Table 1.

Strengths and limitations of approaches to studying the effectiveness of GH treatment as a growth-promoting agent

Study design components	Strengths	Challenges and Limitations
Data collection
Prospective	Data collection can be designed to answer the study question Participants can be randomized to treated and untreated cohorts Equitable recruitment of participants from different demographics is possible	Requires more resources Participants must be recruited prospectively and retained Prolonged study duration, if both groups are followed to adult height Bone age acceleration can lead to overestimated efficacy in short-term studies
Retrospective	Quicker (data already extant) Feasibility	There may be selection bias, as patient characteristics (eg, shorter stature) may have influenced treatment decisions Harder to find a matched control group May be harder to cleanly define the participant characteristics, especially if assays for measuring key biomarkers have changed or if clinician diagnoses are not standardized All the desired data may not be available May be less generalizable, especially if secular changes have occurred
Participant recruitment
Randomized, controlled trial	Least prone to bias Prospective data collection	Must recruit the right kind and matching participants Participants may not want to risk being in the control group (ie, prefer off-label treatment)
Observational cohort	Easier to recruit participants	May be prone to selection bias
Registry	Can accumulate a large study population → increases statistical power Can collect long-term longitudinal data Can collect “real-world” data	May be prone to selection bias May be prone to ascertainment bias No untreated control group to determine effect size May be harder to cleanly define the participant characteristics, especially if assays for measuring key biomarkers have changed or if clinician diagnoses are not standardized
Case series	Feasibility Can be exploratory and provide preliminary data to inform more robust studies	Smaller sample → less power or may not be amenable to statistical analyses at all May be prone to selection bias No control group
Primary outcome
(Near)Adult height	Gold-standard outcome for assessing total height gain from GH treatment	Takes longer to collect data Requires more resources Must recruit/retain participants
Short-term outcomes (height velocity and/or change in height SDS)	Feasibility Can more rapidly discard ineffective treatments	Small measurement errors get amplified in calculating annualized height velocity Can overestimate the impact of GH on adult height when bone age also accelerates
Comparator/control
Untreated participants with same condition	Best group to control against confounders Prospective data collection	May be prone to selection bias Participants may not want to risk being in the untreated group (ie, prefer off-label treatment)
Historical controls with same condition	Data already extant	May be harder to cleanly define the participant characteristics, especially if assays for measuring key biomarkers have changed or if clinician diagnoses are not standardized Similarly affected controls may not be able to match closely Prone to confounding and less generalizability if treatment practices have changed with time
Referent growth chart	Readily available data Provides population-level “normal” ranges Do not need additional participants as controls	Does not take personal genetic growth potential into account
Sex-adjusted midparental height	Growth potential that is more specific to the individual Do not need additional participants as controls	Self- or partner-reported heights are frequently inaccurate Children often do not grow to midparental height (ie, sex-adjusted mean) exactly, especially if there is a large spread between the parents’ heights
Self at baseline: projected adult height	Do not need additional participants as controls	Projected heights have a ± 2-inch statistical error Projected heights vary depending on projection model used
Self at baseline: height SDS or pretreatment height velocity	Do not need additional participants as controls	Small measurement errors get amplified in calculating annualized height velocity Requires lead-in time for pretreatment data; either follow longer to measure or accept clinical height data which may not be as accurate Does not account for bone age delay at start of treatment

Additional challenges and limitations common to all approaches include publication bias and issues related to funding source.

Abbreviations: GH, growth hormone; SDS, standard deviation score.