The Role of Aggressive Corticosteroid Therapy in Patients With Juvenile Dermatomyositis: A Propensity Score Analysis

Abstract

Objective

To compare outcomes at 36 months in patients newly diagnosed with juvenile dermatomyositis (DM) treated with aggressive versus standard therapy.

Methods

At diagnosis, 139 untreated juvenile DM patients were given aggressive therapy (intravenous methylprednisolone or oral prednisone 5–30 mg/kg/day; n = 76) or standard therapy (1–2 mg/kg/day; n = 63) by the treating physician. Aggressive therapy patients were more ill at diagnosis. Matching was based on the propensity for aggressive therapy because propensity scoring can reduce confounding by indication. Logistic regression of the matched data determined predictors of outcomes, controlling for clinical confounders and propensity score. Outcomes comprised Disease Activity Score (DAS) for skin and muscle, range of motion (ROM), and calcification.

Results

Sex, race, and age were similar between groups, and initial DAS weakness and ROM significantly predicted the therapy chosen. Based on propensity scores, 42 patients from each group were well matched. In the matched pairs, there were no significant differences in outcomes. Methotrexate use (odds ratio [OR] 3.6, 95% confidence interval [95% CI] 1.15–11.5) and duration of untreated disease (OR 1.2, 95% CI 1–1.38) were associated with ROM loss, hydroxychloroquine use (OR 11.2, 95% CI 3.7–33) and calcification (OR 6.8, 95% CI 1.8–25.4) with persistent rash, abnormal baseline lactate dehydrogenase (OR 11.2, 95% CI 1.4–92) and age at onset (OR 1.3, 95% CI 1–1.4) with weakness, and duration of untreated disease (OR 1.2, 95% CI 1–1.39) with calcification.

Conclusion

Using a retrospective, nonrandomized design with propensity score matching, there was little difference in efficacy outcomes between aggressive and standard therapy; however, the sickest patients were treated with aggressive therapy and were not included in the matched analysis. Comprehensive clinical studies are needed to determine therapeutic pathways to the best outcome.

Introduction

Juvenile dermatomyositis (DM) is a rare, often chronic rheumatic disease of childhood. It affects ∼3.2 children per million per year (1). The disease is the result of a systemic immune-mediated vasculopathy associated with rash, muscle weakness, and, if there is continued inflammation, development of systemic problems (2,3). Although historically this illness has had a crippling consequence in one-third of children, and has resulted in death in another one-third (4), modern treatments with prolonged corticosteroid therapy (5) or, more recently, corticosteroids in combination with other immunosuppressive agents (e.g., methotrexate [6]) have resulted in a better prognosis. Most children with juvenile DM are now expected to have a good functional recovery (6). However, the disease is often chronic, lasting many years, and may result in severe loss of range of motion (ROM), often involving calcinosis (7).

Preliminary data suggest that early aggressive therapy may lead to a more favorable outcome for patients with juvenile DM (5,8). There are many examples of rheumatic conditions in which the ultimate outcome is affected by early aggressive therapy. For example, in the treatment of rheumatoid arthritis, the Combinatietherapie Bij Reumatoïde Artritis (COBRA) study demonstrated that early aggressive use of corticosteroids in combination with immunosuppressive agents leads to better long-term outcomes (9,10). There is some evidence that adults with DM may also have a better long-term outcome if treated with aggressive therapy early in the disease course (11). However, systematic research has not been conducted to prove this assertion.

It is possible that some of the adverse consequences of chronic juvenile DM may also be affected by aggressive early therapy. Intravenous methylprednisolone (IVMP) administered in a very high dosage (e.g., 30 mg/kg/day, pulse therapy) may lead to a reduction in calcinosis as reported in 2 case series (5,8). This therapy, first reported to be successful in transplant patients (12), is thought to be useful in the treatment of adults with systemic lupus erythematosus with severe renal disease (13) and rheumatoid arthritis (14). IVMP therapy may be cost effective over the medium term when used in patients with juvenile DM (15). One possible rationale for the use of IVMP is that enteral corticosteroids may not be absorbed properly in children with juvenile DM due to proximal gut vasculopathy (16). It is known that the half-life of steroids is decreased in children (17). Furthermore, the plasma levels of corticosteroids achieved with lower oral doses may have differential effects on cell function, compared with the higher doses achieved with IVMP (18).

Randomized controlled trials (RCTs) are the gold-standard method for evaluating new therapies because they allow for an unbiased comparison of the new therapy with either a previous standard therapy or placebo. Perhaps because juvenile DM is a rare disease, no RCTs have been conducted for any therapy for juvenile DM, including high-dose IVMP. Our current understanding of therapy in juvenile DM comes from observational studies. One problem with observational data is that there is usually a reason why patients with juvenile DM are initially treated with IVMP (they are more sick, they have had a refractory course, etc.), and it is often this reason that has more to do with the eventual outcome than any therapy; this is called confounding by indication. Because of this, we need better studies on which to base our understanding of the role of early aggressive treatment of juvenile DM. We chose to test a cohort of patients with juvenile DM to determine if applying a propensity score analysis approach to data from disparate sources will adequately control for their inherent differences.

Propensity score matching is an analytical method that has been developed to reduce bias in the assessment of observationally collected data about new therapies (19). The lack of randomization in an observational study introduces a potential bias in assessing treatment effects due to imbalance of covariates and predictors between study groups (e.g., confounding by indication). Although biases due to unobserved or unmeasured factors cannot be addressed, those due to measured characteristics can be controlled to some extent using matching (20,21). Propensity score is defined as the conditional probability of being treated given an individual's characteristics. This conditional distribution is the same for individuals in the treatment and control groups, and the process results in a subset in which there is an overlap of characteristics between both groups (22). This subset of overlapping subjects can be matched based on propensity score; therefore, the propensity score serves to reduce bias by balancing the 2 groups and adjusting for confounding by indication and allows for meaningful comparisons. Use of this approach is increasing, especially for analyzing treatment data collected in large administrative databases (23–25).

Because of scarce outcome data for juvenile DM, we elected to analyze the previously coded juvenile DM research registry sponsored by the National Institute of Arthritis and Musculoskeletal and Skin Diseases (NIAMS) (1,26,27) to perform a propensity score–matched evaluation of aggressive treatment of juvenile DM with pulse IVMP at diagnosis.

Patients and Methods

Study patients

An inception cohort of untreated children newly diagnosed with juvenile DM were enrolled in a registry study sponsored by the National Institutes of Health (1994–1999) that focused on the epidemiology of juvenile DM (1,26,27). Patients were referred from 45 states, and were within 6 months of the occurrence of their first symptom of rash and/or weakness. The referring physicians were primarily pediatric rheumatologists, but also included neurologists, adult rheumatologists, pediatricians, dermatologists, and other specialists. Once patients were enrolled in the registry and had signed consent forms approved by institutional review boards, contributing physicians completed a standardized form that included physical findings and basic laboratory information. The families then participated in a 1-hour semistructured interview administered by the nurse study manager (1,26,27). Patients were included in the study if they had outcome data available at 36 months after enrollment. Our sample of eligible children comprised 139 patients. Included in the registry were 57 untreated children from 1 center (Children's Memorial Hospital [CMH] in Chicago, IL). Similar data were obtained from these children, but the selection process included screening for myositis-associated antibodies (MAAs) and myositis-specific antibodies (MSAs). MSA/MAA screening was not routinely performed at other centers. If positive for MSA or MAA, children from CMH were not enrolled in the registry study.

As in previous studies (27), identification of the first definite juvenile DM symptom, rash and/or weakness, by the caretaker was designated as the date of disease onset. The diagnosis date was defined as the day when a physician first classified the patient as having juvenile DM.

We categorized the 139 patients into 2 groups based on the dose of corticosteroid therapy administered at treatment onset. Aggressive therapy at diagnosis was defined as IVMP 30 mg/kg/day or oral prednisone 5–30 mg/kg/day. Standard therapy was defined as the prescription of lower dosages of corticosteroids (1–2 mg/kg/day) (5).

Study outcomes

The outcomes of interest were Disease Activity Scores (DAS) for skin and muscle, ROM at 36 months, and calcification. The previously published and validated DAS defines gradations in the extent and severity of skin as well as muscle involvement (28). Loss of ROM was defined as ranges that were less than expected for age-matched children by physical examination performed by physical therapists and occupational therapists, as well as one of the authors (LMP). Calcification was determined by physical examination, radiographic scan, and/or computed tomography.

Statistical analysis

Demographic and baseline clinical characteristics were compared between aggressive and standard therapy groups using chi-square tests for categorical variables and t-tests for continuous variables. The decision to use aggressive therapy was not determined in a random manner; therefore, propensity score analyses were used to compare clinical outcomes between the 2 groups. The propensity score is the likelihood that a patient would have received aggressive treatment based on the patient's observed pretreatment covariates alone. Once the score was assigned to each patient, they were matched and outcomes were compared between the groups, controlling for this probability.

Propensity score and matching

Logistic regression analysis was used to determine the predictors of receiving aggressive therapy. The candidate predictors included sex; race; age at onset; duration of untreated disease; DAS skin, DAS muscle, and total DAS scores at baseline (28); ROM at baseline; and standard muscle enzymes used for diagnosis (29,30). Results of the routine diagnostic laboratory data (creatinine phosphokinase, lactate dehydrogenase [LDH], serum aspartate aminotransferase [serum glutamic oxaloacetic transaminase (SGOT)], and aldolase) were standardized according to a previously established international method, which included cutoffs for normal ranges for each assay (31,32). A composite variable for abnormal laboratory values, defined as abnormal if at least 1 of the 4 muscle enzymes was abnormal, was also investigated. Race was categorized as white versus nonwhite.

The score statistic was used to identify the predictors that were significantly associated with receiving aggressive therapy at diagnosis, and the area under the curve (AUC) was used to quantify the predictive strength of the model. The likelihood or propensity of receiving aggressive therapy based on the identified predictors was computed for each patient.

The greedy matching algorithm (using the GMATCH macro in SAS) was used to match patients in the aggressive therapy and standard therapy groups (33,34). The samples were randomly ordered and the first match that fit the GMATCH criteria was selected. In this process, once a match was made, it remained unbroken. An inefficiency of this process is that it precludes better matches from replacing an existing match. Optimal matching was also considered to determine if the matching could be improved upon. Given the limited study sample available, one-to-one matching was used. Another approach we used was to restrict matches to specified ranges of the predictor variables. That is, an upper limit was specified for the absolute difference between the predictors in each matched pair. The distance measure chosen was the weighted sum of the absolute differences.

Analyses of primary outcomes

Descriptive statistics are presented for the matched sample. Clinical outcomes of interest were ROM, DAS skin and DAS weakness at 36 months, and calcification. In addition to the set of predictors used to model aggressive treatment at diagnosis, whether or not patients were receiving specific medications (methotrexate, hydroxychloroquine, intravenous immunoglobulin) and propensity score quintile were also included as covariates. ROM and DAS weakness at baseline were not included as additional predictors because the propensity score was a function of these 2 covariates. The range for DAS skin was 0–9 and for DAS weakness was 0–11. These data were not normally distributed and transformations did not achieve normality. In a healthy population, DAS skin and weakness are expected to be zero and calcification and loss of ROM are expected to be absent and are the target outcomes. Therefore, these scores were dichotomized as 0 versus ≥1 and logistic regression models were used for analyses. Given the large number of predictors to be studied, preliminary univariate analyses were used as a data-reduction step. At this stage, covariates that were significant at a 0.2 level were considered for the multiple regression models. If these subsets were too numerous for the available sample size, the score statistic was used to identify the best set of allowable number of predictors as determined by the available sample size. Odds ratios (ORs) and corresponding 95% confidence intervals (95% CIs) are presented. All statistical analyses were conducted using SAS software, version 9.1 (SAS Institute, Cary, NC) and conclusions were made at a 0.05 level of significance.

Results

Our 139 patients included 57 from the CMH database and 82 from the national registry. The mean ± SD age at onset was 6.18 ± 3.72 years; 101 (72.7%) were female, and 113 (81.29%) were white. Sixty-three (45.3%) patients were classified as having received standard therapy and 76 (54.7%) received aggressive therapy at diagnosis.

Baseline characteristics of entire sample

Demographic and baseline clinical characteristics and the difference between the study groups are presented in Table 1. There were no significant differences in age at onset, sex, race, and duration of untreated disease. The aggressive therapy group had higher DAS muscle and DAS total scores, and had a higher prevalence of loss of ROM. The aggressive treatment group also had a higher prevalence of abnormal aldolase, LDH, and SGOT (Table 1).

Table 1. Demographics and clinical characteristics of the entire sample^*.

	Total sample (n = 139)	Standard therapy (n = 63)	Aggressive therapy (n = 76)	P
Age at onset, years	6.18 ± 3.72	6.19 ± 4.03	6.17 ± 3.48	0.97
Female sex	101 (72.66)	45 (71.43)	56 (73.68)	0.77
White race	113 (81.29)	51 (80.95)	62 (81.58)	0.92
Duration of untreated disease, months	10.1 ± 17.81	12.32 ± 19.96	8.25 ± 15.71	0.27
Baseline disease status
DAS skin	5.72 ± 1.73	5.68 ± 1.56	5.75 ± 1.86	0.82
DAS muscle	5.71 ± 3.47	4.38 ± 3.46	6.8 ± 3.09	< 0.001
DAS total	11.42 ± 3.83	10.06 ± 3.53	12.55 ± 3.72	< 0.001
Loss of range of motion	81 (58.27)	24 (38.1)	57 (75)	< 0.001
Abnormal laboratory values
CPK, μkat/liter	87 (63.5)	40 (64.52)	47 (62.67)	0.83
Aldolase, nkat/liter	91 (72.22)	29 (56.86)	62 (82.67)	0.001
LDH, μkat/liter	92 (74.19)	31 (60.78)	61 (83.56)	0.004
SGOT, μkat/liter	108 (81.2)	42 (72.41)	66 (88)	0.023

^*Values are the mean ± SD or number (percentage) unless otherwise indicated. DAS = Disease Activity Score; CPK = creatinine phosphokinase; LDH = lactate dehydrogenase; SGOT = serum glutamic oxaloacetic transaminase.

Propensity score matching

Given the differences in baseline severity and predictors, greedy matching was used to obtain a propensity score–matched sample. Logistic regression with a stepwise selection process yielded a model in which baseline DAS weakness and loss of ROM were significant predictors of whether patients received standard or aggressive therapy. Patients with loss of ROM had an OR indicating that they were 3.6 times more likely to receive aggressive therapy over standard therapy (95% CI 1.67–7.72, P = 0.001), and each unit increase in DAS weakness was associated with a 1.2 increased odds of receiving aggressive therapy (95% CI 1.06–1.34, P = 0.004). These 2 covariates accounted for an AUC of 0.76. Using propensity scores from this regression model, one-to-one matched pairs were obtained with the restriction of DAS weakness within 3 points and the same ROM status. This algorithm yielded 42 matched pairs. There were no significant differences in baseline characteristics in the matched sample (Table 2) except for LDH, with a higher prevalence of abnormality in the aggressive therapy group.

Table 2. Demographics and clinical characteristics of the matched sample^*.

	Total sample (n = 84)	Standard therapy (n = 42)	Aggressive therapy (n = 42)	P
Age at onset, years	6.45 ± 3.9	6.14 ± 3.84	6.76 ± 3.99	0.47
Female sex	56 (66.67)	27 (64.29)	29 (69.05)	0.64
White race	67 (79.76)	33 (78.57)	34 (80.95)	0.79
Duration of untreated disease, months	10.69 ± 17.25	13.7 ± 22.9	7.68 ± 7.7	0.48
Baseline disease status
DAS skin	5.63 ± 1.58	5.45 ± 1.31	5.81 ± 1.81	0.3
DAS muscle	5.52 ± 3.34	5.43 ± 3.49	5.62 ± 3.22	0.8
DAS total	11.15 ± 3.65	10.88 ± 3.49	11.43 ± 3.83	0.5
Loss of range of motion	48 (57.14)	24 (57.14)	24 (57.14)	–
Abnormal laboratory values
CPK, μkat/liter	49 (58.33)	26 (61.9)	23 (54.76)	0.51
Aldolase, nkat/liter	55 (71.43)	22 (61.11)	33 (80.49)	0.06
LDH, μkat/liter	52 (70.27)	20 (57.14)	32 (82.05)	0.019
SGOT, μkat/liter	65 (79.27)	29 (72.5)	36 (85.71)	0.14

^*Values are the mean ± SD or number (percentage) unless otherwise indicated. See Table 1 for definitions.

Thirty-six-month outcome comparisons in matched pair groups

Aggressive versus standard therapy at diagnosis

There was no significant association between type of therapy (aggressive versus standard) and outcome. The odds of a worse outcome in the aggressive therapy group compared with the standard therapy group was 1.78 (95% CI 0.69–4.6) for ROM, 0.56 (95% CI 0.24–1.34) for DAS skin, 1 (95% CI 0.38–2.65) for DAS weakness, and 1.83 (95% CI 0.69–4.87) for calcification. Therefore, additional models looking at factors that were associated with outcome did not include therapy type as a covariate.

Range of motion

Of 84 matched patients, 25 (29.76%) had loss of ROM at 36 months. Duration of untreated disease and prescription of methotrexate at diagnosis were determined to be significant predictors. Patients who were started on methotrexate at diagnosis were more likely to have loss of ROM, reflecting either damage and/or disease acuity (80% versus 52.54%; P = 0.029). The mean ± SD duration of untreated disease among patients who did and did not experience loss of ROM was 18.63 ± 28.15 months versus 7.33 ± 7.68 months (P = 0.044). The AUC of this model was 0.75 (Table 3).

Table 3. Results of regression analyses of significant predictors in the matched sample (n = 84)^*.

Outcome	Predictor	P	OR (95% CI)	AUC
Range of motion	Duration of untreated disease†	0.044	1.18 (1–1.38)	0.72
	Methotrexate	0.029	3.63 (1.15–11.53)
DAS skin	Calcification	0.004	6.82 (1.83–25.36)	0.81
	Hydroxychloroquine	< 0.001	11.19 (3.71–33.74)
DAS muscle	Age at onset‡	0.043	1.31 (1–1.44)	0.77
	LDH	0.025	11.2 (1.36–92)
Calcification	Duration of untreated disease†	0.033	1.18 (1.01–1.39)	0.61

^*OR = odds ratio; 95% CI = 95% confidence interval; AUC = area under the curve; see Table 1 for additional definitions.

^†OR estimated for 3-month increment.

^‡OR estimated for 1-year increment.

DAS skin

Of 84 patients, 44 (52.38%) had active skin involvement at 36 months. Ever having had calcification and being treated with hydroxychloroquine at diagnosis were associated with the presence of persistent skin activity at 36 months. Patients who had calcification (78.26% versus 42.62%; P = 0.004) and those started on hydroxychloroquine (76.74% versus 26.83%; P < 0.001) were more likely to have persistent skin activity than patients who did not have calcification and who were not treated with hydroxychloroquine. The AUC of this model was 0.81 (Table 3).

DAS muscle

Twenty-two (26.2%) patients had ongoing muscle inflammation at 36 months. LDH and age at onset were associated with presence of muscle inflammation. Patients with abnormal LDH levels were more likely to have ongoing muscle inflammation (32.69% versus 4.55%; P = 0.025) than patients with normal LDH levels. The mean age at onset among patients who had muscle inflammation compared with those who did not was 5.07 ± 3.41 years versus 6.94 ± 3.98 years (P = 0.043). The AUC of this model was 0.77 (Table 3).

Calcification

Twenty-three (27.38%) patients had calcification during the 36-month followup period. Duration of untreated disease was the best predictor of calcinosis; the mean duration of untreated disease among patients who did and did not have calcinosis was 19.75 ± 29.56 months versus 7.28 ± 6.91 months (P = 0.033). The AUC of this model was 0.61 (Table 3).

Discussion

In our propensity score–matched study of a national inception cohort of patients with juvenile DM, we were unable to reject the null hypothesis that aggressive corticosteroid therapy for moderately ill children at diagnosis does not differ from standard therapy in improving outcomes at 36 months. Furthermore, no standard assessment of the safety of the 2 initial treatments was routinely documented. We were, however, able to determine a number of important factors associated with or predictive of outcome. Most notably, a longer duration of untreated disease at diagnosis leads to unfavorable outcomes, as previously observed (27). However, it is not clear if this longer duration of untreated disease represents an insidious onset subtype with worse outcomes, lack of access to medical care, or a true effect of delayed therapy. Not surprisingly, the requirement for adjunctive treatments such as methotrexate and hydroxychloroquine at diagnosis may identify patients who are more likely to have unfavorable outcomes.

One strength of our study is the availability of a carefully collected data set from a national registry. The demographics of our cohort suggest that our results can be extended to juvenile DM patients in North America (3) and Western Europe.

We used propensity score matching to reduce bias that results from confounding by indication. Our matching algorithm produced a study sample that was very well matched for baseline disease severity. However, several points must be considered when interpreting our findings. First, we were only able to study the impact of a type of therapy at a single point in time (diagnosis) on the outcome at a single point in time (36 months). We were not able to match on other potentially important indicators of disease severity at diagnosis, such as nailfold capillary density or immune parameters.

Second, given our small sample, we were unable to find good matches for those patients with severe baseline disease (who were all treated aggressively) and those with mild baseline disease (who were all treated with standard therapy) at diagnosis. Because we could only match a subset of our initial cohort, our sample size was reduced and our power to find differences was limited. Furthermore, the patient groups may not be fully matched, for some children with MAA or MSA who are therefore more prone to chronic disease activity may have been included in one group and excluded from the other group. Although the null hypothesis may, in fact, be true, we must also consider that our results may reflect a Type II (false-negative) error. However, the confidence intervals around treatment effects are concordant with our clinical observations regarding aggressive therapy at diagnosis.

Third, we can only adjust, using the propensity score matching, for observed and measured covariates. It is certainly plausible that unknown or unmeasured covariates may confound the relationship between aggressive therapy and outcomes, thus leading to bias. An RCT would be the best method to control for unmeasured covariates, and would be inclusive of the full range of patients' disease severity.

Fourth, the fact that the majority of patients who received IVMP were from one institution could have been another source of confounding. However, administration of IVMP was based on indicators of disease severity.

It is important that our data not be used to draw inferences regarding children with severe juvenile DM. The use of methotrexate and hydroxychloroquine at diagnosis was associated with worse ROM and DAS skin score at 36 months, respectively. Children with more severe disease often receive these treatments, which are thought to be effective (5,6,35). A likely explanation for these findings is confounding by indication. For example, hydroxychloroquine is often used to treat patients' juvenile DM with the most resistant or severe skin disease; it is not surprising that those patients are most likely to have ongoing skin activity at 36 months (36). Review of the other medications used to initially treat children with juvenile DM did not identify any significant association with DAS skin score at 36 months.

Increased duration of untreated disease at the time of diagnosis was a significant predictor of worsened ROM and calcinosis. This association has been reported previously from the NIAMS registry (27), as well as in a study of the composition of calcifications, in which chronic skin inflammation was associated with calcifications, as seen in the present study (37).

While our study was observational in nature, we were fortunate to have carefully collected registry data, although it was restricted to data at diagnosis and 36 months thereafter. We were able to successfully apply propensity score matching as a bias reduction strategy. This approach can serve as a model for future studies of therapies for rare disorders, where funding, infrastructure, and patient availability for RCTs may be problematic. Our numbers were small and our matched sample did not include patients with severe disease. Comprehensive studies may require the identification of more sensitive indicators of continuing disease activity that can be used to guide therapy in an objective manner.

This useful and promising method of propensity scoring should be applied to future observational studies that collect larger samples of children with DM in order to determine the safety and efficacy of aggressive treatment.