To provide preliminary validation of the Cutaneous Assessment Tool (CAT), a new tool to assess cutaneous manifestations of juvenile dermatomyositis (DM), and to explore the clinical meaning of CAT scores.
To provide preliminary validation of the Cutaneous Assessment Tool (CAT), a new tool to assess cutaneous manifestations of juvenile dermatomyositis (DM), and to explore the clinical meaning of CAT scores.
Children with juvenile DM (n = 113) were assessed at baseline and 7–9 months later (n = 94). Internal consistency, redundancy, construct validity, and responsiveness of the CAT were examined. CAT scores corresponding to ordinal global assessments were determined.
Item-total correlations ranged from 0.27–0.67 for activity lesions present in ≥10% of patients; item-domain and domain-total correlations ranged from 0.25–0.99. Cronbach's alpha was 0.79 for the CAT activity score and 0.74 for the CAT damage score. As predicted, the CAT activity score correlated strongly with both global disease activity and skin disease activity and moderately with the Childhood Myositis Assessment Scale, whereas the CAT damage score correlated moderately with the physician global disease and skin disease damage scores. Median CAT activity scores of 1, 7, 13, 18, and 31 corresponded to absent, mild, moderate, severe, and extremely severe skin disease activity, respectively. Median CAT damage scores of 0, 1, 2, and 5 correlated with the same descriptions of damage (severe and extremely severe combined).
Preliminary validation of the CAT demonstrated good internal consistency, nonredundancy, good construct validity, and appropriate responsiveness. The CAT is a comprehensive, semiquantitative assessment tool for skin disease in juvenile DM.
Juvenile dermatomyositis (DM) is a serious, idiopathic, autoimmune illness that causes inflammation and dysfunction of muscle (1). Cutaneous involvement is a cardinal feature and can represent both disease activity and disease damage of the underlying myositis syndrome or its treatment (2, 3).
Cutaneous manifestations of juvenile DM are important in the assessment of this illness. We have previously demonstrated that global skin disease activity correlates moderately with measures of physical function in children with juvenile DM (4, 5). In adults with DM, Hundley et al have demonstrated that active skin disease is associated with reduced quality of life (6). Skin damage caused by chronic skin lesions, such as calcinosis, is an important source of morbidity in juvenile DM, and may be associated with pain, disfigurement, and reductions in quality of life and physical function (7). For these reasons, the assessment of skin disease is part of the core sets of outcomes recommended for studies in adult and juvenile DM (8, 9). Both of these proposed core sets recommend the assessment of extramuscular disease in the overall assessment of disease activity and damage. Skin disease is an important component of the measurement of extramuscular disease, and the Cutaneous Assessment Tool (CAT) helps with its comprehensive assessment.
We have previously reported on the development and interrater reliability of the CAT, a comprehensive tool for assessing the cutaneous manifestations of adult and juvenile idiopathic inflammatory myopathies (IIM) (10). The goal of the present study was to provide further data on the validity of this tool, including internal consistency, item redundancy, construct validity, and responsiveness, and to provide additional information regarding the clinical meaning of the CAT including the relationship between skin and muscle disease in children with juvenile DM.
The CAT was developed by an interdisciplinary group that included adult and pediatric rheumatologists and a dermatologist experienced in the cutaneous assessment of myositis and other autoimmune disorders. Lesions were included that the investigators considered to be important in the assessment of skin activity in IIM. These lesions were consistent with the classification of DM skin lesions described by Sontheimer (3). The CAT comprises 21 items, with 10 representing disease activity, 4 representing disease damage, and 7 that include elements of both disease activity and damage, which are scored separately. This tool can be found on the following Web site: https://dir-apps.niehs.nih.gov/imacs/index.cfm?action=home.imacsforms.
Each cutaneous lesion listed in the CAT had a specific definition for the lesion and the characteristics to be assessed. Depending on the lesion, there were between 2 and 7 possible responses corresponding to increasing levels of activity or damage. Lesions were weighted by assigning a priori scores based on consensus expert opinion on the relative importance of individual lesions in contributing to activity or damage. Individual activity and damage item scores were summed to calculate the CAT activity score (range 0–96) and the CAT damage score (range 0–20). Higher scores corresponded to greater degrees of activity or damage.
A total of 113 patients with juvenile DM with definite or probable myositis (based on the criteria by Bohan and Peter ) whose age at onset was <18 years were assessed by 11 pediatric rheumatologists at 10 tertiary care pediatric rheumatology centers. Prior to enrolling patients in the study, all assessors reviewed a detailed 35-mm slide atlas defining the cutaneous manifestations of the IIM. All assessors also attended 1 or 2 training sessions at meetings of the Juvenile Myositis Disease Activity Collaborative Study Group.
This cohort was consecutively enrolled at any point in their disease course, as previously described (4, 5, 12). At the time of enrollment, participants had a median disease duration of 19 months (range 0–137 months, 25th–75th percentile 8–33 months), a median global disease activity measured by a 10-cm visual analog scale (VAS) of 2.1 cm (range 0–9.7 cm, 25th–75th percentile 0.6–4.3 cm), and a median global disease damage of 1.2 cm (range 0–10 cm, 25th–75th percentile 0–1.5 cm). In 94 children, a second assessment was performed 7–9 months later. Institutional review board approval was obtained at each center, and informed consent was obtained from all participants' parents/legal guardians.
At each assessment, standardized history, physical examination, and laboratory investigations were obtained. Physician global assessments of disease activity and damage and skin disease activity and damage were obtained using a 10-cm VAS (12). These were completed by the examining physician at the same time as the CAT. Physician global skin disease activity and damage were also assessed separately with 5-item ordinal scales as previously described (12).
The Childhood Myositis Assessment Scale (CMAS) (5, 13), the Childhood Health Assessment Questionnaire (C-HAQ) (4), and manual muscle testing (MMT) (4) were obtained as previously described. Due to age and time constraints, MMT was not obtained in all children. Serum levels of at least 1 muscle-associated enzyme were measured (creatinine kinase, lactic acid dehydrogenase, or aldolase). To facilitate comparison, all values were divided by the upper limit of normal for the laboratory in which the test was performed.
All analyses were performed using the statistical program SAS, release 8.02 (SAS Institute, Cary, NC). Internal consistency was assessed using 2 methods. First, adjusted item-total, item-domain, and domain-total Spearman's correlations were calculated. These correlations were adjusted by performing the calculation with each item or domain sequentially removed from the domain or total score. Second, Cronbach's alpha was calculated for the total scores, and with each item deleted from the total score. Cronbach's alpha was calculated using standardized variables. Cronbach's alpha scores >0.70 were considered to show good internal consistency (14).
Redundancy of items was assessed using 2 methods. First, Spearman's correlations among activity items and damage items were calculated. Values >0.70 were considered to show high degrees of redundancy between items. Second, Cronbach's alpha was calculated for the CAT activity and damage scores. Values >0.90 were considered to show redundancy within the tool (15).
Construct validity of the CAT was assessed by Spearman's rank correlations of the CAT activity and damage scores with other measures of disease activity and damage. A priori predictions about the strength of correlations of the CAT activity and damage scores with other measures were made. Spearman's correlations ≥0.70 were defined as strong, whereas correlations ≥0.40 and <0.70 were defined as moderate and correlations <0.40 were defined as poor. Specifically, it was predicted that the CAT activity score would correlate strongly with physician global skin disease activity, correlate somewhat less strongly with physician global disease activity, correlate moderately with measures of muscle function (CMAS, C-HAQ, MMT), and correlate poorly with measures of disease damage (physician global skin disease and overall disease damage) and serum levels of muscle enzymes. It was predicted that the CAT damage score would correlate strongly with physician global skin disease damage, somewhat less strongly with physician global disease damage, and poorly with physician global skin disease activity and overall disease activity, based on the predicted poor relationship between skin disease activity and damage. Congruence of observed Spearman's correlations with those predicted was taken as evidence of construct validity (4, 5, 15).
Responsiveness of the CAT was assessed with the standardized response mean (SRM) for the CAT activity and damage scores (16). A physician global assessment of skin disease activity of at least 0.8 cm was considered an external standard of change in activity. A value of 0.8 cm was chosen based on work suggesting that a 0.8-cm improvement in a pain VAS is associated with the perception of improvement in children with arthritis (17). The SRM was calculated for patients who met the external standard of change (CAT activity score only) and for the entire group of children (both scores). For comparison, the SRMs for the physician global disease and skin disease activity and damage assessments were also calculated. SRM values of 0.8 were considered large, 0.5 moderate, and 0.2 small (16).
The clinical meaning of CAT scores was assessed by calculating the CAT activity and damage scores corresponding to the 5-item skin disease activity and damage ordinal scores. Discordance in the degree of skin and muscle disease activity was assessed by constructing 3 × 3 tables of low, medium, and high skin disease activity versus low, medium, and high muscle disease activity for both baseline and followup visits. Discordance was defined as the presence of high or low skin disease activity with low or high muscle disease activity, respectively. For skin disease activity, low was defined as a CAT activity score ≤7 (median CAT activity score for mild skin disease activity) and high was defined as a CAT activity score ≥18 (median CAT activity score for severe skin disease activity). For muscle disease activity, low was defined as a CMAS score ≥45 (median CMAS score associated with mild disability ) and high was defined as a CMAS score ≤30 (median CMAS score associated with moderate disability ). The proportion of patients with less substantial discordance (difference in skin and muscle disease activity of one step only, such as low skin disease activity with moderate muscle disease activity) was also calculated. Marginal homogeneity for the baseline and followup 3 × 3 tables was assessed using the Stuart-Maxwell test, which tests the null hypothesis that the marginal proportions of the table are the same for both variables.
The median CAT activity score was 7 (25th–75th percentile 3–11, range 0–44). The median CAT damage score was 1 (25th–75th percentile 0–1, range 0–13).
Item-total correlations for the CAT ranged from 0.02 to 0.67 for the activity items and from 0.001 to 0.29 for the damage items (Table 1). The items with low correlations were generally those present in few patients, and they improved to a minimum of 0.27 (P < 0.05) for lesions with >10% endorsement. Item-domain correlations for the activity items ranged from 0.25 to 0.99 and increased to a minimum of 0.42 (P < 0.05) for lesions with >10% endorsement. As expected, item-domain and domain-total correlations were higher than item-total correlations.
|Proportion (no. with lesion)||Item-total correlation||Item-domain correlation||Domain-total correlation|
|Characteristic lesions domain||0.70†|
|1A Gottron's papules or sign erythematous changes||0.56 (63)||0.60†||0.85†|
|2A Heliotrope||0.52 (59)||0.67†||0.89†|
|Erythematous lesions domain||0.60†|
|3A Malar/facial erythema||0.52 (59)||0.56†||0.89†|
|4A Linear extensor erythema||0.10 (11)||0.36†||0.42†|
|5A V sign||0.12 (14)||0.46†||0.54†|
|6A Shawl sign||0.15 (17)||0.47†||0.62†|
|7A Non-sun exposed erythema||0.18 (20)||0.44†||0.61†|
|8A Erythroderma||0.04 (4)||0.27‡||0.33‡|
|Vasculopathic lesions domain||0.54†|
|9A Livedo reticularis||0.08 (9)||0.12||0.34‡|
|10A Ulceration||0.04 (5)||0.25‡||0.36†|
|11A Mucous membrane lesions||0.24 (27)||0.39†||0.58†|
|12A Periungual capillary loop changes||0.64 (72)||0.48†||0.83†|
|Hand lesions domain||0.26‡|
|13A Mechanic's hands||0.03 (3)||0.14||0.32‡|
|14A Cuticular overgrowth||0.41 (46)||0.27‡||0.99†|
|Other active lesions domain||0.36‡|
|15A Subcutaneous edema||0.12 (14)||0.37‡||0.86†|
|16A Panniculitis||0.01 (1)||0.02||0.25‡|
|17A Alopecia||0.04 (5)||0.14||0.50†|
|1D Atrophy or hypo/hyperpigmentation in a distribution of Gottron's papule or sign||0.26 (29)||0.22‡||NA|
|2D Atrophy or hypo/hyperpigmentation in a distribution of heliotrope||0.11 (12)||0.001||NA|
|3D Atrophy or hypo/hyperpigmentation in a distribution of malar/facial erythema||0.06 (7)||0.17‡||NA|
|4D Atrophy or hypo/hyperpigmentation in a distribution of linear extensor erythema||0.03 (3)||0.27‡||NA|
|5D Atrophy or hypo/hyperpigmentation in a distribution of V sign||0.03 (3)||0.29‡||NA|
|6D Atrophy or hypo/hyperpigmentation in a distribution of shawl sign||0.04 (4)||0.15||NA|
|7D Atrophy or hypo/hyperpigmentation in a distribution of non-sun exposed erythema||0.03 (3)||0.25‡||NA|
|18D Poikiloderma vasculare atrophicans||0.04 (5)||0.14||NA|
|19D Calcinosis||0.15 (17)||0.07||NA|
|20D Lipoatrophy||0.06 (7)||0.10||NA|
|21D Depressed skin scarring||0.12 (13)||0.13||NA|
The standardized Cronbach's alpha for the CAT activity score was 0.79, demonstrating good internal consistency. When individual items were removed from the calculation of the CAT activity score, the resulting standardized Cronbach's alpha scores ranged from 0.77 to 0.81. The standardized Cronbach's alpha for the CAT damage score was 0.74, also demonstrating good internal consistency. When individual items were removed from the calculation of the CAT damage score, the resulting standardized Cronbach's alpha scores ranged from 0.67 to 0.76.
Spearman's correlations of individual activity lesions with other activity lesions were all <0.60 (range 0.001–0.59), while the correlations of individual damage lesions with other damage lesions were all <0.70 (range 0.02–0.66). As noted above, Cronbach's alpha for the CAT activity and damage scores did not exceed 0.90. These results suggest that the items in the CAT are not overly redundant.
Consistent with our predictions, the CAT activity score correlated strongly with both physician global disease and skin disease activity, moderately with the CMAS and the C-HAQ, and poorly with physician global disease and skin disease damage and serum muscle enzymes (Table 2). Correlation of the CAT activity score with the MMT was poor, which was somewhat lower than predicted.
|CAT activity score||CAT damage score|
|Physician global assessment of skin disease activity||96||0.83†||0.20‡|
|Physician global assessment of disease activity||98||0.77†||0.14|
|Childhood Myositis Assessment Scale||98||−0.48†||−0.19|
|Childhood Health Assessment Questionnaire||96||0.40†||0.13|
|Total manual muscle testing score||50||−0.36‡||−0.38‡|
|Physician global assessment of skin disease damage||95||0.18||0.53†|
|Physician global assessment of disease damage||113||0.13||0.52†|
|Serum creatinine kinase||97||0.03||−0.06|
|Serum lactate dehydrogenase||65||0.37‡||−0.19|
The correlations of the CAT damage score with the physician global disease and skin disease damage scores were moderate, which were somewhat lower than predicted. Correlations of the 2 most frequently observed damage lesions with physician global disease and skin disease damage were no better (rs = 0.20, P = 0.03 and rs = 0.29, P = 0.005, respectively, for atrophy or hypopigmentation/hyperpigmentation in the distribution of Gottron's papules or sign; rs = 0.67, P < 0.0001 and rs = 0.56, P < 0.0001, respectively, for calcinosis). Correlations with physician global disease and skin disease activity were poor, consistent with predictions.
The SRM for the CAT activity score was 0.67 (95% confidence interval [95% CI] 0.42, 0.92) for those children who improved >0.8 cm in physician global assessment of skin disease activity (n = 63). The SRM was 0.52 (95% CI 0.32, 0.72) for all children with a 7–9-month followup evaluation (n = 94). These values represent moderate responsiveness. In comparison, the SRMs for physician global disease and skin disease activity were 0.54 (95% CI 0.31, 0.77) and 0.50 (95% CI 0.27, 0.73), respectively.
The SRM for the CAT damage score was 0.02 (95% CI –0.18, 0.22) for all children with a 7–9-month followup evaluation (n = 94). This represents essentially no responsiveness, which was expected over the duration of this study. Similarly, the SRMs for physician global disease and skin disease damage were 0.12 (95% CI –0.08, 0.32) and 0.25 (95% CI 0.02, 0.48), respectively.
Figures 1 and 2 show the CAT activity and damage scores, which correspond to 5-point ordinal global assessments of skin disease activity and damage. The median CAT activity scores (25th–75th percentiles), which corresponded to “no evidence of skin disease activity,” “mild,” “moderate,” “severe,” and “very severe skin disease activity,” were 1 (0–3), 7 (4–9), 13 (10–20), 18 (12–33), and 31 (27–39), respectively. The median (25th–75th percentiles) CAT damage scores, which corresponded to “no evidence of skin disease damage,” “mild,” “moderate,” and “severe or very severe skin disease damage,” were 0 (0–1), 1 (0–2), 2 (1–4), and 5 (3–6), respectively.
When discordance of skin and muscle disease was considered, the results shown in Table 3 were obtained. Five (5%) of 98 children were discordant at baseline, with either high skin disease activity and low muscle disease activity or high muscle disease activity and low skin disease activity. Forty-six (47%) of 98 patients had minor discordance (for example, low skin disease activity with moderate muscle disease activity); 47 (48%) of 98 patients were concordant at baseline, with the same category of skin and muscle disease activity. At the followup visit, 1 (1%) of 83 patients was discordant, 30 (36%) of 83 patients had minor discordance between skin and muscle disease activity, and 52 (63%) of 83 patients were concordant. The test of marginal homogeneity (Stuart-Maxwell) failed to show a difference in the marginal probabilities at baseline (χ2 = 0.09, 2 df, P = 0.95) but indicated a difference at followup (χ2 = 6.55, 2 df, P = 0.04). This result appeared to be related to a greater number of patients with low skin disease activity (low CAT activity score) while the CMAS continued to show a moderate degree of muscle disease activity. A greater proportion of patients appeared to improve in skin disease activity than in muscle disease activity between the 2 assessments.
|CAT activity score||CMAS score||Totals (%)|
|Low muscle activity (≥45)||Medium muscle activity||High muscle activity (≤30)|
|Baseline visit (n = 98)|
|Low skin activity (≤7)||33||16||2†||51 (52)|
|Medium skin activity||14||11||8||33 (34)|
|High skin activity (≥18)||3†||8||3||14 (14)|
|Totals (%)||50 (51)||35 (36)||13 (13)||98 (100)|
|Followup visit (n = 83)|
|Low skin activity (≤7)||44||18||1†||63 (76)|
|Medium skin activity||7||5||1||13 (16)|
|High skin activity (≥18)||0†||4||3||7 (8)|
|Totals (%)||51 (61)||27 (33)||5 (6)||83 (100)|
The CAT is a new tool for the assessment of skin disease activity and damage in patients with IIM. This tool fills an important gap in the assessment of these diseases by providing a method to comprehensively and quantitatively assess the spectrum of cutaneous manifestations, which are important in understanding the activity and damage associated with these illnesses. In this report, we have provided preliminary validation of this tool in a large population of children with juvenile DM. We have also explored the clinical meaning of CAT scores by determining the CAT scores that correspond to different levels of global skin disease activity and damage and exploring the relationship between skin and muscle disease activity.
The CAT was developed to include lesions that are important in the assessment of IIM. This included both common lesions, such as periungual capillary loop changes and Gottron's papules, and rare manifestations that are important in the assessment of severity and outcome, such as cutaneous ulceration and calcinosis. Therefore, it was not surprising that there was a wide range in the frequency of observed lesions. It was expected that some lesions in this tool would be endorsed rarely, but they were retained based on their significance when present. Children who participated in this study were at varied stages of their illness. This was necessary to capture aspects of both skin disease activity and damage, but may have affected the reported frequencies of individual lesions. It is likely that a cohort of children being studied at disease onset would have higher frequencies of some activity lesions.
Internal consistency, as assessed by Cronbach's alpha, was good for the CAT activity and damage scores. Item-domain and domain-total correlations were moderate to high, and all were significant. Individual item-total correlations for items of the CAT ranged more widely, but were higher and significant for all items observed in at least 10% of the patients. Low item-total correlations for those items not frequently observed are not necessarily evidence that the skin lesion does not perform well in the CAT.
Redundancy of items in the CAT was investigated using both Spearman's correlations and Cronbach's alpha. Although the items in the CAT are certainly related to one another, consistent with clinical experience the results show that the items in the CAT are not excessively redundant.
For the CAT activity score, the predictions used to assess construct validity were generally consistent with the observed correlations between the CAT activity score and other measures of disease activity and damage. As expected, the CAT activity score correlated strongly with physician global assessments of skin disease activity and overall disease activity, and moderately with the measures of muscle strength and function. Overall, these results indicate good construct validity of the CAT activity score. The CAT damage score also appeared to show good construct validity, with the observed correlations of the CAT damage score with other measures being consistent with those predicted. The CAT damage score correlated moderately with measures of disease damage (physician global assessment of skin disease damage and overall disease damage) and poorly with measures of disease activity. Even when only the most common damage lesions were considered, the correlations were no better. It is possible that the correlations of the CAT damage score with other measures may have been somewhat lower due to the small number of study participants who had features of skin damage.
Responsiveness of the CAT activity score was moderate, which suggests that the CAT would be able to detect changes in skin disease activity in children with juvenile DM. Responsiveness of the physician global disease and skin disease assessments was quite similar. This makes the CAT activity score potentially useful in clinical trials and for longitudinal assessment of skin disease activity. In comparison, the CAT damage score exhibited very poor responsiveness, which was similar to that observed for the physician global disease and skin disease damage assessments. This is consistent with expectations, because skin damage develops slowly and would not be expected to change greatly over the 7–9-month followup of this study. The responsiveness of the CAT damage score over longer periods will require further investigation. Due to the significant floor effects (i.e., skewing of CAT data toward relatively low skin disease activity and damage scores), it is likely that our results for responsiveness are conservative. In patients with more active skin disease, such as might be encountered in a clinical trial, it is likely that responsiveness would be greater.
By providing CAT activity and damage scores that correspond to physician global skin disease activity and damage scores assessed using an ordinal scale, we have provided a context in which to understand the meaning of CAT scores. The ordinal scale used descriptors that are easily understood, such as “severe skin disease activity,” thereby helping to give meaning to the CAT scores obtained in patients with juvenile DM.
Our results also show that the relationship between skin and muscle disease in children with juvenile DM is only moderate. We have demonstrated that some children with active skin disease can have low or moderate muscle disease activity and that the opposite can also be true. This result is consistent with the work by O'Connor et al, who demonstrated a correlation between up-regulation of the type 1 interferon-induced gene myxovirus resistance protein and muscle disease in patients with untreated juvenile DM, but no correlation with skin disease (18), and the work by Smith et al, who showed that nailfold capillary abnormalities were associated with skin disease but not muscle disease in juvenile DM (19). This emphasizes the importance of considering skin disease activity independent of muscle disease in the overall assessment of children with juvenile DM. Our results may also suggest a greater tendency for improvement in skin disease activity compared with muscle disease activity in a cross-sectional cohort such as this.
Although we attempted to perform as careful a study as possible, this work has some potential limitations. Some lesions in the CAT were observed rarely, and it is possible that they should be deleted from future versions of this tool. First, however, there is a need to further examine the distribution of these lesions in adult polymyositis and DM, as well as a more active population of patients with juvenile DM. Second, although we studied a large number of children with juvenile DM, they had limited disease damage. This made it difficult to fully investigate the validity of the CAT damage score. Future work will need to involve patients who have had juvenile DM for longer periods, and follow them for extended periods to ensure that changes in skin disease damage have an opportunity to occur. Finally, the CAT is relatively long, occupying 8 pages. Some of the length of the CAT is related to the inclusion of detailed definitions and instructions on the use of the tool, rather than having this information in separate appendices or glossaries. However, this may limit acceptance of the tool in nonresearch contexts. Future work will look at possible changes in format or scoring to determine if the tool may be shortened without negative effects on the measurement characteristics.
In conclusion, we have introduced a new, systematic, and semiquantitative tool for the assessment of both acute and chronic cutaneous changes in patients with juvenile DM. Preliminary validation of the CAT has demonstrated good internal consistency, nonredundancy, good construct validity, and appropriate responsiveness. We have also provided some clinical context for the interpretation of CAT scores and explored the relationship between skin and muscle disease activity in juvenile DM.
Dr. Huber had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study design. Dugan, Lachenbruch, Feldman, Miller, Rider.
Acquisition of data. Dugan, Feldman, Perez, Zemel, Lindsley, Rennebohm, Wallace, Passo, Reed, Bowyer, Ballinger, Rider.
Analysis and interpretation of data. Huber, Dugan, Lachenbruch, Feldman, Miller, Rider.
Manuscript preparation. Huber, Dugan, Feldman, Miller, Rider.
Statistical analysis. Huber, Lachenbruch.
We thank Drs. Edward Cowen and Gregory Dennis for valuable feedback on the manuscript.