Cochrane COVID-19 Study Register
Choe 2020e

Statistical mitigation of COVID-19 pandemic disruptions to growth data collection in clinical studies

  1. Study Type
  2. Modelling
  1. Study Aim
  2. Epidemiology
  3. Prevention
  1. Study Design
  2. Other
  1. Intervention Assignment
  2. Not Applicable

Statistical mitigation of COVID-19 pandemic disruptions to growth data collection in clinical studies

Choe Y, Baggs GE, Steele C, Williams T, Mackey A
Rationale: Conducting clinical trials during the pandemic is fraught with challenges like missed visits/ procedures, visits out of window, and subject discontinuation. We seek to show that predictive growth models are adequate descriptors of infant weight (wt), length (ln), and head circumference (hc) under various patterns of missing and visits outside protocol windows. Methods: Data from two ongoing and two historical studies were bootstrapped and used for simulation. Goodness of fit was evaluated by comparing how close the observed values were with those estimated by the Gompertz model using t-tests. The Gompertz is a nonlinear growth function. The model may be fitted using ≥ 3 points – the start, end, and at least a middle point. We assumed 1,2, or 3 missing middle points under different patterns of missing. To quantify clinical relevance, differences in means between predicted vs. observed for 1000 bootstrapped studies were compared to the maximum allowable differences (MAD) between repeat measurements for acceptable precision: 100 g for wt, 0.7 cm for ln, and 0.5 cm for hc, and to the 1-month WHO weight velocity standards. Results: There were no significant differences between observed and fitted data for the simulated or bootstrapped distributions (all p>0.1006). On average, predicted values differed from observed by no more than the MAD or are less than or close to the WHO 1-month weight increments that infants growing at the 1st percentile should achieve. On a study basis, in ≥ 80% of the 1000 studies, the 95% confidence interval for the mean difference (predicted – observed) were bound by ½ SD on either side. Collectively, these show that differences were not clinically relevant. Conclusion: The Gompertz model has good predictive properties for estimating infant growth at the regulatory required timepoints even in the presence of missing values or collection outside visit windows and provide an analysis strategy to mitigate the impact of Covid-19. References: N/A. Disclosure of Interest: Y. Choe Other: YC is an employee of Abbott Nutrition, G. Baggs Other: GB is an employee of Abbott Nutrition, C. Steele Other: CS is an employee of Abbott Nutrition, T. Williams Other: TW is an employee of Abbott Nutrition, A. Mackey Other: AM is an employee of Abbott Nutrition