Making the Most of Missing Data

Clinical trials are demanding, resource intensive, and essential to the health and well-being of children. In this issue of Pediatrics, Li and colleagues¹ outline the magnitude of an often-overlooked problem within neonatal and perinatal clinical trials. Their results demonstrate that missing primary outcome data are often ignored or mishandled, which decreases statistical power, increases bias, and may jeopardize the validity of clinical trial results. The authors provide practical approaches to avoid missing outcome data through clinical trial design strategies and review sensitivity analyses and imputation methods to mitigate the risk of bias. The problem of missing data is not unique to the neonatal and perinatal population,² and is common in trials focused on pediatric chronic medical conditions, including HIV,³ obesity,^4,5 and asthma.⁶ Thus, the concepts reviewed by Li and colleagues apply broadly.

Although clinical trials in pediatric populations are challenging, trials in the neonatal and perinatal population contend with ethical and practical considerations that make studying this vulnerable population increasingly difficult.^7,–9 With sick and preterm infants surviving hospitalization at higher rates,¹⁰ there is increasing effort from clinical researchers, clinicians, and families to standardize and prioritize the most meaningful outcomes.^11,12 Many of these critical outcomes center on long-term quality of life and neurodevelopment, which may not be observed until well into childhood and have life-long implications. These important outcomes ascertained years after hospital discharge greatly increase the potential of missing data.

Li and colleagues outline how clinical researchers and clinicians should be cognizant of the potential for missing outcome data and take steps to mitigate this risk. These efforts must start during the design stages before the clinical trial begins, when statisticians should be involved early to consider how attrition may affect sample size considerations and to identify sensitivity analyses best suited for the trial. Clinical researchers can consider an alternative trial design, such as a decentralized clinical trial¹³ or direct-to-family trial,¹⁴ to improve the ease of follow-up and decrease trial-related burdens on participant families. Decentralized clinical trials use technology, home health providers, and local clinics for data collection to reduce geographic, time, and travel related barriers for participant families and may improve participant inclusivity. Direct-to-family trials are similar, but also foster the role of parents and caregivers in supporting clinical research activities. During the trial, it is necessary for researchers to engage participants and their families, whereas minimizing burdens and maximizing trial-related benefits. After the trial, sharing results with families helps maintain the relationship and express gratitude for participation in the trial.

Reporting guidelines, such as the Consolidated Standards of Reporting Trials statement, may benefit from including missing outcome data reporting standards.¹⁵ Medical journals should insist as part of the review process that missing outcome data be reported and that best practice sensitivity analyses are used to inform the uncertainty that results from missing data. Clinicians must be sensitive to the risks of missing outcome data when evaluating clinical trials and incorporating knowledge gained from such trials into clinical practice. However, it is necessary to take a nuanced view toward missing data. Indeed, some clinical trials with relatively large amounts of missing outcome data have had a substantial impact on clinical practice, especially when little other evidence exists.¹⁶ Clinicians can gain confidence in the results of trials with missing outcome data if the trials are thoughtfully designed, methods are transparent, and sensitivity analyses are used to mitigate uncertainty.

The problem of missing outcome data are a threat to the validity of neonatal and perinatal clinical trial results and therefore a threat to the health and well-being of children. Although necessary to generate these important results, clinical trials are expensive and resource intensive. Many of the essential interventions outlined by Li and colleagues to address missing outcome data during the trial can further increase costs, including expanded statistician support, more clinical research coordinator resources, flexible home and nonstandard visits for data collection, family and community engagement, and transportation and compensation for families. These costs must be accounted for in budgets upfront, and sponsors should acknowledge that successful trials require sufficient funds to support retention efforts. Clinicians, researchers, families, and policy makers must continue to advocate for the resources necessary to perform high quality, consequential clinical trials to improve the health of children.