Background
Methods
Study design and sample
Article and forest plot screening
In meta-analyses, it is possible to formally test whether an intervention has different effects across subgroups based on patient, intervention, or study characteristics. When investigating subgroup differences in individual trials, subgroup analyses are based on within-trial comparisons. In other words, the subgroups of patients being compared come from the same study and population [35]. In meta-analyses, comparisons can be either within or between-trials. Within-trial comparisons are only possible if meta-analyses have access to individual participant level data and some of the trials contribute patients from every subgroup level considered [35]. However, subgroup analyses in meta-analyses are traditionally based on between-study comparisons, where certain trials only contribute data to one subgroup level [35]. When conducting a subgroup analyses for a meta-analyses, the two main steps are as follows: (1) calculating the effects within each subgroup level and (2) comparing the summary effects across the subgroup levels, using either fixed or random models within and between subgroups (e.g., a study can have random-effects within, and fixed effects between) [36]. In RevMan, the official Cochrane review software, a formal test for interaction is conducted using Cochran’s Q test, which tests the null hypothesis that the subgroup effects are the same and that any variation is no more than what would be expected by chance alone. The results of tests for interaction are typically present at the bottom of forest plots, with the text “Test for subgroup differences” [37]. If subgroup analyses are based on between-trial comparisons, the trials contributing data to different subgroup levels can have different patient, intervention, or study characteristics, which can complicate the conclusions from an interaction test [35]. |
Data abstraction
Statistical analysis
Analysis of clinical relevance of significant age-treatment interactions
Sensitivity analyses
Patient involvement
Results
Search results
Frequency of planned age-treatment subgroup analyses
Characteristics of eligible Cochrane reviews
There were 9 reviews with 32 individual age-treatment subgroup analyses based on potentially overlapping subgroup levels (e.g., mean age < 50 years, mean age 50 to < 65 years, vs. mean age 65+ years). The majority (25 of 32, 78.1%) of these subgroup analyses were in reviews that specified plans to conduct subgroup analyses in their methods section. Almost two thirds (20 of 32, 62.5%) of the analyses reported a P value from an interaction test, of which four (4 of 20, 20.0%) were statistically significant. Standardization by effect measures (mean difference or risk ratio) and a random effects model did not change the number of statistically significant age-treatment interactions; however, using a fixed effects model resulted in three additional statistically significant age-treatment interactions. Among the 12 age-treatment subgroup analyses without a reported P value from an interaction test, five analyses were from three reviews that outlined plans to conduct an age-treatment subgroup interaction test. |
Age-treatment subgroup analyses in Cochrane reviews
Frequency and characteristics of statistically significant age-treatment interactions
No. (%) of statistically significant age-treatment interactions | |||
---|---|---|---|
P value for interaction reported in forest plots | P value for interaction not reported in the forest plots | Total | |
All age-treatment analyses | |||
Using analytical methods reported in forest plotsa | 7/51 (13.7) | 4/14 (28.6) | 11/65 (16.9) |
Standardized using a fixed effects modelb | 8/49 (16.3) | 5/14 (35.7) | 13/63 (20.6) |
Standardized using a random effects modelb | 7/49 (14.3) | 4/14 (28.6) | 11/63 (17.5) |
Corroboration of significant age-treatment subgroup analyses
Clinical translation of statistically significant age-treatment interactions
Comparison | Population characteristics | Outcome | Primary outcome* | Specified in methods?† | Reported age subgroup levels | Effect size [95% CI]‡ | RCTs per subgroup level | RCTs across all levels§ | P value | Biological/clinical rationale¶ |
---|---|---|---|---|---|---|---|---|---|---|
Botulinum toxin vs surgery (2017) | Children and adults with strabismus suitable for treatment with botulinum toxin to align the angle of deviation | Improved ocular alignment ≤ 10 prism dioptres | Yes | No | Children | RR 0.91 [0.71, 1.16] | 2 | 0 | 0.04 | No |
Adults | RR 0.38 [0.17, 0.85] | 1 | ||||||||
Digital intervention vs no/minimal intervention (2017) | Individuals in the community with hazardous or harmful alcohol consumptions directed toward any digital intervention | Quantity of drinking (grams/week), based on the longest follow-up | Yes | Yes | Adolescents/young adults | MD − 13.44 [− 19.27, − 7.61] | 28 | 0 | 0.002 | No |
Adults | MD − 56.05 [− 82.08, − 30.02] | 14 | ||||||||
Supplementary feeding vs comparator by age of children (2012) | Children from low- and middle-income countries born at term, from birth to 5 years old | Weight gain (kg) during the intervention | Yes | Yes | Children younger than 24 months | MD − 0.01 [− 0.09, 0.07] | 4 | 0 | 0.008 | No |
Children older than 24 months | MD 0.22 [0.07, 0.37] | 1 | ||||||||
Diet plus physical activity vs comparator (2017) | Individuals diagnosed with intermediate hyperglycemia or prediabetes at increased risk of developing type II diabetes mellitus | Incidence of type 2 diabetes | Yes | Yes | Ages < 50 years | RR 0.70 [0.57, 0.85] | 2 | 0 | 0.009 | No |
Ages ≥ 50 years | RR 0.50 [0.44, 0.58] | 9 | ||||||||
2-h plasma glucose | No | Yes | Ages < 50 years | MD − 1.62 [− 2.49, − 0.76] | 2 | 0 | 0.003 | No | ||
Ages ≥ 50 years | MD − 0.27 [− 0.49, − 0.05] | 7 | ||||||||
Colony-stimulating factor plus antibiotics vs antibiotics alone (2014) | Individuals undergoing chemotherapy for cancer who experienced neutropenia and fever | Time to neutrophil recovery | No | Yes | Children | RR 0.80 [0.66, 0.97] | 1 | 0 | 0.02 | No |
Adults | RR 0.45 [0.29, 0.70] | 5 | ||||||||
Fluticasone propionate vs beclomethasone dipropionate or budesonide, parallel-group studies: dose ratio 1:2 subgroup by age (2007) | Children (> 2 years) and adults with a clinical diagnosis of asthma | Change in Forced Expiratory Volume 1 compared to baseline | Unclear | Yes | Children | MD − 0.04 [− 0.10, 0.02] | 2 | 0 | 0.02 | No |
Adults | MD 0.04 [0.00 0.08] | 10 |