this apparent gap to (1) undiscovered genetic variances, (2) a heritable epigenetic component or structural variation,43 and (3) gene-gene interactions among known or undiscovered loci.44 Another possible explanation for complex diseases (eg, Crohn disease) is that a proportion of heritability may remain hard to detect because of contributions from rare variants.45 Therefore, our updated estimate of heritability, which is not as high as previously reported,15,16 could partly explain the gap between observed and theoretical variation.
A family history of SLE is also a risk factor for primary Sjögren syndrome, systemic sclerosis, rheumatoid arthritis, multiple sclerosis, myasthenia gravis, and type 1 diabetes mellitus. Previous studies12 have reported that families including individuals with SLE are enriched with cases of rheumatoid arthritis, autoimmune thyroiditis, systemic sclerosis, and polymyositis. Findings of the studies suggest that these autoimmune diseases share part of the pathogenesis of SLE, but the extent of overlapping contributors to disease manifestation differs. This theory is partly supported by the findings46 that some single-nucleotide polymorphisms for risk of immune-mediated disease are associated with multiple autoimmune diseases.
There are several limitations to the present study. First, the classification of cases was based on the diagnosis recorded in the registry of patients with catastrophic illnesses. We do not have detailed information on clinical findings, laboratory testing, and examinations to verify the diagnosis according to formal classification criteria for SLE. Nevertheless, issuance of a catastrophic illness certificate in Taiwan requires strong medical evidence for a diagnosis of SLE that is agreed on by an expert panel, and applications for these certificates are submitted almost exclusively by rheumatologists. Therefore, our case definitions are stringent. However, patients with less severe disease or incomplete lupus are not eligible for a certificate and thus could not have been identified as cases. Furthermore, an alternative case definition, using rheumatologist diagnosis as a case definition for SLE, resulted in findings similar to those of the primary analysis. Second, zygosity of twins is not recorded in the database; therefore, we cannot estimate heritability using a classic twin study design. However, we used 2 of 4 methods most commonly applied to estimate heritability47 and found similar results. Third, because we estimated heritability using the threshold liability model, the results are subject to the assumption that diseases result from underlying liability that is normally distributed in the population. Nevertheless, although this estimation is a potential caveat, data on other diseases, such as schizophrenia, support the validity of this model.48 Fourth, we cannot account for the effects of assortative mating whereby spouses are more similar for a phenotype than they would be if mating occurred at random in the population. If this assortment was not negligible, heritability could have been underestimated.49 However, our model has a theoretical limit of heritability, which cannot be higher than that of familial transmission. Therefore, the heritability of SLE would approach previous estimates only if shared environmental factors were nonexistent. Fifth, the present study was restricted to Taiwan, and different findings may occur in other populations and environments. Therefore, additional studies in other countries are required to determine the generalizability of our findings. Furthermore, cluster effect (ie, correlation between family members) may affect the estimate of RR and its 95% CI. There are 2 approaches to adjust this effect: the marginal and frailty models.30 The frailty model estimates a cluster-specific hazard function through latent variables common to the same cluster, whereas the marginal model focuses on the marginal distribution of the function while separately modeling the association among responses from the same cluster. The current penalized partial likelihood approach to fit the frailty model fails owing to a huge matrix caused by the large number of families involved in our study (4.22 million). However, we randomly selected 10 000 families to compare the results from these 2 models. We found that the familial risk (95% CI) for SLE was 13.92 (6.23-31.14) by the marginal model and 12.24 (5.37-26.29) by the frailty model, suggesting that the 2 models provide similar results. This is consistent with the literature concerning the comparability of these 2 models.30,31
To our knowledge, this is the first nationwide family study confirming that a family history of SLE in residents of Taiwan is one of the strongest risk factors for the disease. Differential risk associated with different kinships suggests a strong genetic component in the susceptibility of SLE. A family history of SLE also exerts an increased risk of other autoimmune diseases. These findings may help inform the design of studies of familial and genetic risk of SLE and may also be useful in counseling families that include individuals with SLE.
Accepted for Publication: June 2, 2015.
Corresponding Author: Chang-Fu Kuo, MD, PhD, Division of Rheumatology, Orthopaedics, and Dermatology, School of Medicine, University of Nottingham, Nottingham NG5 1PB, England (firstname.lastname@example.org).
Published Online: July 20, 2015. doi:10.1001/jamainternmed.2015.3528.
Author Contributions: Drs Zhang and Doherty contributed equally to the study. Dr Kuo had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Kuo, Valdes, Luo, Yu, Zhang, Doherty.
Acquisition, analysis, or interpretation of data: Kuo, Grainge, See, Luo, Yu, Zhang, Doherty.
Drafting of the manuscript: Kuo, Luo, Zhang.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Kuo, Grainge, Zhang.
Obtained funding: Kuo, Yu, Zhang.
Administrative, technical, or material support: Kuo, See, Luo, Yu.
Study supervision: Grainge, Luo, Yu, Zhang, Doherty.
Conflict of Interest Disclosures: None reported.
Funding/Support: This work was funded by the National Science Council of Taiwan (projects 103-2314-B-182A-070-MY2 and 103-2314-B-182-043-MY2) and Chang Gung Memorial Hospital (project CMRPG3D1671) and was supported by the University of Nottingham in methodology and infrastructure.
Role of the Funder/Sponsor: The funding sources had no role in design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, approval of the manuscript; and decision to submit the manuscript for publication.
Disclaimer: This study was based in part on National Health Insurance Research Database data provided by the Administration of National Health Insurance, Ministry of Health and Welfare, and managed by the National Health Research Institutes. The interpretation and conclusions contained herein do not represent positions of the Administration of National Health Insurance or the National Health Research Institutes.