Cluster Analysis of Clinical Remission and Relapse Patterns in Chronic Urticaria: Results from the PREDICT-CU Study

Abstract

Introduction

Patients with chronic urticaria (CU) may have different clinical courses of disease including periods of active CU, clinical remission, and relapse. The objective of this study was to describe representative clinical remission and relapse profiles for patients with CU.

Methods

Adults with a CU diagnosis and confirmation CU diagnosis/CU-related treatment at least 6 weeks later were identified in the Optum® de-identified Electronic Health Record dataset (2007–2018). Active CU was a period during which a patient was not in clinical remission. Clinical remission was defined as at least 12 months without CU diagnosis and/or treatment. Relapse was defined as having a CU diagnosis and/or treatment following clinical remission. A data-driven clustering algorithm grouped patients on the basis of clinical remission and relapse patterns.

Results

The 112,443 patients were grouped into four clusters. Cluster 1 (N = 36,690 [32.6%]) had the shortest median time to clinical remission (4.1 months) and lowest relapse rate (38.0%). Cluster 2 (N = 29,834 [26.5%]) reached clinical remission later (10.0 months), with a higher relapse rate (52.3%). Clusters 3 (N = 24,093 [21.4%]) and 4 (N = 21,826 [19.4%]) had the longest median times to clinical remission (33.8 and 44.6 months) and highest relapse rates (75%). Cluster 4 had the most frequent CU diagnoses and treatments, and highest comorbidity burden, polypharmacy, and resource use.

Conclusions

Patients in Clusters 3 and 4 had the lowest clinical remission and highest relapse rates relative to Clusters 1 and 2; additionally, Cluster 4 had higher resource use, more comorbidities, and polypharmacy. These cluster definitions could be used to develop a model to predict patients with relapsing and remitting patterns associated with higher disease burden who might require enhanced disease management.

Plain Language Summary

Patients with chronic urticaria (CU) may have different clinical courses of disease including periods of active CU, clinical remission, and relapse, but detailed information on patient subgroups is lacking. This study described representative profiles of clinical remission and relapse in a cohort of 112,443 patients with CU in the Optum® de-identified Electronic Health Record dataset (2007–2018). Active CU was defined as a period during which a patient was not in clinical remission (at least 12 months without CU diagnosis and/or treatment), and relapse was defined as having a CU diagnosis and/or treatment following clinical remission. A data-driven clustering algorithm was applied to group patients on the basis of clinical remission and relapse patterns. Cluster 1 had the shortest median time to clinical remission (4.1 months) and lowest relapse rate (38.0%). Cluster 2 reached clinical remission later (10.0 months), with a higher relapse rate (52.3%). Clusters 3 and 4 had the longest median times to clinical remission (33.8 and 44.6 months) and highest relapse rates (75%). Cluster 4 had the most frequent CU diagnoses and treatments, and highest comorbidity burden, polypharmacy, and resource use. The novelty of this study is that it uses a data-driven approach to extract meaningful patterns of clinical remission and relapse from the largest population of patients with CU analyzed to date.

Key Summary Points

Why carry out this study?

Chronic urticaria (CU) is an inflammatory skin condition with a heterogeneous clinical course that imposes a substantial burden on patients and the healthcare system.

This study aimed to describe representative patient profiles for different CU clinical remission and relapse patterns using cluster analysis.

What was learned from the study?

Four distinct clusters of patients with different profiles of initial active CU period, clinical remission and relapse, and different comorbidity and healthcare burden unique to each cluster were identified using a data-driven clustering algorithm.

The defined clusters differentiate various patient profiles that could inform prognostic models to predict high-risk patient profiles. Such models could be a practical tool to support a personalized disease management approach in clinical practice.

Introduction

Chronic urticaria (CU) is an inflammatory skin disorder characterized by persistent itchy wheals (hives), with or without angioedema, lasting more than 6 weeks [1]. CU symptoms have a negative impact on patients’ physical and mental health, personal life, and work [1, 2–3]. CU can co-exist with a broad spectrum of comorbidities, such as allergic rhinitis, asthma, and depression [4]. Thus, CU may impose a substantial burden on patients and society [5].

Published evidence suggests that patients experience different clinical courses of CU, including active CU periods assessed by time to clinical remission, periods of clinical remission or symptom resolution, and relapse or recurrence [6, 7–8]. The variability in the natural history of CU may be attributable to the fact that it is a heterogeneous group of conditions with differing etiologies and pathogenetic mechanisms [2]. Previous studies have shown that the proportion of patients reaching clinical remission varies, with 21–47% reaching remission at 1 year and 34–45% at 5 years [6]. The proportion of patients who relapse after reaching clinical remission also varies, with 31% of patients relapsing after an initial clinical remission period of 21 months on average [8]. In our previous work, we similarly observed varied clinical remission and relapse patterns within our study population. Although the median time to clinical remission was 11 months, 66% of patients who reached clinical remission within 3 months relapsed [6]. However, the study did not assess patterns of multiple remissions and relapses.

Patients may transition between active CU periods, clinical remission, and relapse as a function of different factors [8, 9, 10–11]. Previous studies have shown that patients who are female [12], older at CU onset [9, 12], have a longer history of CU at initial examination [12], or history of atopic dermatitis [13], angioedema [12, 14], systemic hypertension [15], or anti-thyroid antibodies [14] tended to take longer to reach clinical remission.

Our prior work showed that older age at CU diagnosis, higher body mass index, polypharmacy, and comorbidities such as asthma, allergic rhinitis, and chronic pulmonary disease (CPD) in the year before the first CU diagnosis predicted longer time to clinical remission [7]. One study showed that a positive basophil histamine release assay (for autoimmune urticaria) was associated with increased odds of clinical remission [16], while others showed that older age and male sex were associated with lower odds [10, 17]. Moreover, one study showed that receiving treatments beyond recommended first-line options such as anti-inflammatory agents, immunosuppressants, and immunotherapy was associated with higher risk of relapse [11]. Previous studies have examined specific events during the clinical course of CU [8, 11, 17, 18]. However, uncertainty still exists about the extent to which patients sharing comparable clinical remission and relapse patterns may share clinical characteristics. In addition, it is not known whether these events contribute to the burden of CU.

The objective of this study was to describe representative patient profiles for different clinical remission and relapse patterns using cluster analysis, a method used to partition and identify meaningful patterns within complex datasets [19]. The identified clusters were further characterized by burden of comorbidities, treatments, and resource use.

Methods

Data Source

The Optum® de-identified Electronic Health Record (EHR) data set (Q1 2007–Q2 2018) was used, which includes clinical information (e.g., demographics, medications prescribed and administered), healthcare/insurance claims, and other medical administrative data on over 100 million individuals who received treatment at over 2000 care centers (hospitals and clinics) across the USA. Given the variable and poorly understood natural history of CU, prospective studies of clinical remission and relapse patterns are challenging. Instead, the use of large-scale EHR data allows for the retrospective analysis of disease-related associations among a broad range of patients.

Ethical Approval

The study was considered exempt research under 45 CFR § 46.104(d)(4) as it involved only the secondary use of data that were de-identified in compliance with the Health Insurance Portability and Accountability Act (HIPAA), specifically, 45 CFR § 164.514.

Study Design and Sample Selection

A retrospective cohort design was used (Fig. S1 in the Supplementary Material). Adult patients with CU were identified on the basis of an International Classification of Diseases 9th/10th edition (ICD-9/10) diagnosis code for urticaria (Table S1 in the Supplementary Material, excluding secondary codes). A second diagnosis for urticaria and/or angioedema (Table S1 in the Supplementary Material) or a CU-related treatment (Table S2 in the Supplementary Material) at least 6 weeks apart was required as confirmation.

The index date was the date of first urticaria diagnosis. Patients were required to have at least 12 months of data pre and post index date. The baseline period was the 12 months preceding the index date during which patients had no CU diagnoses, and the follow-up period spanned the 12 months or more from the index date until the end of clinical activity, medical insurance coverage, data availability, or death, whichever occurred first. The first active CU period was defined as the period from the index date to the start of first clinical remission. Clinical remission was defined as a period of at least 12 months without CU diagnosis or CU-related treatment, with the clinical remission date defined as the start of this period. Relapse was defined as a CU diagnosis or CU-related treatment occurring after a period of clinical remission. Active CU periods (i.e., first active CU period and subsequent relapses) were those during which a patient was not in clinical remission.

Identification of Cluster Definitions

Clusters were derived using a divisive hierarchical clustering method which categorized patients into groups using one variable at a time and scored clusters using multiple variables [20]. This method was chosen to create interpretable clusters that could be described using a small number of variables. This approach has been shown to perform well for a smaller number of clusters [20].

Patients were grouped according to CU activity, clinical remission, and relapse characteristics, and all active CU periods were considered. Different cluster configurations (i.e., combinations of variables and corresponding cutoff values) were systematically tested. Initially, three variables were considered in the cluster configurations which allowed for clusters with up to three dimensions. However, the model containing the following two variables had the best performance: (1) proportion of total active CU duration during follow-up (one or several active CU periods); and (2) within active CU periods, frequency of active CU encounters (i.e., CU diagnosis/CU-related treatment). The third variable (i.e., total number of remissions and relapses) was not retained as informative, resulting in a two-dimensional model. The proportion of total active CU duration during follow-up described the clinical course of CU while accounting for variable follow-up duration between patients.

The optimal cluster configuration was defined as that which maximized intra-cluster similarity and inter-cluster dissimilarity. This was assessed using key disease characteristics post CU diagnosis which were identified on the basis of previous work on predictors of time to clinical remission [7], published literature [6, 8, 10, 17, 21, 22–23], and clinical input. These variables included age at first CU diagnosis; sex; and Charlson Comorbidity Index (CCI), presence of asthma, and presence of allergic rhinitis in the 6 months post CU diagnosis.

Outcomes

Outcomes included clinical characteristics, resource utilization, and time to first and second clinical remission and first relapse. Clinical characteristics included comorbidities commonly reported in patients with CU [4, 24] and medications found to be predictors of time to clinical remission in our prior work [7]. Resource utilization was proxied by the number of prescriptions and healthcare provider visits. All-cause and CU-related healthcare provider visits were assessed with the latter defined as visits recorded on the date of a CU diagnosis or CU-related treatment.

Statistical Analysis

For the clustering analyses, the k-prototype algorithm was used to combine information from continuous and categorical variables when calculating intra-cluster similarity and inter-cluster dissimilarity [25]. As distance metrics are sensitive to scale [26], variables were rescaled before clustering. Continuous variables were converted to percentiles to rescale both range and variance; range was rescaled so that clusters were not influenced by changes in units or scale, and variance so as not to favor splits on variables with high variance.

Model validation was performed using a holdout set. The nloptr package (version 2.0.3) in R (version 4.0.4) was used to identify the optimal model through non-linear optimization. A complete description can be found in Supplemental Methods in the Supplementary Material.

Clinical characteristics were summarized during baseline and the first 12 months of follow-up. Resource utilization and healthcare provider visits were evaluated during the full follow-up period and reported on a per-patient-per-year (PPPY) basis. Continuous variables were summarized using means, standard deviations, and medians, and categorical variables using frequency counts and percentages. Times to first and second clinical remission and first relapse were described using Kaplan–Meier (KM) analyses.

All outcomes were described for the overall population and each cluster.

Results

Patient Cluster Characteristics Based on Clinical Course of CU

A total of 112,443 patients with CU met selection criteria (Fig. S2 in the Supplementary Material). Patients were classified into four mutually exclusive clusters (Cluster 1: N = 36,690 [32.6%]; Cluster 2: N = 29,834 [26.5%]; Cluster 3: N = 24,093 [21.4%]; Cluster 4: N = 21,826 [19.4%]; Fig. 1). Patients in Clusters 3 and 4 tended to have longer initial active CU periods with longer time to first clinical remission; a lower proportion achieved clinical remission, and their relapse rate was higher relative to Clusters 1 and 2. Patients in Clusters 3 and 4 had similar clinical remission and relapse patterns, but patients in Cluster 4 had more frequent occurrences of CU diagnoses and CU-related treatments during active CU periods (Table 1). As such, patient characteristics and resource use are presented for Cluster 4 relative to Clusters 1 to 3.

[See PDF for image]

Fig. 1

Clusters of patients with CU distinguished by clinical course. a Graphical representation of patient-personas illustrating median clinical remission and relapse characteristics of each cluster profile. Relapse rate was the proportion of relapsed patients among patients reaching remission. The number of red points represents frequency of CU diagnoses/CU-related treatments within active CU periods. b Cluster characteristics. *Duration and intensity. CU chronic urticaria

Table 1. Variables used in the construction of the clustering model

	Overall	Cluster 1	Cluster 2	Cluster 3	Cluster 4
	N = 112,443	N = 36,690 32.6%	N = 29,834 26.5%	N = 24,093 21.4%	N = 21,826 19.4%
Number of clinical remissions, mean ± SD [median]	1.0 ± 0.8 [1]	1.4 ± 0.7 [1]	1.4 ± 0.7 [1]	0.4 ± 0.6 [0]	0.5 ± 0.7 [0]
Patients observed to reach clinical remissiona, N (%)	82,680 (73.5%)	36,690 (100.0%)	29,834 (100.0%)	7746 (32.2%)	8410 (38.5%)
Patients who relapsed, N (%)	41,675 (37.1%)	13,953 (38.0%)	15,616 (52.3%)	5843 (75.4%)	6263 (74.5%)
Number of relapses, mean ± SD [median]	0.5 ± 0.8 [0]	0.5 ± 0.8 [0]	0.8 ± 0.9 [1]	0.9 ± 0.7 [1]	0.9 ± 0.7 [1]
Total duration of active CU during follow-up (1 or several active CU periods, reported as proportion of total time), mean ± SD [median]b	47.1% ± 36.3% [34.7%]	10.5% ± 5.0% [10.3%]	31.4% ± 7.6% [30.5%]	87.8% ± 18.7% [100.0%]	85.4% ± 19.6% [100.0%]
Within active CU periods, frequency of active CU occurrences (i.e., CU diagnosis/CU-related treatment, reported as proportion of total time), mean ± SD [median]c	39.2% ± 23.2% [31.6%]	57.1% ± 25.9% [50.0%]	32.5% ± 14.1% [28.6%]	18.4% ± 4.3% [18.8%]	41.1% ± 16.4% [35.3%]

CU chronic urticaria, SD standard deviation

^aClinical remission was defined as ≥ 12 months without CU diagnosis and/or treatment

^bA value of X% means that X% of the observed follow-up duration was spent in periods of active CU

^cA value of Y% means that among all months during which a patient had active CU, there was ≥ 1 active CU occurrence (i.e., CU diagnosis/CU-related treatment) in Y% of these months

Demographic Characteristics and Comorbidities

The proportion of female patients was similar between clusters (74.4–79.4%). Patients in Cluster 4 were slightly older at first CU diagnosis relative to patients in Clusters 1 to 3 (mean age, Cluster 4: 50.0 years; Clusters 1–3: 46.3–48.4 years; Table 2). Patients in Cluster 4 had shorter median follow-up duration (26.3 months) relative to the other clusters (37.6–55.9 months).

Table 2. Key demographic and clinical characteristics of patients with CU

	Overall	Cluster 1	Cluster 2	Cluster 3	Cluster 4
	N = 112,443	N = 36,690 32.6%	N = 29,834 26.5%	N = 24,093 21.4%	N = 21,826 19.4%
Baseline
Female, N (%)	86,531 (77.0%)	27,284 (74.4%)	23,159 (77.6%)	18,763 (77.9%)	17,325 (79.4%)
Age at first CU diagnosis, mean ± SD [median]	47.7 ± 17.3 [47]	46.3 ± 17.3 [46]	47.0 ± 17.4 [46]	48.4 ± 17.5 [48]	50.0 ± 16.6 [50]
Race, N (%)
African American	14,211 (12.6%)	4262 (11.6%)	3839 (12.9%)	3207 (13.3%)	2903 (13.3%)
Asian	2914 (2.6%)	1079 (2.9%)	768 (2.6%)	627 (2.6%)	440 (2.0%)
Caucasian	88,111 (78.4%)	28,949 (78.9%)	23,334 (78.2%)	18,664 (77.5%)	17,164 (78.6%)
Other/unknown	7207 (6.4%)	2400 (6.5%)	1893 (6.3%)	1595 (6.6%)	1319 (6.0%)
Region, N (%)
Northeast	12,404 (11.0%)	4493 (12.2%)	3149 (10.6%)	2413 (10.0%)	2349 (10.8%)
Midwest	67,002 (59.6%)	21,563 (58.8%)	17,637 (59.1%)	14,420 (59.9%)	13,382 (61.3%)
South	21,172 (18.8%)	6536 (17.8%)	5796 (19.4%)	4726 (19.6%)	4114 (18.8%)
West	7835 (7.0%)	2706 (7.4%)	2164 (7.3%)	1691 (7.0%)	1274 (5.8%)
Other/unknown	4030 (3.6%)	1392 (3.8%)	1088 (3.6%)	843 (3.5%)	707 (3.2%)
Duration of follow-up (months), mean ± SD [median]	48.1 ± 26.1 [44.0]	58.0 ± 25.8 [55.9]	50.2 ± 24.2 [45.8]	35.0 ± 23.6 [26.3]	43.1 ± 24.5 [37.6]
First 6 months after CU diagnosis
CCI, mean ± SD [median]	0.7 ± 1.3 [0.0]	0.5 ± 1.1 [0.0]	0.6 ± 1.2 [0.0]	0.8 ± 1.4 [0.0]	1.0 ± 1.6 [1.0]
Allergic rhinitis, N (%)	22,762 (20.2%)	6246 (17.0%)	5591 (18.7%)	5005 (20.8%)	5920 (27.1%)
Asthma, N (%)	19,132 (17.0%)	4038 (11.0%)	4425 (14.8%)	4623 (19.2%)	6046 (27.7%)

CCI Charlson Comorbidity Index, CU chronic urticaria, SD standard deviation

Cluster 4 had higher comorbidity burden relative to Clusters 1 to 3, with the highest mean CCI (Cluster 4: 1.0; Clusters 1–3: 0.5–0.8; Table 2). During baseline, a higher proportion of patients in Cluster 4 had CPD, asthma, depression, allergic rhinitis, and diabetes relative to patients in Clusters 1 to 3 (Fig. 2). The same trend was observed for comorbidities newly diagnosed during the first 12 months post CU diagnosis, although inter-cluster differences were generally smaller.

[See PDF for image]

Fig. 2

Prevalence of comorbidities of each patient cluster during baseline and follow-up

Medication Use

The most widely reported CU-related medications during baseline and the first 12 months post CU diagnosis were systemic glucocorticoids (Fig. 3a). During both time periods, the proportion of patients receiving treatment was highest in Cluster 4 and lowest in Cluster 1 for various treatment types. Patients in Cluster 4 also tended to have a higher number of unique reports of both CU-related and non-CU-related medications throughout follow-up relative to patients in Clusters 1 to 3 (Fig. 3b). The most frequently reported CU-related medications were systemic glucocorticoids (Cluster 4: 2.7; Clusters 1–3: 0.7–1.4 PPPY), and sedating antihistamines (Cluster 4: 2.5; Clusters 1–3: 0.3–0.8 PPPY). Of note, non-sedating antihistamines were reported less frequently than sedating antihistamines across clusters (Cluster 4: 1.6; Clusters 1–3: 0.2–0.5 PPPY). The most frequently reported non-CU-related medications were proton pump inhibitors (Cluster 4: 2.3; Clusters 1–3: 0.4–1.0 PPPY), sympathomimetics (Cluster 4: 1.6; Clusters 1–3: 0.3–0.7 PPPY), and benzodiazepines (Cluster 4: 1.9; Clusters 1–3: 0.4–0.8 PPPY).

[See PDF for image]

Fig. 3

Reported medication use/prescription in each patient cluster during baseline and follow-up. a Proportion of patients in each cluster with prescriptions for different medication classes during the first 12 months of follow-up. b Number of CU-related and non-CU-related prescriptions PPPY during follow-up. CU chronic urticaria, PPPY per-person-per-year

H₁-antihistamine (H1AH)-based treatments were the most common first treatment post CU diagnosis prescribed for all clusters (60.1%), followed by corticosteroid-based treatments (20.8%); 14.0% of patients had no CU-related treatments within 6 months post CU diagnosis (Table S3 in the Supplementary Material). The proportion of patients receiving treatment was higher in Cluster 4 relative to Clusters 1 to 3 for the majority of treatment types received as the first treatment, including H1AH-based treatments and combinations with H₂-antihistamines (H2AH), doxepin, and leukotriene receptor antagonists (LTRAs). However, the proportion of patients receiving corticosteroids alone without any reported antihistamines as first treatment was similar between clusters.

Healthcare Provider Visits

Relative to patients in Clusters 1 to 3, patients in Cluster 4 had the highest number of all-cause healthcare providers visits during follow-up. The most frequently seen specialties included primary care providers, such as family medicine (Cluster 4: 10.7; Clusters 1–3: 4.6–7.8 PPPY) and internal medicine (Cluster 4: 7.6; Clusters 1–3: 3.3–5.0 PPPY), mid-level providers (i.e., nurse practitioners and physician assistants; Cluster 4: 7.1; Clusters 1–3: 2.5–4.8 PPPY), and emergency medicine (Cluster 4: 2.2; Clusters 1–3: 0.7–1.3 PPPY). Patients in Cluster 4 also had more visits to providers specialized in allergy/immunology (Cluster 4: 2.0; Clusters 1–3: 0.3–0.7 PPPY) and dermatology (Cluster 4: 0.8; Clusters 1–3: 0.4–0.5 PPPY) relative to patients in other clusters. Similarly, a higher proportion of patients among Cluster 4 had at least one CU-related visit relative to patients in Clusters 1 to 3 for all aforementioned providers. For all clusters, the proportion of patients with at least one CU-related visit to allergy/immunology and dermatology was lower than the proportion for other specialties (Fig. 4).

[See PDF for image]

Fig. 4

Resource utilization (healthcare provider visits) for each patient cluster. Proportion of patients with ≥ 1 all-cause (a) and CU-related (b) visit during baseline and the first 12 months of follow-up. c Number of all-cause visits during follow-up. CU chronic urticaria. ¹“Mid-level provider” included nurse practitioners and physician assistants. ²For all providers, the mean number of CU-related visits was ≤ 0.5 (median = 0)

Time to Clinical Remission and Relapse (KM Analysis)

KM analyses were performed to further characterize clusters while accounting for censoring. The patients in the four clusters differed in their clinical course of CU with patients in Clusters 3 and 4 having somewhat similar clinical course of CU. After censoring was accounted for, the median time to first clinical remission was longer for patients in Cluster 3 (33.8 months) and Cluster 4 (44.6 months) relative to patients in Cluster 1 (4.1 months) and Cluster 2 (10.0 months; Fig. 5a). All patients (100.0%) reached clinical remission in Clusters 1 and 2, while only approximately half of patients in Cluster 3 (53.0%) and Cluster 4 (42.7%) reached first clinical remission by 36 months.

[See PDF for image]

Fig. 5

Time to clinical remission and relapse. Time to first clinical remission (first active CU period) (a), relapse (b), and second clinical remission (c). CU chronic urticaria. ¹Dashed line and numbers represent median time. ²Panel b: All patients had ≥ 12 months of clinical remission per study design. Additional time patients remained in remission before the date of relapse is shown

Among patients who reached first clinical remission, time to first relapse was shorter for patients in Cluster 3 (4.2 months) and Cluster 4 (4.3 months) relative to those in Cluster 1 (55.7 months) and Cluster 2 (18.8 months; Fig. 5b). Among patients who relapsed, the median time to second clinical remission was longer for patients in Cluster 3 (36.0 months) and Cluster 4 (40.4 months; Fig. 5c) relative to patients in Cluster 1 (0.0 months) and Cluster 2 (6.2 months).

Discussion

This study used a data-driven clustering algorithm to group patients with CU according to clinical remission and relapse patterns over their clinical course of CU as observed in the EHR. It described representative patient profiles for different clinical remission and relapse patterns, focusing on comorbidities, treatments, and resource use.

The final model distinguished four patient clusters that were distinct in terms of rate and time to clinical remission and relapse, as well as the overall burden of CU. A longer initial active CU period corresponded to a higher relapse rate and shorter time to relapse. Patients in Clusters 3 and 4 had lower rates of and longer times to clinical remission relative to other clusters. As well, among patients who reached their first clinical remission, patients in Clusters 3 and 4 had higher relapse rates and longer time to second clinical remission relative to patients in Clusters 1 and 2. Although patients in Clusters 3 and 4 had similar clinical courses of CU, patients in Cluster 4 had more frequent occurrences of CU diagnoses and CU-related treatments during active CU periods. In addition, patients in Cluster 4 tended to be older, with higher comorbidity burden, higher reported medication use/prescription, and more frequent specialist visits, relative to patients in Clusters 1 to 3.

The differences between clusters observed herein are consistent with a retrospective clustering analysis in patients with CU in Korea [10]. The study by Ye et al. identified clusters differing on clinical remission patterns; patients were clustered into four groups based on disease activity (as defined by a medication score) and other clinical characteristics [10]. In their study, the median time to clinical remission and various clinical characteristics, such as laboratory values and medication use, increased with disease activity, which is consistent with the present study. Of note, relapse was not included in their clustering analysis, while in the present study, relapse was important in describing the clinical course of CU and clustering patients. In addition, the use of population-based EHR data in the present study allowed for the analysis of a larger and more diverse patient population.

The present study’s finding that patients in Cluster 4 were older at CU diagnosis relative to other clusters is consistent with Ye et al.’s where patients in clusters with longer time to clinical remission were older [10], and with other work where older age predicted longer time to clinical remission [7] and a higher probability of relapse [11]. It is possible that older patients may experience generally worse health, including more comorbidities and autoantibodies [27], which, in turn, may impact the clinical course of CU. In fact, the four clusters in the present study differed in their comorbidity burden both diagnosed pre CU and newly diagnosed post CU diagnosis. For instance, the prevalence of respiratory conditions, such as allergic rhinitis, asthma, and CPD, psychiatric comorbidities, such as depression, as well as diabetes was progressively higher as time to clinical remission increased among the clusters (prevalence was highest among patients in Cluster 4 and progressively lower among Clusters 3, 2, and 1) in the present study. This is consistent with prior findings showing that asthma, allergic rhinitis, psychiatric disease, and diabetes often co-occur with CU [10, 21, 22, 28, 29, 30–31], and asthma and allergic rhinitis recorded pre CU diagnosis predicts time to clinical remission [7]. It can be hypothesized that asthma, allergic rhinitis, and CU may share pathogenic mechanisms [10] and that, beyond atopy, urticaria may co-exist with a variety of conditions [4, 28]. There is growing evidence that aging and infections, like SARS-CoV-2 and other respiratory viruses, may lead to the production of autoantibodies [27, 32]. Since CU has been hypothesized to be caused by autoallergic type I or autoimmune type II antibodies [32], it is plausible that aging and a variety of other conditions may be associated with difficult-to-manage CU, as was observed in Clusters 3 and 4 of the present study.

In all, the findings involving comorbidities in the present study are consistent with previous studies. One exception is that Ye et al. reported that the prevalence of allergic rhinitis was higher in clusters with shorter time to clinical remission [10]. It is unclear what may explain this difference, given the differences in study design and patient populations, and overall mixed evidence in the literature on associations between patient and clinical characteristics, clinical course of CU, and treatment response [9, 10, 12, 17, 33]. Overall, however, the results of the present study demonstrate that comorbidities can characterize profiles of patients with CU created using clinical remission and relapse patterns.

Patients in Cluster 4 had higher reported use/prescription of medications and healthcare provider visits relative to other clusters. Patients in Cluster 4 had higher reported usage/prescription of both CU-related treatments, such as H1AH, H2AH, corticosteroids, LTRAs, and omalizumab, and non-CU-related medications, which is consistent with prior work. Ye et al. also reported higher CU-related medication use/prescription for groups with longer time to clinical remission [10], and our previous work found that use of sympathomimetics and benzodiazepines pre CU diagnosis predicted longer time to clinical remission [7]. Notably, the present study observed the reported use/prescription of some treatments that are inconsistent with existing US and European guideline recommendations (e.g., sedating antihistamines) [1, 34]. Prior studies have shown that patients who used alternative agents, such as anti-inflammatories, immunosuppressants, sympathomimetics, and benzodiazepines, had increased risk of relapse [7, 11].

Furthermore, patients in Cluster 4 had the most visits with primary care providers, emergency medicine, and allergy/immunology and dermatology specialists, relative to the other clusters. These findings may reflect the general health of the patients in Cluster 4 who also had higher comorbidity burden and polypharmacy. These findings further highlight that the clinical course of CU may be more complex and challenging than initially reported, and distinct profiles sharing common features can be identified. These disease profiles could suggest that CU management may be particularly challenging throughout the clinical course in the presence of other conditions. It is also possible that patients who have a variety of medical issues may be more likely to seek care.

Strengths and Limitations

This study had several strengths. First, a data-driven approach was used to extract meaningful patterns of clinical remission and relapse from a dataset comprising the largest population of patients with CU analyzed to date. By examining the full clinical course of CU rather than isolated events, the clustering algorithm model summarized a complex disease using a small number of variables. However, the study also had certain limitations. First, ICD codes for urticaria were used as a proxy for CU, which may have resulted in some misclassification. Second, because patients were only observed during periods of insurance coverage, the first observed CU diagnosis may not have been the true first. Third, patients had variable follow-up duration; however, the average follow-up for each cluster was over 24 months, which allowed observation of clinical remission and relapse periods and revealed inter-cluster differences in clinical characteristics. Nonetheless, some patients may not have been observed to reach clinical remission as a result of limited follow-up. Fourth, guideline-specified treatments are not specific to CU and could have been used for other conditions; additionally, as some treatments may have been obtained over the counter, data on medication use might be underestimated, though over-the-counter records were available for some patients. Fifth, the present analyses capture profiles of patients with CU during the study period and may not reflect those of the current population of patients with CU. Finally, there were limitations inherent to EHR data such as limited generalizability to the overall population of patients with CU and the possibility of coding errors and omissions.

Conclusions

This study identified four profiles distinguished by clinical course and burden of CU suggesting that CU has a more complex pattern of active CU periods, clinical remission, and relapses than originally reported [35]. Some patient clusters had short active CU periods, long clinical remissions, no or uncommon relapse, and low rates of comorbidities, reported treatment use/prescription and overall burden, while others had much longer active CU periods, shorter clinical remissions, higher relapse rates, more comorbidities, higher reported medication use/prescription and overall burden. Patients with a longer time to first clinical remission tended to have higher relapse rates, and vice versa. The cluster definitions of this study could be used to develop a model to predict which patients may be at higher risk of experiencing longer disease, relapse, and higher disease burden. Such models may help clinicians better understand the disease course and support personalized disease management for specific patient profiles.

Acknowledgements

Medical Writing, Editorial, and Other Assistance

Medical writing assistance was provided by professional medical writer, Janice Imai, PhD, and analytical assistance was provided by Jonathan Pearce, MSc, Eugene Baraka, MSc, and Matthew Araneta, MSc. Jonathan Pearce and Eugene Baraka are employees of Analysis Group, Inc., a consulting company that has provided paid consulting services to Novartis Pharma AG, Switzerland, which funded the development and conduct of this study and manuscript. Janice Imai and Matthew Araneta were employees of Analysis Group, Inc. at the time that this study was conducted.

Author Contributions

Irina Pivneva, Kathleen Chen, Tom Cornwall, Jimmy Royer, and James Signorovitch made substantial contributions to study conception and design, and data collection, analysis, and interpretation. Maria-Magdalena Balp, Andrii Danyliv, Dhaval Patil, Ravneet K. Kohil, and Thomas Severin contributed to study conception and design, and data analysis and interpretation. Weily Soong and Alexander M. Marsland contributed to study conception and design, and data interpretation. All authors revised the manuscript critically for important intellectual content, provided final approval of the version to published, and agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Funding

This study and Dermatology and Therapy’s Rapid Service Fees were funded by Novartis Pharma AG, Switzerland.

Data Availability

The data that support the findings of this study are available from the Optum® de-identified Electronic Health Record (EHR) data set. Restrictions apply to the availability of these data, which were used under license for this study.

Declarations

Conflict of Interest

Weily Soong and Alexander M. Marsland received a consultancy fee for their medical expertise on this project from Novartis Pharma AG. WS has served on Research, Consultant and Advisory Boards for Novartis, Allakos, Incyte, Regeneron, Sanofi, Genentech, AstraZeneca, and Amgen, and Speaker’s Bureau for Regeneron, Sanofi, AstraZeneca, and Amgen. AM has received grants/educational grants/consulting fees from Almirall, Galderma, Lilly, Novartis, Roche and UCB Pharma. Maria-Magdalena Balp, Andrii Danyliv, and Thomas Severin are employees of Novartis Pharma AG. Dhaval Patil is an employee of Novartis Pharmaceuticals Corporation. Ravneet K. Kohil is an employee of Novartis Healthcare Pvt Ltd. Irina Pivneva, Kathleen Chen, Tom Cornwall, Jimmy Royer, and James Signorovitch are employees of Analysis Group, Inc., a consulting company that has provided paid consulting services to Novartis Pharma AG, which funded the development and conduct of this study and manuscript.

Ethical Approval

The study was considered exempt research under 45 CFR § 46.104(d)(4) as it involved only the secondary use of data that were de-identified in compliance with the Health Insurance Portability and Accountability Act (HIPAA), specifically, 45 CFR § 164.514.

References

1. Zuberbier, T; Abdul Latiff, AH; Abuzakouk, M et al. The international EAACI/GA²LEN/EuroGuiDerm/APAAACI guideline for the definition, classification, diagnosis, and management of urticaria. Allergy; 2022; 77, 3 pp. 734-766. [DOI: https://dx.doi.org/10.1111/all.15090] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/34536239]

2. Sánchez-Borges, M; Ansotegui, IJ; Baiardini, I et al. The challenges of chronic urticaria part 1: epidemiology, immunopathogenesis, comorbidities, quality of life, and management. World Allergy Organ J; 2021; 14, 61:CAS:528:DC%2BB3MXhvFKmtrjP [DOI: https://dx.doi.org/10.1016/j.waojou.2021.100533] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/34221215][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8233382]100533.

3. Kolkhir, P; Giménez-Arnau, AM; Kulthanan, K; Peter, J; Metz, M; Maurer, M. Urticaria. Nat Rev Dis Primers; 2022; 8, 1 61. [DOI: https://dx.doi.org/10.1038/s41572-022-00389-z] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/36109590]

4. Papapostolou, N; Xepapadaki, P; Katoulis, A; Makris, M. Comorbidities of chronic urticaria: a glimpse into a complex relationship. Front Allergy; 2022; 3, 1008145. [DOI: https://dx.doi.org/10.3389/falgy.2022.1008145] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/36465885][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9712803]

5. Gonçalo, M; Gimenéz-Arnau, A; Al-Ahmad, M et al. The global burden of chronic urticaria for the patient and society. Br J Dermatol; 2021; 184, 2 pp. 226-236.1:CAS:528:DC%2BB3MXns1Oisrw%3D [DOI: https://dx.doi.org/10.1111/bjd.19561] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/32956489]

6. Balp, MM; Halliday, AC; Severin, T et al. Clinical remission of chronic spontaneous urticaria (CSU): a targeted literature review. Dermatol Ther (Heidelb); 2022; 12, 1 pp. 15-27. [DOI: https://dx.doi.org/10.1007/s13555-021-00641-6] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/34807372]

7. Pivneva, I; Balp, MM; Geissbühler, Y et al. Predicting clinical remission of chronic urticaria using random survival forests: machine learning applied to real-world data. Dermatol Ther (Heidelb); 2022; 12, 12 pp. 2747-2763. [DOI: https://dx.doi.org/10.1007/s13555-022-00827-6] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/36301485]

8. Stepaniuk, P; Kan, M; Kanani, A. Natural history, prognostic factors and patient perceived response to treatment in chronic spontaneous urticaria. Allergy Asthma Clin Immunol; 2020; 16, 63. [DOI: https://dx.doi.org/10.1186/s13223-020-00459-5] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/32834828][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7371813]

9. Curto-Barredo, L; Archilla, LR; Vives, GR; Pujol, RM; Giménez-Arnau, AM. Clinical features of chronic spontaneous urticaria that predict disease prognosis and refractoriness to standard treatment. Acta Derm Venereol; 2018; 98, 7 pp. 641-647.1:CAS:528:DC%2BC1MXkvFWrt7o%3D [DOI: https://dx.doi.org/10.2340/00015555-2941] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/29648675]

10. Ye, YM; Yoon, J; Woo, SD et al. Clustering the clinical course of chronic urticaria using a longitudinal database: effects on urticaria remission. Allergy Asthma Immunol Res; 2021; 13, 3 pp. 390-403. [DOI: https://dx.doi.org/10.4168/aair.2021.13.3.390] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/33733635][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7984955]

11. Kim, JK; Har, D; Brown, LS; Khan, DA. Recurrence of chronic urticaria: incidence and associated factors. J Allergy Clin Immunol Pract; 2018; 6, 2 pp. 582-585. [DOI: https://dx.doi.org/10.1016/j.jaip.2017.07.012] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/28888844]

12. Gregoriou, S; Rigopoulos, D; Katsambas, A et al. Etiologic aspects and prognostic factors of patients with chronic urticaria: nonrandomized, prospective, descriptive study. J Cutan Med Surg; 2009; 13, 4 pp. 198-203. [DOI: https://dx.doi.org/10.2310/7750.2008.08035] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/19706227]

13. Chung, BY; Um, JY; Kang, SY; Kim, HO; Park, CW. Natural history of chronic urticaria in Korea. Ann Dermatol; 2020; 32, 1 pp. 38-46.1:CAS:528:DC%2BB3cXhtlSrtb3E [DOI: https://dx.doi.org/10.5021/ad.2020.32.1.38] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/33911707]

14. Toubi, E; Kessel, A; Avshovich, N et al. Clinical and laboratory parameters in predicting chronic urticaria duration: a prospective study of 139 patients. Allergy; 2004; 59, 8 pp. 869-873.1:STN:280:DC%2BD2czivFWksA%3D%3D [DOI: https://dx.doi.org/10.1111/j.1398-9995.2004.00473.x] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/15230821]

15. Nebiolo, F; Bergia, R; Bommarito, L et al. Effect of arterial hypertension on chronic urticaria duration. Ann Allergy Asthma Immunol; 2009; 103, 5 pp. 407-410. [DOI: https://dx.doi.org/10.1016/s1081-1206(10)60360-2] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/19927539]

16. Berti, A; Yacoub, MR; Stahl Skov, P et al. Histamine release positive test associates with disease remission in chronic spontaneous urticaria: a proof-of-concept study. Eur Ann Allergy Clin Immunol; 2017; 49, 4 pp. 154-160.1:STN:280:DC%2BC1cfitVChsg%3D%3D [DOI: https://dx.doi.org/10.23822/eurannaci.1764-1489.17] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/28752718]

17. Kim, YS; Park, SH; Han, K et al. Clinical course of chronic spontaneous urticaria in the Korean adult population. Allergy Asthma Immunol Res; 2018; 10, 1 pp. 83-87.1:CAS:528:DC%2BC1MXnsFOhsL8%3D [DOI: https://dx.doi.org/10.4168/aair.2018.10.1.83] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/29178681]

18. Sánchez, J; Álvarez, L; Cardona, R. Prospective analysis of clinical evolution in chronic urticaria: persistence, remission, recurrence, and pruritus alone. World Allergy Organ J; 2022; 15, 10 [DOI: https://dx.doi.org/10.1016/j.waojou.2022.100705] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/36267098][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9554810]100705.

19. Loftus, TJ; Shickel, B; Balch, JA et al. Phenotype clustering in health care: a narrative review for clinicians. Front Artif Intell; 2022; 5, [DOI: https://dx.doi.org/10.3389/frai.2022.842306] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/36034597][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9411746]842306.

20. Chavent, M; Lechevallier, Y; Briant, O. DIVCLUS-T: a monothetic divisive hierarchical clustering method. Comput Stat Data Anal; 2007; 52, 2 pp. 687-701. [DOI: https://dx.doi.org/10.1016/j.csda.2007.03.013]

21. Kulthanan, K; Jiamton, S; Thumpimukvatana, N; Pinkaew, S. Chronic idiopathic urticaria: prevalence and clinical course. J Dermatol; 2007; 34, 5 pp. 294-301. [DOI: https://dx.doi.org/10.1111/j.1346-8138.2007.00276.x] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/17408437]

22. Vadasz, Z; Kessel, A; Hershko, AY; Maurer, M; Toubi, E. Seasonal exacerbation of asthma is frequently associated with recurrent episodes of acute urticaria. Int Arch Allergy Immunol; 2016; 169, 4 pp. 263-266.1:CAS:528:DC%2BC28XhtVSiu7zI [DOI: https://dx.doi.org/10.1159/000446183] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/27240837]

23. Magen, E; Chikovani, T; Waitman, DA; Kahan, NR. Factors related to omalizumab resistance in chronic spontaneous urticaria. Allergy Asthma Proc; 2019; 40, 4 pp. 273-278.1:CAS:528:DC%2BB3cXisVGiurg%3D [DOI: https://dx.doi.org/10.2500/aap.2019.40.4230] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/31262381]

24. Balp M, Geissbühler Y, Mcconnon A, et al. Clinical profiles of patients with CSU-a real-world study in the United States. Poster presented at: GA²LEN UCARE Urticaria Conference 2021; December 9-11 2021, Hiroshima, Japan.

25. Huang, Z. Extensions to the k-means algorithm for clustering large data sets with categorical values. Data Min Knowl Disc; 1998; 2, 3 pp. 283-304. [DOI: https://dx.doi.org/10.1023/A:1009769707641]

26. Mohamad, IB; Usman, D. Standardization and its effects on K-means clustering algorithm. Res J Appl Sci Eng Technol; 2013; 6, 17 pp. 3299-3303. [DOI: https://dx.doi.org/10.19026/rjaset.6.3638]

27. Bastard, P; Gervais, A; Le Voyer, T et al. Autoantibodies neutralizing type I IFNs are present in ~4% of uninfected individuals over 70 years old and account for ~20% of COVID-19 deaths. Sci Immunol; 2021; 6, 62 eabl4340. [DOI: https://dx.doi.org/10.1126/sciimmunol.abl4340] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/34413139][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8521484]

28. Türk, M; Ertaş, R; Zeydan, E et al. Identification of chronic urticaria subtypes using machine learning algorithms. Allergy; 2022; 77, 1 pp. 323-326. [DOI: https://dx.doi.org/10.1111/all.15119] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/34606643]

29. Konstantinou, GN; Konstantinou, GN. Psychiatric comorbidity in chronic urticaria patients: a systematic review and meta-analysis. Clin Transl Allergy; 2019; 9, 42. [DOI: https://dx.doi.org/10.1186/s13601-019-0278-3] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/31462988][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6706894]

30. Zazzali, JL; Broder, MS; Chang, E; Chiu, MW; Hogan, DJ. Cost, utilization, and patterns of medication use associated with chronic idiopathic urticaria. Ann Allergy Asthma Immunol; 2012; 108, 2 pp. 98-102. [DOI: https://dx.doi.org/10.1016/j.anai.2011.10.018] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/22289728]

31. Khan, S. Urticaria in patients with diabetes: adverse drug reaction or relapse of underlying autoimmune urticaria?. Indian J Med Res; 2019; 149, 3 pp. 423-425. [DOI: https://dx.doi.org/10.4103/ijmr.IJMR_1060_17] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/31249210][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6607826]

32. Feng, A; Yang, EY; Moore, AR et al. Autoantibodies are highly prevalent in non-SARS-CoV-2 respiratory infections and critial illness. JCI Insight; 2023; [DOI: https://dx.doi.org/10.1172/jci.insight.163150] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/37943605][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10807721]

33. Sánchez-Borges, M; Caballero-Fonseca, F; Capriles-Hulett, A; González-Aveledo, L; Maurer, M. Factors linked to disease severity and time to remission in patients with chronic spontaneous urticaria. J Eur Acad Dermatol Venereol; 2017; 31, 6 pp. 964-971. [DOI: https://dx.doi.org/10.1111/jdv.14221] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/28299827]

34. Bernstein, JA; Lang, DM; Khan, DA et al. The diagnosis and management of acute and chronic urticaria: 2014 update. J Allergy Clin Immunol; 2014; 133, 5 pp. 1270-1277. [DOI: https://dx.doi.org/10.1016/j.jaci.2014.02.036] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/24766875]

35. Beltrani, VS. An overview of chronic urticaria. Clin Rev Allergy Immunol; 2002; 23, 2 pp. 147-169.1:CAS:528:DC%2BD38XntleltLw%3D [DOI: https://dx.doi.org/10.1385/criai:23:2:147] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/12221861]