Comorbidity burden in adult atopic dermatitis: A population ‐ based study

Background: Atopic dermatitis (AD) is a chronic inflammatory skin disease that has been shown to be associated with allergic comorbidities. However, studies examining comorbidities in patients with AD are incomplete, which may contribute to suboptimal care


INTRODUCTION
Atopic dermatitis (AD) is a common and chronic inflammatory skin disease characterized by dry and inflamed skin and pruritus. 1In Sweden, previous studies have presented 12-months prevalence numbers of 14%. 2 AD causes significant patient burden, impacting quality of life 3 and is associated with significant economic costs. 4urrent disease management aims to improve signs and symptoms and to achieve long-term control.AD has traditionally been viewed as the first clinical manifestation in the 'allergic march' (also known as 'atopic march' including asthma, allergic rhinitis and food hypersensitivity). 51][12] Increased levels of disease severity are associated with higher prevalence of comorbidities 9 and increased costs and resource use. 13Despite the rapid pace of research in AD, observational studies examining the clinical burden of adult AD regarding the assessment of comorbidities are incomplete, which may contribute to suboptimal treatment of patients. 10,12,14Establishing the clinical burden of patients with AD may encourage physicians to proactively screen for clinically relevant and common comorbid conditions and ensure that patients are treated holistically.
This large population-based study examined the clinical burden of AD comorbidities by comparing the time to onset, prevalence, and incidence for a wide range of conditions in adult patients with AD, to that of a matched reference cohort without AD in Sweden.The analyses were controlled for disease severity, including time in remission.

Data and ethics
This large population-based cohort study used linkage data from prospectively collected national and regional registers in Sweden.The registers included the National Patient Registry (NPR), which contains medical information for all in-and outpatient specialist visits, including International Classification of Disease version 10 (ICD-10) codes and dates, the Prescribed Drug Registry (PDR), which includes data for all pharmacy-dispensed medications [Anatomical Therapeutic Chemical (ATC)codes] from both primary and specialist care, including pharmacy dispensation dates, and the Cause of Death Registry (CDR), which contains information on causes of death and corresponding dates.These three databases are managed by the Swedish National Board of Health and Welfare which requires mandatory reporting and thereby provides complete population coverage.
Data were also extracted from regional primary care databases from two regions in Sweden, that is, Västra Götaland and Skåne.These databases cover approximately one-third of the Swedish population and include healthcare visit information (ICD-10 diagnoses codes and corresponding dates for visits).
Socioeconomic information including household income, education and employment status and migration information, was extracted from the Longitudinal Integration Database for Health Insurance and Labor Market Studies (LISA).Last, the Total Population Registry provides data that includes the entire Swedish population and was used to identify a random sample of individuals without AD (i.e., non-AD cohort).Unique personal identification numbers were used to link the data at the patient level from each registry.
Ethical approval for this study was received from the Swedish Ethical Review Board (reference number 2019-03840).Individual patient consent was not required for this type of study.

Study population and study design
A cohort of adult patients (over 18 years of age) with AD identified between 2007 and 2017 was included in this study.AD was defined using only a registered diagnosis of AD (ICD-10: L20) in primary-(Västra Götaland and Skåne) or specialist care (NPR).The index date i.e., start of follow-up was defined as the date of the first observed AD diagnosis during the study period.The inclusion period ended on 31 December 2017 and the study period ended on 31 December 2018, allowing all included study participants to have a minimum follow-up of 1 year.Individuals were followed from index to comorbidity event or censoring [death, emigration, or end of the study period (31 December 2018)], whichever came first.Individuals with a prior record of the evaluated comorbidity were excluded.The study processes are shown in Figure 1.
The non-AD reference cohort was randomly selected from the Swedish population who did not have an ADdiagnosis nor were dispensed any oral or topical corticosteroid (TCS) or a topical calcineurin inhibitor (TCI) before the index date.This minimized the risk of including AD-patients into the non-AD reference cohort which can be considered a 'healthy' representative sample of the Swedish non-AD population.The non-AD reference cohort was matched 1:1 to the AD cohort with replacement, 15 on age at index, gender, and region.Each non-AD reference individual was assigned the same index date as the matched AD patient.

Comorbidity development
This study evaluated time to first diagnosis for an extensive set of comorbid conditions in adult patients with AD and a non-AD reference cohort, aged 18 years and above at index.The following comorbidity endpoints were analyzed separately: asthma, allergic rhinitis, food hypersensitivity, neurological disorders, psychiatric disorders, ocular disorders, infections, immunological & inflammatory disorders, type 1 diabetes (T1D), T2D, endocrine & metabolic disorders, skeletal disorders, malignancies (any type of malignant tumors) and cardiovascular diseases (including coronary artery disease and venous thrombotic events).7][18][19][20][21][22] The specific diseases and ICD-10 codes for all conditions evaluated are enlisted in Supporting Information: Table S1.In each analysis of comorbidity onset, individuals were excluded if they had the specific comorbidity of interest before the index date.
F I G U R E 1 Schematic chart of study design.AD patients were identified at their first AD diagnosis in 2007-2017 and followed until death, emigration, or end of study period (31 December 2018).Patients with a prior record of the evaluated comorbidity were excluded.AD, atopic dermatitis.

Disease severity
][25] However, administrative databases do not include patient-reported outcomes or clinical measures of relevance to define the severity of the disease.7][28][29] Severity in this study was defined according to both (a) type of treatment and (b) potential visits to specialist care for AD.This approach is similar to that used in a previous population-based cohort study from the United Kingdom 14 with modifications to better fit the Swedish setting based on inputs from clinical experts.Patients with a dispensation of a very potent TCS (e.g., clobetasol), dupilumab, systemic immunosuppressants or a systemic corticosteroid prescribed by a dermatologist or phototherapy within secondary care were classified as severe AD.Patients with a dispensation of mild, moderate or potent TCS (e.g., hydrocortisone, betamethasone and mometasone), TCI, emollients, phototherapy within primary care, or those who had a healthcare contact with a diagnosis of AD within primary or secondary care, were classified as mild-to-moderate (M2M) AD.Disease severity for AD was evaluated on an annual basis from index date until the end of the study period, except for the first year when severity level was evaluated from 1 year before index date to-and including-30 days after index.Disease remission for AD was defined as a stop in meeting of the definitions of M2M or severe AD in patients who had been previously identified through an AD diagnosis.Since AD is a chronic disease, these patients were considered to have a nonactive AD.All patients were classified as having either M2M or severe AD in the first year since an AD diagnosis was encompassed in both the inclusion criteria and the severity criteria.See Supporting Information: Table S2 for complete classification criteria including corresponding ICD-10-, ATC-, and procedure codes.

Statistical analyses
Descriptive statistics were computed for patients at risk, stratified by severity and compared with the non-AD reference cohort: Mean and standard deviation (SD) were compared for continuous variables and number and percentage for categorical variables.The risk of onset of comorbidities and the presence of AD (M2M, severe, or AD in remission) were analyzed using Cox proportional hazards models and displayed using forest plots and graphs of proportion of patients with onset by follow-up time (i.e., 1-Kaplan-Meier survival function). 30Cox regression models were used in this study.These included unadjusted models which used the presence of AD as the single covariate when estimating the hazard ratio (HR); and adjusted models estimating the HR for the severity variables while controlling for baseline age, sex, pre-existing conditions profile, disposable household income, education, employment status, region of residence, and index year.
The proportional hazards assumption was tested for all adjusted Cox regression models based on the Schoenfeld residuals. 30Selected variables were used to stratify certain analyses where the proportionality assumption for the Cox model was violated.
The effect of referral bias was tested in a sensitivity analysis in which a subsample of the non-AD reference cohort (who were required to have had at least one in-or outpatient hospital visit or at least two primary care visits within 1 year before the index date) were studied.
All p-values were evaluated assuming two-sided tests, with an a priori statistical significance level of α = 0.05.This study followed the extended STROBE guidelines. 31ata management, statistical analyses and graphics were performed using Stata version 16.

Missing data
Patients with missing socioeconomic or demographic information were excluded from this study.See the attrition figure for details in Supporting Information: Figure S1.

Patient characteristics
This study included a total of 215,548 adult individuals, with 107,774 patients diagnosed with AD and an identical number of individuals without a diagnosis of AD.At index date, 92,413 AD patients were classified as having M2M AD (65% females) and 15,361 patients were classified as having severe AD (62% females).The prevalence for each comorbidity at baseline was below 20% in all cohorts and the greatest differences in prevalence between the AD-and the non-AD reference cohorts were observed for asthma, allergic rhinitis, cardiovascular disorders, immunological & inflammatory disorders, infections, and neurological disorders, see Table 1.

Incidence of comorbidities among the AD cohort compared with the non-AD reference cohort
Total years of follow-up in this study were 648,414 person-years (PYS) for AD patients (mean years of follow-up per patient: 6.02 PYS) and 643,197 PYS for the non-AD reference cohort (mean follow-up per patient: 5.97 PYS). Figure 2 shows the proportion of patients by number of developed comorbidities.In the AD cohort, a total of 23.5% developed one comorbidity during follow-up compared to 20.0% in the non-AD cohort.In addition, 15.9% in the AD cohort developed multiple comorbidities (≥2) compared to 7.2% in the non-AD cohort.This implies that among those patients who developed at least one comorbidity, 39.7% and 26.5%  developed multiple comorbidities (≥2) in the AD cohort and the non-AD cohort, respectively.Neurological disorders (dizziness and giddiness, headache, etc.) were the most common comorbidities in the AD cohort with 8.73% [95% confidence interval (CI): 8.55%-8.91%] of patients getting these events.The most common events in the non-AD reference cohort were skeletal disorders, which occurred in 6.78% of individuals (6.63%-6.94%).See Table 2 for incidence rates for all evaluated conditions and Supporting Information: Table S3 for details for years of follow-up and number of PYS.The incidence of the investigated conditions is graphically displayed using cumulative incidence curves, see Supporting Information: Figures S2-S15.

Risk of comorbidity onset based on HRs
The risk of developing comorbid conditions was analyzed using unadjusted-and adjusted Cox proportional hazard models.The unadjusted models showed a statistically significant increased risk of onset for all analyzed conditions (except T1D) in the AD cohort compared with the non-AD reference cohort, see Table 2.
Figure 3 shows the adjusted HRs for the Cox proportional hazard models.Adult AD patients had a statistically significant increased risk of onset for all analyzed conditions except for T1D and T2D for patients in remission, compared with the non-AD reference cohort.Atopic comorbidities (asthma, allergic rhinitis, and food hypersensitivity) had the highest risk among the analyzed comorbidities, followed by the immunological-& inflammatory disorders and infections.Patients with AD also had a significantly higher risk of developing cardiovascular disease and malignancies compared to the non-AD reference cohort.The AD cohort had more events of malignancies in all subcategories including melanoma, lymphoma and non-Hodgkin lymphoma.See Supporting Information: Table S5 for statistical testing of differences in HRs by AD severity.
A subsample of the non-AD reference cohort [who were required to have had at least one in-or outpatient hospital visit or at least two primary care visits within 1 year before the index date (data not shown)] were studied in the sensitivity analyses to evaluate any possible effect of referral bias.The results from this subsample were consistent with the overall results.
The comparison of AD patients who had an active disease, with AD patients who were in remission (as per definition) showed a statistically significant increase in risk of onset for all analyzed conditions with the only exception of T2D (remission vs. M2M) and skeletal disorders, see lower panel of Figure 3.Although no significantly increased risk of skeletal disorders and AD was found, osteoporosis (a subcategory within skeletal disorders) had a twofold number of events in the AD population compared to the non-AD reference cohort (286 vs. 140 events, statistically significant at α = 0.05).
All models met the proportional hazards assumption based on Schoenfeld residuals after stratification (p < 0.05).Most models needed to be stratified by calendar year of index and region of residence to satisfy the proportional hazards assumption.Detailed Cox regression results can be found in Supporting Information: Tables S5-S8.The distribution of events for each comorbidity endpoint are presented in Supporting Information: Table S9.

DISCUSSION
Using a large population-based cohort of adult AD patients compared to a non-AD reference cohort from real-world clinical settings in Sweden, the present study showed that the comorbidity burden of AD is substantial.Differences in comorbidity burden between AD patients and non-AD reference individuals were present at baseline and continued to develop over time at a higher rate in AD patients.The development of multiple comorbidities, including either allergic or nonallergic, was also more frequent in patients with AD.
The association between atopic disorders and AD is well documented and can aid the physician in diagnosing AD. 9 Our results showed that adults with AD are at an increased risk of many other comorbidities that ultimately have a negative effect on quality of life. 32xtending previous findings, we observed a statistically significant association between AD and nonatopic conditions, including neurological disorders, 16 psychiatric disorders, 17 infections, 18,19 immunological-& inflammatory disorders, 20 endocrine-& metabolic disorders (excluding T1D and T2D), 21 skeletal disorders, 16 ocular disorders, 22 and cardiovascular diseases. 10n almost all analyses performed in this study, the point-estimate of the HR showed that patients with severe AD had higher risk of comorbidity onset and some differences in risk for comorbidity onset between severity levels were statistically significant.Similarly, adults in remission tended to have a lower risk of comorbidity onset compared with those who had active AD.Moreover, this is the first study showing that even AD patients in remission are at higher risk of comorbidity onset compared with individuals without AD which indicated that the risk of comorbidities is not associated with the skin manifestations of AD only.
T A B L E 2 Incidence rates for comorbidities in the AD cohort and the non-AD reference cohort.][35][36][37][38] The results from the present study add to the growing body of evidence suggesting that AD should be diagnosed and managed as a systemic condition, or at least it should be recognized that patients with AD can have multiple comorbidities, including nonatopic conditions. 6,10This highlights the need for physicians to take a holistic approach when treating AD and to ensure that patients are monitored for the development of comorbidities on a continuous basis even when symptoms of AD are in remission.

Malignancy
The relationship between AD and malignancy is complex and remains controversial. 6Multiple studies have analyzed the relationship between AD, AD treatments, and malignancy.A previous meta-analysis showed an increased risk for lymphoma in AD cohort studies, but none in case-control studies. 39The literature on the relationship between AD treatments and cancer shows mixed results.Some studies show no association 34,35,38,40 while other studies cannot reject an association between the AD treatment and elevated risk for cancer. 36,37,40

Cardiovascular disease
Previous studies have reported conflicting results regarding a possible association between AD and cardiovascular diseases.A systematic review and meta-analysis showed that it is unlikely that AD represents an independent and clinically relevant risk factor for cardiovascular diseases. 41The present study could not control some risk factors for cardiovascular disease, such as diet, smoking or low physical activity.These poor health behaviors of patients with AD might have contributed to statistically significantly higher risk of onset of cardiovascular diseases in the AD cohort compared to the non-AD reference cohort.

Immunological & inflammatory disorders
Immunological & inflammatory disorders showed the highest risk of onset in the AD population after the atopic comorbidities.This is perhaps unsurprising as closely related skin conditions are included in this group but the higher risk was also driven by noncutaneous diseases such as ulcerative colitis, rheumatoid arthritis and Crohn's disease.A previous Danish study showed that adult AD was significantly associated with 11 of 22 examined autoimmune diseases 20 and AD has been shown to be a significant risk factor for the development of rheumatoid arthritis and inflammatory bowel disease. 42The full mechanisms linking AD to immunological and inflammatory disorders are not completely understood although known shared susceptibility loci between AD and some inflammatory bowel diseases may explain some of the higher risks.

Infections
The results from this study showed that infections had the second highest risk aside from the atopic comorbidities.These results are supported by other studies showing significantly higher odds of infections in AD patients. 43,44Skin barrier defects are potentially contributing factors for the increased risk of skin infections in AD. 18 Moreover, Th2/Th1 imbalance which is typically present in patients with AD may contribute to the higher risk of infections as Th2-driven responses are believed to impair the innate immunity defenses against bacterial infections. 45,46I G U R E 3 Forest plot of the adjusted Cox regression models, showing hazard ratios for (a) AD severity and remission in comparison to the non-AD reference cohort, (b) AD severity in comparison to the AD remission cohort: The Cox regression models used the displayed condition as endpoints and included the covariates: M2M and severe AD evaluated annually from index date until the end of the study period, except for the first year when the severity was evaluated from one year before index date to-and including-30 days after index.Sociodemographics were defined using age, sex (measured at index), education and household income (measured during the calendar year before the index date).Pre-existing conditions profile was measured from 1 July 2005 to-and including-the index date.The calendar year for index date, and the region of residence were also included as covariates.95% confidence intervals are presented in the parentheses.Patients were followed from the index date to comorbidity event or censoring [death, emigration, or end of study period (31 December  2018)].AD, atopic dermatitis; End., endocrine; imm., immunological; inf., inflammatory; M2M, mild-to-moderate; met., metabolic; T1, type 1; T2, type 2.

Diabetes
The significant higher risk of T2D for patients with M2M and severe AD observed in our study differentiates from the results in a Danish study.The Danish study suggests that the association between AD and diabetes could be linked to other factors which the present study could not control for. 41The imbalance between Th2 and Th1 in patients with AD may have a protective effect on T1D, a Th1-driven disease and our results support this hypothesis.

Strength and limitations
A major strength of this study is the comprehensive assessment of comorbidities in adult AD patients in this population-based registry study.This study included over 200,000 individuals (AD patients and non-AD reference individuals combined), more than 1.2-million-person years of follow-up and almost 100 individual conditions grouped into 14 comorbidity-categories.The main results were based on comparisons between the AD cohort and a matched non-AD reference cohort.In addition to these results, we also compared patients with an active AD disease to patients in remission to demonstrate the robustness of the findings and we observed similar results.Moreover, all models included age, sex, income, education-level, employment status, region residence as well as the presence of other comorbidities which address some of the risk of cofounding that generally exists in observational studies like this.
The longitudinal Swedish register data used in this analysis allowed for the evaluation of a wide range of comorbidities in a single cohort with a high level of coverage.Both primary and specialist care registers were utilized in the present study to identify AD patients, allowing the inclusion of the mildest to the most severe AD patients.
Despite the comprehensive databases used in the present study, certain limitations still exist.Most importantly, data on clinical assessments of severity of AD were not available.Instead, disease severity was assessed through a case-finding algorithm that relied on registry-based diagnosis and medication data based on treatment guidelines as a proxy. 47isclassification of AD was possible since this study lacked complete coverage of primary care data in Sweden.Patients with an AD diagnosis from a primary care region which was not included in this study may have been included in the non-AD reference cohort.However, while sampling of the non-AD reference cohort, an inclusion criterion of no dispensation of TCSs and TCIs was used, therefore the number of misclassified AD patients in the non-AD reference cohort should be low but cannot be completely ruled out.The ICD-10 code used for AD (L20+) identification in this study has a positive predictive value of 92% as validated in a Danish study which can be generalized to a Swedish setting. 48isclassified AD patients in the non-AD reference cohort would likely bias the results towards smaller differences being observed between cohorts, when the true difference may be higher.Finally, given the observational design of the present study, the observed association between AD and various comorbidities might have also been confounded by factors that were not observable in this study like smoking and alcohol consumption or body weight.Nonetheless, the study did control for various sociodemographic factors like income, education and employment status.

Future research
The associations between AD severity and remission with the onset of a large number of comorbidities may be attributable directly to disease severity or mediated indirectly by treatment decisions.Due to the current fast-paced development of pharmaceuticals in AD, the interplay between comorbidity, disease severity and treatments should be further investigated.Both adult and pediatric patients with AD have been shown to have a high comorbidity burden with large differences between the incidence of many comorbidities. 5,10,17dditional investigations of these differences may yield data that could be used to further improve the long-term management of AD.

CONCLUSIONS
The clinical burden of AD is substantial, and adult patients are at an increased risk of developing a large number of comorbid conditions that extend well beyond the atopic disorders.This study highlights the need for interdisciplinary approaches in the management and care of adult AD patients.

F
I G U R E 3 (See caption on next page).
Patient characteristics at baseline.
T A B L E 1 Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/jvc2.303 by Umea University, Wiley Online Library on [20/03/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Abbreviations: 95% CI, 95% confidence interval; AD, atopic dermatitis; PYS, patient-years.aIndividuals with the condition of interest at baseline were excluded.bAssessed from index to comorbidity event or censoring [death, emigration, end of study period (31 December 2018)], whichever came first.cUnadjustedCox regression models used the presence of AD as the single covariate when estimating the hazard ratio.