Longitudinal course of cognitive impairment in patients with atopic dermatitis

Atopic dermatitis (AD) is a common inflammatory skin disease. Cognitive dysfunction was recently demonstrated to be increased in adults and children with AD. However, little is known about the longitudinal course of cognitive impairment in AD and its relationship with pruritus. To investigate the longitudinal course and predictors of cognitive impairment in AD a prospective dermatology practice-based study was performed using questionnaires and evaluation by a dermatologist (n = 210). Patients with ≥ 2 visits were included (mean follow-up time: 318 days). Cognitive function was assessed using the Patient-Reported Outcomes Measurement Information System (PROMIS) Cognitive Function 8-item Short-Form. At baseline, 20.85% of patients had PROMIS T scores ≤ 45, indicating cognitive impairment (CI). Among patients with CI at baseline, 34.09% had persistent CI, 47.72% had a fluctuating course, and 18.18% had sustained improvement of cognitive function. In repeated-measures regression models, cognitive function scores declined overtime in patients with worse AD severity [SCORing Atopic Dermatitis (SCORAD): p = 0.01, Atopic Dermatitis Severity Index: p = 0.001], increased itch (p = 0.01), skin pain (p = 0.001), and sleep disturbance (p = 0.001). In multivariable logistic regression models, persistent CI was associated with female gender and depressive symptoms [moderate-to-severe Patient Health Questionnaire-9 score (PHQ9)]. Latent class analysis identified two classes of cognitive dysfunction: normal cognition (77.23%), moderate dysfunction (16.21%) and severe impairment (6.55%). Black/African-American race (p = 0.02), moderate-to-severe SCORAD (p = 0.03), dermatology life quality index (p < 0.0001), PHQ9 (p < 0.0001), itch (p = 0.02) and skin pain (p < 0.0001) were more likely to experience moderate dysfunction or severe cognitive impairment. AD is associated with a heterogeneous longitudinal course of cognitive function in adults, with some patients experiencing persistent CI over time.

Cognitive function is an important neuropsychiatric domain. It encompasses perception, attention, memory, decision making, and language comprehension. Cognitive function heavily influences quality of life and health behaviors [17,18]. Clinically, cognitive dysfunction reduces adherence to medications, self-care and hygiene, ability to follow clinician instructions, and increases injury [19][20][21].
Previous studies showed higher rates of cognitive dysfunction in adults and children with AD [22,23]. However, none assessed the longitudinal course and persistence of cognitive dysfunction, and how cognitive function fluctuates with dynamic AD severity. Some aspects of cognitive dysfunction may be modifiable with treatments that reduce AD severity. However, others may suffer cognitive dysfunction secondary to neuropsychiatric disorders that do not improve with the treatment of AD. Improved understanding of the phenotypical and temporal patterns of cognitive function and their relationship with AD severity may help differentiate patients whose cognitive function can be improved in the dermatology setting versus those who warrant referral to a specialist. This study examined the longitudinal course of cognitive function in adult patients with AD and associations of persistent cognitive dysfunction over time.

Study design
A longitudinal, dermatology practice-based study was performed between 2013 and 2019 with patients ≥ 18 with AD defined by the Hanifin-Rajka diagnostic criteria [24]. Participants were enrolled from an eczema clinic at an academic medical center. Exclusion criteria included those without a definitive diagnosis of AD (determined by the Hanifin-Rajka criteria), an alternative diagnosis (determined by skin biopsy and/or epicutaneous patch testing), or an inability to complete the questionnaires. Virtually all (> 99%) patients who were invited agreed to participate. Patients enrolled in this study were evaluated at baseline and received follow-up visits every 6 months. Throughout the study, patients received standard of care including emollients, topical and systemic medications, and phototherapy as indicated. This study was approved by the institutional review boards of Northwestern University.

Outcome measures
Self-administered questionnaires were completed by patients of the eczema clinic at an academic medical center before their encounter. Questionnaires included the following: Numeric Rating Scale (NRS) for average and worse itch, NRS skin pain, Patient Oriented Eczema Measure (POEM), Dermatology Life Quality Index (DLQI), and with a fullbody skin examination by one of the authors, a dermatologist (JIS). Eczema Area and Severity Index (EASI), SCORing AD (SCORAD) and Atopic Dermatitis Severity Index (ADSI) were the clinician-reported or composite outcomes examined. Surveys were administered between January 2017 and February 2019. The Northwestern University Institutional Review Boards approved the study, and informed consent was obtained electronically.

Statistical analysis
Summary statistics were estimated for baseline population characteristics. Clinical cognitive impairment was defined by a PROMIS score of 45 or lower. Logistic regression models were constructed with persistent impairment as the binary dependent variable, and age, gender, race, education, Patient Health Questionnaire (PHQ)-9 score as the independent variables.
Repeated-measures linear regression models were constructed to examine the baseline predictors of declining PROMIS cognitive function T scores over time. Assumptions for linear regression models were all met, including linearity, homoscedasticity, independence, and normality. Histograms of residuals and quantile-quantile plots showed approximately normal distribution. Models were constructed with the PROMIS Cognitive Function T scores as the dependent variable and age, gender, race, education, anxiety, depression, atopic conditions, symptoms, composite signs and symptoms, and AD signs as the independent variables. Multivariate models adjusted for age, gender, race, education, anxiety and depression. Adjusted β coefficients and 95% confidence intervals (CI95) were estimated.
Sankey plots were generated to present the fluctuation of cognitive impairment over time, and for cognitive impairment-itch intensity over time. The analysis was performed using multiple imputations by chained equations (MICE). Assumptions were met, including missing observations that were missing at random. All variables from the repeatedmeasures analysis model were included in the imputation model.
Latent class analysis (LCA) was used to examine phenotypical patterns of binary variables of each component in the PROMIS cognitive function test. LCA uses observed categorical or binary data to identify patterns, termed latent classes [25]. Subjects with similar response patterns are categorized into specific classes. Conditional probabilities were estimated using maximum likelihood to characterize the latent classes by indicating the chance that a member would give a certain response (never/rarely, sometimes, or often/very often) for the specific item. Conditional probabilities are included, where probabilities closer to 0 or 1 indicate lower or higher chances, respectively. The number of latent classes and best-fitting models were selected by minimizing the Bayesian information criterion (BIC), Akaike information criterion (AIC) statistics and interpretability. The statistical analyses were performed in SAS 9.4 software (SAS Institute, Inc, Cary, NC). Chi-squared models were utilized to examine the differential effects of individual variables across classes.

Population characteristics
Overall, 192 patients were included in the study [mean ± std. age: 46.28 ± 18.40 years; 55.68% female; 65.76% white race; 56.35% had a college degree and baseline SCORAD: 32.29 ± 16.99 (Supplemental Table 1)]. At baseline, 79.15% of patients had normal PROMIS cognitive function T scores, 20.85% of patients met the criteria for clinical cognitive impairment.

Heterogeneous time course of cognitive impairment in AD
Among patients with baseline normal cognitive function, 12.57% had impaired cognition at ≥ 1 follow-up visits. 83.83% continued to have normal cognitive function, 14.37% had a fluctuating course, 1.80% consistently declined (Table 1). One in five (20.85%) patients had cognitive impairment at baseline (PROMIS cognitive function T score ≤ 45). Of these, 59.09% experienced cognitive impairment at ≥ 1 follow-up visit. Approximately one-third (34.09%) experienced a persistent course, almost half (47.72%) a fluctuating course, and 18.18% experienced consistent improvements in cognition (Table 1). Substantial fluctuation of cognitive function occurred over time (Fig. 1A). There was a complex relationship of itch severity and cognition over time (Fig. 1B). Most patients experiencing cognitive impairment concomitantly reported moderate-severe itch at all times. Though, after 12 months of follow-up, all patients reporting cognitive impairment reported moderate-severe itch.

Two distinct phenotypes of cognitive dysfunction in adults with AD
LCA was used to identify patterns of cognitive function. The best-fit model had three classes based on minimal Bayesian information criterion (BIC), and Akaike information criterion (AIC), and interpretability. Class 1 consisted of normal cognition, defined by > 90% reports of never/ rare dysfunction in all variables including brain fog, brain slowing, trouble shifting, diminished concentration, inattention, trouble forming thoughts, having to work harder to keep track, and impact on quality of life (77.23% of patients) (Supplemental Table 2). None met the criteria for clinical impairment of PROMIS cognitive function T score ≤ 45. Class 2 consisted of moderate dysfunction, defined by moderate difficulties with brain fog, brain slowing, trouble shifting, diminished concentration, inattention and having to work harder to keep track, but never/rarely struggled with forming thoughts (16.21% of patients). Class 3 was consistent of severe impairment, defined by frequent dysfunction in all variables (6.55% of patients). There were significant differences of PROMIS cognitive function T scores between the two latent classes (moderate: 42.26 ± 1.89, severe: 34.25 ± 5.21, Kruskal-Wallis, P < 0.0001). Membership in class 2 or 3 (moderate dysfunction or severe impairment) was significantly increased in Black/African-American vs. White patients (P = 0.02), moderate-severe SCORAD (P = 0.03), DLQI (P < 0.0001), PHQ9, NRS itch and skin-pain (P < 0.0001 for all) ( Table 3).

Discussion
This study identified two phenotypes of cognitive dysfunction in adults with AD, including moderate dysfunction and severe cognitive impairment. Moderate and severe symptoms of cognitive dysfunction, including difficulty with attention and concentration, commonly occur in individuals with ADD/ADHD. Indeed, previous studies found that AD is associated with higher prevalence of ADD/ADHD in both children and adults [1,22,26]. AD patients with Black/African-American race and moderate-severe SCO-RAD, DLQI, PHQ9, NRS itch and skin-pain were more likely to have moderate or severe cognitive dysfunction. The association of Black/African-American race with moderate or severe cognitive dysfunction coincides with previous reports, indicating that older Blacks/African-Americans may be more vulnerable to mild cognitive impairment [27]. The association of Black/African-American race with cognitive impairment may be due to under treatment of AD due to underappreciation of AD severity in skin of color, and to an underlying more severe AD phenotype. Indeed, previous studies found more persistent and severe AD in Black/ African-American children [28,29]. Previous studies also found that African-American race was one of the strongest predictors of a pattern of persistently active AD during adolescence and early childhood [30]. African-American patients are also more likely to present with a treatmentresistant lichenified phenotype, and some studies suggest that reliance on erythema scores may mask the severity [31,32]. Additionally African-American patients has been observed to display higher values for CRP, ferritin, and blood eosinophils suggesting a higher propensity for broad systemic inflammation [33]. In addition to a larger degree of systemic inflammation, some studies suggest that AD in African-American patients tends to be skewed toward more of a Th2/Th17/Th22 profile, while others suggest that  Th17 is absent in African-American AD [33,34]. If AD in African-American patients is in fact skewed toward a Th17 profile, the differential cytokine expression may additionally contribute to potential differences in the cognitive burden of AD. Moreover, previous studies posed a potential biological connection for stress with racial disparity in cognitive dysfunction [35]. Future efforts are needed to identify the causes of and address disparities in AD among other health disorders.
The mechanism of association between AD and cognitive dysfunction is not fully elucidated, but may be bidirectional. Chronic pain is associated with elevated levels neuroinflammation and central sensitization [36]. Central cytokines and chemokines that are upregulated in AD are powerful neuromodulators and may play a role in inducing hyperalgesia and allodynia [36]. Neuroinflammation may increase itch, pain and cognitive dysfunction [37]. In the brain, microglia activation and aberrant cytokine signaling are implicated in itch, pain, and cognitive impairment. Multiple cytokines and chemokines may play a central role in itch and cognitive dysfunction [1,37]. Additionally, cognitive dysfunction may worsen AD by reducing adherence to trigger avoidance and treatment [19,20,23]. In addition to reducing the physical burden of AD, early and aggressive treatment may improve the quality of life and avert the dynamic interplay of AD inflammation and cognitive dysfunction.  (4) 25 (13) AD was associated with a heterogeneous longitudinal course of cognitive impairment. For most patients, cognitive dysfunction was responsive to improvements of itch and AD severity [23]. This was not the case for a small but substantial proportion of patients. This novel subset of non-responders warrants further investigation. It may be that such patients have cognitive impairment for reasons unrelated to AD. Alternatively, cognitive dysfunction secondary to AD may resolve with prompt AD treatment. However, chronic cognitive dysfunction associated with long-standing uncontrolled AD may persist even after eventual treatment of the AD. This is concerning because mild cognitive impairment has been linked to increased risk of permanent dementia, especially in individuals with older age, lower educational attainment and mental health symptoms [38]. Persistent cognitive impairment in the setting of AD and perhaps other chronic inflammatory diseases may increase risk of dementia.
This study has some limitations, including being performed in a single urban academic, dermatologic setting and may not be generalizable to the US population. Some outcomes were assessed by self-report, including history of asthma, other atopic comorbidities, depression, anxiety, and sleep disturbances. Data were not collected for potentially confounding factors, including history of cognitive dysfunction, consumption of alcohol or use of medications that might cause cognitive dysfunction. Larger-scale multicenter studies including controls without AD are needed to confirm the results of this study.
In conclusion, AD is associated with a heterogeneous longitudinal course of cognitive function in adults, with some patients experiencing persistent CI over time.
Declining cognition was associated with moderate-tosevere depression, Black/African-American race, moderate-severe SCORAD, DLQI, PHQ9, NRS itch and skin pain. These results provide demographic and AD characteristics that can help identify patient populations that may be higher risk to suffer from cognitive dysfunction as a comorbidity and/or complication of their AD.
Author contributions LJC was responsible for data analysis, manuscript drafting.JIS was responsible for data collection, supervision, manuscript drafting. Funding This publication was made possible with support from the Agency for Healthcare Research and Quality (AHRQ) grant number K12 HS023011, and the Dermatology Foundation.
Data availability The participants of this study did not give consent for their data to be shared publicly. So, supporting data is not available.