Clinical and Economic Burden of Mild-to-Moderate Atopic Dermatitis in the UK: A Propensity-Score-Matched Case–Control Study

THIN is an electronic medical records (EMR) database including anonymised general practitioner (GP) patient records in the UK [33]. THIN provides data on demographic and clinical characteristics, HCRU and drug acquisition, for a representative sample of around 5.7% of the UK population [34, 35]. The database collects primary care patient information from practices that use Vision; a general practitioner software package developed to facilitate and support practice management and patient care. The database is regularly updated and currently contains inputs from data collected in over 550 general practices [35].

Clinical data in THIN are catalogued using Read codes, a comprehensive and searchable classification scheme for medical conditions, symptoms, and important background information. THIN has been widely used for epidemiological research, and prior studies have validated algorithms that allowed identifying patients with AD in THIN [36]. This study was conducted using THIN version 1809 and analyses were performed in June 2019.

IQVIA Medical Research Data (IMRD), incorporating data from THIN, a Cegedim Database is a collection of de-identified patient records collected from primary care.¹ The data collection scheme is approved by the UK Research Ethics Committee (reference number: 18/LO0441). The protocol for this study was also reviewed and approved by an independent Scientific Review Committee ([SRC] Reference Number 19THIN033), and the study was conducted in accordance with the ethical standards of the 1964 Declaration of Helsinki and its later amendments. Individual patient consent is not required for this type of study.

This retrospective analysis used a cross-sectional design to describe demographic and clinical characteristics of AD patients and to estimate HCRU by analysing data from the most recent complete years available (2013–2017) in THIN. Each individual calendar year, as well as the entire 5-year period, were evaluated to assess the cyclical nature of HCRU in AD. The study focused on the most recent individual calendar year (2017) to assess the nature of HCRU in AD.

AD patients in each time period were identified in THIN based on GP diagnosis of AD, continuous disease activity from 2013 to 2017 and sufficient patient status to fully validate data integrity for the entire time period (Table 1). Continuous disease activity was based on previously validated algorithms, which assessed the later of practice acceptable mortality recording, the information in Vision general practice system or patient registration date, and the earlier of patient transfer out date to practice last collection date [37].

As structured EMR lack information on disease severity, AD patients were stratified by disease severity using treatment as a surrogate severity measure. A published algorithm defined in a UK-based study by Silverwood et al. [38] was used to exclude severe patients and capture mild-to-moderate AD patients. Severe AD patients were excluded based on three criteria: systemic immunosuppressant treatment, a phototherapy code in the Clinical Practice Research Datalink (CPRD) or Hospital Episode Statistics; or AD-related referrals. Silverwood et al. considered patients to have mild AD by default and classified moderate AD based on a second potent TCS treatment within one year or a first topical calcineurin inhibitors (TCI) treatment.

Matched non-AD controls in each time period were identified in THIN based on the same criteria as AD patients except for the presence of AD diagnosis. To achieve this, a propensity-score-matching methodology was used. A logistic regression was performed on all patients (AD patients and matched non-AD controls, combined) to derive the propensity scores. The model included the variables age group, gender, and practice location; the propensity scores were the estimated probabilities that a patient belonged in the non-AD control group. The method of matching used was a greedy-match technique, with a calliper of 0.2 times the pooled standard deviation of the propensity scores, and with matching resulting in equal numbers in each group (PROC LOGISTIC and PROC PSMATCH in SAS version 9.4 were used).

In the UK, NES guidelines on AD treatment reserve very potent TCS for patients with severe AD[12], a criterion that was not captured in the Silverwood et al. algorithm. These guidelines state that AD patients who have received at least one TCS defined in the UK as ‘very potent’ (e.g. clobetasol propionate 0.05% and diflucortolone valerate 0.3%) would have severe disease [12]. Therefore, a fourth criterion that excluded patients who were treated with very potent TCS and who most likely had at least moderate AD as per the guidelines, was applied and explored in a secondary analysis.

Details of the variables included in this THIN analysis are available in Supplementary Tables S1–3. Demographic characteristics included age, sex, urban/rural classification, practice ID and Townsend code. Clinical characteristics included AD diagnosis and AD-related comorbidities. In order to assess the most relevant comorbidities in AD, common AD-related comorbidities were identified in the literature, validated through clinical expert insights, and aligned with corresponding frequencies in THIN during the entire study period. Subsequently, AD-related comorbidities in this study were confined by considering only comorbidities with a prevalence of 2.00% or higher (a threshold selected by consensus of all co-authors) in THIN.

HCRU was evaluated based on GP visits, total referrals, non-AD-related dermatology referrals, and AD-related prescriptions. GP visits included home, nurse, and telephone consultations. Non-AD related referrals were excluded using a specific THIN variable to align with the Silverwood et al. algorithm. Additionally, AD-related prescriptions were captured for emollients, TCI, TCS, topical antibiotics and topical antivirals. The drugs prescribed for AD patients were based on THIN drug codes using the British National Formulary (BNF) codes.

Means, medians, and standard deviations were provided for continuous variables when performing descriptive analysis of continuous data, while numbers and percentages were provided for dichotomous and polychotomous variables when performing descriptive analysis of categorical data.

A generalised linear model was fitted for each of the HCRU variables (separate models), applying the normal distribution and the identity link. The independent variables included patient group (AD or control), practice location (England, Scotland, Northern Ireland and Wales), and their interaction; as well as age group (age 0–7, age 8–11, age 12–17, age 18+), and its interaction with patient group. Furthermore, gender and practice location were included as classification variables. Additionally, two summary measures of comorbidity were considered including ‘at least one metabolic-and-lifestyle’ comorbidity as a classification variable and Charlson Comorbidity Index (CCI) as a continuous variable. Mean differences were calculated (e.g. between AD patient and control group), along with 95% confidence intervals and associated p-values. It is noted here that the comorbidities had been aggregated and categorized into larger groups including (I) allergic comorbidities for AD, and (II) non-allergic comorbidities for AD. This latter was then divided into the following categories: (1) neuropsychiatric, (2) cardiovascular, (3) metabolic and lifestyle, (4) malignancies: lymphoma (adult), (5) skin infections, (6) eye disease and (7) autoimmune disease. All of these had been considered as classification variables but including them all into the models led to an ill-condition results (e.g. estimates were inflated). Through fitting multiple models and comparing results across dependent variables (in order to have a single set of independent variables for all of the dependent variables), ‘Metabolic and lifestyle’ alone remained in the models as a meaningful, statistically significant predictor. Based on clinical expert opinion and findings from the literature [21, 22], the metabolic and lifestyle category was considered as clinically relevant and predictors for the additional comorbidities included in the analysis (e.g. allergic and neuropsychiatric comorbidities).

Statistical analysis was developed and conducted in SAS.

The GP visit unit cost was identified from the Personal Social Services Research Unit [39] while referrals unit costs were derived from the National Health Services (NHS) Reference Costs from 2017 to 2018 [40]. Drug unit costs were sourced from the BNF 2018 [41] and the Monthly Index of Medical Specialties database 2018 [42].

Additional targeted literature reviews, using UK-specific sources, were conducted to identify productivity loss and comorbidity costs. The Ovid search platform was used to conduct the literature searches. The following databases were identified and used to conduct the searches: Excerpta Medica dataBASE (EMBASE), MEDLINE®, Cochrane Library, EconLit and NHS Economic Evaluation Database. The search strategies were defined in terms of the patient population, intervention and comparator, outcomes and study design (PICOS) framework. All literature searches were conducted on 25 June 2019. Details of the economic inputs, inclusion/exclusion criteria and the search terms are available in the Supplement and Tables S4–S10.

A cost-of-illness model was constructed to extrapolate the clinical and economic burden of mild-to-moderate AD and matched non-AD controls in THIN to the UK population in 2017 and subsequent years (2018–2022). Population estimates were obtained from the Office for National Statistics (ONS) [43, 44]. The primary analysis considered the burden of mild-to-moderate AD by excluding severe patients defined in the Silverwood et al. algorithm [38] and including HCRU and drug costs. The impact of disease severity was evaluated in a secondary analysis by excluding patients treated with very potent TCS who most likely had at least moderate AD as per the NES guidelines. Additionally, the burden of AD-related comorbidities and the impact of productivity loss due to AD were assessed in scenario analyses.

All costs have been inflated to 2018 and converted to euros (€) using the average ONS conversion rate (1.13) for 2018 [45].

The sample size and the baseline characteristics of the mild-to-moderate AD patients and non-AD patients prior to the propensity score matching in THIN are available in Table 2.

In 2020, mild-to-moderate AD patients incur substantially higher HCRU costs compared with matched non-AD controls encompassing GP visits (€202.37M versus €130.36M) and total referrals (€81.09M versus €68.51M). The costs per patient per year are shown in Fig. 2.

In 2020, mild-to-moderate AD patients also incur substantial drug acquisition costs accounting for €8.04M, based on a per-patient cost of €9.36. Overall, the total HCRU and drug costs for mild-to-moderate AD patients compared with matched non-AD controls were €291.51M and €198.87M, respectively. The excess drug and HCRU cost at the UK population level was €462.99M over the total projected 5-year period (Table 6).

In 2020, excluding AD patients treated with very potent TCS, who most likely had at least moderate AD, decreased the total costs of mild-to-moderate AD (n = 802,278) to €269.65M based on a per-patient cost of €336.11 (Table 6 and Fig. 3). The excess drug and HCRU costs decreased to €417.35M after excluding these patients compared with the primary analysis for the projected 5 years at the UK population level. The clinical and economic burden of comorbidities also decreased for these patients (Supplementary Tables S12–15 and Fig. S1).

In 2020, the burden of mild-to-moderate AD substantially increased when considering the financial impact of AD-related comorbidities (Tables 7, 8 and 9 and Figs. 4–5). Overall, the total costs of mild-to-moderate AD patients increased to €591.59M. The excess cost of mild-to-moderate AD patients compared with matched non-AD controls increased to €1.21B for the projected 5-years at the UK population level.

The impact of the productivity loss also had a substantial impact on the burden of mild-to-moderate AD (Table 9 and Fig. 6), increasing the associated total costs of mild-to-moderate AD to €8.06B aggregate for the projected 5 years at the UK population level.

This study demonstrated that mild-to-moderate AD is associated with an increased public health burden given its high prevalence among the overall THIN population (1.28%). However, the overall prevalence estimated from this study was lower than previously reported estimates, ranging between 1.62% and 15% [3‐5]. Compared with previous studies reporting AD prevalence, this study focused specifically on mild-to-moderate AD patients using EMR. Given the differences between the self-reported AD prevalence in the open population compared with physician-diagnosed disease in general practice, it is challenging to establish the true prevalence of mild-to-moderate AD [47]. Therefore, mild-to-moderate AD prevalence may be underestimated by only using physicians’ consultations. It can be suggested that clinicians should aim for improved identification and recording of AD and its severity in the primary care setting in order to ensure optimal management of the disease [32, 48].

The prevalence obtained from this analysis was the highest in childhood (5.11% among children between 0 and 7 years old) aligning with previously reported prevalence of children (5.9% and 14.2%)[6]. Given that parents and other family members are commonly involved in the care-giving of children with mild-to-moderate AD, the previous literature suggested that the burden of AD is shared by the patients and their families [49]. The detrimental impact of AD on the quality of life, the social, academic and occupational aspects on both patients and their families, as well as the burden on society due to higher costs and decreased productivity, have also been recognised in previous studies [46, 49]. Hence, the high prevalence of mild-to-moderate AD in children and the associated impact of the disease on the entire family unit, should be recognised when assessing the true burden of mild-to-moderate AD.

In this analysis, mild-to-moderate AD patients reported substantial HCRU and drug acquisition costs compared with matched non-AD controls. The main drivers of HCRU were GP visit costs, consistent with literature findings [50]. However, the AD-related dermatology referrals costs were not included in this evaluation to align with the algorithm by Silverwood et al., which may have underestimated the calculated HCRU. Hence, as mild-to-moderate AD is commonly treated in primary care, reducing GP visits through effective treatments might decrease primary care demand and also reduce the burden on the entire healthcare service [51].

The costs substantially increased when considering AD-related comorbidities, particularly asthma, bacterial skin infections, depression and sleep disorder; the impact of these comorbidities on the burden of AD has been recognised in previous studies [14, 52‐54]. Despite its low frequency in THIN (1.38%), sleep disorder is estimated to account for substantial costs (€14,667,476.37 in 2020) at the UK population level. Additionally previous studies have shown that sleep disorders may result in increased levels of pruritus with associated poor school performance, family dysfunction and high Dermatology Life Quality Index scores [55‐58], and may also worsen with disease severity[59]. Studies have also found that children with severe and persistent AD may face an increased risk of developing asthma and allergic rhinitis at a later life stage [60, 61]. Secondary skin infections caused by Staphylococcus aureus and Herpes simplex have also been associated with AD flares [62].

Based on the assumption that AD patients treated with very potent TCS had at least moderate disease, this analysis showed that more severe AD is possibly associated with a greater economic burden of comorbidities compared to milder AD. These findings are consistent with the conclusions from previous studies where increased disease severity was shown to be associated with a substantial increase in the frequency of comorbidities, which further associate with detrimental impact on quality of life and productivity loss [27, 63]. This notion would support the hypothesis that early diagnosis and treatment of mild-to-moderate AD may prevent the future development of associated comorbidities, which represent a significant burden to the health service [64].

The main limitation of this study is related to the definition of severity used which required stratification of AD patients by disease severity using treatment as a surrogate measure for severity rather than using an objective clinical measure of severity. Data using established disease severity scoring measures, such as the Investigator Global Assessment (IGA), Eczema Area and Severity Index (EASI) or SCORing Atopic Dermatitis (SCORAD) are not routinely collected in the THIN database and therefore were not available for use in this study. Therefore, this study used a previously published algorithm for severity that was reviewed and supported by clinical expert opinion.

Based on further insights provided by clinical experts, this analysis assessed the impact of considering a potentially ‘milder subgroup’ on the burden by excluding patients treated with very potent TCS. Although this analysis did not allow for controlling for the impact of cofounding factors including site of application, patient age and previous treatment [12], the burden observed substantially decreased when excluding patients treated with very potent TCS. Therefore, it might be suggested that these patients have an increased clinical and economic burden, although the impact of cofounding factors on the burden remains uncertain.

Furthermore, this study focused on the most recent complete years available (2013–2017) in THIN at the time of the analysis and, hence, could not account for the use of dupilumab, which was recommended in the UK after the time period of our study [65].

As a targeted literature review did not identify any UK-specific sources reporting the incidence of mild-to-moderate AD, the AD population was estimated using the prevalence obtained from THIN. The economic burden of AD might be underestimated in this study given the previously reported incidence of 1.10% for mild-to-severe AD patients in England and Wales [5], which was not considered. When including this incidence in an exploratory analysis, it yielded a substantial increase in the economic burden of AD to €1.65B [5]. Therefore, it can be argued that the true clinical and economic burden of mild-to-moderate AD lies between €462.99M and €1.65B.

Additionally, THIN collates medical records collected at primary care, therefore this study could not capture secondary care data depicting the burden of moderate-to-severe AD. Previously, THIN data have been linked with Hospital Episode Statistics (HES) data, which provide the potential for linking primary and secondary care [35]. However, this was beyond the scope of the current study, which originally aimed to explore the burden of mild-to-moderate AD given its high prevalence. It can still be suggested that the estimates on the burden of mild-to-moderate AD are generalisable for the UK population as approximately 97% of AD patients are managed by their GP in the UK [7, 66], with THIN capturing primary care data for a representative sample of around 5.7% of the UK population [34].

Furthermore, while a comprehensive range of Read codes were applied to identify obesity in this analysis, the estimated frequency remains lower than expected. This underestimation is likely due to obesity being under-managed and recorded despite guidelines. This was concluded in a previous review which included studies that measured the proportion of adult patients with documented body mass index (BMI) or weight loss interventions in the UK across a range of regional and national databases [67]. This under-reporting is likely to underestimate the economic burden of AD-related comorbidities. Comorbidities such as asthma and ischaemic heart disease may be recorded more accurately given the nature of these conditions.

Finally, the costs of comorbidities, which were identified from the targeted literature search, entailed different cost components, and therefore the true burden of comorbidities remains uncertain. Given that mild-to-moderate AD patients had substantially higher comorbidity rates compared with matched non-AD controls, it can still be argued that the incremental burden of AD-related comorbidities remains substantial.