Abstract
CONTEXT: Infantile hemangiomas (IH) may be associated with significant functional impact.
OBJECTIVE: The objective of this study was to meta-analyze studies of pharmacologic interventions for children with IH.
DATA SOURCES: Data sources were Medline and other databases from 1982 to June 2015.
STUDY SELECTION: Two reviewers assessed studies using predetermined inclusion criteria.
DATA EXTRACTION: One reviewer extracted data with review by a second.
RESULTS: We included 18 studies in a network meta-analysis assessing relative expected rates of IH clearance associated with β-blockers and steroids. Oral propranolol had the largest mean estimate of expected clearance (95%; 95% Bayesian credible interval [BCI]: 88%–99%) relative to oral corticosteroids (43%, 95% BCI: 21%–66%) and control (6%, 95% BCI: 1%–11%). Strength of evidence (SOE) was high for propranolol’s effects on reducing lesion size compared with observation/placebo. Corticosteroids demonstrated moderate effectiveness at reducing size/volume (moderate SOE for improvement in IH). SOE was low for effects of topical timolol versus placebo.
LIMITATIONS: Methodologic limitations of available evidence may compromise SOE. Validity of meta-analytic estimates relies on the assumption of exchangeability among studies, conditional on effects of the intervention. Results rely on assumed lack of reporting bias.
CONCLUSIONS: Propranolol is effective at reducing IH size compared with placebo, observation, and other treatments including steroids in most studies. Corticosteroids demonstrate moderate effectiveness at reducing IH size/volume. The meta-analysis estimates provide a relative ranking of anticipated rates of lesion clearance among treatments. Families and clinicians making treatment decisions should also factor in elements such as lesion size, location, number, and type, and patient and family preferences.
- BCI —
- Bayesian credible interval
- IH —
- infantile hemangioma
- RCT —
- randomized controlled trial
- SOE —
- strength of evidence
Infantile hemangiomas (IH) are the most common tumors of childhood. IH are benign but possess potential for permanent local tissue damage, ulceration, infection, bleeding, functional impact, and pain. Because of historical inconsistencies in naming conventions, the true prevalence of IH is unclear, but it is estimated that they affect ∼4% to 5% of children,1 with higher prevalence in females and Caucasians.2,3 In most children, IH will become apparent in the first few weeks of life and reach 80% of total size by ∼3 to 5 months.4,5 With expectant observation, many children may experience a complete or near complete involution without significant sequelae; however, IH frequently occur in cosmetically and functionally sensitive areas. Even with complete involution, some patients have permanent disfigurement and functional compromise.6 Furthermore, some lesions are particularly aggressive or morbid and can cause severe pain, ulceration, and bleeding even in early stages.7,8 Early assessment of the extent of the hemangioma, and early, appropriate treatment of IH may potentially mitigate these complications; however, in a large multicenter treatment analysis, the first specialist visit for infants and children did not occur until a mean age of 5 months.5 The rapid growth of IH leaves little time for prospective observation to determine which IH will lead to complications and require specialist attention and treatment before complications begin to manifest.
Specific disease characteristics, such as lesion size, location, rate of growth, and persistence as well as child age, functional impact, and IH subtype influence whether children are treated with pharmacologic agents or surgically. Lesions that possess immediate risk for morbidity or mortality, such as IH obstructing the airway or visual axis, may require immediate surgical intervention. Lesion characteristics such as size, location, and type (eg, superficial, deep) also influence treatment choice. Both medical and surgical treatment paradigms contain significant variability and lack of consensus.
The β-blocker propranolol was approved by the US Food and Drug Administration for use in IH in March 2014.9–11 Propranolol was historically used in children for cardiac conditions, and off-label use to treat IH began after the serendipitous discovery of its effects on IH lesions in 2008.12 Before this discovery, corticosteroids were the drug of choice, but propranolol has become the typical choice for initial medical management in children without contraindications. Steroids may be used in children with contraindications to β-blockers or who do not respond to β-blockers.
This meta-analysis, 1 component of a broader systematic review of pharmacologic and surgical interventions for IH funded by the Agency for Healthcare Research and Quality, summarizes the evidence for the effectiveness of pharmacologic treatments for IH. The full review, which addresses surgical interventions and harms of all treatments, and review protocol (PROSPERO registration: CRD42015015765) are available on the agency’s Effective Health Care Web site.13
Methods
Information Sources
We searched the Medline database via PubMed, the Cumulative Index of Nursing and Allied Health Literature (CINAHL), and Embase from 1982 to June 2015 by using a combination of controlled vocabulary and key terms related to interventions for IH (eg, infantile hemangioma, corticosteroid, propranolol). We also hand-searched the reference lists of included articles and recent reviews of interventions for IH to identify potentially relevant articles. Complete search strategies are available in the full review.
Eligibility Criteria
We developed inclusion criteria in consultation with an expert panel of clinicians and researchers (Table 1). We included comparative study designs (eg, randomized controlled trials [RCTs], prospective or retrospective cohort studies) to address the effectiveness of treatments and comparative studies and case series with at least 25 participants to evaluate adverse effects. Studies were included in the meta-analysis subset if they satisfied the following additional inclusion criteria:
Inclusion Criteria
Outcomes were reported quantitatively, using an objective metric for reporting intervention effects that could be converted into a proportion of IH clearance.
One or more study arms evaluated a single intervention; study arms in which ≥2 treatments were applied were excluded.
Reported outcomes were accompanied by an associated measure of variation or precision.
Noncontrol pharmacologic treatments could be reasonably classified into 1 of the following classes of agents: oral, intralesional, or topical propranolol; intralesional triamcinolone; topical or ophthalmic timolol; and oral steroid.
Data Extraction and Analysis
One team member initially extracted study design, study population characteristics (age, gender, IH site and type), intervention characteristics (dosage, duration, route), and baseline and outcome data on constructs of interest from eligible studies. A second team member reviewed extracted data for accuracy and completeness.
Study outcomes were reported in a variety of ways. Most identified an arbitrary threshold of IH clearance (eg, >75%) as a positive outcome or divided the continuous clearance measure into a small number of categories. Others reported visual analog scale scores, either for entire study arms or for individual patients within study arms. To incorporate as many studies as possible, by minimizing the number excluded due to technical constraints on statistical integration, we constructed a Bayesian latent variable model.14 This model allowed several types of outcome data and a suite of pharmacologic interventions to be analyzed in the same model, thereby maximizing the power for estimating parameters precisely. The estimands of interest were the expected proportion of clearance for each intervention agent, along with associated posterior uncertainty.
For studies that reported the number of outcomes below or above a prespecified threshold, the proportion in each group were incorporated into the analysis using the inverse of the normal cumulative distribution function. For a given clearance threshold, there is a corresponding expected number of observations below and above that threshold, under a particular model for the distribution of clearance rates, which affords an opportunity to use these data to inform the parameters of that model. We used a logit-linear mixed-effects model, with intervention type as a fixed effect and study label as a random effect, combining to predict the distribution of clearance rates. Outcomes from studies reporting mean reduction in size or surface area were included directly in this model, rather than indirectly via the inverse cumulative distribution function method described earlier.
Study Selection and Study Characteristics
We identified 4132 potentially relevant titles or abstracts, with 1273 proceeding to full text review (Fig 1). After dual screening, 18 met all criteria for quantitative synthesis and were included in the meta-analysis (Table 2). Studies included 2 RCTs and 1 cohort study evaluating oral propranolol (doses ranging from 1–4 mg/kg/day) and placebo or observation15–17; 1 RCT compared different forms of propranolol (oral 2 mg/kg/day, intralesional 1 mg, topical 1%).18 Two RCTs and 2 cohort studies compared oral propranolol (2.0–2.7 mg/kg/day) and oral prednisolone or oral prednisone (2–4 mg/kg/day).19–22 One cohort study compared oral propranolol (2 mg/kg/day) and intralesional bleomycin (0.5 mg/kg)23; and 1 RCT and 1 cohort study evaluated oral propranolol (2–3 mg/kg/day) and other β-blockers (1–4 mg/kg/day).24,25
Disposition of studies identified for this review. *Numbers do not tally as studies could be excluded for multiple reasons.
Characteristics of Included Studies
Three cohort studies and 2 RCTs assessed topical timolol (0.5%) compared with placebo or observation or another agent,26–30 and 1 RCT and 1 cohort study evaluated different steroids, including oral prednisone and intralesional triamcinolone.31,32 Studies included a total of 1265 children with IH (ages ranging from 2 weeks to 9 years), and most were fair quality.16–18,20–22,25,28,30
Studies compared pre- and posttreatment photographs to assess improvement in IH and reported change using varied metrics. Three studies used a visual analog scale.22,24,28 Ten studies used a subjective rating of response (eg, good, fair, poor; complete, partial, no response), typically with a percentage improvement related to each grade (eg, good ≥75% size reduction).15,16,18,20,23,25,26,29–31 Two studies rated improvement as greater or less than either 50%32 or 75%.21 The additional studies reported percent reduction,27 change in total surface area,19 or mean percentage decrease in size.17 Fourteen studies explicitly reported using masked assessors to rate changes in IH.15–17,19–22,24,25,27,29–32 All studies included IH in multiple anatomic locations, and lesion types (when reported) also varied, with the exception of most studies of topical timolol, which generally included only superficial lesions. Supplemental Table 5 outlines key study characteristics.
Assessment of Study Quality and Strength of Evidence
Two investigators independently evaluated the methodologic quality of studies using separate tools appropriate for specific study designs (eg, the Newcastle-Ottawa Quality Assessment Scale for cohort studies).33–35 A senior reviewer resolved discrepancies in quality assessment. We considered 4 studies as good quality15,19,27,31 and 5 as poor quality.23,24,26,29,32 Two investigators also graded the strength of the body of evidence (confidence in the estimate of effect) by using methods based on the Methods Guide for Effectiveness and Comparative Effectiveness Reviews.36 The full team reviewed the final strength of evidence (SOE) designation with discussion to reach consensus as needed. Table 3 describes these ratings.
SOE Grades and Definitions
Results
We calculated mean expected clearance rates associated with each agent and confidence bounds around the estimates (Fig 2). The expected efficacy of control arms was estimated to be 6% (95% Bayesian credible interval [BCI]: 1%–11%), that is, we would expect to see, on average, 6% clearance of IH in children who receive placebo or no treatment during the study period. All noncontrol treatments were estimated to have a larger expected clearance than control. The largest mean estimate of clearance was for oral propranolol (95%, 95% BCI: 88%–99%). Clearance associated with the use of oral steroids was 43% (95% BCI: 21%–66%), thus providing a clearance rate intermediate to control and use of β-blockers. Triamcinolone, an intralesional injectable steroid, had a higher clearance rate than oral steroids, with wide BCI (58%; 95% BCI: 22%–99%). Few data were available for intralesional propranolol, which is reflected in its larger credible interval (estimated clearance: 9%; 95% BCI: 0%–45%). With fairly wide confidence bounds and limited data in some areas, the relative differences among these estimates are of greater importance than absolute effects in interpreting these results.
Estimates of expected IH clearance. Estimates of expected IH clearance are expressed as percent clearance relative to initial condition for each treatment, along with associated posterior interquartile range (thick lines) and 95% credible interval (thin lines).
Figure 3 presents the variability in effects seen across the patient populations in terms of percent clearance. Oral propranolol was estimated to have the largest variability in clearance rate, with some patients experiencing much greater clearance than others (σ = 2.5, 95% BCI: 2.1–2.9) and timolol (σ = 1.5, 95% BCI: 1.4–1.6), intralesional triamcinolone (σ = 1.8, 95% BCI: 1.3–2.3), and oral steroids (σ = 1.3, 95% BCI: 1.1–1.6) yielding similar, lower estimates. All of the estimates of effect SD were at least nominally higher than the control SD, which may be a reflection of the heterogeneity of the study population in terms of response of IH to treatment.
Estimates of the variation of each treatment. Estimates of the variation of each treatment are expressed as SD, along with associated posterior interquartile range (thick lines) and 95% credible interval (thin lines).
Because of relatively sparse information from several treatment agents, we were unable to separately estimate variance parameters for all of the interventions and instead fit a simplified model that assumed variances were equal. To check the validity of this assumption, we also fit a model on the subset of interventions with sufficient numbers of studies (>3) to estimate variance parameters and noted that the variance estimates ranged from 1.3 (1.1–1.6) to 2.6 (2.2–2.9) on the logit scale. This was reasonably close to the 1.8 (1.1–2.6) estimated as the pooled variance. To assess for methodologic heterogeneity, we ran additional models with only RCTs and with only good and fair quality studies included. Estimates did not differ markedly when poor-quality studies were removed, although BCI typically widened; thus, we report the model with poor-quality studies included. To examine the possible effect of bias due to the inclusion of cohort studies, we fit the same model to RCTs only. The resulting estimates were similar to those of the model fit to all studies but with much wider posterior credible intervals. Because there was no obvious systematic bias due to study design, we report the model estimates based on the entire body of evidence.
On the basis of meta-analysis results and results of individual studies included in the full review, our assessments of SOE ranged from low to high for the interventions and comparators evaluated (Table 4). We considered the SOE to be high for the greater effectiveness of propranolol compared with placebo or observation and moderate for the superiority of propranolol compared with steroids in achieving IH clearance. We also considered SOE as low for topical timolol versus observation.
SOE
Discussion
The literature on pharmacologic approaches for the treatment of IH is heterogeneous in terms of populations, interventions, comparators, and outcomes. Comparative studies included individuals with ages of <1 month to 9 years, and lesion types and locations varied across studies. Most studies included children with IH in multiple anatomic locations and of multiple types (eg, deep, superficial) typically without stratifying outcomes on these characteristics. Most studies did not clearly describe diagnostic criteria and few clearly noted whether previous treatment had been administered (n = 4 or 18 studies). Comparators also varied across studies and included placebo, observation, historical control groups, and other active interventions. Outcome measures similarly differed. Although studies generally assessed change in lesion size or appearance, scales and methods varied.
Until fairly recently, corticosteroids were the treatment of choice for IH. Corticosteroids demonstrate some effectiveness but are associated with clinically significant side effects. More recently, β-blockers, and propranolol specifically, have been studied and recommended for use. Studies of propranolol compared its effectiveness with placebo/observation, with corticosteroids and other modalities, and with other β-blockers. Relative to observation or placebo arms, oral propranolol was consistently superior in individual studies and in our meta-analysis. Relative to other active modalities, we find that propranolol is generally superior with the exception of no significant differences in reducing lesion size in 1 study comparing it with steroids19 and 1 comparing it with bleomycin.23 Finally, given that propranolol has been demonstrated to be associated with positive outcomes, the question of whether effectiveness is associated with propranolol specifically or β-blockers in general has been studied. Although there were only 2 small studies of these agents in our meta-analysis, early results for atenolol and nadolol are as positive as those noted for propranolol, and we believe that they suggest that these and potentially other β-blockers may also be effective, potentially with fewer side effects. These findings, however, are preliminary and require further investigation.
Studies of topical timolol, typically used to treat superficial IH, also reported greater effectiveness for timolol compared with placebo/observation in reducing IH lesion size.26,27 One study comparing ophthalmic timolol and imiquimod reported no differences in effects.28 Two studies using laser as a comparator reported mixed results: no differences in average overall improvement in 1 study comparing timolol and timolol plus laser modalities30 and greater response to timolol in superficial IH, with greater response of mixed IH to timolol plus laser in another.29
In our network meta-analysis specifically, all noncontrol treatments were estimated to have a larger expected clearance than control arms. The largest mean estimate of expected clearance was for oral propranolol (95%, 95% BCI: 88%–99%), followed by timolol (62%, 95% BCI: 39%–83%) and intralesional triamcinolone (58%, 95% BCI: 22%–93%), albeit with wider confidence bounds. Oral steroids had a clearance rate of 43% (95% BCI: 21%–66%). The preponderance of available evidence used in the meta-analysis was derived from studies of propranolol and corticosteroids.
Our findings are limited by several factors. We included only studies published in English and did not seek unpublished data. We were also dependent on the characterization of IH as presented in each study. Given changes in nomenclature and variations in the way IH are described and treated, it may be that some studies included non-IH lesions or that some used older treatment regimens that may no longer be commonly used. We also note that other approaches to meta-analysis could be used but that our estimates of a high anticipated response to propranolol largely align with those in other reviews of propranolol.38–41
The evidence base for IH treatment is limited by a small number of comparative studies including a limited number of participants. A growing number of studies address β-blockers, but current studies are limited by a general lack of long-term follow-up and analyses to explore differences in response among subgroups. Few comparative studies addressed steroids, and indications for steroid treatment compared with β-blockers are unclear. Studies are also limited by the use of multiple and variable outcome measures to assess resolution of lesions. Because no objective laboratory value or other measures exist to determine size changes, investigators have developed multiple techniques, and studies did not always report scales or other approaches clearly. The variety of scales (eg, percentage change, mean change, visual analog scale) makes combining outcomes challenging. Similarly, studies typically included multiple lesion types in multiple locations, which also makes understanding potential differences in response difficult. In addition, few studies reported baseline physical characteristics of the lesion, so understanding the magnitude of change reported is challenging. Most studies included children with problematic IH (10 of 16 clearly reporting this information), so change was likely substantial, and parents and children may value any lessening of lesion size or change in color or texture.
The most important deficiency in the reported outcomes across studies is the tendency for the reporting of discretized outcomes, when the underlying outcome is a continuous variable. Specifically, although outcomes are likely recorded as a continuous measure (ie, the proportion of an existing lesion that is cleared or reduced in size after treatment), authors often chose an arbitrary cutoff proportion (or a small number of “bins”) and reported only the numbers in each of the resulting categories. This results in an immediate and unrecoverable loss in power for any quantitative meta-analyses. Researchers should be encouraged to report outcome variables as they were recorded, without transforming them in such a way that information is lost.
Other areas for future research include a better understanding of appropriate dosing and timing of propranolol treatment and optimal duration of β-blocker use. In addition, characteristics such as lesion size, location, and persistence, as well as modifiers such as patient age, functional impact, and IH subtype influence whether children are treated with pharmacologic agents or surgically. Lesion characteristics also influence the choice of specific pharmacologic agents. Most studies included multiple lesion types and in multiple locations, and few included specific modifier analyses or reported outcomes by lesion characteristics. Research to improve understanding of which lesions are likely to respond best to specific agents is critical, especially given that understanding of the effectiveness of β-blockers in the involution phase is limited. Optimal treatment in the proliferative phase may be key to maximal resolution of IH.
Conclusions
Corticosteroids demonstrate moderate effectiveness in reducing IH size/volume. Propranolol is effective at reducing the size of IH, with high SOE for effects on reducing lesion size and compared with placebo, observation, and other treatment methods, including steroids, in most studies. The meta-analysis estimates reported here provide a relative ranking of anticipated rates of lesion clearance among treatment options. Families and clinicians making treatment decisions should also factor in elements such as lesion size, location, type, and number, which may affect choice of treatment modality, as well as patient/family preferences.
Acknowledgments
The authors thank the technical experts and review team who contributed to the full report. The authors gratefully acknowledge the valuable input and guidance from Christine Chang, MD, MPH, Task Order Officer with the Agency for Healthcare Research and Quality.
Footnotes
- Accepted November 2, 2015.
- Address correspondence to Sivakumar Chinnadurai, MD, MPH, Department of Otolaryngology, 702 Doctor’s Office Tower, Vanderbilt University Medical Center, Nashville, TN 37203. E-mail: sivakumar.chinnadurai{at}vanderbilt.edu
Dr Chinnadurai contributed to the conceptualization and design of the original review and meta-analysis and helped assess studies, verify data, develop the analysis, and draft the initial manuscript; Dr Fonnesbeck contributed to the conceptualization and design of the original review and meta-analysis, conducted the meta-analysis, and helped verify data and draft the initial manuscript; Drs Likis and McPheeters contributed to the conceptualization and design of the original review and meta-analysis and helped assess studies, develop the analysis, and draft the initial manuscript; Drs Morad and Snyder contributed to the conceptualization and design of the original review and meta-analysis and helped assess studies, verify data, develop the analysis, and draft the initial manuscript; Ms Sathe contributed to the conceptualization and design of the original review and meta-analysis and helped evaluate studies, extract and verify data, develop the analysis, and draft the initial manuscript; and all authors approved the final manuscript as submitted.
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose. The authors of this report are responsible for its content. Statements in the report should not be construed as endorsement by the Agency for Healthcare Research and Quality or the US Department of Health and Human Services.
FUNDING: Supported by the Agency for Healthcare Research and Quality, US Department of Health and Human Services (contract HHSA290-2012-00009-I).
POTENTIAL CONFLICT OF INTEREST: The authors have indicated they have no potential conflicts of interest to disclose.
References
- Copyright © 2016 by the American Academy of Pediatrics
In this issue
