Abstract
BACKGROUND: An important step toward improvement of the conduct of pediatric clinical research is the standardization of the ages of children to be included in pediatric trials and the optimal age-subgroups to be analyzed.
METHODS: We set out to evaluate empirically the age ranges of children, and age-subgroup analyses thereof, reported in recent pediatric randomized clinical trials (RCTs) and meta-analyses. First, we screened 24 RCTs published in Pediatrics during the first 6 months of 2011; second, we screened 188 pediatric RCTs published in 2007 in the Cochrane Central Register of Controlled Trials; third, we screened 48 pediatric meta-analyses published in the Cochrane Database of Systematic Reviews in 2011. We extracted information on age ranges and age-subgroups considered and age-subgroup differences reported.
RESULTS: The age range of children in RCTs published in Pediatrics varied from 0.1 to 17.5 years (median age: 5; interquartile range: 1.8–10.2) and only 25% of those presented age-subgroup analyses. Large variability was also detected for age ranges in 188 RCTs from the Cochrane Central Register of Controlled Trials, and only 28 of those analyzed age-subgroups. Moreover, only 11 of 48 meta-analyses had age-subgroup analyses, and in 6 of those, only different studies were included. Furthermore, most of these observed differences were not beyond chance.
CONCLUSIONS: We observed large variability in the age ranges and age-subgroups of children included in recent pediatric trials and meta-analyses. Despite the limited available data, some age-subgroup differences were noted. The rationale for the selection of particular age-subgroups deserves further study.
- CDSR —
- Cochrane Database of Systematic Reviews
- CENTRAL —
- Cochrane Central Register of Controlled Trials
- GA —
- gestational age
- IQR —
- interquartile range
- RCT —
- randomized clinical trial
There is a dearth of evidence for the therapeutic efficacy of medical interventions for several clinical conditions affecting children. Pediatric evidence is often extrapolated from adult trials.1,2 This might not be justified because important discrepancies in the comparative effectiveness of medical interventions between adults and children have been observed.3 Recently, an international pediatric research network in the United States, Canada, Australia, Europe and Asia4 was created to provide guidance and improve pediatric clinical research design, conduct, and reporting. One important step forward is the standardization of the ages of patients to be included in pediatric trials and the optimal age-subgroup analyses to be performed to gain insight on the efficacy of diverse interventions in children. We set out to evaluate empirically the age ranges of children, and age-subgroup analyses thereof, reported in recent pediatric randomized clinical trials (RCTs) and pediatric meta-analyses. This empirical evaluation may help us understand the current status of the use of age groups and ranges in pediatric evidence and highlight the current inconsistencies and difficulties that remain to be overcome to optimize the design, collection, and analysis of age-related information in pediatric trials and meta-analyses.
Methods
Search Strategies
The following 3 search strategies were used.
First, we searched in PubMed for RCTs published in Pediatrics during the first 6 months of 2011. The following search terms were used: 2011 (day of publication), Pediatrics (source) and randomized controlled trial (type of article). The search was last updated on September 10, 2011. We excluded secondary publications and studies that did not use children as the unit of randomization (ie, cluster RCTs for pediatric practices). Eligible articles were reviewed in full text, and the following information was extracted: (1) age range of included children (minimum, maximum, median, interquartile range [IQR], and range); (2) whether stratification by age was done during randomization; (3) whether subgroup analyses for different pediatric age groups were done (and whether significant age-subgroup differences were identified); (4) whether treatment effect was adjusted for age; (5) whether subgroup analyses were done for other factors (besides age or study center); and (6) whether treatment effect was adjusted for these other factors.
Second, we also screened 300 pediatric RCTs published in 2007 in the Cochrane Central Register of Controlled Trials (CENTRAL) that had been collected for a previous evaluation of pediatric trials,5 and information on age ranges and age-subgroup analyses for included children was extracted. The CENTRAL comprises records of studies indexed in Medline and Embase, as well as hand-search results, gray literature, and the trial registers of Cochrane Review Groups. In 2007, a total of 2832 articles were retrieved. These were randomly ordered, using a computer-generated list, and screened consecutively for relevance. Trials published in the English language, including participants aged 0 to 18 years or participants with upper age limit ≤21 years (for trials including both children and adults), were considered for further analysis. The first 300 RCTs meeting these criteria (∼10%) were considered eligible. These were further screened, and RCTs revealing the age range of included children (n = 188) and age-subgroup analyses (n = 28) were included in this report.
Third, we also screened the Cochrane Database of Systematic Reviews (CDSR) for pediatric meta-analyses published in 2011. The following search terms were used: children (title), Cochrane Database of Systematic Reviews (source limit), and 2011 (year limit). The search was last updated on September 10, 2011. We excluded systematic reviews without forest plots and protocols. From eligible pediatric meta-analyses, we extracted information for the following: (1) age-subgroup analyses and age-subgroup differences and (2) subgroup analyses for other factors-besides age and subgroup differences. Furthermore, we screened the age ranges of individual RCTs included in a random sample (n = 14) of 30% of identified eligible pediatric meta-analyses.
Results
Pediatric RCTs
Age Ranges of Children Included in Pediatric RCTs
Among the 49 articles identified with the first search strategy, 24 eligible RCTs published in Pediatrics during the first 6 months of 2011 were considered for further evaluation (Table 1). After excluding 8 RCTs in newborns/preterm infants (only gestational ages [GAs] on admission were provided), the age range of included children varied from 0.1 to 17.5 years (median age: 5 years; IQR: 1.8–10.2 years). The minimum age of included children in these 24 trials ranged from 0 to 15 years (median age: 5 years; IQR: 0.6–7.5 years); the maximum age of included children ranged from 0.3 to 21 years (median age: 12 years; IQR: 6–17 years).
Age Ranges in the 24 Screened RCTs Published in Pediatrics During the First 6 Months of 2011
In the 188 eligible RCTs that reported age ranges (Appendix) identified in the CENTRAL 20075 with the second search strategy, the age ranges of the included children were very varied, as shown in Fig 1.
Age ranges of included children in 188 RCTs from the CENTRAL 2007 (that reported age ranges).
Age-Subgroup Analyses in Pediatric RCTs
Six trials (25%) among the 24 eligible RCTs6–29 published in Pediatrics during the first 6 months of 2011 (identified with the first search strategy) presented age-related subgroup analyses (treatment by age interaction)6,9,13,15,23,28 (Table 2), and only 3 trials (13%) reported that they adjusted the treatment outcome analysis for age at baseline.9,10,29 Nine trials (38%) evaluated subgroup effects for other factors besides age and study center; the most common factor being gender (n = 5).9,11,21,23,29 Eight trials (33%) adjusted the treatment outcome analysis for other factors; the most common again being gender (n = 3)9,21,29 (Table 2).
Subgroup Analyses (by Age and by Other Factors Besides Age) Reported in RCTs Published in Pediatrics During the First 6 Months of 2011
Some examples of reported age-subgroup differences in outcome were the following (Table 2): (1) Failures were observed more frequently in patients younger than 3 years compared with patients older than 3 years, in a trial evaluating equimolar nitrous oxide/oxygen versus placebo for procedural pain in children.6 (2) An obesity intervention trial (comparing diet alone, activity alone, and activity plus diet), evaluating changes in metabolic parameters for participating children9, identified a statistically significant effect in the 24-month low-density lipoprotein cholesterol levels (among 12 other metabolic variables studied) in age-adjusted analyses. (3) A trial evaluating the immunogenicity of a monovalent 2009 pandemic influenza A/H1N1 MF59-adjuvanted vaccine13 in children aged 6 to 23 months who had different GAs at birth identified that the geometric mean titers in the subjects with the lowest GA aged 6 to 11 months were significantly lower than those in the subjects with the same GA but aged 12 to 23 months. However, the authors did not draw any conclusions based on this age-subgroup difference. Their overall conclusion was that a single dose of the 2009 influenza vaccine evoked an immune response in children aged 6 to 23 months (including those with a GA of <32 weeks) that can be considered protective. (4) A trial examining the analgesic effects of EMLA cream and oral sucrose during venipunctures15 evaluated the correlation between postnatal age at venipuncture and pain scores during venipuncture. There was no significant correlation identified (Spearman ρ coefficient was 0.207; P = .073). (5) A trial evaluating cephalexin versus clindamycin for uncomplicated pediatric skin infections23 identified that more children >1 year of age had improved at 48 to 72 hours (primary outcome) than children 6 to 12 months (97% vs 76%, respectively; P = .004). (6) A trial evaluating the effects of serving high-sugar cereals on children’s breakfast-eating behavior28 identified that there was a main effect of cereal consumption according to age. Older children (8–12 years) consumed 30% more cereal in total than did younger children (5–7 years) (P = .03). Additional sugar poured from packets did not differ according to age, and there were no significant interactions between age and condition. Younger children in the low-sugar condition consumed more orange juice compared with younger children in the high-sugar condition, but older children consumed similar amounts in both conditions. Older children consumed 39% more calories in total than did younger children (P < .01).
With the second search strategy, we identified 28 RCTs30–57 (among the 300 screened trials published in the CENTRAL 20075) that reported age-subgroup analyses for the included children. The reported age-subgroups (and the age ranges of the included children) varied considerably across those trials (Fig 2).
Age-Subgroup Analyses in Recent Pediatric Meta-analyses (CDSR 2011)
Among 48 pediatric meta-analyses58–105 identified with the third search strategy in the CDSR (Table 3), only 11 had performed subgroup analyses for the treatment effect according to different age groups. These pertained to the following: (1) acyclovir for treating varicella in otherwise healthy children and adolescents58; (2) antibiotics for preventing complications in children with measles62; (3) antibiotics for the prevention of acute and chronic suppurative otitis media in children64; (4) gastroesophageal reflux treatment of prolonged nonspecific cough in children and adults67; (5) antifungal therapy in infants and children with proven, probable, or suspected invasive fungal infections75; (6) corticosteroids for the prevention and treatment of postextubation stridor in neonates, children, and adults80; (7) home fortification of food with multiple micronutrient powders for health and nutrition in children under 2 years of age81; (8) oral zinc for treating diarrhea in children94; (9) vitamin A for preventing acute lower respiratory tract infections in children up to 7 years of age103; (10) vitamin A for treating measles104; and (11) vitamin A supplementation for preventing morbidity and mortality in children from 6 months to 5 years of age.105
Subgroup Analyses by Age and by Other Factors Besides Age Reported in Pediatric Meta-analyses From the 2011 CDSR
The age ranges of the age-subgroup analyses in those pediatric meta-analyses varied substantially as described in Table 3.
In 6 of these meta-analyses, the reported age-subgroup analyses were done by using patients from different studies and among those, age-subgroup differences were reported in 5 meta-analyses (Table 3). Most of these observed differences were not beyond chance, and the 95% confidence intervals of the treatment effects estimates in the age subgroups were substantially overlapping. Specifically, (1) acyclovir nominally statistically decreased the “time to no new varicella lesions” only in the age-subgroups 2 to 12 years and 13 to 18 years but not in the subgroup 5 to 16 years.58 However, there was no comparison of nonoverlapping age groups to test if they have formally different treatment effects, and any claims for subgroup differences can be spurious. (2) Antibiotics prevented acute otitis media or chronic suppurative otitis media only in the subgroup of patients >12 months, but the difference found was not beyond chance.64 (3) Corticosteroids significantly prevented stridor incidence only in children but not in neonates; however, this observed difference was not beyond chance.80 (4) Home fortification of foods improved hemoglobin levels only in the 6 to 11 months subgroup and mixed-ages subgroup but not for the 12 to 23 months subgroup; however, this observed difference was also not beyond chance.81 (5) Oral zinc significantly decreased diarrhea by day 7 in the subgroup more than 6 months but was apparently not effective for the subgroup less than 6 months. The difference between these 2 age-subgroups was beyond chance, but all patients included were from different studies.94
Within-same-study age-subgroup analyses were done in only 3 meta-analyses,62,104,105 and among those, 2 meta-analyses revealed age-subgroup differences. However, none of those differences was necessarily beyond chance. Specifically, (1) antibiotics significantly increased complication rate from measles in the subgroup of children aged less than 2 years, whereas there was no statistically significant effect in children aged <1.62 However, there was no comparison of nonoverlapping age groups to test if they have formally different treatment effects; (2) vitamin A significantly reduced mortality in children with measles in the subgroup younger than 2 years, but there was no significant effect in the subgroup of children older than 2 years; however, the difference was not beyond chance.104 Thus it seems that, similar to other fields, investigators misleadingly focus on showing nominally significant results for 1 subgroup and no nominally significant results in another, instead of properly testing for formal treatment-age interaction to prove that the difference of effect sizes is beyond chance.
Subgroup Analyses by Other Factors Besides Age in Recent Pediatric Meta-analyses (CDSR 2011)
Moreover, subgroup analyses by other factors besides age were performed in 9 meta-analyses62,64,81,87,91,94,103–105 (Table 3). Among those, only 5 meta-analyses reported at least 1 subgroup analysis in which analyzed patients were from the same studies,64,87,94,103,105 and of those, 2 meta-analyses claimed subgroup differences. These pertained to the following: (1) antibiotics prevented acute otitis media or chronic suppurative otitis media only in the subgroup of patients with high antibiotic compliance, but the difference was not beyond chance,64 and (2) vitamin A prevented acute lower respiratory tract infections in the subgroup of underweight and stunted children, whereas it was harmful for the subgroup of normal weight children.103 This difference was beyond chance; nevertheless, only 1 study was included in this meta-analysis. All the other reported subgroup analyses were performed by using patients from different studies.
Age Ranges of Included/Participating Children in Recent Pediatric Meta-analyses (CDSR 2011)
We also screened the age ranges of individual RCTs included in a random sample of ∼30% (14 of 49) of the previously identified pediatric meta-analyses from the 2011 CDSR. Fourteen pediatric meta-analyses with a total of 196 included pediatric RCTs were reviewed (Table 4). The age ranges of included/participating children were reported in 116 of 196 RCTs (according to the CDSR reports). The minimum age ranged from 0 to 13 years (median age: 1.5 years; IQR: 0.5–3 years); the maximum age ranged from 0.3 to 24.8 years (median age: 10 years; IQR: 5–14 years); and the age range varied from 0.17 years to 24.3 years (median age: 7.29 years; IQR: 4–11 years).
Age Ranges of Children in the 116 RCTs Included in 14 Meta-analyses From the 2011 CDSR (116 of 196 Included RCTs Reported Age Ranges)
Discussion
We observed large variability in the age ranges of children considered appropriate by different research teams to be included in recent pediatric trials and pediatric meta-analyses, even for trials targeting similar interventions and similar conditions.
Variability was also detected in the definitions of reported age-subgroup analyses. There was a dearth of evidence for age-subgroup analyses, and only 1 in 5 pediatric meta-analyses reported such analyses. Apparently, the power to detect formal age-treatment interactions must have been limited for single trials. Power may be better for meta-analyses, but still the majority of pediatric meta-analyses simply used patients from different studies for age-subgroup analyses, and it was rare for meta-analyses to be able to split trial populations into age subgroups that could then be consistently synthesized across trials.
In age-subgroup analyses derived from different trials, it is possible that the presence or absence of age-subgroup differences might be confounded by the effect of several other factors that might be different between the included studies. Moreover, all the age-subgroup differences found, except for 1 pediatric meta-analysis,94 were not beyond chance, and the 95% confidence intervals of the treatment effects estimates in the age subgroups were substantially overlapping. Even for this single case found, the clinical significance of the finding is unclear because only different studies were included in the age-subgroup analysis. Furthermore, some reported age-subgroup comparisons had overlapping age groups in the compared subgroups. Any age-subgroup difference claims from such comparisons can be spurious; moreover, one may wonder whether those were selected posthoc.
There was also a dearth of data for subgroup analyses for factors other than age. Different studies were included in the majority of the subgroup analyses for other factors in the studied pediatric meta-analyses. We found only 1 meta-analysis103 that identified a subgroup difference beyond chance by using same-study patients; however, the significance of this finding is unclear because only 1 study was included.
Despite the limited available data, some age-subgroup differences were noted. The validity and clinical importance of those reported age-subgroup differences remains to be established, and the rationale for the selection of the particular age-subgroups deserves further study.106 These important issues will only be able to be fully assessed when trials in children use consistent, clearly reported age groups.
Appendix: References for the 188 RCTs identified in the CENTRAL 2007
Acknowledgment
The authors thank Annabritt Chisholm for her assistance in data extraction and verification.
Footnotes
- Accepted March 23, 2012.
- Address correspondence to Despina G. Contopoulos-Ioannidis, MD, Division of Infectious Diseases, Department of Pediatrics, Stanford University School of Medicine, 300 Pasteur Dr, Stanford, CA 94305. E-mail: dcontop{at}stanford.edu
Dr Contopoulos-Ioannidis, Dr Williams, and Ms Seto wrote the first draft of the article; Dr Hamm, Ms Thomson, Dr Hartling, Dr Ioannidis, Dr Curtis, Dr Constantin, Dr Batmanabane, and Dr Klassen contributed to the writing of the article and provided input via meetings and e-mail; Dr Contopoulos-Ioannidis, Ms Seto, Dr Hartling, and Ms Hamm participated in identifying and extracting the data; and Dr Contopoulos-Ioannidis, Ms Seto, Ms Hamm; Ms Thomson, Dr Hartling, Dr Ioannidis, Dr Curtis, Dr Constantin, Dr Klassen, and Dr Williams agreed with the final version.
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
References
- Copyright © 2012 by the American Academy of Pediatrics
In this issue
